|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2005.03.3399 on March 20 2006 © 2006 American Society of Clinical Oncology. Tumor Transcriptome Reveals the Predictive and Prognostic Impact of Lysosomal Protease Inhibitors in Non–Small-Cell Lung Cancer
From the Oncology Research Group, Department of Medicine and Therapeutics, Institute of Medical Sciences, School of Medicine, University of Aberdeen; Department of Oncology, ANCHOR Unit, Aberdeen Royal Infirmary; Department of Pathology, University of Aberdeen, Aberdeen, United Kingdom. Address reprint requests to Elaina Collie-Duguid BSc (Hons), PhD, or Russell Petty, BMSc (Hons), MBCh, BMRCF (UK), Department of Medicine and Therapeutics, IMS, University of Aberdeen, Aberdeen, AB25 2ZD, United Kingdom; e-mail: e.collie-duguid{at}abdn.ac.uk or r.d.petty{at}abdn.ac.uk
PURPOSE: Insight into clinical response to platinum-based chemotherapy (PBC) in non–small-cell lung cancer (NSCLC). METHODS: Matched tumor and nontumor lung tissues from PBC-treated NSCLC patients (four nonresponders and four responders) and tumor tissue from an independent test set (four nonresponders and four responders), were profiled using microarrays. Lysosomal protease inhibitors SerpinB3 and cystatin C were highly correlated with clinical response and were further evaluated by immunohistochemistry in PBC-treated patients (36 prechemotherapy and 13 postchemotherapy). Investigation of the pathogenic and prognostic significance of SerpinB3 was performed in 251 primary tumors, with 64 regional lymph node pairs, from chemotherapy-naïve NSCLC patients using immunohistochemistry. RESULTS: Bioinformatic analyses of gene expression in the training set identified a gene set (n = 17) that separated all patients in the training and test sets (n = 16) according to response in hierarchical clustering. Transcriptome profiling revealed that SerpinB3 mRNA was highly correlated with degree of response (r = –0.978; P < .0001) and was a clear outlier (nonresponders:responders > 50-fold). SerpinB3 protein expression was correlated with clinical response in PBC-treated NSCLC patients (P = .045). Expression of SerpinB3 and cystatin C, relative to the target, protease cathepsin B, was independently predictive of response (odds ratio, 17.8; 95% CI, 2.0 to 162.4; P = .01), with an accuracy of 72%. High SerpinB3 expression levels, invariably associated with chemoresistance, had contrasting prognostic impact in untreated squamous cell carcinomas (hazard ratio [HR], 0.43; 95% CI, 0.18 to 0.93) or adenocarcinomas (HR, 2.09; 95% CI, 1.03 to 4.72). CONCLUSION: This provides the first comprehensive molecular characterization of clinical responsiveness to PBC in NSCLC and reveals the predictive and prognostic impact of two lysosomal protease inhibitors, potentially representing novel targets for NSCLC therapeutics.
Non–small-cell lung cancer (NSCLC) is the most common cause of premature death as a result of cancer.1 Improved understanding of the precise molecular mechanisms that underlie NSCLC biology and determine clinical outcomes will facilitate improved therapeutic approaches.2 The diversity of individual clinical response to therapy is readily evident, and the importance of unraveling the elaborate, complex and heterogeneous molecular networks behind this response is clearly apparent.2-7 This would provide further insight into both the biologic features of the disease that may be successfully exploited therapeutically and may allow prediction of response, thereby facilitating a paradigm shift towards individualized therapy. Global transcriptome profiling provides a means of addressing the complexity and heterogeneity of the molecular mechanisms governing tumorigenesis and underlying individual clinical behavior. This technology has been used successfully in NSCLC and other malignancies to stratify each disease into new molecular subgroups, providing novel insight into the mechanisms of disease etiology and pathogenesis,2,8-13 and identifying prognostic molecular signatures that outperform current clinicopathologic methods for a variety of tumor types.9,14-17 Other, mostly preclinical, studies have provided novel insight into molecular mechanisms of response or resistance to cytotoxic agents.18-21 In this study (AMLUCS-1; Aberdeen Microarray in Lung Cancer Study-1), we hypothesized that global transcriptome profiling would identify important biologic differences between cytotoxic chemotherapy responding and nonresponding NSCLC. Analysis of gene expression profiles has identified a predictive molecular signature for individual patient response. Novel predictive and prognostic biomarkers have been identified that may have a role in the pathogenesis and therapeutic responsiveness of NSCLC and represent putative novel therapeutic targets for this disease.
