Identifying Optimal Methods for Clinical Quantitative Flow Cytometry Hyatt Regency Hotel 1800 Presidents Street Reston, Virginia April 10, 2003 P R O C E E D I N G S DR. MARTI: Good morning. Good morning, everyone. Welcome to the second day of this conference to try and define what the problems are and some appropriate solutions for clinical quantitative flow cytometry or quantitative flow cytometry in a clinical setting. We thought that in order to start the session today, we would start out with compensation, have part of the discussion about compensation. DR. VOGT: As opposed to decompensation? DR. MARTI: As opposed to decompensation. Recently we had a couple of experiments in our lab, one involving the CE79D. As you came acutely aware that this particular reagent -- I don't remember whose company it was or was. I don't remember what the fluorochrome is. We realized that the nonspecific background was pretty high, that the t-cells were staining more than we were used to. And it reminded us that we hadn't titered that reagent. When we titered it, it behaved very nicely. The second thing in this series of experiments, as we were moving from four-color to six-color, we carried out an experiment on the LSR2. We dutifully were doing a comparison between manual compensation and automated compensation. We constructed for this experiment all of our single tube controls using CD3: CD3 FITC, CD3 PE, CD3 PE 5.5, APC, the six reagents that we were using. And when we did it, we ran those. And we were comparing the six-color to the four-color. We essentially had such terrible overcompensation on the FITC PE that we actually didn't know whether we had made a mistake in the sense that the fluorescence intensity of our controls was the same or lower than the test panel or that conceivably that it might be above or failure in the software. That is an easily testable problem, and we are looking into it hopefully this week. However, in the process of this, it made me think once again about compensation controls. When we were discussing this with Steve Perfetto in Mario Roederer's lab, he mentioned to us that he had received some microbeads from BD. I don't know what division, whether it was the West Coast or Europe or somewhere in this BD operation. We tracked down these microbeads that contained I think goat anti-mouse kappa. And, as you know, most murine monoclonals are kappa, kappa Verity. And we thought, "Gosh, that's a wonderful idea to make a compensation control." And we thought that, you know, we have had a little experience with antibody capture beads. So the first word out of my mouth was, "Well, how long did you incubate?" Oh, 15-20 minutes. And what else did you do? Did you keep buying them? Well, no. We fixed them. And, you know, all the time we used the antibody capture beads, I don't think it ever occurred to us to fix them. We looked at them over an hour, four hours, one day, one week. So we got a sample of these microbeads and stained them. When we stained the first set of these microbeads, we were kind of surprised to see that with the FITC fluorochrome PE and PE-Cy7 that it wasn't quite the homogeneous peak that we thought we were going to get. We didn't know quite what this represented. I mean, is the antibody coming off of the beads? Apparently you have to be very careful with some of these tandem conjugates. They actually form end-to-end dimers. In fact, you know, there is a very ancient procedure in immunology where when you take your reagent out of the refrigerator, you are supposed to take an aliquot of it and walk over to the microcentrifuge and do three minutes or something. In this area, these colors, when you get up above four or six, I think we are going to have to really pay attention to that. DR. LENKEI: The picture is still very bad. It's not a clean picture. DR. MARTI: I couldn't agree with you more. So we got a fresh sample. That is four or six of the eight, four of the six. I overlapped it this morning because we just did this experiment. Now I was similarly impressed with these antibody-coated beads because the first thing was they looked awfully bright to me, which suggested that that might guarantee more often that the brighter your antibody in your control, compensation control, the better chance you had of not having it be the same level of fluorescence intensity or, God forbid, less. I think if it's less and you put it through the automated algorithms, and you are going to be in big trouble, big trouble. The second thing -- and I haven't had a chance to look at the statistics. It sure seems like an awfully tight CV to me for an antibody-captured bead. The difference between this sample and the previous sample was two months in terms of sitting in the refrigerator. So I suspect that there may be some shelf life in and out of the refrigerator, et cetera. DR. QUINTANA: Can you do it the same way, this one and the previous one? DR. MARTI: Well, you tell me. I mean -- DR. MUIRHEAD: There is or isn't. DR. MARTI: You can take your pick. I mean, we thought that was the most representative of what was there. I think there is a lot more aggregation in the over-sample compared to this. DR. SCHWARTZ: Actually not. The bigger peak is a doublet peak. DR. MARTI: Well, nice that you should point that out, but walk your eye up from the lower one to the top one. That doublet has disappeared. And I would say that the side scatter signal has increased slightly. What's the only difference between these three or, actually, between these two and that one is that this required a different laser excitation. So I think some of this scatter difference is wavelength-dependent or it's a different optical pathway of the second laser through the optics. Howard? DR. SHAPIRO: What's that little peak below the main peak in the PECy7? That's what I'm saying. The problem is basically I think what you are looking at there is that the gating is not the same for that because you can see if you look, there is a lower side scatter peak to the left of your gating region in there. And my impression would be that those would be the singlets and that you are basically discriminating orientations of doublets. DR. MARTI: You think that all represents doublets? DR. SHAPIRO: Yes, because what do you have, sperm beads, in there? I mean, how do you get -- DR. SCHWARTZ: Have you looked at it in the microscope? DR. MARTI: Not yet. Well, not so much shame on me. It's that we didn't have time to set up to do the -- I do want to photograph these because we've always photomicrographed beads that we've used, and it's always very enlightening to see poikilocytosis and variation of hypochromia. DR. SHAPIRO: If it will break streptococcus and staphylococcus, we may have more than a vested interest in staying together of the beads. DR. MARTI: We just took these out of the bead bottle. You drip them out into the tube and "stain them like cells." I only brought these. We have hardly had time to think about this experiment. But I thought -- and perhaps you will correct me -- that it might be a good way to introduce the topic of compensation. DR. SCHWARTZ: How are you going to compensate with one population? DR. MARTI: Well, that's a good question. The fact is the Purists chided us -- and I haven't shown it here to you chided us for even thinking about mixing these individual ones together as a task at the end. DR. SCHWARTZ: You cannot. It's true. DR. MARTI: Why? DR. SCHWARTZ: Because you have overlaps in things, but you should have two populations if intensities could be able to DR. PURVIS: You've got to know what the background of the bead is. DR. SCHWARTZ: No, not background. A second population of the same label fluorochrome. And if you have two of them that look like they're compensated -- DR. HOFFMAN: This is not in the same tube for this software. DR. SCHWARTZ: It doesn't have to be, but it won't hurt it because there is no mixing of fluorochromes. There are just two separate populations at different intensities. DR. SHAPIRO: If you actually run two or more of those populations, you could basically do a regression line so you get a much more accurate measure of the contribution. You basically need to define a single number, but you probably can define it somewhat more precisely if you have two intensity levels or more intensity levels. DR. SCHWARTZ: No, that is not the way this -- DR. SHAPIRO: Yes, I understand that, but how do you solve the problem overall? I mean, the software that we were playing with years ago to do this, we actually did regression lines and -- DR. HOFFMAN: This doesn't improve the matrix, the actual compensation. Basically you measure the spillover of every fluorochrome or DR. SCHWARTZ: Right. But the space from one population is more comfortable if it's on two or more, I would say. DR. HOFFMAN: Well, it takes software to do that. It's a measure of each reference of each fluorochrome with all of the other fluorochromes in the map. DR. MARTI: I don't think it's -- DR. SHAPIRO: Actually, for purposes of compensation, it doesn't matter what you deem are doublets because what you really want to see there is to what extent the fluorescence from the individual is bleeding to the other. And, yes, if you want a single value, you probably are better off having your compensation sample as bright as you can. DR. SCHWARTZ: As a matter of fact, put a gate around both of them. You do have your two populations. You can solve it. DR. MARTI: So, again, I brought this map not to be definitive or anything. It's just I grew used to having a mixture of MESM, FITC, PE beads. And I like the opportunity to do compensation across the whole scale. So it wouldn't be impossible to use this system, but I do think, as Bob pointed out, that the algorithm requires or works with just a single peak. And with that peak being brighter than your test reagent, you should be okay. Now, I understand there are further discussions about which type of algorithm is being used. But obviously since we can't discriminate singlets or doublets, we're not ready to look at differences between algorithms. One would hope if that were necessary -- and, actually, it has become necessary because, as I said, our first experiment was so disappointing nothing turned out right except the FITC PE. So, anyway, what kind of controls if you were running your lab for compensation would you use or recommend? If you were doing a CLIA inspection and you walked in, what would you want to see in the SOP? DR. HOUTZ: Biologicals and set of beads. DR. MARTI: Biologicals? DR. HOUTZ: Biologicals and set of beads, yes. The spectral difference between beads themselves is pretty low. DR. QUINTANA: We have a fluorescence issue also with the FITC. DR. HOUTZ: A lot of fluorescence differences. DR. QUINTANA: There is a little bit of difference on the FITC, not on the longer wavelengths but on the FITCs, you see some difference between the beads themselves. It depends on the prep method, too. DR. HOFFMAN: I think you should be able to set up the compensation with beads as well as a test method using the full matrix. But what is really important is that, especially for any tandem fluorescence, it be the same lot of the reagent that you're going to be using in the sample. Especially in the PE, I am not sure how it could affect the FITC too much, if you compensated with a different antibody or a different lot of antibody for any of the tandems, PE tandems, that could affect your spillover matrix in every -- DR. MARTI: Every dimension. DR. QUINTANA: But we are looking at the issue with the tandems because there is a type. It depends on the anti-bacterium specter when you prepare the tandem dye in terms of the bleed-over. You have to have a very tight specification. And we did a study looking at full matrix comp looking at different tandem degradations. And we find still the full matrix system we have been using gives us a little wiggle room that it's about a five percent difference in the bleed-over that really has very little effect on the compensation at all in terms of moving the populations. I agree with Bob. You can use the exact dye lot. It's the same. What happens is over time, some of the tandem dye lots do degrade from time zero to, let's say, eight or nine months. So it's one of the things that you have to stay on top of, especially even mixing manufacturers' tandem dyes. You really can't because there are different specifications. DR. MARTI: I think the bottom line is -- and I think you folks said it more eloquently than I probably will -- whatever you stain yourself with and whether you use cells as a compensation control or bead, it has to be stained not only with the same antibody but with the same antibody, particularly tandem conjugate. If you have the same lot of monoclonal antibody but the conjugate was prepared at two different times and, therefore, two different sublots, we got burned in a lot of these things in a big hurry. It didn't take long. It just jumped out at us and kind of beat us up. DR. FISCHER: So, Bob, I have a question for you. You say you prefer beads because of the automated software. DR. HOFFMAN: I didn't say I preferred beads, but -- DR. FISCHER: Well, that's what I heard. And so I guess that's what I -- DR. MARTI: No. He said there would be no reason why beads shouldn't work. DR. HOFFMAN: And they do. The software works about the same as -- DR. FISCHER: My question is, how are we in the clinical labs running the new software with the compensation? DR. HOFFMAN: They're doing one or two-color. And they have the option of using the automated, fully automated. I sure wouldn't do it. Doing two-color compensation actually is pretty easy. DR. FISCHER: What's the software that does that automatically? DR. HOUTZ: Fax Comp. DR. FISCHER: Well, Fax Comp works on calibrated beads. My understanding is it doesn't work on NESF beads. DR. LENKEI: No. In the preliminary means, still there are some variations among colorblind beads. DR. FISCHER: It's close, but it's still not best. From everything you just said, it's got to be the right