BARSeq Reveals the Brain Is Like a Pointillism Painting

The cortex of the brain is made up of various different types of neurons, which are organized into defined regions and drive specific functions. Neurons of the visual cortex, for example, receive and integrate inputs from the retinas. In the motor cortex, neurons are responsible for organizing and executing voluntary movements.

Whether or not the spatial organization of these neurons is reflected in their transcriptomics signatures – and how such signatures arise in neurodevelopment – has not been well established. This piqued the interest of researchers at the Allen Institute for Brain Science, who also wanted to explore the extent to which peripheral sensory inputs could play a role in shaping the cellular composition of different cortical areas.

Dr. Xiaoyin Chen, assistant investigator at the Allen Institute for Brain Science, develops and applies sequencing-based approaches to explore the wiring logic of neuronal cell types across brain development and evolution. He was part of the team that developed BARseq, a high-throughput sequencing method that allows scientists to read and localize RNA barcodes in situ.

In a new study published in Nature, Chen and colleagues applied BARseq to interrogate gene expression patterns over four million cortical neurons across nine mouse forebrain hemispheres, at cellular resolution. They found that the transcriptomic signature of cortical neurons is highly predictive of their cortical area identity.

What’s more, when the researchers induced blindness in the mice, the transcriptomic composition of the cortical cell types shifted, becoming similar to neurons from adjacent areas. This provided evidence that peripheral inputs can indeed alter the identity of cells during development.

Technology Networks interviewed Chen, who is the lead author of the study, to lean more about the development of BARseq, its application in this study and why the brain can be thought of as a pointillism painting by Georges Seurat.

Molly Campbell (MC): Can you talk about the development of BARseq? How does it work and why is it possible to sequence areas of the brain at such a high speed?

Xioayin Chen (XC): BARseq is a way to sequence RNA molecules directly in tissue using an approach that is based on conventional high-throughput sequencing. In conventional sequencing, DNA molecules are spread out in a flow cell and amplified. Their sequences are then read out through a series of chemical reactions followed by fluorescent imaging. In BARseq, we basically perform the same procedures, but directly in situ. This allows us to read out the RNAs that we are interested and see where they are directly in tissue.

Because BARseq, like conventional sequencing, is based on fluorescent imaging, anything that make imaging faster also makes BARseq faster. We achieved high-throughput by making the signals very bright and by optimizing the microscope to maximize how much tissue we can image in a given amount of time.

We developed BARseq when I was a postdoc in Professor Anthony Zador’s lab at Cold Spring Harbor Laboratory. Initially, it was designed to sequence random RNA sequences (i.e., RNA barcodes) that we introduced into neurons in the brain to map axonal projections. Using this approach, we revealed a lot of diversity in the projections of cortical neurons, and we wondered how the differences in projections correspond to differences in gene expression.

Since BARseq is based on in situ sequencing, we optimized it to read out both the mRNAs of many genes and RNA barcodes together in a second study. We realized that BARseq is extremely fast and low-cost compared to other spatial transcriptomic techniques. These advantages are ideal for understanding the cellular organization of large brain structures like the cortex, because it requires reading out gene expression in detail across large spans of tissue. In the current study, we optimized BARseq as a standalone spatial transcriptomic technique.

MC: What inspired you to apply BARseq to explore the impact of sensory stimuli on brain development? 

XC: The cortex contains many areas that are specialized in different functions; for example, the visual cortex processes vision and the motor cortex plans movements. As recent transcriptomic studies have suggested that different cortical areas can share similar cell types, we were motivated to understand how cortical areas can be different from one another if they have the same cell types.

First, we focused on one brain and found that even though cortical areas can have the same cell types, the proportions of those cell types can be different from area to area; in other words, the compositional profiles of cell types are like signatures for each cortical area. We wondered how these differences were established in development.

This has been an important question in the field for the past few decades. Researchers already knew that peripheral sensory inputs play an important role, but we didn’t know how they would affect the signatural cell type compositions of cortical areas. Since BARseq can be applied quickly across whole brains, this approach gave us an opportunity to address this question with unprecedented detail.

