Design, Optimization, and Validation of Multiplex Immunofluorescence Panels

Hello, and a very warm welcome to everyone joining us for today's SelectScience webinar, titled Design, Optimization, and Validation of Multiplex Immunofluorescent Panels. My name is Charlie Carter, and I will be moderating today's presentation. I'm delighted to be joined by Michael Surace from AstraZeneca, who will be exploring the biological and technical considerations when designing a panel, describing a linear process for developing, testing, and optimizing a panel, and providing an approach for technical validation of multiplex panels that detects and addresses known risks. Please feel free to ask your questions for the Q&A session at any time during the webinar. You can submit your question at the left of your screen. Without further delay, I would like to hand over to Michael.

Thank you very much, Charlie. Hi, my name's Mike Surace, and I'm a scientist here at AstraZeneca in Gaithersburg, Maryland, where I build and validate multiplex IF panels. Today, we're going to take some time to discuss our strategy here for design, optimization, and validation of multiplex IF panels. When we first started working in this area, panel development was somewhat non-linear, and validation was ad hoc. So, we worked to linearize the process and set clear criteria for validation, including considerations for known artifacts as well as less characterized risks. What I'm going to walk you through today is our process for design optimization and validation. I'm going to break this up into five segments. I'll touch briefly on available multiplexing technologies, just to give some context to the platform that we use now. Some of the controls and artifacts are common to all multiplex imaging platforms, and some are not. Panel design, I'll describe considerations and the appropriate parties to consult at the outset. In development and optimization, this is where most of the work will be, but there are approaches to managing this stage. Panel testing, I've separated out from validation, even though they're related and overlap somewhat. Panel testing, or what we sometimes call pressure testing, is a set of controls that we run as a final test of optimization to check for the presence or absence of known artifacts that can complicate multiplex IF staining and imaging. And finally, panel validation is a planned experiment validating the reproducibility of the multiplex IF at a single site or multiple sites, as well as comparing the multiplex to chromogenic IHC, which is currently the gold standard for clinical-grade diagnostic staining.

But I want to take just one slide to go over the use case. For anyone unfamiliar with what's driving a lot of multiplex IF development, the field of immune-oncology is relatively new and concerned principally with understanding how the human adaptive immune response reacts to cancerous lesions. In many cases, the immune response can eliminate tumors, but in other cases, it permits tumor growth. Describing how this happens requires a comprehensive understanding of the immune response in and around, as well as at sites distant from the tumor. We can measure this with flow cytometry, gene expression profiling, proteomics, monoplex IHC, metabolomics, et cetera, but multiplex IF combines two critical elements, which are multi-marker phenotypes and spatial resolution from the same sample. This makes multiplex IF useful for mechanism of action studies, as well as development of critical testing for immuno-oncology drugs, including checkpoint blockade, targeted therapies, and cellular therapies.

And just a bit about multiplex technologies. Multiplex with chromogenic substrates on the left side of the slide is possible and commonly used to detect three and upwards of maybe four or five targets, but it's limited by the translucency of the chromogens, which have to be separated with software following imaging and can't really be completely separated during the imaging process. Also, overlapping targets present challenges because chromogenic is conducted in bright field, and substrates are fundamentally light absorbing. Sequential chromogenic detection using ethanol labile substrates like AEC is conducted well in several labs, but it's labor intensive and requires multiple rounds of staining and imaging. This approach allows you to look at effectively unlimited number of targets in the same tissue. So, it avoids many of the challenges of staining and imaging all at the same time, but it's not quite ready for prime time. On the right side of the slide, we have CyTOF and MIBI, which are two devices which both exploit mass spec to separate signals as opposed to visible light. Briefly, antibodies are labeled with rare earth metals, and 30 to 50 can be applied to a piece of tissue in a single staining step. The devices use either a laser or an ion beam to ablate or remove tags from the tissue in a raster fashion, measuring all species of metal tags representing antibodies at the same time using mass spec. These devices both have distinct advantages, but the panels can be more challenging to develop and validate, and the throughput is low compared to other multiplex staining technologies.

