Patch-seq and behavior

Patch-seq delivers three modes of data: morphology, electrophysiology and transcriptomic data. These are helpful for analyzing brain cell type diversity, brain architecture, even behavior.
Patch-seq and  behavior
Like

Behavior is, of course, hard to parse and analyze. It also tough to stack multi-modal data from Patch-seq to analyze behavior.

For Nature Methods, I recently wrote about Patch-seq, a technique that delivers multimodal data and that is part of a larger trend in neuroscience toward reaping and analyzing multi-modal approaches. Here is the story Patch-seq takes neuroscience to a multimodal place. here are several protocols and the protocols have been tweaked, for example by teams at The Allen Institute for Brain Science ad by individual labs. Some additional reporting was about how Patch-seq can help to study behavior and the brain.

In the Morris water maze, rats swim in a round tub of water. There’s an island in the tub for them, a platform, submerged right below the water surface and the rat can stand on it. In the second phase of the experiment, the rats that had experienced the tub of water and found the platform, swim again in the tub. This time, the water is rendered opaque with coloring. The platform is no longer visible. What this maze tests is the ability of these animals to remember where the platform is and to track how they navigate toward finding it.

 When analyzing an animal’s brain after such an experiment, says University of British Columbia neuroscientist Mark Cembrowski, researchers get “a snapshot” of a quite dynamic process the animal was involved in. Cembrowski is in the university’s department of cellular and physiological sciences and also at the Djavad Mowafaghian Centre for Brain Health.

“The beautiful thing about transcriptomics is, when we take something that looks structurally to just be a relatively homogeneous population of cells, and look at gene expression, there's a huge variation in gene expression,”  says Mark Cembrowski. 

Cembrowski is keen on better understanding the hippocampus, a region of the cortex that helps rodents--and people--to navigate, to track the passage of time, and it appears involved in fear, stress and anxiety. As he and his colleagues point out in one paper, “it is difficult to reconcile the putatively simple architecture of the hippocampus with such a diversity of associated functions.”

 The hippocampus is populated by CA3 and CA1 neurons and the subiculum, which is part of the hippocampus has pyramidal cells. “If you crack open a text book on the hippocampus,” says Cembrowski, it will say there is one cell type of excitatory neuron, a pyramidal cell and it is what allows the subiculum to perform various roles in memory, cognition and behavior.

“The beautiful thing about transcriptomics is, when we take something that looks structurally to just be a relatively homogeneous population of cells, and look at gene expression, there's a huge variation in gene expression,” he says. This pyramidal cell is actually a bundle of different, spatially organized, cellular subtypes, hooked into a variety of local and long-range circuits.

In his view, transcriptomics helps to begin understand the rules that govern this circuit. The idea is to learn and mechanistically understand what role cells play in memory, cognition and behavior.

What helps further in analysis is to apply Patch-seq, which delivers three types of data; morphology, electrophysiology and single-cell gene expression data all from the the same neuron.

 Under a microscope a scientist will navigate to a cell and record electrophysiology data while infusing a dye that will later be used to characterize the cell’s shape using immunohistochemistry. After those steps, using the same pipette, a scientist will remove the nucleus, extract genetic material and sequence it.

“I quicky realized I have mathematician hands,”  says Mark Cembrowski.

Extracting the nuclei from neurons in a brain slice and then performing single-cell RNA-seq, one can obtain a transcriptome. It represents gene expression in a cell in “one window in time,” says Cembrowski. One can interrogate gene expression at different times in begin animals, which is a way to resolve some of the dynamics of the behavior.

 Patch-clamping is old-school neuroscience. “Patch clamp electrophysiology is half skill and half magic,” he says. The way transcriptomics now informs on the brain and the nervous system, is, says Cembrowski, a “merging of old world and new world neuroscience.” “It has a lot of variability, too. When he first tried patch-clamping, “I quicky realized I have mathematician hands,” he says laughing. His lab uses both patch-clamping and Patch-seq. “We do a lot of patching here in the lab, of both mouse brain sections and human brain sections.”

 Patch-seq gives three types of data from that cell but it means doing all those measurements at once. “The Goldilocks Point” get the methods “have to play nice with each other” combine them all and compromise each one the least.

Tug at the nucleus of a neuron such that the membrane tips and does not re-seal all the dye leaks out of the cell. “There’s this very difficult balance of pulling out the nucleus, the main part of the cell body, and at the same time having the cell go back together so that you can see the morphology after the fact.”

In his view, the Allen Institute for Brain Science, which is also scaling up the method, leads the way in applying this method. “This is a technique that's not really amenable to doing at high throughput unless you're at the Allen where they built up all this infrastructure that surrounds it,” he says.  “It's such a challenging and niche technique that they'll probably be largely the only Institute or lab that does this sort of thing with a few exceptions in there.”

