In Situ Sequencing: 10x Genomics, NanoString, and Vizgen $txg $brkr
Lawsuits, Sensitivity/Specificity, and Mouse Brains
I wrote a few weeks ago on Illumina and its single-cell/spatial opportunity. Currently, the player to beat here is 10x Genomics. 10x offers a lot more on the single-cell side, has a spatial platform already up and running, and offers in situ analysis, an area where Illumina hasn’t indicated an interest in playing.
As a brief review, analyzing gene expression levels has historically been done using bulk-RNA sequencing. This method reads out the average gene expression levels from a sample, but doesn’t allow you to zoom in and examine individual cells. This lack of granularity presents an issue when rare cell types are driving a disease. Single-cell RNA sequencing (scRNAseq) enables researchers to look at gene expression on the individual cell level. Standard scRNAseq methods don’t, however, include spatial information on these cells, and so researchers are still left in the dark on how these cells may be interacting with one another. This can again be quite relevant when examining things like tumor resistance, which at times is driven by the microenvironment a tumor resides in rather than the tumor’s transcriptome.
In situ analysis goes a step further than both single-cell and spatial, giving researchers a sub-cellular (rather than single-cell) level of resolution with spatial context. This comes with trade-offs: it’s more time consuming/expensive, you can’t analyze anything close to the whole transcriptome, and you need to know ahead of time which genes you want to target (so you can’t use it to discover a novel transcript you didn’t know was there). Put more simply, in situ’s a great tool to zero in on a very particular area at a level of detail that regular spatial analysis doesn’t allow. For example, you might use spatial analysis to figure out where gene expression changes are taking place in lung cancer tumors, and then zoom in with an in situ lung panel to examine a particular tumor subtype. Importantly, in situ analysis is imaging-based rather than NGS-based, making it the one product in the single-cell/spatial world where Illumina and other sequencing providers don’t benefit from a growth in its usage and don’t have a front-row seat to how the technology is used.
There are four in situ providers one frequently comes across: 10x Genomics, through its Xenium instrument, NanoString Technologies (now a part of Bruker) through its CosMx SMI Instrument, Vizgen, through its MERSCOPE instrument, and Resolve Biosciences, through its Molecular Cartography Platform. Resolve Biosciences hasn’t updated its publications page since June 2023, went through a leadership change in 2023, and has no clear way on its website to contact a sales representative. Consequently, I’ve mostly excluded Resolve from the following discussion.
Beyond sensitivity and specificity, studies comparing different instruments’ performance focus on a few key factors:
Transcript counts per gene – the number of transcripts a platform picks up associated with a specific gene. This matters because so much of RNA analysis is about quantifying gene expression! All else equal, an instrument that can identify more transcripts per gene than a competing offering will give researchers a better window into the questions they’re trying to answer.
Plex – the number of distinct genes an in situ platform can target at one time. An 1000 plex panel allows researchers to pinpoint 1000 genes in a tissue sample; a 300 plex panel would only allow researchers to pinpoint 300.
Cell Segmentation Performance – the ability of a platform to accurately draw cellular boundaries. If boundaries are too wide transcripts end up getting assigned to the wrong cell.
I looked at three different pre-print studies comparing in situ platforms. The first analyzed the performance of Xenium, CosMx, and Merscope when examining FFPE samples from various human tissue types; the second analyzed the performance of Xenium and CosMx when examining only FFPE prostate samples; the third analyzed the performance of Xenium, Molecular Cartography, Merscope, and three non-commercial methods when examining mouse brains. It’s of course worth noting that these studies are all just pre-prints, and that 10x, Vizgen, and NanoString have all improved their respective technologies since these papers were published.
Xenium emerged superior for the two studies examining FFPE tissues, capturing the highest number of transcript counts per gene with high sensitivity and specificity. That’s not to say Xenium won on every metric: its cell segmentation abilities were the worst of the bunch, and it and Merscope’s panels were much lower plex than what CosMx offered (CosMx was 1000 plex; Xenium and Merscope were more than half that).
Poor cell segmentation seems to be forgivable in the FFPE cases: Xenium’s default segmentation algorithm may be lacking, but it resulted in a minimal number of incorrectly assigned transcripts. Furthermore, one isn’t obligated to use an instrument provider’s algorithm, and as in situ usage increases there’s reason to think independent algorithms will get better and better.1 Put more simply, if an instrument has poor sensitivity/specificity there’s no way for a researcher to fix that; it’s inherent to the instrument. If a provider’s segmentation algorithm isn’t great, however, a researcher is free to use a different one.2
Xenium was not the winner when it came to mouse brains. In this study Merscope won out (followed by Xenium), and Xenium’s poor cell segmentation was flagged as leading to “a substantial increase of incorrect molecular assignment.” Merscope may be the optimal choice when it comes to mice! This leads into a related point, which is that the right instrument may just depend on the type of sample studied.
