10x Genomics recently held their second annual Xperience this past month. During the event, which can be watched here, announcements and updates were provided for both the single-cell analysis and spatial businesses within 10x Genomics. On the single-cell side, fixed RNA profiling, single-nuclei isolation kits, CRISPR 5’ screening, barcode enabled antigen mapping (BEAM), and ATAC v2 were introduced. For Visium, further announcements regarding the upcoming CytAssist and Visium HD were discussed, and the combined gene and protein expression assay was also announced. Xenium, originally teased during last year’s Xperience, was covered in greater detail. Following the event, I had the opportunity to discuss the spatial announcements with Michael Schnall-Levin. You can find a transcript of our discussion below.
Michael Schnall-Levin, PhD, is the SVP of R&D and founding scientist of 10x Genomics, helping guide the company’s research divisions since its inception a decade ago. Prior to co-founding 10x Genomics, Michael worked as the Chief Product Officer for Medigram and as a Computational Biologist at Foundation Medicine.
Michael, thanks for joining me. I really appreciate getting the opportunity to chat about the announcements made at Xperience. 2021 marked another big year for 10x Genomics, and we at DeciBio have observed the large growth in access and use of 10x Genomics’ platforms over the past year since the first Xperience and years prior. Starting with Visium, the combined spatial gene and protein expression assay is expected to be released mid-year. What customers and applications, in particular, did 10x Genomics have in mind when developing the combined assay, and do you anticipate any differences in comparison to the original transcriptomics assay?
In general, I think the world is really beginning to appreciate multiomics and we’ve been able to observe that on our single-cell platforms. You’re now able to profile T-Cell receptors, and surface proteins, and antigens on our 5’ assay. It’s the way the world is moving and it’s something we’ve been thinking about for a while here at 10x for the spatial platforms. For Visium, moving into multiomics with gene expression and protein, we’re going after certain kinds of cells and cell states where proteins are the key markers. A lot of people are really interested in getting that whole transcriptome view, but also want specific information on protein markers. I think this expansion deepens our ability to address immuno-oncology and other applications in which you are looking at tissue infiltrates, whether that’s in cancer or auto-immunity. Looking forward, I think there is an opportunity to expand the protein content, and break into neuroscience where researchers are quite interested in better understanding protein expression, on top of the RNA layer.
Of course, and looking down the road towards translational and clinical applications, most of those biomarkers are proteins today. Building on Visium with Visium HD, it’s now expected to be released in the latter half of the year, pushing resolution to the single-cell level and allowing for highly detailed tissue analysis. For researchers considering Visium and Visium HD, what research questions or tradeoffs do you see catering to one assay or the other?
We’re going to have to see how it plays out ourselves, right? So, part of it will come when it gets out on the market and we get to see what people want to do with it. A lot of people are really excited about Visium, and there are a lot of questions you can answer with it. There are somewhere around 200 publications and preprints already, so there’s been very strong adoption. However, there are certain questions where you want to probe even deeper than the the multi-cellular level of Visium, and get down to the single-cell level. It cuts across a large swathe of biology, and, again, if you’re looking for tumor infiltrating immune cells and you have smaller isolated cells, or you’re looking for fine structure within tumors and tissues, those will be enabled by Visium HD. There are things you can do well with Visium, but people are excited to get to the higher resolution. I think there are more questions in developmental biology and tissue architecture questions which can already be addressed by Visium, and will only be improved on by Visium HD. It won’t necessarily turn on new applications but will expand the questions which can be addressed within the spaces we already cover.
Based on my discussions with researchers, people are definitely very excited by the increased resolution coming with Visium HD. Building on the relationship of Visium and Visium HD, during Xperience, Visium was highlighted as a complementary, and confirmatory, assay alongside Xenium. How do you expect 10x Genomics’ spatial platforms and single-cell platforms to be used alongside each other once on the market? Do you see them continuing to work in this complementary manner?
Just to clarify, we don’t necessarily see Visium as confirmatory, but an upstream and discovery platform. Because Chromium and Visium are both whole transcriptome and you don’t have to make any decisions on the genes you’re looking at, they really cater to discovery. On the other hand, Xenium, we see as more of a downstream platform once you’ve made those discoveries. More broadly, the reason we’ve invested in all three of these approaches and are excited by them is because we see them as inherently very complementary. You already see that out in the literature. I think there was a paper on the cover of Cell a few years ago that used both spatial transcriptomics and an earlier version of in situ sequencing, there was a fetal cell development atlas that used both single cell, spatial transcriptomic data, and in situ sequencing. We think there are a lot of cases where people will utilize information from all three platforms. A lot of the work people want to do with, say, Xenium, needs to have a starting point that is unbiased, and that will frequently come from prior studies with Chromium and / or Visium. We’ve had customers asking us to help go between their single-cell data and their Visium data and transitioning that into an “optimal” Xenium experiment. It signals to us that there’s a really strong coupling across these platforms.