dioxin. It's got to be the right things for you to -- DR. SCHWARTZ: It has to be the right intensities for their compensation. DR. FISCHER: Right. That's what I mean. So -- DR. HOFFMAN: Less right, correct intensities. But with calibrated beads, they don't have to be exactly the same intensity as antibody clonal beads. We tried putting the correction factor in there that will get you close. And that actually is I think the right bead. DR. FISCHER: So to come back to the question of who in here will have the automated software, I have seen them on the SR2. It looks really nice, but my calendar doesn't run that. So we have to go back to the question Jerry posed at the beginning of how do we do our compensations when we're doing them essentially by eye. You're still adjusting it with your eye and looking at the numbers a lot of times DR. MARTI: Randy, you can collect uncompensated data on the fax calendar and then analyze it with flowcharts. I don't know about Winlist. I assume you can. I would assume you can. I think that the whole idea of collecting uncompensated list mode data in the clinical lab -- and people correct me if I am wrong. There is nothing wrong with doing that, and I think there is some experimental data to support that that is a pretty good idea. I think that they have to understand that the clinical budget is going to go up because all of those reagents that are being used will have to be run as single color controls. DR. PURVIS: Not only that. Also, you have multiple instruments, and you're doing samples. You've got to know which instrument it came off of. I mean, you've got to look at the serial number embedded each time that you pull up a file. DR. FISCHER: Let's talk to someone like Mary Alice who runs a clinical lab. Are you going to want your people to be taking all of their data uncompensated onto the FlowJo and doing all of this -- DR. STETLER-STEVENS: We're considering it. I've wanted to do it for quite a while. We haven't been able to move the force to do it because, first of all, they're saying, "Oh, God, no." Another thing, what about the Coulter's FC500? They have a different setup for compensation. We bought one, and it seems to work pretty nicely. I'd like to hear about -- DR. QUINTANA: The FC500 can do compensation that points at data. So all the compensation assistance is done in linear. It's not done on log. It's displayed with a lot of echoes with a front log look-up table. The FC500 has this ability to do post-acquisition compensation, where you can actually take all of your data uncompensated, you can run your compensation setup any time during the day as long as it is the same setting, and then you can apply that compensation to the linear data. It also gives you the capability to reanalyze the data to manually compensate it. It's not something we recommend, but using the compensation setup, it allows you to go back. An application for that would be to say you've looked at it and you have a concern. We use CD45 single colors along with a control cell as setup for the compensation. What we found is that with the control cells and running those control cells, it gets us a lot closer of a fluorescence measure to at least the samples that we have run through our light systems. It's as close as it is of any of the other systems that we have looked at. We can go back and actually set up setup panels and go back. You can reanalyze the data by rerunning a different lot of CD45, let's say one of the tandem dyes. If there is a different manufacturer, you can actually re-create a panel and reanalyze the data and apply the compensation to it. That offers a lot of flexibility, but it still keeps the operators from having to do any manual compensation because you will apply the standard auto-setup panel to run the data. DR. LAMB: You know, I've got a real problem with post-acquisition compensation, especially in a clinical setting, because you can make things what you want them to be. What's the difference? I mean, you get your instrument setup parameters that you use. If you're doing a leukemia, for instance, you set up your instrument according to manufacturer's instructions, which basically you are tied to if you are going to have the same thing from day to day to day and pass your CAT inspection. Maybe I am living in the past, but "It is what it is," to quote Popeye. This is a problem I had with people at one point in time who bought a Becton Dickinson instrument and didn't like to buy PerCP. So they bought the other manufacturer, Coulter's or Caltag or somebody like that, or FACSCalibur. And then they ran all of their leukemias uncompensated and came back to Winlist and fixed it later that day. DR. MARTI: Okay. He had his hand up for a while. DR. SCHWARTZ: This is ten years ago. I remember one of the same concerns about doing this is should we have list mode files at all. You should just get what it is is what it is. And we are going to take that histogram, and that is all we need. And now I don't think there is anybody here who does approach it like that. DR. QUINTANA: The post-acquisition color comp is not something that we would intend for the clinical labs. DR. SCHWARTZ: But it can be -- DR. QUINTANA: When you have the clinical setup software, it is going to run through the algorithms and do what it needs to do. DR. STETLER-STEVENS: Although there are situations when you have -- DR. SCHWARTZ: You want to tweak. DR. STETLER-STEVENS: -- a very small specimen that are patients who have been treated and you have weird autofluorescence and you have funny findings and the compensation comes out bad. But we have like 10,000 cells in the whole system. We can't adjust the compensation. You can, but if you run it, it's gone. It's done with. And sometimes it would be nice -- well, I compensate with my eyes on that point. I go, "All right. I know what that is. And I figure it out." But sometimes I think there would be a place where post-acquisitions could go in and toggle their little -- DR. WOOD: I think that there needs to be a clarification here, and that is what it is that we are storing. We want to store data. We don't want to store information. Once you have compensated it, you are now storing information. And you are imparting a bias field already. You want to store the most unbiased version of what is coming off the flow cytometer. So ideally what you would like to have is raw linear data that hasn't even been transformed by any type of a log display, but not all instruments have that. It's not unlike what was available with Electric Desk. I don't know if people remember that from way back where you could run a virtual flow cytometer at your desk that data was collected and then you could go back and rerun the experiment, adjust settings, and so forth, and deal with the fact that your operator didn't necessarily know how to set all of the fine adjustments. What you did there is you stored the data, not information. So I think that we need to be aware of that and actually break away from the old paradigm of when you store histograms. The histogram and display are one and the same. What we need to be looking at is there is data that is stored on the computer. And then there is a display that can be generated from the data, but the display is not the actual data. It's a way for you to visualize it on your standard. And the two of them are separate. DR. FISCHER: It's my understanding -- and I can't speak to the Coulter's because I have not -- it sounds to me like they are also now saving it as uncompensated data, but I'm not positive on that. I know that certainly on the new instrument, all of our data saved uncompensated. It's just the way they designed the storage. And then you come back and actually do the compensations post-acquisition because the data now is all saved. I can't speak as to whether it's there. I wish Gordon were here so he could speak to that. Phil, you might know. I don't know. DR. McCOY: I think it's log. Actually, you can store both, store in one list mode panel both compensated and uncompensated. DR. QUINTANA: The FC500 does the same thing. It saves a 10-bit compensated and a 20-bit length of 20-bit linear data. You have both models. DR. FISCHER: And I know that mine does the same thing. The Calibur does both. I'm assuming at some point in time they may decide to switch that to a clinical instrument. I am not sure they are going to want to do that. The research instruments obviously are all going to be that way because the research -- DR. HOFFMAN: Any future instruments are going to be that way. Any future instruments are going to have digital high-resolution linear data on that. DR. TAMUL: But that's the future. I think that we need to get back to the real world here for a second. The majority of clinical labs that are going to try to do this are not going to use FlowJo. It's too complex to carry out in three to five years. But everyone in here is running high-powered labs with really serious effects. The rest of the world out there is not going to quite be so sophisticated and able to do it. The only other thing I wanted to mention was that perhaps a definition or a little bit of change in semantics might help. When I was in the lab, what we used to do is called instrument setup or calibration with beads and instrument setup, then compensation adjustments and any other adjustments with biological control as optimizing for the samples in vivo. That might help clarify. DR. STETLER-STEVENS: I have to say if you're going to compensate online, you need to have a really experienced tech doing it and that -- DR. TAMUL: That CID up there. DR. STETLER-STEVENS: Yes. Out in the community, when you have people who don't know what they are doing, compensate -- DR. QUINTANA: The software system has the -- DR. STETLER-STEVENS: You have to go with something that is automatic. DR. HOUTZ: And you can do it automatic after acquisition, post-acquisition to be automatic. DR. MARTI: Howard and then Bruce. DR. SHAPIRO: I apologize for delaying this, but I don't see any way around it. First of all, when we say "post-acquisition compensation," we're really covering a moment to things because of the different ways in which different instruments process the data. If we are talking about the instruments that are in the majority of clinical labs -- and, you know, there is a dichotomy here. We have got Scans or FACSCaliburs, which are probably the most recent clinical version of the old style instrument. That is, the Calibur has got, what, 10-bit A&D converters? So basically you've got 10-bit data. You have got 10-bit linear data. And if you want to do post-acquisition compensation to that, you have to convert the log to linear and back. And you lose data. The thing is kind of granular. To make the data look halfway decent when you display them, you've got to diddle and add random numbers and things like that. So, in essence, on the other hand, if you were doing a three-color measurement on a Scan or Calibur, you could certainly compensate a two-color measurement. A lot of people can compensate a three-color measurement. Four colors is dicey. I don't think anybody can really compensate four colors effectively by eye. When you go to more colors, you can simply forget it. Now, if you look at the major competitor of the Calibur, which is the XL, Coulter XL, Coulter XL has done away with the log amps. It is collecting 20-bit linear data, which until recently was not accessible to the users. The newest software of Coulter gets that data out there. And the FC500 has got the digital software. So when you do that, when you collect 20-bit linear data, there is no log amplifier involved. You have got the raw data. So you can shift from linear log and back. You don't lose any precision in the measurement. And it's perfectly feasible to do post-acquisition compensation. Remember, if you are doing more than four colors, you basically have to do matrix compensation if you really want to compensate the data, certainly if you are doing four or more. We could argue about three. First of all, what we are talking about here, we are talking about doing quantitative, clinical quantitative, flow cytometry. It is not going to be possible to do that, certainly not on three-color instruments, probably on our four-color instruments, out there in labs in community institutions. it's just not going to happen. We now are all pretty much used to doing four colors for CD4, CD8. We know that is an improvement. And yes, you can still get by the guidelines and do it with a two-color instrument. Some people are doing that. But if you are trying to be on the cutting edge, you have got to be doing four colors. My guess is that to do what we need to do with the gating parameters that are involved, we are talking more than four colors. And almost all the instruments that are doing that are going to be doing digital processing. Okay. I didn't mention Cytomation. Cytomation has got log amplifiers, but they convert the log data to 16-bit data, which allows them to move it back and forth between linear and log without screwing it up too much. And they also can compensate for their log amp response. And BD with the Vivo Electronics, again, it's got a high-resolution linear data. So basically you can always store the raw data in a form in which you can shift back and forth between linear and log and which you can shift back and forth between compensated and uncompensated without degrading the number that you have. That is the important point. By the time we actually get through -- when we start doing this, if we set up a protocol now, the labs that did it would all have the wherewithal to do it the right way, even with the high-resolution linear data. And it wouldn't really matter. It doesn't matter if you run the compensation samples first and you collect the data compensated. You still have access to the uncompensated data. So whether you run pre or post-acquisition compensation is not going to make that much difference. DR. DAVIS: My only point is I think we haven't asked the basic