By integrating neuroscience, engineering and data science, the Kavli Institute at Johns Hopkins University aims to fuel new discoveries about how the brain functions.
NEUROSCIENTISTS TODAY ARE WIELDING ever-more-powerful tools for understanding the mysteries of the brain. A suite of new approaches is allowing researchers to listen in on brain activity and to measure the molecular, cellular, and structural changes that underlie complex behaviors as well as neurological disorders such as Alzheimer’s and Parkinson’s disease. But this technological burst – spurred especially by the BRAIN (Brain Research for Advancing Innovative Neurotechnologies) Initiative – has also raised a major challenge in brain research: wrangling the deluge of data that these novel approaches produce.
The Kavli Neuroscience Discovery Institute (Kavli NDI) at Johns Hopkins University, launched on October 1, 2015, will promote collaboration between neuroscientists, engineers and data scientists in their efforts to organize, analyze and manipulate the wealth of data emerging at different scales of inquiry – from single cells to behavior – in order to build a unified understanding of brain function. Recently its director, Richard Huganir, joined with co-director Michael Miller to discuss the new institute’s vision, the changing face of brain research and the value of sharing data in this new era of neuroscience.
Richard Huganir, PhD – Director of the Kavli NDI. Huganir is a professor at Johns Hopkins University and has been director of the department of neuroscience since 2006. He is also the co-director of the Johns Hopkins Brain Science Institute. His work has helped lay the foundation for understanding the molecular basis of brain plasticity.
Michael Miller, PhD – Co-director of the Kavli NDI. Miller is a professor of biomedical engineering at Johns Hopkins University and director of the school’s Center for Imaging Science. He has been a pioneer in the field of computational anatomy, which uses algorithms constructed through structural and functional imaging to characterize the trajectory of neurodegenerative diseases and their correlation with psychiatric disorders.
The following is an edited transcript of the discussion. The participants have had the opportunity to amend or edit their remarks.
To understand the brain, the Kavli NDI aims to create a framework for organizing, analyzing, and manipulating the vast quantities of data being generated by neuroscientists. Why is this data wrangling so necessary today and how will the Institute make that happen?
Richard Huganir: Researchers in this new age of neuroscience are really coming to appreciate the full complexity of the brain, and a host of new techniques and technologies are generating huge amounts of data from the micro scale to the macro scale. So the Institute’s primary goal is to contribute to gathering that data. But as you say, there’s just so much data that it’s hard to analyze and really make discoveries from it.
Michael Miller: When you think back to 25 years ago, we were all working at our own scale – on synapses, or on recording single cells as they fired, or on behavioral types of experiments. But today we’ve gone from single-cell recordings to being able to measure hundreds of neurons simultaneously. Soon we’ll be measuring thousands of neurons simultaneously. We’re asking, what information is being processed by those neurons? What does that information correspond to? And how are we going to go from explaining what’s happening at the synaptic level, all the way to the behavioral constructs?
This all comes down to figuring out how to deal with these massive data sets and how to extract the information that’s relevant to the scale at which we’re measuring, so that we can understand the function of an organism. That is a major challenge to all of us, and we believe that the next ten years in neuroscience will largely consist of many different groups connecting information across the spatial and temporal scales that we’re all measuring individually. Then we can start to understand and model across those scales so that we can go from the molecule to behavior.
There’s a growing sense in neuroscience that individual labs working in isolation aren’t going to be able to crack their walls and make the new wave of discoveries. With Kavli NDI, you’re intentionally creating an environment that fuses biology, engineering, and data science. In fact, you seem to take for granted this is essential to neuroscience.
Huganir: We have a very long history in neuroscience at Hopkins, with pioneers in the field going back 100 years. We also have incredible strengths in biomedical engineering and in data science. But most of us have been working in our own areas, and there has not been as much interaction between these domains as there could be. Over the past couple years, though, the BRAIN Initiative really has emphasized and encouraged the funding of technology development for tracking the activity of thousands of neurons or of reconstructing the connectome – a complete set of neural connections in the brain.