There are differing opinions on the importance of CosMx’s better plexity. On the one hand, plex obviously matters. It enables researchers to visualize a greater number of genes, which is clearly useful for answering scientific questions! That said, higher plexity is less useful if it comes with much lower sensitivity and specificity. In situ loses its utility if it’s identifying transcripts that aren’t there (poor sensitivity) or missing transcripts that are (poor specificity). Researchers are excited about single-cell/spatial/in situ methods because granular gene expression data that moves beyond bulk methods can be so informative. As one paper put it:
“A key observation from this analysis has been the importance of assay sensitivity. Simply, many analyses depend on the robust detection of relevant transcripts. Although CosMx data has consistently higher noise that masks lowly expressed transcripts, the challenges that emerge in accurate cell typing and data interpretation tend to come from the weak detection of highly expressed cell type markers….. it is critical to underscore that many biological questions will depend on the accurate detection of a small number of genes: tumor classification based on the presence of a single biomarker, distinguishing between subtly polarized cell states, the spatial mapping of cells expressing immunosuppressive factors, and more.”
Again, none of the above is to say that plex isn’t a relevant factor. CosMx now offers 6000-plex RNA panels, a 6-fold increase from when these studies were done. 10x offers 5000-plex panels, which interestingly come with a lower-sensitivity per gene than its lower plex methods. 10x’s decision to offer increased plexity at the cost of lower sensitivity seems to indicate that some researchers are happy to make that trade-off (although presumably only up to a point). It also means that CosMx’s plex is now only 20% better, rather than 2.5x better as it was the time of these preprints.
I’d be remiss to study the differences between these technologies without examining the companies these technologies reside in. A reasonable takeaway from looking at Vizgen, 10x Genomics, and NanoString is that they enjoy suing each other. NanoString went into Chapter 11 proceedings after a lawsuit with 10x over its GeoMx NGS-based spatial analysis tools didn’t go its way. For a period, NanoString was also unable to sell its CosMx in situ products in Europe following injunctions from the European Unified Patent Court and a court in Germany. Bruker acquired NanoString’s assets out of bankruptcy for 392.6mm, or ~2.3x 2023 revenues in May of 2024. Those 2023 figures are somewhat misleading: Other than a consumables carve-out for those who purchased an instrument before November 2023, NanoString is now barred from selling its GeoMx products in the US. Bruker management have said the NanoString business is currently running at a bit under 10mm a month, and they don’t expect revenue to return to pre-Chapter 11 levels in ’25.3 There’s a second, continuing, lawsuit between 10x and Bruker over patent infringement/anti-trust violations on the in situ side of things, with the trial set for August 4th of this year. Until recently, Vizgen and 10x Genomics were suing each other over their in situ technologies. 10x was suing Vizgen for patent infringement and Vizgen was countersuing 10x for antitrust violations. Those suits were settled in early February.
I think there’s a question as to how hard Bruker will push to see NanoString’s in situ business succeed. Bruker’s 2024 revenues were 3.37B, but operating income dropped from 437mm in 2023 to 253mm for 2024, its cash & equivalents are down from 488.3 in 2023 to 183.4mm in 2024, its stock is down 50% over the past year, and it’s been on a partially debt-fueled acquisition spree. NanoString succeeding isn’t of existential importance to the business, and spending too much time in litigation is unlikely to be an activity investors have much patience for.
Vizgen and 10x Genomics, on the other hand, have much more of an incentive to fight for success. Vizgen’s capabilities are more limited when it comes to both plexity and pre-designed gene panels. It offers 1000-plex panels as compared to Xenium’s 5k plex panels, and 3 pre-designed panels against Xenium’s 11, but that doesn’t mean it can’t succeed as a business. That’s especially the case with the ill will 10x created through initiating so many lawsuits. I’m not sure it's fair to view 10x as a totally bad actor here: patents exist to protect innovation, and so while it may be true that suing competitors can put them into bankruptcy (as it did with NanoString) or force them to restructure and do layoffs (as it did with Vizgen), it’s also true that it’s important to live in a society where IP is protected. That said, that doesn’t mean a lot of academic researchers were pleased by 10x’s tactics, which could end up negatively affecting 10x’s brand with critical buyers.
The other question to mull over when thinking through the future of in situ is whether improved NGS-based spatial analysis technologies will cannibalize some use cases. Until very recently, it was the case that NGS-based spatial analysis wasn’t able to give you information at a single-cell level. 10x’s Visium HD, the company’s most recent NGS-based spatial offering, now offers this. Given Visium’s far greater throughput (it can do the whole transcriptome rather than target 5000 genes), this enhanced resolution will likely be enough to keep a portion of researchers from moving towards in situ. That doesn’t really pose a problem for 10x, but it poses a very real one for Bruker, now barred from selling NanoString spatial products in the U.S., and Vizgen, which doesn’t offer an NGS-based spatial solution.
Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence
From the second pre-print: “Cell segmentation is a common feature compared between platforms. However, it is noteworthy that this is the one technical component of the data that is not inherent to the platform itself. Independent segmentation algorithms can be applied to data from all iST platforms, and as new tools are developed, this may become common practice.”
More recently, 10x has taken steps to improve Xenium’s cell segmentation.
Bruker Q4 2024 Earnings Call