Purpose: To determine the efficacy of AZD0530, an orally active small molecule Src inhibitor, in human pancreatic cancer xenografts and to seek biomarkers predictive of activity.

Experimental Design: Sixteen patient-derived pancreatic cancer xenografts from the PancXenoBank collection at Johns Hopkins were treated with AZD0530 (50 mg/kg/day, p.o.) for 28 days. Baseline gene expression profiles of differently expressed genes in 16 tumors by Affymetrix U133 Plus 2.0 gene array were used to predict AZD0530 sensitivity in an independent group of eight tumors using the K-Top Scoring Pairs (K-TSP) method.

Results: Three patient tumors of 16 were found to be sensitive to AZD0530, defined as tumor growth <50% compared with control tumors (100%). Western blot and/or immunohistochemistry results showed that AZD0530 administration resulted in the down-regulation of Src, FAK, p-FAK, p-paxillin, p-STAT-3, and XIAP in sensitive tumor xenografts compared with control tumors. The K-TSP classifier identified one gene pair (LRRC19 and IGFBP2) from the 16 training cases based on a decision rule. The classifier achieved 100% and 83.3% of sensitivity and specificity in an independent test set that consists of eight xenograft cases.

Conclusions: AZD0530 treatment significantly inhibits the tumor growth in a subset of human pancreatic tumor xenografts. One gene pair (LRRC19 and IGFBP2) identified by the K-TSP classifier has high predictive power for AZD0530 sensitivity, suggesting the potential for this gene pair as biomarker for pancreatic tumor sensitivity to AZD0530.

Translational Relevance

Given that Src inhibitors are in clinical trials, the ability to predict which patients would benefit most from Src inhibitor therapy would be a great clinical asset and may ultimately improve the grim prognosis of pancreatic cancer patients. To this end, we evaluated the in vivo efficacy and pharmacodynamic effect of AZD0530, a small molecule inhibitor of Src in a large collection of pancreatic tumor xenografts and used a novel approach to predict biomarkers for AZD0530 sensitivity. In this study, we show, for the first time, that the gene expression profiles of baseline tumors could be used to predict AZD0530 sensitivity in pancreatic tumors. Here, we used K-TSP algorithm, a novel machine learning approach, to identify one gene pair that has a high predictive power for AZD0530 sensitivity and that may have potential as a biomarker for predicting pancreatic tumor sensitivity to AZD0530 in the clinic.

A number of studies have shown that Src tyrosine kinase activity is frequently activated and elevated in many human tumor types where it seems to correlate with disease stage and patient survival (1, 2). The major consequence of increased Src activity is to promote an invasive tumor phenotype characterized by breakdown of cell-cell adhesion, increased cell-matrix adhesion, and formation of focal adhesions (3, 4). Accordingly, inhibition of Src activity in preclinical models restores cell-cell adhesion, inhibits cell migration and invasion, and reverses the Src-modulated invasive phenotype (5, 6).

Although a direct role of Src in regulating the cell cycle has been suggested, there is compelling evidence that Src affects cell adhesion, invasion, and motility in cancer cells during the later stages of cancer progression (4). Clearly, Src does not act alone but works in concert with a large number of substrates to orchestrate the process of cancer development and tumor progression (7). Src activates focal adhesion kinase (FAK) and signal transducers and activators of transcription 3 (STAT-3) and their linked activities act to control cell migration through the turnover of focal adhesions and the suppression of cell-cell contacts (810). Also, Src is an important mediator of many downstream effects of receptor tyrosine kinases including the epidermal growth factor receptor family (11). In addition, there are extensive data supporting the influence of Src on the development of a metastatic phenotype (12). Collectively, these findings linking Src to multiple processes that determine the clinical outcome of a tumor, have served as an impetus for the development of Src inhibitors, several of which are currently in clinical trials for treatment of a variety of human cancers (1315).

Pancreatic cancer has the worst prognosis of any major malignancy, and the annual death rate due to this disease approximates its annual incidence rate (16). At advanced metastatic stages, pancreatic cancer can almost never be controlled by any of the available therapeutic options, as reflected by an extremely low estimated 5-year survival rate of <2% (17). Clinical benefit of gemcitabine as a systemic agent for the treatment of advanced pancreatic cancer results in a median survival of <6 months (18). Improvements in therapy have been modest with the addition of erlotinib to gemcitabine in combination, resulting in improved median survival in the order of weeks (19). The paucity of breakthroughs in treatment regimens and continued poor survival shows an acute need for improvement in therapy for this lethal malignancy.

The extensive presence of activated/overexpressed Src in pancreatic cancer and its potential role in tumor development and progression makes Src an appealing target for pancreatic cancer drug discovery (20, 21). In the present study, we explored the in vivo efficacy and pharmacodynamic effects of AZD0530, an orally active small molecule Src inhibitor, (22), in a collection of patient-derived pancreatic tumor xenografts (PancXenoBank; ref. 23). In addition, we attempted to identify potential predictive biomarkers of AZD0530 using K-Top Scoring Pairs (K-TSP) of baseline gene expression profiles of tumors as well as pancreatic cancer cell lines.

In vivo efficacy studies. All animal experiments were conducted following approval by the Johns Hopkins University Animal Care and Use Committee. Fresh pancreatic cancer tissues operated from patients at Johns Hopkins Hospital during Whipple procedure (pancreatoduodenectomy) were propagated as subcutaneous tumors in 6-wk-old female athymic nude mice (Harlan) as a live PancXenoBank according to an Institutional Review Board–approved protocol (23). The patients had not undergone chemotherapy or radiation therapy before Whipple procedure. Xenografts derived directly from patients were reported to have better retention of the morphologic and molecular markers of the source tumors, despite serial passing across several generations of mice (2426). Tumors from each treatment case were allowed to grow on both flanks of mouse from the same patient xenografts until the tumor reached a volume of ∼200 mm3. Mice were randomized to treatment groups, with 5 to 6 mice (8-10 evaluable tumors) in each group: (a) control; (b) AZD0530 50 mg/kg/d by oral gavage for 28 d. Mice were monitored daily for signs of toxicity. For assessing relative tumor growth index (TGI), tumor size was measured twice per week throughout the study and tumor volume was calculated using the formula: tumor volume = [length × width2]/2. TGI was calculated using the formula: (mean tumor volume of drug-treated group / mean tumor volume of control group) × 100. Cases with a TGI of <50% were considered sensitive, a TGI of >50% were considered resistant to AZD0530. Fifty percent cutoff value was used as a criterion for sensitivity in prior published studies using pancreatic tumor xenografts (27).