Patients The study was performed within the guidelines and with approval of the regional research ethics committee. All patients presented and were treated within the Departments of Oncology or Cardiothoracic Surgery at Aberdeen Royal Infirmary, and full clinicopathological details are provided. Snap-frozen tissues, collected at surgical resection between April 1996 and March 2003, were utilized in the microarray studies. All NSCLC patients who had received neoadjuvant platinum-based chemothearpy (PBC) and for whom we had snap-frozen tumor and matched adjacent nontumor tissues of suitable quality were used in the training set for microarray analysis (Fig 1A). All NSCLC patients who received PBC (either neoadjuvant [early-stage patients] or at time of relapse [advanced-stage patients]), with only snap-frozen tumor tissue available, were used in the test set (Fig 1B). Criteria for inclusion of PBC-treated NSCLC patients in the IHC study are provided in Tables 1 and 2 (immunohistochemistry). For immunohistochemical analysis of response, tumor tissues were collected as part of routine diagnosis or surgical resection between April 1996 and March 2003. All patients for whom sufficient tumor sections were available for analysis of all three proteins were included in the study.
Preoperative staging was with mediastinoscopy and computed tomography (CT) scan of the chest and upper abdomen; in advanced disease, staging was performed with CT scan of the chest and upper abdomen (as well as CT head and isotope bone scan as directed by clinical features). All stage information is clinical stage (unless otherwise indicated) according to International Union Against Cancer TMN Classification of Malignant Tumors.22 Response to chemotherapy was assessed according to Response Evaluation Criteria in Solid Tumors (RECIST) criteria23 (ie, complete response [CR], disappearance of all target lesions; partial response [PR], 30% decrease in sum of longest diameter of target lesions; progressive disease [PD], 20% increase in sum of longest diameter of target lesions; stable disease [SD], small changes that do not meet preceding criteria; CR and PR are responders and PD and SD are nonresponders).
Gene Expression Profiling Extraction and purification of total RNA was performed using TRIZOL (Invitrogen, Carlsbad, CA) and RNeasy Minikits (Qiagen, Venlo, the Netherlands), respectively, according to the manufacturer's instructions. Reverse transcription of cDNA from 8 µg of total RNA (Superscript II Kit, Invitrogen) and synthesis and amplification of biotin-labeled cRNA by in vitro transcription (RNA Transcript labeling kit, ENZO Diagnostics, Farmingdale, NY) was performed according to the manufacturer's instructions and standard protocols as provided by Affymetrix (Santa Clara, CA). Quantification of total RNA and biotin-cRNA was performed by spectrophometry and the 260:280 ratio was between 1.9 and 2.2 for all samples. A Bioanalyser 2100 (Agilent Technologies, Palo Alto, CA) and Test 3 GeneChips (Affymetrix, Santa Clara, CA) were used to assess total RNA (28S:18S > 1.5) and cRNA (Actin 3':M < 3 and GAPDH 3':5' < 3) quality, respectively. Biotin-labeled cRNA was fragmented and hybridized HGU133A GeneChips (Affymetrix; 10 µg) for gene expression analysis, according to standard protocols. Initial quality control analysis and normalization of data were performed using Affymetrix Microarray Suite (MAS) version 5.0 software. Subsequent filtering of data was performed using Affymetrix MicroDB version 5.0, Affymetrix Data Mining Tool (DMT) version 3.0, and additional threshold and probabilistic filtering, supervised analysis using gene ontologies, hierarchical cluster analysis and leave-one-out cross validation using Fisher's exact t test hypergeometric probability; K nearest neighbors was performed using GeneSpring version 6.1 (Silicon Genetics, Redwood City, CA). NetAffx Analysis Center (http://www.affymetrix.com/analysis/index.affx; Affymetrix) was used in supervised analysis of the data. Gene expression signals were normalized using scaling of all probe sets to an arbitrary target signal of 100 (MAS version 5.0). These data were utilized for quality control and generation of "detection call" and "change call" gene lists (MicroDB version 3.0 and DMT version 3.0). Further details of these algorithms have been described previously24 and are available from Affymetrix. Additionally, the signal was transformed, and per chip and per gene normalization steps were performed in GeneSpring version 6.1 before detailed data analysis: Signals less than 0.01 were transformed to 0.01 to allow more efficient analysis of log-transformed data; per chip, each measurement on the array was normalized to the 50th percentile of all measurements on the array; and each gene was normalized to its median value across all arrays in the experiment to compare the relative gene expression changes of each gene in different samples. In tumor-nontumor comparisons, normalized gene expression data for individual tumors was expressed relative to the normalized gene expression in the matched uninvolved healthy lung control.