In the middle column, you can see there's a proliferation of technologies in the midplex and fluorescence. Akoya, which used to be owned by Perk and Elmer, or part of it, offers the Opal Vectra system, which we employ here. This is based on Tyramide signal amplification, TSA, as is the Ventana 5-Plex immune fluorescent panel. AltaView, Codex, and the Comet, which are not yet available, all use TNA barcodes to allow for the imaging of a small number of targets. usually three to four, in a batch fashion, making it possible to image also an effectively unlimited number of targets in a sequential manner. The advantage here over sequential chromogens is that each of these leverages an automated platform of some kind for the staining. They require fewer steps than chromogenic sequential staining. And they use fluorescence, which is more suitable for quantitative work. The Multiomyx Celldive and Miltenyi MACSima, which is also, I believe, not yet on the market, are both sequential staining and imaging devices which use antibodies with direct conjugated floors. And finally, CellIDX uses haptic labels on the primaries to avoid species cross-reactivity with secondaries, which allows for the staining all at once with multiple antibodies of the same isotype.

So here, we leverage the Opal Vector system quite a bit for several reasons. First is that TSA allows for sequential staining of multiple overlapping targets with spectrally distinct fluorophores because TSA, once activated, is covalently bound to proteins, allowing for the removal of the primary and secondary antibodies after each staining round. The staining can be conducted on the BOND RX auto stainer, which allows for sequential staining of multiple targets with the same or different hosts of antibodies. This improves not only the reproducibility, but also the throughput, which is very important for us, because we use this to stain a large number of clinical trial samples. The Vector Polaris imaging platform allows for the spectral separation of up to eight protein targets, plus NAPI and autofluorescence removal, which totals 10 spectrally distinct channels. And this system actually strikes a really good balance between throughput, target count, and reproducibility that works for our needs in research and clinical trials.

So, panel design and development. At the outset, it's important to nail down your study goals and reportable metrics. MidPlex IF platforms are not the best place to go fishing. You want to identify what's your top line question is and what are the components that you need to answer that question. For example, what phenotypes do you expect and what spatial features are you interested in? This also is important to consider in marker selection. Be sure that you designate positive and negative controls early on. Gene Investigator, ATCC, the Human Protein Atlas can all be good sources of information. Try to stick to controls that express endogenous levels of a target, and don't rely on overexpression models. So first, we developed monoplex chromogenic solutions for each target, testing multiple clones and primary antibody concentrations and other conditions. While still in chromogenic development, we determined the order of staining. Since we stained one target at a time in sequence, we had to determine which targets are robust enough to be stained last and which ones must be stained early on. After the order of staining is established, we determine fluorophore selection. Here we maintain the same antigen retrieval, blocking, and primary conditions and test different fluorophores with each target. Once we've selected the fluorophores, we determine the proper dilution of each to match the intensity of each signal across all targets and all channels. This is extremely important, and it's one of the major levers that you're going to pull during development to avoid both spectral bleed and TSA blocking artifacts, which I'll describe later.

So here, I want to mention that we should assess not only signal intensity, but signal-to-noise ratio as well. In the final multiplex, we're going to use these measures to rebalance the signals in an appropriate and useful way. So, it's important to include, if at all possible, a translational scientist or another lead scientist in your team. This is the individual who can provide the study goals, and the scientist might be supplemented by an immunologist. In fact, quite normally, we have an immunologist as a consultant on all of our panel design here. Pathology input is indispensable when interpreting staining patterns and signal distribution. Pathologists are also well acquainted with common IHC and IF artifacts and can help in troubleshooting and also help determine appropriate reportable metrics. The multiplex scientist is you. That's the individual developing assay. And you stand in the middle of a group which also includes the image analyst. You know, the image analyst may also be you, but it's important to include this individual early on in the process in order to get feedback on what will work for the image analysis purposes going forward. This is a sketch of a panel-building exercise. We have a clear hypothesis and a justification for each target. You can see where we've denoted expected cell types and colocalizations, as well as interesting distances which we'd like to report. At the bottom, we have a list of reportable metrics. And this is the product of perhaps a two to three-hour meeting, but it's really the bare minimum that you should have before developing a panel. Even if you're working alone or in a small team, this is a useful exercise.

