Cell Culture Drug Resistance Testing (CCDRT) Cell Death Assays:
Misconceptions Versus Objective Data
What the Critics Say: F. The "Oncotech Reporter," Issue 5, July, 1999. Semantic distinctions between "drug resistance" and "chemosensitivity."
Recently, Oncotech, Inc. has self-published an analysis of (sic) "Crucial differences between the Oncotech Extreme Drug Resistance (EDR) Assay and Chemosensitivity Assays" (Oncotech Reporter, No. 5, July, 1999). As I played a key role in the development of the EDR assay, and, in fact, defined the assay features of EDR and coined the term "EDR" (in my capacities of co-founder, Laboratory Director, and Scientific Director of Oncotech; refs: Kern,DH and Weisenthal, LM. J Natl Cancer Inst 82:582,'90; Weisenthal and Kern, Oncology 5:93-103,'91; Weisenthal, et al. Adjuvant Therapy of Cancer, VI, 1990, WB Saunders Co., Philadelphia,pp. 334-9), I am in a position to clarify certain misleading statements in this position paper.
Although Oncotech attempts to make a distinction between Oncotech's assay - ostensibly for drug resistance - and all other assays which it lumps together as "chemosensitivity" assay, this is purely and simply a semantic distinction; not a real biological distinction. To understand this, one must review the history of the development of the Oncotech assay.
The Oncotech assay was developed by Tanigawa and Kern at UCLA as a part of an early 1980s NCI contract to use the Hamburger-Salmon clonogenic assay for new drug screening. The clonogenic assay was based on preparing single cell suspensions, plating these suspensions in soft agar (which retards, but which does not invariably prevent fibroblast growth, see below), and allowing the single cells to grow into multicellular colonies, which were then counted either manually or with a computerized image analyzer. This endpoint suffered from practical disadvantages, and Tanigawa substituted tritiated thymidine labeling for colony counting as the assay endpoint. This endpoint was adopted by the NCI screening program not because "it identified drug resistance with an accuracy of >99%" (as claimed by Oncotech), but simply because the endpoint was felt to offer practical advantages for high volume drug screening. In fact, Tanigawa, Kern, and associates published many clinical correlation papers with the thymidine assay prior to 1990. In all of these papers, the assay was specifically described as a "chemosensitivity assay." In none of the papers does the label "drug resistance assay" appear, nor is the concept of "extreme drug resistance" described.
When I recruited Dave Kern to come to Oncotech in 1987, he brought with him a database with all of the assay/clinical correlations accumulated with the soft agar assays while he was at UCLA. These data were in the custody of the UCLA Department of Biomathematics, and Kern received permission from UCLA to take the database with him and to re-analyze the data for publication at Oncotech. Kern then gave me the dBaseR file with the raw data for a total of 450 individual assay/treatment correlations, and I spent about 6 months analyzing the data. It is important to note that most of the data had been previously published in a series of clinical correlation papers (again describing the assay as a "chemosensitivity assay"). It is also crucial to examine Oncotech's claim that "the EDR assay used tritiated thymidine incorporation as the assay endpoint, rather than colony counting," as Oncotech goes on to describe all of the artifacts with the colony endpoint which may contribute to false resistance, with the implication that the thymidine endpoint is a crucial and unique advance necessary for achieving high specificity for drug resistance. In point of fact, of the 450 total clinical correlations, 331 of the correlations were with the classic Hamburger-Salmon colony endpoint and only 119 were with the thymidine endpoint. More critically, of the 127 correlations for "EDR" (defined below), only 25 were with the thymidine endpoint, while 102 were with the colony endpoint. I was always very careful to make this distinction in my publications and presentations, but this point has become increasingly obfuscated since I left Oncotech in 1992.