Of course. In Xperience I believe it was more confirmatory specifically for the development process, but regardless, one last note on Visium, this has been a product which has so far enjoyed a complementary relationship with a lot of the high-plex proteomics platforms on the market. How do you anticipate this changing with the launch of the combined gene and protein expression assay that we discussed at the top of the call?
I wouldn’t necessarily expect a massive shift, to be honest. With this new version of Visium, we’re really going after multiomics, so if you want RNA plus the protein, you can have it. We’re not going after protein-only solutions. I think it will remain complementary with high-plex proteomics approaches, but we’ll see how all this shakes out over time. In general, we see that people, at least for discovery and translational questions, want to get more data. A lot of people want that RNA and protein data together, and we hope that the solutions we provide will get good adoption for that.
We’ve seen the market over time shift from proteomics to the transcriptomics hype spurred by Visium to now converging around multiomics in this next generation of platforms. With that, shifting focus to Xenium, the platform is expected to enable single-molecule RNA and protein analysis at a sub-cellular level and is launching with a set of curated panels. What research applications do you expect to be key for Xenium, and how may these differ from Visium and Visium HD?
Again, I think, at the highest level, it will cut across large swathes of biology, and hopefully every aspect of biology eventually. Neuroscience is an area where people have been extremely excited about in situ, as well as immuno-oncology, and anything else looking towards those infiltrates or tissue architecture. So, at the highest level, it’s quite similar to Visium. Where it ultimately gets differentiated is around the kind of questions you’re asking. If you’ve used Chromium or Visium to discover some really interesting biomarkers, you can then have more targeted panels on the Xenium system and scan through a large number of samples to see how the biomarkers correlate with things like treatment of disease, progression of disease, and heterogeneity of disease across individuals. So, it’s neuroscience, immuno-oncology, oncology, immunology, developmental biology—pretty much every area of biology eventually, but these are where we’d focus initially.
Well, I would similarly hope that these platforms make it into all aspects of biology. As a follow-up on these off-the-shelf panels, how have the applications and targets been prioritized for these assays with the understanding that it is an ongoing process?
Absolutely, this is 100% an ongoing process. It’s pretty simple, we talk to customers. We’re engaging with the first set of Xenium customers, and a lot of leaders in this space, and we’re discussing what they want to do with the platforms and what the research sphere overall would want to do. What emerges from that is, at least for the top set of tissues and content, a pretty clear prioritization. You can always quibble over 3rd place versus 4th place, but the top ones are fairly clear. We are continuing to engage with early adopters on this, and I think that’s one of the fun aspects of the development, as we’ll get a lot of feedback from people on what’s working well, what isn’t, and where they want improvements, and we’ll be able to iterate quickly on that.
Of course. FISSEQ, fluorescent in situ sequencing, is one of the methods Xenium is based on, and had historically been claimed to enable the analysis of RNA, DNA, proteins, and therapeutic compounds in situ when originally commercialized by ReadCoor as the RC2 prior to your acquisition of the company. As discussed, at launch, Xenium will include 10x-developed panels, but flexibility of the platform and the opportunity for custom panels were also stressed during Xperience. What levels of flexibility can customers expect, and will there be a consideration or ability to expand analyte types in the future?
I’m glad you asked about the customization because I forgot to add that in my last response. We’re working on these panels because a lot of customers like having the resources for pre-defined panels across a segment of applications, but we also hear from customers that they want the ability to add in customization for their specific biological questions. That’s been built in very strongly from the start, and we’ve spent a lot of time thinking about how to expand on the customization capabilities and anticipate what customers’ asks will be in the future, to sort of create a roadmap on customization.
One clarification is that, what was originally described as FISSEQ is quite different from the method powering Xenium. Obviously, we acquired ReadCoor and Cartana AB, which was a smaller startup in Stockholm, and Xenium is a bit of a mishmash and a bit of completely new stuff from 10x Genomics. There was a ton of know-how from both companies and their foundational technologies. ReadCoor had been working on every aspect of in situ for 10 years and had a plethora of IP, which played into how we thought about this as well. I wouldn’t want people to think that this is just FISSEQ, because if you look back to the original Science paper, Xenium looks pretty different.
For the question on multiple analytes, of course. Today RNA and protein are the top focus, and we’ve considered DNA and methods to implement it, and while it is of interest, it’s not an immediate focus area. We have also thought about other analytes as well, and you can imagine being able to detect all sorts of compounds which people likely can’t taxonomize properly today. Decades down the road you could foresee wanting to look at every molecule in the cell instead of just the nucleic acids and / or proteins. That’s something that’s inspiring to us on the multi-decade timescale. There’s no shortage of very interesting biology, metabolites, and lipids which could be measured with the right tools.
