Prediction Analysis Of Microarrays Research Articles

Classification of Hepatotoxicants Using HepG2 Cells: A Proof of Principle Study

With the number of new drug candidates increasing every year, there is a need for high-throughput human toxicity screenings. As the liver is the most important organ in drug metabolism and thus capable of generating relatively high levels of toxic metabolites, it is important to find a reliable strategy to screen for drug-induced hepatotoxicity. Microarray-based transcriptomics is a well-established technique in toxicogenomics research and is an ideal approach to screen for drug-induced injury at an early stage. The aim of this study was to prove the principle of classifying known hepatotoxicants and nonhepatotoxicants using their distinctive gene expression profiles in vitro in HepG2 cells. Furthermore, we undertook to subclassify the hepatotoxic compounds by investigating the subclass of cholestatic compounds. Prediction analysis for microarrays was used for classification of hepatotoxicants and nonhepatotoxicants, which resulted in an accuracy of 92% on the training set and 91% on the validation set, using 36 genes. A second model was set up with the goal of finding classifiers for cholestasis, resulting in 12 genes that appeared capable of correctly classifying 8 of the 9 cholestatic compounds, resulting in an accuracy of 93%. We were able to prove the principle that transcriptomic analyses of HepG2 cells can indeed be used to classify chemical entities for hepatotoxicity. Genes selected for classification of hepatotoxicity and cholestasis indicate that endoplasmic reticulum stress and the unfolded protein response may be important cellular effects of drug-induced liver injury. However, the number of compounds in both the training set and the validation set should be increased to improve the reliability of the prediction.

A 16-Gene Signature Distinguishes Anaplastic Astrocytoma from Glioblastoma

Anaplastic astrocytoma (AA; Grade III) and glioblastoma (GBM; Grade IV) are diffusely infiltrating tumors and are called malignant astrocytomas. The treatment regimen and prognosis are distinctly different between anaplastic astrocytoma and glioblastoma patients. Although histopathology based current grading system is well accepted and largely reproducible, intratumoral histologic variations often lead to difficulties in classification of malignant astrocytoma samples. In order to obtain a more robust molecular classifier, we analysed RT-qPCR expression data of 175 differentially regulated genes across astrocytoma using Prediction Analysis of Microarrays (PAM) and found the most discriminatory 16-gene expression signature for the classification of anaplastic astrocytoma and glioblastoma. The 16-gene signature obtained in the training set was validated in the test set with diagnostic accuracy of 89%. Additionally, validation of the 16-gene signature in multiple independent cohorts revealed that the signature predicted anaplastic astrocytoma and glioblastoma samples with accuracy rates of 99%, 88%, and 92% in TCGA, GSE1993 and GSE4422 datasets, respectively. The protein-protein interaction network and pathway analysis suggested that the 16-genes of the signature identified epithelial-mesenchymal transition (EMT) pathway as the most differentially regulated pathway in glioblastoma compared to anaplastic astrocytoma. In addition to identifying 16 gene classification signature, we also demonstrated that genes involved in epithelial-mesenchymal transition may play an important role in distinguishing glioblastoma from anaplastic astrocytoma.

Research-Based PAM50 Subtype Predictor Identifies Higher Responses and Improved Survival Outcomes in HER2-Positive Breast Cancer in the NOAH Study

We report a retrospective exploratory analysis of the association of the research-based prediction analysis of microarray 50 (PAM50) subtype predictor with pathologic complete response (pCR) and event-free survival (EFS) in women enrolled in the NeOAdjuvant Herceptin (NOAH) trial. Gene expression profiling was performed using RNA from formalin-fixed paraffin-embedded core biopsies from 114 pretreated patients with HER2-positive (HER2(+)) tumors randomized to receive neoadjuvant doxorubicin/paclitaxel (AT) followed by cyclophosphamide/methotrexate/fluorouracil (CMF), or the same regimen in combination with trastuzumab for one year. A control cohort of 42 patients with HER2-negative tumors treated with AT-CMF was also included. The PAM50 subtypes, the PAM50 proliferation score, and the PAM50 risk of relapse score based on subtype (RORS) and subtype and proliferation (RORP) were evaluated. HER2-enriched (HER2-E) tumors predominated within HER2(+) disease, although all PAM50 intrinsic subtypes were identified across the three cohorts. The OR for achieving pCR with trastuzumab-based chemotherapy for HER2(+)/HER2-E and HER2(+)/RORP-high were 5.117 (P = 0.009) and 8.469 (P = 0.025), respectively, compared with chemotherapy only. The pCR rates of HER2(+)/HER2-E and HER2(+)/RORP-high after trastuzumab-based chemotherapy were 52.9% and 75.0%, respectively. A statistically nonsignificant trend was observed for more pronounced survival benefit with trastuzumab in patients with HER2(+)/HER2-E and HER2(+)/RORP-high tumors compared with patients with HER2(+)/non-HER2-E and HER2(+)/non-RORP-high tumors, respectively. As determined by EFS and pCR, patients with HER2(+)/HER2-E tumors, or HER2(+)/RORP-high tumors, benefit substantially from trastuzumab-based chemotherapy. The clinical value of this genomic test within HER2(+) disease warrants further investigation.