In my analysis of the UCLA/Kern database, I made two important observations. First, there was a striking relationship between assay results and pre-test/post-test response probabilities which were precisely in accord with predictions from Bayes' Theorem (Kern and Weisenthal,J Natl Cancer Inst 82:582-8,'90; Weisenthal and Kern, Oncology 5:93-103,'91; Weisenthal, L.M. Developments in Oncology. Kluwer Academic Publishers,Dordrecht. 64:103-50,91). Second, if one were to draw a cut-off line with either the thymidine or colony endpoint, there were virtually no responses in patients with assays showing drug resistance at a level greater than one standard deviation below the median of the comparison, database assays for each drug or combination tested. This observation led me to coin the term "extreme drug resistance," and to define this as an assay result one standard deviation or more resistant than the median or mean of a dataset of comparable assays with each drug or drug combination in question. Note that nothing in this definition applied uniquely to a particular endpoint (in this case colony versus thymidine). It was simply a statistical definition to be applied to a cell culture assay in which continuous drug exposures and high drug concentrations were used. But the critical point, for the time being, is that there were only a total of 25 clinical correlations for "EDR" with the particular endpoint (thymidine incorporation) used by Oncotech.
As shown in Figure 1 and Table 1, there is a high degree of concordance between assay results with different cell death endpoints, which exceeds the concordance between thymidine and colony results (JNCI 82:582,'90). Thus, it is more justifiable to combine results from studies of cell death assay endpoints for meta-analysis than to combine results for colony and thymidine assays (see also discussion below). Three different groups published correlations between patient response to chemotherapy and extreme drug resistance (as defined above) in cell death assays (Bosanquet,AG. Lancet 337:711,'91; Larsson,R. and Nygren,P. Anticancer Res 13:1825,'93; Wilbur et al. Br J Cancer 65:27,'92). In this subset of publications describing clinical correlations with the EDR endpoint, there were a total of 224 correlations between assay results and patient treatment results. There were 35 cases with an "EDR" result (16%, or precisely that which would be expected by drawing a cut-off line one standard deviation below the median/mean in a normal distribution). None of these patients responded to chemotherapy. Thus, the EDR concept, according to the definition which I originated, has been shown to apply to the cell death endpoint, as well as to the thymidine and colony endpoints.
The Oncotech Reporter article draws a careful distinction between colony and thymidine endpoints. The claim is that the thymidine endpoint is less subject to errors which may result in false resistance.
Human tumor specimens most often contain a mixture of normal cells and tumor cells. Most human tumor specimens contain copious numbers of connective tissue cells, which, particularly in the cases of tumors from the breast and lung often vastly exceed the quantity of actual tumor cells. But virtually all tumor specimens contain these normal cells. In addition, some tumors contain large numbers of lymphocytes, which may spontaneously transform in culture and malignant effusions often contain mesothelial cells, which may also synthesize DNA during culture. The potential problem with the connective tissue cells is that they contain fibroblasts which can and do synthesize DNA in culture.
Now, what Oncotech does is to slice off a piece of the submitted tumor (which is itself most often a slice of a tumor which has already been processed for surgical pathology by the submitting institution). They then fix the specimen, imbed it in paraffin, and process it for standard surgical pathology, have a contract pathologist read the slide, and bill for this as a Part B physician's service.
It is important, however, to note that all the pathologist is saying is that, yes, some amount of tumor happens to be present, along with whatever else is also present (often in vastly greater amounts) in the submitted specimen. The rest of the specimen is digested with enzymes and plated in soft agar for culture with and without drugs. At the conclusion of the culture period, radioactive thymidine is added to the cultures. This is incorporated into the DNA of cells which are actively synthesizing DNA at the time the cultures are processed.
So here is the critical question: Is the incorporation of radioactive thymidine in these cultures specific for tumor cells? If it IS specific for tumor cells, then formally duplicating the surgical pathology is not necessary. All that should be necessary is an inspection of slides prepared from the enzyme-disaggregated tissue by a trained and experienced observer and a later comparison of this with a copy of the official pathology report from the submitting institution. If, however, it is NOT specific for tumor cells, then merely confirming the presence of some tumor cells in the submitted specimen does nothing at all to assure anyone that the cells actually incorporating thymidine at the conclusion of the culture are, indeed, tumor cells.