Immunoblots. Tumor tissues (50-75 mg/mouse) from control and AZD0530-treated mice for 28 d were minced on ice in prechilled lysis buffer (20 mmol/L sodium phosphate buffer, 150 mmol/L NaCl, 5 mmol/L EDTA, 1% Triton X-100, and 0.5% sodium deoxycholate) in the presence of EDTA-free protease and phosphatase inhibitors (Roche Molecular Biochemicals) and sodium orthovanadate (1 mmol/L; pH 7.4). The minced tissue was homogenized using a Dounce homogenizer and centrifuged at 16,000 × g at 4°C for 10 min. Protein content in the supernatants was measured with the Pierce Protein Assay kit using bovine serum albumin as a standard. Forty micrograms of sample were electrophoresed on 4% to 12% Bis-Tris precast gels (Bio-Rad Laboratories, Inc.). After electrotransfer to Immobilon-P membranes (Millipore), membranes were blocked at room temperature with TBS [10 mmol/L Tris-HCl (pH 7.5), 0.5 mol/L NaCl, and 0.1% (v/v) Tween 20] containing 5% nonfat milk (Pierce) for 1 h. Primary antibodies for Src, p-Src Tyr416, FAK, STAT-3, p-STAT-3 Tyr705, Akt, p-Akt Ser473, mitogen-activated protein kinase, p-MAPK Thr202/Tyr204, epidermal growth factor receptor, c-PARP, XIAP, matrix metalloproteinase-2 (Cell Signaling Technology), and p-FAK Tyr861 (BioSource International) were diluted at 1:1,000 in TBS containing 5% protease-free bovine serum albumin (Sigma-Aldrich), and the membranes were incubated with primary antibodies overnight at 4°C with rocking. After washing thrice with TBS, the membranes were incubated for 2 h at room temperature with enhanced chemiluminescence (Amersham) anti-rabbit/mouse IgG horseradish peroxidase–conjugated antibody (GE Healthcare) at a final dilution of 1:2,000 in TBS containing 0.01% Tween 20 and 5% nonfat dry milk. After washing thrice with TBS, bound antibodies were detected by enhanced chemiluminescence (GE Healthcare).

Immunohistochemistry. Tumor tissues from control and AZD0530-treated mice were fixed in formalin immediately after surgical excision and processed in to paraffin wax blocks. Sections were deparaffinized using standard histologic procedures, and an antigen retrieval method (pressure cooker and high EDTA buffer) was used to ensure optimal antigen integrity and expression. A suite of immunohistochemical (IHC) biomarkers across several biological effect areas (Table 1) was screened against these tissue sections, using the Envision+ (DAKO) detection system. IHC staining was scored by eye (from a minimum of 1,000 cells for the % markers) by a blinded pathology concordance panel.

Table 1.

List of IHC markers and method used for scoring

BiomarkerBiological effect areaScoring method
Caspase 3 Apoptosis % Positive cells 
M30 Apoptosis % Positive cells 
Ki-67 Proliferation % Positive cells 
Phospho HH3 Proliferation % Positive cells 
MCM2 Proliferation % Positive cells 
CD31 Vascular Microvessel density 
CD105 Vascular Microvessel density 
Phospho FAK Invasion Total H-score 
Phospho Paxillin Invasion Total H-score 
Cyclin D Cell cycle % Positive cells 
Cyclin E Cell cycle % Positive cells 
Phospho H2AX DNA damage % Positive cells 
BiomarkerBiological effect areaScoring method
Caspase 3 Apoptosis % Positive cells 
M30 Apoptosis % Positive cells 
Ki-67 Proliferation % Positive cells 
Phospho HH3 Proliferation % Positive cells 
MCM2 Proliferation % Positive cells 
CD31 Vascular Microvessel density 
CD105 Vascular Microvessel density 
Phospho FAK Invasion Total H-score 
Phospho Paxillin Invasion Total H-score 
Cyclin D Cell cycle % Positive cells 
Cyclin E Cell cycle % Positive cells 
Phospho H2AX DNA damage % Positive cells 

Gene expression profiles. RNA isolated from baseline (untreated) tumors were profiled using Affymetrix U133 Plus 2.0 gene arrays at least in duplicates. This gene array has ∼54,000 probes comprising ∼20,000 genes. Sample preparation and processing procedure was done as described in the Affymetrix GeneChip Expression Analysis Manual (Affymetrix, Inc.). Gene expression levels were converted to a rank-based matrix and standardized (mean, 0; SD, 1) for each microarray.

K-TSP classifier. We used the K-TSP algorithm (28) to construct a discriminative classifier in predicting tumors sensitive to AZD0530. In brief, the algorithm exploits the information contained in the rank-based matrix by focusing on “marker gene pairs” (i, j) for which there is a significant difference in the probability of the event (RI < Rj) across the N samples from class Y = 1 (AZD0530 sensitive) to Y = -1 (AZD0530 resistant), where the event (RI < Rj) is equivalent to the rank of gene i is less than the rank of gene j if and only if gene i is expressed less then gene j (relative expression). Here, the quantities of interest are pij(m) = Prob(Ri< RjY = m), m = (18), i.e., the probabilities of observing Ri< Rj in each class. These probabilities are estimated by the relative frequencies of occurrences of Ri< Rj within profiles and over samples. Let Δij denote the “score” of gene pair (i, j), where Δij = ∣ pij(1) − pij(−1)∣. A score Δij is computed for every pair of genes i, j ∈{1,…,P}, ij. Gene pairs with high scores are viewed as most informative for classification. Using an internal leave-one-out cross-validation, the final k-TSP classifier uses the k disjoint pairs of genes, which achieve the k best scores from the training set. In this study, maximum number of pairs (kmax) was fixed as 10.

