Epigenetics in the nervous system

Video Transcript

• 00:00 - 00:11:  Hello, everybody. Thank you all for joining. I hope you're all comfortable and settled
• 00:11 - 00:16:  in. My name is Kumaran, and on behalf of Abcam, I would like to welcome you to our Epigenetics
• 00:16 - 00:21:  in the Nervous System virtual talks. This is the third and final live session in our
• 00:21 - 00:26:  Spotlight on Neuroscience event and follows on from our Adult Neurogenesis and Building
• 00:26 - 00:32:  and Repairing the Brain sessions. With the current pandemic continuing to cause havoc
• 00:32 - 00:38:  with our working lives, we at Abcam are pleased to support our global neuroepigenetics community
• 00:38 - 00:45:  by providing this digital platform to help you stay connected and continue to share and showcase
• 00:45 - 00:51:  your amazing work and ideas. As with the previous sessions, I have a quick 30-second poll to kick us
• 00:51 - 00:56:  off. We have a truly global audience with us online today and with participants from all over
• 00:56 - 01:01:  the world, and so we thought it would be really interesting to see how COVID is still impacting
• 01:01 - 01:08:  our working lifestyles and our ability to return back to normal. We are also interested in your
• 01:08 - 01:18:  background and what position in science you currently occupy. Great, thank you for answering
• 01:18 - 01:23:  those questions. As you can see, the results are on your screen now and are in line with
• 01:23 - 01:28:  what we've seen in the previous sessions, with the majority of us only partially being able
• 01:28 - 01:34:  to return back to work. Nearly 30% of us are still just fully working from home,
• 01:34 - 01:40:  but there is a lucky 18% that have managed to return back to work fully, so great for them,
• 01:40 - 01:46:  and it looks like the tide is slowly turning. Right, so now back to the science. So today's
• 01:46 - 01:52:  session will focus on some of the cellular and molecular strategies used to explore
• 01:52 - 01:59:  the biological mechanisms that dynamically control gene activity, expression, and function. This is
• 01:59 - 02:05:  an area of science that Abcam is proud to support, and following this event, you will receive an
• 02:05 - 02:10:  email with a link to both our neuroscience and epigenetics landing page where you can explore
• 02:10 - 02:14:  some of the tools and reagents that we make to support researchers in this field.
• 02:16 - 02:24:  Today, this afternoon's session on epigenetics in the nervous system will be chaired by Dr.
• 02:25 - 02:32:  Gonzalo Castiello Branca and Professor Anna Pombo. Dr. Gonzalo is an Associate Professor
• 02:33 - 02:38:  of Neurobiology at the Karolinska Institute in Stockholm. His research group focuses on the
• 02:38 - 02:44:  molecular mechanisms regulating the epigenomic states of oligodendrocyte lineage cells in
• 02:44 - 02:50:  neuroinflammation and demyelinating diseases. His long-term goal is to build a solid platform
• 02:50 - 02:56:  of convergent knowledge and know-how on the epigenetics of remyelination and neuroinflammation
• 02:56 - 03:01:  and to promote the establishment of innovative regenerative strategies for neurological
• 03:01 - 03:06:  disorders. He received his PhD in Medical Biochemistry from the Karolinska Institute
• 03:06 - 03:11:  in Sweden, after which he completed his postdoc training at both the Karolinska Institute and
• 03:11 - 03:16:  the University of Cambridge before eventually moving back to Stockholm to start his lab in 2012.
• 03:17 - 03:21:  He has received numerous awards and grants including the European Research Council
• 03:21 - 03:27:  Consolidated Grant and the Swedish Society for Medical Research 100 Years Jubilee and Prize.
• 03:29 - 03:35:  His co-chair, Anna Pombo, is a Deputy Scientific Director of the Berlin Institute for Medical
• 03:35 - 03:41:  Systems Biology and the Scientific Program Deputy Chair for the MDC. She pioneered the
• 03:41 - 03:47:  development of the genome architecture mapping and orthogonal technology to map the 3D structure
• 03:47 - 03:53:  of chromatin. Her lab investigates the mechanisms that regulate 3D genome folding and gene expression
• 03:53 - 03:58:  during mammalian development and in disease. After receiving her doctorate from the University of
• 03:58 - 04:03:  Oxford, she was awarded a Royal Society Dorothy Hodgkin Fellowship to continue her work on the
• 04:03 - 04:10:  spatial organization of transcription in mammalian nuclei. In 2000, she started her lab at the MRC
• 04:10 - 04:16:  London Institute for Medical Sciences before being appointed Professor at Imperial College London in
• 04:16 - 04:22:  2011. Then in 2013, she received a Helmholtz Association Professorship Award and moved her
• 04:22 - 04:28:  laboratory to the Max Delbrück Centre for Molecular Medicine and was co-appointed Professor for
• 04:28 - 04:34:  Epigenetic Regulation and Chromatin Architecture at the Humboldt University of Berlin. Professor
• 04:34 - 04:42:  Pombo received the Robert Fugger Prize in 2007 and became an elected EMBO member in 2018.
• 04:43 - 04:47:  And with that, I am pleased to hand over to both Gonzalo and Anna. Thank you.
• 04:48 - 04:59:  Okay. So, thank you, Kumaran, for the introduction. And in my name, in Anna's name, I would like to
• 04:59 - 05:05:  welcome you all to this virtual Abcam meeting on epigenetics in the nervous system. This meeting
• 05:05 - 05:12:  is a follow-up of our first meeting that Anna and I organized with Abcam in Stockholm
• 05:12 - 05:22:  two years ago in October 2018. And there will be also a third iteration of the meeting in Berlin
• 05:22 - 05:32:  in June 2022. So, in two years. So, stay tuned for that meeting. So, this year's meeting,
• 05:33 - 05:39:  we are very happy that Abcam is co-organizing with us this meeting this year and we will cover
• 05:40 - 05:46:  new exciting areas emerging in neuroepigenetics, including new technologies in the single-cell
• 05:46 - 05:54:  space, novel histone modifications, also insights into how chromatin architecture affects neural
• 05:54 - 06:01:  activity and expression. And apart from these basic science approaches, the meeting will also
• 06:02 - 06:09:  some translational aspects, such as the impact of DNA methylation and the mutations at histone tails
• 06:10 - 06:13:  in neurological diseases such as Alzheimer's disease and also in gliomas.
• 06:14 - 06:21:  So, Anna and I are very happy for the fantastic line-up of speakers that have accepted
• 06:21 - 06:27:  our invitation to present at the meeting. So, we are looking forward to an afternoon
• 06:27 - 06:33:  or morning for many of you of excellent science. And with that, I'll now pass over to Anna,
• 06:33 - 06:38:  who will share and moderate the first session. So, hello everyone. Thank you for joining. It's
• 06:38 - 06:44:  really a delight to be here. And without much further ado, I start by introducing the first
• 06:44 - 06:51:  speaker, who is Dr. Jennifer Phillips-Cremins. She's an Associate Professor and Dean's Faculty
• 06:51 - 06:56:  Fellow in Engineering and Medicine at the University of Pennsylvania, where she leads the
• 06:56 -  3D Epigenomics and Systems Neurobiology Lab. She did her PhD in Biomedical Engineering at
• 07:03 -  07:11:  Georgia Institute of Technology, co-mentored by Job Decker and Victor Corses. She, since then,
• 07:11 -  07:18:  has received many awards. For the sake of time, I mention a couple. The NIH Director New Innovator
• 07:18 -  07:24:  Award, as well as, more recently, a Chan Zuckerberg Neurodegenerative Disease Grant.
• 07:24 -  07:34:  And she is also a very main member of the 4D Nucleome, where I meet her very often. And she's
• 07:34 -  07:40:  been one of the few, I think, to get two of the awards in the Phase II, which I think just shows
• 07:40 -  07:48:  how wonderful her work is in that area. As probably many of you know, she's interested
• 07:48 -  07:57:  in understanding epigenetic mechanisms that regulate neuronal phenotypes, neuronal commitment,
• 07:57 -  08:04:  not only in health, but also when they go wrong during neurodevelopment and neurodegeneration.
• 08:04 -  08:10:  And so, without much ado, and I think just on time, I pass to Jen. So, welcome, Jen. Thank
• 08:10 -  08:18:  you so much for joining us. Thank you tremendously to both Anna and Gonzalo, as well as the
• 08:19 -  08:24:  Abcam team. Good morning, or good afternoon, everyone. I'm very happy to be here today.
• 08:24 -  08:31:  And so, sort of generally speaking, I think the area that I and my team are fascinated with at
• 08:31 -  08:36:  the moment is this notion of just simply how complex neural connections truly are in the
• 08:36 -  08:42:  mammalian brain. And then, our specific sort of effort is in thinking about understanding the
• 08:42 -  08:48:  true functionality of key epigenetic and genetic features and how they may govern properties of
• 08:48 -  08:54:  synaptic plasticity, not only in the healthy developing brain, but also how these elements
• 08:54 -  09:00:  may go awry in neurological disorders. And speaking of function, I think one thing that
• 09:00 -  09:06:  has been emerging is that over time, we have grown as a community in the ability to map
• 09:06 -  09:12:  various genetic or epigenetic features and to begin to correlate with neurological phenotypes.
• 09:12 -  09:17:  But the name of the game these days is truly function. And I think many high-throughput and
• 09:17 -  09:23:  local approaches begin this perspective of perhaps gain or loss of function of a particular
• 09:23 -  09:30:  epigenetic feature with CRISPR, and then asking how this might affect expression of a nearby gene.
• 09:30 -  09:36:  So, we take a little bit of an intermediate and top-down approach before doing the functional work,
• 09:36 -  09:42:  which is to first understand how the genome folds in three dimensions within the larger nucleus,
• 09:42 -  09:47:  and then to computationally identify the spatial touch points and to begin to think about how
• 09:47 -  09:53:  epigenetic or genetic features may work through long-range regulatory mechanisms. And for those
• 09:53 -  09:59:  not in this particular field or subfield of epigenetics, I think what has emerged in terms
• 09:59 -  10:05:  of its utility is this notion that perhaps if one is to only look at this location independent of
• 10:05 -  10:10:  the rest of the genome, it would rule out the possibility that megabases and megabases away,
• 10:10 -  10:16:  there are other features that might be indeed in nuclear space spatially connected to the primary
• 10:16 -  10:22:  site, and thereby regulating the particular gene. And so, looking at all of the DNA in aggregate in
• 10:22 -  10:27:  the spatial connections has been a very fruitful area of inquiry for understanding genome function.
• 10:28 -  10:32:  And so, my group is at the current time sort of divided into three areas. We have a heavy
• 10:32 -  10:37:  technology development component, which I will not talk about today. We're also very interested
• 10:37 -  10:42:  in understanding how the 3D genome might be miswired in neurological disorders. And then,
• 10:43 -  10:48:  sort of the new area that we're beginning to push on is thinking about synaptic plasticity and how
• 10:48 -  10:53:  epigenetic marks through 3D regulation may play a role in synaptic properties. And that's
• 10:54 -  10:59:  a story that I'm going to talk about today. And so, why did we think that 3D genome folding would
• 10:59 -  11:05:  be relevant in any way, shape, or form in the case of neurons and neuronal activity? And I think
• 11:06 -  11:11:  one thing that inspired us was in digging back through the older literature, one can very quickly
• 11:11 -  11:17:  find examples such as this one. These are classic EM images in a particular neuron in which we're
• 11:17 -  11:25:  looking at a high-resolution view of the chromatin. And what we see is that upon a very brief pulse,
• 11:26 -  11:32:  less than 30 seconds of depolarization, there is a marked global change in chromatin density.
• 11:33 -  11:39:  This continues to be altered upon additional depolarization. And then, perhaps most
• 11:39 -  11:45:  interestingly, upon removal of the stimulus, there is indeed some form of reversal, and chromatin
• 11:45 -  11:51:  structure goes back to its original density, but not fully. So, there are some elements that appear
• 11:51 -  11:57:  to remain, and this is also of great fascination to us. And so, many of you know that our approach
• 11:57 -  12:03:  at the current time is to focus on both genomics and imaging approaches to understand the 3D genome.
• 12:03 -  12:10:  And we just simply wanted to know if we applied either 5C or Hi-C technologies to this problem,
• 12:10 -  12:15:  could we then begin to really understand the precise high-resolution structures which may
• 12:15 -  12:21:  or may not be changed upon stimulation of neural circuits? And so, for those that are new to this
• 12:21 -  12:27:  field, this is a classic Hi-C map, and the structure that I will be sort of focusing on
• 12:27 -  12:32:  today are these so-called long-range looping interactions. These loops are made manifest in
• 12:32 -  12:38:  Hi-C maps as these sort of spherical structures of pixels that are adjacent to each other and show
• 12:38 -  12:45:  an increased interaction frequency relative to the local subtad and tad background. And we can
• 12:45 -  12:52:  computationally identify such clusters, and then ultimately, in principle, what we believe these
• 12:52 -  12:57:  reveal is the possibility that this particular genomic fragment and this particular genomic
• 12:57 -  13:04:  fragment would interact in physical space to ultimately loop out the intervening DNA.
• 13:06 -  13:10:  And so, there's another reason that we thought that this one might be interesting. It was really
• 13:10 -  13:16:  highlighted to me by the PhD student that led this project, John Began. He recently graduated
• 13:16 -  13:21:  from my group and has gone on to do a postdoc at Yale in Daniel Cohen and Ramos' lab. And at the time,
• 13:21 -  13:26:  John really pointed out to me that the time scale that neurons regulate gene expression
• 13:27 -  13:31:  is extremely rapid, and this is quite interesting in terms of a structural perspective.
• 13:32 -  13:38:  Very briefly, upon a stimulus which leads to depolarization of a particular neuron,
• 13:38 -  13:44:  there is an activation of signaling cascades which, as a first pass in a protein-independent
• 13:44 -  13:50:  manner, will upregulate extremely quickly on the time scale of seconds to minutes a class of so-called
• 13:50 -  13:57:  immediate early genes. And the classic examples in this case are FOS or ARC. And then there is a
• 13:57 -  14:02:  second wave, and this occurs on the time scale of minutes to hours. You can see it's quite delayed
• 14:02 -  14:08:  in...as compared to the immediate early genes. And again, these are so-called secondary response
• 14:08 -  14:14:  genes that include hallmarks such as BDNF, and the upregulation occurs in a protein-dependent
• 14:14 -  14:20:  manner, so clearly a mechanistically different wave of activation. And so, the questions that we had
• 14:20 -  14:25:  were, first, are genome folding patterns altered upon stimulation of neural activity
• 14:25 -  14:32:  if we look by high-resolution genomics methods? Are the 3D epigenetic mechanisms different
• 14:32 -  14:37:  for immediate early versus secondary response genes? In a way, could we explain the timing
• 14:37 -  14:43:  of activation? And then, finally, what is the timing of any possible activity-dependent loops?
• 14:43 -  14:48:  How fast do they occur, and do they oscillate in any way, shape, or form? And I will begin
• 14:48 -  14:52:  today by providing some preliminary data toward that last question.
• 14:53 -  14:58:  And so, this is, in principle, what we really would want, which is the ability to continue to
• 14:58 -  15:03:  track neuronal firing in vivo, to watch the dynamics of these neurons fire, to keep the
• 15:03 -  15:10:  circuits intact but acquire their epigenetic information, and then ultimately take many
• 15:10 -  15:17:  rapid time points over a series of time. And ultimately, we can't have exactly what we want
• 15:17 -  15:22:  yet. And so, we began this work by beginning with in vitro systems in which we use pharmacological
• 15:22 -  15:30:  activation or inhibition of mammalian neurons. And so, this is the first result that we had.
• 15:30 -  15:36:  These are mouse cortical neurons, 5C or Hi-C maps at very high resolution, about 1 kb.
• 15:36 -  15:40:  And we're looking at either pharmacological inhibition or induction. And if you take a
• 15:40 -  15:46:  global view without looking at the fine-grained details, the surprising observation was that,
• 15:46 -  15:52:  essentially, the TADs and subtads were largely unchanged at the regions we queried. And so,
• 15:52 -  15:58:  it was difficult to reconcile this with the EM images. And we began to really say that beyond
• 15:58 -  16:03:  this sort of global qualitative observation, let's begin to identify quantitatively the
• 16:03 -  16:08:  differential changes. And these are the hypotheses that we had from that global map.
• 16:08 -  16:13:  The first is that in the case of neuronal circuits where things have to happen very rapidly,
• 16:13 -  16:18:  perhaps this is a case where distal enhancers do not function. And in fact, enhancers are
• 16:18 -  16:24:  directly adjacent to the genes that they regulate. And that is their mode of activation.
• 16:25 -  16:30:  The second hypothesis is that loops are already preformed to the genes that need to be activated
• 16:30 -  16:36:  on very short timescales. And then, enhancers would simply come into the preexisting prewired
• 16:36 -  16:42:  loops upon neural circuit activation. And then, in a third model, there would be a subset of
• 16:42 -  16:48:  loops that are indeed activity-dependent. And in parallel with the acquisition of a distal enhancer,
• 16:48 -  16:51:  a loop would be formed upon neural circuit activation.
• 16:52 -  16:58:  And so, to test these ideas, the first thing that we did was look at the first possibility. And this
• 16:58 -  17:04:  is sort of an internal control, so no surprises here. We looked at the promoter signal of H3K27
• 17:04 -  17:11:  acetyl between active and inhibition states. And ultimately, we saw, as expected, an extremely
• 17:11 -  17:15:  strong correlation with the alteration of gene expression upon neuronal activity.
• 17:16 -  17:24:  Then next, we looked at the change of H3K27 acetyl at proximal enhancers, the enhancers closest to
• 17:24 -  17:30:  the genes that they regulate. And quite surprisingly in this case, we saw no correlation
• 17:31 -  17:37:  to the activation of genes. Whereas if we then use loops, and we use the loops to target the
• 17:37 -  17:43:  particular enhancers that might be linked to their target genes, ultimately what we saw is that we
• 17:43 -  17:49:  could find an additional layer of correlation between H3K27 acetyl acquisition of the enhancer
• 17:49 -  17:56:  and the target gene. And this was able to bump up our model to explain over 75% of the variance.
• 17:57 -  18:02:  And so, from this data, I think what it enabled us to do was perhaps rule out the possibility that
• 18:02 -  18:07:  in neurons, enhancers only work through a proximal mechanism. And then we wanted to begin to explore
• 18:07 -  18:14:  the possibility of pre-wired versus de novo looping interactions. The first thing that we
• 18:14 -  18:19:  did was look at the possibility that there could be or could exist de novo looping interactions.
