From: Dave Coder (dcoder@u.washington.edu)
Date: Mon Apr 13 1998 - 12:43:09 EST
To Mario's thoughtful response, I'd toss in a couple of ideas. That people are thinking of using experimental controls (in larger sense) is good and should be encouraged. But the crux of this whole discussion is the nature of the appropriate control. A couple of things to remember: 1. Flow cytometric measurements are measurements of light intensity. Further, the measurements are comparative not absolute. 2. How you design the experiment depends on the question you wish to answer. Given that measurements are comparative, you have a couple of choices: 1. choose a good control, or 2. calibrate your instrument response with a standard that yields biologically sensible units. As Mario pointed out, there are number of factors make the selection of a good--i.e., proximal--control difficult. (It's more a philosophical discussion as to whether you can ever get the absolutely proper control. It may possible with an individual animal if control serum is taken prior to immunization with the specific antigen of interest. Of course, you would take preimmune serum only after having done a sham immunization with any adjuvants or carriers used when challenging with antigen; your antigen should be absolutely pure, too. Then you can do the fluorochrome conjugation to give exact dye/protein ratios, and use precisely the same same protein concentration for each, etc. Most antibodies, however, are monoclonal so you really can't get the "perfect" control.) Even if you select a proximal control, there is no guarantee that it performs properly. (I have heard of isotype-matched control antibodies that have different isoelectric points, perhaps explaining some nonspecific binding differences.) More attractive is calibrating the fluorescence intensity scale to a unit that answers your question. Conceptually this is very attractive, and currently there's some agreement about how to go about this, but there is a lack of an independent standard. Point #2: How you proceed depends on the question you want to answer. Consider the following: A. How many cells of a certain type are present? B. What is the proportion of cells in the population that express an antigen? C. How many antigen molecules does a cell have? D. How many antigen molecules does the "typical" cell have? These are all different questions and different approaches may be used to obtain answers. The first can be very complex taxonomically, but in practice it is fairly simple. Perhaps 4 or 5 parameters are measured, and gating may be used to define a cell type as a subset of all cells examined. Where subsets are clearly present (discrete not continuously variable parameters), controls may not be needed. (A related example is setting compensation without single-labeled controls. You can do this easily on normal peripheral lymphocytes whose surface CD8 or CD4 are labeled with FITC or PE. You know in advance what the population distributions are so you know what a properly compensated bivariate distribution looks like.) B. If you wish to know the proportion of some cell type among all cells in the population, then the procedure is simple. But (there's always a "but" isn't there?), if and only if, the population subset of interest is discrete. If you can define the population using +/- classification, then quadrant measurements of distribution are useful. All bets are off, however, if any parameter distribution used to define the subset is continuously variable. Take activation markers for example. C. and D. This is like the old scholastic question: "How many angels can dance on the head of a pin?" Quantitation of antigen expression is a bit sticky as noted above, because the means for doing calibration and standardization are not universally accepted. Continuously variable distributions of antigen expression are best described if you can measure their number. In contrast to quadrant measurements (or other proportional measurements) that disregard the thee level of expression, quantitative cytometry takes advantage of the instruments' capabilities and is one the chief powers of the technique. D. This brings us to the often discussed issue of measuring the "average" cell. Without going into detail, I'll only point out the distribution of cell surface antigens is not always log normal. (In fact, I have yet to see any that are demonstrably log normal. See Coder et al. Cytometry 1994 18(2) 75-8 for discussion.) If a distribution is skewed, then the median is a simple indicator of the "typical" cell. Some classifier range is probably better, but how do you chose the limits to such a range? Quantiation of a cellular property could define a range, but then how much of the fluorescence is due to background? Hence, you need a good control. And I'll stop the ramble here for now. Dave dcoder@u.washington.edu -----Original Message----- From: Mario Roederer <Roederer@Beadle.Stanford.edu> To: cyto-inbox Date: Saturday, April 11, 1998 5:06 PM Subject: Isotype "controls" OK, first let me say that I put in the statement > Besides, "we" should all stop using isotype controls to set gates. primarily to see how awake everyone was. I must admit, I was (pleasantly) surprised by the level of attention! Of course, now I feel guilted into actually responding. Especially after Alice's recent posting. First, let me thank Phil McCoy for pointing people to the published paper about the use of isotype controls. This is an old topic, older than FACS technology, and it has been dealt with many times over the years. What follows below is my own discourse on the topic, uncolored by the rational arguments put