So, once again, is the incorporation of radioactive thymidine in these cultures specific for tumor cells? The literature would indicate that, no, it is not.
Lawton PA, Hodgkiss RJ, Eyden BP, Joiner MC. Growth of fibroblasts as a potential confounding factor in soft agar clonogenic assays for tumour cell radiosensitivity. Radiother Oncol 1994 Sep;32(3):218-25
Stausbol-Gron B, Havsteen H, Overgaard J, Fibroblast growth in the soft agar clonogenic assay for cervix cancer radiosensitivity. Br J Cancer 1998 Aug;78(4):550-3
Stausbol-Gron B, Nielsen OS, Moller Bentzen S, Overgaard J Selective assessment of in vitro radiosensitivity of tumour cells and fibroblasts from single tumour biopsies using immunocytochemical identification of colonies in the soft agar clonogenic assay. Radiother Oncol 1995 Nov;37(2):87-99
Torry DJ, Richards CD, Podor TJ, Gauldie J. Anchorage-independent colony growth of pulmonary fibroblasts derived from fibrotic human lung tissue. J Clin Invest 1994 Apr;93(4):1525-32
Peehl DM, Stanbridge EJ. Anchorage-independent growth of normal human fibroblasts. Proc Natl Acad Sci U S A 1981 May;78(5):3053-7
Greco RM, Ehrlich HP. Differences in cell division and thymidine incorporation with rat and primate fibroblasts in collagen lattices. Tissue Cell 1992;24(6):843-51
The last of these papers is particularly relevant, as the authors document that normal human fibroblasts grown in anchorage-independent conditions (the basis for selective identification of tumor cells in the Oncotech thymidine assay) have inhibited growth (cell division) but do not have inhibited DNA synthesis/thymidine incorporation. Essentially, human fibroblasts in anchorage independent conditions are blocked at the G2/M boundary, but not in S phase, so they continue to replicate DNA and incorporate thymidine, even when they are growth arrested.
Therefore, it cannot be concluded that the tritiated thymidine incorporation assay in soft agar is specific for tumor cells; in fact, a properly-controlled clonogenic assay may be more specific for tumor cells. Obviously, measuring normal fibroblasts rather than tumor cells could produce false results, including false results for resistance. Which is why the number of correlations specifically obtained with the thymidine endpoint is of such importance. By this measure, the DISC endpoint, MTT endpoint, and other related cell death endpoints have been validated to a much greater extent in the published literature than has the thymidine endpoint.
Does this mean that the Oncotech assay is invalid or unreliable? Certainly not. The validation of the Oncotech assay is in the published clinical correlations (the majority of which were with the colony endpoint - where fibroblast growth is at least partially inhibited by growing cells in anchorage-independent conditions; but 119 of which were with the thymidine endpoint). However, it is not accurate to claim that there is superior validation generally for the Oncotech assay, nor is there necessarily superior validation for the "EDR" component of this assay, over that published for the cell death assay endpoints.
So this all becomes a semantic obfuscation. What is abundantly clear is that the results of cell death assays divide drugs into two groups: those associated with an above-average probability of working and those with a below-average probability of working. This has been shown in every study ever published in which fresh tumors were exposed to drugs and measured for loss of viability several days following the initial exposure to the drugs. This has been shown for a wide variety of drugs in a wide variety of tumor types. This has been shown in a wide variety of laboratories. That there should be such a correlation is a perfectly ordinary expectation; but it is supported by an extraordinary level of proof.
So call it "chemosensitivity" or call it "drug resistance" (as I personally prefer). It all describes the same phenomenon: in vitro drug effects clearly predict for clinical drug effects. And there are no data to indicate that the thymidine endpoint is in any way superior or advantageous over the cell death endpoints.