• 18:19 -  18:24:  And what we did was ultimately using statistical models that we and others have developed,
• 18:24 -  18:30:  we identified that nearly 10% of the loops in our systems were activated upon neuronal
• 18:30 -  18:36:  stimulation. And when we looked at the base of these loops, I'm boiling down many, many,
• 18:36 -  18:41:  many months of work into this simple cartoon, but ultimately what we saw is that in many,
• 18:41 -  18:47:  many cases, loops are induced de novo upon neuronal activation, and then there are activity-dependent
• 18:47 -  18:51:  enhancers that occur at the base of such structures. Here is one example of the classic
• 18:51 -  18:57:  FOS gene. FOS is a very small gene, and it's shown here. You can see that in the inhibition state,
• 18:57 -  19:03:  not surprisingly, it's not forming any particular structures. And then upon induction of activity,
• 19:03 -  19:09:  there is sort of an acquisition of the spherical loop-like structural feature, which I explained
• 19:09 -  19:14:  previously. And if you look at the base of that structure, there is also the acquisition of a
• 19:14 -  19:19:  particular enhancer. And so this would be an example of an enhancer that we would
• 19:19 -  19:24:  deem as extremely interesting for understanding its functionality in regulating FOS timing.
• 19:25 -  19:30:  So what about this idea of pre-poised loops? In this particular case, to clarify,
• 19:30 -  19:35:  these would be loops that are already existing in an inhibition or untreated state.
• 19:35 -  19:41:  They remain tethered upon activity induction, but there is one key change, and that is at the
• 19:41 -  19:47:  base of this already pre-tethered anchor, we do see the activation of an activity-dependent
• 19:47 -  19:52:  enhancer. And just to prove that we sort of stratified these possible looping interactions
• 19:52 -  19:58:  correctly, you can see that the activity-induced or de novo loops show a strong increase in
• 19:58 -  19:59:  interaction frequency as expected.
• 20:00 -  20:11:  Whereas this other group of two activity-invariant loops are already pre-tethered in the inhibited state and then essentially remain at the same interaction frequency.
• 20:11 -  20:18:  And so the question that we really want to know is, so what is the correlation to gene expression? And this was a bit of a surprise to us.
• 20:18 -  20:30:  And so activity-induced loops showed a marked fold change correlation to gene expression, in fact, up to 24-fold change of genes that engage in these activity-dependent loops.
• 20:30 -  20:41:  Whereas in the case of the pre-poised loops, although they have activity-dependent enhancers, we only saw a very small effect size in terms of the induction of gene expression.
• 20:42 -  20:58:  And so from a bigger picture from these correlations, ultimately the model that we have been building is this notion that poised loops, in fact, are extremely common and all over the genome, whereas dynamic loops induced by activation are extremely rare.
• 20:58 -  21:08:  And where this is sort of being manifest is in thinking about the effect size. So although poised loops are extremely common, their effect size on gene expression is quite modest.
• 21:08 -  21:19:  Whereas loops that are induced de novo are the ones that we would predict with our modeling would indeed have a strong effect on activity-dependent gene expression.
• 21:20 -  21:31:  So here is one example of what these loops look like. This is a case of secondary response genes. And another question we had was, okay, I just showed you an example of FOS.
• 21:31 -  21:45:  Quite intriguingly, there was just a singular loop, and it was very, very short. And what we noticed and what stood out to us is that in the case of BDNF, a secondary response gene, we saw a very, very different type of structural pattern.
• 21:45 -  22:01:  Here, you might notice that the map seems a lot bigger, and that is because it's zoomed out to two to three megabases. And what we ultimately saw was many, many, many different loops, upwards of 30 looping interactions, all with sizes that were much larger than the FOS locus.
• 22:01 -  22:17:  And these looping interactions were quite complex. Here's a zoom-in of just one particular loop that's about a megabase in size. And we still do see this sort of activity-dependent increase in looping interaction with the BDNF promoter with activity-dependent enhancers at the base.
• 22:17 -  22:38:  But this is one particular loop in addition to many, many other even longer structures that form with this secondary response gene. And so this led to a working model that we had that, indeed, perhaps immediate early genes, their timing is governed by the fact that they are very, very short, and there's only a few of them, if not only one.
• 22:39 -  22:51:  Whereas in the case of BDNF, we see many, many different longer-range loops. And this led us to perhaps a hypothesis that this is connected to the rapid versus delayed activation of these two gene classes.
• 22:51 -  23:18:  So here's our working model. And again, this is a model that has been published at this point, but now we're working on how we can test the functionality. And that is that immediate early genes generally that occur through, that are activated through protein-independent mechanisms, it may be that they're very, very short-range singular looping interactions play a functional role in helping to facilitate that very rapid activation.
• 23:18 -  23:37:  Whereas the case in secondary response genes, we might imagine that there would be more loops, they would be longer, there would be a much more complex environment of regulatory enhancers. And this would perhaps functionally contribute to the delayed kinetics and the protein-dependent manner of secondary response genes.
• 23:37 -  24:05:  And so this led us to one of our final questions, which is that those singular response genes now make us think that perhaps the kinetics of loops really, really matters. And so what we did was go back to the beginning and now take very fine-grained time points. Here you can see the total mRNA levels of CFOS, essentially, upon the activity response, gaining expression to the max at about 60 minutes, and then subsequently going down.
• 24:05 -  24:29:  And when we begin to compare to the dynamics of looping interactions, what stands out to us is that the looping strength is actually maximum prior to the max mRNA levels. In fact, loops occur as early as five minutes, and then ultimately obtain their maximum strength at 20 minutes. And we're particularly interested in the loop remaining tethered after the activation signal has been completed.
• 24:29 -  24:57:  The other question is the enhancers. So do they resemble the dynamics of the loops or the gene expression? And in fact, this was a surprise to us. And so the enhancer signal is shown here, overlaid over the gene expression. And ultimately, what we observed is that the enhancer signal also seems to take a max around 20 minutes in parallel with the looping interaction, but then it goes down quite sharply. And it's the gene expression mRNA levels that are a bit delayed past the enhancer and the looping interaction.
• 24:57 -  25:09:  And I would say that at the current time, one thing that we're really focused on is in looking into using Proseq and other methods for nascent transcripts to understand truly if this observation will continue to hold.
• 25:09 -  25:33:  Finally, we then asked, well, what about the dynamics of BDNF? In this case, the loops, our model might predict, would take a lot longer to form. And in fact, that is indeed what we saw. So no surprises here for the transcripts. What we see is that as opposed to CFOS, there's quite a steady, slow rise in expression, with the maximum expression or mRNA levels occurring at 24 hours.
• 25:33 -  25:45:  And then you can see that the loop slowly acquires its strength and ultimately really only is made manifest in the full looping strength in parallel with the activation of the gene.
• 25:45 -  25:53:  Okay, so I know we're getting close to time, and I'll just summarize here that this work has just very recently been published.
• 25:53 -  26:03:  We've identified that more than 10% of loops in the regions that we queried in mouse cortical neurons occurred de novo upon activation. We identified two types of loops.
• 26:03 -  26:19:  We see that activity-inducible enhancers can engage in either pre-existing loops or rare de novo-formed looping interactions. And depending on which type of structure, this has a massive difference on the correlation to the fold change in the activity-dependent gene expression response.
• 26:19 -  26:33:  Again, with the de novo activity-dependent loops having the highest fold change response in gene expression. We also see that immediate early genes seem to form simple structures, singular loops or doublets that are quite short in genomic size.
• 26:34 -  26:58:  Whereas in the case of secondary response genes, they seem to engage in an extremely complex network of long-range interactions engaging in multiple enhancers. And it is our hypothesis that the length of these loops that would require much longer timescale formation might indeed be functionally connected to the timescale at which they are activated in neural circuits.
• 26:59 -  27:18:  With that, I'll close here by just a couple of slides of where we'd like to go from here. Very recently, we've been interested in building tools to begin to control loops on demand. And one such particular tool among others that others are also developing is so-called light-activated dynamic looping.
• 27:18 -  27:32:  And the idea here is simply that at a gene such as FOS, we could now in wild-type cells or in unstimulated circuits, begin to engineer this loop on very, very short timescales and understand the functionality on nascent transcripts.
• 27:32 -  27:53:  I would say that as a larger view in the 3D genome folding fields, really getting at the genome structure-function relationship now that we can create such elegant and beautiful folding maps is really where I think a lot of fun is to be had and a lot of insight is to be gained. And so I really look forward to working alongside people like Anna and others in this next wave.
• 27:54 -  28:09:  And finally, another future direction that we envision, and this is at the end of the paper that was recently published, is that we became even more excited about these long-range looping interactions when we looked at common variants that have been linked to both autism and schizophrenia.
• 28:10 -  28:25:  And in this case, for example, autism, what we noticed is that many of the disease-associated common variants were often at the so-called class II pre-poised loops, indeed connecting them over long ranges to target genes.
• 28:26 -  28:35:  And what was very interesting is when we looked at the target genes that are connected, essentially skipping over the adjacent genes, what we found is that oftentimes those genes were activity-dependent.
• 28:36 -  28:45:  And so we think this might be a really interesting way in which common variants might work to govern perhaps changes in the activity-dependent response in disease.
• 28:46 -  28:47:  With that, I'll just stop here.
• 28:48 -  28:58:  I'd like to thank everyone in my lab who has done the work, in particular, John Began, who was an incredible sort of energy force behind all of this work.
• 28:59 -  29:07:  This work was done in collaboration with a colleague, Jason Shepard, and a postdoc in his lab, Elisa Pastusin, with key contributions from many others in the lab.
• 29:08 -  29:10:  With that, I'll stop here, and I'll be happy to take questions.
• 29:11 -  29:12:  Thank you so much, Jen.
• 29:13 -  29:17:  I encourage everyone to use the Q&A box to send your questions.
• 29:18 -  29:19:  I'll just start with one.
• 29:20 -  29:31:  So you showed us that the pre-existing loops seem to have a lesser effect, but could it be that they are there to cope with all the changes that are ongoing?
• 29:32 -  29:35:  So having those loops pre-established is what really matters.
• 29:36 -  29:43:  And do we know anything about the genomic variants hitting these more strongly than the others?
• 29:44 -  29:50:  What do you predict if you delete the pre-existing or one of the induced ones?
• 29:51 -  29:52:  So could you expand on this?
• 29:53 -  29:54:  Well, thank you for raising this.
• 29:55 -  29:57:  I think this is why we're so excited about this area.
• 29:57 -  30:03:  There is a reason why the genome has pre-wired such structures.
• 30:04 -  30:13:  And although the effect size on gene expression might be low, we don't want to make any comment about the critical nature of what these loops might represent in neural circuits.
• 30:14 -  30:25:  So I think it's just careful dissection of the loops and the enhancers, both gain-of-function and loss-of-function, would probably be needed before we determine their particular functional role.
• 30:25 -  30:28:  But to your second point, essentially you hit it dead on.
• 30:29 -  30:41:  So that last slide I showed with the autism-associated variants, nearly all of the variants that we looked at, there was an extremely strong enrichment specifically for the class you raised, which is those loops that are pre-poised.
• 30:42 -  30:48:  And they're already pre-poised to their target gene, and then the activity-dependent enhancer comes in at that particular loop anchor.
• 30:49 -  30:52:  That's where all of the autism-associated variants were located.
• 30:52 -  30:58:  And why we thought that was very interesting is when we moved over to a disease like schizophrenia, this was not the case.
• 30:59 -  31:05:  So schizophrenia was in a very different looping class that had a very different functional effect on gene expression.
• 31:06 -  31:11:  So we thought it was just very interesting that autism variants in particular would be at the base of these looping structures.
• 31:12 -  31:18:  And then these would be loops that connect to activity-dependent genes that go up upon neural circuit activation.
• 31:18 -  31:22:  So there could be guesses and predictions of what could happen to such loops in autism.
• 31:23 -  31:24:  Thank you, Jen.
• 31:25 -  31:27:  So we also have a question from Rita Souza Nunes.
• 31:28 -  31:29:  Hello, Rita.
• 31:30 -  31:35:  Did you see any correlation between the localization topology of the loops and their response?
• 31:36 -  31:38:  Formation, immediate, early, secondary?
• 31:39 -  31:46:  For example, near the nuclear envelope or nuclear pore complexes or elsewhere?
• 31:46 -  31:47:  Yeah, thank you.
• 31:48 -  31:59:  So Rita is asking about the link of our looping interactions from the Hi-C map to global nuclear architecture placement with respect to lamina or nucleolus or other structures.
• 32:00 -  32:02:  And we haven't delved into that yet.
• 32:03 -  32:15:  But I think we or perhaps others, this seems to be a very important area for the Hi-C field generally is to begin to think about linking the large-scale nuclear architecture.
• 32:16 -  32:20:  Nuclear organizational properties to the fine-scale looping events.
• 32:21 -  32:24:  And I just think it's going to require an integrative approach.
• 32:25 -  32:26:  And it's not something we've done yet.
• 32:27 -  32:30:  But I believe that many, including Anna's lab, are thinking about such a problem.
• 32:31 -  32:34:  So now we have two questions which more or less are asking the same thing.
• 32:35 -  32:38:  One from Kumaran from Abcam and the other from Giuseppe Lupo.
• 32:39 -  32:46:  And I think Kumaran is asking whether there are chromatin remodeling complexes that are recruited to these long-range looping interactions.
• 32:47 -  32:49:  And Giuseppe about chromatin marks.
• 32:50 -  32:57:  For example, histone modifications that accompany the looping dynamics for especially the secondary response genes.
• 32:58 -  33:02:  Yeah, so both questions are alluding to now the mechanism of the two formations.
• 33:03 -  33:06:  And could it be related to chromatin remodelers?
• 33:06 -  33:12:  Are there particular epigenetic marks beyond K27 acetyl that could lead to some predictive power?
• 33:13 -  33:23:  Or I would even add on to that, are there different architectural proteins or architectural mechanisms that would yield very different mechanistic phenomena for these two looping classes?
• 33:24 -  33:29:  Thus far, the majority of what we've done is begin to understand the motifs at the base of such loops.
• 33:30 -  33:34:  And to begin to hypothesize how we would work in the perturbative sense.
• 33:34 -  33:40:  So I suppose in our group, before we pursue mechanism, we want to make sure something is functional.
• 33:41 -  33:47:  But I appreciate that one could easily reverse that approach and begin to understand mechanism and assess function later.
• 33:48 -  33:49:  Both are probably complementary.
• 33:50 -  33:52:  At the current time, we're really focused on how do we kill these loops?
• 33:53 -  33:56:  How do we kill these enhancers to assess if they even matter for the activity response?
• 33:57 -  34:00:  So we have, I think I'm going to allow one more question.
• 34:01 -  34:02:  So we stay on time.
• 34:02 -  34:03:  Elisa Moreno.
• 34:04 -  34:05:  And thank you.
• 34:06 -  34:07:  Very interesting talk.
• 34:08 -  34:11:  Did you ever think of addressing the role of activity-dependent loops in interneurons?
• 34:12 -  34:18:  For example, in the olfactory bulb where activity is crucial in promoting the maturation and localization of granule cells.
• 34:19 -  34:21:  So neuroscientist question.
• 34:22 -  34:23:  It's a really cool idea.
• 34:24 -  34:28:  And there's just not enough projects and time to do everything.
• 34:29 -  34:30:  But I think it's pretty amazing.
• 34:30 -  34:34:  Stavros Lombardis has done quite cool work in this particular area.
• 34:35 - 34:41:  And also Anna's group is beginning to crank out some amazing publications to begin to look at rare neuronal populations.
• 34:42 - 34:47:  And I think the GAM technology is particularly wonderfully suited for rare populations.
• 34:48 - 34:53:  And especially like in situ, in place, beginning to think about how these neurons might function.
• 34:54 - 34:55:  So it's a great question.
• 34:56 - 34:59:  And if they function the same as excitatory neurons, it would be very interesting.
• 35:00 - 35:03:  Could you do in 20 seconds about CTCF?
• 35:04 - 35:07:  So what is CTCF doing or not doing?
• 35:08 - 35:10:  And I think we have to finish this.
• 35:11 - 35:15:  I think this probably is something that the 3D genome aficionados could guess.
• 35:16 - 35:21:  But what we see is that CTCF is often at the base of nearly all of the prepoised looping interactions.
• 35:22 - 35:23:  No surprises.
• 35:23 - 35:30:  However, when we look at the de novo interactions, at FOSS, at ARC, at others, very rarely do we see CTCF.
• 35:31 - 35:35:  And so that would allude to the possibility that there's a very different mechanism going on.
• 35:36 - 35:38:  And we don't know yet if they're cohesin-dependent or not.
• 35:39 - 35:44:  But that is something that perhaps someone like Markus Merkenschlager will be able to shed light on in the coming weeks.
• 35:45 - 35:46:  Fantastic. Thank you so much, Jennifer.
• 35:47 - 35:48:  Thank you so much, Yen, also for joining.
• 35:49 - 35:52:  I've been wanting to hear you give a talk for ages.
• 35:53 - 35:56:  It happens to be in this conference that we organized.
• 35:57 - 36:03:  So Yen is an associate professor at the Icahn School of Medicine at Mount Sinai.
• 36:04 - 36:11:  He initially studied microbiology at Ohio State University before moving to neurosciences,
• 36:12 - 36:16:  joining Eric Nestler's lab at Mount Sinai School of Medicine.
• 36:17 - 36:26:  He then, after this step into neurosciences, went to David Allis's lab to work more on the epigenetic side.
• 36:27 - 36:32:  And he then started his lab at the Icahn School of Medicine at Mount Sinai in 2014.
• 36:33 - 36:39:  He has also received several awards, including the NARSA Young Investigator Award,
• 36:39 - 36:42:  the Harold and Golden Laporte Basic Research Award,
• 36:43 - 36:47:  and in 2019, the Presidential Early Career Award for Scientists and Engineers.
• 36:48 - 36:55:  And his lab investigates chromatin and histone regulatory mechanisms contributing to neuroplasticity and disease.
• 36:56 - 37:01:  And I think your work is extremely original, and I look forward, I think we all look forward to hearing your talk.
• 37:02 - 37:03:  Thank you, Yen, for joining us.
• 37:04 - 37:05:  All right, perfect.
• 37:05 - 37:09:  So first, of course, I'd like to thank the organizers for inviting me here to speak today,
• 37:10 - 37:11:  and of course, for all of you for coming.
• 37:12 - 37:13:  I should apologize in advance.
• 37:14 - 37:18:  I'm dealing with a little bit of a cold, non-COVID-related, but if I break into a coughing fit, you know why.
• 37:19 - 37:26:  Okay, so today I'm going to tell you a little bit about some of the work that we've been involved in over the last few years now,
• 37:27 - 37:33:  which has been focused on studying novel neurotransmission-independent roles for biogenic monoamines,
• 37:33 - 37:39:  things like serotonin and dopamine, as direct regulators of transcriptional plasticity within the nervous system.
• 37:40 - 37:44:  And this, of course, is occurring through the direct covalent modification of histone proteins.
• 37:45 - 37:49:  And so today, I'm going to be giving you a bit of an update on an extension project,
• 37:50 - 37:54:  where we've now identified and will be studying histone H3 histaminylation,
• 37:55 - 38:01:  and then studying potential roles for this modification as a potential novel regulator of circadian rhythmicity in the brain.