forth over the past decades. Ray Hicks succinctly addressed some of the problems with isotype controls; I'll provide a little more detail. There are two principle issues: (1) the concept of isotype controls, and (2) the use of isotype controls to set gates. First of all, let's consider the whole point of "isotype controls." They are meant to approximate the background binding of your conjugated antibody to cells that wouldn't specifically bind your antibody (i.e., don't express the antigen). But this means that we would have to use a control antibody that is (1) the exact same isotype; (2) conjugated to exactly the same degree; (3) has the same background binding characteristics as your antibody; and (4) is used at the same concentration. Rarely is more than the first criterion met. (1) Exact same isotype. OK, how many of you actually purchase DIFFERENT isotype controls and use them for every different isotype in your experiment? I would wager a beer at the next ISAC that not a single lab on this planet does this. While most reagents are IgG1, there are plenty of G2 (a or b), some G3, etc. And there are some IgM's--arguably with enormous differences in background binding compare to IgG's. (2) Conjugation. In trying to estimate the background, obviously the F/P (fluor to protein) ratio is crucial. After all, if you double the number of fluors on your conjugate, you will double the background. Therefore, the isotypes should have the same conjugation ratio as the antibody you are controlling. While this conjugation ratio MIGHT be consistent for reagents from a single manufacturer (and it rarely is, at that), it certainly will be different for reagents from different manufacturers. (3) Furthermore, there is the problem that even at the same F/P ratio, the conjugates could be significantly different. It is quite possible that in one antibody there is a fluor at a critical "background" binding site; on another antibody, this site is unconjugated. This could significantly change the "stickiness" of an antibody. For example, when you conjugate antibodies to Texas Red, you can get hugely different "background" binding characteristics depending on the reactive form of TR that you use--even after getting exactly the same F/P ratios. Clearly, the sites on an antibody that are conjugated are different by these different reactive forms of TR, and those differences translate into different "background" binding characteristics. (4) Concentration. OK, what concentration do you choose for an antibody in an experiment? For a regular antibody, you choose the "saturating" concentration. For a control, there is no such thing as saturation; the more antibody you use, the more background you get. Therefore, the pragmatic approach is to use the same concentration as in your original reagent. And there's the rub! Each conjugated reagent has been carefully titred (hopefully) to be used at the proper minimal saturating concentration. Therefore, every different antibody can potentially be used at a different concentration! Do you therefore prepare a different isotype stain for each different concentration of conjugated antibody? Of course not. So how can you claim that the isotype control is even valid as a control? Calman Prussin and Brent Dorsett ask about isotypes having higher backgrounds. Recently I spoke with someone who had this same question. This researcher contacted the manufacturer of the isotype control, who told him that he should simply dilute the isotype control until the background was down! This smacks of homeopathy, doesn't it: "Our isotype control works better the more you dilute it!" As Ray asserts, now one has to use a little black magic in waving ones hands and ignoring the isotype control for these samples but not for others! This brings me to the second major problem, the use of isotype controls to set gates. Unfortunately, this is not a problem that will go away when people stop using isotype controls; most will simply use unstained cells to set gates. (Right now, however, the isotype controls lend an inappropriate air of validity to setting the quadrant gates). By the way, "we" should stop using quadrant gates! The problem with using isotype controls (or unstained cells) to set gates blindly is that many antigens do not should bimodal expression patterns that are either "on" or "off". Many are expressed even on "negative" cells; and, the brighter your reagent is, the more off of the bacgkround these cells will be! An excellent example of this is the expression of CD45RA on T cells. In the CD4 population, there are 2 reasonably distinct populations, RA+ and RA-. In the CD8, there are also two populations, but the lower population expresses a reasonable amount of CD45RA. If you were to use an isotype or background control to gate on CD45RA, you would include many of the "dim" population (which are memory cells) in the CD45RA+ gate (with which you are trying to select naive T cells). Furthermore, the gate that you need to use to distinguish the CD45RA populations is very different for CD4 cells and for CD8 cells. There is no easy solution to these problems. Gating is an art, one which requires considerable experience and knowledge of the system. (Hence job security for FlowJocks). Blindly using isotype gates or background, or blindly using quadrant gates (because you are lazy) can only lead down the path marked "Artefact". I do want to reiterate what Ray Hicks said: Isotype controls can be a valuable tool for rooting out problems. However, it is a rare problem that will be solved with isotype controls. mr
This archive was generated by hypermail 2.1.6 : Thu Jan 01 2004 - 17:35:17 EST