Among other facets, the prep needs to be “ultra-clean, the experiments are challenging.” The idea of sampling cells in a high throughput way across a learning and memory paradigm or an aspect that is relevant to behavior, that helps to shed light on how these cells transform during behavior “those are the angles that get really, really neat,” he says.

Neuroscience is getting good at understanding the “naïve brain.” What’s incumbent upon us, he says, is to understand how the brain changes. That can include changes due to disease or disorders and physiological processes such as learning and memory. Patch-seq can help to offer informative angles such that one can identify cells that participate in a memory. Scientists can ask: How does their gene expression change? How does their potential local micro-structure change? How do their electrophysiological properties change? He says. “Patch-seq is very well powered at a technical level to do all of those things,” he says.

 He works with mice and studies the mouse brain as a model for the human brain. One can work with animal models in ways one cannot, also for ethical reasons, do with people. But he wants to study both mouse and human brains in a way “that they can mutually inform one another,” to get at general principles that underpin the brain. With fear and anxiety, you get the molecular basic, but transcriptomics datasets can indicate targets which might help with treatment development because it’s information that “tells you how the brain does something.”

Cells can look like they have similar looking spike patterns but the gene expression properties differ widely. Alternatively, spike patterns that look very different can come from what appears to be the same transcriptomic battery or signature. “My guess is that it’s just the biology does not yield itself to a perfect interrelationship,” he says.  

Data stacks

Patch-seq is a method he thinks highly of. In a perfect world, he says, “all of the properties would co-vary,” Says Cembrowski, gene expression would predict morphology and predict electrophysiology. “But this is not the case,” he says. studying the brain in ever greater multi-modal detail can, for example, reveal, that cells ‘classically’ lumped together are actually different cell types. It’s then possible to query these multi-modal data: “To what extent are there variations? And what are the rules that govern those variations?” he says.

Knowing the answers contributes to studying some of the neuronal underpinnings of behavior, memory, spatial navigation and learning, among many other brain feats “and we're very far away from what we would hope would be this beautiful one to one mapping, where any modality can explain any other modality,” he says. “Instead, there's a huge amount of variability.”

Cells can look like they have similar looking spike patterns but the gene expression properties differ widely. Alternatively, spike patterns that look very different can come from what appears to be the same transcriptomic battery or signature. “My guess is that it’s just the biology does not yield itself to a perfect interrelationship,” he says.  

“I don’t think there can ever be too much data,” says Brandeis University neuroscientist Christine Grienberger. 

Patch-seq delivers “three different beautiful, complementary datasets” they are not all of the different perspectives needed. Much still is missing such as local microcircuit computation. The RNA being analyzed is in the cell body but plenty goes on distally far from the cell body. “There are things that you miss in these recordings,” he says. Having those extra data would be an informative way of further classifying these cells and understanding it, “but you just don't have access to it.” This means, he says, it probably precludes a very nice relationship between one type of property and another and coming up with unifying principles across these datasets.

A cell type may not be burst firing as expected says Nathan Gouwens, a computational neuroscientist at The Allen Institute for Brain Science. Burst-firing is when action potentials fire in very close sequence. The transcriptomic data might indicate “there’s actually something interesting biologically going on there,” he says.

 Patch-seq generates a lot of data and of three types: electrophysiology, transcriptomic and morphology.  Given that the brain is a complex system, “I don’t think there can ever be too much data,” says Brandeis University neuroscientist Christine Grienberger. When electrophysiology and transcriptomic data from Patch-seq experiments don’t align readily, “I think this is where it gets interesting,” she says. Perhaps, neurons can generate similar action potential output in different ways and these ways can be shaped, for example, by ion channel content, morphology, synaptic input pattern.

He sees neuroscience efforts increasingly layer data on top of transcriptomics, such as the data types Patch-seq offers. But then, when aligning and interpreting such multi-modal data, “you have to choose your battles,” says Baylor College of Medicine neuroscientist Andreas Tolias.

As Baylor College of Medicine neuroscientist Andreas Tolias explains, even with some multimodal data on hand, it’s hard to analyze behavior all the way down to its molecular facets. And that is the ultimate goal, to understand something about the brain mechanistically, such as a psychiatric disorder. One would need to travel in one’s analysis from the cellular level to the wiring to the computational level. It’s “a must,” he says. But it’s also true that not all scientific questions require this kind of comprehensive journey through a multi-modal data stack.

Tolias sees neuroscience efforts increasingly layer data on top of transcriptomics, such as the data types Patch-seq offers. But then, when aligning and interpreting such multi-modal data, “you have to choose your battles.” One day, it may become possible to predict neuronal function from the transcriptome. ‘How far you can take this?’ is one of many open questions, he says.

 The Allen Institute for Brain Science has eight Patch-seq rigs, and has been using these rigs to obtain multi-modal data from thousands of cells. Eight rigs, that’s indeed a lot, says Tolias. But it’s also true that some questions, such as those related to behavior, might need much more data. “Maybe we need 100 rigs to figure something out,“ he says. The idea, he says, is to potentially apply many Patch-seq rigs or other high-throughput multi-modal profiling to study different cell types, across natural behaviors and in experiments with stimuli or in which animals learn tasks.    

