Cell Culture Drug Resistance Testing (CCDRT) Cell Death Assays:
Misconceptions Versus Objective Data
Data Validating the Biologic and Clinical Validity of CCDRT With Cell Death (Apoptotic) Assays
What the tests can realistically do
Before going on to review the voluminous data which clearly support the use of these assays in cancer medicine, I'd like to present the following analogy. The use of these assays is very analogous to using a barometer to predict for rain. A barometer is a precise instrument, as the assays are precise instruments. A barometer accurately measures barometric pressure, in the same way that drug resistance assays accurately measure in vitro drug resistance. Barometric pressure correlates with weather, in a similar way to which in vitro drug effects correlate with clinical drug effects. A falling barometer in Paris in the spring is usually accurate in predicting for rain. A falling barometer in Los Angeles in the summer will usually be wrong. But it is always more likely to rain in a given location when a barometer is falling than when it is rising. Thus, it is a useful tool in the hands of an experienced meteorologist, but it would be absurd to use the tool in a vacuum, ignoring all other data and knowledge.
Technical confusion: Not all cell culture drug resistance tests are alike
When most academic oncologists refer to "chemosensitivity testing," they are virtually always referring to and thinking about "the human tumor stem cell assay" or "clonogenic assay," as promoted and published by Sydney Salmon of the U of Arizona and Daniel Von Hoff of the U Texas San Antonio. The work of these authors is often used to support the view that cell culture drug resistance testing doesn't work. Yet this technology hasn't been used by any private sector laboratory for more than 10 years. Nor have I ever advocated the clonogenic assay as the best cell culture assay (my criticisms of this assay date back more than a dozen years (e.g. Weisenthal, LM. Human tumor stem cell assay. New Eng J Med 308:1478-79, 1983; Weisenthal, LM, Shoemaker, J, Marsden, JA, Dill, PL, Baker, J, and Moran, EM. In vitro chemosensitivity assay based on the concept of total tumor cell kill. Rec Results Cancer Res 94:161-173, 1984; Weisenthal, LM and Lippman, ME. Clonogenic and non-clonogenic in vitro chemosensitivity assays. Cancer Treat Rep 69:615-632, 1985).
Another assay is often called a "tumor chemoresistance assay," and is based on a technology originally described by Tanigawa and Kern. This is a soft agar, non-clonogenic assay for solid tumors, currently provided by Oncotech. This assay is quite complex in its own right, but it is not germane to the technologies to be discussed below. The Tanigawa/Kern/Oncotech assay will be discussed in Part III, Section F.
Most so-called "authorities" omit any mention or reference at all to the non-clonogenic, "total cell kill" or "cell death/apoptotic") assays which have been most extensively studied and reported on in recent years and which are described below. These assays are the ones in use at private laboratories such as the Weisenthal Cancer Group, Rational Therapeutics, Anticancer, Inc., as well as reported in major studies by European and Japanese investigators and by the National Cancer Institute. One of these assays (the DISC assay) is also used in hematologic neoplasms by Oncotech.
Total Cell Kill (Cell Death or Apoptotic) Assays: CCDRT Technology which works
I would now like to proceed with a consideration of the data supporting the clinical validity of a particular class of assays. These are the "total cell kill" or "cell death" assays that I have been advocating, studying, and applying since 1979. These are the assays which have received, by far, the most amount of published study during the past 10 years. Remarkably, they are never or barely mentioned by critical university-based "authorities."
Cell death (CD) assays are different than both the "human tumor stem cell" ("chemosensitivity") and Oncotech "extreme drug resistance" ("chemoresistance") assay. Both of these latter assays are based on measuring inhibition of cell proliferation. The CD assays, in contrast, measure total tumor cell killing or cell death (CD). In practice, all of these cell death assays are quite similar. Suspension cultures are studied in the case of discohesive hematologic neoplasms, such as leukemia and myeloma. Three dimensional microclusters or macroclusters are studied in the case of solid tumors disaggregated by enzyme solutions and not subjected to mesh filtration (this latter procedure unfortunately carried out in the "tumor stem cell" assay).
