Chemicals and supplies
Bicinchoninic acid and bFGF monoclonal antibody were purchased from Sigma Chemical Co. (St. Louis, MO), Streptavidin-Biotin detection kit from Dako (Carpiteria, CA), cefotaxime sodium from Hoechst-Roussel (Somerville, NJ), DAB (3,3 diaminobenzidine tetrahydrochloride) substrate kit from BioGenex (San Ramon, CA), and collagen gel from Ethicon (Somerville, NJ). All other cell culture supplies were purchased from GIBCO (Grand Island, NY). All chemicals and reagents were used as received.
Development of image analysis method
The image analysis-based measurement was accomplished in 4 steps: (a) Identified the parameters for image analysis and developed a macro that automated the procedures. (b) Identified the human xenografts tumors that yielded a 100-fold range of bFGF levels as measured by ELISA. (c) Comparison of bFGF levels in xenograft tumors measured by image analysis and ELISA, to obtain standard curves. (d) Using the standard curve to convert the image analysis readings to protein levels.
Human tumor specimens
Archived tissues of tumors previously studied for paclitaxel sensitivity were used [20]. Of the original 96 patient tumors described in the previous study, 87 (15 bladder, 14 breast, 22 head and neck, 13 ovarian, and 23 prostate) contained sufficient materials for the current study. The pharmacodynamics of paclitaxel effects in these tumors were obtained from previous report [20]. The present study used the control, untreated samples to determine the baseline bFGF level.
Animal protocols
Establishing standard curves for quantifying the bFGF level required renewable tumor source. This was accomplished by using human xenograft tumors maintained in immunodeficient mice. We screened several human tumor cell lines, i.e., prostate (PC3), pancreatic (MiaPaCa-2 and Hs766T), colon (HT29), ovarian (SKOV3), renal cell carcinoma (RCC), and pharynx (FaDu) (American Type Culture Collection, Rockville, MD). Pilot studies indicated a ~100-fold range in the bFGF levels in these cell lines. The four cell lines that yielded the greatest dynamic range of bFGF levels, i.e., FaDu, HT29, PC3, and MiaPaCa-2, were selected for further studies. Cells were cultured at 37°C in a humidified atmosphere containing 5% CO2, in culture medium (PC3 and MiaPaCa-2 in DMEM medium, HT29 in McCoy's 5A plus 0.1% non-essential amino acids, and FaDu in MEM medium plus amino acids) supplemented with 10% fetal bovine serum, 90 mg/ml gentamicin, 2 mM L-glutamine, and 90 mg/ml cefotaxime.
Five-week old mice were purchased from the National Cancer Institute (Bethesda, MD), housed and cared for in accordance with institutional guidelines. Tumors cells were harvested from sub-confluent cultures using trypsin, suspended in serum free medium, and implanted subcutaneously into the flank on both sides of a mouse (2 × 106 cells/200 μl per injection site). PC3 cells were implanted in male Balbc/nu.nu mice, HT29 in female athymic nude mice, and MiaPaCa-2 and FaDu in male athymic nude mice. Tumors (5–7 mm in length) were harvested and, after removing the non-tumor tissues, were each divided into two halves. One-half was fixed in 10% formalin and embedded in paraffin for immunohistochemical staining for bFGF and image analysis. The other half was used to analyze for bFGF level using ELISA.
ELISA analysis of bFGF levels
Xenograft tumors were weighed, minced, and sonicated in a mixture of ice cold RIPA lysis buffer (containing 150 mM NaCl, 50 mM Tris (pH 7.4), 1% NP-40, 1% deoxycholate, 0.1% sodium dodecyl sulfate, 2 mM EDTA, 50 mM NaF; Upstate Biochem., Lake Placid, NY), 0.2% Protease Inhibitor Cocktail Set III (CalBiochem, San Diego, CA), and 1 mM phenylmethylsulfonyl-fluoride. Tumor lysates were centrifuged at 14,000 rpm for 15 min and the supernatant was stored at -70°C and subsequently analyzed for bFGF level using an ELISA kit (Oncogene, Cambridge, MA) according to manufacturer instructions. The antibody was murine monoclonal anti-bFGF antibody conjugated to horseradish peroxidase. Detection limit of the assay was 2.5 pg/ml.
Quantification of FGF levels in tumor tissues using image analysis
Sections of xenograft tumors and archived patient tumors were stained for bFGF using immunohistochemical methods as previously described, with the exception that we used only one antibody dilution (1:50) [20, 21]. The inter-day variations in staining intensity were established using the same standard curve samples stained on different days; the average variations in the intensity in individual samples (5 repeats) were 28.1% and the average variation in the resulted standard curve slope was 18.5%.
Images were captured at a magnification of 400× using a Hamamatsu (Hamamatsu-City, Japan) color chilled 3CCD camera attached to a ZEISS (Thornwood, NY) Axiovert 35 microscope, saved in 8-bit TIF format and analyzed in the HSI format (hue, saturation, intensity). The HSI format was preferred over the standard RGB format due to the better separation of the brown (bFGF staining) and blue (hematoxylin counterstain) colors. A macro was written in Optimas® (version 6.51, Media Cybernetics, Silver Spring, MD) to enable unsupervised, high throughput processing of the images. The macro opened each image in turn, applied a user defined threshold to distinguish the brown from the blue staining, extracted the sum of the optical densities (OD) for each image, and converted the data to the corresponding bFGF level using a standard curve.
We quantified 10 images for each of the standard curve tumor samples, and on average 4 images per patient tumor (determined by the amount of available tissues).
bFGF standard curve
To correct for day-to-day variations in staining intensity, the four standard curve xenograft tumor samples were sectioned and processed together with patient tumor samples. Tumor sections were photographed and the sums of OD were calculated using the macro. For the xenograft tumor samples, the image analysis results (sum of OD values) were plotted against the ELISA results on bFGF levels in respective tumors, to generate a standard curve.
Statistical analysis
Coefficients of determination (r2) were calculated using linear regression (SAS Inc., Cary NC, USA). Confidence intervals around r2 were estimated using Fisher's Z transformation [22].