question, if we're really going to do quantitative data, should we do any compensation? I mean, my data with what we do is if you compensate, there is no way you are going to get the same value from different instruments. And if the whole idea is to do quantitation where one lab gets the same answer as another, you should at least start with a base protocol where you understand the effect of compensation between instruments and all of that because if somebody -- DR. SHAPIRO: I agree with you that if the channel that you use for your quantitative measurement is one in which you don't expect any cross-talk, then a quantitative measurement is going to be cleaner. But that is a separate issue from the fact that in the context in which we are talking about doing quantitative measurements, we are almost certainly going to have to use enough gating parameters that we have got to compensate to get the population on which you are going to do the quantitative measurement. DR. MARTI: Bob, then Jim. DR. HOFFMAN: First of all, there are not that many. With the existing instruments, your situation is limited to what accommodations of fluorochromes you can use for specifically reliable quantitation. So like with the BD Systems -- people are going to ask. They're going to say like "Who's this?" You set up a system that is robust and reliable. We would use a PE. If we are going to quantitate with PE, we would use PerCP or PerQM. ABC doesn't have any. So you use accommodations other than FITC that don't have any cost docked in the PE. The only two fluorochromes that Billy talked about, at least in the near future, quantitating, are FITC and PE. Those actually are the two that you can pretty reliably compensate for on any system. DR. LENKEI: But you have to compensate. DR. HOFFMAN: So you're just more limited to what fluorochromes can be compensated on and what combinations of fluorochromes you would use in multi colors, for instance. DR. MARTI: Jim? DR. WOOD: Well, one of the things I just wanted to bring up in terms of doing your post-acquisition compensation; that is, some instruments only have what I have called subtractive compensation. With that type of compensation, the only thing you can do is two colors. Those two colors are sought exactly by subtractive compensation. Three-color cannot be. Four-color can be if you have, for example, colors 1 and 2 interacting and colors 3 and 4 interacting, but you can't have 2 and 3 interacting. So taking that into mind, then some of the older instruments really can't do compensation at all beyond two colors. So you have to do post-acquisition unless you choose the colors real carefully, in which case, say, for example, if you are doing three colors, the third color didn't need compensation to begin with. Then in terms of choosing colors, one of the things to keep in mind is trying to minimize the cross-terms that exist. If, for example, you can keep one and two so that they are only interacting, three and four interacting, you don't have much two and three, then the compensation matrix will be very simple. And the instrument or the map won't be working quite as hard trying to compensate. DR. MARTI: Kathy? DR. MUIRHEAD: To get back to Jerry's original question, which is what do we need to worry about, I think the answer is if you under-compensated in your sample, it probably doesn't matter. If you, without realizing, over-compensated, you're going to lose data, and you're not going to know it, you're not going to -- DR. FISCHER: I've actually seen that stuff come through from other laboratories. They've sent it through me and asked and said, "Well, you know, we are having a problem. We know you do a lot of flow. Could you tell us what is wrong with this thing?" And I look at them and say, "It's compensated. Where is your data?" My Calibur saves all my data compensated. It's gone, which is why I think they went to the uncompensated because now if you screw it up when you do the printout, you can just simply go back -- DR. MUIRHEAD: It's arguably more time, especially if you have got a system that enables -- DR. SCHWARTZ: What takes more time, doing that or another sample? DR. MUIRHEAD: Not being able to get the sample. DR. FISCHER: The sample is not available. DR. SCHWARTZ: That is even worse. DR. FISCHER: Compensated, you have a chance of going back and doing things with the data. DR. SCHWARTZ: Yes. DR. LENKEI: Compensation can be dangerous because then you don't see. If you gate on subsets of cells and aren't interested in double positive cells, then if you type the compensation, I don't see the double positive. If you type the compensation, then I see application of double positive cells. So it depends from the clinical point of view. If we send a Branch Lodicom to measure and the others forget, then it is important. Otherwise, probably to look at one pair and one a cell, there's a lot being compensated, but it's not because it's not effective. But we studied that, and I checked both sets. I think we have a very good compensation, even checking on the negatives. DR. HOFFMAN: The Calibur doesn't do full-matrix compensation, but PerPC or PerPC10, which doesn't have any spillover back into the FL3, what we don't compensate is -- so it's really the FL3 channel that will never be perfectly compensated by most -- DR. LENKEI: And the problem is that you are inferencing Calibur. And then if you use Calibur as compensation, sometimes you can get into trouble. It happens. So from this point of view, if you have CD3 concomitantly with Calibur, then it's no risk. Then you know how Calibur is looking, but you see still that it's a good compensation. DR. SCHWARTZ: Let me ask a question that is kind of rhetorical. When you compensate, do new populations appear or what is the uncompensated is still the compensated, but it looks pretty because it is -- DR. SHAPIRO: The only pseudo new populations are the ones on the bottom of the scale. There is some work being done to deal with that. I don't know. Maybe this would be an appropriate time to show that slide. DR. SCHWARTZ: I mean, the compensation doesn't make things appear that weren't there before. So other than making it look pretty, why bother? You can do gates. DR. HOFFMAN: That's actually not quite true. I mean, the example I used of being very densely stained, double stained, CD45, FITC, and a very dim DE. In a normal dot plot, in that example, you have to resolve those uncompensated. DR. SCHWARTZ: So things do appear when they're compensated? DR. HOFFMAN: Resolution of the -- DR. SHAPIRO: No, they don't really. DR. MUIRHEAD: Well, I think he's asking, do they become more discrete? DR. SCHWARTZ: Could you -- DR. SHAPIRO: The mathematical distance between those in the linear space -- DR. SCHWARTZ: But could you find the population so you can do quantitation on it? That is the question. DR. FISCHER: The answer is no sometimes. DR. MUIRHEAD: But if you I think don't -- DR. WOOD: That's not really true. DR. LENKEI: Yes, you can, but it depends on the -- DR. MUIRHEAD: You can. DR. WOOD: If it's not distinguishable when it's uncompensated, it will not become distinguishable. DR. FISCHER: The problem is if you have the bleed-over, how can you tell if bleed-over if you're looking at it from the quantitation, until we get to the point where we can do two-color quantitation -- DR. WOOD: Compensation is a crutch. Okay? Once you understand what the problem is and understand what the clusters are, then you can actually solve the problem on a routine basis with uncompensated data. DR. SHAPIRO: Yes, but compensation isn't a crutch for a human in the sense that when we are working, we start working. The data that we collect are collected in what I will call a color space. So just you use a three-color thing. We're basically using green fluorescence and yellow fluorescence and orange fluorescence. What we want to get from the green fluorescence -- or let's say green, yellow, and red. What we want to get from green, yellow, and red is fluorescein, phycoerythrin, and PerCP or PECy5 or something like that. The green signal is mostly fluorescein, but it's got some PE and maybe a -- well, it probably doesn't have very much PerCP or PECy5. The yellow signal is mostly PE, but it's got the chain and probably maybe some of the other things. The red signal is probably going to have a little PE in it, probably not very much fluorescein. The compensation gets us closer to the numbers that we expect to quantify the individual labels, which, in turn, is what we need to quantify the antibodies found per cell. And so in that sense, the compensation is not a crutch. The compensation is necessary to get us the information. Compensation was basically invented at Stanford to be able to sort cells using that kind of a gate. DR. STETLER-STEVENS: Can I make a comment about observations on compensation that it is not perfect when you are doing it manually? I have someone really good doing compensation, and I have really intelligent people. But it's not perfect. And you have a range of fluorescence that you're looking at. One of the cases I showed you early on was when we were looking at an antigen trying to quantitate it, when we kappa in one tube and lambda in another tube, we see differences. We see a difference when we have an extremely high level of bright staining with the light chain in the one tube and it's negative in the other. The antigen that we're quantitating changes. It shifts. And this has to do with compensation issues. Yet, it looks well-compensated. Yet, it's enough of a shift to change the numbers. If you're not interested in quantitating, you say, "Look, it's positive." And that is your answer. So if we're doing clinical, you know, leukemia immunophenotyping, it is perfect. But when it comes to quantitate, you see shifts in the numbers of the fluorescence. It has to do with it has to be compensation because it is another antigen that is either positive in one tube and negative in another. You see differences. DR. SHAPIRO: I think the summary compensation is -- we were talking about English majors the other day. This is not who wrote Shakespeare. Compensation is a mathematical process. You know, there is an equation to solve, and there is a right answer. And that is the way it's done. In terms of what we need to do if we are going to do quantitative flow cytometry, clinical or otherwise, you have to use matrix compensation for just the reason that you said. Manual compensation will give you data with which you feel comfortable, but you have no guarantee that you have the right numerical answers. So we don't need to discuss this anymore because basically if you want to do this right, then you have to do it on an instrument which allows you to do matrix compensation, preferably on one that keeps high-resolution digital data. I would venture to say that the labs that pilot this are all going to have access to that equipment. And the labs that want to prove this in the clinic are going to have to acquire that equipment. It would be very nice if the manufacturers made the software, made the use of those instruments as easy as possible, but, as Einstein said, no easy. DR. HOFFMAN: Just a small correction. You have to do correct compensation. Okay? Like in the case of the -- DR. SHAPIRO: Okay. Yes, yes. DR. HOFFMAN: It does not have to be matrix as long as you are correctly compensating the channel fluorescence that you are compensating. DR. STETLER-STEVENS: Which is -- DR. HOFFMAN: But it does have to be corrected, and it can't be to whatever your taste is as to what the data should look like. DR. LENKEI: But it just may take you to use adjust to correct for this type of high-intensities in some because you make the compensation adjusting to one tube which you run. So it would be the same problem. DR. SHAPIRO: No, no. There would be a right way to do it. I wouldn't worry about that. I don't think the intensity is as big an issue as -- DR. LENKEI: No because you titrate the in-tube art. Then you cannot -- DR. HOFFMAN: No. There is only one correct setting. If you don't change your PMTs and you are using the MPE, there is only one correct setting. The data may not look the way you want it to look. If they would be broad for -- I agree, but that is the way it should work. DR. MARTI: And then I would like to try and summarize this so we can move on to the next comment. DR. D'HAUTCOURT: I will make one more distinction. It is not only manual versus automatic compensation, but it is also hardware versus software. In my opinion, we must agree with the finding about three-color. There is no art work possibility. So there is only one way to do multi-color. It is to do the inverted matrix. This is only done with software. DR. SHAPIRO: You can actually implement that matrix in hardware. DR. D'HAUTCOURT: Yes. DR. SHAPIRO: So yes, you're right. That's correct. Basically hardware compensation is going the way -- DR. MARTI: You know, several years ago Carlton wrote a paper on compensation and pointed out very well the application that you get in hardware if you try to do that subtraction first. Of course, there was an exchange of letters between him and Mario over exactly what that meant. Then Mario published three papers back to back in Cytometry dealing with not only compensation but data exploration. I think that you have all been in the field long enough that you must have seen publications in the New England Journal of Medicine and Blood that have lousy histograms. When you start looking at four-color data, it is frightening. I mean, it sends shudders. I mean, there is probably only ten percent of four-color that is probably just going to be correct. DR. SHAPIRO: The question I wonder about is clearly -- you said a journal. A good journal sends something out to maybe three referees. A paper these days, it's got gels. It's got microarrays. It's got flow. It's got this. It's got that and the other thing. So I don't think flow is isolated. I think there is a lot of crappy array data out there and a lot of crappy gel stuff. DR. VOGT: Yes, right. DR. MARTI: If you think standardizing flow is great, you ought to try 18,500 spots on the microscope. DR. VOGT: Gridlock is right. There is, by the way, an NCCLS microarray committee, which is moving quite expeditiously or they exchange e-mails regularly. I have seen them. I haven't looked at them, but we hope to have some forum for discussing that at the CCS meeting in November. DR. MARTI: Just to see if I have interpreted or heard correctly, at some level, at perhaps three but certainly four and more, this group would recommend automated compensation inside if that's possible. And ideally somebody made a comment that the clinical labs that want to move toward quantitation in the setting of three or four or more colors is going to have to acquire -- DR. SHAPIRO: Let's say high-resolution because, in fact, I think that if you talk about the accuracy of present methods, that the Coulter approach, which does not do digital pulse processing but which gets a 20-bit number, is probably slightly more accurate than the digital pulse processing approaches. And as we are able to convert more bits worth of data, the distinctions will go away because ultimately full digital is better once you get enough resolution, but we don't need to argue about that because you can get high-resolution linear data. And, actually, as I said, even with cytomation, with log amplifiers, if you digitize log data to high resolution, then you also get to the point where you can convert to linear log and back. Ultimately what we really need is for the instruments to have the gains calibrated so you can change the gain and figure out what the new compensation settings are going to be. That is doable, but how soon it will take to implement that, I don't know. DR. HOFFMAN: I agree with that statement. If what you're saying is that you want to have the capability, the flexibility to quantitate on any of your fluorochromes, that is true. In any of the existing systems, even with four colors, if you choose the right combination of fluorochromes and the right fluorochrome is quantitated on -- DR. LENKEI: Yes. DR. HOFFMAN: I mean, you can do that perfectly well within your equipment. DR. LENKEI: If I had been stuck, you told me that after four colors, the matrix compensation was really good. I think on my Calibur, that is the best I can do, quantitation. DR. MARTI: The fact is that's where you can automatically start looking for bad data in the literature. DR. LENKEI: Yes. DR. MARTI: The PECy monitor. DR. LENKEI: Yes. And I don't think that any contents would be recommended for quantitation because too many subsets of data. DR. QUINTANA: What about correct compensation? Correct compensation can be correctly set with the lower-level instruments if it's done properly and it's the right fluorochrome. Is there any other form of algorithm that has a different name matrix? Coulter is using a matrix. So that seems to be what is -- DR. QUINTANA: It's available in the XL now as well as the FC500. DR. MARTI: Are there any other mathematical methods? DR. HOFFMAN: The correct compensation actually covers -- DR. QUINTANA: The XL software -- DR. HOFFMAN: It does either what we call conventional -- DR. QUINTANA: Right. DR. MARTI: Okay. On that note, I am going to conclude that -- DR. SHAPIRO: There's something I want to show related to compensation. And it's also related to log scales. I think it's important. So it's in a late-breaking news section of the fourth edition. DR. VOGT: Does compensation work if you don't have the cables hooked up? DR. SHAPIRO: I think there are some people in my lab who think so. DR. VOGT: Maybe it works better. DR. SHAPIRO: Here it is. DR. VOGT: Good. DR. SHAPIRO: One of the problems that you run into, one of the problems that you run into, with high-resolution digital data or not quite high enough resolution digital data, at the bottom end of the log scale when you do a conversion is that you run into what we call the picket fence, in fact. And they just start to get kind of grainy. This is particularly bad with compensated data. With compensated data, you have an additional problem. The problem is whenever you see a cluster, you always tend to draw data around it. So we tend to say, "This is different from these guys," but, actually, these are all the negatives. It's hard to see them on a dot slide because you have got a substantial chunk of the negatives clustered against your axes. The reason for that is that when you do this subtraction, we are taking a deterministic number and subtracting it from a random number. Down in this neck of the woods is where all the noise lives. So every so often when you do the subtraction, you end up with a negative number because you're really looking down around zero. And it's a fluctuation on that zero. So you can see this in uncompensated data. And it typically gets worse when you compensate data. So what happens is that things fail ungracefully. If you draw a contour around it, it looks like you have three clusters here when there is actually one only cluster. One of the problems here is that the log of the negative numbers is undefined. The log of zero is minus infinity. So when you try to display this stuff on a log scale, there is no place to go. The log scale is just a convenient dualization. The advantage of having high-resolution data is that you can do your statistics on the linear numbers. The log scale is very nicely displayed, but it has this failure in that if you have got numbers that go down below zero, it chokes up. So the question is, is there another scale that you can use that has the advantage of compressing the high-end data and expanding the low-end data but not blowing up when you hit zero. There are a few. At the cytometry development workshop, which is an annual event for serious hardware jocks and software jocks, that is held in Asilomar, California every fall, this subject has been kicked around for a couple of years. There were several approaches that were suggested. Jim Woods suggested that we do what they do in audio and use what is called a compander. The compander turns out to have the same problems as a log amp when you get to zero. Mario Roederer, who is now at NIH, and Dave Parks and Wayne Moore and some other folks at the Herzenberg lab and Adam Treister and Tree Star, who was also in the Herzenberg lab, were kicking around a biexponential transform. This is a transform that has got a lot of coefficients. And it basically uses hyperphonic science and you don't want to know. If you take this transform and apply that -- so the reservation I had about this is that, you know, if you give people access to these coefficients, they are going to want to tweak them the way they tweak the knobs out of a machine. That's bad. For sure, at least half the people who want to do the tweaking shouldn't be doing it. So if you can make a one size fits all or one size fits nearly all version of this thing, then this is good. So this is actually done with a beta version of FlowJo. I think Dave Novo is working to put the same stuff into MCS Express. I don't know about Winlist. And logical is Wayne Moore's term for this transformation. When you get here, you get the scale where you can accommodate a broad range of data. And your negatives now are from one cluster. The dotted lines show you where the negative values are, zero points. So everything looks good, looks like the kind of cluster you want to see. And we should all live happily ever after. Now, whether it some weird things start happening on higher ends of the scale, I don't know. My approach, my counter-proposal to this is if, for some reason, we can't get this out in the field, my counter-proposal is when you see a picket fence, just remember that that picket fence has a sign on it. And it says, "Beware of the data." If this doesn't work out, then I think what we need is a data that grays out everything. You can put the numbers there, but you can't show the dots. That way nobody is tempted to gain on junk. Bob? DR. HOFFMAN: Joe Hart has been working with the guys at Stanford University. DR. SHAPIRO: Okay. DR. HOFFMAN: And they have been trying, basically tweaking the knots on that algorithm and see maybe what setting of that transformation works best and doesn't give you artificial sort of looking data on all sorts of different situations. The next I think version of the software that comes out will have a compromised algorithm in it like this. So we won't have the picket fence or at least you will have the option of using a different axis to do this transform display. DR. PURVIS: This is really, though, just an axis. DR. HOFFMAN: It's just a display. DR. PURVIS: So you're still collecting linear data, displaying the linear results on a transformed axis, correct? DR. SHAPIRO: Yes. What this means, what this reflects is that the linear values at the low end, some of them are negative. And that is more so in compensated than uncompensated. DR. HOFFMAN: But it's not just them. You're displaying data on a log scale. It can give you artificial peaks, -- DR. SHAPIRO: Yes, yes. DR. HOFFMAN: -- even if there is no peak -- DR. SHAPIRO: Right. DR. VOGT: It is true that that is still for the -- I mean, that still log scale in terms of the peak recognition would be the same. DR. HOFFMAN: No, it's not a log scale. DR. SHAPIRO: It's not a log scale. I mean, you notice that. DR. VOGT: Well, that's not trying to go to 100. It is quite different than -- DR. SHAPIRO: It's a compressed scale, but the idea, the point is half the time until we had digital processing, the log scale reflected -- the log scale wasn't really a log scale because the log amp was not really linear along that entire scale. So the log data were always a little bit bent out of shape. The point is that this is a display. This facilitates visualization. All data processing is done on a linear -- DR. HOFFMAN: What I am getting at is peaks wilt and look different than they would -- DR. SHAPIRO: The numbers are all the same. The data -- DR. HOFFMAN: No. The data is not what I am talking about. No. All I am saying is with beautifully normally distributed in log space and, you know -- DR. SHAPIRO: It will still log -- DR. HOFFMAN: They disappear. And it may be shouldered. DR. SHAPIRO: As you know, it is really not -- DR. VOGT: Holding the low end. DR. WOOD: It is kind of what you are doing is you are moving in the low channels from something that is more linear looking to something that is exponential as you go further out. Actually, in the histogram over here on the left, the buildup on the axes is really a function of the fact that the channels are varying in width. That is, any time you have a histogram where the channel width goes from something that is infinitesimally small up to larger and larger values. Your peaks out in the middle, which looks like a separate population, is really the second part of the distribution. That is, half is on the axis. Half is out in the second population. And, practically speaking, that peak occurs about one SD away from the point where -- one SD away from the mean because your mean is actually there at zero on your noise. And so that is really a -- the reason that you have the split there is because of the fact that the channels closest to the axis are so narrow. They're infinitesimally small. And then as you go away from the axis, as you usually go out, they become broader and broader so that they collect more -- DR. FISCHER: We understand that. Try telling that to somebody who is not used to it. DR. VOGT: Well, I guess my question is I would be inclined to agree with Howard that it's better to not play these tricks and get -- I mean, if you look at that nice little cluster at the right end there, you really are not getting a good picture of the measurement process. DR. SHAPIRO: No. You're getting a fine picture of the measurement process. The linear data are the same. This is a display scale. What we are saying is that these are the optical illusions. I mean, that is closer to the real state of affairs than either of the distributions on the left. DR. VOGT: Because it is not bounded by zero? DR. SHAPIRO: Well, because it shows the bad because it does not make a distinction that shouldn't be there. DR. WOOD: If you think of the -- DR. MARTI: It establishes they're negative. I think this is a wonderful improvement, Howard. DR. SHAPIRO: Well, I had nothing to do with it. DR. VOGT: But they're negative because the compensation algorithm is -- DR. SHAPIRO: They're negative. DR. MARTI: Well, they're negative. DR. SHAPIRO: The compensation -- DR. VOGT: They are negative. DR. SHAPIRO: The compensation algorithm -- this happens with uncompensated data. You are more likely to end up with negative values with compensated data than with uncompensated data. But if you have values that are down near zero -- DR. WOOD: But negative values are real. DR. SHAPIRO: Yes. DR. WOOD: Once it restores, is this trying to take the baseline level of noise and put it at zero, which means there is stuff on the other side? You're trying to set your zero at the center of the -- DR. SHAPIRO: Bear in mind that if you are dealing with a scale of -- well, this goes from 100 to 105. If you know of an analysis that goes from a 4-decade window, a million modules down to 100 modules, this stuff is down in models. I mean, there aren't that many instruments that are going to take care of 100 models. DR. MUIRHEAD: Is it fair to say that this is a way of trying to visualize populations that are there without creating artificial ones? It's not really important for gating but that in terms of how you are going to quantify it, you're going to have an effect because you're not quantifying based on that transformer. You are quantifying based on the underlying data. DR. FISCHER: Mostly for reporting out to people who aren't used to looking at the data and understanding where the problems lie. DR. WOOD: It's a display function solely. DR. VOGT: Right, but it will be used for data. I mean, that's I think -- DR. FISCHER: Yes, that