Gene set enrichment analysis. Gene set analysis was done using the GSEA software Version 2.0.1 obtained from the Broad Institute.5

Genes represented by more than one probe were collapsed using the Collapse Probes utility to the probe with the maximum value. Gene set permutations were done 500 times for each analysis and the pathway/gene set list was sorted by the Normalized Enrichment Score. We used the 199 pathways defined by Kyoto Encyclopedia of Genes and Genomes (KEGG) database as the gene set in this study. Human pathway annotations were downloaded from KEGG (August 2007 release). The KEGG human pathways used in this study include metabolism, genetic information processing, environmental information processing, cellular processes, and human diseases. One hundred and sixty-six gene sets passed the gene set size filter criteria (min, 10; max, 500).

Quantitative real-time reverse transcription-PCR analysis. Total RNA was extracted from eight baseline tumors of predicted cases (A6L, Panc140, Panc194, Panc159, Panc294, JH010, JH069, and JH131) using the RNeasy Mini kit (Qiagen). cDNA was synthesized using the iScript cDNA synthesis kit (Bio-Rad), following the manufacturer's instructions. cDNA (25 ng) from each sample was amplified with specific primers for LRRC19 and IGFBP2, and the housekeeper gene ubiquitin from Applied Biosystems Taqman probes using an iCycler iQ real-time PCR detection system (Bio-Rad). Accumulation of the specific PCR products was detected as an increase in fluorescence that was plotted against cycle number to determine the cycle threshold (CT) values. CT reflects the cycle number at which the fluorescence generated within a reaction crosses the threshold. Relative expression of the mRNA analyzed was estimated using the formula: 2ΔCT, where ΔCT = CT (mRNA) − CT (Housekeeper). Samples were analyzed in a blinded manner and plotted as log-scale.

Statistical analysis. All error bars are represented as the SE. Significance levels for comparison between groups were analyzed using unpaired Student's t test and Kruskal Wallis statistic. The differences were considered significant when P value was <0.05.

AZD0530 slowed the growth of pancreatic tumor xenografts. AZD0530 exhibited a range of in vivo tumor growth indices between 40% to 142% in the 16 human pancreatic xenografts (Fig. 1A). Using TGI <50% as a cutoff value for sensitive, three cases (Panc291, Panc410, and Panc420) showed sensitivity to AZD0530 treatment (Fig. 1A). AZD0530 did not result in tumor regression even in the most sensitive cases. Rather, the established tumors showed decreased growth compared with controls in sensitive cases (Fig. 1B). This is consistent with previously published studies of AZD0530 in prostate cancer models (22) as well as other Src inhibitors in colorectal, breast, and pancreatic cancer mouse models (2830).

Fig. 1.

Efficacy of AZD0530 on the tumor growth of in vivo pancreatic adenocarcinoma xenografts. Patient-derived low passage pancreatic cancer xenografts were implanted in athymic mice. Animals with established tumors were treated with AZD0530 50 mg/kg/d by oral gavage for 28 d. Tumor size was evaluated twice per week by caliper measurements using the formula: tumor volume = [length × width2]/2. Relative TGI was calculated by relative tumor growth of treated mice divided by relative tumor growth of control mice × 100. Cases with a TGI of <50% were considered sensitive, TGI of >50% were considered resistant to AZD0530. A, antitumor effect of AZD0530 on the tumor growth of 16 xenografts. Points, mean (n = 8 to 10 tumors per group); bars, SE. Dotted line, 50% TGI. *, significance (P < 0.01) compared with vehicle-treated tumors. B, growth curves representative of tumor sensitive to AZD0530 (Panc410) and resistant (Panc253) to AZD0530 treatment.

Fig. 1.

Efficacy of AZD0530 on the tumor growth of in vivo pancreatic adenocarcinoma xenografts. Patient-derived low passage pancreatic cancer xenografts were implanted in athymic mice. Animals with established tumors were treated with AZD0530 50 mg/kg/d by oral gavage for 28 d. Tumor size was evaluated twice per week by caliper measurements using the formula: tumor volume = [length × width2]/2. Relative TGI was calculated by relative tumor growth of treated mice divided by relative tumor growth of control mice × 100. Cases with a TGI of <50% were considered sensitive, TGI of >50% were considered resistant to AZD0530. A, antitumor effect of AZD0530 on the tumor growth of 16 xenografts. Points, mean (n = 8 to 10 tumors per group); bars, SE. Dotted line, 50% TGI. *, significance (P < 0.01) compared with vehicle-treated tumors. B, growth curves representative of tumor sensitive to AZD0530 (Panc410) and resistant (Panc253) to AZD0530 treatment.

Close modal

Protein expression by Western blot and immunohistochemistry. Src is an integral part of several signaling pathways and signaling through Src can mediate mitogenic, proinvasive, antiapoptotic, and proangiogenic effects. We therefore analyzed whether AZD0530 treatment affected tumor proliferation, invasion, apoptosis, and angiogenesis in vivo in xenografts using Western blot and quantitative IHC techniques. AZD0530 administration clearly down-regulated Src, FAK, p-FAK, and p-STAT-3 expression in the sensitive tumor (Panc410) compared with control tumors (Fig. 2). In addition, AZD0530 administration resulted in the down-regulation of XIAP as evidenced by the immunoblot of Panc410 (Fig. 2). However, there were no obvious differences in the protein expression in the AZD0530-treated resistant case (Panc265) compared with controls (Fig. 2).

Fig. 2.