So, step one of the development of chromogenic solutions has been extensively described elsewhere, but I want to point out a few things about this stage. You want to make sure that you develop your solutions so that they're able to be easily ported to or easily developed on the automated staining platform. In this case, it's the BOND RX. You have two options for antigen retrieval, ER1 and ER2. ER1 is this citrate-based PH6 buffer, and ER2 is PH9 and it's EVTA based. ER2 is generally more stringent, more aggressive antigen retrieval. But during this stage you want to develop all of the antigen retrieval, dewaxing, blocking, protein blocking, peroxidase blocking, primary antibody, clones, diluent and exposure time to the primary antibody. Don't worry too much about the secondary, the DAB, or the hematoxylin, because those are all going to go away when you move over to monoplex immunofluorescent development. After the antigen retrieval, the blocking, the primary concentration, the clone, and the diluent are all optimized for signal and signal to some of these ratios. Staining order can be addressed. This is still in chromogenic IHC. So, some targets are going to suffer from over-retrieval, and some will become stronger. Targets which require this lighter citrate-based retrieval, such as FOXP3 or CD68, should be placed in early positions or the signal will diminish. Targets which require more aggressive AR should be placed later. If you don't have any information available, then you may simply choose to test each target in each position. Antibody retrieval is achieved by 20-minute incubation in ER1 at 95 degrees centigrade. So, include the appropriate number of antigen retrieval steps to test each marker at each position if you're going to take this route. So, this is the kind of experiment that is quite straightforward on the bond to complete, but it's labor intensive when conducted manually. Assessment at this stage is qualitative, but the goal is to have all targets as close to the monoplex solution as possible. After primary conditions and the staining order are established, change from chromogenic to fluorescent monoplex assay development. At this point, everything from the secondary antibody on will change, but everything before the secondary antibody should remain unchanged. Why is this? Because we've established that the antigen retrieval, blocking, and primary conditions in the standing order are suitable to achieve both sensitivity and specificity for the specific clones that you've chosen on your epitopes in the tissue of interest.

There are two parameters to optimize in fluorescent monoplex. The first is fluorophore selection, and the second is fluorophore concentration. We use both of these parameters to control our positive signal intensity, maximize our signal-to-noise ratios, and most importantly, to balance the signals across all channels. There are eight TSA-linked floors in the OPAL system, and depending on your imaging and standing protocols, you may have six, seven, or eight in a panel. In this example, we're using six. The simplest approach is to test each floor with each target, but this may be excessive due to practical considerations. Some floors are brighter, and some are dimmer. Highly expressed targets and antibodies which yield a strong signal should be paired with weak or dimmer fluorophores and vice versa. So, you may not have to, in practice, run all 36 of these potential combinations. When it comes to assessing the intensity of each signal in monoplex fluorescence, we’re able to do this quantitatively as opposed to qualitatively. This is due to the fact that spectral and mixing and image analysis software Inform allows us to determine the intensity of each signal on a pixel-by-pixel and channel-by-channel basis. So raw counts as well as normalized counts can be assessed. Normalized counts are calibrated for gain, binning, bit depth, and exposure time, yielding a value which can be compared across channels and across slides.

So, in the image, you can see for PDL1, we have a normalized counts value of 16.6. Even though some of the floors are brighter in an absolute manner as compared to other floors, the way that normalized counts are calculated, you can compare from one channel to the next. The target range for normalized counts is between 5 and 30, normalized counts in the positive signal. And we want to achieve a signal-to-noise ratio of 10 to 1 or greater. So, the positive signal is measured in a positive cell, and the negative signal is measured in an area with no cells or an area that doesn't-- or a cell that doesn't express the marker that you're assessing. I should also note that some fluorophores, especially one that's not pictured here, called Opal 780, is a far-red floor. It's a long wavelength and a low energy wavelength. So, this is routinely dimmer. So, you can adjust your expectations somewhat for the signal, as long as the signal-to-noise ratio is acceptable. And artifacts like spectral bleed are not apparent. If normalized counts are too high or too low, then we can adjust the intensity of the signal using TSA concentration, as indicated here by dilution factor. Here, you can see that over selection of non-small cell lung carcinoma and tonsil samples, we assessed positive signal intensity of various concentrations of each fluorophore. And we're able to use this information to select final concentrations again. We prefer to change the TSA concentration first to balance the signal intensity and leave the primary antibody conditions alone, because these are the conditions under which we have ideal binding of the primary antibodies. If the TSA concentration is too high, I don't like to dilute it anything less than 1 to 50, or too low, I prefer not to exceed 1 to 1,000, then the next lever to pull is the secondary antibody using a more or less amplifying secondary antibody. And if that can't balance your signals, then you can pivot to changing some primary conditions.