is the -- DR. VOGT: That's how it could influence your results. It would in that case not be just a display tool. It's actually part of the analytical process. DR. FISCHER: What we are dealing with here, I call it -- DR. WOOD: What this is is cluster knowns for the human being. Okay? That is, the human being is expecting certain clusters to occur based on the antibodies that are being used. So if you orthogonalize everything, then you can think in antibody space. And a human can then do clustering and group things based on what they feel intuitively is going to happen. Now, as far as a computer is concerned, it doesn't matter what the real orientation is. The computation on the raw data can be done in any space. DR. VOGT: Given that the pathologist or cytometrist is used to looking at certain clustering displays and that's actually -- I have compared it to looking at tissue through a microscope. I mean, you see these patterns, and it is very important. So is it true that this display would not too much disturb or perturb the people, that they will still be able to pick out their leukemic clusters in the -- DR. WOOD: Yes because what is happening here is if, say, for example, you were to have the luxury of having a full linear display and you could actually look at the big wraparound display and pick out the clusters. You would see what you, too, want to see there. What this does is it goes from at the lower channels to something that is linear because if you take the log function at the low channels, it's almost linear. And then it starts taking off and doing its bends, as do the compression. And so at the low channels here, you are almost linear in your approach or your display. And then as you go further and further out, it starts compressing it. It actually goes according to the amount of information that is out there based on the standard deviation. So at the low channel, where you need the fine resolution to see the small clusters, you have it; whereas, out at the far end, at the high end with high gate, the high values, you don't need the resolution out there. And so, therefore, you can start bidding things in bigger groups. And so in this sense, it goes based upon close to the information content, if you will, of what is out there. DR. LENKEI: What about the sources of error when it's from the clinical diagnosis? What will be the cost for us to know the process and it's correct? DR. VOGT: Well, that's sort of what I was asking. I mean, if you found that you have to pick out these clusters in order to -- DR. SCHWARTZ: It's going to look the same. DR. SHAPIRO: Zoom display. At the bottom, you look in at 100 channels. At the top, you can see every data point is on the channels. DR. VOGT: Right. I understand that. I'm just stating an operational question. Okay. Well, that's interesting. DR. MARTI: Howard, thank you for that. I wonder if before we take a coffee break, if we could get started on what was the next topic, which is QC. How are you doing now? How are you going to go about QC in your laboratories for clinical quantitative blood? Don't everybody speak at once. DR. PURVIS: QC3 bead setup standard sessions. DR. MARTI: Maybe this would be a good time for you -- DR. PURVIS: I would like to hear what everybody else has to say, instead of standing up and saying what I do. DR. SCHWARTZ: I think somebody should know what you do. DR. MARTI: Everybody in favor of having Norm lead the discussion? DR. SCHWARTZ: That's it. Yea, Norm. DR. MUIRHEAD: Somebody has to start. Go for it. DR. SCHWARTZ: Show us how it should be done. DR. PURVIS: I would like to have heard some of your comments before -- DR. MUIRHEAD: You will. You will. DR. PURVIS: I did all of this on the plane Wednesday night flying in here. So essentially what I did was took a bunch of my other presentations that I have given at other times, compiled them all, and they're up here now. So some of this is probably not going to fit right in with just QC. So we will scan through them real quick. This is comments that I have had before. Why do we have to standardize? I think it is pretty self-explanatory. We want to be able to do it intra-laboratory as well as inter-laboratory standardization. Here is an example of why. CD8 FITC and CD4 PE measured on two different instruments. Can you tell me whether that is the same sample? Can you make any kind of assessment off of that? Quantitative? I mean, we could say something about percentages probably, but could we directly prepare the intensities off of this? I think this is all about consensus here. Basically I base my decisions and my interpretations on the historic data that I have seen, what my personal experience is. Yours is going to be different. So it's a group of people sitting around a slide looking at the scope and, yes, all in favor. Next one. Well, here is the data we distributed from those same two sets of plots in a quantitative space. It's putting it on a quantitative scale: a, b, c's. What we have there is the same sample run on two different instruments with two drastically different instrument setups. So it kind of goes back to what we were talking about with Jerry's stuff the other day. His ended up that we had different samples. And I didn't catch that in his introduction to the data. This one truly was an instrument setup issue. Just by running MESF beads to knowing what that piece of these antibodies were, I can do this. Okay? And that's what we want to be able to get to so we can compare data. DR. VOGT: So, Norm, then the blue data was set up with presumably a lower PMT so that the resolution -- DR. PURVIS: What we lost was the negative information. What we lost was the negative. Therapeutic capabilities, this is why we are looking at quantitation. And we have therapeutic monoclonals, cell signal, drug pathways. I think we just need to be focused on surface markers, but there is also a distinct need to be able to do quantitation for cytoplasma as well and nuclear. So there are a number of important issues. You have got to standardize. I would go so far as to say we need to standardize how we do our instrument setup, but if we are going to go to absolute quantitation, as I showed you a while ago, if what we are interested in is the positive expression and not so much negative is negative, then, even by applying quantitation, we can standardize. Okay? I am going to go back and say this over and over again. There has to be agreement in the standards. I know that we can dead with bias. BD has a bias in one direction for PE and Bangs has a bias in the other direction for the PE, but I think that there needs to be the true absolute quantitative value that we are working with so that apples are apples. This here, well, I don't know. Yes, it would be great if we could define what the lower limit of quantitation needs to be and then set up our instruments and standards capable of resolving that. That is into the future. Okay? DR. VOGT: But, Norm, suppose I said, actually, the lower one were closer to the last one and getting closer to the middle one and furthest away from the top one. DR. MARTI: Yes. Why is that? Everyone wants to set up their -- well, you know, from a regulatory standpoint, it is the old issue. The minute you mention the word "standardize," then you take flexibility out of the system. Maybe that is what people don't even know that they are opposed to. DR. VOGT: I think that is a major part of it. I think those are three very good pullout points from this conference. DR. PURVIS: So, as I've given some other talks, essentially what we have got to do is concentrate in this one. Set up standardization, quality control and quality assurance programs, quantitative reagent quality control and standardization. The standardization there is something. It's not absolute standardization. It is how we are going to evaluate it; and then procedural standardization. We have discussed that numerous times already. DR. VOGT: As in preps and that kind of thing? DR. PURVIS: I put this slide up mainly to get across the point that we are a laboratory that has seriously been performing standardization for a long time. We have multiple sites. We have multiple instrument platforms that we are working with. We are in a process. Currently we have got 15 Epics going, but we have 6 FC500 installed. And we're in the process of replacing the XLs with the FC500s for a number of reasons that have already been pointed out here with the digital processing, the 20-bit linear data access. There's a number of reasons that we wanted to be able to go onto this platform, not to mention resolution, sensitivity issues that we have got. Anyway, the other thing that we do and that I have been required to do is to go out to principal investigator sites, go in and set up their instruments, and allow them to participate in some of the clinical trial work that we do. So I have got to set up their instrument and also train them on the procedures that we use in the actual processing of the samples because it is not just instrument standardization. You have to standardize the procedures. I think that is what we are going to have to do here. But not all assays are standardized the same way. So we are going to have to look at a number of variables. So it does become experimental. It is an experimental process, each one. DR. VOGT: Just to stop there for a second, where do people learn how to prep? I mean, basically in the laboratory they happen to start working in, I would guess. We haven't talked at all in this conference about there are a lot of didactic exercises in instrumentation and setup and none in preparation. DR. QUINTANA: What happens, Bob, for some of the training -- BD does the same thing -- is that they have customers like that would be running the automated CD4 systems. And they come for training. They get trained on those processes, what they do. They might get training in general but then on the individual. So I would imagine when clinical assays, clinical quantitative assays, come out, that we will be training people, purchasing or using, starting to get new instruments and stuff like that, that they are going to be trained at the manufacturers also on the specific methods that they use for setup. DR. VOGT: I think that would be very important for deliberations of this type of thing to have communications or some impact on those training programs. And then also the people who already have prep training and buy a new instrument probably don't either get that much or get trained that much when they come to learn about their new instruments. DR. PURVIS: And I think one of the other things that needs to be pointed out -- and I normally do this -- I can't use your automated instrument setup. Because I have BD, I have Coulter, they don't give me the same instrument setup. So I have to go to another program. DR. HOFFMAN: You use the BD kit to set up the instrument and try to use the same kit to set up Coulter. DR. PURVIS: And you're automated, your compensation, everything that you all are doing, I've got one over here and one over here. How am I going to make these instruments look the same? DR. FISCHER: Is that because we don't have one uniform standard that everybody should be using for setup of instruments? DR. QUINTANA: They are different protocols. DR. PURVIS: And there's a number of differences within instruments. DR. QUINTANA: They're optimized reagent systems and software, other systems. DR. PURVIS: I am not going to take credit for it. The man sitting right back there is the one. DR. HOFFMAN: I didn't take credit. The fact that you set up systems for each of their instruments, you need to set up a system that is probably -- DR. PURVIS: It covers everything that I can possibly go into. So there are still some other instruments that are sitting out there that I have to go into principal investigator sites and set them up as well. So we have had a tremendous challenge in doing this. We will go on to the next slide. DR. MARTI: I don't think I would hold my breath waiting for the manufacturers in their training courses to do much about standardization procedure. I would think about going to the lab where the procedure is being done and seeing if I could transport it from that lab back to mine. I think given the experience that we have had with Abe's course, both in this country and internationally, that the kind of stuff that you are talking about, when I am in the setting for those courses, the questions I get from the users would certainly suggest that those questions weren't answered when they went to the manufacturer's course, such as "What is the difference between a log scale and a relative log scale, relative fluorescence scale? Which scale are you using?" DR. PURVIS: Which scale am I using? DR. MARTI: No, no. When you just ask, most people don't know. I mean, 60 percent of the AIDS cohort -- that was a factor that emerged -- did not know which scale they were using. They still don't know. They use whatever the default one is. DR. PURVIS: There is so much terminology that is used there, RCI -- DR. MARTI: Oh, yes. They keep inventing new ones. DR. PURVIS: I think we have got to come up with a name so that we know whether we are talking about channel numbers, log channel numbers, or actual fluorescence intensity, I think, simplifying those terms down so that we are at least talking about either intensity or channel number. When you're in the linear scale, they're one and the same. It's a relative intensity. When you're in a log, log channel numbers do not reflect the actual relative intensity, fluorescence intensity. We have got to come up with some consistent terminology to use. DR. HOFFMAN: I think we should stop using the term "log channel number." It's really outdated. We don't need it anymore. We know that in the current software, if you are going to be using log amplifiers, the data is as linear units. DR. PURVIS: We still use log channel numbers in doing our QC. DR. HOFFMAN: We do, too. DR. VOGT: Well, Bob, I think this will be an unnecessary conversation