AZD0530 treatment inhibits Src and Src pathway proteins. Three separate tumors from AZD0530 and vehicle-treated animals were homogenized. Lyates were resolved in SDS-PAGE and probed with specific antibodies against indicated proteins. β-Actin was used as a loading control. AZD0530 treatment in sensitive tumor (Panc410; TGI, 46%) resulted in the down-regulation of Src, p-FAK, FAK, p-STAT3, and XIAP compared with tumors in the vehicle-treated animals. Protein expression in the resistant tumor (Panc265; TGI, 111%) was not altered by AZD0530 treatment.

Fig. 2.

AZD0530 treatment inhibits Src and Src pathway proteins. Three separate tumors from AZD0530 and vehicle-treated animals were homogenized. Lyates were resolved in SDS-PAGE and probed with specific antibodies against indicated proteins. β-Actin was used as a loading control. AZD0530 treatment in sensitive tumor (Panc410; TGI, 46%) resulted in the down-regulation of Src, p-FAK, FAK, p-STAT3, and XIAP compared with tumors in the vehicle-treated animals. Protein expression in the resistant tumor (Panc265; TGI, 111%) was not altered by AZD0530 treatment.

Close modal

There was a very clear treatment effect from AZD0530 on both p-FAK and p-paxillin as shown by IHC (Table 2; Fig. 3A, B, and C). Boxplots of the levels of expression of these biomarkers showed clear differences (Fig. 3A), with well-separated median H-scores in each arm. There was a down-regulation of proliferative (Ki-67) and vascular markers (CD31) with AZD0530 treatment (Fig. 3A). However, the values are not statistically significant (Table 2). There were no significant differences in the expression of caspase3, M30, p-HH3, MCM2, p-H2AX, CD105, Cyclin D, and Cyclin E with AZD0530 treatment (Table 2).

Table 2.

Summary of the statistical significance in the level of expression of each biomarker between control and AZD0530 treated samples

BiomarkerControl NAZD0530 NFixed effect test PKruskal Wallis statisticKruskal Wallis P
CD105 18 18 0.855 0.07 0.788 
CD31 18 18 0.369 1.6 0.205 
Caspase 3 18 18 0.242 1.61 0.204 
Cyclin D 18 18 0.845 0.26 0.613 
Cyclin E 18 17 0.105 2.06 0.151 
Ki67 18 18 0.051 3.25 0.071 
M30 18 17 0.808 0.47 0.495 
MCM2 18 18 0.572 0.13 0.716 
pH2AX 18 18 0.778 0.961 
phosHH3 18 18 0.613 0.18 0.668 
pFAK 15 17 <0.001 7.75 0.005 
pPaxillin 17 16 0.001 8.91 0.003 
BiomarkerControl NAZD0530 NFixed effect test PKruskal Wallis statisticKruskal Wallis P
CD105 18 18 0.855 0.07 0.788 
CD31 18 18 0.369 1.6 0.205 
Caspase 3 18 18 0.242 1.61 0.204 
Cyclin D 18 18 0.845 0.26 0.613 
Cyclin E 18 17 0.105 2.06 0.151 
Ki67 18 18 0.051 3.25 0.071 
M30 18 17 0.808 0.47 0.495 
MCM2 18 18 0.572 0.13 0.716 
pH2AX 18 18 0.778 0.961 
phosHH3 18 18 0.613 0.18 0.668 
pFAK 15 17 <0.001 7.75 0.005 
pPaxillin 17 16 0.001 8.91 0.003 

NOTE: Deparaffinized sections from control and AZD0530-treated xenografts (2-3 samples each) from Panc291, Panc420, Panc185, Panc265, JH033, and JH024 were used for the IHC. Values of all six cases were pooled together for each biomarker to calculate the significance. AZD0530 treatment resulted in the significant down-regulation of pFAK and pPaxillin. Entries in bold have statistical significance.

Fig. 3.

IHC analysis for the expression of invasive (p-Paxillin and p-FAK), proliferative (Ki-67), and vascular (CD31) markers. Paraffin-embedded tumor sections from two cases sensitive (Panc291 and Panc420) and two resistant (Panc265 and JH024) cases to AZD0530 were stained with antibodies against p-Paxillin, p-FAK, Ki-67, and CD31. A, IHC boxplots combining the p-Paxillin, p-FAK, Ki-67, and CD31 of two sensitive (Panc291 and Panc420) and two resistant (Panc265 and JH024) cases. Strong immunoreactivity scores for p-Paxillin, P-FAK, Ki-67, and CD31 observed in the vehicle-treated tumors were reduced in the AZD0530-treated tumors. H-score means % of cells positively stained X staining intensity (0, 1, 2, and 3). B, representative micrographs of p-FAK from Panc291 showing low immunoreactivity for p-FAK in the AZD0530 treated tumor compared with vehicle treated tumor. C, representative micrographs of p-Paxillin from Panc291 showing low immunoreactivity for p-Paxillin in the AZD0530-treated tumor compared with vehicle-treated tumor.

Fig. 3.

IHC analysis for the expression of invasive (p-Paxillin and p-FAK), proliferative (Ki-67), and vascular (CD31) markers. Paraffin-embedded tumor sections from two cases sensitive (Panc291 and Panc420) and two resistant (Panc265 and JH024) cases to AZD0530 were stained with antibodies against p-Paxillin, p-FAK, Ki-67, and CD31. A, IHC boxplots combining the p-Paxillin, p-FAK, Ki-67, and CD31 of two sensitive (Panc291 and Panc420) and two resistant (Panc265 and JH024) cases. Strong immunoreactivity scores for p-Paxillin, P-FAK, Ki-67, and CD31 observed in the vehicle-treated tumors were reduced in the AZD0530-treated tumors. H-score means % of cells positively stained X staining intensity (0, 1, 2, and 3). B, representative micrographs of p-FAK from Panc291 showing low immunoreactivity for p-FAK in the AZD0530 treated tumor compared with vehicle treated tumor. C, representative micrographs of p-Paxillin from Panc291 showing low immunoreactivity for p-Paxillin in the AZD0530-treated tumor compared with vehicle-treated tumor.