Semiquantitative Reverse Transcriptase Polymerase Chain Reaction Validation of Profiling Data Immunohistochemistry. All sections were obtained from archived formalin-fixed, paraffin-embedded tissues from NSCLC patients. All cases were reviewed by a consultant pulmonary pathologist (K.M.K.). In all cases, antigen retrieval was performed by microwaving in 10 mmol/L citrate (pH, 6.0) for 20 minutes. An autostainer (Dakocytomation, Glostrup, Denmark) was used for detection of proteins using either the CSAII detection system (fluroscein-labeled tyramine-amplification; SerpinB3) or Chemate-Envision detection system (Dakocytomation; cystatin C, cathepsin B) according to the manufacturer's instructions. All sections were double scored by two independent investigators (K.M.K., and R.D.P. or P.H.R.), who were blinded to the clinical data. Overall, more than 80% agreement in scoring was observed for each molecule. Scoring discrepancies were resolved by examination of sections at a double-headed microscope. SerpinB3 (1:400) mouse monoclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA) was used. The sections were scored as follows: no staining, 0; more than 0% to 10% positive tumor cells, 1; more than 10% to 50% positive tumor cells, 2; or more than 50% positive tumor cells, 3. All SerpinB3 positive cells demonstrated strong staining, and four clear and distinct patterns of staining were observed, forming the basis of the scoring system (Fig 2). Cystatin C (1:125) rabbit polyclonal antibody (Dakocytomation) and mouse monoclonal cathepsin B antibody (1:200; Abcam, Cambridge, United Kingdom) were used. Because of observed variation in both the number and intensity of positive tumor cells for each protein, sections were scored for percent of positive staining tumor cells (<10%, 0; 10% to 30% positive, 1; 30% to 70% positive, 2; and more than 70% positive, 3) and intensity of staining (none, 0; weak, 1; moderate, 2; or strong, 3). The final score for cystatin C or cathepsin B was obtained by addition of the percent and intensity scores.
Identification of Genes Highly Correlated With Clinical Response to PBC in NSCLC Gene expression levels of more than 22,000 transcripts in a training set of eight patients (four responders and four nonresponders) with resectable NSCLC who underwent three cycles of PBC before surgical resection of their primary tumors, were profiled using Affymetrix HGU133A GeneChip oligonucleotide microarrays (Table 3; Fig 1A). Tumor and adjacent uninvolved lung tissue samples were profiled for each patient (Fig 1A). Bioinformatics analysis (Fig 3) identified a set of genes (n = 17) whose expression was highly correlated with clinical response in the training set. Gene expression levels of this gene set (Table 4; Fig 3) classified all tumors in the training set correctly according to response in hierarchical clustering (Fig 4A) and using leave-one-out cross-validation.
Examination of the normalized expression of the 17 genes revealed a gene set whose expression is highly discriminatory with regard to clinical response (Table 4; Fig 4). Generally, the genes show a four- to 10-fold change in their mean expression between responders and nonresponders (Table 4; Fig 5A). However there is one distinct outlier, encoding the cross-class lysosomal protease inhibitor SerpinB3,25,26 showing a 50-fold change (Table 4; Fig 5A). SerpinB3 gene expression shows a highly significant correlation with the degree of response seen clinically on CT scan (Fig 5B). SerpinB3 has been implicated in cancer cell line studies as a negative regulator of programmed cell death (PCD) in response to both cytotoxics and radiation.27,28
To further confirm the importance of these genes in clinical response determination and to achieve a proof-of-principle that the derived gene set predicts clinical response, gene expression levels in an independent test set of NSCLC primary tumor tissues (four responders and four nonresponders), were profiled (Table 5; Fig 1B). The independent test set was not utilized in generation of the predictive molecular signature, included patients with early-stage disease, who received three cycles of PBC before surgical resection of their tumors, and patients with metastatic disease who received up to four cycles of PBC on relapse (Table 5). Primary tumor obtained at surgical resection was used for profiling (Fig 1B).
All tumors in the independent test set (eight of eight) clustered appropriately according to response in hierarchical clustering using standard correlation of the expression levels of the predictor genes (n = 17) in the eight training samples and each of the independent test samples (Fig 6A -C).