• 38:01 - 38:04:  And so, I guess I have to use this.
• 38:05 - 38:10:  Okay, so before I start, though, of course, it'd be remiss to not mention my postdoc, Ryan Bastl,
• 38:11 - 38:12:  who has really spearheaded this project in my lab.
• 38:13 - 38:15:  He's absolutely phenomenal.
• 38:16 - 38:20:  Okay, so as some of you may or may not be aware, over the last couple years,
• 38:21 - 38:27:  we've begun to publish on the discovery and initial characterization of numerous new histone modifications.
• 38:27 - 38:30:  Again, these so-called histone monoaminylation states,
• 38:31 - 38:36:  whereby serotonin, dopamine, other neurotransmitters can effectively serve as donor sources
• 38:37 - 38:39:  for the covalent modification of histone proteins.
• 38:40 - 38:44:  And these are kind of interesting because these different histone monoaminylation modifications
• 38:45 - 38:46:  exist only on one histone.
• 38:47 - 38:48:  They're found on histone H3.
• 38:49 - 38:52:  They're found at one far-end terminal glutamine residue, glutamine 5.
• 38:53 - 38:56:  And of course, glutamine modifications on histones are generally pretty rare.
• 38:57 - 39:00:  And we started to do some initial characterizations to try to understand
• 39:01 - 39:03:  what types of proteins may deposit these modifications.
• 39:04 - 39:07:  We've identified that there's a protein called the tissue transglutaminase 2 enzyme
• 39:08 - 39:09:  that can put on these modifications.
• 39:10 - 39:12:  And we've also started to try to understand mechanistically
• 39:13 - 39:15:  what they may be doing in the context of transcriptional plasticity.
• 39:16 - 39:20:  And so using a variety of different systems that range from early cellular differentiation
• 39:21 - 39:24:  or brain development systems all the way into adulthood plasticity models,
• 39:24 - 39:27:  we've been able to show that these different monoaminyl modifications,
• 39:28 - 39:30:  at least in the case of serotonylation and dopaminylation,
• 39:31 - 39:35:  they can exist in conjunction with the adjacent lysine 4 trimethylation,
• 39:36 - 39:38:  which as many of you probably know is an important modification
• 39:39 - 39:43:  in the recruitment of the pre-initiation complex to drive transcriptional permissiveness.
• 39:44 - 39:47:  And we've basically found that the addition of these monoaminyl marks
• 39:48 - 39:51:  in the context of H3K4 trimethyl serve generally two roles.
• 39:51 - 39:55:  One of them seems to be to hyper-recruit or increase the binding
• 39:56 - 39:58:  of the general transcription factor complex TF2D.
• 39:59 - 40:01:  This happens through a particular protein called TAF3, which is a subunit of that complex. And this seems to
• 40:02 - 40:04:  TF2D, has a high level of transcriptional permissiveness.
• 40:04 - 40:09:  basically potentiate transcription. And more recently in a paper currently under revision,
• 40:09 - 40:14:  we've found that not only does this serotonin mark, for example, increase the recruitment of TF2D,
• 40:14 - 40:21:  but it also antagonizes the ability of erasers like KDM5 proteins, which take off K4 from
• 40:21 - 40:25:  trimethyl, from actually taking it off. And so we've again, of course, you know,
• 40:25 - 40:29:  we're still studying this a lot mechanistically, but we've started to look in the context of
• 40:30 - 40:36:  adult environmental exposures that may regulate donor availability. So serotonin release dynamics,
• 40:36 - 40:40:  dopamine release dynamics. And we've began to publish on some of these in the case earlier
• 40:40 - 40:44:  this year, looking at histone dopamineylation and showing that it is indeed responsive
• 40:44 - 40:49:  to withdrawal from drugs or abuse, for example, and that it seems to be functionally important
• 40:49 - 40:55:  both for transcriptional plasticity and also behavior. So of course, given the fact that
• 40:55 - 41:00:  we know that things like serotonin and dopamine can be transaminated to this histone H3 protein,
• 41:00 - 41:06:  we were interested early on in whether or not, you know, this is a classical mechanism of all
• 41:06 - 41:11:  linear hydrophobic monoamine neurotransmitters or not. And so early on, we had also been
• 41:11 - 41:16:  investigating histamine. Histamine, like the other monoamines, is known to be found
• 41:16 - 41:21:  in nuclei of the brain. Kind of surprisingly, when I started these projects, I didn't imagine
• 41:21 - 41:25:  that you would have so much extra vesicular monoamine present in the nucleus, but it turns
• 41:25 - 41:29:  out that there's a lot of precedence for this. You know, even going back to some very nice
• 41:29 - 41:34:  papers from the early 1970s, this is an example from Ann Young and Saul Steiner, where they were
• 41:34 - 41:39:  actually measuring histamine content in the nuclear and cytoplasmic fractions during embryonic
• 41:39 - 41:44:  and early postnatal brain development. And what they had found is that actually during early
• 41:44 - 41:48:  postnatal phases, approximately 90% of all the histamine content in the brain was found in the
• 41:48 - 41:54:  nucleus. And this reduces a bit as you get into our later postnatal phases. However, there is still
• 41:54 - 41:58:  a significant amount there. Now, in these early papers, of course, they didn't describe what they
• 41:58 - 42:02:  thought it may be doing in the nucleus. They just showed that it was there. And of course, I should
• 42:02 - 42:07:  mention that we've observed nuclear histamine content in the brain as well, using cellular
• 42:07 - 42:14:  fractionation and ELISA-based assays. So as my lab does to try to figure out then whether or not,
• 42:14 - 42:19:  theoretically, this can happen, we started with some basic enzymatic assays in vitro.
• 42:20 - 42:24:  Given that we know the depositing enzyme, the writer enzyme, this tissue transglutaminase 2,
• 42:25 - 42:29:  what we decided to do is start just on peptide reactions, where we synthesize
• 42:30 - 42:34:  histone-tail peptides, we perform in vitro transamination reactions, and then we ask,
• 42:34 - 42:39:  one, can it happen? And two, on which sites do we see this modification? And long story short,
• 42:39 - 42:42:  similar to what we found with histone serotonylation and dopaminylation, we found
• 42:42 - 42:48:  that histone serotonylation can only occur on the far terminal H3 tail at glutamine 5,
• 42:48 - 42:53:  and we have good MS/MS data to support that. But I think most importantly, and more recently,
• 42:53 - 42:57:  in collaboration with Simone Cedoli at Albert Einstein College of Medicine, we've been able
• 42:57 - 43:03:  to show that this H3Q5 histamine modification not only exists in vitro when we force the system,
• 43:03 - 43:09:  but it also exists endogenously within cells, indicating that there could be some functional
• 43:09 - 43:18:  significance to its deposition. So as we do, we decided then to develop some tools so we could
• 43:18 - 43:24:  study this in a more of a dynamic fashion, as opposed to doing mass spectrometry on every
• 43:24 - 43:29:  sample. So we generated rabbit polyclonal antibodies that are site-specific for H3Q5
• 43:29 - 43:34:  histidine. We've recently started projects to make monoclonals as well. And we were able to show
• 43:34 - 43:38:  that, of course, these antibodies are very specific, just like our antibodies for serotonylation and
• 43:38 - 43:43:  dopaminylation. There's no cross-reactivity with other monoaminyl states. We can get a signal to
• 43:43 - 43:48:  occur enzymatically on peptides in the context of nucleosomes with no cross-reactivity. And even in
• 43:48 - 43:53:  brain nuclear extracts, we're able to see that the modification exists, and it can be appropriately
• 43:53 - 43:59:  blocked through peptide competition assays. So of course, using this novel tool, we now wanted to
• 43:59 - 44:03:  try to understand its dynamics and understand what it may be doing functionally. So very quickly,
• 44:03 - 44:07:  I just want to mention a little bit about histamine and its distribution in the brain.
• 44:08 - 44:13:  So histamine is produced in a very small subpopulation of posterior hypothalamic neurons
• 44:13 - 44:17:  in what's called the tuberomammillary nucleus. These are histidine decarboxylase-containing
• 44:18 - 44:23:  neurons. And just to show you in a mouse brain, this is a sagittal section, coronal section,
• 44:23 - 44:28:  that when you stain for HDC, you find that it only exists in this very small nucleus, as I said,
• 44:28 - 44:35:  the TMM. But interestingly, the histaminergic neurons, while relatively sparse, project broadly
• 44:35 - 44:40:  throughout the brain with innervation to the cortex, hippocampus, amygdala, striatum, and so
• 44:40 - 44:46:  forth. So it does exist broadly in terms of at least its release dynamics. So based on that,
• 44:46 - 44:50:  we started by just basically running brain-wide blots, looking at different brain regions,
• 44:50 - 44:56:  trying to identify where the H3Q5 histamine modification may enrich. Like with the other monoaminyl
• 44:56 - 45:00:  modifications, we do see that there is some ubiquitous expression, although perhaps very
• 45:00 - 45:05:  low expression across the brain. However, we do find that in the TMM, where we have the
• 45:05 - 45:09:  histaminergic neurons, that not surprisingly, we get a higher enrichment for this particular
• 45:09 - 45:14:  modification. So this is going to be the area that we're going to be focused on for the remainder of
• 45:14 - 45:19:  the studies I talk about. Okay, so the other interesting thing about histamine and its
• 45:19 - 45:26:  release dynamics is that histamine is very heavily correlated to circadian rhythmicity and locomotor
• 45:26 - 45:31:  activity in animals across the light-dark cycle. Now, of course, keep in mind that mice are
• 45:31 - 45:36:  nocturnal animals, so they show elevated activity during their dark phase and reduced activity
• 45:36 - 45:41:  during their light phase. Obviously, this is the opposite in humans. But what's been shown
• 45:41 - 45:46:  many times over is that correlating with this increased activity during their active phase,
• 45:46 - 45:51:  we also see quite a bit of histamine being released from these TMM neurons. And then as
• 45:51 - 45:56:  you enter this more asleep phase, you start to see histamine release dynamics shutting down.
• 45:56 - 46:00:  This, of course, makes sense to anyone who's taken antihistamine drugs, and they make them a little
• 46:00 - 46:07:  bit drowsy. So the goal was to basically take TMM tissues from mouse brain across the entire
• 46:07 - 46:11:  Zeitgeber. The Zeitgeber is going to be basically just the entire light-dark cycle,
• 46:11 - 46:17:  and then measure for H3Q5S levels to see whether or not these release dynamics, these changes in
• 46:17 - 46:22:  release dynamics that happen over the circadian time course, whether they lead to changes or
• 46:22 - 46:28:  correlate with changes in the modification itself. And so what we found is basically just that.
• 46:28 - 46:35:  So what we're doing is taking tissue across the Zeitgeber every four hours, basically for 24 hours.
• 46:35 - 46:41:  This is a high end, 17 to 19, so we feel very strongly in these trends. And basically
• 46:41 - 46:47:  what we found is that H3Q5S does indeed show this somewhat cyclical pattern of regulation
• 46:47 - 46:51:  that's significant by cubic trend analysis. And what's interesting is that when you look at these
• 46:51 - 46:54:  phases, this awake phase, where you have a lot of histamine being released from these
• 46:54 - 46:59:  histaminergic neurons, we're actually seeing that the modification is showing a loss of expression,
• 46:59 - 47:03:  which may indicate that with loss of donor availability, you're losing the mark.
• 47:03 - 47:09:  And then as you enter the inactive phase, where histamine release dynamics go down,
• 47:09 - 47:13:  you start to see that there's a slight gradual accumulation of the modification
• 47:13 - 47:18:  over that time point. And what's important here too, is that the TMN, like many tissues and many
• 47:18 - 47:22:  different cell types, I think this has been beautifully shown by the late Sasson-Corsi,
• 47:22 - 47:29:  Paolo Sasson-Corsi's lab and others, that the TMN itself does display very robust patterns of
• 47:29 - 47:33:  circadian gene expression. So in this case, this is basically the same Zeitgeber analysis,
• 47:33 - 47:38:  but now using RNA sequencing and then using a method adapted from what's called JTK cycle,
• 47:38 - 47:42:  which is an algorithm for detecting rhythmic components in genomic data sets. And in doing
• 47:42 - 47:47:  so, we're able to identify that the TMN displays very robust circadian gene expression. When you
• 47:47 - 47:53:  map using platforms like ChIA to try to identify what upstream regulators may be of these circadian
• 47:53 - 47:59:  genes, not surprisingly, we find that things like CLOCK, which is a dominant transcription factor
• 48:00 - 48:04:  regulating circadian expression, it itself is, of course, enriched as a potential upstream regulator.
• 48:04 - 48:08:  And then as a sanity check, we can go in and dig into certain specific circadian genes,
• 48:08 - 48:13:  things like the period genes, period one, period two, CLOCK, BMAL, and so forth. And again,
• 48:13 - 48:19:  even in our TMN data set, we do see that there are these circadian fluctuations. And so of course,
• 48:19 - 48:24:  the question is, might this H3Q5 histaminyl modification be linked to circadian gene
• 48:24 - 48:31:  expression in this brain region? And then, of course, if so, how? So I should also just quickly
• 48:31 - 48:35:  mention that some of our previous reports, as I said in the second slide, have been focused on
• 48:35 - 48:41:  these combinatorial interactions between K4 trimethyl and Q5 dopamine and serotonin. We also
• 48:41 - 48:47:  designed antibodies that allow us to specifically target those dual modifications. We were able to
• 48:47 - 48:52:  show in vitro that they can at least in theory exist. We've made these dual modified antibodies
• 48:52 - 48:57:  that will be publicly released soon. They're very specific. But very interestingly, what we found is
• 48:57 - 49:02:  that this dual histaminyl modification doesn't actually show any circadian arrhythmicity,
• 49:02 - 49:07:  unlike the single modification, which may suggest that the histaminyl dynamics that we're seeing
• 49:07 - 49:12:  are independent of its adjacent role or combinatorial role with H3K4 trimethyl.
• 49:13 - 49:17:  And of course, I should mention that the transglutaminase enzyme itself is not circadian
• 49:17 - 49:21:  in this region. And we have profiled Q5 hist across other non-histaminergic brain regions,
• 49:21 - 49:24:  and we don't see any circadian pattern or expression.
• 49:25 - 49:28:  So of course, then the question was, I think the obvious question was, what about
• 49:28 - 49:33:  alterations in circadian rhythms in sleep? Can these things themselves affect the expression
• 49:33 - 49:38:  of H3Q5 hist? So one of the first experiments that Ryan did was he wanted to perform an experiment
• 49:38 - 49:43:  where we gave animals Ambien or Zolpidem. And the idea here was to give them this drug during their
• 49:43 - 49:48:  typical active phase, and then collect tissues and see whether or not the histaminyl mark may
• 49:48 - 49:55:  be responsive to that kind of sleep promotion. And so you can see here
• 49:55 - 50:01:  very clearly, when you give animals Ambien, you see a very rapid drop in their overall locomotor
• 50:01 - 50:06:  activity. This is sustained for some period of time. And so the idea was to give them Zolpidem
• 50:06 - 50:10:  and then basically eight hours into their active cycle, take tissue and see whether or not the mark
• 50:10 - 50:18:  may show a response to that. So again, just kind of validating what we had seen before, we took
• 50:18 - 50:23:  tissues from this eight-hour time point versus 20 hours. This is a time in which we previously saw
• 50:23 - 50:28:  that the mark goes down. And again, we were able to replicate this in an independent cohort of
• 50:28 - 50:32:  animals. What's interesting is that if you promote sleep during that normal active phase, you
• 50:32 - 50:39:  basically take the levels of H3Q5 hist and you revert them back to what they would be during a
• 50:39 - 50:44:  typically inactive state. So that's kind of interesting. The other thing that we wanted to
• 50:44 - 50:48:  look at is what about in a genetic model of circadian defects, you know, where of course
• 50:48 - 50:55:  circadian rhythms are dramatically altered. So in this case, we were using the CLOCK Delta-19
• 50:55 - 50:59:  mice. These were developed by Joe Takahashi, and the mice were given to us by Colin McClung at
• 50:59 - 51:05:  University of Pittsburgh. And the basic idea here was to compare wild type versus mutant animals
• 51:05 - 51:10:  where clearly their overall circadian rhythmicity is greatly affected. And so again, what we wanted
• 51:10 - 51:15:  to do was take tissue from this period during their active state, eight hours into their active
• 51:15 - 51:20:  state, and just ask whether or not CLOCK mutants show some kind of alteration in their overall Q5
• 51:20 - 51:25:  hist levels. And basically that's what we found similar to the Zolpidem experiment. We were able
• 51:25 - 51:29:  to see that at this time point, there's a significant elevation of the Q5 hist mark in the
• 51:29 - 51:34:  CLOCK mutant animals, which I think is encouraging. Now we're going to basically look at it across the
• 51:34 - 51:42:  entire Zeitgeber, but I do think that it's intriguing to begin. Excuse me. Okay, so then the
• 51:42 - 51:46:  question would be then what might H3Q5 hist be doing mechanistically that could regulate
• 51:46 - 51:51:  circadian gene expression if that's what it's doing? And so as many of you are probably aware,
• 51:51 - 51:56:  circadian gene expression is tightly controlled by a series of transcriptional feedback loops.
• 51:56 - 52:01:  Again, this is true in brain tissues and in other cellular systems. The basic idea being that you
• 52:01 - 52:05:  have these master regulators of circadian gene expression, things like CLOCK and BMAL.
• 52:05 - 52:09:  They dimerize to bind to EBOX promoters, and of course they drive the expression of many of these
• 52:09 - 52:14:  downstream circadian genes that we all know and love, things like period, cryptochrome, etc.
• 52:14 - 52:21:  The idea being that as these circadian genes are translated, they will dimerize in the cytoplasm
• 52:21 - 52:25:  through phosphorylation events. They're shuttled into the nucleus to then inhibit CLOCK and BMAL,
• 52:25 - 52:29:  and so starts the cycle of being able to regulate gene expression. And this has been implicated in
• 52:29 - 52:35:  everything from sleep-wake cycles to body temperature regulation, metabolism, and so forth.
• 52:36 - 52:41:  What's also interesting, and this was a beautiful work from Sasson-Corsi's lab,
• 52:41 - 52:46:  is that there is an interplay between CLOCK regulation of circadian gene expression and
• 52:46 - 52:50:  the recruitment of the MLL1 complex, which is an important histone methyltransferase complex
• 52:51 - 52:56:  involved in the deposition of H3K4 methylation, right? And so the idea was that CLOCK and MLL1
• 52:56 - 53:01:  are co-recruited to circadian promoters in a cyclic manner, and in the case of, for example,
• 53:01 - 53:06:  the Delta-19 mice, they can't interact with MLL1, the complex is not recruited, and therefore you
• 53:06 - 53:15:  lose this normal rhythmic fashion of circadian expression. And so what is MLL1? So MLL1 and its
• 53:15 - 53:23:  associated complex, CET1, are again this kind of common complex that's involved in the ability to
• 53:23 - 53:29:  add methyl groups to H3 lysine 4. While the methyltransferase itself, whether it's CET1 or
• 53:29 - 53:34:  MLL1, may differ between the two complexes, there are common subunits between these complexes,
• 53:34 - 53:41:  and in particular there's a protein called WDR5, which contains WD40 repeats, and basically its
• 53:41 - 53:47:  job is to present an unmodified lysine 4 side chain, and this allows basically MLL1 or CET1
• 53:47 - 53:52:  to come closer and promote the methylation and therefore transcriptional activation.
