2022 High Throughput Imaging Characterization of Brain Cell Types & Connectivity-Day 1, Part 1

Transcript

NED TALLEY: It's my great pleasure to introduce the High Throughput Imaging and Characterization of Brain Cell Types and Connectivity Workshop. The purpose of the workshop is to discuss unmet needs and emerging imaging approaches to characterize brain cell types and their connectivity in humans and other mammals. The workshop goals are to discuss how benchmark existing technologies identify critical gaps and barriers of the imaging pipelines and to discuss potential solutions through innovation and collaboration. We will also be assessing ways to streamline and standardize methods. And we'll be discussing expectations for data sharing and reuse by the research and education communities.

Next slide, please, Laura. I see you're still working out how to adjust the black boxes, but I think we can keep going in the meantime. So general guidelines, if you're not a panelist, please keep yourself muted and with video off. If you have questions, feel free to place them in the chat, and we might have extra time during the plenary talks for answering questions. But more likely, what we would prefer is to have the speakers actually answer the questions in the chat, just in terms of time constraints. And if you have any technical issues, contact Laura Reyes or Sandeep Kishore. They are both on the line and will respond to emails. And let's see. So before we move on, I just want to acknowledge that Yong Yao and Laura Reyes have really essentially done this workshop by themselves, and that's quite an achievement. And I think that we all owe them a debt of gratitude for the work that they've put in and the efficiencies that they've gotten for it.

And I also want to thank Sandeep, as I mentioned, and Michele Pearson who is also on the line helping out in terms of managing the various Zoom settings. Okay. Next, I need to introduce John Ngai. Is there a separate slide for that, Laura? Okay. So John Ngai, he was named director of the BRAIN Initiative about three years ago. And it's now grown to over $600 million in annual funding, so he's in control of more money than a lot of NIH institute directors. But don't tell anyone. He has spearheaded the launch of BRAIN 2.0 transformative projects for mapping and manipulating the brain cell types and their connectivity, including the BRAIN Initiative Cell Atlas Network, which has a goal of generating a complete reference atlas of cell types in the human brain across the lifespan. He's also overseeing the rollout of the BRAIN Initiative's new approach to diversity and inclusion in which investigators' plans for enhancing diversity are scored elements in their research proposals. And this is a big deal. It's a paradigm that other parts of the organization are picking up and other federal agencies are mimicking. So it represents a real shift in NIH and federal policy with respect to the expectations for workforce diversity. I think that is probably a good enough introduction for someone who's my boss. I hope that I get credit for that later in the year.

JOHN NGAI: Always, Ned. Thanks, Ned. And thank you all for being here. Special shout out to Yong and Laura for putting together what I'm sure will be a really exciting workshop that we will all benefit from. When Yong asked me to give a few introductory remarks-- reflecting back on the last, I don't know, 8 or 10 years, we've really seen remarkable advances in our ability to characterize cell types throughout the body, and particularly the brain, the most complex organ in the body. And this, no doubt, has been fueled by a revolution in several areas, most notably in single-cell analysis. Now, it's not all about the genes and molecules that are being expressed within cells, but it is a great place to start. And this way, we can get a handle for getting an accumulation on integration of other modalities that really tell us not only just what a cell is but what it's doing and its role it plays in the nervous system that we're interested in, underlying complex behaviors.

Now, these advances, these revolutions were not only on the technical weapon side but it was the integration of very sophisticated analytical and theoretical techniques that allowed us to not just making sense of the data but to design better ways of understanding what cells look like and what they do. So I think what we're going to hear about today is this next stage in the revolution, which is really trying to understand what cells are and what they do in both space as well as in time. And you can think of this in whatever dimension you want. It's all going to be applicable. We're really leaving it up to the field to determine where the best opportunities are and where the greatest ideas are, and to dig in and really try to develop better tools and resources for understanding neurocircuit function based on a better understanding, not just of cell types but again, where they are in the brain in space and time and how they connect with each other.

So I will leave it to you. I encourage you to listen well. Unfortunately, we're not in person where perhaps some side conversations could happen in the hallway, at the bar, over dinner, over lunch, or whatever. But I do encourage you folks to talk to each other based on what you hear. Again, we are really looking for a diversity of expertise and approaches to solve really, really tough problems. Judging by the acceleration of knowledge that we have seen over the last 8 years, I'm really quite hopeful that over the next 5 to 10 years, we're going to understand how the brain works in a way that we couldn't imagine today. So with that, I'll leave it to everybody here. We have some great talks coming up, some really great breakout sessions, flash talks, whatever, and I look forward to seeing what everybody comes up with at the end of these two days.