in two years or so, but back when everything was analogued, I objected to the use of linearization because it made this presumption in channeling. I thought the raw data is how many events fell in this particular bin after it had been log-amplified and that was the real data. And it's all coming out on this convenient linearization is misleading. Now we have gone through a transition point here where some analog, some digital. When we come out the other end in full digital, then I think you would be exactly right. And I guess we're not far away from that now. And that may make things a lot easier. From the standpoint of fluorescence intensity, the definition of fluorescence intensity that is in the NCCLS guideline is the reading on the instrument. I mean, that is actually what you get. It can be a needle going from zero to whatever. DR. PURVIS: But that's some relative intensity. DR. VOGT: They're all intensities. DR. PURVIS: That needs to be pointed out, though. DR. VOGT: Well, the fluorescence intensity is -- DR. PURVIS: If I could change that number that you've given in the result by changing the voltage -- DR. VOGT: Right, but the fluorescence intensity still is the instrument reading. The fluorescence radiance is the actual amount of fluorescence. And how the two of them relate is the subject of quantification. That is what you are trying to establish. Any reading off any instrument is a fluorescence intensity. And you don't use the term "relative," but they are all relative. The log scale is relative, too. It's just that it's probably quicker. So all fluorescence readings are relative. And they're all fluorescence intensity. And then the way that relates to the actual amount of fluorescence is -- DR. PURVIS: What did you call it? DR. VOGT: The radiance. DR. SHAPIRO: Actually, most of the scales of optical measurements are really photoelectron scales because they're what comes out of the detectors. DR. VOGT: Right, but that is still considered fluorescence intensity. DR. SHAPIRO: Yes, yes. DR. VOGT: The equations that put those things into -- DR. SHAPIRO: That's true, but we are not getting into figuring out how many watts or even how many photons there were because there is a quantum efficiency thing there that we -- DR. VOGT: Right, but there -- DR. SHAPIRO: Well, yes. What we know is coming out, we know that somewhere in the detector, there are photoelectrons coming out. And everything proceeds from that. DR. VOGT: Right, right. DR. SHAPIRO: I do think it would be -- I mean, when I took a Coulter course years ago, there was no discussion whatsoever of the scale per se, what is it that you are measuring. And I know that none of the manufacturers have time with new users particularly to train them in optical physics and electronics and so on and so forth. There was no discussion of what this scale means. DR. HOUTZ: I think with the methods in the software dissertation, you see the educational benefits. I mean, it's a whole week's worth, and it's pretty intense. But it's still much more basic than what we are talking about here. DR. VOGT: Well, is it all basic or is it just more useful? See, I have always argued -- DR. HOUTZ: Specific to the product, too, though. DR. VOGT: And it's completely to the product. I think this stuff can be presented in an hour, the basics of what it is. I don't think people even know when they come to this course what the -- I don't think they think of it as a fluorimeter. They think of it as a flow cytometer. I think one hour would be plenty of time to explain the flow cytometer as a fluorimeter. That would include and probably end with a discussion of the scale. DR. HOUTZ: To a basic user? DR. VOGT: Yes, yes. I think people are unnecessarily afraid of this. You know, it ain't the question of the origin of the universe. It's just a tool. DR. HOFFMAN: I was really talking about the physics, about what is laser, what is NP. There are some slides. We can introduce those. DR. VOGT: Right. DR. FISCHER: The thing is that most of the education programs, -- and I have been through several of BD's, one of Cytomation -- the main focus is on how can you get back to your lab and plug a sample on that instrument and get data out. Whether you understand how it is doing it or not is not the big thing. I would argue from a lot of the users I have worked with I don't want them to understand a lot of this stuff because they are going to want to start monkeying around with instruments they have no business even touching. DR. VOGT: That becomes a self-fulfilling prophecy. And you wind up with a dumbed-down group of people operating your instruments. I argued against that 15 years ago, and I will argue against it now. DR. STETLER-STEVENS: We talked about details of education -- DR. VOGT: Wait. Mary Alice, I think it is a forum to talk about it because this is what you are trying to do. And it can start with the pathologists would be my suggestion, that they don't know what they are doing either. DR. STETLER-STEVENS: I think that the place to discuss it -- we do need more education in so many areas, period. But are we going to solve this problem right now? DR. VOGT: No, but here's the thing, that it's all quantitative flow. DR. STETLER-STEVENS: Then let it be more in education. DR. VOGT: Everyone is saying, "Oh, God." It's all quantitative. I mean, we're putting these clusters in places because it's quantitative. If it wasn't quantitative, we wouldn't have clusters. And I'm saying that you can begin there and in an hour, you can get it over with. And if you did that, you would have a community that understood that the cluster doesn't appear there because it's a cluster. It appears there because this is brighter than this and it was measured by the instrument that way. It doesn't matter if it is PMT or an APD. I would spend no time discussing PMTs. DR. FISCHER: Yes. We're not arguing with that. DR. VOGT: Right. But I never heard it presented that way. And, actually, I still don't, you know. And in the NCCLS process, we kind of pick through it, that, you know, this isn't that hard. It should fall out. As Jerry always said, the instrument should fall out of the equation. DR. SHAPIRO: Well, at this meeting that you very nicely got me sent to in Belgium last week, this is a meeting to try and define standards for biomedical metrology of all kinds, flow cytometry being a relatively minor thing here. What is metrology? Metrology is measurement science. The answer is that everybody who is doing research is measuring something. They all think of what they're measuring, but they don't think of the common measurement issues. So nobody knows what metrology is, but everybody is measuring something. DR. FISCHER: All the metrology we did when we worked for a prior company, everything got traced back. Whether we measured temperature, whether we measured weight, whether we measured volume, everything went back to the standard that is kept at the National Institute of Standards and Technology. Somewhere back along the line, someone made the decision that this is what we are going to measure everything by. Time is involved in that as well. If you look certainly in any GOP laboratory, it ought to have the little stickers on things that say that they can back to a NIST-traceable standard. There is nothing for setting up our instrument that goes back to a NIST. This is instrument qualification and operational qualification, instrument and operational qualification for these instruments. We don't have that for a flow cytometer, although I understand from BD that they're now going to be putting one of those out. We don't have that. And that is why a lot of the differences that we have come up with now are theirs because we don't have any standards that have already been set up with the instruments. DR. STETLER-STEVENS: Can I say that for education purposes, why don't we, you and whoever else wants to, come up with a list of what you think needs to be addressed in education, maybe not the first week? When you're just starting to learn how to turn the flow cytometer on might not be the best time, but maybe at the cytometry meetings, we should suggest to manufacturers courses that we think would be beneficial or we could even within the format of the society, the Clinical Cytometry Society. They have educational workshops. We could address this issue by forwarding it to these groups what we think needs to be addressed in education. DR. FISCHER: There's a lot of enough local user groups out there that they could run a half-day educational meeting. Instead of always having Howard come down and present data to us, they would have Howard come down and teach us some of this stuff. DR. SHAPIRO: The idea is that the book has always been designed to include pretty damn near everything you would want to know. DR. FISCHER: Yes, but, you know, Howard, an awful lot of people won't pick up a book. DR. SHAPIRO: Well, I suppose education -- DR. FISCHER: I hate to say it. I love to read, but I hate to say it. I pick your book up mostly when I'm sitting in the lab going, "I don't remember that." Then you go find Howard's book. I'm not taking it home with me to read. And a lot of these people, that's the only time. They're going to pick up the journal article that is concerned with their science before they are going to pick up the book to figure out how to make their science work right. DR. SHAPIRO: There were a few people who wanted it to be an electronic book, but most people didn't. So in the introduction there, I said, "The bottom line is you can still read it in the bathroom." DR. STETLER-STEVENS: So, Randy, you are on the Education Subcommittee. Let's move on. DR. VOGT: See how easy this is, Norm, for you to stand there. DR. SHAPIRO: Who else needs to check out? DR. STETLER-STEVENS: Should we put Howard on it, too? DR. VOGT: All right. DR. SHAPIRO: Well, balance me with Alice. DR. MARTI: Let's take a break. (Whereupon, the foregoing matter went off the record at 10:08 a.m. and went back on the record at 10:40 a.m.) DR. VOGT: This is it, Norm, is that right? DR. PURVIS: Yes, sir. DR. VOGT: And so where did we want to go on this? I've forgotten. Yes. We have gotten through what, four slides? Is that right? Eight? Hey, we're really moving here. DR. PURVIS: I guess continuing now, I agree that education is something we need to work with. There are a number of different setup options available to us. BD has one. Immunotech has one. I'm sure Tri-Dimension has one. Ortho had their own. So everybody has their own idea about how you are going to set up. One of the problems is that they are going to teach you their method, but that doesn't work when you have multiple instruments, multiple assays that you are working on. DR. Schwartz came up with QC Windows -- how long ago was that? -- as a way of standardizing our flow cytometers. I think it is a very good approach, an approach that we have been using forever now. We have modified it some, but for the most part, it is the same general idea that Abe was putting forth two years ago. So we do an initial instrument setup using QC Windows now from Bangs as well as other labeled beads and stained normal donor leukocytes, healthy volunteers. I won't claim that I'm normal or anybody else in my lab is normal. We go through an initial setup. We lock everything in. We don't set up our instrument unless it's -- we develop a very good QC program around our instruments to verify that our instruments are within our specifications, performing properly. And that is all based on our daily QC. We use here a combination of QC3 beads, full spectrum beads, CalBRITE APC beads depending on whether we are doing a Calibur where we use APC signals; FC500s, where we are using the flow set beads, Flow Set 660s and 770s. So it is customized, but everything is truly being based on the standardized setup in many of the different channels as I possibly can get a standardization problem. We do multi-color staining and do our compensation on that, verify our compensation on a daily basis. So not only are we looking at our beads and knowing that our compensation is turned on when we are looking at the beads and they should be falling in a certain range, but we also go back and multi-color the same sample so we can verify that that is good. Then we also took -- DR. STETLER-STEVENS: Do you approach this -- DR. PURVIS: A combination of both. A combination of both. For the most part, we found in our clinical area the samples that are -- because those are in PECy5, Cy7, our procedure was to -- the clinical trials area is essentially freshly stained samples that they're doing the basis off of because the samples are generally -- DR. LENKEI: Stained for five days. We stain always on a Monday, but it's for the four-color. It's four colors. I would check repeatedly. It's no change for this type. DR. PURVIS: And I've seen that with this. I would say you could do that without any problems. To tandems, no. Even after they have been put into a fixative and run, for the most part, 24 hours later, you can still run them in the compensation that you see initially compared to the 24 hours. It doesn't change. But you will start seeing that tandem degradation occur once you go beyond. So there we do make sure that we are dealing with fresh stained samples or something that is no more than 24 hours old. DR. TAMUL: Norman, do you use normal leuko cells or do you do any of the commercial controls? DR. PURVIS: The clinical lab does run some commercial controls. DR. TAMUL: Like CD checks? DR. PURVIS: Yes, CD checks or the -- what are they called? DR. TAMUL: Immunotrope. DR. PURVIS: Immunotrope. Primarily, it really goes with our lymphocyte cell- setting stuff that we're doing for the leukemia/lymphoma. There is really not something that is available for us in that. There are some stem cell controls that we do use as well that are standardized. They have a percentage that we ought to be hitting. We are using those, but that kind of goes along with the assay QC and not really the instrument QC. DR. TAMUL: Okay. DR. PURVIS: Okay. The other part of what we do is our linear characterization. We can do that either with quantitative or qualitative beads. So I've listed those up here. And I've put Quantum 1000, but you can also use the QuantiBRITE beads, trying not to just show what I do but give you all your different options. That's essentially our setup in a nutshell. So here is our modified QC Window setup. So here is our modified QC window setup, pros and cons. What we are trying to get here is a common window of analysis on all of our instruments and as many of the different detectors as we possibly can. We are going to base our compensation on biologic samples. I have a problem with using beads to base my compensation. I think for the most part they work all the time or they work most of the time, but there are cases where I have seen the bead compensation methods not work. So I tend to stay away from that. We do one instrument setup from all biologic samples. And DNAs fall outside of this because you have to run DNAs special. Some of the cell line works that we are working with, you can't run cell lines at this high setting. Your cells' auto-fluorescence will be up on the second decade, and everything else is off-scale. So in some cases, we do have stuff that we will run at some other settings, but for the most part, we establish our settings. We do our QC at those settings, verify our compensation is working at those settings, and that is what everything is being on that. DR. LAMB: Where are your -- when you say, "follow-up samples," that thing that comes to mind is that -- DR. PURVIS: That's what we use. DR. LAMB: You have it done every day? DR. PURVIS: Yes. And then we also use some of the patient samples that we do. We process some 20,000 LL beads a year just in the clinical lab. And then we have other patient samples coming in through all of the other labs. So we have a steady source of samples that we go to. And we also go back and reconfirm by re-staining them the next day if we needed to. It begins to address this approach. The modified approach, using the beads, the linearity beads, begins to address our resolutions. I think that Jim and Bob have both worked out a spreadsheet for actually looking at this in a much better way using the rainbow beads or some other beads to be able to characterize it. So that is something that I would like to incorporate in my QCs to look at that and see if maybe not on a daily basis but at least on a frequent basis, go to it and see if my instrument is changing because your low-end resolution is very important as well. So we want to be able to go towards that, but we haven't at this time. It allows us to do four-color instrument setup. Actually, that is moving to five-color now that we have moved to the FC500s. We're doing seven-color on our Altra. So this does work. We just have to have all of the materials. Colors. It does not standardize your background. Okay? You can have an instrument -- the Ortho instrument was one of them. The XL is one of them. The low-end resolution, sometimes we're pulling the negative cells out because we have really standardized on the high fluorescence intensity and tried to hit a mark there. And it will pull it out. That is something that you have to be aware of and look at. If your instrument is bad, you need to address it. And it's time-consuming. The good thing is we do it once. And I have instruments that have stayed in this QC setup for over three years without having to go back and make other changes, even after having a major PM. As long as the alignment has been put back in, they come right back into place. As long as you are keeping your instrument good and clean, you don't have flow problems, you don't have fluidics issues that are affecting it, this does work. So I am not running an automated system. Yes? DR. ORFAO: Do you also mean that they remain stable for three years for a standard -- DR. PURVIS: Our standard is standardized as well. We standardize our scatter off of the lymphocytes and -- DR. LENKEI: Even when you have an instrument service? DR. PURVIS: Yes. If the engineers do their jobs properly, yes. DR. LENKEI: Yes, they do that. DR. HOFFMAN: Do you leave it the same and then you monitor what the -- DR. PURVIS: No, I do not adjust. So I am expecting it to fall into an acceptable range. And as long as it is doing that and my compensation is also working properly based on our verification, then I am confident that the instrument is good to go. You will have some slight variations, but for the most part, from those slight variations and drops, temperature can have effects. There are a number of things that can have an effect here. The absolute change that it causes into the compensation is not enough to warrant going through and resetting it up for our clinical applications without a doubt. Some of our more quantitative assays, there have been some questions as to whether we should be adjusting that. DR. DAVIS: How do you define your acceptable range? DR. PURVIS: This was something that we defined years ago. I being an engineer would like to tighten up our acceptable range more than everybody else, but it was based on looking at the variance in the instruments, a very good instrument, and then defining it. I think we are using plus or minus 15 channels right now, but that is log channels. I think that at bright intensities, those log channels are problematic. So I would rather go to the intensities and say, "We are going to allow this much of a variance." I would like to tighten it up. I would vote for five or ten. DR. HOFFMAN: Well, that is going to depend on whether the manufacturer can actually make the instrument and has designed it to be that stable over whatever you are offering. DR. PURVIS: That has been a problem, but I would still think that tightening up probably for me would be something that I would suggest I would like to see immediately. DR. DAVIS: So you can look at it statistically in terms of establishing 2SD replicates. DR. PURVIS: We have data that we can go back and do that and actually tighten it up based on our historic data. And if I were to do that based on my historic data and call it a 2SD cutoff, then we would probably be down around a five range. The clinicians really felt like I was tightening it up and taking an instrument off line, that from their perspective, they weren't seeing enough variance in the data or they couldn't even see the variance in the data to warrant it. So on the quantitation side, maybe there is a need to tighten it up and expect it to be more stringent than that, but we are probably talking about 2SDs at the five-channel range. It depends on which channels you are looking at as well. If you're multi-laser and time-delayed, that has been problematic. I definitely couldn't expect that on my sorters, I don't think, multiple laser sorting. DR. HOFFMAN: Just one other question. DR. PURVIS: Sure. DR. HOFFMAN: Do you also leave the compensation setting the same or do you change that each day? DR. PURVIS: Don't be silly. It's all kept. It's locked in. DR. HOFFMAN: So then you just check that you have got appropriate compensation after? DR. PURVIS: Right. Now -- DR. SCHWARTZ: In fact, if it meets his criteria, most likely the compensations are okay because if the compensations weren't, it would screw it severely. And then he checks it again with cells to make absolutely sure. DR. PURVIS: And we're going to go through these quickly. If you want me to comment more, than I will. But the initial instrument setup is truly based on a fluorescence PMT setting using QC Windows, where I have a target channel for a bead that has been given to me in a COA. So this is where I will still be setting this instrument up to. One of the things that we do in our lab is when we get in a set of beads, we compare it to the old set of beads that we have been running with to make sure that we are not changing our instrument setup based on the manufacturer has made a mistake or something has occurred during transportation, which has happened in the past. So we verify compared to the old lot as well as our historical values. We know that we should be running very consistent. So we adjust the voltage to obtain a target channel. And this is initially done with the compensation turned off. So this is the initial instrument setup. We have done the fluorescence intensities now. We need to get the scatter detector settings standardized as well so we base that on our lymphocytes. Here we are using healthy volunteer whole blood. So for the most part, we are sitting in a very consistent pattern here. Most everybody's lymphocytes are good enough for us to work with. We have our PMT voltage gain settings. We have our scatter. Now we need to go through and look at compensation. Now, this is where I will differ from Mario Roederer and some of the other people that are out there because I have pathologists that I have to make happy. When we do compensation, for the most part, they don't want to see the flare at the end resulting in sales in a second decade or third decade or whatever it ends up being at. So they won't need to bring it down so that it looks orthogonal and everything has a flat top to it. DR. HOFFMAN: So you're overcompensating DR. PURVIS: I am overcompensating to a certain extent, yes. I walk a fine line with them. I try to still have some events moving up into the second decade. And then realizing that for the most part, my actual test antibodies are going to be much dimmer than what I set up here, I don't have to worry about that, seeing that flare. But I want to make sure that if I did get something biologically, plasma cell or whatever, that really jumps out there, that I am not going to make the pathologists panic because they have seen something that is like a 56-positive or whatever we have got, the combinations, going in there. They chase their tail, and then they come after me. All right. We do a biologic compensation. I have listed here how we approach that is to get a biologic continuum. Each tube has CD3, CD4, CD8 in it, ECD, PECy7, antibody combinations that are Ortho combinations that we might be using. What you end up with, b-cells, nK cells, and RT-cells, it makes it a very nice biologic continuum there. So as we start to adjust down here, I can see what my compensation is actually doing. Okay? Some kinds of ideas of what we were talking about before. You need two peaks or two populations to base it on. I've got essentially a continuum that I can base it on. Go down again. The previous one I was under-compensating. This one here I have overcompensated. I know all of you are saying, "Why did he make it so bright that they're going out in the last channel and piling up out there? That is causing all kinds of problems." I want to see them out there. I want to know what is going on out in that last channel or the channel before I hit the last channel so 1,022, I want to know what is going on out there. I have to ignore those that have piled up in the last chapter. I don't know what their intensity is. The instrument doesn't know what the intensity is. It can't properly compensate. So I ignore those from the standpoint of looking at mediums or whatever and trying to get my compensation set up perfectly. Next one. This would be fairly close to what I would call our compensation. I have gotten my team pass to accept that if they get something out there at this end, there is going to be a tail up there. They are really looking for a distinct population above that now. So they are accepting it. You can see here what it is in the last channel. They are tailing way up there. They have to be aware of it also because that population, if you did something like a plasma cell that goes off-scale, it will look over in another. If you are just looking at single-parameter histograms, it will look like you have got a population sitting out there. And you have to go back and look and see if it is truly a compensation issue where something has gone off-scale and you haven't properly compensated it. That's essentially how we go through the compensation. Next slide. I put this one in here because I think there is a lot of confusion on the users as to, man, let's use this log bias. Let's put this artificial signal in here and use that while we are setting up a compensation because it makes it look so much better. DR. HOFFMAN: Is that an offset that you put in your software? DR. PURVIS: Well, you could actually do it on the Coulter software. You can turn on a log bias. It's not stayed down into the list mode file. DR. HOFFMAN: Is that an offset on the -- DR. PURVIS: Yes. You can also do it in Winlist. And it's just essentially they're taking and redisplaying it into the events that are down on the axis. They're redisplaying up into this first decade. So this is our normal display. This is with the log bias turned on. If I start compensating down, you can see that here, all of a sudden, I started getting events that had gone on to the axis but are being redisplayed back out. So it's showing a double population there that really doesn't make any sense. Here that looks beautiful. I mean, I would love to see compensation that looked that way. If you look at it, really, what has happened is I have got a huge number of events down onto the axis so we don't see them there any more. And they have been redisplayed here. Same here. I still probably significantly overcompensated there based on what I am seeing. This is probably getting pretty close. So it is somewhere in between this one and this one is what I would really be -- these are slides I made a long time ago, before my pathologists kind of loosened up a little bit. Next slide. Data QC. At this point, we have got everything set. I can go through this setup in about a two-hour process. Now I would average what my daily targets need to be for my daily QC program. So now I am going to start running with my voltages set on, turned on, compensation turned on, everything set up the way it needs to be, or how I am going