• 53:53 - 53:58:  So the first thing we did is we wanted to take a look at the different methylation states on
• 53:58 - 54:02:  H3K4, either mono, di, or trimethyl, to try to figure out whether or not they themselves may
• 54:02 - 54:08:  show circadian patterns of expression similar or opposite to Q5 His. I do want to just mention,
• 54:08 - 54:12:  we looked at monomethyl, it's not shown here, we don't see it regulated in a circadian fashion,
• 54:12 - 54:17:  but it's also regulated by a different set of MLL proteins. Now interestingly in our hands,
• 54:17 - 54:21:  we don't see any change in H3K4 trimethyl across the Zeitgeber, this is consistent I think
• 54:21 - 54:26:  with our dually modified antibody, or dual modification specific antibody data I showed
• 54:26 - 54:32:  you before, but very interestingly this dimethyl state of H3K4 does indeed show a very nice
• 54:32 - 54:38:  circadian pattern of expression, and it's a seemingly opposite pattern to that of H3Q5
• 54:38 - 54:42:   Okay, so this is of course intriguing, it makes us wonder whether or not there may be some
• 54:42 - 54:50:  antagonistic relationship between the two. So to test this initially, we decided to go into
• 54:50 - 54:56:  the brain and basically do a dominant negative-like approach, where we can express a histone
• 54:56 - 55:02:  variant H3.3, which is one of the histone H3s that can actually incorporate into neuronal chromatin,
• 55:02 - 55:06:  where we express either a wild type version of this H3.3 that could be histaminylated,
• 55:06 - 55:11:  or one in which we've converted this glutamine to an alanine, and therefore it no longer has this
• 55:11 - 55:16:  monoaminylation capacity. And so what we did is we injected our viruses into the TMN,
• 55:16 - 55:21:  validated of course that they express well, and then we micro-dissected those tissues and blotted
• 55:21 - 55:27:  for either Q5 His or H3K4 methylation states. And in doing this, as predicted, we're able to
• 55:27 - 55:32:  show that we can, through this dominant negative approach, reduce H3Q5 His levels in the TMN,
• 55:32 - 55:38:  and this very interestingly corresponded to an elevation in H3K4 dimethyl states,
• 55:38 - 55:43:  all right, so again supporting this potential idea of an antagonistic relationship between the two.
• 55:43 - 55:48:  And I should mention, although not shown here, that neither H3K4 mono nor H3K4 tri
• 55:48 - 55:53:  are actually affected by these Q5 His reductions, so we do believe there's specificity to this
• 55:53 - 55:57:  system. So very briefly, I don't want to belabor this other than to say that there are differences
• 55:57 - 56:03:  between the different methylation states on H3K4. So H3K4 trimethyl is found typically near the
• 56:03 - 56:08:  transcriptional start site, near promoters. It's very important for transcriptional initiation.
• 56:08 - 56:13:  H3K4 monomethyl has been associated with enhancers and more distal elements,
• 56:13 - 56:17:  and H3K4 dimethyl is actually found more broadly distributed throughout
• 56:17 - 56:23:  the coding region of active genes, and people believe that it may be a mark of transcriptional
• 56:23 - 56:26:  fidelity. There's a few different thoughts on this. I won't go into it too much other than
• 56:26 - 56:32:  to say that some people believe that the dimethyl state is actually serving a repressive role
• 56:32 - 56:36:  because it can recruit HDACs. Some people, the Shalottaford Lab, for example, has shown that
• 56:36 - 56:41:  it may serve to protect regions of the genome from being K27 trimethylated,
• 56:41 - 56:46:  promoting a repressive state through inhibition of PRC2 and so forth. But in any case, I think
• 56:46 - 56:49:  its role is a little less characterized, certainly, than that of the trimethyl state.
• 56:50 - 56:54:  So if there is this antagonistic relationship, the question is, how might this work?
• 56:55 - 57:01:  So what we did is we dug a little bit into the structure of WDR5, again, this protein within
• 57:01 - 57:06:  the MLL1 complex that's known to bind to the tails in order to promote these methylation states.
• 57:06 - 57:12:  What we noticed that was very interesting is that when the unmodified H3 tail is bound to WDR5,
• 57:12 - 57:18:  to its WD40 repeats, the glutamine 5 residue would actually be predicted to lie outside of
• 57:18 - 57:24:  the binding pocket, actually out onto the surface. When we modeled this, it looked like that glutamine
• 57:24 - 57:29:  should be very close to a lysine residue. Now, this is interesting because unlike serotonylation
• 57:29 - 57:33:  and dopaminylation, histaminylation, histamine itself, is actually positively charged.
• 57:33 - 57:37:  And so the thought would be that maybe the Q5 hist, the presence of this mark,
• 57:37 - 57:45:  would be predicted to basically promote some kind of repulsion, electrostatic repulsion,
• 57:45 - 57:49:  because if you have a positively charged glutamine 5 with histamine, you may actually
• 57:49 - 57:56:  repel that lysine residue. So to test this, we basically purified the WD40 repeats from WDR5
• 57:56 - 58:01:  and performed isothermal titration calorimetry against either unmodified or Q5 histamine-related
• 58:01 - 58:06:  peptides for H3 in order to ask whether or not there was a decrease in binding. So what you can
• 58:06 - 58:11:  see clearly here is that the histamine peptide actually does show about a five-fold decrease
• 58:11 - 58:17:  in binding. If we neutralize that positive charge in WDR5, we basically can equalize the binding.
• 58:17 - 58:22:  So it does suggest that that interaction is important. And then finally, of course,
• 58:22 - 58:26:  we can convert it to a glutamic acid, adding a negative charge and even get a potentiation.
• 58:26 - 58:30:  So with all of this being true, of course, the big question is, well, if you have a histamine-related
• 58:30 - 58:35:  peptide, is MLL1 less able to deposit these methyl states? Would it antagonize that
• 58:36 - 58:42:  interaction? And I just want to mention that we perform MALDI-TOF using purified MLL1 complexes.
• 58:42 - 58:47:  These are in vitro reactions, adding SAM and the different peptides. And this allows us to measure
• 58:47 - 58:52:  each methylation state as they're being added in a time course. And so again, a lot of data here,
• 58:52 - 58:57:  I just want to point out that if you compare, for example, an unmodified histone tail where
• 58:57 - 59:02: You start this methylation reaction; you can very quickly go from unmodified to mono and
• 59:02 - 59:07: dimethylated in the system. You see that serotonylation has no impact on that process.
• 59:07 - 59:12: However, Q5 histamination does indeed slow that process way down. So you can see, for example,
• 59:12 - 59:19: that even at 24 hours in the unmodified peptide state, you're completely at dimethylated K4.
• 59:19 - 59:25: However, it's stalled to an extent in the Q5 hist. Okay. So just the last two slides,
• 59:25 - 59:29: very quickly. The question then is, of course, does this have functional significance in the
• 59:29 - 59:34: context of circadian rhythmicity? I just want to mention that, you know, we started by performing
• 59:34 - 59:39: ChIP sequencing across the Zeitgeber in the TMN, looking at H3K4 dimethyl. We wanted to understand
• 59:39 - 59:44: if it itself was circadian as may be predicted from our Western blotting data. And then we also
• 59:44 - 59:50: wanted to overlay these data, again, using the JTK output algorithm with circadian genes to try to
• 59:50 - 59:55: identify how many of the circadian genes that we see show circadian regulation of that modification.
• 59:55 - 59:59: So we find that about 10% of all of them do. And this is just looking at heat maps. I think
• 60:00 - 60:05: they cluster very nicely. And then this is just showing you that of these overlapping
• 60:05 - 60:10: genes, we find many different circadian genes, things like Period 1, Period 2, DBP, BMAL,
• 60:10 - 60:14: etc. And again, corresponding to these alterations in expression over the Zeitgeber
• 60:14 - 60:18: from our data set. I don't know if you can appreciate it here, but you are seeing fluctuations
• 60:18 - 60:22: where the marks enrichment of K4 dye is showing reductions and then enhancements over that
• 60:22 - 60:26: time course. Now, I wanted to be able to show you Q5S ChIP sequencing today. Unfortunately,
• 60:26 - 60:30: we don't have it, but stay tuned. It is in the works. And that will obviously bring this
• 60:30 - 60:36: all together. And then the final step, of course, was we wanted to use our Q5A dominant
• 60:36 - 60:41: negative approach to then manipulate the modification within the TMN and then ask of the differentially
• 60:41 - 60:45: regulated genes through this manipulation, how many of them are circadian and what types
• 60:45 - 60:50: of genes are they? Long story short, we see many genes that are dysregulated. Many of
• 60:50 - 60:54: them are predicted to be clock regulated genes themselves. So they do have circadian function.
• 60:55 - 60:58: And there's a lot of interesting things going on with respect to a lot of these circadian
• 60:58 - 61:02: genes. We're looking at transition periods here where a lot of them are reduced. And
• 61:02 - 61:06: we see that a lot of immediate early genes are also circadian in nature, show these global
• 61:06 - 61:11: reductions. And so we're starting to tease that apart a little bit. And then finally,
• 61:11 - 61:17: we've also started to look at locomotor behavior across these circadian time courses. Simple
• 61:17 - 61:21: locomotor assay in which animals have been injected intra-TMN with either our wild type
• 61:21 - 61:26: or Q5A viruses. We start them out in normal light/dark conditions, and then we switch
• 61:26 - 61:30: them to dark/dark so we can actually see pure circadian behavioral phenotypes. And
• 61:30 - 61:34: what we're starting to find is that as you extend into these dark/dark periods, we're
• 61:34 - 61:39: starting to see these shifts where transitions into what would normally be their inactive
• 61:39 - 61:43: state seem to be delayed. But of course, to tease that apart more, we need to do more
• 61:43 - 61:49: sophisticated analyses. And so Ryan is taking monumental efforts now to perform these EMG
• 61:49 - 61:54: EEG recordings so we can really tease apart different aspects of circadian rhythmicity,
• 61:54 - 61:59: sleep, separating REM, non-REM sleep, and so forth. And so just a last slide to end,
• 61:59 - 62:03: you know, the basic model is this idea that during transitions into wake phases,
• 62:03 - 62:09: the Q5S levels will increase. This itself will reduce the binding of WDR5, thereby allowing
• 62:09 - 62:13: for lower levels of the methylated state. Maybe there's an interplay with protein,
• 62:13 - 62:17: the protein complexes like PRC2, we don't know. But then I think most importantly during the
• 62:17 - 62:22: transition into sleep, we see that these levels decrease. And this seems to allow for greater
• 62:22 - 62:26: binding of WDR5 to promote those methylation states. And I just want to highlight the most
• 62:26 - 62:31: important thing that I can't show you today because it's not quite ready for publication,
• 62:31 - 62:35: but soon, is of course, the big remaining question as to how does this modification
• 62:35 - 62:40: cycle so quickly? How does it turn over within 24 hours? Is there an active D-monoamine oxidase
• 62:40 - 62:43: that takes it off? This has not been published yet. And I just want to tell you that the answer
• 62:43 - 62:48: is absolutely yes, we know what it is. And so please stay tuned. That will probably be
• 62:48 - 62:53: part of this overall paper. So on that, many people to thank. I will just let people
• 62:53 - 62:58: look at it. Again, just want to say, Ryan Bassel, huge efforts in this, really phenomenal.
• 62:58 - 63:03: We have amazing collaborators from all over the world, Tom Muriel, David Haital, Lee Simone
• 63:03 - 63:07: and so forth. And of course, all my funding sources. Thank you.
• 63:07 - 63:14: Thank you so much, Ian. This was very exciting. We're a bit over time, so we may not be able
• 63:14 - 63:22: to do so many questions. But so, Sarah Villa asked, is this modification cell type specific?
• 63:22 - 63:26: And I guess, you know, does it exist also not only in neurons, but in other cells of
• 63:26 - 63:29: the body, in particular, maybe the immune system?
• 63:29 - 63:33: Yeah, the interesting thing about these modifications is contrary to my initial expectations, is
• 63:33 - 63:38: that they're not specific to the monoaminergic cells in which the monoamine is produced.
• 63:38 - 63:42: So we find serotonylation in non-serotonergic cells. We find dopaminergic and non-dopaminergic
• 63:42 - 63:48: cells and histamination as well. We have some interesting ideas about how that works mechanistically,
• 63:48 - 63:52: what types of transporters actually take them into other cells. But 100 percent, it is not
• 63:52 - 63:56: cell type specific. You will see it more broadly. And even in cell culture experiments, tissue
• 63:56 - 64:01: culture. Remember, when you're culturing in serum, serum contains all of these different
• 64:01 - 64:05: monoamines, and actually we see them even in HeLa cells and so forth. So, yeah, they're
• 64:05 - 64:08: more broadly distributed than what we might expect.
• 64:08 - 64:15: And we have another question from Vijay Kiwari. Hello, Vijay. So Q5 ser appears to be a modification
• 64:15 - 64:21: that acts to alter specific protein interactions with other histone marks. Do we know anything
• 64:21 - 64:28: about the mechanisms through which the Q5 ser mark is targeted to specific genomic loci
• 64:28 - 64:31: and reach for these other distinct histone marks?
• 64:31 - 64:36: So that's a phenomenal question and one that we're super interested in. So right
• 64:36 - 64:41: now I have a postdoc who's very interested in trying to identify the TGM2 complex in
• 64:41 - 64:45: the nucleus to try to really figure out how it may get targeted. But I can say at this
• 64:45 - 64:50: point, it's very unclear. The transglutaminase protein is seemingly quite promiscuous. And
• 64:50 - 64:54: so, you know, there's structural information about it, but not really in the context of
• 64:54 - 64:57: histones. And so I think there's a long way to go. But I do believe there is going to
• 64:57 - 65:02: be a targeting mechanism likely through a protein complex.
• 65:02 - 65:08: So we have one more question. Spencer Halls is asking. Very exciting. Do you know what
• 65:08 - 65:17: the relative stoichiometry of H3Q5 histamine is to H3K4 monomethyl and dimethylation at
• 65:17 - 65:20: different stages of the light/dark cycles?
• 65:20 - 65:24: That's a perfect question and exactly what we're trying to do right now. So I can tell
• 65:24 - 65:28: you that the histamine mark is in relatively, in the one snapshot in the cells we looked
• 65:28 - 65:35: at, it's a relatively similar stoichiometry to H3K4 dimethyl. However, we haven't looked
• 65:35 - 65:38: at it over the circadian time course. That's an experiment ongoing right now that Simone
• 65:38 - 65:43: Isabel and Einstein is helping us with. So again, stay tuned, but that's a great question.
• 65:43 - 65:53: So there's a quick one, I think. So Andrei Alendar, what are the readings of H3K5 histamine?
• 65:53 - 65:56: Another great question. So, you know, we've done a lot of legwork to identify how these
• 65:56 - 65:59: different monomethyl modifications affect the recruitment of things that are known to
• 65:59 - 66:05: bind to either unmodified tails or adjacent modifications. We have identified a couple
• 66:05 - 66:09: things; I'm not ready to talk about them publicly, that we do believe are specific readers of
• 66:09 - 66:15: serotonin versus dopamine versus histamine. So I guess my argument would be, I think there
• 66:15 - 66:19: are specific readers. I just can't talk about them quite yet.
• 66:19 - 66:26: Very good. It's a very big open space. I think we'll see a lot more about this in two years
• 66:26 - 66:33: time in our next meetings. Thank you, Ian, so much for joining us. And I'll ask the last
• 66:33 - 66:43: speaker of this first session, Marek Bartoszowicz, to join us. So, hello. Marek is currently
• 66:43 - 66:50: a postdoctoral fellow at the Lab of Gonçalo at the Karolinska Institute in Stockholm.
• 66:50 - 66:59: He completed his PhD in RNA biology at Masaryk University in the Czech Republic before moving
• 66:59 - 67:05: to the field of molecular neurosciences in the Lab of Gonçalo. And he recently developed
• 67:05 - 67:11: and applied a very exciting single-cell CUT&Tag method. And he has performed single-cell profiling
• 67:11 - 67:16: of histone modifications and transcription factor binding in the central nervous system.
• 67:16 - 67:19: So I look forward to your talk, Marek. Thank you so much.
• 67:19 - 67:24: Thank you so much for a wonderful introduction. You should be able to see my screen.
• 67:24 - 67:25: Yes.
• 67:25 - 67:26: Yeah.
• 67:26 - 67:27: Good.
• 67:27 - 67:33: So, yeah, this is a great opportunity to present our most recent efforts in developing this
• 67:33 - 67:42: method, single-cell CUT&Tag. And we use this method to profile the histone modifications
• 67:42 - 67:49: in the mouse brain. So to start, all right, to start with a very brief introduction, obviously,
• 67:49 - 67:56: everybody is very familiar with this, but I will be talking a lot about the histone modifications,
• 67:56 - 68:00: various kinds. So basically, if you zoom out from the histone modifications and look at
• 68:00 - 68:04: the individual roles, you have two basic classes. You have the active and inactive
• 68:04 - 68:12: histone modifications. Inactive, mostly typical examples like H3K9 trimethylation or H3K27
• 68:12 - 68:17: trimethylation. And these modifications are crucial to form these very compact heterochromatic
• 68:17 - 68:23: structures. And it's super important for repression of gene expression. And these
• 68:23 - 68:27: heterochromatic regions also tend to localize close to the nuclear periphery.
• 68:27 - 68:36: On the other hand, you have the active histone modifications, which are histone modifications
• 68:36 - 68:40: which specify this active chromatin. And this active chromatin is characteristic of
• 68:41 - 68:46: highly expressed genes. It has a lot of RNA Pol II binding transcription, also binding of
• 68:47 - 68:51: a lot of transcription factors. And this also specifies a lot of open chromatin
• 68:51 - 68:58: and higher expression. But of course, this is very clear. But then what I want to, my point that
• 68:58 - 69:04: I want to make with this slide is that you have all these various modalities in the inactive and
• 69:04 - 69:10: active. You have multiple tens of different histone modifications. Then you have some
• 69:10 - 69:15: additional features such as accessibility, chromatin architecture, like we heard about
• 69:15 - 69:20: today. And then, of course, many, many more. And all these epigenetic layers together
• 69:20 - 69:27: specify the gene expression programs during different processes, such as development or
• 69:28 - 69:32: disease as well. So this, the measuring these gene expression profiles at a single-cell level has
• 69:32 - 69:39: been, has had a great boom in the recent, recent five, ten years. So what is fantastic about this
• 69:39 - 69:45: approach is that when you measure gene expression of single cells, you can find relationships
• 69:45 - 69:50: between the individual cells very easily. So basically, this allows you to sample, for example,
• 69:50 - 69:54: in this case, a sample of the embryonic development at different time points.