The cell death assays are furthermore broadly similar to one another in that a short-term culture (typically 3-6 days) in the continuous presence of drugs to be tested is utilized. The drug concentrations tested are broadly similar. The assay results are categorized as "sensitive", "intermediate," or "resistant," based on comparisons with other assays in the database, with "sensitive" neoplasms falling at one end of the spectrum, and "resistant" neoplasms at the other. Although in practice results clustered around the median are typically labeled as "intermediate," published studies are forced to draw distinct cut-off lines, usually at or near the median, which results in an unavoidable degree of non-concordance, in the case of results clustered near the median, but falling on one side or the other in individual assays (which is why such results are more appropriately reported as "intermediate" in the clinical setting, though not placed into this category in published research studies, where statistical data analysis often calls for a single "sensitive"/"resistant" cut-off line).
Finally, the assay endpoints all measure cell death or some surrogate for cell death. This is of great importance, as the field of apoptosis research is clearly establishing the more important drug effects to be mediated through induction of "cytolytic" apoptosis (programmed cell death), rather than through "cytostatic" inhibition of cell proliferation.
The four most widely studied of CD endpoints have been:
1. The prototypic DISC assay, in which living and dead cells are differentially stained (allowing identification of tumor cells in mixed populations of tumor cells and normal cells) and where the endpoint is delayed loss of membrane integrity. This was the endpoint used in the NCI lung cancer study described later. The assay was first reported in the literature by me in 1983. It has since been the subject of numerous clinical studies.
The labor intensive nature of the DISC assay led to the adoption of more automated assay endpoints in studies in which the assay conditions first described by me were largely preserved. These automated or semiautomated endpoints all measure the loss of some enzymatic or metabolic function associated with cell death. They do have the advantage of being at least semi-automated, but the disadvantage of not distinguishing between tumor cells and normal cells in mixed populations. These endpoints include:
2. The MTT assay, in which loss of mitochondrial Krebs cycle activity associated with cell death is measured colorimetrically with the MTT reagent.
3. The Fluoroscein diacetate assay, in which the endpoints are the loss of cell membrane associated esterase activity and the loss of membrane integrity with cell death.
4. The ATP assay, in which the endpoint is the loss of cellular ATP with cell death.
Table 1 shows that these endpoints give substantially similar results in cases where "pure" tumor cell populations are studied. Most investigators and clinical laboratories perform cell isolation procedures designed to obtain reasonably "pure" tumor cell preparations or else restrict their studies to cases in which "pure" tumor cell populations exist. In addition to these studies, we have performed direct comparisons between the DISC and MTT assays in more than 3000 different fresh human tumor specimens with a wide range of drugs and have documented the comparability of these endpoints (e.g. Figure 1). Thus, published results with the above assay endpoints may be grouped for meta-analysis. The results presented below are being prepared for submission for publication, in collaboration with Dr. Peter Nygren of the Uppsala University Medical School in Sweden.
Table 2 shows the raw published data from which the meta-analysis results were taken, with literature references.
Figure 2 shows the results of each individual study, arrayed in order of increasing response rates in the total patient population studied. In every single case, without exception, assay "sensitive" patients were more likely to respond than the total patient population as a whole and assay "resistant" patients were much less likely to respond than the patient population as a whole. In every case, patients treated with assay "resistant" drugs were considerably less likely to respond than patients treated with assay "sensitive" drugs. This should not be a surprising finding. Intuitively, tumors relatively resistant to drugs in vitro would seem, on the whole, to be less likely to respond to the same drugs in vivo. This is precisely what the published data show.
Figure 3 shows the "relative risk" for response to chemotherapy according to the cumulative meta-analysis method of Lau, et al (NEJM 327:248-54,1992).
Individual studies are arrayed on the X axis in chronological order of publication date. Each hash mark represents a different study. The top solid line shows the "relative risk" for response, or the probability of an assay-"sensitive" tumor responding to chemotherapy, relative to the study population taken as a whole. The shaded area represents the 95% confidence intervals. Likewise, the bottom solid line also shows the "relative risk" for response, but, in this case, the probability of an assay-"resistant" tumor responding to chemotherapy, relative to the study population taken as a whole. The shaded area again represents the 95% confidence intervals.