Close modal

Global analysis of gene expression profiles. Baseline whole-genome transcriptome profiles of the 24 xenografts were obtained by microarray. By comparing the gene expression profiles between the AZD0530-sensitive and AZD053-resistant xenografts, a clear pattern of differentially expressed genes was observed. Figure 4A depicts the heatmap of the top 100 differentially expressed genes between these phenotypes selected by signal-to-noise metric. These data suggest that we can identify a predictor for AZD0530 sensitivity from baseline gene expression profiles.

Fig. 4.

Baseline gene expression profiles provide a clear pattern of differentially expressed genes between the AZD0530-sensitive and AZD0530-resistant xenografts. A, heatmap of the top 100 differentially expressed genes from the 24 baseline pancreatic xenografts. Five xenografts sensitive to AZD0530 (Panc291, Panc410, Panc420, Panc194, and JH131) have a different profile from the rest of 19 cases. B, heatmap of LRRC19 and IGFBP2 expression in the 16 training set by microarray. C, heatmap of LRRC19 and IGFBP2 expression of eight validation set predicted by the K-TSP classifier. High (red) and low (green) gene expression are shown, respectively. Note: Panc159 which turned to be resistant in the validation efficacy experiment, although predicted sensitive, have a profile similar to five sensitive cases.

Fig. 4.

Baseline gene expression profiles provide a clear pattern of differentially expressed genes between the AZD0530-sensitive and AZD0530-resistant xenografts. A, heatmap of the top 100 differentially expressed genes from the 24 baseline pancreatic xenografts. Five xenografts sensitive to AZD0530 (Panc291, Panc410, Panc420, Panc194, and JH131) have a different profile from the rest of 19 cases. B, heatmap of LRRC19 and IGFBP2 expression in the 16 training set by microarray. C, heatmap of LRRC19 and IGFBP2 expression of eight validation set predicted by the K-TSP classifier. High (red) and low (green) gene expression are shown, respectively. Note: Panc159 which turned to be resistant in the validation efficacy experiment, although predicted sensitive, have a profile similar to five sensitive cases.

Close modal

Training K-TSP classifier for predicting AZD0530 sensitivity. One of the goals of this work is to construct an accurate classifier in predicting AZD0530 sensitivity. We used the 16 tumors treated with AZD0530 as the training set. Tumors with TGI of <50% were considered as sensitive (n = 3), and tumors with TGI a >50% were considered as resistant to AZD0530 (n = 13). We used the K-TSP algorithm in this study as it has proven to be a discriminative classifier in various studies (31). From this training step, the K-TSP classifier identified a gene pair (LRRC19, IGFBP2) as the most discriminative features in predicting AZD0530 sensitivity (Fig. 4B). The decision rule for the classifier is as follows: If LRRC19 > IGFBP2 then sensitive to AZD0530, otherwise resistant to AZD0530. The interpretation of this rule is that the relative expression of LRRC19 higher than IGFBP2, the query case was predicted as sensitive to AZD0530, otherwise the cases were predicted as resistant to AZD0530 (Fig. 4C). The estimated leave-one-out cross-validation accuracy for the K-TSP classifier based on the training set is 97.8%.

K-TSP predictor achieves high accuracy in predicting AZD0530 sensitivity. To evaluate the predictive power of the two-gene predictor, we tested and validated the K-TSP prediction on eight independent patient xenografts that were not included in the training set (A6L, Panc140, Panc159, Panc194, Panc294, JH010, JH069, and JH131). We used the two-gene predictor to predict AZD0530 sensitivity on the baseline gene expression profiles of these xenografts. Three cases (Panc194, JH131, and Panc159) that were predicted as sensitive to AZD0530 and the other five cases (A6L, Panc140, Panc294, JH069, and JH010) were predicted as resistant to AZD0530. The eight cases were treated with AZD0530 (50 mg/kg/day, p.o.) for 28 days. TGI for two xenografts (Panc194 and JH131) was <50% (sensitive); the other six xenografts (A6L, Panc140, Panc294, Panc159, JH069, and JH010) had TGI >50% (resistant; Fig. 5A). The two-gene predictor achieved 87.5% (7 of 8) overall accuracy on the test set (95% confidence interval, 73.4-100%), specificity (83.3%; 95% confidence interval, 62.9-100%), and sensitivity (100% 95% confidence interval, 66.8-100%). The overall accuracy on the test set is similar to the estimated leave-one-out cross-validation accuracy, indicating that the classifier is not overfitted during the training step. The confidence level was estimated by leave-one-out cross-validation of the training data.

Fig. 5.

Gene pair identified by K-TSP algorithm is helpful in predicting AZD0530 sensitivity. A, prospective validation of efficacy in eight xenografts predicted by the K-TSP classifier. The eight xenografts predicted for AZD0530 sensitivity were treated with AZD0530 50 mg/kg/d by oral gavage for 28 d, and TGI were calculated. Three xenografts predicted sensitive to AZD0530 (Panc194, JH131, and Panc159) had TGI of 42%, 44%, and 71%, respectively. Five cases predicted to be resistant to AZD0530 (Panc294, Panc140, JH069, A6L, and JH010) had TGI of 61, 64, 90, 91, and 137%, respectively. Columns, mean (n = 8 to 10 tumors per group); bars, SE. Dotted line, 50% TGI. *, significance (P < 0.01) compared with vehicle-treated tumors. B, RT-PCR validation of LRRC19 and IGFBP2 expression in eight xenografts in the validation set predicted by the K-TSP classifier. Relative expression of LRRC19 to IGFBP2 correlates with the microarray data.

Fig. 5.