• 69:55 - 69:58: You measure the single-cell transcription profiles, then you can find the
• 69:58 - 70:03: relationships, the cells that have very similar transcriptomes tend to, if you,
• 70:03 - 70:10: if you project them on this 2D embedding, the cells with very similar transcriptomes cluster
• 70:10 - 70:16: close to each other. And then you can kind of, you can reconstruct altogether this lineage
• 70:16 - 70:22: development tree. But what is, what is behind these gene expression profiles, like I said,
• 70:22 - 70:27: in the previous slide are all these epigenetic modalities. And if you, if you think about,
• 70:27 - 70:32: or if you talk about single-cell epigenomics, this nowadays is almost synonymous with single-cell
• 70:32 - 70:41: ATAC-seq. Basically, this is the most widespread method used to measure the single-cell level
• 70:41 - 70:46: epigenetic states. So in single-cell ATAC-seq, it takes advantage of this fantastic enzyme,
• 70:46 - 70:53: which is called Tn5 transposase, a hyperactive transposase. And if you add it to intact nuclei,
• 70:53 - 70:58: it's going to insert its cargo. It's carrying the DNA cargo, which is inserted into the open
• 70:58 - 71:03: chromatin regions. And this is an incredibly efficient reaction to a level where you can,
• 71:03 - 71:08: you can measure this insertion at the single-cell level. Then you have multiple single-cell
• 71:08 - 71:13: approaches to barcode the individual cells either by combinatorial indexing or other
• 71:13 - 71:19: strategies. Also, one of the most common is this 10x Genomics, a droplet barcoding strategy of
• 71:19 - 71:26: microfluidics, which are used to barcode individual cells. And then basically, you can convolute,
• 71:26 - 71:30: you get the bulk profile, and then you can convolute the bulk profile into these
• 71:30 - 71:34: profiles into individual cells. And then you can really see at the single-cell level,
• 71:34 - 71:42: the accessible chromatin in these single cells. So this is quite a routine thing that you can do
• 71:42 - 71:47: now, where you can buy it commercially. It's very widespread also. And then we,
• 71:48 - 71:54: there is another technology which kind of occurred recently, which is called CUT&Tag. CUT&Tag
• 71:54 - 71:59: is a way how you measure, how you profile histone modifications or transcription
• 71:59 - 72:04: factor binding. And this is great work coming from the lab of Steve Henikoff. Basically, CUT&Tag
• 72:04 - 72:10: takes advantage of the same Tn5 hyperactive transposase, but instead of
• 72:10 - 72:14: overloading the nucleus with the Tn5 and augmenting everything that is accessible,
• 72:14 - 72:20: you use specific antibodies for specific epitopes. It could be the transcription factors
• 72:20 - 72:25: or it could be modified histones. And then you use this small trick, which you have this
• 72:25 - 72:32: fusion protein A to the Tn5 transposase, which tethers the protein A, part of the fusion binds
• 72:32 - 72:38: to the antibody and this tethers the Tn5 only to the regions which have the epitope present.
• 72:38 - 72:43: And then you wash away all the excess Tn5 and basically have a similar workflow
• 72:43 - 72:49: to ATAC-seq. So, I mean, this technology existed at a bulk level and there were some first
• 72:49 - 72:54: reports of single-cell versions of this protocol, but we asked the question whether if we substitute
• 72:54 - 73:00: this ATAC-seq segmentation for the bulk CUT&Tag segmentation, and then couple it to the 10x
• 73:00 - 73:05: Genomics single-cell barcoding protocol, whether we can basically follow exactly the same
• 73:05 - 73:11: workflow and get single-cell profiles of, for example, some modifications. So what we did is we
• 73:11 - 73:17: took a mouse brain. So we are working in, the lab is mostly working in the development
• 73:17 - 73:22: of the oligodendrocyte lineage. We took the interesting developmental region for us, which
• 73:22 - 73:29: are P15 and P25, dissociated basically the whole brain, the suspension nuclei and ran the 10x
• 73:29 - 73:36: protocol with the ATAC-seq with some modifications of optimization that we made. And in fact, we can
• 73:36 - 73:41: pretty much do the same thing, like we can do with ATAC-seq just by substituting the segmentation step.
• 73:42 - 73:48: We're able to deconvolute the single-cell profiles from the data, and then to use some
• 73:48 - 73:55: dimensional reduction techniques to cluster the data, we're able to obtain the single-cell profiles
• 73:55 - 74:00: for all the major cell types within the mouse brain at this stage. So we can find,
• 74:00 - 74:07: we can find neurons, we can find all of the glia populations, oligodendroglia,
• 74:07 - 74:13: both OPCs, which are coming mostly from the P15 brain, mature oligodendrocytes from P25,
• 74:13 - 74:19: we can find a huge proportion of astrocytes, we can find immune cells, microglia, vascular cells,
• 74:19 - 74:23: and this specific population of olfactory ensheathing cells.
• 74:25 - 74:31: What is interesting with ATAC-seq, you can only profile active regions, so highly accessible
• 74:31 - 74:37: regions. We can also do the same for inactive regions. So we have done single-cell CUT&Tag
• 74:37 - 74:43: for H3K27 trimethylation as well. So you use an antibody against these repressive modifications,
• 74:43 - 74:48: and the Tn5 is active enough to tag even these heterochromatic regions, and basically obtain
• 74:48 - 74:53: very similar quality data to what you can do with active modifications. Again, you can, you can,
• 74:53 - 74:59: again, see all the major populations of the cell types in the mouse brain at this stage.
• 75:00 - 75:09: So to show you a little bit more detail, the data, here I'm focusing on one
• 75:09 - 75:16: important developmental gene with SOX10. As I said, we work mostly with the oligodendrocyte
• 75:16 - 75:20: lineage, and this is a critical transcription factor, important for the specification and
• 75:20 - 75:25: differentiation of the oligodendrocyte lineage. And then you can, when you look at the promoter here,
• 75:25 - 75:31: you can see that the promoter is H3K4 trimethylation, which is, this is a mark of
• 75:31 - 75:38: active promoters, is modified on the SOX10 in the mature oligodendrocytes. We have also signal in
• 75:38 - 75:43: the olfactory ensheathing cells, and then these are the oligodendrocyte progenitor cells. And then
• 75:43 - 75:48: all the other cell types, so just like astrocytes, VLMCs, neurons, they don't have this modification
• 75:48 - 75:55: on the SOX10 promoter. If you look on the other side with the repressive modifications,
• 75:55 - 76:01: we can do the single-cell deconvolution, and actually you see exactly the inverse profile.
• 76:01 - 76:06: So all the cell types that don't have the H3K4 trimethylation on the promoter, on SOX10,
• 76:06 - 76:12: have the H3K27 trimethylation, which mediates the repression of the gene. And then if the cell
• 76:12 - 76:17: clusters that have the modification, have the H3K4 trimethylation, don't have any H3K27 trimethylation.
• 76:17 - 76:24: So basically there is zero bivalency at this gene in this terminology differentiates the cell types.
• 76:24 - 76:30: So to also show you some of the markers, so this is very typical for the single-cell RNA-seq.
• 76:30 - 76:36: When people show the single-cell RNA-seq profiles, we can do the same.
• 76:36 - 76:43: So we look at the whole gene body and promoter, and then we count how many reads in individual
• 76:43 - 76:49: cells we find in these gene bodies and promoters. So we do that for H3K4 trimethylation, and you can
• 76:49 - 76:55: find, for example, this gene RFX4 is a super, very nice marker for astrocytes. If we look on
• 76:55 - 77:00: the repressive modification, H3K27 trimethylation, it's completely missing from
• 77:00 - 77:07: astrocytes, and this modification is there to repress the RFX4 expression in all the other
• 77:07 - 77:13: cell types, except for OPCs, which also still, even though they should not have the modification,
• 77:13 - 77:18: the H3K4, so much there, they also don't have the H3K27 trimethylation. You can look at some
• 77:18 - 77:24: other genes, markers for other cell populations, for example, MOG is a really well-established
• 77:24 - 77:32: marker for oligodendrocytes. OLIG2 is a super important transcription factor for glial cells
• 77:32 - 77:39: development and glial lineage specification, and you see that OLIG2 is quite strongly repressed
• 77:39 - 77:46: in all the other cell types, except for the glial, astrocytes, oligodendrocyte progenitors, and
• 77:46 - 77:54: mature oligodendrocytes. Then I'm showing also some markers of neurons, so RBFOX3, a panneural
• 77:54 - 78:00: marker, H3K4 trimethylated in all the neuron populations, and this is NIR2D2, which is not
• 78:00 - 78:07: repressed in these subpopulations of neurons and repressed everywhere else. What is really
• 78:07 - 78:14: unique about this method is that you can also construct these bulk profiles,
• 78:16 - 78:21: and under normal circumstances, if you do a bulk CUT&Tag or ChIP-seq for all these
• 78:21 - 78:26: different populations, you would have to basically find markers for the populations,
• 78:26 - 78:32: set up a FACS sorting strategy, and then do all these experiments, which would be like a
• 78:32 - 78:37: long-term project, whereas in the single-cell version of this protocol, you can basically just
• 78:37 - 78:42: take a bulk heterogeneous population of cells, and then you deconvolute the data
• 78:45 - 78:49: during the analysis, so in silico, and then you don't need to have any a priori knowledge about
• 78:50 - 78:55: the cell heterogeneity. You can find all the cell heterogeneity there in the data,
• 78:55 - 79:01: so this is a really fast way to generate these profiles for these really highly heterogeneous
• 79:01 - 79:09: samples, such as the brain, but it also finds application in other processes, which are highly
• 79:09 - 79:15: dynamic, and then you have perhaps very short-term modifications, such as in development
• 79:15 - 79:18: or in cancer, where you also sometimes don't know what the heterogeneity might be.
• 79:25 - 79:33: Like I said, we work in the oligodendrocyte lineage in the differentiation of oligodendrocyte
• 79:29 - 79:35: progenitor cells to the mature oligodendrocyte, and from the previous publications from our lab,
• 79:35 - 79:39: we know we have done a single-cell RNA-seq study, where we know that we can
• 79:40 - 79:44: find a gene expression profile where OPCs, oligodendrocyte progenitor cells,
• 79:44 - 79:49: differentiate through all these intermediate states, COP-NFOL, we call them committed
• 79:49 - 79:54: oligodendrocytes, and newly formed oligodendrocytes to this heterogeneous population of mature
• 79:54 - 79:59: oligodendrocytes. But with the CUT&Tag data, we only saw the clustering only.
• 80:00 - 80:04: We showed us two populations, two major populations, which is the OPCs and the mature OLs.
• 80:04 - 80:15: So we're wondering whether there is no continuity in the epigenetic profiles, and then the cells
• 80:15 - 80:20: either have a state where they're OPCs and the epigenetic state is now the mature oligodendrocytes,
• 80:20 - 80:25: or whether there is really a continuity of the histone marks.
• 80:25 - 80:28: In this case, it's H3K4 trimethylation.
• 80:28 - 80:34: We integrated the single-cell RNA-seq data with the H3K4 trimethylation data, and actually
• 80:34 - 80:41: we found that using this supervised way to cluster the cells, so you are basically telling
• 80:41 - 80:47: the algorithms to use the cells, the genes that are important, and then determine them
• 80:47 - 80:52: from the single-cell RNA-seq to use these genes in the clustering of the H3K4 trimethylation
• 80:52 - 80:53: CUT&Tag.
• 80:53 - 81:00: We could reconstruct the whole lineage, and when you look at the markers of these transitional
• 81:00 - 81:05: states and also at the markers of the terminal differentiation state, you can see that these
• 81:05 - 81:11: cells in here are enriched for genes characteristic of OPCs.
• 81:11 - 81:16: Then you can find these intermediate populations with expression of genes that are marked for
• 81:16 - 81:25: these intermediate states, and we can even deconvolute the heterogeneity in the mature
• 81:25 - 81:27: oligodendrocyte populations.
• 81:27 - 81:34: We have cells that mostly have H3K4 trimethylation of the OL1-specific genes here, OL2-specific
• 81:34 - 81:38: genes here, and OL3-specific genes here.
• 81:38 - 81:43: One more thing that you are able to do, you can look at very dynamic processes using this
• 81:43 - 81:44: method.
• 81:44 - 81:48: You have this phenomenon called H3K4 trimethylation spreading, so H3K4 trimethylation is a mark
• 81:48 - 81:51: of promoters again.
• 81:51 - 81:52: You have these two...
• 81:52 - 81:56: Basically, there's a paper saying that you have these two states; you have these narrow
• 81:56 - 82:02: H3K4 trimethylation peaks, and then you have these broad H3K4 trimethylated domains, and
• 82:02 - 82:06: this is associated with expression level, but also, more importantly, with transcriptional
• 82:06 - 82:07: consistency.
• 82:07 - 82:12: You have very often these very broad marks on important developmental regulators and
• 82:12 - 82:17: transcriptional factors, and they need to be consistently expressed throughout the development.
• 82:17 - 82:23: We tried to look at this in our OPC differentiation again, so we ordered, again, the OPCs as they
• 82:23 - 82:29: differentiate our OL states, and then we sorted the cells in this heat map, and you
• 82:29 - 82:33: can nicely see that as the cells progressed, or as the more differentiated states, you
• 82:33 - 82:39: see the H3K4 trimethylation mark both increasing in intensity but also increasing in the breadth
• 82:39 - 82:40: of the H3K4 trimethylation.
• 82:40 - 82:43: This was never possible before with bulk approaches.
• 82:43 - 82:46: You need a single cell to look at it.
• 82:46 - 82:50: We are sort of, I think, running out of time.
• 82:50 - 82:51: Try to...
• 82:51 - 82:52: I don't know how many more slides...
• 82:52 - 82:53: It's the last slide.
• 82:53 - 82:54: Okay, great.
• 82:54 - 82:55: Sorry.
• 82:55 - 83:01: I just wanted to advertise also this web resource that we also deployed, where you can also
• 83:01 - 83:02: look at your favorite genes.
• 83:02 - 83:05: We have all these different cell types and different modifications.
• 83:05 - 83:08: We have four histone marks and two transcription factors.
• 83:08 - 83:11: I didn't mention we could also do this on transcription factors.
• 83:11 - 83:17: And then you can visualize the enrichments on all these on the UMAPs and then also on
• 83:17 - 83:18: the genome browser.
• 83:18 - 83:22: And we also recently posted the preprint into the BioRxiv server, so if you're more interested
• 83:22 - 83:25: in the technology, check the paper out.
• 83:25 - 83:30: And yeah, so with that, thanks to Gonçalo's group and Gonçalo for giving me the opportunity
• 83:30 - 83:33: to work in his lab, being a fantastic PI.
• 83:33 - 83:39: And then I would like to only highlight Mukund, who is a Pearson lab, who collaborated with
• 83:39 - 83:44: me on this project, and then Bastian, who constructed the Shiny web resource.
• 83:44 - 83:47: And thanks for the attention.
• 83:47 - 83:48: Thank you so much, Marek.
• 83:48 - 83:51: This was beautiful work and a beautiful talk.
• 83:51 - 83:52: Sorry I rushed you.
• 83:52 - 83:55: As you say, we are a little bit late.
• 83:55 - 84:00: So we have a couple of questions that I'll try to fit in if everyone allows me.
• 84:00 - 84:01: So one is from Sara Vila.
• 84:02 - 84:06: How many nuclei would you need to carry this type of experiment or minimum of nuclei per
• 84:06 - 84:07: cluster?
• 84:07 - 84:10: So try to give a short answer and then I'll give you another question.
• 84:10 - 84:15: In principle, the more nuclei you have, the better you can identify the clusters.
• 84:15 - 84:20: But I would say you need roughly a few hundred nuclei of the same cell type to identify the
• 84:20 - 84:21: cluster.
• 84:21 - 84:23:  And another question from Jay Divakar.
• 84:23 - 84:24:  Hi, Marek.
• 84:24 - 84:25:  Thank you.
• 84:25 - 84:30:  Is it possible to combine your workflow with the new 10x platform that allows ATT&CK and
• 84:30 - 84:35:  RNA from the same cell in order to obtain single cell ATT&CK plus RNA together?
• 84:35 - 84:37:  Yeah, this is super exciting.
• 84:37 - 84:40:  So we already have the kit and we are trying it out.
• 84:40 - 84:41:  I don't know yet.
• 84:41 - 84:44:  We don't have the results yet, but very soon we are going to do that.
• 84:44 - 84:50:  And for me, does it work with RNApol2 antibodies or its chromatin K9?
• 84:50 - 84:53:  So is it, do you have trouble with some other?
• 84:53 - 84:55:  Yeah, we haven't tried RNApol2.
• 84:56 - 85:02:  In the papers from Steve Henikoff's lab, they have done RNApol2.
• 85:02 - 85:03:  So it works in bulk.
• 85:03 - 85:06:  So I don't see a reason why it wouldn't work in the single cell.
• 85:06 - 85:09:  And with K9, we haven't tried.
• 85:09 - 85:11:  We didn't have a good antibody.
• 85:11 - 85:13:  We have tried, but it was not a great antibody.
• 85:13 - 85:15:  So we have to repeat it again.
• 85:15 - 85:16:  So we also don't know.
• 85:16 - 85:25:  I mean, it's a pleasure to introduce you to Dr. Zabadul, who is a professor in pediatrics
• 85:25 - 85:31:  at McGill University and a pediatric neuro-oncologist at the Montreal Children's Hospital.
• 85:31 - 85:38:  So Anaza began her career at McGill in 2003 with her own group, pioneering a research
• 85:38 - 85:42:  program on pediatric brain tumors.
• 85:42 - 85:49:  And her group has uncovered pediatric high-grade astrocytomas as molecularly diverse and genetically
• 85:49 - 85:54:  distinct from adult tumors.
• 85:54 - 86:00:  More importantly, Anaza's group identified a new molecular mechanism driving pediatric
• 86:00 - 86:10:  HGA, specifically somatic-driven mutations in the tail of histone H3 variants, H3.3 and
• 86:10 - 86:11:  3.1.
• 86:11 - 86:19:  And these were actually the first reported histone mutations in human disease.
• 86:19 - 86:23:  So I'm sure that Nada is going to talk some more about this.
• 86:23 - 86:29:  Nada was also named in 2015 Fellow of the Royal Society of Canada in Life Science Division.
• 86:29 - 86:35:  So we are very happy that you could join and very eager to listen to your talk.
• 86:35 - 86:38:  Thank you for inviting me, and I'm very eager to present.
• 86:38 - 86:42:  This is very new data, actually, and I thought it would be of interest to share it with this
• 86:42 - 86:43:  audience.