TKF: And how do you want to spark this exciting dynamic?
Miller: One major mechanism for promoting this type of interaction at the Kavli NDI will be what we call pipeline grants, where we will support collaborations between these three domains by building a pipeline, from gathering the data to conducting the analysis. The grants will fund projects in which neuroscientists team up with engineers who are developing novel techniques for imaging or recording, and also with data scientists who will analyze that data and be able to interpret large data sets. The idea is to fund maybe three investigators who would probably never have worked together before.
Theory and computation have always gotten a bit of a short shrift in neuroscience. How have these two levels of analysis come together in your own work?
Huganir: One major effort in my lab these days is in vivo imaging of synapses in the brains of awake, behaving animals. We’re actually able to watch synapses change as an animal is learning something. So far, we can record a couple neurons at a time, which include a couple thousand synapses. But we’re generating techniques to actually image a whole field of synapses – tens of thousands of them – within the rodent cerebral cortex. This will simply be too much data for us to analyze. So it’s a perfect example of where we need the data scientists to help us parse out patterns and information from that data.
So far, we’re imaging mostly the cortex to look at cortical plasticity. But we would also like to go into deeper structures, such as the hippocampus and the amygdala, to see how synapses in those areas change with learning. That will require fiber optic techniques for getting a look at those deeper structures, so we’re collaborating with engineers who can help us with that.
How will having data from a whole field of synapses change the kinds of questions you are able to ask?
Huganir: In the past, these types of studies were done ex vivo, which means in slices of brain tissue that are artificially kept alive, or in cultures of neurons. But imaging in live, behaving animals gives us a direct window into the brain. That means we can look in detail at how plasticity occurs within the connections made by the dendrites – the signal-receiving protrusions – of a single neuron. Does plasticity cluster within dendritic regions of the brain? How do the connections onto dendrites change across cortical regions in real time? Answering these questions this will give us much more information about brain mechanisms at the synaptic and circuit level that allow an animal to decode memory.
Dr. Miller, you describe yourself as a computational anatomist and pattern theorist – a somewhat unusual specialty. Can you explain what this is and put it into the context of big data in neuroscience?
Miller: Computational anatomy is about understanding why the human body’s shape and form varies and developing theories about this based on computational models. The underlying theory of computational anatomy is that tissues and particles develop along highly curved and predictable trajectories, but are locally influenced by their neighboring environment. With the brain, studying this allows us to study connectivity, folding, development, and neurodegenerative changes.
One of the organizing hypotheses in several neurodegenerative illnesses is that the disease starts at some focal point in the brain and then travels along circuit connections, manifesting in target locations. So to understand the spread in space and time we have to understand neurodegeneration in associated groups of networked structures, and the circuitry that connects them. We can do that by imaging at the scale of 50-100 microns to 1 millimeter – which we call the mesoscale. At that scale, we can computationally define how the developmental changes, folding, atrophy, and connections we’ve observed with brain imaging are predictive of changes in cognitive and developmental phenotypes.
How will computational anatomy contribute to the Kavli NDI’s vision?
Miller: Well, researchers are building what we call brain clouds – basically, online image databases – at scales ranging from well below a micrometer all the way to a millimeter. Through the Kavli NDI, we’re going to be able to start bringing these online clouds together, so that we can have integrated methods for neuroscientists to navigate across these scales. You want to be able to ask, “Is there any relationship between a particular behavior in the animal, and a specific synaptic density or some change at the place where the receptors that we are interested in are located?”
We have to have technologies for carrying information from different coordinate systems, and for retrieving information from particular locations in the brain for a close-up look. This integration of navigation – in the face of vast anatomical variation between brains – is one of the key technologies in computational anatomy. The huge amounts of data involved make it a major challenge.
As this new data-based vision of neuroscience continues to evolve, how will the Kavli NDI help the next generation of researchers comfortably weave between disciplines and across scales of analysis?