Gene pair identified by K-TSP algorithm is helpful in predicting AZD0530 sensitivity. A, prospective validation of efficacy in eight xenografts predicted by the K-TSP classifier. The eight xenografts predicted for AZD0530 sensitivity were treated with AZD0530 50 mg/kg/d by oral gavage for 28 d, and TGI were calculated. Three xenografts predicted sensitive to AZD0530 (Panc194, JH131, and Panc159) had TGI of 42%, 44%, and 71%, respectively. Five cases predicted to be resistant to AZD0530 (Panc294, Panc140, JH069, A6L, and JH010) had TGI of 61, 64, 90, 91, and 137%, respectively. Columns, mean (n = 8 to 10 tumors per group); bars, SE. Dotted line, 50% TGI. *, significance (P < 0.01) compared with vehicle-treated tumors. B, RT-PCR validation of LRRC19 and IGFBP2 expression in eight xenografts in the validation set predicted by the K-TSP classifier. Relative expression of LRRC19 to IGFBP2 correlates with the microarray data.

Close modal

Translational bioinformatics. We performed reverse transcription-PCR (RT-PCR) on the two genes in the eight independent test set. The RT-PCR results showed the relative expression of LRCC19 and IGFBP2 was the same as the microarray gene expression in all cases tested (Fig. 5B). This highlights two translational values of this work: (a) the K-TSP algorithm identifies two genes from high-dimensional microarray data that are highly accurate in predicting AZD0530 sensitivity; and (b) the relative expression of the two genes can be rapidly tested by RT-PCR in the clinic without costly microarray gene expression profiling.

Gene set analysis reveals the pathways enriched in AZD0530 sensitive cases. Finally, we sought to identify the underlying pathways enriched in AZD0530 sensitive and resistant cases. Gene set enrichment analysis was done on the expression profiles of the xenografts by comparing the 5 sensitive to the 19 resistant cases. The top 10 pathways enriched in the AZD0530-sensitive phenotype were mostly metabolic pathways (Supplementary Table S1A). In contrast, basal cell carcinoma, cell cycle, and hedgehog signaling pathways per KEGG definition were enriched in the AZD0530-resistant phenotype (Supplementary Table S1B).

Predictive biomarkers, which would assist in patient selection for clinical trials, would be of utmost importance in the development of a potentially effective drug. Given that at least four Src inhibitors are in active clinical trials, the ability to predict which patients would benefit most would be a great clinical asset. This is especially true in pancreas cancer, which has an extremely poor prognosis and a paucity of active therapeutic agents. To this end, we evaluated the in vivo efficacy and pharmacodynamic effects of AZD0530, a small molecule Src inhibitor in pancreatic tumor xenografts, and used a novel approach to predict biomarkers for AZD0530 sensitivity in pancreatic cancer.

Src is an integrator of divergent signal transduction pathways that has been implicated in oncogenesis from proliferation to invasion and metastatic spread (32). Observations that Src expression increases with disease progression suggests that Src may be more active in invasion and metastasis than in tumor initiation (2). The contribution of Src in pancreatic tumor progression has been firmly established in an orthotopic implantation model done using human pancreatic tumor cells, in which Src expression was down-regulated by siRNA, resulting in a significant reduction in the incidence of metastases (26).

Given its well-defined role in cancer growth, therapeutic targeting of Src kinase has been an area of intense investigation. Recently, Src inhibition with a Src family kinase inhibitor alone or in combination with gemcitabine, inhibited growth and metastases of orthotopically implanted human pancreatic carcinoma cells and reduced tumor cell proliferation, tumor microvessel density, and increased apoptosis in vivo (6). Evidence from several lines of promising preclinical research has suggested that Src kinase activity influences a plethora of mechanisms that allow tumor cells to proliferate and migrate (33). Consequently, strategies to target Src kinase activity alone and in combination with cytotoxic agents are undergoing investigation in most major tumor types including breast, lung, prostate, pancreatic, and colorectal carcinomas (34).

The present preclinical study used large number pancreatic tumor xenografts to test the effect of Src inhibitor in pancreatic cancer. In this study, we show, for the first time, that the gene expression profiles of baseline tumors could be used to predict AZD0530 sensitivity in pancreatic tumors. Here, we used the K-TSP algorithm, a novel machine learning approach that used a small number of genes in its decision rules. The classification is based on comparing the relative expression of gene pairs, which is relatively easy to interpret and facilitates translational work as shown by RT-PCR validation in this study (Fig. 5B) and others (35).

The genes in the K-TSP predictor are the leucine rich repeat containing protein 19 (LRRC19) and insulin-like growth factor binding protein 2 (IGFBP2; Fig. 4A). Global analysis also identified LRRC19 and IGFBP2 in the top 100 differentially expressed genes. These two genes have been implicated to play essential roles in the progression of various cancer types (36, 37). LRRC19 encodes a leucine rich surface protein where up-regulation is associated with hypoxia (36). Recently, using a glial-specific transgenic mouse model, IGFBP2 has been implicated to play a key role in promoting the development and progression of glioma. Most importantly, activation of two separate pathways (combined expression of IGFBP2 or Akt with K-ras) is necessary for glioma tumorigenesis (37). Overexpression of IGFBP2 may indicate activation of the cell cycle via the Akt pathway, and may make the tumor less responsive to AZD0530. This complex interaction between IGFBP2 and the cell cycle was revealed by the two-gene predictor and gene set analysis. This indicates that interpretation of a predictor also plays an important aspect in building an accurate classifier, which is often neglected by many bioinformatics analyses. From this study, the higher relative expression of LRCC19 to IGFBP2 indicates AZD0530 sensitivity (Fig. 4B and C). This may indicate that the K-TSP algorithm helps to identify key genes that drive important cellular processes in response to AZD0530.

The two-gene predictor achieved 83% overall accuracy in the validation experiments. Panc159, which was predicted sensitive (TGI, <50%) based on the predictor turned out to be resistant (TGI, 71%) in the test validation. Differences in response based on predictive biomarkers among patient tumors may be attributed to a wide variety of tumor factors including the sequestration or efflux of drugs, down-regulation of apoptotic pathways that likely act downstream of drug targets and the stimulation of growth/survival pathways (38). In addition, continued exposure of tumor cells to chemotherapy agents can often result in selective amplification of specific genes whose products promote drug resistance (38). The present study used a small sample size for test and prediction. Significantly larger sample sizes will be required to further validate the prediction before translating the prediction to select patients in the clinic.