• 86:43 - 86:46:  And actually, she had to ask the last question is really an integral and a major part of
• 86:46 - 86:48:  this data.
• 86:48 - 86:51:  So the title of my talk is A Case of Mistaken Identity.
• 86:51 - 86:55:  And actually, it showcased how using the tools that we have that we've described throughout
• 86:55 - 87:02:  the different talks that were here can potentially help us orient what a mutation and how a disease
• 87:02 - 87:04:  is coming and where it's coming from.
• 87:04 - 87:08:  So I have no disclosure, except that I thought that I was living in the Ice to Stone Age
• 87:08 - 87:13:  and with all of those next-generation technologies that are happening, the single cell, super
• 87:13 - 87:18:  exciting the single cell data on epigenetic markers, beauty.
• 87:18 - 87:20:  And I really hope to dig even further into that.
• 87:20 - 87:23:  So that's not going to be in this talk, but hopefully in another.
• 87:23 - 87:29:  And I'll start by showing how I think those brain tumors that I study are so closely linked
• 87:29 - 87:33:  to brain development that it's really unreal.
• 87:33 - 87:38:  The histone mutation that we described, the lysine 27 to methionine substitution in any
• 87:38 - 87:44:  of the histone 3 variants and the glycine 34 to arginine or valine exclusively, and
• 87:44 - 87:50:  this is still true up to now in histone 3.3, are really tightly associated with age and
• 87:50 - 87:52:  the brain location.
• 87:52 - 87:58:  If you have a midline tumor, and what I mean by brain midline is the spine, the pons, the
• 87:58 - 88:03:  thalamus, anywhere within this midline structure is going to be a lysine 27 mutation in any
• 88:03 - 88:05:  of the histone 3 variants.
• 88:05 - 88:06:  And it's going to be associated with age.
• 88:06 - 88:10:  The younger in the pons, the older going to the thalamus and the spine.
• 88:10 - 88:15:  So really very tightly based on the age, I could potentially identify what's the mutation
• 88:15 - 88:19:  and what's going to be the best combination of mutation that we're going to be finding
• 88:19 - 88:22:  in the tumor just based on age and tumor location.
• 88:22 - 88:28:  We go to the cortex starting at the age of 12, and this is where in kids, they start
• 88:28 - 88:32:  to join the adult world and they have cortical tumor, tumors in the cortex.
• 88:32 - 88:38:  And those that are in the temporoparietal cortex are exquisitely between the age of
• 88:38 - 88:45:  12 and 35, and will be a glycine 34 to arginine and valine mutation on histone 3.3.
• 88:45 - 88:48:  And they're always associated with TP53 and ATR-X.
• 88:48 - 88:53:  ATR-X is a chromatin modeler, and I don't have time today to kind of delve into why
• 88:53 - 89:00:  this association, but this trio is almost all the time present over there and exclusively
• 89:00 - 89:05:  found between 12 and 35 years old and exclusively in this brain area.
• 89:05 - 89:09:  The IDH mutation that people have known for a long time about are more in the frontal
• 89:09 - 89:14:  lobe and sometimes can go a little bit to the temporoparietal, but it's really a very
• 89:14 - 89:17:  unique distribution within the brain.
• 89:17 - 89:23:  And we identified truncating mutation in PDGFRA, and PDGFRA is a lysine 36 trimethyltransferase,
• 89:23 - 89:26:  the only lysine 36 trimethyltransferase.
• 89:26 - 89:32:  And we've shown that actually the lysine 36 on the histone 3 tail seems to be quite important
• 89:32 - 89:36:  for those brain tumors within the cerebral cortex.
• 89:36 - 89:42:  So what we think is happening that the brain is just split into a lysine 27 for the midline
• 89:42 - 89:48:  and a lysine 36 potentially on histone 3.3 in your cortex, and that those pediatric and
• 89:48 - 89:52:  young adult astrocytomas are developmental defects.
• 89:52 - 89:56:  And more to that, our studies on other brain tumors that are quite unique to kids and younger
• 89:56 - 90:02:  children have shown a pattern really where we found a fusion that led to the expression
• 90:02 - 90:08:  of a specific isoform of DNMT3B that's a fetal isoform and exclusively expressed in the brain
• 90:08 - 90:10:  between 8 and 12 weeks.
• 90:10 - 90:17:  It really shows that within the brain, if you hit at the given time point with the right
• 90:17 - 90:21:  oncogene, you may lead to tumor formation.
• 90:21 - 90:23:  And this is mainly through probably stalled development.
• 90:24 - 90:26:  Okay, so why am I speaking about that?
• 90:26 - 90:32:  And I'm going to be speaking about the glycine 34 mutation, it's just that this is understanding where
• 90:32 - 90:37:  they arise, where those tumors arise in, gives you a little bit of a hint of what happened
• 90:37 - 90:41:  over there, and maybe how you can potentially treat them, because this is our A.
• 90:41 - 90:45:  So almost like Sherlock Holmes, we went through a detective story.
• 90:45 - 90:51:  A lot is known about the lysine 27M, not a lot is known about glycine 34 to an arginine or valine
• 90:51 - 90:52:  substitution.
• 90:52 - 90:53:  So how does this work?
• 90:53 - 90:57:  I'll start by the most important part is a collaboration between my lab and Claudia Klein's
• 90:57 - 90:58:  lab.
• 90:58 - 90:59:  Claudia is here.
• 90:59 - 91:04:  She's an amazing person, a bioinformatician that loves biology, and her group has become
• 91:04 - 91:06:  mine and this is really an ongoing love story.
• 91:06 - 91:09:  The story started with Carol Shen's finding.
• 91:09 - 91:13:  She's a postdoctoral fellow in my lab of PDGFRA mutation.
• 91:13 - 91:18:  And more than that, Celine Jessa, who does a lot of the single cell and the deconvolution
• 91:18 - 91:20:  is an integral part of that.
• 91:20 - 91:26:  Shreya, who is a graduate student, an MD-PhD student in my lab, worked a lot on the CRISPR
• 91:26 - 91:31:  and the identification and a lot of the different techniques and deconvolutions.
• 91:31 - 91:37:  Jihad worked on the Hi-C data that we had generated, and he had asked that question.
• 91:37 - 91:41:  And Leonid Klyzy, who was co-supervised by Claudia and me, worked on the single cell.
• 91:41 - 91:47:  Those are the people that should be talking, but I'm acknowledging them at first hand.
• 91:47 - 91:52:  When you look at the glycine-34 mutation and the tumor, the high-grade glioma, there is a conundrum.
• 91:52 - 91:54:  Those people actually have a dual phenotype.
• 91:54 - 92:00:  They're either neuronal, and for glioma, it's like, oh my God, they were actually misdiagnosed
• 92:00 - 92:05:  as primitive neuroectodermal tumors, which are neuronal tumors, or they have a mixed
• 92:05 - 92:10:  glial astrocytic component, and the glial component could be a little bit more rich.
• 92:10 - 92:14:  But having this, if you look at the DNA methylation, which really kind of tells you if it's the
• 92:14 - 92:20:  same biological entity, yes or no, whatever their phenotype under the microscope is, they
• 92:20 - 92:28:  cluster with the same DNA methylation cluster, which shows that they are one unique entity,
• 92:28 - 92:31:  and they share the TP53 and ATRX mutations.
• 92:31 - 92:35:  But it kind of raised a lot of questions, why this dual composition?
• 92:35 - 92:42:  So what we did is we kind of went and canvassed and acquired a lot of samples from adults,
• 92:43 - 92:49:  when above the age of 18, go to adult centers, so we acquired samples from adult patients
• 92:49 - 92:51:  and from pediatric patients.
• 92:51 - 92:57:  We did what we could on bulk transcriptomics, single-cell transcriptomics, a lot of epigenetic
• 92:57 - 93:02:  profiling on the primary tumor and on the very rare cell lines that you could derive
• 93:02 - 93:08:  from those patients, and we did CRISPR-Cas editing and intrauterine modeling in mice
• 93:09 - 93:13:  to see if we could understand a little bit better those gliomas.
• 93:13 - 93:19:  The first hit came when Carol kind of came with those PDGFRA mutations.
• 93:19 - 93:23:  PDGFRA is a major gene in brain development.
• 93:23 - 93:28:  It's really very important for glial development, and for oligodendrocyte precursor cells, Gonzalo would know even better
• 93:28 - 93:30:  about that than me.
• 93:30 - 93:38:  And these PDGFRA is amplified mainly in the K27M setting, and there is in the IDH
• 93:38 - 93:44:  mutation, there is a TAD that actually favors the expression, the overexpression of PDGFRA.
• 93:44 - 93:49:  The mutations are present but are very sparse, up to 5 to 7 percent, but nothing more.
• 93:49 - 93:56:  And here, in those glycine 34 to arginine or valine, just on 3.3 tumors, they were up
• 93:56 - 94:01:  to 50 percent, 50 percent at diagnosis, and they could go up to 80 percent at recurrence,
• 94:01 - 94:03:  so there is an enrichment at recurrence.
• 94:03 - 94:08:  And here, it's shown that the K27M has a few, but three were nothing to compare with
• 94:08 - 94:15:  the glycine 34 mutations, and the other IDH or wild type have none of that.
• 94:15 - 94:19:  And when we looked at the gene expression, not only the mutation is there, but there
• 94:19 - 94:28:  is a higher expression of PDGFRA in those G34R mutant gliomas, and this is even more
• 94:28 - 94:31:  when you carry the mutation.
• 94:31 - 94:34:  And the other striking thing, and I won't go about it, this is something that we're
• 94:34 - 94:39:  working on more, is that those mutations are identified in the extracellular domain.
• 94:39 - 94:45:  They're predicted to be activating to render this to be ligand-independent or to continue
• 94:45 - 94:50:  signaling throughout the endocytic pathway, and so those mutations really are quite interesting
• 94:50 - 94:55:  in that they're mainly seen, if not exclusively seen, in brain tumors.
• 94:55 - 95:01:  So they are brain-specific PDGFRA mutations, and the kinase mutant domain is really barely
• 95:01 - 95:03:  seen in 2 percent.
• 95:03 - 95:08:  So even if it's seen, and this one is the one that's in GIST and other cancers where
• 95:08 - 95:10:  it's been better characterized.
• 95:10 - 95:16:  So all of this, okay, PDGFRA, what's happening in those glycine 34 mutations?
• 95:16 - 95:22:  So we went back to our RNA-seq data on bulk tumors on the cell lines that we have, and
• 95:22 - 95:27:  what we've been doing, and we've published about that, is trying to create a blueprint
• 95:27 - 95:30:  of gene expression for brain development, so we're not the only one.
• 95:30 - 95:36:  This is just a very small part of a larger initiative, and what we had done is that we
• 95:36 - 95:42:  had compiled, because those tumors occur either in the cerebellum or the pons or the cortex,
• 95:42 - 95:51:  to try and gather very early stages of single-cell data and bulk transcriptomic data from mouse
• 95:51 - 95:52:  and human brain.
• 95:52 - 96:00:  In the mouse, we went as early as E6.5, and we went postnatally, and to characterize what
• 96:00 - 96:05:  are the lineage that could, or what is the gene signature for specific lineages within
• 96:05 - 96:11:  this, the transcription factor that would potentially identify one lineage versus the
• 96:11 - 96:16:  other, and we used a lot of the other data sets that are being, every day there's a new
• 96:16 - 96:21:  data set that we try and incorporate to enrich our signature for those specific neuronal
• 96:21 - 96:29:  populations, and we projected single-cell data that we acquired on those glycine 34 mutant
• 96:29 - 96:36:  high-grade glioma or bulk RNA-seq to try and see if they could match better a signature
• 96:36 - 96:41:  of a given lineage, and this is where we had the first surprise.
• 96:41 - 96:46:  When we looked compared to the other hemispheric tumors, whether, or the midline tumors, the
• 96:47 - 96:55:  high-grade glioma K27 and mutant, the glycine 34 mutant were highly enriched for inhibitory
• 96:55 - 97:03:  progenitor neurons, prenatal, not postnatal, and those, if you look at the inhibitory progenitor
• 97:03 - 97:08:  neurons, they're also enriched in the subventricular zones, and what we did is we acquired a recent
• 97:08 - 97:14:  data set that had used a subventricular zone, and we saw that they also matched to this
• 97:14 - 97:19:  inhibitory interneuron progenitor signature that's either in the ganglionic eminences, and
• 97:19 - 97:25:  I'll come to that, in the ventral brain, or postnatally in the subventricular zone where
• 97:25 - 97:29:  there is a pool of those progenitors that remains there.
• 97:29 - 97:33:  In the normal developing brain from, this is very simplified, but I need to simplify
• 97:33 - 97:38:  it in order to understand, from a ventral radial glial cell that's progenitor cells,
• 97:38 - 97:44:  you could either become an oligodendroglial progenitor or an interneuron progenitor,
• 97:44 - 97:51:  and this fate, you go one way or the other, you cannot go back, and this is actually achieved
• 97:51 - 97:57:  by specific transcription factors that dictate where you would go, and that would prevent
• 97:57 - 97:59:  you from going back to another one.
• 97:59 - 98:04:  In the dorsal cortex where you have your excitatory neuron, those actually arise from the lateral
• 98:04 - 98:05:  ventricle.
• 98:05 - 98:13:  The interneurons arise from the ventral forebrain, from those ganglionic eminences, and at E13.5,
• 98:13 - 98:20:  this is where you have the major transcription factors that actually mediate these interneuron
• 98:20 - 98:27:  lineage, and those are GSX2 and DLX1 and 2, which actually will push your radial glial
• 98:27 - 98:33:  cell into becoming an interneuron, and while they do that, they actively inhibit OLIG2,
• 98:34 - 98:41:  which would specify an OPC lineage, and actually, OLIG2 would do the reverse if that's the case,
• 98:41 - 98:46:  and the postnatal brain, very early in the subventricular zone, you still have the same
• 98:46 - 98:51:  transcription factors that dictate this interneuron origin.
• 98:51 - 98:57:  So what we did is we looked at the expression of those transcription factors in our data
• 98:57 - 99:02:  set of high-grade glioma, and we saw that exquisitely in the glycine 34 mutant, you had
• 99:02 - 99:08:  increased expression of GSX2, DLX1, and DLX2, and this work was done by Celine, who really
• 99:08 - 99:16:  did those projections that I showed earlier, and this type of analysis, and Carol and Shreya
• 99:16 - 99:23:  in my lab looked at the epigenome, and I apologize for this little complex 3D figure, where the
• 99:23 - 99:30:  size is actually of the dots, is the level of expression, and what you see here on this
• 99:30 - 99:36:  axis is a level of glycine 27 trimethylation, and here the level of glycine 27 acetylation,
• 99:36 - 99:43:  and the genes that were pasted with glycine 27 and expressed were actually those DLX1, GSX2,
• 99:43 - 99:48:  and DLX2, and OLIG2 we know is DNA methylated, because it's actually probably the cell of
• 99:48 - 99:55:  origin that's not there, it's not an oligodendroglial lineage, so the OLIG2 gets methylated,
• 99:55 - 100:00:  and other markers of excitatory neurons and astrocytes who are more
• 100:00 - 100:06:  suppressed by K27me3 deposition. And when we looked further into markers that would give you
• 100:06 - 100:14:  that were genes that were up in glycine in G34R mutant tumors, we saw that those were interneuron
• 100:14 - 100:22:  progenitor markers, and what was really down was markers that would differentiate the interneuron.
• 100:22 - 100:29:  So what we think is happening is similar to what we see in K27M, the glycine 34 mutation is blocking
• 100:30 - 100:35:  further differentiation of a progenitor that acquires it. So what we think is happening is
• 100:35 - 100:39:  that this mutation is occurring in either this early progenitor, the radial glial that already
• 100:39 - 100:45:  committed to interneuron, or in the interneuron progenitor, and the mutation by changing the
• 100:45 - 100:51:  epigenome, and we have some data to that effect, is preventing from further maturation into a more
• 100:51 - 100:58:  mature interneuron. So how do we consolidate PDGFRA and those DLX1 and 2 mutations,
• 100:59 - 101:05:  and this is where we went back and looked at our census, and what we saw is that while PDGFRA
• 101:05 - 101:10:  alpha, as expected, is highly expressed in early glial progenitors, it's nowhere
• 101:10 - 101:16:  expressed in interneurons, whereas GSX2, DLX1, and DLX2, as expected, are present here.
• 101:17 - 101:21:  And the other thing is that if you look at the kinetics of PDGFRA, it's present throughout,
• 101:21 - 101:27:  and it's inversely correlated with GSX2. So how those two are present at the same time,
• 101:27 - 101:34:  in the same cell, in those gliomas. This is where we had the best surprise ever. When we looked,
• 101:34 - 101:41:  we saw that GSX2 and PDGFRA are located on the same chromosome within a closed chromosomal
• 101:41 - 101:46:  region, and we had Hi-C data that we had gathered at some ages in one of the fishing
• 101:46 - 101:53:  experiments that we did, and GHAD helped us immensely analyze it better. And what we see,
• 101:53 - 101:57:  what I'm showing here, this is a wild-type high-grade glioma, this is another wild-type
• 101:57 - 102:03:  high-grade glioma, and this is a K27M mutant glioma. And those two are cell lines from
• 102:05 - 102:12:  glycine 34 arginine mutant glioma with a PDGFRA mutation, and one that's wild-type for PDGFRA.
• 102:12 - 102:18:  In both those gliomas, in everywhere, you could see the first stat that's being described
• 102:18 - 102:24:  by Bradley Bernstein's group, that associate, that anchors the PDGFRA promoter to another
• 102:24 - 102:30:  CTCF site and leads drive potentially in some glioma PDGFRA expression. So this is present
• 102:30 - 102:38:  throughout, okay. But exquisitely in the G34, we saw increased enhanced interaction between
• 102:38 - 102:45:  the promoter of PDGFRA and an enhancer upstream of GSX2 that's known to drive the
• 102:45 - 102:52:  expression of GSX2. So this is a virtual foresee that shows the anchor and how this is, and this
• 102:52 - 102:59:  is statistically significant, this is anchored. And what we see is that the lysine 27 acetylation
• 102:59 - 103:05:  is basing throughout here, this area, whereas in the IDH mutant or in the wild-type, you don't have
• 103:05 - 103:10:  a lot of acetylation at that site. And what's driving the expression of GSX2 is potentially
• 103:10 - 103:19:  bleeding and driving the expression of PDGFRA that are very close by interacting within
• 103:19 - 103:24:  the type. So we went back and said, okay, is this interaction something that's specific or is it
• 103:24 - 103:30:  something that was present before? So we went back and we had data from mouse embryonic stem cells
• 103:30 - 103:38:  and required data from a cortex at E13.5 and the ganglionic eminence at E13.5. E13.5 is the time
• 103:38 - 103:44:  where GSX2 expression is there in peaks. And what we could see exquisitely is that you had
• 103:44 - 103:51:  this interaction quite increased in embryonic stem cells, but mainly between PDGFRA and GSX2,
• 103:51 - 103:58:  and that both PDGFRA promoter and GSX2 promoter in the mouse embryonic stem cells are
• 103:58 - 104:06:  actually bivalently marked, silenced by K27me3 and pasted with a lysine 4 me3 and not expressed.
