The Wistar Institute of Anatomy and Biology Annual Progress Report: 2011 Nonformula Grant

Transcription

1 The Wistar Institute of Anatomy and Biology Annual Progress Report: 2011 Nonformula Grant Reporting Period July 1, 2012 June 30, 2013 Nonformula Grant Overview The Wistar Institute of Anatomy and Biology received $991,900 in nonformula funds for the grant award period June 1, 2012 through May 31, Accomplishments for the reporting period are described below. Research Project: Project Title and Purpose Diagnostic Markers for Early-stage Lung Cancer in PAXgene Blood Samples This proposal will verify the hypothesis based on preliminary results (published and unpublished) that tumorspecific signatures induced by a cancer in the lung are present in the peripheral blood and can be identified by global gene expression analyses. The commercialization of this technology requires that the samples be collected in a manner that stabilizes the nucleic acids in the samples at the time of the blood draw. A commercially available blood collection system, the PAXgene tube developed by PreAnalytX, meets this requirement. The purpose of the present study is to confirm the presence of the tumor-specific signatures in blood samples collected using PAXgene samples. This simplification of the blood collection will facilitate commercialization of our blood-based diagnostic platforms. Anticipated Duration of Project 6/1/2012 5/31/2014 Project Overview Our novel finding is that mononuclear white blood cells (PBMC) from lung cancer patients contain a tumor-relevant gene signature that accurately predicts early-stage lung cancer. Using microarray data from 267 individuals, we developed: 1) a 29-gene signature that predicts the presence of lung cancer with 86% accuracy when compared to a wide variety of controls with non-malignant, smoking-related lung disease, and 2) a 24-gene signature that distinguishes control nodules from cancers with an apparent specificity of 80% using a subset of the controls (41 controls) that included only those with various types of lung nodules. The product we propose to develop is a gene-expression-based signature derived from an easily obtained peripheral blood sample. In the first application, this blood test would be used to provide additional information as to whether a suspicious lung nodule identified by a spiral Cancer Diagnostics and Therapeutics Page 1

2 computerized tomography (CT) scan is malignant. Because CT scans cannot identify whether a nodule is malignant or benign, the genomic information provided by this test would help to reduce unnecessary surgery or biopsies and their associated expense and risks. We have already demonstrated that our diagnostic system works accurately using RNA from PBMC collected under research conditions under peer review. The next step, the development of a clinically useful test, requires the simplification of the sample collection to one that is appropriate to a variety of clinical settings including clinicians offices and more remote clinical settings. The commercially available blood collection tubes (PAXgene) from PreAnalytiX perfectly meet this requirement. These tubes provide rapid stabilization of RNA in whole blood at the moment of the blood draw so that a reproducible and highly standardized blood collection system can be established. Samples are stable for several days at room temperature and for more than a year at -20 to -80 C. Our preliminary results from microarray studies on 48 cancer and control samples in PAXgene support the viability of this approach. These results support the feasibility of moving our PBMC results to a commercially viable platform. The validation of our preliminary studies using the PAXgene collection system is the object of this proposal. We have proposed to analyze 600 PAXgene samples to test and validate a blood-based approach to develop an accurate lung cancer diagnostic. Specific Aims: Aim 1: Collect patient and control samples in PAXgene from two collection sites. Aim 2: Process 600 PAXgene RNA samples (~300 from cancer patients) and analyze gene expression profiles. Aim 3: Identify the minimal number of genes and mirnas that provide the most accurate classification of cancers. Aim 4: Assess new and old IP with Business Development. Principal Investigator Louise C. Showe, PhD Professor The Wistar Institute of Anatomy and Biology 3601 Spruce Street Philadelphia, PA Other Participating Researchers Michael K. Showe, PhD; Andrei V. Kossenkov, PhD; Qin Lui, MD, PhD; George D. Hobbs, JD employed by The Wistar Institute of Anatomy and Biology Steven M. Albelda, MD employed by the University of Pennsylvania School of Medicine Gerard J. Criner, MD employed by Temple University Thomas L. Bauer, MD employed by the Helen F. Graham Cancer Center at Christiana Care, Newark, DE Cancer Diagnostics and Therapeutics Page 2