Well, I wouldn’t have expected 10x Genomics to just rebrand the RC2 as Xenium and call it a day. I imagined there was some additional R&D magic behind the curtain, and I look forward to seeing how it works when the method comes out. Continuing with Xenium, while Visium launched as an assay which plugs into an existing install base of NGS instruments, Xenium is launching as a stand-alone, fully automated, high-throughput platform. How important was it to develop a sample-to-answer solution for in situ, particularly with analysis built in, and how important will these simplified workflows be for driving adoption?
It was absolutely critical for in situ, and that’s partly because of the nature of the product. You’re basically in this decoding process going through complex fluidics, imaging, and registering, overlaying, and compiling all these images together. This is something people can do with their own microscopes, and we’ve seen people do that, but it’s nowhere near optimal. When you do this in an end-to-end instrument, you get massive gains in making it a lot easier for customers, but also get gains in throughput and robustness. So, I think, it was critical, and as we drive adoption for any of these platforms across early adopter technologists to more general biologists who don’t have as much experience with novel technologies, the platforms need to become increasingly turnkey. We think about this a lot, and it’s the path that a lot of these technologies follow. To maximize your adoption curve, you need to make them easier to use as time goes on.
Running with that theme of ease of use, one thing that has long been difficult in this industry is analysis. Improvements and expansions to the Loupe Browser and the Cloud Analysis solution were also noted during Xperience in addition to the built-in analysis capabilities on Xenium. Michael, most of our readers likely aren’t familiar with your background, which is in mathematics and includes the development of algorithms which detected mutations in tumor sample sequencing data. What do you see as the key issues within bioinformatics and analysis when it comes to spatial omics, particularly as 10x Genomics pushes into multi-analyte combinations at sub-cellular resolution, and how important do you expect AI- or ML-based analysis to become?
Totally. There are a few different challenges. First, there’s just the practical challenge, which is especially true for Xenium, where you get a ton of data out, and in its raw form, no human can understand it. That’s also true with NGS, but there’s a more mature tool chain now, so it’s not quite as overwhelming for people. One of the first things we’re really focused on is being able to analyze that data rapidly, robustly, and well enough to get it to output formats for hand off points to more extensive analysis. For advanced labs that want to go and do their own software development and their own bioinformatics, we can, similar to Chromium and Visium, provide the initial data processing and image generation aspect of the method and make it as easy as possible to hand off the data in clear, open file formats.
There is then a challenge for the 95% of researchers who don’t want to do the software development themselves. Even once you get the analysis to a point where you know where every molecule is, there are still open questions. You want to ask, where are all my various cell types? How does this group of cells differ from this other group? Where are the most differentiated genes? How do the cell types correlate with morphological data? There are a whole host of questions which get back to the biology, and our goal is to keep taking on these core, central questions which touch a lot of customers and streamline those.
For the question on AI and machine learning, I think it’s very interesting, and I think there’s a lot of opportunity. We’re doing some of it already for things like cell segmentation with deep learning models internally to generate training data to look at where different cells are located and how the images of cells look across tissue types. That’s just one early area where there’s a lot of power. Downstream, I’m really excited about the opportunity to do a lot more machine learning approaches which are able to make interesting queries on things like case progression in these rich images and rich datasets. But that’s something that’s quite long-term and will develop over many years.
You mentioned open file formats just now, and Kamila Belhocine, who covered Xenium during Xperience, noted a desire to develop and embrace open standards for image generation and analysis on Xenium. With how many different platforms and analysis methods are available today, how do you see this standardization developing across the spatial industry?
NGS is a good parallel here. In what ultimately won out, there were standardized file formats that could then be handed off to whatever data analysis you wanted to do. I would expect in spatial, particularly in major applications, to have open standards. We’re in favor of that. When companies try to do very proprietary file formats with customers’ data, customers don’t like it and it impedes progress in the field. I think that is how it will evolve, and it’s certainly something we’re committed to, and that’s true across all of our businesses with software and output. Given the way other technologies have evolved, open formats are the way spatial will go.
Lastly, to round out our conversation, the number of publications utilizing Visium has grown precipitously since launch. Thinking back over the publications you’ve seen using 10x Genomics’ spatial platforms, which have been the most exciting to you?
Certainly. With Visium, we’re now starting to see the first publications come out using FFPE samples, as that product was launched in the middle of 2021. We’re starting to see those publications make it through the peer review process now. There was one which just came out in Immunity (DOI: 10.1016/j.immuni.2022.02.001), which looked at tertiary lymphoid structures—which I had never heard of before and are these conglomerations of immune cells that develop within an area of inflammation, like in a tumor. That’s a pretty exciting paper which is using both FFPE and fresh samples with Visium.
There was also another publication on the cover of Science Translational Medicine (DOI: 10.1126/scitranslmed.abj8186) recently, which looked at the locations of pain receptors across the brain and how that differed across rodents and primates and related to pain perception. Those are just two, but there is a ton of exciting research coming out and we’ve been able to see a lot of preprints recently of what’s to come.