NED TALLEY: Thanks, John. Great introduction. I think next up is Yong Yao to talk about the goals and the details of the workshop. Yong, take it away.

YONG YAO: Thank you, Ned. Good morning, good afternoon, and good evening. Welcome to NIH BRAIN Initiative Cell Atlas Network BICAN workshop on high throughput imaging and characterization of brain cell types and connectivity. I would like to put this workshop into perspective of the overarching goal of the BICAN that is to build reference brain cell atlases in human and other mammalian species, which will be widely used throughout the research community, providing a molecular and anatomical foundational framework for the study of brain function and disorders. The BICAN goal is aligned with the first two research priorities recommended by advisory committee to the NIH director in their BRAIN 2025 report, first discovering brain cell diversity, and second, mapping neural connectivity, and more recently, the recommendation by the BRAIN 2.0 report to establish the human brain cell atlas as a transformative project.

What is brain reference cell atlas? The NIH BICAN program envisioned that the reference brain cell atlas should be an open and a broadly accessible digital product of a BRAIN Initiative data ecosystem that can link data across different levels and scales on distinct features and phenotypes of brain cells, including genome sequencing, DNA maceration and folding, chromatin modification, and the states on the isoforms and quantification, protein expression, and small molecules as well as cell location, size, morphology, tissue center composition, cell-cell communication, connectivity function, pathology, and the medical conditions. The analogy to geographic information system that can capture, store, and display seemingly unrelated data according to spatial coordinates such as latitude and longitude, physical addresses, and zip codes. The reference brain cell address will be developed by establishing anatomical and spatial common coordinate frameworks that this workshop will discuss. In addition, the reference brain cell atlas will also use genome coordinates to connect DNA, RNA, and the protein expression data, and establish brain cell ontology as well as common data elements.

Next. So the BRAIN Initiative has so far awarded 11 BICAN grants with a total of about $100 million per year for the next five years, supporting a variety of assay methods as listed on the right side of the slide. The founded BICAN projects will provide comprehensive single-cell transcriptome and the epigenome profiling of brain cell types in human, mouse, marmoset, macaque across lifespan in a set of spatial transcriptomic maps in whole brains of mouse, marmoset, and the macaque, and the select brain regions in human. At this time, it remains challenging to generate high comprehensive and high-resolution human brain cell atlas. That's why this workshop will address this unmet needs and gaps.

New BICAN RFAs now published to support UM1 centers that will adopt high throughput imaging technology platforms and R01 specialized collaboratories that will complement UM1 centers as well as cloud-based common imaging data processing pipelines with an upcoming application due date of February 1st, next year. So the purpose of this workshop is to tackle critical gaps and barriers in the imaging technology pipelines for generating comprehensive and high-resolution brain cell atlases in human and the other species and to streamline individual steps from human brain specimen, acquisition of processing, multiplexed labeling, high throughput imaging through scalable computing and data management.

Today's workshop has nine-- the workshop has nine topic panels, and we'll start in a minute with two keynote presentations on human brain and a mouse brain cell atlases, which will be followed by 12 flash presentations. Then after a half-hour break, there will be three concurrent panel discussions. Session one, breakout room one, the acquisition and processing of human brain specimens. Breakout room two, tissue clearing, molecular labeling, and scalable regions. And the breakout room three on brain histology and cytoarchitecture, cell morphology, and other anatomical phenotypes. Then after a short 10-minute break, three additional, concurrent panels will-- room one will discuss optical and non-optical imaging platforms. Room two will discuss common coordinate frameworks. And room three, use cases of brain cell atlases, math research, and medical needs. And tomorrow, workshop will start with report back and a panel summary of today's discussion, followed by a keynote presentation on Google Earth Engine platform, and then followed by three panel discussion on data preprocessing, data models, labels, visualization of summary, and data infrastructure. I'd like to thank the organization committee of this workshop, and this workshop is coordination with more than 80 workshop participants. Thank you.

NED TALLEY: All right. Thanks, Yong. That's terrific. I think it's going to be a great meeting. Now, if she is ready, I think that the next item is keynote speaker, Dr. Katrin Amunts. Katrin, do you have video and audio?

Katrin AMUNTS: Yes.

NED TALLEY: All right. So Dr. Amunts is director of the Vogt Institute for Brain Research at the University of Dusseldorf and a director at the Julich Institute of Neurosciences and Medicine. And she's also a scientific research director of the EU flagship, Human Brain Project. And she's been at the forefront of human brain mapping at multiple scales, using a variety of cutting-edge techniques, incorporating cytoarchitectonics, connectivity, and genetics to model brain structure, function, relationships, and their diversity. And she's going to be telling us about some of that work today. So Katrin, please take it away.