Both clinical and preclinical work have suggested the importance of PCD in the mechanism of action or resistance to cytotoxic drugs.29-31 A global supervised analysis of all cell death pathways and 1,007 probe sets (Fig 7) identified key cell death genes and pathways associated with sensitivity or resistance to PBC in NSCLC (Table 6). These response-associated cell death genes included SerpinB3 (upregulated in nonresponders), which has a previously documented role as a negative regulator of cell death,27,28 and another cross-class lysosomal protease inhibitor, cystatin C (CST3; downregulated in responders), an inhibitor of the cysteine protease, cathepsin B, which has a documented role in bid cleavage and cell death.32-34 Gene expression levels of key genes previously reported to play a role in platinum resistance were evaluated, and expression in nonresponding patients was consistent with a platinum-resistant phenotype (Appendix Fig A1, online only).
Semiquantitative real-time polymerase chain reaction (PCR) was performed to validate gene expression over a wide and representative range of raw signals and fold changes measured on the microarrays. Gene expression levels measured using either microarray or real-time PCR analysis were highly correlated (Appendix Fig A2, online only).
Lysosomal Cysteine Protease Inhibitor Protein Expression in PBC-Treated NSCLC Patients Scoring systems were derived for each protein (SerpinB3, cystatin C and cathepsin B), representative of the range of staining patterns seen within the full set of patients (described in the Immunohistochemistry subsection under Semiquantitative Reverse Transcriptase Polymerase Chain Reaction Validation of Profiling Data in Methods). The scoring system was designed to reflect protein levels within tumor cells. Staining for SerpinB3 was exclusively within the cytoplasm of tumor cells (Fig 2). Staining for cystatin C and cathepsin B was within the cytoplasm and on the membrane of tumor cells, but staining within the tumor stroma was also observed (data not shown). Scoring was representative of staining only within tumor cells. In prechemotherapy tumor biopsies (n = 36), a significant association between SerpinB3 protein expression and response was demonstrated (P = .045; Table 7; Fig 8), and SerpinB3 expression in more than 10% of tumor cells (IHC score > 1) was invariably associated with chemoresistance (Fig 8). In postchemotherapy tumor biopsies (n = 13), a significant association between SerpinB3 protein expression and response was also demonstrated (P = .01; Table 7). A combined IHC score designed to reflect protease inhibitory activity (SerpinB3 + cystatin C/cathepsin B) revealed a highly significant relationship between clinical response to PBC in both prechemotherapy (n = 36) and postchemotherapy (n = 13) specimens (P = .007 and P = .021, respectively; Table 7; Fig 8B).
A statistically significant increase in cathepsin B protein expression was observed after PBC (P = .021), although paired prechemotherapy and postchemotherapy tumor tissues were available for only a small subset of the NSCLC patients (n = 8). No significant difference was observed in SerpinB3 or cystatin C protein expression in these paired samples. In multivariate analysis of the prechemotherapy specimens (n = 36), including the clinical variables sex, age (> or < 70 years), smoking history, weight loss (<or > than 10%), WHO performance status (0 v 1), stage (early v late), histologic type (SCC v AC), histologic grade (poorly v moderately/well-differentiated), and chemotherapy regimen, a combined IHC score (SerpinB3 + cystatin C/cathepsin B) using a threshold cutoff of 2.0 was an independent predictor of response (odds ratio, 17.8; 95% CI, 2.0 to 162.4; P = .01). A combined IHC score of more than 2 was almost invariably associated with resistance to PBC, and provides sensitivity for response of 94%, specificity for nonresponse of 91%, and an overall accuracy of 72% (Fig 8B). Specificity for response (64%) and sensitivity for nonresponse (53%) are limited; however, the specificity for nonresponding patients is high, suggesting that this test could identify 53% of nonresponders, with few responders being missed.
SerpinB3 Protein Expression in Chemotherapy-Naïve NSCLC Patients
SerpinB3 staining was positive more commonly in SCC than AC (P < .001; Table 9). There was no significant association between SerpinB3 protein expression and any clinicopathologic variable (histologic type, International Union Against Cancer stage, grade, sex, smoking history, age).
As described herein, a SerpinB3 IHC score of more than 1 (> 10% of tumor cells positive for SerpinB3; Fig 8) was invariably associated with chemoresistance. Applying this as a cutoff in the chemotherapy-naïve NSCLC patients reveals contrasting prognostic impact of SerpinB3 in SCC and AC (Fig 9A -B).
