In the latest of our digital pathology interview series, we spoke with James Mansfield, Senior VP, Research Business Development at Visiopharm to discuss the intersection of multiplex tissue analysis and digital pathology.
• It is not just clinical validation, but clinical utility, that is preventing more widespread uptake of multiplex tissue analysis in the clinic. With some exceptions (e.g., transplant panels), tests of >3 plex have limited actionability vs. their single plex counterparts.
• On the research side, biopharma is frequently using multiplex technologies to analyze patient samples, sometimes at multiple levels per patient (e.g., H&E, low-plex IHC, high-plex spatial analysis). Analytical tools, such as those offered by Visiopharm, will be vital for making sense of the increasingly complex, high-plex output.
• There is a significant unmet need for tissue staining QC, which can be accomplished via digital pathology tools. Internal variation in staining is commonplace in clinical labs and can have an impact on therapeutic recommendations.
Jim, thank you for joining us. To start, could you share your background as it relates to digital pathology and multiplexing?
I wound up in digital pathology by accident, I suppose. My original background is as a spectroscopist, and then I started building multispectral cameras. After a few years of that, I started working for a little company called CRI, Cambridge Research and Instrumentation. It was really the company that started the whole multiplexed imaging field in pathology…At the same time, there was this digital pathology field which was just starting to take hold, so we were talking to companies about incorporating a multispectral camera into their whole slide scanner.
Then in around 2010 or 2011, there was a big shift because I realized that imaging of immune cells was really tricky in tissue sections because you need multiple markers to phenotype immune cells. I did the first application that was looking at multi-marker phenotyping of immune cells in situ in cancer; and it really took off. CRI was then acquired by Caliper Life Sciences, who was acquired by PerkinElmer. PerkinElmer then spun that technology out to Akoya Biosciences – that's a technology that is now in the Vectra Polaris and PhenoImager and their inForm software.
About a year and a half ago, I joined Visiopharm, which is an image analysis company. They don't make any hardware, which for me was a bit of a switch because I'd always done hardware-oriented stuff, but I’d come to the conclusion that multiplex staining and imaging hardware was becoming a bit ubiquitous and the analysis was the bottleneck. I started at Visiopharm to work on the software side of their business, where I have been doing business development for analysis of higher plex images.
It's interesting how your career path from hardware to multiplexing to image analysis has matched the development of the field in a way. How would you characterize the state of multiplex digital pathology today, both in research settings and in the clinic, to the extent relevant?
The move to the clinic is always top of mind, I think it's really what everybody wants. Clinically, as far as I know, there are a few labs – I'll say 5 or 10 maybe, but it might be more – who have taken some of these multiplexed imaging systems, typically the Akoya Vectra products, and made their own lab developed tests (LDTs). They've done all the internal development in their lab to show that it's valid and it can be used clinically. I know that there are some of those that are used regularly, but none of it has FDA approval.
I think it's going to be a while before there's a fully FDA approved multiplex method. Part of it is just the validation, but part of it is also the clinical need. When will somebody have a test that actually changes clinical practice? Then you need to get the approved CPT codes and need insurance companies to pay for that. I think we're still four or five years away from true clinical multiplexed imaging, where it's all approved, and it's all paid for. I think it'll slowly trickle its way in because LDTs are getting more common. Maybe some multiplexed chromogenic work, like low level duplex/triplex/four-plex stuff, and chromogenic brightfield and microscopy will make it to the clinic first, as there is a lower bar to entry for that.
But having said all that, the vast majority of multiplexing is in research, whether that's translational research or basic research or drug development or biomarker exploration… there's a lot of areas. But that's 98-99% of what people are doing right now.
We’ve seen that multiplexing is largely used in biopharma / research. How do you see biopharma using these tools in their research? And what are the unmet needs or pain points that multiplexing and digital pathology are addressing?
I think it's an interesting transition time because three years ago, there was really just Vectra and other 6-plex, 7-plex platforms (what we would now call low-plex); and pharma uses those a lot. There are a lot of those systems out there for biomarker discovery, clinical trial support, and translational work. Now there are even higher plex tools, whether it's imaging mass cytometry or cyclic staining systems. There's this natural tradeoff between the number of markers that I can do versus the amount of sample that I can image in a given amount of time. The cost is a factor too, of course, because buying 100 antibodies for one slide can be very expensive. So, I think there'll be an interesting balance between really large areas of sample at lower ‘plexes, like 10 or less, versus doing smaller areas of sample at 40 to 50 to 100 plex.
Of course, to go with that, there's this whole new area – I'll just call it spatial biology – where it's not necessarily even imaging the tissue, but they can do 1,000 plex analysis on spatial areas of a sample without forming a highres image of that area.