So, after the primary conditions, the standing order, the fluorophore selection, and the fluorophore dilution are determined in that order, then all the targets can be combined into the initial version of your multiplex. Any changes in signal from the monoplex should be addressed preferentially by changing the TSA fluorophore concentration. Normally, when you put together a multiplex, you may see a slight decrease in some channels of intensity in your positive signal. So, this can be addressed by increasing the TSA concentration. There's two specific artifacts that I want to describe and tell you how to detect and address them. TSA blocking occurs when an earlier stained target prevents detection of a co-localized target, which is stained later. In this example, PD-1 is expressed on CD3 positive T cells in a tertiary lymphoid structure in non-small cell lung carcinoma. PD-1 is stained first on the right, but on the left, we omit the PD-1 primary antibody. You can see that the CD3 on the T cells is not detected in the area of high PD-1 expression. TSA binds to tyrosine residues, so this may be due either to full occupancy of all of the available tyrosine residues in the area where those two antibodies are binding, or by occlusion of an epitope, in this case for the CD3 antibody containing a tyrosine. This can be addressed by decreasing the blocking TSA concentration, or by changing the staining order. But it must be detected first, and this is something that is not trivial to identify if you don't know where to look for it. So that's where pressure testing becomes useful.

The other common artifact is spectral bleed. In this case, the Opal 540 signal, which is CD8, is bleeding into the Opal 570 channel, producing a dim, characteristically CD8 pattern in the channel where we only expect nuclear staining. This is normally due to a signal of a shorter wavelength and higher energy being too bright. And it's addressed by reducing the concentration of the TSA in the contaminating channel. which can be a little bit easier to detect. But the best practice in order to-- when you have a membranous and a nuclear target in adjacent channels, it's easy to detect. But this is one good reason not to put two co-localizing targets. So, we wouldn't put CD8 and then PD1 in spectrally adjacent channels because that can make it more challenging to detect spectral bleed.

So, a simple method for detecting these and other issues before going into a complete validation is to perform a set of what we call drop controls. These are similar to FMO controls in flow. In the full multiplex, all antibodies and fluorophores are included. In each sequential slide, the entire multiplex is stained, omitting only one primary antibody. This is going to tell us two things when we analyze these. First, we can tell if we have spectral bleed because the drop channel should have no cells. If there are cells, then we know that the signal comes from some source other than our primary antibody binding to the appropriate epitope. Secondly, we can tell if we have any TSA blocking by counting cells in drop controls as compared to the full multiplex.

First, I'm going to show you what the slides look like. So, this is the full multiplex. You can see a TLS on the right and tumor epithelium on the left, so you know we're near invasive margin. The TLS contains proliferating cells, some of them clearly high in PD-1 and some PD-L1 positive macrophages. Also, you can see some normal lung epithelium staining positive for CK. And as we look to the left, you can see a lot of cyanotoxic CD8 positive T cells in the stroma and PD-L1 positive CK positive tumor epithelium. In the PD-L1 drop, we lose that green signal on the tumor epithelium and on the macrophages. In the CD8 drop, we lose those stromal cytotoxic T cells. In the Ki-67 drop, we lose those proliferating yellow nuclei. And in the CD68 drop, we lose our macrophages. Decay drop, we lose all of the epithelium. And in the PD-1 drop, we lose those magenta PD-1 positive cells, especially in the tertiary lymphoid structure. It's apparent. So, viewing them this way is straightforward, but it's not easy to separate the signals for the human eye.