In AC, high SerpinB3 protein expression (IHC score > 1) is a factor that indicates poor prognosis (Fig 9A). In multivariate analysis with the clinical variables stage, grade, sex, smoking history, and age (> or < 70 years), high SerpinB3 protein expression (IHC score > 1) is an independent prognostic marker for poor 5-year survival in NSCLC patients with AC (Fig 9A). In SCC, a contrasting prognostic impact is evident. High SerpinB3 protein expression in SCC (IHC score > 1) is a factor that indicates a favorable prognosis. In multivariate analysis of SCC with the clinical variables stage, grade, sex, smoking history, age (> or < 70 years), SerpinB3 protein expression is an independent factor for favorable prognosis for 5 year survival (Fig 9B). However, the prognostic impact varies with nodal status, so that in N0 and N1 tumors high SerpinB3 protein expression remains associated with favorable prognosis (Fig 9C), whereas in N2 tumors it is a factor that indicates poor prognosis (Fig 9D). The distinct prognostic impacts of SerpinB3 expression in AC and SCC may be explained by different putative roles for SerpinB3, including negative regulation of both invasion and metastasis,35-37 and cell death.27,28 Consistent with this hypothesis, in the matched primary AC and regional lymph nodes there was no significant change in SerpinB3 expression in tumor cells (Fig 10B), suggesting that SerpinB3 does not inhibit invasion and metastases in AC of the lung. This is in contrast to downregulation of SerpinB3 in metastatic lymph nodes of SCC (P = .003; Fig 10A), which suggests a role in invasion and metastases in this histologic subtype.
To provide new insight into NSCLC treatment, we have performed a global molecular characterization of clinical response to PBC in NSCLC patients, and have identified genes that are strongly associated with clinical response. The importance of key biomarkers has been confirmed in independent sets of patients using gene expression profiling and immunohistochemical analysis of protein expression in PBC-treated NSCLC patients. The correlation of these markers with response to cytotoxic therapy in NSCLC patients persists independent of diverse clinical and pathologic parameters, including pre- and postchemotherapy tissues, different platinum-based regimens, different histologic types of NSCLC, and different clinical stages of disease. Together the data provide an important proof of principle that global gene expression profiling can be used to derive a molecular signature capable of predicting individual patient response to systemic treatment in NSCLC. Additionally, our approach has identified novel molecules and pathways that may be important mechanistic determinants of clinical response or resistance, disease pathogenesis, and therapy-independent prognosis, thereby providing targets for further investigation as novel therapeutics. The cross-class lysosomal protease inhibitor SerpinB3 was identified as a biomarker with both predictive value for response to PBC and also independent prognostic value in untreated patients with resected NSCLC. Gene and protein expression levels for SerpinB3 each demonstrated a strong correlation with clinical response in PBC-treated NSCLC patients. Immunohistochemical measurement of protein expression levels of SerpinB3 with another lysosomal protease inhibitor identified in this study, cystatin C, and its main physiological target, cathepsin B, which has a documented role in PCD,32-34 provides an independent predictor of response to PBC (hazard ratio, 17.8; 95% CI, 2.0 to 162.4; P = .01). This assay provides an accuracy of 72% and suggests that 53% of patients who are unlikely to benefit from PBC may be prospectively identified, while missing very few patients who may benefit by maintaining a high sensitivity for responding patients (94%). Potentially, this could allow first-line treatment with platinum-independent or novel therapies, although this would need further evaluation in a larger cohort. The highly accurate performance of the 17-gene predictive set (Table 4) with the independent test set of NSCLC patients (Fig 6) illustrates that sensitivity for nonresponding patients may be improved by incorporation of further markers, but additionally demonstrates the undoubtedly multifaceted nature of clinical chemoresistance in NSCLC. Nevertheless, the importance of these two lysosomal cysteine cathepsin protease inhibitors is suggested and represents the first report of a putative role of lysosomal proteases and their inhibitors in response or resistance to cytotoxic therapy in NSCLC patients, and suggests a role of for the recently described pathway of lysosomal cathepsin protease-mediated cell death. The likely physiological role of SerpinB3 is to protect against leaked lysosomal proteases, and it has also been shown to be a negative regulator of PCD in tumor cell lines in response to cytotoxic drugs and ionising radiation.27,28 The role of SerpinB3 in cancer pathogenesis is not fully defined, but it was initially described as a serum marker for SCC of the cervix38 and overexpression has also been demonstrated in SCC of lung and head and neck, and hepatocellular carcinoma.38-41 In primary head and neck tumors, SerpinB3 expression is a factor that indicates poor prognosis.40 In addition to SerpinB3's role in negative regulation of cell death,27,28 in vitro studies have suggested a role in inhibition of tumor cell invasion and metastasis, a process in which there is evidence of a role for several lysosomal cathepsin cysteine proteases, including the SerpinB3 substrate cathepsin L.35-37 In NSCLC, there have been many contradictory reports of the diagnostic and prognostic value of SCCA (SerpinB3/SCCA1 and squamous cell carcinoma antigen 2).42-44 Our data demonstrate that intracellular SerpinB3 protein expression in tumor cells has contrasting prognostic value in untreated resected patients with NSCLC of different histologic types (SCC v AC), which may explain the contradictory findings in previous studies.