Katrin AMUNTS: Thank you very much for this introduction and thank you for having me in your meeting, which is, of course, extremely exciting for me as well. And there is, I think, a lot where we could think of collaborating in the future and where we can perhaps benefit from each other. So many aspects that I will show today have been indeed developed in the context of the Human Brain Project, but this work started earlier. It started, I would say, 25 years ago with Karl Zilles in Dusseldorf. And what you see today is the result of this very long year approach. And let me start with this picture, which you, of course, all understand and all know. It shows the multilevel organization of the human brain, starting from the genetic level to circuits to large networks to the whole brain level. And the brain is so complex because it has this multilevel organization, organized in space and in time. And what can be the approaches of analyzing the brain? Well, these are classical, I would say, empirical methods, but what we see is also that more and more analytical methods based on AI modeling and simulation are coming in. And this comes not for free but comes also with a certain requirement in terms of compute resources.

Okay, and how to-- I mean, how to tackle this complexity of the human brain? And here, we are in a situation that we can look back to Vogt who was the first director of the Vogt Institute that I'm leading now today. And Vogt and also Korbinian Brodmann, and Cecile Vogt, of course, they had a very modern concept of brain research, and my view is it's still true in many respects, and this is also what we are seeing here. So if you look to this first institute of Vogt in the beginning of the '30s of the last century, and you have a look to the different departments of this institute, then you see that indeed, there is neuroanatomy, architectonics, histology, histopathology, physiology, even human genetics, and experimental genetics. Yeah, and we speak about 1930. There's also psychology and phonetics, so a very modern approach that the Vogts have followed and that they have realized in their institute. So I think we are quite on a good way here together and having this very comprehensive approach.

What is our particular way of contributing to this work? And Julich-Brain is our aim. And we want to create a coherent atlas framework on cyto, fiber, and receptor architecture to approach brain organization. And there are certain conditions that is we use cytoarchitectonic maps as references. This is a choice where we started. And we use these cytoarchitectonic maps having in mind that there are, of course, levels with higher spatial resolution, cellular, genetic, molecular level, on the one hand. And there's also the other side of the spectrum, where we talk about large functional systems, areas, the brain as a whole, and so on and so forth. So we start a little bit in the middle, and this was illustrated in this first slide. We feel it's absolutely necessary to integrate intersubject variability also at the microstructure level. Brains are very variable. And in order to understand normal brain function model, in particular, also to understand the disease brain, we need to understand intersubject variability. And variability differs with respect to the level that we analyze and also with respect to the region, for example.

So we use the same logic for fiber and receptor architecture and have introduced BigBrain as a cellular template to bridge the different scales between the macroscale, I would say, and some microscale. Interspecies comparison are absolutely necessary. We cannot do everything in the human brain, and we should make conclusions based on rock and rodents or in nonhuman primates. Last but not least, we need innovative tools for atlasing and analysis to study this structure function relationships. So let me start with cytoarchitectonic maps. And here, we have developed more than 20 years ago, a method that allows to define borders between cortical areas in a more observer-independent way based on image analysis and statistical criteria. And I think this is absolutely a necessary step when we want to come up with reproducible atlases, that we have these objective tools. It's more difficult in subcortical nuclei, but for the cortex, I think there are good methods available, developed long time ago. And of course, we also take increasingly advantage and benefit from deep learning methods that help in this kind of work.

Cytoarchitectonic variability, well, we found out very early that this plays really a significant role. And this is one of my first papers showing that the primary and secondary visual cortex in 10 human postmortem brains really differs quite a lot in shape and size and position. And there's also a variability between the brains regarding the sulci, for example, the calcarine. And there is a variation in the relationship between borders and sulci. So it's a complex phenomenon, this intersubject variability. And gyri and sulci are only approximate predictors of areas. And we can see that functionally, different areas can be located within a gyrus. This is an example of the superior temporal gyrus, where we see a whole bunch of different areas. And one of the same areas can be also located on two sides or one in the single sulcus, and they can be functionally quite distinct. And this is not true only for primary areas. It's also true for higher associative areas in particular.