MC: Can you discuss your findings that demonstrate how vision is important in the development of the brain?

XC: There were two findings that were particularly striking to me. First we found that, in the visual cortex, cell types across all cortical layers were affected by the loss of visual peripheral inputs. This effect was much stronger than what people have seen previously by manipulating visual inputs at a later stage.

Second, beyond the strong effects seen in the visual cortex, many areas outside of the visual cortex also showed changes in cell type proportions. Thus, it appears that vision affects not only the parts of the brain that directly process visual signals, but also other parts of the brain that are involved in other types of perception and behaviours.

MC: Can you discuss the implications of your results?

XC: Our main findings can be summarized in two points. First, we revealed that different areas in the cortex can share similar building blocks – cell types – but they are different in the proportions of these building blocks. In other words, the brain can be thought of as a pointillism painting by George Seurat; in Seurat’s paintings, people and objects are painted with combinations of colored dots (analogous to cell types in the brain). In different parts of the painting, one can usually find dots of similar colors, but because the proportions of those dots vary, those parts of the painting take on different shades and colours to our eyes.

Second, we showed that these proportions of cell types are shaped by vision, but the cell types themselves remain largely the same. Using Seurat’s painting as an analogy again, without vision, one could still paint with dots of the same variety of colors, but it would be difficult to find the right proportions of dots to achieve the perceived colors one has in mind. As a result, even though the colors of the dots remain the same, objects and people would become dull and murky.

Technically, our study demonstrates a new way of examining how gene expression and cell types are shaped by development systematically.

In previous studies, people were able to use similar approaches to reveal what gene expression was like in a whole “standard” or “model” brain. However, because one could only apply these approaches on a brain-wide scale to a small number of brains, they could not reveal how this model brain organization is shaped by biological processes. Separately, other studies focused on smaller brain regions and could reveal changes across many brains, but these studies may miss other changes in other regions of the brain. Our study combined with the benefits of both approaches to reveal developmental changes over the whole cortex.

We focused only on looking at how perturbing vision changes cortical cell types and areas, but similar approaches can be easily applied to many different questions, for example, how brains age evolve and change under disease conditions.

MC: What would you describe as the key challenges in this study, and how did you overcome them?

XC: To compare cell types across brains, we had to collect high-quality data that were consistent across all the brains. These datasets were collected across different days, in different experiments and by different experimenters. This required us not only to make BARseq extremely robust (meaning that the data quality does not change dramatically with small variations in how people perform the experiment), but we also had to “standardize” procedures to make sure that all experimenters perform the experiments the same way.

This project was a large collaboration across five labs with complementary expertise, and it came with all the challenges associated with coordinating across a large group. We were extremely fortunate that all of our collaborators shared our vision of the project and were highly motivated to push the project across the finish line.

MC: Are there any limitations to BAR-seq that need to be addressed?

XC: As a next step, we are really excited to combine BARseq’s ability to reveal the interplay between gene expression and projections in development – how incoming projections shape cortical areas, and how gene expression contributes to establishing specific outgoing projections. Achieving this goal will require optimization and adaptation of BARseq projection mapping for developing animals, and we will need to develop computational approaches to analyze these new types of data.

BARseq is straightforward to perform and requires only commercially available reagents and equipment, but it is a specialized technique that only a handful of labs can perform currently. To help others use BARseq, we are building an open-design automated BARseq machine and an open-source data processing pipeline.

In the meantime, we are providing students with hands-on training in BARseq, along with other modern neuroanatomical approaches at the Cold Spring Harbor Laboratory High Throughput Neuroanatomy Course. With these training efforts and automation, our goal is to make it possible for any lab to establish BARseq in their labs and use BARseq routinely in their everyday research.

Dr. Xiaoyin Chen was speaking to Molly Campbell, Senior Science Writer for Technology Networks.

About the interviewee:

Chen is an assistant investigatory leading the Barcoded Connectomics project at the Allen Institute for Brain Science. Chen and his team are developing and applying sequencing-based neuroanatomical approaches to understand the wiring logic of neuronal types across development and evolution.