The patient numbers were too small in this pilot study for a definitive analysis of the impact of the predictive markers on outcome in PBC-treated patients, because the 36 patients from whom tissues were obtained before PBC included patients with different stages of disease. However, analysis of the subset of 25 advanced-stage patients within this set suggested there may be a trend for poorer overall survival (OS) and TTP in patients with high SerpinB3 expression (mean OS, 14.5 v 13.1 months; mean TTP, 10.2 v 8.3 months; Cathepsin B has been reported to have prognostic value in NSCLC patients,45 which may be related to its role in invasion and metastases. Although cathepsin B protein expression alone was not correlated with response in our study, analysis of this protease with both its inhibitor, cystatin C, and SerpinB3 was strongly correlated with response to PBC in NSCLC patients, suggesting that it may also have a role in cytotoxicity of PBC in NSCLC patients. Cathepsin B can mediate caspase-independent cell death in response to cytotoxics in NSCLC cell lines.33 SerpinB3 protein was detected in our study in both AC and SCC of the lung. Although SerpinB3 expression is associated with SCC at various sites including lung, this is the first detailed report of SerpinB3 expression in AC. The contrasting prognostic impact of SerpinB3 expression in AC and SCC, may relate to a dual pathogenic role of SerpinB3 in NSCLC of different histologic types, which is consistent with the suggested molecular and cellular functions of SerpinB3 and one of its protease substrates cathepsin L.27,28,35-37 In patients with resected tumors, particularly in the absence of any chemotherapy, the major determinant of long-term cancer-specific survival is expected to be micrometastatic disease, which would account for the association of high SerpinB3 expression as a factor indicating favorable prognosis in N0/N1 SCC through a putative role in negative regulation of invasion and metastases. Similarly, its association with poor prognosis in N2 disease may relate to a predominant role of SerpinB3 as a negative regulator of cell death in these advanced stage NSCLC patients. Our data demonstrate that SerpinB3 expression is reduced in metastatic nodes compared with paired primary SCC (P = .003), supporting a role for SerpinB3 in inhibition of invasion and metastasis. In AC, SerpinB3 expression levels in the nodal tumor cells are not significantly different from that in the paired primary tumor, suggesting that SerpinB3 does not have a role in invasion and metastasis. High SerpinB3 expression is a factor that indicates poor prognosis in AC, and SerpinB3 may potentially function primarily as a negative regulator of cell death in this histologic type. Strong SerpinB3 protein expression invariably associated with resistance to PBC may provide a useful biomarker in the selection of patients who would or would not benefit from adjuvant chemotherapy. For example, strongly SerpinB3-positive stage IB and IIA SCC are likely to be chemotherapy resistant, but our data suggest a favorable prognosis in these early-stage SCC patients after resection in the absence of any therapy. This group is therefore unlikely to benefit from adjuvant chemotherapy, but under current practice, many oncologists would treat all such patients with adjuvant chemotherapy. In contrast, strongly SerpinB3-positive AC and N2 SCC are unlikely to benefit from PBC, but these clinical and histologic subgroups are likely to have a poor prognosis if untreated. Therefore alternative therapeutic approaches would be indicated in these patients, who could accordingly avoid ineffective and toxic PBC. The contrasting prognostic impact of SerpinB3 in SCC and AC of the lung suggests that histologic type must be considered in any future prospectively designed clinical trial utilizing our expression signature or IHC-based assay for treatment decisions in the clinical management of NSCLC patients. Importantly, the data presented here have implications for novel therapeutic approaches in NSCLC treatment. The mechanisms by which these lysosomal proteases and their inhibitors may mediate chemoresistance remain unresolved but there are several possibilities. SerpinB3 is cytosolic and can be secreted, cystatin C is secreted and is functional extracellularly, but can be reinternalized into the endolysosomal compartment and may have as yet uncharacterized intracellular roles.46 Localization of both cathespin B and cystatin C has been reported on the surface of tumor cells and in juxtanuclear vessels.47 Additionally, cystatin C may play an important role in preventing necrotic tumor cell death following the release of lysosomal proteases in chemotherapy-treated NSCLC patients. Necrosis is certainly observed in chemotherapy-treated lung cancers.48 SerpinB3 may prevent PCD occurrence by the recently described lysosomal pathway, in which a triggering event is the release of lysosomal cathepsin cysteine proteases into the cytosol, an event that may be induced by cytotoxic chemotherapy.29,32,49-50 Overall, our study has identified predictive and prognostic biomarkers and has suggested the importance to clinical response of a previously unsuspected pathway and group of proteins. The potential involvement of lysosomal proteases and their inhibitors in clinical response to cytotoxic therapy, cell death, invasion, and metastasis mean that they are intriguing targets for novel therapeutics, particularly SerpinB3. We suggest that these molecules and this pathway warrant further mechanistic and clinical investigation, and may hold promise as a source of novel therapeutic targets for this devastating disease.