• 104:06 - 104:13:  In the cortex, you don't find those interactions. They are not present anymore. But in the ganglionic
• 104:13 - 104:20:  eminences, not only are they present, but you start seeing how the GSX2 gets actually acetylated
• 104:20 - 104:28:  while the promoter of PDGFRA retains the trimethylation. So what we're seeing is actually
• 104:28 - 104:36:  a bivalent demethylation of this mark and acetylation of the PDGFRA in the context
• 104:36 - 104:42:  of the G34 tumor that takes advantage of pre-existing interaction in these ganglionic
• 104:42 - 104:48:  eminences, potentially to drive the expression. And what's quite key is that I'm showing mouse
• 104:48 - 104:53:  data and human data. This region is conserved in evolution, which tells you it's important.
• 104:54 - 104:58:  So what we think is happening that usually in the embryonic stem cells,
• 104:58 - 105:03:  you have this confirmation, this looping that's happening, and it's bivalently silenced.
• 105:03 - 105:06:  In the cortex, it doesn't need the GSX2 because you don't have interneurons
• 105:07 - 105:13:  that are generated in the embryonic cortex. This loop doesn't exist, but the ganglionic
• 105:13 - 105:20:  eminences keep it and activate using this enhancer, the GSX2, which drives the expression
• 105:20 - 105:25:  in a normal interneuron progenitor of the transcription factors that will further
• 105:25 - 105:31:  specify those interneurons, while the PDGFRA locus retains its silencing and should not be
• 105:31 - 105:39:  acting in those interneurons. And there's a kind of an addiction. If you go and look further at
• 105:39 - 105:47:  the mutant versus the wild type G34 mutant, you see that in the mutant tumor, further interaction,
• 105:48 - 105:53:  further downstream from this enhancer that we know is activating the GSX2 and the PDGFRA
• 105:53 - 105:58:  are getting acquired almost as if there is a kind of a feed, positive feedback,
• 105:58 - 106:04:  and you're addicted to this. Using single cell data that we acquired on 20 tumors, including
• 106:05 - 106:10:  primary and recurrence pairs, what we could see, this is showing the full heat map of the different
• 106:10 - 106:16:  tumors. What we could see exquisitely, anything that's blue is neuronal. Anything that's red
• 106:16 - 106:22:  is more astrocytic. In the wild type, you had way more neuronal subgroups that were present in
• 106:22 - 106:28:  your single cell data. And in the mutant, it seems that they were more astrocytic. Even though you
• 106:28 - 106:34:  still had some neuronal features, there was a huge increase of the astrocytic signature within
• 106:34 - 106:41:  those single cell data. And in the primary tumor that didn't have a PDGFRA mutation that
• 106:41 - 106:46:  acquired it, you have a kind of a shift and you have even more of an astrocytic signature
• 106:46 - 106:51:  that's arising as you acquire the PDGFRA mutation. And when we looked at the different
• 106:51 - 106:59:  lineages, really there is zero oligodendrocytes in the G34RNV tumors. And this is something that's
• 106:59 - 107:04:  expected if they are interneuronal, they should not at all, and expressing GSX2, they should not
• 107:04 - 107:11:  have this oligodendrocyte precursor cell signature. Whereas in every single other high-grade glioma,
• 107:11 - 107:15:  you have oligodendrocytes that are present. And this is something that was really, we couldn't explain it before,
• 107:15 - 107:20:  but now we know it's just because the cell lineage of origin is potentially an interneuron progenitor or
• 107:20 - 107:26:  something that's committed to going there. And the transcription factor program is actively
• 107:26 - 107:32:  repressing any oligodendrocyte differentiation in those progenitors. And when we look at those
• 107:32 - 107:37:  primaries, this is just to showcase that when you get the recurrence you have, your astrocytic
• 107:37 - 107:44:  signature gets more increased. And when we looked at the mutation, it's really a swiping
• 107:44 - 107:49:  addiction because it becomes the dominant clone the moment you acquire this mutation. This is the
• 107:49 - 107:54:  only tumor that didn't acquire it, but it acquired a mismatch repair that gave it a lot of other
• 107:54 - 108:00:  mutations as seen in other high-grade gliomas. And the other thing that we saw is that even though
• 108:00 - 108:06:  they are called astrocyte or interneuron, they're neither astrocyte nor interneuron because they're
• 108:06 - 108:11:  expressing something that's completely abhorrent. Markers of astrocytes and interneurons that are
• 108:11 - 108:18:  never seen to exist together. So we CRISPR-ed out this mutation and this was done by Shreya
• 108:18 - 108:24:  in my lab and we analyzed the dataset that we acquired from there. CRISPRing out the mutation
• 108:24 - 108:30:  didn't do anything. Those tumors when they have acquired the other TP53, PDGFRA,
• 108:31 - 108:36:  ATRX mutations, they don't seem to care anymore if the G34 mutation is here or not in contrast
• 108:36 - 108:42:  actually to K27M where we saw a drastic increase in survival when we removed it.
• 108:42 - 108:48:  And when we modeled this tumor, whereas the G34 are very poor oncogene, you get it,
• 108:48 - 108:54:  the PDGFRA mutant by themselves in association with ATRX and TP53 drive very aggressive tumors.
• 108:54 - 108:59:  So what we think is really that what's happening, you have the mutation occurring in an
• 108:59 - 109:05:  interneuron progenitor that confers, really it shapes the epigenome slowly but surely,
• 109:05 - 109:12:  leads to PDGFRA being expressed. PDGFRA drives the fire, meaning more proliferation and
• 109:12 - 109:18:  this astrocytic signature that have them labeled as gliomas. And this is really the oncogenic
• 109:18 - 109:25:  component because G34R is only here to set the epigenome. And once those progenitors go and
• 109:25 - 109:31:  migrate, they're so stuck that it's very hard to go back after that. And those mutations are the
• 109:31 - 109:36:  ones that we could potentially target today, either downstream, the MAP kinase signaling pathway or
• 109:36 - 109:41:  upstream where those mutations are located and hopefully help those children while we find
• 109:41 - 109:47:  better ways to kind of unlock this G34 mutation. So this work that I spoke to you about, I
• 109:47 - 109:51:  acknowledge the people that did the work, but there are many collaborators that provided samples,
• 109:51 - 109:57:  expertise, Paolo Salomone's group at Bournemouth University with Mana Petania that has his own lab
• 109:57 - 110:06:  now at UCL in the UK, really worked a lot on the intrauterine explorations and helped us with
• 110:06 - 110:11:  those experiments. The different partners across the world and of course our funding sources.
• 110:11 - 110:18:  And I'm happy to take any questions. Okay, thank you very much, Nada, for a really exciting work.
• 110:19 - 110:27:  We have a question from Mark Lee Jr. Congratulations, which sub-ventricular zone
• 110:27 - 110:32:  Wall macrodomain was the focus of your transcriptomic evaluation?
• 110:33 - 110:38:  So we didn't do ourselves the transcriptomic evaluation. This was a dataset that was collated
• 110:38 - 110:44:  from recent work in the literature. So they used it at P7, if I remember correctly. And we're doing
• 110:44 - 110:50:  it again because when we stain for the GSX2, not everywhere stains, but we're using something
• 110:50 - 110:57:  that's more, we obtain a GSX2 transgenic mice from Ken Campbell, who is really the leader in
• 110:57 - 111:04:  the GSX2 field. And we're trying to separate, to just really look specifically at these progenitor
• 111:04 - 111:08:  cells, either immediately after birth or a little bit later, just to see if it correlates better,
• 111:08 - 111:13:  because we would like to time it either to some ventricular zone or the ganglionic eminences.
• 111:13 - 111:20:  And just based on how messy it is to get single cells from there, I think this transgenic mice
• 111:20 - 111:26:  might be of help if it works. A question that I have is, so it's quite interesting that,
• 111:28 - 111:32:  so you have increasing proliferation of OPCs, but then you don't have differentiation
• 111:32 - 111:37:  totally with interneurons, but you get astrocytes instead, right? We don't have proliferation of
• 111:37 - 111:44:  OPCs. They're not OPCs. They're interneurons. They are, and PDGFR-alpha is giving them a little
• 111:44 - 111:50:  bit like the credential of an astrocyte or an OPC, because PDGFR-alpha drives this astrocytic
• 111:50 - 111:55:  signature. And it's invariantly expressed because usually interneurons should never express it,
• 111:55 - 112:02:  but it's so happened that the PDGFR-alpha and its promoter are really downstream of the main
• 112:02 - 112:08:  transcription factor, GSX2, that drives the OPC. What we think the G34 mutation is doing
• 112:08 - 112:16:  is making the promoter of PDGFR-alpha accessible. And this way you bleed in there.
• 112:17 - 112:23:  But then there is also the, I mean, the pre-OPC or the OPC would have the choice to become an
• 112:23 - 112:28:  oligodendrocyte or an astrocyte. Usually in that domain, it would go to an oligodendrocyte. So
• 112:28 - 112:34:  there must be something else downstream that is really driving the astrocytic fate, right?
• 112:34 - 112:36:  Have you, do you have any clues of what could it be?
• 112:36 - 112:43:  There is a lot of MAP kinase signaling that we see, and it's, we don't have, again,
• 112:43 - 112:50:  if you just overexpress PDGFR-alpha or those mutant PDGFR-alpha, you drive astrogenesis to no end.
• 112:50 - 112:56:  I'm not saying oligodendrogenesis. I'm saying astrogenesis. I don't think they're OPCs because
• 112:56 - 113:02:  they don't have Olig-1 and Olig-2. They have really zero oligodendroglial markers. Tox10
• 113:02 - 113:09:  that you see is not there. So it's, they're really, this is what people are, it's misleading.
• 113:09 - 113:15:  People call them OPC progenitor. It's very true for K27 and probably this is where it arrives.
• 113:16 - 113:24:  But the Glycine 34, no way. They have nothing in terms of OPC. They have an astrocytic,
• 113:24 - 113:30:  I'm calling it for lack of a better word, signature, glial signature that they,
• 113:30 - 113:33:  it's just PDGFR-alpha expression probably that's giving them that.
• 113:34 - 113:39:  And is there any other kind of tumors that might have shared this kind of origin?
• 113:39 - 113:45:  So we tried to see, and there is none that express that to that level of GSX2 or DLX,
• 113:45 - 113:50:  because we're using that as a surrogate to find for now, but you never know. I, again,
• 113:50 - 113:56:  it's all of the high-grade glioma and all of the lower-grade glioma, they don't have this.
• 113:56 - 114:02:  And we have the dataset. There's nothing there. Okay. Okay. Thank you very much, Nada for the
• 114:02 - 114:10:  presentation. So it's my pleasure to present Dr. Phil DeYager, who is a well-granted professor
• 114:10 - 114:16:  of neurology at Columbia University Medical Center. And he's also chief of the division
• 114:16 - 114:21:  of neuroimmunology and directs the Center for Translational and Computational Neuroimmunology.
• 114:24 - 114:31:  So Phil's research program combines methods of human immunology, genomics, and computational
• 114:31 - 114:41:  biology to investigate several neurodegenerative diseases, such as Alzheimer's disease,
• 114:41 - 114:47:  which is going to be the topic of today's talk, but also Parkinson's and also multiple sclerosis.
• 114:48 - 114:55:  So Phil graduated from Yale University and received his MD and PhD from Rockefeller University and
• 114:55 - 115:03:  Cornell University Medical College. And he completed his neurology residency at Massachusetts
• 115:03 - 115:10:  General Hospital and Brigham and Women's Hospital. So really great that you could join us, Phil,
• 115:10 - 115:16:  and we're looking forward to your presentation. Great. Well, thank you for the kind introduction.
• 115:17 - 115:23:  So today I'll be talking to you about a new story in the lab that's just under review and
• 115:24 - 115:32:  that where we basically took an epigenomic approach to try to identify factors that may
• 115:32 - 115:37:  attenuate the effect of known risk factors as a way to try to identify mechanisms that we could
• 115:37 - 115:44:  then target for drug development. And particularly here, we decided to focus on one very well, of
• 115:44 - 115:50:  course, validated risk factor for Alzheimer's disease, which is APOE4, a haplotype that is
• 115:50 - 115:54:  common in the general population, about 20% frequency, depending on the exact population,
• 115:54 - 116:02:  higher in other populations, and again, for which a lot is known. So
• 116:03 - 116:09:  sorry, these are my disclosures. So APOE4 is a very strong risk factor, but it's not deterministic.
• 116:09 - 116:15:  And I think this is nicely shown in this paper, actually, from the late 90s,
• 116:17 - 116:26:  where the investigators plotted the disease-free survival for different subjects. And you can see
• 116:26 - 116:32:  that, you know, the subjects that have no E4 are the squares at the top. And so, you know, over time,
• 116:33 - 116:39:  some of them do decline and develop Alzheimer's disease, but, you know, by the age of 93 or so,
• 116:39 - 116:46:  it plateaus and there are no more cases. And APOE4, which are the black circles, those are the
• 116:46- 116:49:  homozygotes, and the E4 heterozygotes are the triangles. And you can see that both classes of
• 116:53 – 117:00:  subjects plateau. So meaning that even as they get older, a substantial proportion of the
• 117:00 - 117:07:  individuals who have an E4 haplotype do not develop Alzheimer's disease. So again, so while
• 117:07 - 117:11:  this is a very strong risk factor, it's not definitive, and many people with APOE4 do not
• 117:11 - 117:20:  develop dementia. Another way to show this in this paper was to see, to plot the effect size of the
• 117:21 - 117:29:  different haplotypes over time. And so you can see that this is the different ages from 40 to 90 years
• 117:29 - 117:35:  old on the x-axis, and the odds ratio is plotted on the y-axis. And you can see that the largest
• 117:35 - 117:42:  effect size is seen in people in their 60s. And afterwards, the risk actually begins to decline.
• 117:42 - 117:50:  So basically, if you survive to your later 80s, the effect size of APOE4 is much diminished. And
• 117:50 - 117:55:  again, suggesting that there is a window of risk, and that there are other factors that are
• 117:55 - 118:02:  interacting with APOE4 to cause dementia. Another sort of aspect of this is we're looking for
• 118:02 - 118:08:  interaction with other factors. So in this case, actually, it's the reverse. So then this group
• 118:08 - 118:13:  looked at the effect of certain genetic variants, which are listed here in rows. So there's CR1,
• 118:13 - 118:20:  BIN1 cluster, and the MS4 region. These are four well-validated susceptibility loci for
• 118:20 - 118:27:  Alzheimer's disease. And you can see, if you look at the effect size and the p-values for individuals
• 118:27 - 118:32:  who have E4 or who don't have E4, there's actually, for these loci, there's a substantial difference.
• 118:32 - 118:37:  And these are just four of what are now about 30 different loci involved in AD.
• 118:38 - 118:43:  Again, so here there's interaction, meaning that some of the effects are different in the two
• 118:43 - 118:49:  groups. And I think probably one of the more interesting ones may be the first one, CR1,
• 118:50 - 118:56:  where the effect size and the p-value is much stronger in E4 subjects than in E4 negative
• 118:56 - 119:02:  subjects, while the E4, of course, are only about a quarter or less of the total population. So this
• 119:02 - 119:06:  is a very small number. So statistically, you have less power in this group to detect an effect
• 119:07 - 119:14:  than in the E4 negatives. And yet, the association is much stronger. Again, suggesting that there
• 119:14 - 119:19:  is some level of interaction between the CR1 variant, perhaps, and the APOE4.
• 119:19 - 119:24:  So evidence of this type of interaction exists, but there are a few cases so far.
• 119:25 - 119:31:  So a postdoctoral fellow in my lab, Dr. Yi Mao, who is now an instructor and developing her own
• 119:31 - 119:37:  research portfolio, had an interesting idea, which is basically to use our epigenomic data
• 119:37 - 119:45:  to try to see whether we can identify factors that would attenuate the effect of E4. And again,
• 119:45 - 119:50:  the idea there being if we can identify those factors and understand their mechanism, perhaps
• 119:50 - 119:59:  we can mimic the effect using medications over time. So this is what she undertook.
• 120:00 - 120:08:  To do this, we used two cohorts of aging as our discovery study, and these are superb
• 120:08 - 120:14:  cohorts led by my colleague, Dr. David Bennett, who designed and runs these studies, the Religious
• 120:14 - 120:19:  Order Study and the Memory and Aging Project. In both cases, these are older individuals
• 120:19 - 120:27:  who are living in group communities. They're non-demented at the onset of the study, and
• 120:27 - 120:32:  over time, they're observed and some develop dementia and others don't and pass away with
• 120:32 - 120:38:  no cognitive impairment. After death, Dr. Julie Schneider is the neuropathologist working
• 120:38 - 120:43:  with these two cohorts, and she does a very structured neuropathological examination to
• 120:43 - 120:48:  derive both a pathological diagnosis of Alzheimer's, but also a number of different quantitative
• 120:48 - 120:55:  measures of pathology. And altogether, now there are over 3,700 participants, over 1,600
• 120:55 - 121:01:  now are deceased. And so we have a large number of brains available, and we have generated
• 121:01 - 121:06:  a number of different data types. We'll be talking about three of those data types today.
• 121:06 - 121:12:  On this slide, I'm simply showing you some of the different data types that we have generated.
• 121:12 - 121:18:  We focus on one large region of the cortex, the dorsolateral prefrontal cortex, to be
• 121:18 - 121:23:  specific, to generate many of our data. And the data that I'll be talking about today
• 121:23 - 121:29:  all comes from the same individuals sampled in the same region. So we have genotype data,
• 121:29 - 121:36:  we have demethylation data measured by the Illumina 450 human methylation array, and
• 121:36 - 121:41:  also we have RNA-seq profiles, again, from the same individuals from the same brain region.
• 121:41 - 121:47:  So this forms the basis of data that we're going to be analyzing.
