Abstract. Subcellular distribution of mass can be analyzed by a technique that involves culturing cells on interferometers and digitizing their interference contours. Contour sampling resulted in 102 variables per cell, which were predictors of oncogenic transformation. Cell phenotypes can be deconstructed by use of latent factors, which represent the covariance of the real variables. The reversal of the cancertype phenotype by a combination of microtubule-stabilizing and -depolymerizing agents was described previously. The implications of these results have been explored by clinicians who treated patients with the combination of docetaxel and vinorelbine (Navelbine ® ). The current study was performed to determine the effects of different combinations on phenotype and in phases of the cell cycle other than mitosis. Combinations of paclitaxel with either colchicine, podophyllotoxin, nocodazole, or vinblastine caused phenotype reversal. Paclitaxel analogue, 7-deoxytaxol, by itself caused reversal. Factors #4, (filopodia), #5 (displacement and/or deep invaginations in the periphery), #8, and #12 took on values typical of normal cells, whereas the values of #7 (p21-activated kinase), and #13 (rounding up) shifted toward the cancer-type. All combinations altered microtubule arrangement at the cell edge. Delivery schedules and drug ratios used in clinical studies were subjected to analysis. Clinical response rates were better when the combination was not interspersed with a single agent (P=0.004). The results support the idea that efficacy depends upon simultaneous exposure to both agents, and suggest a novel mechanism for combination therapies. These therapies appear to restore in transformed cells some of the features of a contactinhibited cell, and to impede progress through the cell cycle even when provided at nanomolar concentrations.

The laboratory employs an assay for cell shape phenotypes, which is based on an analysis of contours obtained from cells that are grown on solid substrate interferometers. The shapes of several contours are analyzed by computing the values for a total of 102 variables. A few of these variables, e.g. shape factor, are dimensionless at the outset, and the remainder are rendered dimensionless by normalizing them either to dimensions of the cell or the ellipse of concentration. The variables' values comprise a unique description of each cell. Using various classification methods, we studied cell lines that tested negative for tumorigenicity in whole animals at the outset, but gradually became tumorigenic [1]. The results suggest that lines becoming cancerous share some common changes, even if they originated from different tissues. One of the drawbacks of this assay, however, is that the variables are not intuitively related to morphological features of cells. This was addressed by making different combinations of variables to extract factors, which are based on common changes in variance of the variables' values. Of 20 factors extracted from a database of morphometric information, several correspond to specific features. When shape changes are broken down by factors, the biggest single, quantitative difference between normal and cancer cells is in factor #4 (microspikes/filopodia) [2].The goal of the current work was to determine whether a morphological feature corresponding to factor #7 could be found. The important variables used in computing #7 were ALTI (mean altitude of projections), LNNC (mean length of negative curvature regions), MEDN (mean length of projection medians), WDTH (mean width of projections at base), and MINP (area of polygon formed by joining local minima). Some variables that were heavily weighted in computing #4 values had little weight in computing #7. WDTH was entered in both cases but with opposite sign (positive with #7 and negative with #4). The identification of #7 with a morphological feature was suggested by data obtained from a preneoplastic cell line, which had been derived and followed through the time course to tumorigenicity as described above. Cells transiently took on a phenotype like that of tumorigenic cells following treatment with a tumor promoter [3]. They showed enhancement of ruffling activity over a period of 0.5 to 5 hours, followed by ruffling suppression [4]. Visualizing actin in the cells, we found that the actin filament arrangement coinciding with the phase of ruffling suppression was suggestive of stress fiber formation (FIG. 1). Data on the frequency of cells exhibiting stress fibers over the time course confirmed this result (FIG. 2). Stress fibers form in response to activation of the RhoA GTPase downstream of EDG receptors. Since these receptors can be stimulated by adding lysophosphatidic acid (LPA) to the culture medium, further work was done to analyze cells treated with promoter alone or with