It can be seen that patients treated with assay-"sensitive" drugs have a 1.44-fold greater probability of response (95% confidence interval 1.36 to 1.52) than the population taken as a whole, while patients treated with assay-"resistant" drugs have only about one-fourth the probability of response (relative risk = 0.23, 95% confidence interval 0.18 - 0.29), relative to the study population taken as a whole. These differences between assay-"sensitive" patients, relative to all patients, and assay-"resistant" patients, relative to all patients, are each significant at the two sided P < 10-8 level.
In practical terms, the overall study population had a 56% response rate. Patients treated with assay "sensitive" drugs had an 81% response rate. Patients treated with assay "resistant" drugs had a 13% response rate. Dealing only with solid tumors, the overall study population had a 45% response rate. With a "sensitive" assay, there was an 80% response rate. With a "resistant" assay, there was an 8.6% response rate. With solid tumors, the average advantage to receiving treatment with an assay "sensitive" regimen compared with an assay "resistant" regimen was a 9.3-fold advantage.
Assay Results in the Context of Bayes' Theorem
The absolute predictive accuracy of the tests varies according to the overall response rate in the patient population being studied, in accordance with Bayesian principles (e.g. see Figure 4). This is a very complex topic, on which I have written extensively. However, the crucial things to remember are:
2. Assay-"Resistant" drugs are significantly less likely to work than a clinician would otherwise expect.
3. Assay-"Sensitive" drugs are MUCH more likely to work than assay "Resistant" drugs.
Figure 5 shows the correlations between assay results and treatment, broken down as to histopathologic diagnosis. These are also arrayed in order of increasing overall response rates of the patient populations under study. In each case, assay "sensitive" patients were more likely to respond than the overall patient population and assay "resistant" patients were less likely to respond. In every case, patients treated with assay "sensitive" drugs were more likely to respond than patients treated with assay "resistant drugs." The only "near exception" to this point was in the case of head and neck cancer, in which results were available only from a single study, in only a handful of patients. Of note is the documentation in more than 100 breast cancer patients.
It may further be concluded from Figure 5 that cell death assays are broadly applicable to a broad range of neoplasms. This does not prove, for example, that the assays are clinically valid for a given rare tumor, such as esthesioneuroblastoma, but there is no reason to expect that the cell death assays should not be valid in any given type of neoplasm.
Figure 6 is of greatest importance, and is well worth considering. If one understands this figure, one goes a long way to understanding how the results of these assays should be used in patient management. The solid and dashed lines in this figure show the theoretical expectations for the cell death assays, based on Bayes' Theorem, applied to assays with an overall specificity for drug resistance of 0.91 and an overall sensitivity of 0.72, which represent the overall findings from the studies included in the meta-analysis. The circles show the actual response rates of patients with different types of neoplasms, given that either "sensitive" or "resistant" results were obtained. It may be seen that, in every case, the actual performance of the assays in each type of tumor precisely matched predictions made from Bayes' Theorem, projected from the overall assay sensitivity and specificity.
The findings in Figure 6 show conclusively that the cell death assays are broadly applicable to a wide range of human neoplasms, ranging from low response rate tumors, such as pancreatic cancer and cholangiocarcinoma (group 1, the non-colon, non-stomach GI adenocarcinomas) to acute lymphoblastic leukemia (group 11), and including breast cancer.
Of equal importance, this figure shows how the assay may be best applied to patient management decisions. It is obvious that, in high response rate neoplasms, there will be many "false negative" predictions. No one should ever use these assays to deny chemotherapy to such patients, if chemotherapy is otherwise indicated, any more than one should deny antibiotics in an infection with an in vitro drug resistant bacterium. But the assays could be appropriately used to identify patients with poor clinical prognoses if treated with a given therapy. In cases where more than one acceptable regimen exists, the physician could select the regimen containing the most favorable drugs and avoid the regimen containing the most unfavorable drugs. This would apply to clinical decisions at all points along the curve. Thus, the absolute probability of response with assay "sensitive" and "resistant" drugs varies according to the overall prior response probability in the patient population studies, but, at all points, assay "resistant" patients have a below average probability of response and assay "sensitive" patients have an above average probability of response and treatment with assay "sensitive" drug(s) is more likely to be associated with a favorable outcome than treatment with assay "resistant" drugs.