I do think the biggest unmet need right now is image analysis. And I don't mean that just as in how to analyze your image, but what to do with the data. Once you've found all your cells and all your phenotypes, how do you do the spatial analysis of that? What are the distances between cells and the metrics that are useful? And how do you combine phenotyping results with some of the spatial biology stuff like GeoMx, or Visium, which are large-area imaging?
What do you perceive to be the applications for the different plex levels? As you alluded to, these lower plex technologies are more clinically relevant, whereas the larger plex techs are much more research focused while we try to figure out what to do with the data. How does that paradigm evolve?
I think given the natural technical tradeoffs between ‘plex level, area, cost, and time, the requirement to image smaller areas of the sample when doing really high-plex analysis will be there for a long time. What I think we will wind up with is having three levels of imaging: you have an H&E and IHC, standard pathology imaging and analysis. Then you'd have a lower plex, 4-8 plex or so on a whole slide, and you can use that to do some simple phenotyping of cells where you might see a cell relative to macrophages relative to tumor boundaries. And because you can do that on a whole slide, you can then use that to help guide where you will do the region of interest analysis at 40-100+ plex. Then you can merge all those datasets into one big data set that contains the information from them all.
I think that's one of the things that we are going to be trying to work on. We can already merge them, but how can we analyze them together to take advantage of all that? I think that stepwise, first you look at H&E, then you look at a lower plex, then you look at your high-plex, and then you put all that back together. I think bringing it together, all these old and new technologies, into one unified way to look at a slide is probably where it's going. I think there's a lot of work that has to be done for that.
When do you expect these multiplexing applications to transition into the clinic? We've seen some of the barriers, including things like reimbursement, infrastructure, clinical utility. What does the transition to the clinic look like?
Leaving aside the barrier of clinical utility and payment – those are sort of their huge issues, but they’re very hard to speculate about – one thing that I think is going to affect the transition of multiplexing into the clinic is what you mentioned: infrastructure.
There are way more brightfield scanners than fluorescence in the clinic. I think the first multiplexing tests that go through the clinic will be brightfield -- maybe you take two serial sections and we put three or four markers into each one and you merge those into one six-plex or one eight-plex slide, and then you do the analysis. That way, it can be run on standard clinical brightfield pathology hardware like Roche Ventana, Leica, and Agilent - the stuff that most institutions have. I think that really cuts down the cost of entry for a lot of labs to get into this.
Then there will probably be some low-plex fluorescence that can take advantage of standard fluorescence whole slide scanning systems, like 3DHistech and Olympus, that go up to six-plex or eight-plex. I do know some pathologists who use multiplex fluorescence LDTs today. UPenn, for example, has a really nice assay that they use for every transplant biopsy, but it does require that the pathologist be pretty savvy about what they're looking at. Not every pathologist has learned how to look at these things. I think there's going to be a big gap there, and it will take a while. Whether it goes beyond that is really hard to say. Do you really need a 20-plex? Can you not narrow that down to five markers or two slides refined? I think the practicalities of it are going to be a big part, even once there's clinical utility and a desire to pay.
Do you expect a more centralized or more decentralized commercialization pathway will be more successful, considering the infrastructure requirements?
I think the decentralized pathway in pathology is always going to be tough because it removes the pathologist. In genomics, you can put in a sample, and they can give you a big table of which genes are there, and then somebody can interpret that. But in pathology, you almost always need somebody to look at the images. No matter how successful Visiopharm is at making an algorithm that analyzes something, all you can do is say, “Mr. Pathologist, here's what we're going to suggest to you. It should be this percentage; these areas are hotspots; have a look at them.” Nobody is going to sign off on it unless they actually look at it. So, if it was centralized, somehow it has to come back to the pathologist to look at, which is harder than the genomics pathway.
If the role digital pathology plays within multiplexing is expected to stay in research in the near term because of the infrastructure requirements, what do you think the main driver of clinical digital pathology will be?
It's a tough question to answer - I tried to think about this a lot over the years. I have what might be a bit of an unpopular opinion, but I think one of the reasons digital pathology hasn't gone further is that the business model has not worked very well. [For digital pathology] … you have to buy the storage, the software, incorporate it into the hospital image management and patient management system. In other words, you need to spend more money to get to the same place. Whereas radiology was the other way around, and I think that's been a big limiting factor.
Now the hope is that there'll be some value-add that digital pathology can provide that isn't just simply replicating glass on a microscope. Maybe that's hotspot detection, maybe that's taking some of the boring things away from pathologists, like counting mitosis on a whole slide. I think that by automating some of those boring tests, you can add value by, for example, highlighting areas that they should look at, rather than having to scan through a whole slide to look for something that might be invasive. I know there's companies working on all of those parts of things, Visiopharm amongst others.
It's getting to the point now where pathologists come out of school and all they've done is looked at images on a computer, then they have to use an actual microscope in their practice. I think digital pathology is going to be common in five years, but I think it's really this value-add part that is going to put it over the hump.