3 Expected Research Outcomes and Benefits Lung cancer remains the primary cause of cancer-related deaths. In part, this is due to the lack of viable early detection protocols and the fact that common early symptoms are easily ignored by the individual. We know that a very large (>95%) percentage of lung cancers are the result of smoking, with additional risk factors that include exposure to asbestos, radon, and second-hand smoke. Thus, there is a large and identifiable population at risk for developing lung cancer. Deaths from lung cancer are higher for minorities, higher in men than women, and higher in underserved populations. In a ranking of lung cancer incidence by state into four classes, with the fourth class being the highest incidence class, Pennsylvania s lung cancer rate is in the third class (70 cases/100,000, with a range from 22-98/100,000). Neighboring Delaware, where many Pennsylvania workers live, is in the fourth or highest incidence class. In the rankings for lung cancer deaths, Pennsylvania is in the second class and Delaware is in the fourth. These numbers are striking given the fact that both of these states are located in a corridor of prestigious academic, medical, and biotech organizations. Although the use of CT scans is projected to increase significantly for lung cancer detection, access to this technology will be much less available in rural settings and less likely to be available, and also perhaps less desirable, to underserved and ethnically diverse communities. Blood tests, by contrast, are routine and well-accepted occurrences associated with a visit to a doctor s office. The collection system we propose could also be easily used in community settings where, for example, flu shots are given or blood donations are taken. Such a test can be an early warning for imaging tests that would not otherwise be warranted; or it can be used in combination with imaging results that are inconclusive, essentially providing a second opinion based on an independent assessment. It is clear from years of lung cancer statistics that early detection of lung cancer is the most viable way to prevent lung-cancer-related deaths and to minimize the costs of treatment, the costs to employers, and the tolls on families of the affected individuals. Recent advances in sample collection and storage and more sensitive and reliable platforms for testing multi-gene diagnostic panels suggest that the development of such platforms is now feasible. Summary of Research Completed Aim1: Collect New Samples for Analysis in PAXgene Tubes: We proposed to collect 300 samples from the Helen Graham Cancer Center (HGCC) and Temple University (TU) by 6/30/2013. After some small issues, we established a Standard Operating Procedure (SOP) for sample collection, storage and annotation. The collection of samples from HGCC has been very efficient with new samples being picked up by the Wistar courier on the last Friday of each month. We have collected 136 samples to date. Our sample accrual from TU has been less successful with only ten samples collected during the first year. To meet our projected accrual, we found it necessary to identify another Pennsylvania collection site to replace TU. Dr. Marjorie Clapper, Professor and Co-leader of the Cancer Prevention and Control Program at Fox Chase Cancer Center (FCCC), now a part of TU, will join this study. Dr. Clapper s research focuses on the role of estrogen in smoking-related lung carcinogenesis. Moving from a mouse Cancer Diagnostics and Therapeutics Page 3

4 smoking model to human studies, she has been archiving mononuclear white blood cells (PBMC) from lung cancer and lung nodule patients for her ongoing studies and has agreed to now also collect PAXgene samples from these patients for our biomarker study. FCCC has had a well-organized Biorepository for several years, and Dr. Clapper has agreed to collect 100 samples for this project in the next year. Dr. Clapper s CV is attached, and both IRB and MTA approvals are in progress. The budget has been re-organized to address this change. Dr. Criner has been informed of these problems and is aware that the short duration of this award required samples to be collected in a timely fashion. In addition to the sample collection supported by this grant, we have accrued 133 samples from the University of Pennsylvania (Penn). In addition to the 133 Penn samples (site A), 60 samples were collected at Roswell Park (site B) both supported by an R21 grant (193 samples). However, the site B samples turned out to have significant problems as discussed below and they eventually could not be included in the mrna or mirna expression studies. More than 500 samples have been received from New York University (NYU) to include in this study. The NYU samples were collected as part of a smoker CT screening study, and samples primarily include controls with and without nodules and a few cancers. Dr. Rom has engaged the collaboration of Dr. Harvey Pass to address the collection of additional lung cancer samples at NYU for these studies, and we expect ~100 cancer samples from that site. Although all of the NYU samples cannot be processed within our DOH array budget, they will provide us with a cushion to meet our collection goals. Overall we have met our sample procurement goals for this period (193 from A+B, 146 from HGCC+TU and 500 from NYU). Aim 2: Process 600 PAXgene RNA Samples for Gene and mirna Expression: We proposed to process samples for gene and mirna expression in year 1. To reach this goal in one year, as described in the proposal, we had to be able to process samples collected prior to the start of the grant in addition to the new samples. For our initial study, we first processed 193 previously collected PAXgene samples from sites A and B that included cancers and controls, with essentially equal numbers of cancers and controls. We also collected and processed 146 new samples from HGCC and TU and 36 samples from NYU for a total of 375 samples. All samples were processed by the Wistar genomics facility using highly standardized methods for RNA preparation and microarray processing. RNA was prepared from the PAXgene tubes as recommended to isolate both mrna and mirna species. The collection process is described in Figure 1. The site A and site B samples (193 samples) were processed and analyzed early on as a pilot study as they had already been collected by the start of year 1. Our initial efforts to combine data from A&B clearly identified a problem with the samples from the B site that was evident in our quality control assessments and more evident in effects on classifier performance. There are too many control studies to report here; but, to our best estimation, site B samples had a collection-related problem. Going forward we focused just on the site A samples although this decreased the size of the pilot study. Aim 3: Identify minimal classifiers for cancers vs. all non-cancers and cancer vs. non-malignant nodules: We compared gene expression between 69 cancers and 64 at-risk controls, including 32 controls with high risk lung nodules that warranted further study and were histologically verified as benign. The analysis was carried out using SVM-RFE (Support Vector Machine with Cancer Diagnostics and Therapeutics Page 4