Western blots of sensitive tumor xenografts indicated that AZD0530 led to significant down-regulation of FAK and p-FAK (Fig. 2). It has been reported that Src and FAK cross-activate, and enhanced migratory activity is linked to increased FAK expression and activation (39). Src activates FAK and STAT-3 and their linked activities act to control cell migration through the turnover of focal adhesions and the suppression of cell-cell contacts (810). p-STAT-3 inhibition observed in this study with AZD0530 is also consistent with the p-STAT-3 inhibition reported with other Src kinase inhibitors (40). Interestingly, it has recently been shown that Src-dependent induction of vascular endothelial growth factor expression is mediated via STAT-3 (41). AZD0530 treatment down-regulated XIAP, a member of the apoptosis inhibitor family (Fig. 2). These results are consistent with the reported results of AZD0530 in an ovarian tumor model, where AZD0530 treatment inhibited the expression of survivin, another member of the apoptosis inhibitor family (42).

Immunohistochemistry results clearly show a treatment effect for AZD0530 on the Src substrates key to the invasive phenotype. There were significant decrease in the posttreatment levels of expression of both p-FAK and p-paxillin compared with controls for both sensitive and resistant tumors. This work in pancreatic explants agrees with and reinforces the clinical work in man with the same biomarkers (43). There was a very clear treatment effect of AZD0530 on both p-FAK and p-paxillin with high values of the test statistics. These data provide evidence supportive of earlier clinical results in man demonstrating the effect of AZD0530 on these biomarkers (43).

In summary, we show that the Src inhibitor AZD0530 displays preclinical activity in a subset of pancreatic tumor xenografts. Antiproliferative activity of AZD0530 results in the specific down-regulation of Src, FAK, XIAP, p-FAK, p-paxillin, and p-STAT-3. Finally, our study identified one gene pair (LRRC19 and IGFBP2) that has a high predictive power for AZD0530 sensitivity and that may have potential as a biomarker for predicting pancreatic tumor sensitivity to AZD0530 in the clinic.

M. Hidalgo, commercial research support, Astra Zeneca. C. Womack, H. Wombwell, S. Morgan, J. Walker, and T.P. Green, employees, Astra Zeneca. The other authors declare no potential conflicts of interest.

Grant support: AstraZeneca, Alderley Park, Macclesfield, UK to Manuel Hidalgo. Presented in part at European Organization for Research and Treatment of Cancer-National Cancer Institute-AACR Molecular Targets and Cancer Therapeutics symposium, October 21-24, 2008, Geneva, Switzerland (Abstract # 77).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

N.V. Rajeshkumar and A.C. Tan contributed equally to this work.