 As part of the Brain Initiative Cell Census Network, the Allen Institute and colleagues at a number of institutions have built atlases such as of the motor cortex. It’s a transcriptomic atlas built from the analysis of thousands of disassociated cells.

 The Tolias lab and the Allen Institute teams are using Patch-seq to explore the logic of how cells in the brain are organized. In such an analysis, such as the one that led to the ability to discern elongated neuroglia from cells (eNGCs) and single bouquet cells (SBCs), which are both inhibitory neurons, the data grouped in large clusters, with the data visualization looking a bit like “a large banana tree,” he says with a few large leaves, or larger clusters, and within each cluster there is transcriptomic diversity.

In some instances, morphology data might segment into clusters that can be grouped by size. If one were to use only a few thousand cells, one lacks the data to see “the details of the leaf.”

 For analysis, when working with dissociated cells, says Tolias, the disassociation can alter the transcriptome. Most of the data they recorded were from cells in vitro. Ideally scientists want to get in vivo data, he says. For example, he and his team recorded electrophysiology data from L1 interneurons in vivo in anesthetized animals and extracted cell content for RNA-seq.

But the resting membrane potential fluctuated too much to use that data alone for cell type classification. The transcriptomes, however, let them classify the two molecular interneuron groups that correspond to eNGCs and SBCs, using a few thousand highlight variable genes and identified two molecular interneuron clusters that correspond to eNGC and SBC.

 “Ideally what you want is in vivo,” he says, referring to in vivo Patch-seq. With in vivo experiments one is bound to encounter more contamination on the pipette tip. He and his team recorded electrophysiology data of nearly three dozen L1 interneurons in vivo from in anesthetized animals, extracted cell content for RNA-seq. The resting membrane potential fluctuated a lot so it was not possible to classify the cells based on these data, but they were able to cluster the interneuron transcriptomic data using a few thousand highly variable genes and identified two molecular interneuron clusters that correspond to eNGC and SBC.

Depression research, a hypothetical scenario

Patch-seq, says Tolias, is one technique of many and it’s still quite a journey to connect data it yields to behavior. It might at one point help to model depression which is quite difficult to model in an animal such as a mouse.

One might, says Tolias, try to model a genetic component of depression. One could use Patch-seq to try and build a transcriptomic atlas of the brain of a ‘depressed’ mouse and a non-depressed’ mouse. Then also build an atlas of brain shapes, which would be a morphology atlas of the brain of a ‘depressed and ‘non-depressed mouse.’ This is all just a hypothetical example to sketch out potential paths to take.

 Then one would—in this hypothetical scenario—try to link ‘connectivity map’ of these two types, tracking, for example, if some genes changes because of genetic background, because of different connectivity, because perhaps there is a higher level of inhibitory neurons firing a certain pattern in the brain of a depressed mouse. “I’m just kind of making up a story,” he says.

 These insights, such as genes mutated in the ‘depression’ condition and expressed in a specific cellular subtype, might have elevated activity. With such results, that is when work on a drug candidate might unfold, one that decreases this elevated gene expression and that could perhaps address depression. This, says Tolias is the kind of theoretical roadmap scientists think about when it comes to multi-modal data and complex behaviors.

 Learning about learning

Christine Grienberger at Brandeis University is a new principal investigator who runs a small lab focused on how the brain learns and looks at a number of different levels: the level of cells, synapses and circuits. She also studies the hippocampus, notably the entorhinal-hippocampal circuit that is involved in spatial learning. She uses optogenetics and patch-clamping. She has tried Patch-seq for in vivo recordings, to try and identify cell type based on the previous RNAseq work and to start to understand the role of cell-type diversity in the learning task she was assessing.

 “Unfortunately” she says, tissue contamination, “or so I assume” made it too challenging to interpret the RNA-seq data so she decided to put those experiments on hold for the moment. Until she and her team have addressed how to best do the experiment, one plan is to do the in vivo recordings from fluorescently labeled neurons and also do the other parts of the Patch-seq analysis, dye infusions and single-cell RNA–sequencing.  

Patch-seq and other techniques can deliver reference datasets, which are “incredibly useful” says Grienberger. The community can use them and such datasets help standardize nomenclature or technology or other aspects across the field. “However, I don’t think they will be able to address all the scientific questions that are being worked on,” she says. In the area of her interest learning and memory, different types of learning might affect transcription differently in different cell types. In this case, experiments will probably need to be tailored to the question. “Large institutional efforts and small lab research complement each other here, which is great and will be to the benefit of the entire community,” she says.

Says Grienberger, “this is an incredibly fascinating time to be a neuroscientist, with all the experimental and computational tools that are available to us. Together with the push towards open science, I am just excited to see all the different datasets that are being generated.”

(Getty Images/iStockphoto)

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in