So, one other way to view these same images I've just shown you is not only by separating the slides but separating the channels. So, what we've done here is each row is one of the slides I just showed, and each column is one of the different channels. So, in blue, you can see the drop channels in the slide in which that antibody was dropped. So, these should all be clean and have no signal. These are the channels you can use to detect spectral bleed. In red, we have the CD8 channel in the full multiplex and in the PD-1 drop. PD-1 was stained before CD8 and does co-localize in some cells, so there's potential for blocking here. We can take the two slides and count cells using image analysis software in matched regions. And if we have the same number of CD8 positive cells in both slides, then that means it's very unlikely that PD1 is blocking CD8 signal. Now, if you do generate data in the pressure test that indicates you've got one of these artifacts, then you can address it before performing the validation. But here you can see the two red bars are essentially the same number of cells, indicating that we have the same number of CD8 positive cells in the full multiplex and in the PD-1 drop. So it's unlikely that PD-1 is blocking CD8, and in the CD8 drop, you can see that there are no CD8 cells detected, which means that all of the signal that we see for CD8-positive cells is dependent on the presence of that anti-CD8 antibody.

So, we've talked about the approach for the team approach for design and optimization of multiplex IF, the streamlined approach for optimization, so chromogenic, monoplex, then order of staining, floor combinations, and floor dilution, and then pressure testing the panel once you've got it built, looking for TSA blocking and spectral bleed using drop controls. So next, once you have a panel that passes the pressure test, you can move on to validation. So, we're not performing a validation of any of the antibodies here. This validation is focused on the technical aspects of the panel. You see three sections here. The first is intermodal consistency. That measures each channel of the multiplex against the chromogenic monoplex. We want to be sure that our multiplex channels report the same information as the monoplex chromogenic gold standard. Then we assess intra-user reproducibility. So that measures how reproducible the assay is on a single device at a single site with the same scientist. And then I call it inter-user. This could be inter-user, inter-site, inter-device reproducibility. This can measure one user against another, one bond against another, one site against another. The specific comparisons are going to be determined by the needs of your particular validation.

And this is a general slide plan for a single case. So, in the chromogenic monoplex comparison, slides are leveled so that each chromogenic monoplex can be compared to an adjacent multiplex for intermodal consistency. This is levels 1 through 9. We conducted this image analysis with Definiens. So, slides are compared by being-- they're co-registered, which means they are-- the two sections are aligned, the two sequential sections are aligned. Then a matched grid is superimposed on the slides to be compared. And this produces a large number of small fields in which all cells can be counted and compared from side to side. As compared to pooled whole slide analysis, this approach has the advantage of allowing us to calculate correlation coefficients with a relatively few number of cases. Ideally, we prefer 30 or more cases. But in this case, we made the indicated comparisons between the stain slides for five non-small cell lung carcinoma cases.

So here, we've calculated both Pearson's and Spearman's correlation coefficients, and we're showing the full slide and tile-based comparisons. So, what these numbers indicate is how well does the appropriate channel in my multiplex IF compare to the monoplex chromogenic solution. So, we're very happy with these values, although CD68 does suffer somewhat as compared to the other markers, likely due to imperfect cell segmentation. And this ends up being a common theme for us. If we compare this to a similar validation with some shared markers, which was published in 2017, we observed a marked improvement in performance, which is likely creditable in part to the advantages of automated staining. Levels 11, 12, and 13 are for days 1, 2, and 3 for intra-user reproducibility. I say days here, but really, they just need to be different staining runs. Here we have very high reproducibility across the board. And when we compare this to the validation connected with manual staining, we again see a marked improvement with automated staining as opposed to manual. Level 14 can be stained by a second scientist or at a second site and compared to level 13. We did this with two scientists here in Gaithersburg, and we produced, well, we produced acceptable coefficients. What we did observe were lower values again for CD68. I want to point out that the Lin's correlation coefficients here in this comparison, in addition to Pearson's and Spearman's, and I'm going to talk about precision and accuracy for just a moment. So, we calculated Lin's in this case, because Lin's accounts both for precision and accuracy. Pearson's and Spearman's can yield a perfect correlation coefficient, even if one rater or data set is systematically over-counting or under-counting. Lin's correlation coefficient accounts for precision by including a penalty for departure from the x equals y line. And in fact, intraclass correlation coefficient does this as well, but allows for more than two raters, so we're moving in this direction with ongoing and future validations of using intraclass correlation coefficients where possible, because they're more stringent.