Shifting gears, how important are partnerships in terms of building out the digital pathology ecosystem? Is it something you expect to continue? And are there certain types of partnerships that you see as being the most valuable or critical?
It's very important. No one company can do everything, and every company has their specialty. We've partnered with a lot of companies on the image management side, clinically, because we don't do hospital image management integration or hospital LIMS integration at all. And vice versa, you know. They're not artificial intelligence, deep learning pathology experts, and so for them to replicate that would be very hard. So, it's a natural thing.
In the multiplexing world, we're partnering with a lot of the hardware companies. They create these fabulous imaging systems that make amazing images. But after a few months, their customers are starting to get frustrated because looking at pictures isn't enough; they are needing analysis to go with that. You'll see a bunch of partnership announcements for Visiopharm coming out over the next year or two, and that’s why we have launched our new Phenoplex™ workflow for multiplex/highplex image analysis for the research market.
Over the longer term, what's the level of consolidation that you might expect? And why partnership versus buy or build?
Good question. I think in the brightfield world, we've already seen a lot of consolidation. The number of companies that sell whole slide scanners has plateaued. At first, there was big growth and then some of them got acquired or some of them died. I think for the regular digital pathology world, the scanner market isn't changing too much. The pie might be growing a bit and the shares of that pie might be shifting a bit, but it's mostly the same year over year.
I think multiplexing right now is the Wild West. Five years ago, it was just Akoya, or PerkinElmer at the time, but now I think at last count, I had 15 companies in the multiplex imaging space. It's going to be interesting to see who does well, who bows out, who gets acquired.
I know that there are some of those companies who are trying to do it all; they want to do the imaging, they want to do analysis, they want to own everything. Having been in industry for 20 years, I just have to say good luck to them, because you can't do everything. And if you do one thing really well and the other part not very well, your customers are not going to be happy, and word of that will spread. I think every company needs to focus on what it's really good at.
Are there any other topics that are top of mind for you today?
There's one other topic I think it's just coming to light now that I wanted to comment on - the role of digital pathology quality control metrics in stain management. There's been a lot written about how different labs stain differently – just look at PD-L1 for instance, everybody's using different antibodies. Even for one antibody, different people get different stain results entirely from that same antibody. But even more importantly, what we hadn't fully appreciated until we started looking into it, was just the day-to-day differences in the labs staining with their own protocol or their own systems. Particularly as pathologists are being pushed more and more into these in-between cases, the more the staining matters. For example, if it's a little bit darker today than yesterday, you're going to give a different answer than you would have.
Visiopharm has put a lot of research and energy into the analysis of this. Some labs were pretty good, but every once in a while, they'd find a big outlier. When they went back to look at it, it turned out it was because they had changed the lot of antibodies or something of that sort – there was generally some reason for why the staining went off. Once you know that your staining is off, you can do something to fix it. There were also those labs that were all over the place, and if you just give them a metric, they can ask the right questions to improve their consistency. People are discovering that stain consistency makes a bigger difference than they thought and there are ways to address it.
Do you think this lack of consistency is due to lack of awareness? Lack of perceived impact? Or is it due to necessity since it's hard to compare without digital tools? And is there a role that regulatory bodies are going to play in pushing digital tools to be used to supplement manual reads?
On the latter point, I hope so! I hope regulatory bodies push this. I do have a few CROs now that are using this as a selling point – as in, “We quality control our stain management, and don't you want the best staining?”
But I think to your first question, it's really more that the human eye is just really bad at measuring intensity. When people are doing their own visual quality control, they're just left with how good they are at assessing how dark something is and that's generally not easy. When they're given a tool that assesses not just how dark it is, but patterns, and all the little things that go into that – are the same things being stained, are we only staining the right cells, are there other subtle things in there? Until they get a tool that looks at that, they really don't know that anything's wrong.
Do you think that that is a compelling enough argument for your average pathologist sitting in a hospital lab to reconsider adopting digital pathology tools for a second look? Or do most pathologists say, “Yeah, it’s not great in some cases, but I can still do my job for 98% of the cases that I'm signing out and this really only comes into play more as we get into complex, multiplex reads”?
I think unfortunately, as much as I have wanted the digital pathology world to take off and become ubiquitous, I have to be honest with myself, and I think it's more the latter. Most pathologists spend 95% of the time looking at H&E, and for them, it's just a relative look at things, and they do it very quickly and they do it very well. Only in certain cases might they need help with that, such as maybe screening through for metastases or confirming a negative finding.
But there are more and more areas where pathologists are being asked to be quantitative - counting mitoses, counting lymphocytes, or scoring. In particular, not just scoring really positive and really negative but rather the in-between cases. I think they would benefit a lot from digital pathology with an analysis tool that at least gives them a place to start. Even if the pathologist looks at it and says it's negative, they can then pass it to the digital method. And if that turns up something, they'll take a closer look.