So sulcus-based methods, which are very popular and which are, of course, very easy to use, they are not good predictor for biologically meaningful processes. This is at least our conclusion, doing the work for the last years. And as a consequence, we have developed Julich-Brain cytoarchitectonic probabilistic maps. These are three-dimensional maps that consider intersubject variability and space and extent. And we are offering this atlas through the EBRAINS platform together with a bunch of tools, and also as input data for neuroimaging and simulation. And to map one region is extremely time-consuming, and we are not so much faster now than 20 years ago, honestly. And most of the areas that we have here are new areas. They have never been described. Of course, not by Brodmann and also not by other researchers. We are now in the next few days publishing version 3.0 with 157, I think, areas contributing to a map like this, and 175 which have more detailed and fine-grained localization of the areas. And the next portion will follow, I think, when OHBM is coming with about probably again more than 40 different areas.

So use the same logic for fiber architecture. Our approach to fiber architecture is based on polarized light imaging. This is a optical technique, as many of you know, that uses birefringent properties of myelin sheaths surrounding axons in order to identify directions of axons in 3D space with a very high spatial resolution, sorry, with 1.3-micrometer end plane or 64-micrometer end plane, as can be seen here. And the advantage is that it's really a method that has a third dimension, the direction within the section. And this allows us to 3D reconstruct polarized light imaging data. And this is so important because we want, of course, not only to have some very detailed connectivity information in a cortical layer or in a microcircuit, but we also want to capture long-range connections that are going 5 or 7 or 12 centimeters long. And PLI is a method that bridges the different scales.

We are able to study cyto and fiber architecture in one and the same section. And this is an image of a section coming from the human hippocampus. And we are able to identify the different regions in both types of specimen and really in an unambiguously way too, related to each other. We are now moving forward to add to the very same section, also immunohistology. This is a proof of principle. These are only a few one, but it's a triple staining. And that means in one and the same large human brain section, we have polarized light imaging, fiber tracts, cytoarchitecture, and the triple staining for parvalbumin, calretinin, and calbindin positive cells. And this is some collaboration with our friends from Amsterdam. So this approach allows us really to approach the connectome as a nested connectome, having hierarchies. And we are, of course, most interested in understanding what are the rules of connectivity that we see in the human brain, and can we really see repetitive structures on the different spatial levels? And what makes a human brain network so specific to be so great, for example, to identify faces or places while other jobs can be done in a much more efficient way by a computer, for example? So what are the structural differences that make networks so superior for certain types of activities and less for others?

The third element on which we are building is the receptor architecture. And here, we are mapping different receptor binding size for different neurotransmitter systems. We started this again 20 years ago, and now, Nicola Palomero and our Canadian friends are doing the first 3D reconstruction of receptors. So we will have a whole hemisphere reconstructed, volumes of receptors for a whole human brain. To bridge the different levels, the BigBrain plays a very important role, and we are populating the BigBrain with different kinds of areas, cortical layers, for example, and also provides these maps to the scientific community. And this BigBrain allows to include volumes of regions of interest. So for example, here is a combined volume of interest, where we have diffusion, tensor imaging, 3D PLI, and two-photon imaging. And we are collaborating here with our friends in Florence, Irena and Francesco Pavone, for example, and with our French colleagues regarding the PLI. And it is important also to approach such connectivity questions in one and the same tissue blocks. And I think this is a very essential step when we really want to understand this relationship considering the large amount of intersubject variability.

Interspecies play an important role. Of course, we cannot do everything, but we are identifying, I would say, strategic questions of comparison and using, for example, PLI in different species in order to come up with identical organizational principles. So all this comes hand in hand with a number of tools that have been developed mainly in the Human Brain Project. And it's available mainly at the EBRAIN's platform. And you can see here that we are really at a good level already now to include, for example, intracerebral electrodes, to ask questions about receptor distribution in a certain cortical area about cytoarchitecture, about the connectivity. So this is really a platform for multimodal data integration, and data are annotated in a very rich and comprehensive way. These are fair data, so they can be used for collaboration.

And since there is not one fits all template, we are working mainly with three templates. In some templates, MNI single-subject Colin, and the BigBrain as a template. And last but not least, we have also introduced now as a surface model, in order to communicate also with the neuroimaging community and with people working and developing FSL, for example. How to get the data where one can just go and click through the atlas, but we have also the siibra toolsuite and interactive atlas explorer available. This is a Python library, and with a few clicks and lines, it's possible to extract in an automatic way different features like here, for instance, the receptor distributions in a certain area for different receptor types. There are other data that are available and preprocessed. For example, based on cortical cell detection and layer annotation, we can identify density measures of cells in different cortical layers and different cytoarchitectonically identified areas. And quite some progress has been made in the recognition of cell bodies and these data again available.