Summary, we went over panel design and the team approach, panel development from the monoplex chromogenic standing order and fluorophores, pressure testing, panel validation, and correlation coefficients, and considerations around precision and accuracy. So, I want to acknowledge my colleagues at AstraZeneca, Definiens, Koya, and at Leica. And I also want to make everybody aware of two opportunities if you're interested in learning more. I lead a user group of lab-level multiplex IF scientists. We have a monthly call on which we discuss challenges and solutions to delivering multiplex IF in academic, government, and industry settings. And I also co-direct with my pathologist colleague, Jaime Rodriguez-Canales, a symposium and workshop about multiplex IF in immune oncology. We're planning the meeting for the third year in a row in Germantown, Maryland, just outside DC in September. Please take a look at the website and consider joining us there this year. And finally, I also want to point to some R scripts that the very talented Kent Johnson at Akoya has developed, which are useful for assessing some of the artifacts and signal intensities that I mentioned in panel optimization. And with that, I thank you very much for your time, and I'll be happy to take any questions.

Thank you, Michael, for that interesting presentation. Now it is time for our live Q&A session. And to kick off the Q&A, we have the question, how does the technology you describe compare to other multiplex immunofluorescent technologies?

Hi, certainly. Thanks a lot. So, I touched on this a bit during, I think, the intro. And I think the way that I would put the technology that we're using here for the design and optimization and validation of these protocols and panels at AstraZeneca sits squarely in the middle. So, we're balancing throughput and as well as plexing and getting those two types of data, the spatial and the phenotypic in the same data set. So, for us, there are a couple of important things about using this technology as opposed to some of the other available multiplex IF technologies. One is that we do, in order to validate this against the chromogenic, which is our gold standard, we like to be able to use unmodified antibodies. Now, this is not a real problem to use a bar-coded or direct conjugated antibody. But for us and for the ability to rapidly change panels, it really helps to use unmodified antibodies. It also helps us get a little bit closer to an apples-to-apples comparison. when we're doing the validation across modalities. Yeah, now please, do we have any other questions here?

Yes, yeah, we do. So, someone has also asked, is this type of staining suitable for the clinic?

Oh, sure. Yeah, so at AZ, we use this for clinical trials. And I know that at multiple academic centers, this is used in the clinic. And the only problem is that this can't be used for diagnostic purposes or primary endpoints in clinical trials, because the entire pipeline for staining, imaging, and image analysis isn't to the stage of development where it can be validated for that level of an assay. But it can certainly be used for discovery and informational purposes in the clinic and in clinical trials. And it is routinely, yes.

Lovely. Thank you. So, we do have some more questions on that. Someone has asked, how do you go about analyzing the data that comes from the images you get.

Oh, right. So, data analysis is really the bottleneck in any multiplex endeavor. So, this webinar was all about how to develop the panels for standing. and about how to validate those. But once you have a validated panel and you know that you're staining consistently, then the real fun is in determining how you're going to do the analysis. So, we break the analysis down really, really roughly, I can describe it as three tiers. The first tier is really basic data. That's cell phenotypes in areas of tissue. The second tier would be where you can answer sort of hypothesis-driven questions involving specific distances between specific phenotypes or thresholds for populations. And the third tier, which is the most interesting, is, of course, when we get into the AI space for data analysis. And that's a lot of what I spend my time thinking about. So this type of data contains multitudes, of course, and we're interested in understanding-- not necessarily going on a fishing expedition, but understanding a little bit more about what the data can tell us, as opposed to just the questions that we can come up with to ask. So, I think at the end of the-- on the final slide, I put a link to the workshop that we're having in Germantown, Maryland, in September. And the topic of the workshop is going to be based around how to validate analysis methods for this type of data. Really, everything that involves multiple phenotypes and spatial information. So, the analysis methods will apply equally to things like hyperplexing with the Hyperion and MIDI, and Low plex multiplex chromogenic.