Huganir: It comes back to the interdisciplinary issue, where we are pairing up these biologists and data scientists and developing these projects over many scales. That has a training dimension, too. Postdoctoral researchers, students, and others involved in these collaborations will gain experience in many aspects of a project and so will be more broadly trained.
You mentioned the link between the institute’s data aims and the use of data collection in astronomy. Can you elaborate on this connection?
Huganir: People often say that the brain is the most complex structure in the universe, which I think is an overstatement, but certainly the scale of the data and information that neuroscientists and astronomers deal with is really huge.
The astronomy department here at Hopkins is very strong – the astrophysicist Adam Riess is a Nobel Laureate, and there’s a long history of working with large data sets in astronomy, starting with the Sloan Digital Sky Survey. We’re trying to get people with that big data background in astronomy and engineering more interested in neuroscience. There seems to be a natural affinity of physicists and astronomers to neuroscience – we see a lot of graduate students and postdocs coming from physics programs.
Miller: The way in which astronomers use their data sets is a model for what we’d like to create. It was Google that first developed an approach for indexing huge databases online. That was really a major breakthrough in computing because their system gave researchers an infinite ability to expand the amount of knowledge stored online.
Then as astronomers began conducting sky surveys, many realized the power of making those surveys available in an indexable way and giving other astronomers the opportunity to put their data in, too. Historically, data sharing wasn’t common in that field, just like it hasn’t been for neuroscientists, because data was the most valuable thing a researcher had. But astronomers started doing it because once they had their data in this big store, borrowing information from other astronomers’ results would make their own data even more valuable.
How will the Kavli NDI bring that experience back to neuroscience?
Miller: The astronomy folks here appreciated how important this sharing was and they creating the field of social astronomy. The approach was virally adopted as physicists realized how much value add they would get. In the same way, we hope that these kinds of initiatives with our data clouds for the brain will invite neuroscientists in because of the value they will get. We are now indexing brains, making them searchable. If we provide good navigation tools so that Rick, say, can quickly find the area of the brain he’s interested in, then it’s likely that neuroscientists will start to share their data through mechanisms like this. I’m sure we won’t be the only cloud but we will have the technology built up that will allow us to easily connect to other clouds. That’s really the ecosystem that you see building up around the Internet today.
Once we apply all of these data and engineering technologies, how do you envision the core questions about the brain changing over the next two decades?
Huganir: Well, I think that’s tough to predict. With the types of techniques we’re trying to create at the Kavli NDI, it’s possible that we can really understand how changes at a molecular level in the brain affect circuit connectivity, motor output, and obviously, behavior. That’s a big goal, but I think it’s something we can aspire to in ten or 20 years.
Miller: While writing the proposal for the Institute, we had a long discussion about how we’re always overly optimistic about short-term goals and never anticipate the discoveries that are game-changers in the long-term. It’s likely that there’s going to be an incredible breakthrough, a disruptive technology, that we’re not even thinking about, in three to five years.
To wrap up, how does the history of eminent neuroscience at Johns Hopkins set the stage for what you hope to accomplish with the Kavli NDI?
Huganir: Two great icons of the department, Solomon Snyder and Vernon Mountcastle, actually epitomize two ends of the spectrum of neuroscience history. It started in the 1950s with Vernon, who is really the father of systems neuroscience. Actually, Torsten Wiesel and David Hubel were here as well. Then, in the 1960s, Sol Snyder, who is one of the pioneers of molecular neuroscience, came along. They founded the neuroscience department in 1980 – Sol was the chair but Vernon was his partner in crime. Sol was interested in molecules and signaling and receptors whereas Vernon was interested in behavior and circuits. And to be honest, they lived in two different worlds, scientifically. Over the last 30 years the department has grown to bridge those two worlds from molecular to rodent systems to circuits to behavior and primate research. I think that the Kavli NDI will be the next step in using technical and analytical techniques to really merge these areas of research.
— Alla Katsnelson, October 2015