A conventional paradigm in cancer drug development is to identify worthy molecular targets and the corresponding intervening agents in preclinical models, followed by clinical evaluations in human patients. With a few exceptions, the clinical development was done without the knowledge whether the intended targets were present or important for the patients enrolled in clinical trials. This generalist approach has not been highly productive. For example, from 1996 through 2002, 209 anticancer drugs or treatments aiming at 18 newly identified molecular targets (e.g., growth factors, angiogenesis, DNA structure modifications, extracellular matrix proteins, apoptosis-regulatory proteins) entered clinical evaluation, and only 12 drugs/treatments or less than 6% produced survival benefits [1].

An emerging paradigm of matching molecular targeted therapy with patients or diseases with the intended targets has yielded some successes. The most astounding example is imatinib, which has shown significant activity in chronic myelogenous leukemia and gastrointestinal stromal tumor, the two diseases that have the two intended targets of imatinib (Bcr-Abl tyrosine kinase and c-Kit tyrosine kinase) as the respective key lesions [2, 3]. On the other hand, most human cancers have multiple lesions in multiple signaling pathways and would be less likely to respond to a single agent targeting a single aspect in the faulty pathways. A more likely scenario is where the intended molecular target can be readily identified and used to preselect patients for evaluation. An example of success in this area is trastuzumab, a humanized monoclonal antibody that binds to the HER2/neu (erbB2) receptor and thereby prevents signal transduction. In HER2-positive metastatic breast cancer patients, addition of trastuzumab to chemotherapy significantly improved the time to progression, response rate, and overall survival [4]. Conversely, while gefitinib, an inhibitor of epidermal growth factor receptor (EGFR) typrosine kianse, improved the objective response rate in non-small cell lung cancer patients, it did not produce survival benefits. A subsequent study identified several qualitative (EGFR mutation status) and quantitative markers (number of EGFR gene copies, EGFR protein level) as potentially important prognostic indicators for response rate and survival [5]. Taken together, these examples illustrate that successful translation of molecular discoveries to useful clinical interventions is possible. The gefitinib example further highlights the potential importance of quantifying the levels of molecular markers.

Our laboratory is interested in evaluating fibroblast growth factors (FGF) as potential targets for overcoming chemoresistance. This is based on our finding that extracellular basic FGF (bFGF) induces broad spectrum chemoresistance in cultured rodent and human prostate cancer cells [6]. This finding is consistent with the findings in small cell lung cancer cells, bladder cancer cells, chronic lymphocytic leukemic cells and fibroblasts, where bFGF causes resistance to multiple chemotherapeutic drugs including etoposide, cisplatin, fludarabine, doxorubicin, methotrexate, hydroxyurea, 5-fluorouracil, paclitaxel, N-(phosphonacetyl)-L-aspartic acid [7–9]. On the other hand, bFGF has also shown the opposite effect and sensitizes breast, prostate, ovarian and pancreatic cancer cells to different chemotherapeutic agents including cisplatin, etoposide, 5-fluorouracil, doxorubicin, carboplatin, and docetaxel [7]. In cancer patients, the role of bFGF expression in clinical prognosis is controversial. In pancreatic cancer, there is no relationship between the intra-tumoral level of bFGF and postoperative recurrence and survival, but an increased FGF receptor expression is associated with shorter survival [10]. A similar observation was made in non-small cell lung cancer patients [11]. There are also reports indicating an opposite relationship in patients for several cancer types. One study showed an association between increased bFGF expression in tumors and shorter survival in node-negative breast cancer patients [12], whereas other studies showed an association between increased intra-tumoral bFGF expression and better prognosis in breast cancer patients [13–17]. Similar (contradictory) results between bFGF expression and prognosis have also been reported for ovarian cancer [18] and pediatric high-grade gliomas [19]. The reasons for the apparently contradicting relationships between bFGF level and prognosis are not clear.

bFGF is a pleotropic growth factor and its levels is affected by multiple processes (e.g., during would healing, angiogenesis). Previous studies on the relationship between bFGF level and chemosensitivity typically test bFGF levels in plasma or urine. Analysis of bFGF level in tumors is theoretically more likely to correlate with the tumor chemosensitivity, but is limited by the lack of accessibility to patient tumor samples. Hence, a method that allows the use of archived tissues may have greater utility, especially if the evaluation is performed retrospectively for, e.g., the purpose of preliminary testing whether a laboratory-generated hypothesis may have clinical utility. A typical method for studying archived tissues is immunohistochemical staining the protein-of-interest and scoring the staining intensity by visual examination. We applied this method to study the relationship, and established an inverse correlation, between intra-tumoral bFGF level and tumor sensitivity to paclitaxel in 96 patient tumors [20]. Because the visual examination method is highly dependent on investigator-defined parameters, there is the potential for unintended bias. Furthermore, the scoring method primarily describes the outcome in discontinuous terms (i.e., yes vs no, more vs less), and therefore has limited utility in situations that require the numerical value of the biomarker.