Also, within the last few months, we have 3D reconstruct the first volumes at one-micrometer spatial resolution, so these are one-micrometer isotropic volumes of interest. And the accuracy is such that it is possible to match bisected cells. Bisected why? Because during the cutting process of the tissue, we have bisected cells into two parts. And using image analysis deep learning again, we are able now to stitch them together and to create such volumes. Also, these volumes are available. What are they good for? Well, for example, they can inform models like the virtual brain here developed by Viktor Jirsa in Marseille. He is using individual neural imaging and electrophysiological data of patients undergoing surgery for epilepsy. And then he is using atlas data from the BigBrain in order to inform the model, and a clinical trial is running in France. This is not something in the future. This is really already ongoing. That simulation helps to develop personalized methods for brain medicine.

Genes are extremely important, of course, in order to identify cause and mechanisms. And a couple of years ago together with our colleagues at the Allen Institute, we have developed a little tool that allow us to analyze gene expression data coming from the Allen Brain regarding cytoarchitechtonic maps. And again there is a tool available that makes this kind of analysis very smooth. And is this important? I think yes because we need really to compare the single-cell genomics at the end, but here only single-cell-- sorry, the gene expression data of those genes that are coding for a certain receptor type. And if you look specifically in cytoarchitechtonically identified area are four major functional systems. And the left graph shows the analysis or the representation of the gene expressions in different cytoarchitechtonically identified areas in the different functional systems. And what you can see is that there is a certain hierarchy from higher visual-- well, from primary visual area or C1 to higher visual areas. We see same hierarchies of flows of information in the auditory system, also in somatosensory.

And interestingly, the receptor pattern matches very nicely the genetic pattern, and that's interesting because sometimes we have genes that are coded or we have receptors that are coded by different genes and not only by one. So to have here really makes it possible to tell a more comprehensive story also, how the human brain is organized in terms of functions and system of information processing. What are the similarities and what are perhaps also the dissimilarities? All this is provided at EBRAINS infrastructure that HBP has built, and I invite you to have a look through it. This is the atlas as we have it now, still version 2.9. I was hoping to show you a 3.0 today, but the DAE was not ready yet, but it will come in the next few days and will include, really, a large number of new areas. I would also like to invite you to our website, where we explore our idea a bit more in detail and also put the latest events. And saying that, I would like to thank you for your attention, and of course, happy to answer questions.

NED TALLEY: Thank you, Katrin. So David Kleinfeld asked three questions in the chat. I think I can just read them out to you. I think they're general interest. I think the first question is about the resolution of the polarized light imaging, in particular, in the axial direction. And does that depend on the thickness of the slices?

Katrin AMUNTS: Okay. So the thickness of the slices for the human brain is 50 micrometer and the in-plane resolution is down to 1.3. So it's an isotropic voxel that we are using, and the reason for this is that when the slices are thicker, then the signal is stronger in order to identify, really, the direction of the fibers, so. Forgetting the thought I mentioned, the thickness is very important as an isotropic section.

NED TALLEY: Okay. We have another minute. I want to remind everyone to put their questions in the chat. Alison, I see your hand but I'm not sure. I think a lot of people want to know what-- so Katrin, what do you think the implications are for DTI-based tractography and improvements to that based on the data that you're collecting?

Katrin AMUNTS: So I think, in particular, in regions where-- I mean, DTI is great in the living human subject, obviously, and for long fiber tracts. But there are very tricky regions where you have a very rich, I would say, topography of crossing, kissing fibers. And there are also problems when the fibers are entering the cerebral cortex. So sometimes they are doing quite sharp turns, I would say. And DTI can help to inform-- PLI can help to inform DTI in such regions and, in particular, in the cerebral cortex but also within subcortical nuclei. It's a very strong method. So I would think both can benefit from each other. Yeah. Both have their strengths in terms of spatial resolution, and it would be great to combine it and then to avoid wrong tractrograms, for example.

NED TALLEY: Thank you so much. So we are out of time, and I hope, Katrin, that you're able to answer some of the questions by replying in the chat. And thanks for a terrific talk. That was really quite interesting, and it's terrific to see the progress that you're making. So next up, we actually have a joint presentation from Doctors XIAOWEI ZHUANG and HONGKUI ZENG. And I think what I am going to do is just introduce them both now. And, Xiaowei, I'm guessing you need to share your screen and get the slides started if you're going to go first. Am I right about that? And do we have both of you guys on video and audio? I'm not hearing anybody.

HONGKUI ZENG: Yeah. I'm here.

NED TALLEY: Okay. Great.

XIAOWEI ZHUANG: Yeah. I'm here too, sorry.