Known or Hypothesized Markers of Platinum Action or Resistance We investigated the expression profiles of 39 molecules previously shown to be important for platinum mechanisms of action or resistance in either cell lines or in clinical studies. Where more than one probe set was present, the unique "_at" set has been used, with the exception of MDM2. If there was no "_at" set, then all probe sets for that gene were used. Figure A1 shows the fold change for the mean expression in the training set nonresponders to responders. If a cutoff of 1.5-fold change is used (log2 ± 0.58), then nine genes are identified (red). With the exception of BRCA1 and MDM2, the direction of fold change is consistent with the previously reported roles of these molecules in platinum action or resistance. This identifies the importance of several previously implicated pathways in determination of clinical response to platinum-based chemotherapy in non–small-cell lung cancer in our series, as would be expected. The explanation for BRCA1 (Fig A1, both probe sets show opposite patterns of expression to that expected) may be that the probe sets are "s_at" and "x_at" probe sets that recognize several reported transcripts for the gene, and alternate splicing with functional consequences for cancer pathogenesis has been documented. Accordingly, the results shown in Figure A1 may represent a composite of several alternative transcripts with different functional roles and, therefore, consequences for chemotherapy resistance. Additionally translational regulation and post-translation modification may be of importance. For MDM2, three probe sets are shown in Figure A1; the unique "_at" probe set is not changed more than 1.5-fold, but the two "s_at" probe sets show opposite patterns of expression. The data for all three are shown because of the opposing patterns of expression visible, which was not the case for other examples where there were several probe sets (both "_at" or "_s_at"/"_x _at"; see BRCA1, GGT-1, BAD, and FAS in part B). Similar to that for BRCA1, the explanation may lie in the recognition of alternatively spliced transcripts that would be recognized by the "s_at" probe sets. More than 40 alternatively spliced transcripts of MDM2 expressed in human cancers have been shown, and may have different functional roles and therefore consequences for chemotherapy resistance.
Validation of Gene Expression Changes Using Real-Time PCR Real-time reverse transcriptase PCR was performed using the Opticon system and software and SYBR green fluorecent label, according to manufacturer's recommendations (MJ Research, Watertown, MA; Fig A2). cDNA from clinical specimens was prepared from the total RNA prepared for microarray analysis, using the Superscript-II kit (Invitrogen) according to manufacturer's instructions. cDNA prepared from total RNA extracted from MCF-7 cells was used as a standard and a single batch of cDNA was used in all experiments for this purpose. Each PCR reaction utilized 20 ng of cDNA.
Each tumor and matched nontumor lung sample from all eight patients in the training set utilized in the microarray analyses was included on each 96-well plate, and each sample was analyzed in quadruplicate. On each 96-well plate, the MCF-7 cDNA was used to generate a standard curve from the mean of triplicate wells of each concentration. The optimal melting temperature for specific and efficient amplification was identified using the gradient block on the Opticon system. The specificity of each set of primers was confirmed by running an aliquot of PCR amplified MCF-7 cDNA on a 1% agarose gel before real-time PCR analysis. The software was utilized to calculate threshold cycle values and determine the appropriate melting/read temperature, to enable accurate measurement of specific product not influenced by primer-dimers or nonspecific products. The primers used were as follows: cystatin C (CST3) 5', AACAAAGGCCGCCTGCTGCCTTCTC-3' and 5'-GCAGGGCACAATGACCTTGTCGAAA-3'; TNFRSF21 5', AACTGAGCATTAGAAGGTACATTTG-3' and 5'-TCAATAGGTCCAATCTGCTCTCAAG-3'; TOMM7 5', GCTTTATCCCTCTTGTGATTTACCT-3' and 5'-GTGAAGAGCCTTGTGCCATCCAACT-3'; and GAPDH 5', ACATGGCCTCCAAGGAGTAAGACCC-3' and 5'GGTACTTTATTGATGGTACATGACA-3'.