5 recursive feature elimination) as described in detail in the original proposal and in our previous Department of Health (DOH) supported lung cancer publications. The proposal may be viewed at select Search Public Contracts and enter in the Enter specific contract keyword box. We used 80% of the 133 samples (108) for training and gene selection and 20% (25) for testing. The classification based on the SVM scores and a 162 gene signature is shown in Figure 2A, with samples getting a negative score being classified as non-cancer and samples with positive scores being classified as cancer. Sensitivity and specificity values are reported for the training and the test sets (Figure 2B). The cancers are slightly better classified compared to controls (83.9% vs. 78.8%) in the training and 69.2 vs. 91.7% in the test set. If we separate controls into patients without nodules and those with nodules, we see that the large nodules are more difficult to distinguish from the cancers. This is a problem we addressed in our previous studies by analyzing only cancers and nodule controls to identify a nodule specific classifier. We achieved slightly higher accuracy with this approach, but this dataset was too small to make that separation. We will continue to analyze all controls but also analyze different control groups separately to identify the best group-specific classifiers. Although this classifier is larger than our PBMC classifier, we expect this will decrease as we analyze more samples. The new power calculation performed on the 133 samples, shown in Figure 3, estimates that we would need approximately 550 samples to reach our classification goal of 90% accuracy. In addition, based on previous analyses, we estimated that we would need 110 nodule controls to reach our classification goal; and we included only 32 in this study. We expect to surpass the 110 goal in the next six months. We anticipate ~140 additional samples from HGCC in the next six months (15 to 25 samples each month), and ~ 1/3 of these are nodule controls. We also expect 100 samples of patients and controls from Dr. Clapper at FCCC with approximately 30 with nodules, and the NYU samples include many patients with lung nodules. We will also get additional cancer and nodule samples from NYU through Dr. Rom and his associate Dr. Harvey Pass. New MTAs and IRBs are in progress for NYU. Our sample size calculation in our proposal estimated that 500 samples divided between cancers and all controls would be required to achieve a classification accuracy of 90%. We projected to process 600 in the proposal, and we will easily exceed that goal. MiRNA expression studies were also completed on 154 samples from the A and B collection sites as all of the samples from these two sites were processed at the same time. Using the whole data set, we identified a panel of 12 mirnas that classified the training samples from A with a Sensitivity of 71.7%% and a Specificity of 81.3%. When applied to the test set from site B, the sensitivity was similar at 76.7% but the specificity dropped to an unacceptable 23%. Analyses of any combination of sites A and B samples gave similarly poor results confirming our original assessment from our mrna studies that the samples were not collected properly. We then carried out the analysis using only 94 samples from site A using an 80/20 split of the data. While the sensitivity and specificity in the training drops slightly to 67.6% and 71.8% respectively, due to the smaller training set, the specificity in the test set is 77.8% compared to 23% in the A+B analysis and sensitivity is 77.8%. Although our year 1 plans had originally included using a larger number of samples for our pilot study, overall this pilot validates our approach for Cancer Diagnostics and Therapeutics Page 5