Excellent. Thank you. So, someone has asked about the timeline that you have. So about how long is the timeline for testing a new panel?

So yeah, this really depends on the level of validation that you want to do. I think that a technician in the lab who's familiar with the technology can develop a new panel that's sort of fit for use. for lab investigations and research in a matter of one to two months if you have good monoplex chromogenic solutions already in place. The development of a fully validated panel, such as the one I've described here, especially if you don't have good chromogenic solutions, I would start with my estimate, around six months, and it can go up or down from there based on how challenging the targets are, how many times you have to re-optimize the staining order or the intensities of each of the floors. And the actual validation, the staining takes about a month, and the analysis can take one to two months, depending on who's doing the analysis and how many parameters you want to report. So, I realized that was not a very direct answer. For our research panel, one to two months. For a panel that you want to have fully validated, I'd say somewhere around six months is a reasonable timeframe.

Great, thank you. So, sort of rolling on from that, someone has asked how long can you use TSA fluorophores for and how can storage conditions and temperature influence the specificity or quality of these?

Wow, okay, so I see the name next to this is Murph. Murph, it sounds like you may have been listening in on a call that we had yesterday with the Time Group. We were talking about stability of reagents, so also stability of epitopes. But on the reagent side, there's kind of two things to consider. The TSA comes lyophilized, and then it's resuspended in DMSO. So lyophilized, there is, I think, there's something like a year of storage at minus 20 or minus 30. Although lyophilized, I think in practice, we can go longer than that. In terms of after they're resuspended, the floors are stable for a number of months, but we haven't tested long term storage conditions of resuspended floors. There may be more data out there on this, but I don't have any directly. In terms of the second aspect of this is floor stability, the TSA floor stability on the tissue, and I'm not sure if this is what you meant, but I'll touch on this briefly. We found that the stained tissues, once cover slipped, the storage is very, the ability to reimage them multiple times, they're pretty robust once on the tissue. So, I’ve re-imaged multiple times for multiple months. So, some of the considerations with floor stability with older floor sets don't apply, or they're less of a consideration with this particular generation of floors and binding chemistry. I hope that was a useful answer.

Lovely. Thank you, Michael. So, we've got time for one more question. And this one is, how many samples do you usually test to cover the variability in expression levels between patients?

Sure. So, there's two ways to think about this also. One is the number of samples that you have to have in the validation. This is normally calculated empirically from an observed smaller set. So, I think I mentioned we normally start with five patients, five cases when we're doing a validation. Really, I think it would be ideal to do, I think, so I have a colleague at Definiens who is very, very skilled with stats. And he suggested somewhere upwards of 30 based on the degree of variability that we were observing. However, you're not always going to be able to catch everything. So, five is what we normally do for validation. 30 is what we ideally should do in terms of number of patients. that we start to feel comfortable with once we've gone through staining for actual analysis. It's hard to get a sense of that on multiplex. And I think the reason is that the analysis is never really complete until you've gone through an entire study. We have between 50 and 200 patients. on different studies that we've stained, but I don't have an answer for the number of patients that you need to go through to get a sense of the variability. It's not really different than the number of patients that you would need to go through to get a sense of the variability for each of the targets of interest. I’m going to try to look through some of my data and come up with a better answer for you. Looks like I can see who asked these questions. So, I'll either reach out directly to-- I think that was Elena de Miguel, or I'll go ahead and send out an answer that's broadly accessible to the attendees. I'm not sure what the mechanism for that will be, but yeah, let me circle back to that question and try to give you a better answer.

Yes, of course. We can make sure that once you've got an answer ready, we can make sure that everyone gets to see that, especially Elena, thank you for that question. So that's all we have time for today. So, thank you again, Michael, for your informative presentation. And thanks to everyone who joined us online today. If you do have any other questions, please feel free to e-mail me at editor@selectscience.net, and I will follow these up with Michael. And remember, you can also download a certificate of attendance in the related resources tab at the bottom left of your screen. And if you had any difficulty connecting and viewing today's webinar, or if you would like to listen again, it will become available to watch on demand in a few days' time. So goodbye and thank you once again for joining us.
Thank you very much, everyone.