to be actually running the instrument. I am going to start running my bead sets back through. I am going to do this over multiple days. And I may even make multiple measurements each day. When I establish what my target range needs to be, what my mean of that needs to be, and then we allow that plus or minus range to deal with, it is essentially just running those through and then also running our biologic samples and verifying that we have a consistent compensation that is going to work, there are not going to be any problems. And then this establishes our daily QC specifications that we have to be on a particular instrument. One of the things that I think a lot of people have made a mistake of doing -- and I know that this is the case in some of the principal investigator labs that we have gone into -- is they will see this. They have a target range that they want to hit. All of a sudden, something happens one day. They come in. They turn on their instrument, let it warm up, and it doesn't hit the numbers that it is supposed to hit. Automatically because we told them, "You are going to follow this if this is outside of this, you have got to reset up your instrument." They are going to go through some simple troubleshooting issues to see if that is what is going on. Most times or one of the big problems that we were having was bleach was left in the line or something else. And the beads would hit that bleach before they made it into the flow cell, and things were all messed up. So there is some simple troubleshooting that you can do. In 99 percent of the time, just by going back through and flushing the system, doing a prime, clearing any bubbles that might have gotten in the flow cell or maybe it was our sheath tank cap wasn't screwed down tight enough and we weren't pressurizing to the same extent. Simple troubleshooting, 99 percent of the time, will get us right back in our range, and we can keep going. So that is why I am saying if we clean our instruments and stuff, there is no problem getting this to run for a long term. Next slide. We will go over it. We don't need to go over this. You all know the differences. Next slide. Okay. Resolution and sensitivity. This is just a comparison. One of the things, when you are setting up an instrument, you have got to know what you are dealing with and what you are trying to be able to measure. So here are some rainbow beads run on the primary instruments that are out there. I know that LSRN and LSRII are out there, but I don't have access to those. So I ran the ones that I had access to minus the FACScan. You can see FACSCalibur results, the populations. XL has a real big problem down here in the low end resolving the populations. FC500 is actually doing a better job than the Calibur for this particular caliber, but this is very nice resolution. I am very happy. So when we are talking about quantitative, I want to be working on this instrument or this instrument. Low-level I don't want to be trying to do it on the XL here. DR. HOFFMAN: Have you cross-Calibured the rainbow beads to -- I mean, those differences in filters can give you -- because of the rainbow beads spectrum, especially in DR. PURVIS: We see the same with Quantum beads, the FITC. I don't think he calls them Quantum anymore. I think they're going by the catalog number or something. That can have a problem. That can introduce some -- DR. HOFFMAN: I wouldn't necessarily call it a problem, but just for making comparisons, especially across instruments that are -- DR. PURVIS: I know that I would have the same filters in these two. Correct me if I am wrong. Jorge is here. DR. HOFFMAN: Is it 30/30? DR. QUINTANA: It's 525, I think, times 1015, yes. DR. PURVIS: I'm not sure what the caliber is. It's been such a long time. DR. HOFFMAN: 30/30. DR. PURVIS: So you're actually picking up a little bit broader ranger here. DR. SCHWARTZ: That's when you get a better resolution on some of them. DR. PURVIS: All right. But, anyway, there are other routines that we can actually look at and can actually calculate. That will give us the A and B value that the two of you all have worked out. I would love to be able to start implementing that as part of my initial instrument setup and QC programs. Next slide. Log linearity. I think it's important to know what you are working with and how well you can use it for quantitation. Here I have taken the rainbow beads and just used the old technique that I want to start with a low voltage and I want to increment my voltage up over and over and over again. It should have eight peaks here. I've gated it, each of those peaks, off of one of the other detectors. Now I am looking at it over here in FL2. So I don't adjust this, and I am only adjusting FL2 up. So I can identify each of these eight populations based on the one detectors and then follow it on the other detector as I walk it up. So I get very, very good information as to the median values on those as I walk it up. And at some point, I am going to start going off-scale. I will be walking some of the populations off. Next slide. Doing this on these three instruments, this is the type of information that you will see. This is one of the problems that we have with the old log amplifiers. It approximates a linear process, but it is not linear. What I am displaying here is the delta between those peaks, Peak 1 and Peak 2. In fact, this one was Peak 8 minus Peak 7, 7 minus 6, or 6 minus 5, 5 minus 4. I did it on each one of them. So I have got multiple plots here, and I can see what they are doing as they walk through. The XL has a problem. Down here when they were in that first decade and a half where I had problems with resolution, you can see that I have got some weird behaviors going on. Sometimes I would come across XLs that bode down. Other times I would find ones that bode up. So it just depends on what is going on in that calibration that they have on the detectors. Here on the FC500, you are getting pretty good linearity across the entire range. DR. SCHWARTZ: When you did that walking, was it like one volt? At every volt you made an increment? It almost looks like that. DR. PURVIS: No, no. This was five volts. DR. SCHWARTZ: Five volts? DR. PURVIS: Yes, worked it up. I think it was 5 volts, may have been 25. There was a lot more on this one, but I did very small increments so I could actually see what I was getting there. This here causes a problem with our quantitation because our four bead types that we are using to get our calibration line, where you are on that log scale can affect, give you a little bit of a wobble in, your calibration curve. So you may get one set of beads that will give you this but have a different set of beads that when you have different values, they fall a little differently on the scale, it causes problems. You have the same problem going on on the XL if you have a peak that is falling in that lower second decade. It's getting in there. It causes some problems. The FC500 so far I haven't been able to really see that. I do have this as being shown slightly. And I think that is more a function of where I was in that first decade and whether I was really resolving that population, I was getting a mass measure of the median channels, yes. Next slide. DR. VOGT: Norm, before you leave that one, that's an old technique and looks really good. I just am curious about whether the folks here, particularly engineer types, would agree that this still may be the state of the art or way too preliminary. We did guideposts. How do you do linearity? Of course, we use neutral density filters because you can get them, and they literally have eight or ten decimal places with beautiful density value. And they are NIST-certified. So there is reference material for that. And you can put those new Kolinsky filters in front of you, Dector door in front of your expectation energy. But then the discussions with Bob pointed out that you are also filtering out the noise, the optical noise, that you would get from the particle in the flow cytometer. I guess especially toward the lower end of the curve, that will show you a linear response, but it won't have an anomaly there caused by a noise contribution. Is that correct? DR. HOFFMAN: Also, to do the whole scale, you need an awfully expensive instrument, and I don't know if you necessarily want people going in and trying to -- DR. SHAPIRO: It filters a degree or two. Generally speaking, when the stuff is with the optical elements in the back -- DR. VOGT: Then you screen things out. DR. SHAPIRO: Even though you may have many decimal places -- DR. VOGT: Okay. So that's -- DR. SHAPIRO: This is a nice method. When you start getting down to the low end of things, you may actually get back -- DR. VOGT: That was going to be my next question. Are there some PMTs that you can have responses that your PMT is constant across your different voltages you are studying? DR. SHAPIRO: We are looking at it. Essentially, the channel's difference on the log scale is a ratio between fluorescence intensities of the two peaks on a linear scale. DR. VOGT: Right. DR. SHAPIRO: That is going to be constant. So that is what should be constant across the board. The ideal response curve is one of the most horizontal lines. The thing is that there are different kinds of log amps. The kind of log amps that are in Calibur are basically the modules. One takes over from another. That's why you have the waves across the top. On the FC500, where the whole process is digital, then that's pretty close to the horizontal line. DR. VOGT: So we, then, agree that if we are evaluating linearity response on a flow cytometer, this is probably still the best available method? DR. SHAPIRO: The way I see it. And the thing is that some log amp response curves are stable and some are not. So if you have a stable log amp response, having done that curve, you can calibrate the log amp on a channel by channel basis. And that is simply the way standardization is going to the whole issue of calibration and correction curve for the log amps. DR. PURVIS: Are they doing that digitally up at the software collection site? DR. SHAPIRO: Well, I think what happens is that somewhere in the digital processing, they have the correction factor built in. I can't give you the details for that, but I know that they tend to do that. Otherwise if you are doing digital processing, you will be better off over the range you are going to use. You still want to check the amps. DR. HOFFMAN: The only thing that I would differently is in the plot, plot the mean channel of, say, the lowest to the highest bead versus the delta, rather than versus -- DR. PURVIS: Sorry. Plot the what? DR. SHAPIRO: Plot the mean. DR. HOFFMAN: Then you can see where your actual variation is on your scale. DR. PURVIS: Okay. DR. VOGT: So then the lines will no longer be horizontal. DR. SHAPIRO: No. DR. MUIRHEAD: No it should be flat if you have been reading it out on your intensity scale, instead of on a -- DR. HULTIN: My assignments are pretty uniform assignments around a flat line for delta between the peaks. They don't generally go up at an angle like that. They're fairly horizontal. DR. HOFFMAN: I am wondering why it's falling off so quickly. DR. HULTIN: Probably where the place is running together and then when it's coming off scale or -- DR. PURVIS: It's not actually running together because the way I gave it, it went back to the previous slide. I am using one of the other detectors holding it constant and getting a region around it and explaining that region into another histogram all by itself. So there is nothing in that region or that population showing up there. So what you are actually getting into is that there is something going on once you get out to the very end on the FC500 and XL. I will do that in the future, put it on the -- DR. HULTIN: The future will be halfway through the last decade. After that, you have to -- well, you just can't use the quantitation. DR. VOGT: The other point Norm made but I just want to emphasize this is that you can have properly calibrated standards. And depending on where you put your standard curve, you might see shifts in the slope because if they fall into one of the bumps or wells along that -- DR. PURVIS: What happens if your residuals are affected. DR. VOGT: Your residuals are affected. And then imagine what it is like when those standards themselves were calibrated on instruments with some misbehavior of this sort because depending on whether that standard in that batch fell into a bump or a well on that instrument that was used to translate values, that can have the effect of -- DR. HOFFMAN: With what's there -- and that's the delta channel. If you drew a sort of best fit straight line through there, the variation with what you're seeing is what I've seen. Usually it's like plus or minus five channels, which is about five percent. So that is the extreme variation with the mean log amps. I looked at 40-some calibers coming through the back gate at one point, just comparing, at least as they left the factory, how the log amps were set up, basically doing the same thing, comparing the means or channels of rainbow beads. I was surprised how consistent they were. DR. VOGT: And then two operational questions. How far apart should these be ideally? And I forget the second question. First, how far apart should these -- DR. SHAPIRO: A factor of two would be okay, one and a half. DR. VOGT: So a factor of two would give you about 3.5. DR. SHAPIRO: You don't want them to be much more than a factor of two. DR. VOGT: Right. You don't want a real broad range because then you will miss bumps or things in between. Shoot, I forget my last question. DR. SHAPIRO: For the stereo bands here, you might could say 20 hertz with 20 kilohertz plus or minus half a decibel. That usually sounds like a pretty good spec. But if you do the math, a half a decibel is about six percent. That, in fact, is the variability that you get on the decibel log amps. So if you try and quantify and you want to quantify to