The present report described the development of a quantitative image analysis-based method to analyze bFGF levels in archived patient tumor tissues. The image analysis results indicated an inverse relationship between intra-tumoral bFGF levels and paclitaxel sensitivity, and enabled the identification of four pathobiological parameters (stage, p53 and aFGF status, higher-than-median bFGF level) that affected the relationship between bFGF levels and paclitaxel sensitivity.

Biomarkers are gaining importance in the development and usage of molecular medicines. The present study presents a computerized image analysis method for quantifying protein biomarkers in archived, paraffin-embedded clinical samples. The two advantages of using immunohistochemical staining for protein-of-interest are the availability of archived samples and, because the 3-dimensional structure is maintained, the spatial or subcellular distribution information that is not available in methods that use tissue homogenates or extracts (e.g., ELISA, Western blotting).

Our image analysis results indicate statistically significant correlation between bFGF level and the antiproliferation effect of paclitaxel (measured as E_MAX and IC₃₀), but no relationship between bFGF level and the maximum apoptotic index. Both findings are consistent with our earlier findings using the conventional, visual scoring method. However, while both methods yielded the same finding of an inverse correlation between bFGF level and tumor sensitivity to the antiproliferation effect of paclitaxel, the image analysis method provided more robust data and presented several additional advantages, as follows.

The visual scoring method was semi-quantitative and based on the differential staining intensity at three different antibody concentrations. For example, a tumor showing no staining at the highest antibody concentration was scored as negative, a tumor showing positive staining at different antibody concentrations were scored as one, two or three pluses. In comparison, the current method required using a single antibody dilution, and, because the quantitative image analysis results were based on the results obtained from simultaneously stained standard curve samples, was not subjected to or affected by inter-day variability in staining intensity. Furthermore, the image analysis method applied the same analysis parameters to all samples and therefore minimized the potential operator bias.

The image analysis method, by measuring bFGF levels in individual tumors, provided values for each of the 87 tumors. In comparison, the visual examination results provided readings of relative intensity which provided only 4 scores (negative and three scores of positive staining). This, in turn, enabled the analysis of the relationship between bFGF level and tumor sensitivity to the antiproliferation effect of paclitaxel in subgroups of tumors with different pathobiological properties; the results show better correlations in four pathobiological subgroups (late stages, positive p53 staining, negative aFGF staining, higher-than-median bFGF level), compared to all other groups. These findings, in turn, improved our ability to extract information and enabled the generation of the following hypotheses. As p53 immunostaining is generally accepted as indicative of mutated p53 [23] and high stage tumors are generally more aggressive compared to low stage tumors, we propose that the effects of bFGF on paclitaxel resistance are more pronounced in the more aggressive tumors with p53 mutation, or, alternatively, the bFGF effects are masked by the presence of functional, wild type p53. The finding of the apparently better correlation of bFGF level and chemosensitivity in aFGF-negative tumors was unexpected in view of our earlier finding that aFGF amplifies the chemoresistance conferred by extracellular bFGF, in cultured monolayer cells [24]. Multiple possible reasons can explain this difference, including the microenvironment and heterogeneity in human tumors compared to cultured cells. Based on the better correlation in the high bFGF-expressing tumors, we speculate that there is a threshold bFGF level above which bFGF has a role in chemoresistance.

Our finding of a lack of correlation between bFGF level and the apoptotic effect of paclitaxel in patient tumors suggests that bFGF level may not predict for tumor shrinkage, a conventional method of measuring patient response to chemotherapy. Finally, the current finding that the relationship between intra-tumoral bFGF levels and tumor sensitivity to the antiproliferation effect of paclitaxel was dependent on the status of multiple tumor pathobiological parameters suggests that the previous contradictory findings on the merit of using plasma or urine bFGF level as a prognostic indicator may be partly due to heterogeneity in patient tumors.

The present study established a quantitative image analysis method that enabled the measurement of intra-tumoral bFGF level in archived tumor samples. The ability to quantify a potential biomarker provided the opportunity to study the relationship between the biomarker and chemosensitivity in subgroups with different tumor pathobiological parameters and thereby enabled further hypothesis generation, a process that may have implications for translational research.