6 developing blood based mrna and mirna lung cancer biomarkers using samples collected in PAXgene (Figure 4). As planned, we also tested whether we could combine mrna expression and mirna expression for a more robust classification. Again, using an 80/20 split, we selected a gene classifier of 73 genes and one mirna, which returned accuracies of 82.9 and 83.3% on training and testing sets, respectively. The signature distinguishes cancers and controls with sensitivity and specificity in the training set at 81.1 and 84.6%, respectively, but the test set has a sensitivity and specificity of 77.8 and 88.9%, respectively, with four out of the five controls with nodules being classified correctly. With the additional samples now available for testing, we expect to improve this analysis and reduce the signature. To avoid repetitive reanalysis of the data, the individual and combined gene and mirna analyses will be carried out when arrays for 2/3rds (400) of the projected samples are completed. Aim 4: Patent Application Progress: The PBMC/blood gene expression test system is the subject of PCT/US2008/ No changes to this application or development of new applications have been taken at this time, but will be discussed as the majority of our specimens are analyzed in the next months. PCT/US2009/ is a patent application owned by Wistar and directed to the use of a mirna expression level or an expression level profile of multiple mirnas to access lung disease in a patient, including lung cancer. No licenses or options to this patent application have been granted to third parties. We recently replied to patent office actions on this application. Commercialization possibilities are presently in discussion with Oncocyte Corporation a subsidiary of BioTime. We recently met with Joe Wagner, the CEO of Oncocyte, to discuss their interests in our lung cancer studies and a subsidized research agreement. George Hobbs, our Executive Director of Business Development, and Joe Wagner have successfully worked together in the past. We have recently successfully moved a biomarker panel developed on microarrays to the quantitative PCR OpenArray platform that we suggested to use for validation of this study with excellent validation of the original array data. All Aims: The Scientific Advisory Board (SAB) for this study met on May 29 at HGCC. SAB members Drs. Altieri and Petrelli were present. We presented the progress on our lung cancer study and discussed the shortfall with the sample accrual from TU (described under Aim 1). They agreed with our decision to add an additional site for sample collection at FCCC/ TU. Cancer Diagnostics and Therapeutics Page 6

7 Figure 1: Diagram showing the workflow for processing PAXgene samples for RNA. 2.5 mls are collected and processed for mrna and mirna studies. All of the isolations are carried out under standardized conditions in the Wistar Genomics facility. A 2 Training Set Test Set Svm score B Cancer Controls Figures 2A-2B: Distinguishing cancers and controls. SVM-RFE analysis of 69 cancer patients and 64 controls identified a 162-gene signature that distinguished cancers from controls. (A) shows the SVM scores for each patient in the study. 80% of the samples were used for training and 20% for testing. Each bar represents a single patient. A positive score indicates a patient has cancer and a negative score indicates a control. The height of the column is a measure of how well a sample is classified as one or the other by the 162-gene signature. Samples in the training set are on the left and the test set is on the right. (B) The table shows the Sensitivity and Specificity values for the analysis. Sensitivity decreases somewhat in the test set from 83.9% to 69.2%. Line 2 shows the Specificity for all the controls used for the analysis, and it increases in the test set. Lines 3 and 4 divide the controls by whether or not they were from patients with benign nodules and shows that nodules are somewhat less well classified in this data set. Cancer Diagnostics and Therapeutics Page 7

8 Error rate, e Test Classifier Error rate approximation Error rate approximation, lower MAD Error rate approximation, upper MAD e = n Number of training samples, n Figure 3: Power calculation for nodule classification. The power function was fit by selecting different training sets sizes from the overall data and plotting it against the corresponding error rate of the classification. The relationship between the number of samples used for training and the error rate shows that, by increasing the training set size, we can achieve higher accuracies in the classification of Non-Small Cell Lung Cancer (NSCLC) versus controls with and without nodules. A 90% classification accuracy can be achieved by using a training set containing approximately 550 samples. MAD: median absolute deviation across 50 re-samplings. A 2 Training Set Test Set Svm score B Cancer Controls Figure mirnas Distinguish Cancers and Controls: Training (80%)-Testing (20%). The upper panel shows the SVM scores for the individual samples with cancers indicated by a positive score and controls by a negative score. Accuracy for controls as a whole and separated into those with and without nodules are given. Cancer Diagnostics and Therapeutics Page 8

9 A 2 Training Set Test Set Svm score B Cancer Controls Figure 5. Combining mrna and mirna signatures. This figure shows the classification accuracy using expression data from 73 genes and one mirna. Training and testing results are shown for cancers and all controls and for the two control groups separately Cancer Diagnostics and Therapeutics Page 9