• 121:47 - 121:54:  The study design is shown here. And so on the left side is the detailed one that we
• 121:54 - 121:59:  have as a figure in our paper, but it's sort of a stripped-down version on the right side
• 121:59 - 122:05:  for the purpose of presentation. So again, stage one is our DNA methylation-wide association
• 122:05 - 122:11:  study, trying to identify factors that are associated with the traits of interest in
• 122:11 - 122:18:  the ROS and MAP studies. Now, we usually consider going straight to Alzheimer's disease,
• 122:18 - 122:25:  but the problem is that this is a categorical variable, and so it has much less power than
• 122:25 - 122:29:  some of the quantitative traits that we have available in these subjects. And so because
• 122:29 - 122:34:  of that and because of the fact that, again, the APOE4 subset of individuals is small and
• 122:35 - 122:42:  has limited power, we focused the stage one, the discovery stage, on two separate analyses
• 122:42 - 122:50:  identifying CpG dinucleotides where the methylation level was associated with either the burden
• 122:50 - 122:56:  of tau pathology present in the brain or with the burden of neurotic amyloid plaque in those
• 122:56 - 123:06:  brains. And so out of 420,000 sites attested, we identified 25 that met our predetermined
• 123:07 - 123:14:  threshold of significance, and these were then examined both in ROS-MAP in relationship to AD
• 123:14 - 123:20:  pathology, and of course tau and amyloid are the two components with which we make a diagnosis
• 123:20 - 123:25:  of Alzheimer's disease pathologically. And so we took the ROS-MAP subjects and then two replication
• 123:26 - 123:31:  from the London Brain Bank and the Mount Sinai Brain Bank, and then we did a meta-analysis,
• 123:31 - 123:38:  and of these 25 CpGs that have been prioritized in stage one, they passed at stage two, and four
• 123:38 - 123:43:  CpGs were validated. We then went on to stage three where we accessed our RNA-seq data and
• 123:43 - 123:50:  some immunohistochemistry data to expand the story and functionally annotate our results.
• 123:50 - 123:56:  So this I'll now take you through the different results. This is simply the methylation-wide
• 123:56 - 124:03:  association study for neurotic plaque and tau burden. Again, the tau results are on the superior
• 124:03 - 124:09:  aspect of the figure, the inferior aspect in green shows the results for neurotic plaque,
• 124:10 - 124:15:  and you can see the x-axis has the physical position along the genome. Each dot here is
• 124:15 - 124:21:  one CpG, and the threshold of significance is the red line. And you can see the 25 different
• 124:21 - 124:28:  CpGs that were associated significantly with one or the other trait in our analysis. So you can
• 124:28 - 124:32:  see they're distributed throughout the genome, and of course there aren't so many, so it's
• 124:32 - 124:39:  relatively sparse. Now after the meta-analysis where we again bring in two additional datasets
• 124:40 - 124:47:  to validate our results, we find that four of these CpGs are validated in this meta-analysis
• 124:47 - 124:54:  at a corrected level of significance, and these are shown here. One point I'll try to emphasize
• 124:54 - 125:00:  right now is that, again, these four CpGs are shown here in red. Again, physical position is
• 125:00 - 125:06:  on the x-axis, the y-axis is the level of significance of the association, and the four
• 125:06 - 125:11:  CpGs that are significant are in red. You can see that they're located on different chromosomes.
• 125:11 - 125:18:  So again, we now have identified four different loci in different parts of the region of the genome
• 125:18 - 125:27:  that are associated with a diagnosis of Alzheimer's disease, in this case APOE4
• 125:27 - 125:34:  subjects. So this is important, so this is basically within the subjects that have APOE4,
• 125:34 - 125:39:  these particular sites are influencing the likelihood of having a diagnosis of Alzheimer's
• 125:39 - 125:45:  disease. Just to be a little bit more granular, here are the forest plots, so you can actually
• 125:45 - 125:49:  see that the effects of the CpGs are quite consistent across the different datasets.
• 125:50 - 125:55:  If you focus on the superior aspect of the figure here, where you have all subjects initially, you
• 125:55 - 126:00:  can see one row for ROS-MAP, one for the London Brain Bank, and one for the Mount Sinai Brain Bank.
• 126:00 - 126:05:  These are much smaller datasets than the ROS-MAP, where we had a much larger number
• 126:05 - 126:10:  of individuals, but you can see that the effect sizes are quite consistent, and in blue we have
• 126:10 - 126:16:  the meta-analysis for just the replication dataset, and then in red we have the meta-analysis
• 126:16 - 126:22:  with all the different replication data, with all the discovery and the replication datasets
• 126:22 - 126:28:  together. The effect of these four loci, so each column here is one locus, which is listed at the
• 126:28 - 126:36:  top, is shown for E4 here, and for E4 negative individuals. Of course, as expected because of
• 126:36 - 126:41:  our design, we see a strong effect with the E4 population, but what's interesting is that the
• 126:41 - 126:47:  E4 negative subjects, which are much more numerous, about more than double the number of the E4
• 126:47 - 126:56:  positive ones, the effect size is much smaller, in some cases still significant, but less pronounced
• 126:56 - 127:02:  than in the E4. So clearly there's something going on with E4, and at the bottom we show the
• 127:02 - 127:08:  interaction term between the CpG of interest and E4. Again, there's a significant interaction for
• 127:08 - 127:15:  these four CpGs, for their level of methylation in relation to Alzheimer's disease,
• 127:16 - 127:18:  in relative to the risk for Alzheimer's disease.
• 127:19 - 127:25:  So the other thing that was interesting, and this is shown here, is that three of the loci
• 127:25 - 127:30:  are strongly correlated to one another. Again, they're located on different chromosomes, so it's
• 127:30 - 127:34:  not simply the fact that they are in proximity to one another. These are, you know, on different
• 127:34 - 127:39:  chromosomes, and yet their level of methylation is strongly correlated. I would also note that
• 127:39 - 127:44:  RIN3 is located within an Alzheimer's disease susceptibility locus itself, so it's also,
• 127:44 - 127:52:  you know, quite interesting. And you can see that, again, three of these, the RIN3, MPL, and Tom20
• 127:52 - 127:58:  loci, those CpGs are strongly correlated to one another, and they're anti-correlated with the
• 127:58 - 128:03:  level of methylation of a different CpG of a different locus on a different chromosome,
• 128:03 - 128:12:  which is the LPP locus, which is at the top left. And so, again, we have three of the CpGs that are
• 128:12 - 128:18:  basically have very similar behavior to one another, and the fourth one is anti-correlated
• 128:18 - 128:26:  to those three. So this can also be visualized here, where we can look at these effects
• 128:28 - 128:33:  with the different principal components, if we just focus on the four CpGs. And as expected,
• 128:33 - 128:38:  three of the CpGs are very similar to one another. You can see here, the vectors are listed for each
• 128:38 - 128:44:  of the CpGs in relationship to the different cohorts. So, again, red for ROS-MAP, the London
• 128:44 - 128:50:  Brain Bank in green, and the Mount Sinai Brain Bank in blue. And you can see, again, that the
• 128:50 - 128:57:  results are quite consistent. Now, because these three CpGs are correlated, what we did is to
• 128:57 - 129:03:  actually collapse them into one principal component, principal component one, and so this is,
• 129:03 - 129:09:  therefore, a summary measure that we used in further analysis, that we're basically assuming
• 129:09 - 129:14:  that the different CpGs are all capturing the same effect slightly differently, but the underlying
• 129:14 - 129:23:  process may be the same. So what does this effect of PC1 actually look like? And so,
• 129:23 - 129:29:  trying to visualize the effect on Alzheimer risk, we end up with some of the visualizations shown
• 129:29 - 129:36:  here. At the bottom left, we have the proportion of individuals who have Alzheimer's disease at
• 129:36 - 129:41:  the time of death. You can see that for those individuals who have a low PC1 expression,
• 129:42 - 129:46:  you know, there's a certain proportion of individuals in blue who are E4 negative,
• 129:46 - 129:52:  and as expected, you know, more people who are E4 positive have Alzheimer's at the time of death
• 129:53 - 129:59:  in people who have low PC1. So the relationship between…the effect of E4 is clearly present in
• 129:59 - 130:04:  these individuals, but it's less than in the high PC1 individuals. You can see here that
• 130:04 - 130:12:  for both E4 negative and E4 positive, the risk of AD goes up if you have a high level of methylation
• 130:12 - 130:22:  or the high PC1 score. And it's also…the effect on E4 risk actually is not linear. It's actually
• 130:22 - 130:28:  much larger, so there's an interaction term. And we can also visualize this on the right side,
• 130:28 - 130:34:  where the blue dots represent the E4 negative subjects, E4 positives in orange, and you can
• 130:34 - 130:43:  see that as the level of the PC1 methylation goes up, the risk of AD, it goes up. Now, for
• 130:43 - 130:49:  the E4 negative individuals in blue, that risk goes up modestly as the PC1 increases,
• 130:49 - 130:54:  but the effect is much more striking in the subjects that are E4 positive, so their risk
• 130:55 - 131:03:  expands rather rapidly. So we now have this PC1, which is, again, sort of an epigenomic factor
• 131:04 - 131:10:  that is affecting multiple loci throughout the genome, but what exactly is it? Of course,
• 131:11 - 131:15:  the genes themselves were not terribly informative, so to try to understand
• 131:15 - 131:22:  some of the biology, we accessed our RNA-seq data, and we had 421 individuals which had both
• 131:22 - 131:31:  PC1 measurement and also transcriptome data of over 17,000 genes from the same brain region and
• 131:31 - 131:36:  the same individuals. And so the results are shown here with the beta for the association
• 131:37 - 131:42:  between PC1 and the expression of a gene on the x-axis and the significance on the y-axis.
• 131:42 - 131:48:  So this volcano plot shows very nicely that there's a substantial number of genes, 71,
• 131:48 - 131:53:  that are significantly correlated with PC1 in terms of their level of expression.
• 131:54 - 132:00:  So we push this a little further. We actually went into two other datasets. The Mount Sinai
• 132:00 - 132:04:  Brain Bank also had RNA-seq data on a smaller number of individuals, but it had both methylation
• 132:04 - 132:10:  and RNA-seq data. We also accessed a second replication dataset, which I won't show today,
• 132:10 - 132:15:  but from the Mayo Clinic, which had also methylation data and RNA-seq data, and the
• 132:15 - 132:20:  results are very consistent. Basically, the vast majority of the 71 genes that we find to be
• 132:20 - 132:27:  associated with PC1 in the ROS-MAP subjects are also seen to be associated in the same direction
• 132:28 - 132:36:  in the other datasets. Now, what are all these genes? So we have 71 genes. That's not a bad
• 132:36 - 132:42:  number to try to do some enrichment analyses. You can see some of the results here. If you take a
• 132:42 - 132:49:  quick look at the list of pathways that are enriched with this particular set of genes,
• 132:49 - 132:55:  you can clearly get a flavor that there's an enrichment for innate immune genes, particularly
• 132:55 - 132:59:  myeloid cell-related genes. This is seen with the phagosome, for example,
• 133:00 - 133:05:  TB, and some of the other diseases which have a large component of myeloid cells being involved.
• 133:07 - 133:12:  So the next thing we did was to, okay, still within the RNA-seq data, we had previously
• 133:13 - 133:19:  reduced the dimensionality of those data, so we went from over 17,000 genes to 47 modules of
• 133:19 - 133:24:  co-expressed genes. So this is, again, just using co-expression to empirically define groups of
• 133:24 - 133:31:  genes that have a similar pattern of expression across all of our individuals that have an RNA-seq
• 133:31 - 133:37:  profile. And out of these 47 modules, so groups of genes, seven of them were enriched for
• 133:37 - 133:42:  microglia. And so, of course, microglia are the resident immune cell of the brain, and they are
• 133:42 - 133:47:  myeloid cells, so they're very similar to monocytes, macrophages, and the other myeloid cells.
• 133:48 - 133:52:  So this was a way to try to get at the same question but in a slightly different fashion.
• 133:52 - 133:59:  And what you see here is, on the x-axis, we have the number of genes from our list that are
• 133:59 - 134:06:  part of the module of interest, and then the significance and the enrichment on the y-axis.
• 134:06 - 134:12:  And you can see that module 116, for example, at the top right, has close to 60 genes from our
• 134:12 - 134:20:  71 genes that were associated with PC1 that are actually found within the module. And then we
• 134:20 - 134:24:  have some other modules which have fewer numbers but still significantly enriched for this.
• 134:24 - 134:29:  So again, what we're seeing is that there's some enrichment for a microglial transcriptional
• 134:29 - 134:37:  program. Finally, we also accessed a different type of data. We have measured, using immunohistochemistry,
• 134:37 - 134:43:  the counts of microglia inside the tissue in the same individuals in the same brain region.
• 134:44 - 134:47:  And so here we have, and actually in multiple different brain regions,
• 134:47 - 134:52:  and so what we see here is that the PC1 is correlated to the proportion of activated
• 134:52 - 134:57:  microglia defined morphologically in the same tissue. And we had previously shown that this
• 134:58 - 135:02:  measure, the proportion of activated microglia, is also associated with Alzheimer's disease.
• 135:04 - 135:09:  So just to summarize the results, so we've identified four CpGs that are associated with
• 135:09 - 135:15:  AD in APOE4 subjects. So the effect of the three of these CpGs can be summarized as this PC1,
• 135:15 - 135:22:  and as interaction between PC1 and APOE4, where each decrease of one unit of PC1 attenuates the
• 135:22 - 135:27:  odds of Alzheimer's by over 50%. So that's a large number, but just gives you an idea that
• 135:27 - 135:32:  the effect of this PC1 is actually quite substantial, and if we can begin to understand
• 135:32 - 135:38:  what is involved, how is it related to the microglial function, we may be able to mimic
• 135:38 - 135:46:  its effect with chemicals. So I think that's all I have for you today, and I'd like to acknowledge
• 135:46 - 135:54:  particularly Dr. Yi Mao, who designed the study and is now developing the story further
• 135:55 - 135:59:  as she develops her own lab, and of course my colleagues at Rush, David Bennett and Julie
• 135:59 - 136:10:  Schneider. Thank you. Thank you very much, Phil, for very interesting data. So while I'm waiting
• 136:10 - 136:15:  for more questions, I'm just wondering, you find a strong correlation between the CpGs, right?
• 136:17 - 136:23:  Do you, is there any characteristic that might link them that like,
• 136:25 - 136:30:  sequence-wise, or like other transcription factors, any properties that can,
• 136:32 - 136:38:  might lead to a common regulator of these CpGs? Yes, it's a good question. So we've looked at
• 136:38 - 136:44:  this. Of course, we only have three CpGs so far. I suspect there are probably many more,
• 136:44 - 136:48:  but we have to go back and identify maybe a larger collection that are enriched,
• 136:48 - 136:52:  and I think at that point we may be able to do a more powerful job at trying to understand,
• 136:52 - 136:58:  like you say, whether there's a shared regulon or shared sequence feature. So far, again,
• 136:58 - 137:06:  with only three CpGs, we don't see any clear pattern that emerges from them. At one level,
• 137:06 - 137:13:  I think we're basically picking up something from the microglia, but it's not simply the total
• 137:13 - 137:19:  count of microglia in the tissue. So it's really more specific than that. And again, one nuance is
• 137:19 - 137:25:  that it, while there's a correlation with the proportion of activated microglia, PC1 is not
• 137:25 - 137:30:  a proxy for that. So when you account for the proportion of activated microglia, PC1 still
• 137:30 - 137:34:  has an effect. So it's capturing something slightly different, but it's related to activated
• 137:34 - 137:35:  microglia.
• 137:39 - 137:47:  Great. Okay, so I think we will thank you very much, Phil, for the presentation.
• 137:47 - 137:54:  And I think now we'll move, and thank you all the speakers, we'll now move for the closing remarks
• 137:55 - 137:55:  from Ana.
• 137:57 - 138:02:  Hello, thank you, Phil. I actually have a very quick question, whether you could comment on
• 138:02 - 138:09:  the effect of age. So now that you discover all these associations, can you go back and sort of
• 138:09 - 138:17:  comment on the bell shape you found for age? Yeah, that's a great question. It's a little
• 138:17 - 138:23:  difficult from the ROS-MAP data, because these individuals on average are 88 years old at the
• 138:23 - 138:27:  time of death, so they're quite old in general. They're not particularly representative of the
• 138:27 - 138:32:  entire population. So I think we have to expand, this is actually part of what Yi is planning to
• 138:32 - 138:37:  do, to expand this into other cohorts. We're currently working with Dr. Richard Miu and his
• 138:37 - 138:43:  cohorts called YCAP and IFEGA, where the samples are actually younger. And I think then we're
• 138:43 - 138:51:  going to get a better idea of how PC1 behaves over time and age. In general, we do know that
• 138:51 - 138:56:  aging has a big effect on microglia. So that's, in some ways, it's not surprising. It's very
• 138:56 - 139:03:  intriguing that, you know, we may be capturing that effect with the PC1. But again, I think we
• 139:03 - 139:08:  unfortunately have more data. The other aspect to this is also we're trying to find a proxy for PC1
• 139:08 - 139:14:  in the periphery. And so Yi has some preliminary data from peripheral blood data that correlates to
• 139:14 - 139:20:  some degree with PC1. So that would enable us also to look at a much wider number of individuals.
• 139:20 - 139:26:  Thank you so much. This was really exciting. So it's a great pleasure to thank, of course,
• 139:26 - 139:31:  all the speakers for making the time to join us and for their really exciting talks,
• 139:31 - 139:38:  sharing new data. I'd also like to thank all the attendees for registering, joining, and
• 139:38 - 139:44:  participating with their questions. And finally, the Upcom team, especially Lucy Perser, whom
• 139:44 - 139:50:  Gonçalo and I worked through to organize the event, and also Polly, Kaylee, and Kumaran for
• 139:51 - 139:57:  helping and for hosting it. And finally, just to remind you that we are, Gonçalo,
• 139:57 - 139:59:  Lucy, and I are already organizing the
• 140:00 - 140:07:  next one for June 2022. This is on the 8th to the 10th, and it's going to be located here in Berlin
• 140:07 - 140:13:  in the institute where I work, right in the center of Berlin. And so we really look forward to seeing
• 140:13 - 140:18:  you all here. And so with this, thank you so much, and I'll hand back to Kumaran, and I think he's
• 140:18 - 140:25:  just going to say bye on behalf of Upcom. Thank you. Yeah, thanks, Ana. Yeah, I definitely would
• 140:25 - 140:31:  like to echo your thoughts, and thank you all for your truly fascinating talks. Yeah, as you
• 140:31 - 140:36:  mentioned, all that is left now is for me to kind of close out today's meeting, and on behalf of
• 140:36 - 140:41:  Upcom, I'd like to thank you all for your participation. For any questions that we were
• 140:41 - 140:49:  unable to get to during the live Q&A, we will reach out to you after this event. Yeah, so finally,
• 140:49 - 140:55:  I'd like to thank our presenters, Marak, Philip, Nada, Ian, Jennifer, and Dominic, and of course
• 140:55 - 141:02:  our moderators, Gonçalo and Ana, for doing such a wonderful job chairing and helping to put together
• 141:02 - 141:08:  the session. We at Upcom really do hope you've enjoyed our live sessions within the spotlight
• 141:08 - 141:13:  to a neuroscience event, and please, the fun continues with some of our live on-demand
• 141:13 - 141:19:  webinars, so please do check them out. We have David Rubenstein presenting on autophagy and
• 141:19 - 141:24:  neurodegeneration, and Dr. Malou Tansy presenting on neuroinflammation and neurodegenerative
• 141:24 - 141:35:  disorders. Yeah, with that, take care, everybody, and have a great day.