1
Yeatman TJ. A renaissance for SRC.
Nat Rev Cancer
2004
;
4
:
470
–80.
2
Summy JM, Gallick GE. Src family kinases in tumor progression and metastasis.
Cancer Metastasis Rev
2003
;
22
:
337
–58.
3
Warmuth M, Damoiseaux R, Liu Y, Fabbro D, Gray N. SRC family kinases: potential targets for the treatment of human cancer and leukemia.
Curr Pharm Des
2003
;
9
:
2043
–59.
4
Frame MC. Src in cancer: deregulation and consequences for cell behaviour.
Biochim Biophys Acta
2002
;
1602
:
114
–30.
5
Ito H, Gardner-Thorpe J, Zinner MJ, Ashley SW, Whang EE. Inhibition of tyrosine kinase Src suppresses pancreatic cancer invasiveness.
Surgery
2003
;
134
:
221
–6.
6
Yezhelyev MV, Koehl G, Guba M, et al. Inhibition of SRC tyrosine kinase as treatment for human pancreatic cancer growing orthotopically in nude mice.
Clin Cancer Res
2004
;
10
:
8028
–36.
7
Irby RB, Yeatman TJ. Role of Src expression and activation in human cancer.
Oncogene
2000
;
19
:
5636
–42.
8
Avizienyte E, Frame MC. Src and FAK signalling controls adhesion fate and the epithelial-to-mesenchymal transition.
Curr Opin Cell Biol
2005
;
17
:
542
–7.
9
Hiscox S, Jordan NJ, Morgan L, Green TP, Nicholson RI. Src kinase promotes adhesion-independent activation of FAK and enhances cellular migration in tamoxifen-resistant breast cancer cells.
Clin Exp Metastasis
2007
;
24
:
157
–67.
10
Yu CL, Meyer DJ, Campbell GS, et al. Enhanced DNA-binding activity of a Stat3-related protein in cells transformed by the Src oncoprotein.
Science
1995
;
269
:
81
–3.
11
Scaltriti M, Baselga J. The epidermal growth factor receptor pathway: a model for targeted therapy.
Clin Cancer Res
2006
;
12
:
5268
–72.
12
Talamonti MS, Roh MS, Curley SA, Gallick GE. Increase in activity and level of pp60c-src in progressive stages of human colorectal cancer.
J Clin Invest
1993
;
91
:
53
–60.
13
Lee D, Gautschi O. Clinical development of SRC tyrosine kinase inhibitors in lung cancer.
Clin Lung Cancer
2006
;
7
:
381
–4.
14
Golas JM, Arndt K, Etienne C, et al. SKI-606, a 4-anilino-3-quinolinecarbonitrile dual inhibitor of Src and Abl kinases, is a potent antiproliferative agent against chronic myelogenous leukemia cells in culture and causes regression of K562 xenografts in nude mice.
Cancer Res
2003
;
63
:
375
–81.
15
Ottmann O, Dombret H, Martinelli G, et al. Dasatinib induces rapid hematologic and cytogenetic responses in adult patients with Philadelphia chromosome positive acute lymphoblastic leukemia with resistance or intolerance to imatinib: interim results of a phase 2 study.
Blood
2007
;
110
:
2309
–15.
16
Jemal A, Siegel R, Ward E, et al. Cancer statistics, 2008.
CA Cancer J Clin
2008
;
58
:
71
–96.
17
Rosenberg L. Pancreatic cancer: a review of emerging therapies.
Drugs
2000
;
59
:
1071
–89.
18
Burris HA III, Moore MJ, Andersen J, et al. Improvements in survival and clinical benefit with gemcitabine as first-line therapy for patients with advanced pancreas cancer: a randomized trial.
J Clin Oncol
1997
;
15
:
2403
–13.
19
Moore MJ. Brief communication: a new combination in the treatment of advanced pancreatic cancer.
Semin Oncol
2005
;
32
:
5
–6.
20
Coppola D. Molecular prognostic markers in pancreatic cancer.
Cancer Control
2000
;
7
:
421
–7.
21
Lutz MP, Esser IB, Flossmann-Kast BB, et al. Overexpression and activation of the tyrosine kinase Src in human pancreatic carcinoma.
Biochem Biophys Res Commun
1998
;
243
:
503
–8.
22
Chang YM, Bai L, Liu S, Yang JC, Kung HJ, Evans CP. Src family kinase oncogenic potential and pathways in prostate cancer as revealed by AZD0530.
Oncogene
2008
;
27
:
6365
–75.
23
Rubio-Viqueira B, Jimeno A, Cusatis G, et al. An in vivo platform for translational drug development in pancreatic cancer.
Clin Cancer Res
2006
;
12
:
4652
–61.
24
Bankert RB, Egilmez NK, Hess SD. Human-SCID mouse chimeric models for the evaluation of anti-cancer therapies.
Trends Immunol
2001
;
22
:
386
–93.
25
Perez-Soler R, Kemp B, Wu QP, et al. Response and determinants of sensitivity to paclitaxel in human non-small cell lung cancer tumors heterotransplanted in nude mice.
Clin Cancer Res
2000
;
6
:
4932
–8.
26
Rubio-Viqueira B, Hidalgo M. Direct in vivo xenograft tumor model for predicting chemotherapeutic drug response in cancer patients.
Clin Pharmacol Ther
2009
;
85
:
217
–21.
27
Song D, Chaerkady R, Tan AC, et al. Antitumor activity and molecular effects of the novel heat shock protein 90 inhibitor, IPI-504, in pancreatic cancer.
Mol Cancer Ther
2008
;
7
:
3275
–84.
28
Trevino JG, Summy JM, Lesslie DP, et al. Inhibition of SRC expression and activity inhibits tumor progression and metastasis of human pancreatic adenocarcinoma cells in an orthotopic nude mouse model.
Am J Pathol
2006
;
168
:
962
–72.
29
Jallal H, Valentino ML, Chen G, Boschelli F, Ali S, Rabbani SA. A Src/Abl kinase inhibitor, SKI-606, blocks breast cancer invasion, growth, and metastasis in vitro and in vivo.
Cancer Res
2007
;
67
:
1580
–8.
30
Golas JM, Lucas J, Etienne C, et al. SKI-606, a Src/Abl inhibitor with in vivo activity in colon tumor xenograft models.
Cancer Res
2005
;
65
:
5358
–64.
31
Tan AC, Naiman DQ, Xu L, Winslow RL, Geman D. Simple decision rules for classifying human cancers from gene expression profiles.
Bioinformatics
2005
;
21
:
3896
–904.
32
Finn RS. Targeting Src in breast cancer.
Ann Oncol
2008
;
19
:
1379
–86.
33
Boyer B, Bourgeois Y, Poupon MF. Src kinase contributes to the metastatic spread of carcinoma cells.
Oncogene
2002
;
21
:
2347
–56.
34
Kopetz S, Shah AN, Gallick GE. Src continues aging: current and future clinical directions.
Clin Cancer Res
2007
;
13
:
7232
–6.
35
Price ND, Trent J, El-Naggar AK, et al. Highly accurate two-gene classifier for differentiating gastrointestinal stromal tumors and leiomyosarcomas.
Proc Natl Acad Sci U S A
2007
;
104
:
3414
–9.
36
Neri D, Bicknell R. Tumour vascular targeting.
Nat Rev Cancer
2005
;
5
:
436
–46.
37
Dunlap SM, Celestino J, Wang H, et al. Insulin-like growth factor binding protein 2 promotes glioma development and progression.
Proc Natl Acad Sci U S A
2007
;
104
:
11736
–41.
38
Parissenti AM, Hembruff SL, Villeneuve DJ, Veitch Z, Guo B, Eng J. Gene expression profiles as biomarkers for the prediction of chemotherapy drug response in human tumour cells.
Anticancer Drugs
2007
;
18
:
499
–523.
39
Slack JK, Adams RB, Rovin JD, Bissonette EA, Stoker CE, Parsons JT. Alterations in the focal adhesion kinase/Src signal transduction pathway correlate with increased migratory capacity of prostate carcinoma cells.
Oncogene
2001
;
20
:
1152
–63.
40
Johnson FM, Saigal B, Tran H, Donato NJ. Abrogation of signal transducer and activator of transcription 3 reactivation after Src kinase inhibition results in synergistic antitumor effects.
Clin Cancer Res
2007
;
13
:
4233
–44.
41
Niu G, Wright KL, Huang M, et al. Constitutive Stat3 activity up-regulates VEGF expression and tumor angiogenesis.
Oncogene
2002
;
21
:
2000
–8.
42
Chen T, Wang Z, Green T, Taylor C. Src tyrosine kinase as a therapeutic target in ovarian cancer and multidrug-resistant ovarian cancer. In: AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2007 Oct 22–26 San Francisco, CA; 2007.
43
Tabernero J, Cervantes A, Hoekman K, et al. Phase I study of AZD0530, an oral potent inhibitor of Src kinase: First demonstration of inhibition of Src activity in human cancers. In: ASCO Annual Meeting Proceedings; 2007 June 1st - 5th.; Chicago, Illinois; 2007. p. 3520.

Supplementary data