Although all authors completed the disclosure declaration, the following author or immediate family members indicated a financial interest. No conflict exists for drugs or devices used in a study if they are not being evaluated as part of the investigation. For a detailed description of the disclosure categories, or for more information about ASCO‘s conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.
Dollar Amount Codes (A) < $10,000 (B) $10,000-99,999 (C)
Bioinformatics: Bioinformatics or computational biology is the use of techniques from applied mathematics, informatics, statistics, and computer science to solve biologic problems. Biomarker: A functional biochemical or molecular indicator of a biologic or disease process that has predictive, diagnostic, and/or prognostic utility. Hierarchical clustering: An analytical tool used to find the closest associations among gene profiles and specimens under evaluation. Molecular signature: With the advent of bioinformatics, molecular signatures are a new discipline that uses a variety of approaches to generate a global view of mRNA, protein patterns, and DNA alterations in various cell types. Thus, molecular signatures of disease processes may be seen as distinct from healthy cells, and therapeutic approaches may be tailored on the basis of molecular signature. Microarray: A miniature array of regularly spaced DNA or oligonucleotide sequences printed on a solid support at high density that is used in a hybridization assay. The sequences may be cDNAs or oligonucleotide sequences that are synthesized in situ to make a DNA chip. Transcriptome: The complete expressed product of the entire genome in a particular cell, tissue, or biofluid at a specific point in time.
We thank Diane Stewart and Nicky Fyfe for technical assistance, and Gordon Prescott, BSc (Hons), MSc, Medical Statistics Division, Department of Public Health, University of Aberdeen (United Kingdom) for statistical advice.
R.D.P. is supported by a Clinical Research Fellowship from The Friends of ANCHOR, Aberdeen, United Kingdom. E.C.D. is supported by the Association for International Cancer Research. This work was also supported by grants from National Health Service Grampian University Hospitals Trust Endowments, Tenovus Scotland, and The James Alexander Mearns' Trust. Terms in blue are defined in the glossary, found at the end of this article and online at www.jco.org. Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. International Agency for Research on Cancer: GLOBOCAN 2000: Cancer Incidence, Mortality and Prevelance Worldwide, Version 1.0, IARC Cancerbase No 5 (ed 5). Lyon, France, IARC Press, 2001 2. Petty RD, Nicolson MC, Kerr KM, et al: Gene expression profiling in non-small cell lung cancer: From molecular mechanisms to clinical application. Clin Cancer Res 10:3237-3248, 2004 3. Balmain A, Gray J, Ponder B. The genetics and genomics of cancer. Nat Genet 33:238-244, 2003 (suppl)[Medline] 4. Cree IA, Neale MH, Myatt NE, et al: Heterogeneity of chemosensitivity of metastatic cutaneous melanoma. Anticancer Drugs 10:437-444, 1999[Medline] 5. Mercer SJ, Somers SS, Knight LA, et al: Heterogeneity of chemosensitivity of esophageal and gastric carcinoma. Anticancer Drugs 14:397-403, 2003[CrossRef][Medline] 6. Whitehouse PA, Knight LA, Di Nicolantonio F, et al: Heterogeneity of chemosensitivity of colorectal adenocarcinoma determined by a modified ex vivo ATP-tumor chemosensitivity assay (ATP-TCA). Anticancer Drugs 14:369-375, 2003[CrossRef][Medline] 7. Paik S: Incorporating genomics into the cancer clinical trial process. Semin Oncol 28:305-309, 2001[CrossRef][Medline] 8. Beer DG, Kardia SL, Huang CC, et al: Gene-expression profiles predict survival of patients with lung adenocarcinoma. Nat Med 8:816-824, 2002 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
INTRODUCTION
2 or Fisher's exact test were used for binary categoric variables, Pearson 
) leave-one-out cross-validation based on Fisher's exact t test hypergeometric probability and k nearest-neighbor classification (k = 3) correctly classified all tumors in the training set according to response. MAS v5.0, Affymetrix (Santa Clara, CA) Microarray Suite software version 5.0; MicroDB v5.0, Affymetrix MicroDB software version 5.0; DMT v3.0, Affymetrix Data Mining Tool version 3.0; GeneSpring v6.1, GeneSpring software version 6.1 (Silicon Genetics, Redwood City, CA).
) Using gene ontologies (GeneSpring software version 6.1, Silicon Genetics, Redwood City, CA) and literature searches (PubMed [
2; nonresponder, > 2) provides an accuracy of 72%, with a sensitivity for response of 94%. IHC, immunohistochemistry.
$100,000 (N/R) Not Required