Gene expression profiling analysis reveals that DLG3 is down-regulated in glioblastoma

Glioblastoma multiforme (GBM) is the most malignant glioma. In the current study, 149 astrocytoma gene expression datasets were classified by prediction analysis of microarray. Strikingly, disks large homolog 3 (DLG3), a membrane-associated guanylate kinase-family gene, had the highest score in the GBM subset. DLG3 mRNA expression is significantly down-regulated in GBM relative to normal tissue and grade II or grade III astrocytoma according to the results of real-time polymerase chain reaction, and its protein expression shows an obvious difference by immunohistochemistry. Further assays show that DLG3 over-expression induces mitotic cell cycle arrest and apoptosis, and it inhibits proliferation and migration. However, DLG3 over-expression has almost no affect on invasion. The DLG3 protein expression in human brain GBM tissue and its effects on GBM cell invasion were not expected. Our data suggest that DLG3 is down-regulated in this cancer type. To our knowledge, this is the first report to clearly demonstrate the possible involvement of DLG3 in GBM.

Gene expression profiling of colorectal tumors and normal mucosa by microarrays meta-analysis using prediction analysis of microarray, artificial neural network, classification, and regression trees.

Background. Microarray technology shows great potential but previous studies were limited by small number of samples in the colorectal cancer (CRC) research. The aims of this study are to investigate gene expression profile of CRCs by pooling cDNA microarrays using PAM, ANN, and decision trees (CART and C5.0). Methods. Pooled 16 datasets contained 88 normal mucosal tissues and 1186 CRCs. PAM was performed to identify significant expressed genes in CRCs and models of PAM, ANN, CART, and C5.0 were constructed for screening candidate genes via ranking gene order of significances. Results. The first screening identified 55 genes. The test accuracy of each model was over 0.97 averagely. Less than eight genes achieve excellent classification accuracy. Combining the results of four models, we found the top eight differential genes in CRCs; suppressor genes, CA7, SPIB, GUCA2B, AQP8, IL6R and CWH43; oncogenes, SPP1 and TCN1. Genes of higher significances showed lower variation in rank ordering by different methods. Conclusion. We adopted a two-tier genetic screen, which not only reduced the number of candidate genes but also yielded good accuracy (nearly 100%). This method can be applied to future studies. Among the top eight genes, CA7, TCN1, and CWH43 have not been reported to be related to CRC.

Abstract P6-05-02: Molecular genetic grading in breast cancer: Development and validation of a prognostic marker

Abstract Background: Histopathological grading is an important part for the characterization of breast cancer with regard to prognosis and treatment options. 30-60% of breast cancer patients have a grade 2 (G2) malignancy resulting in an intermediate prognosis and the challenge of finding individualized adequate treatment options. Moreover, the lack of objectivity of the grading system is criticized. With the genetic analysis of breast cancer, the deciphering of its genes and gene expression profiling using microarray technology is a new way of classifying breast cancer Patients, Materials, and Methods. Tumor tissue from 229 breast cancer patients collected from January 2009 - June 2010 was examined. A histological grading of the tumors was determined and gene expression profiling was performed by microarray technology (Agilent Technologies). Gene expression data were reduced and a novel molecular grade index (MGI-9) was defined as a G1 versus G3 classifier by SAM (Significance Analysis of Microarray) and PAM (Prediction Analysis of Microarray). The validity of the newly defined MGI-9 as a prognostic factor was tested by applying it to three published data sets with external follow-up information for a total of 760 patients Results: The significance analysis of gene expression profiles of 229 breast cancer patients using SAM showed more than 11,000 significantly differentially expressed genes according to the grading classes G1-2-3. The number of differentially expressed genes was reduced to 228 genes by SAM analysis using only the G1 vs. G3 cases (63 patients). PAM analysis resulted in a more reduced classifier of 44 molecular probes and we could show that a subset of 9 genes gave similar classification results and could reliably used in FFPE Material. The new MGI-9 index was tested in thee external data sets (GSE1456, GSE2034 and GSE3494) and could subdivide the histological G2 group in two molecular subgroups G2A and G2B with significant impact on survival in all test scenarios. Conclusion: The results show that MGI 9-gene molecular grade index is more informative than the conventional histopathological grading system. In order to validate MGI-9 a prospective, randomized study with long-term follow-up for local recurrence rate and distant metastases is urgently needed. Citation Information: Cancer Res 2013;73(24 Suppl): Abstract nr P6-05-02.

Real-time PCR assay based on the differential expression of microRNAs and protein-coding genes for molecular classification of formalin-fixed paraffin embedded medulloblastomas

Medulloblastoma has recently been found to consist of 4 molecularly and clinically distinct subgroups: WNT, Sonce hedgehog (SHH), Group 3, and Group 4. Deregulated microRNA expression is known to contribute to pathogenesis and has been shown to have diagnostic and prognostic potential in the classification of various cancers. Molecular subgrouping and microRNA expression analysis of 44 frozen and 59 formalin-fixed paraffin embedded medulloblastomas from an Indian cohort were carried out by real-time RT-PCR assay. The differential expression of 9 microRNAs in the 4 molecular subgroups was validated in a set of 101 medulloblastomas. The tumors in the WNT subgroup showed significant (P < .0001) overexpression of miR-193a-3p, miR-224, miR-148a, miR-23b, and miR-365. Reliable classification of medulloblastomas into the 4 molecular subgroups was obtained using a set of 12 protein-coding genes and 9 microRNAs as markers in a real-time RT-PCR assay with an accuracy of 97% as judged by the Prediction Analysis of Microarrays. Age at diagnosis, histology, gender-related incidence, and the relative survival rates of the 4 molecular subgroups in the present Indian cohort were found to be similar to those reported for medulloblastomas from the American and European subcontinent. Non-WNT, non-SHH medulloblastomas underexpressing miR-592 or overexpressing miR-182 were found to have significantly inferior survival rates, indicating utility of these miRNAs as markers for risk stratification. The microRNA based real-time PCR assay is rapid, simple, inexpensive, and useful for molecular classification and risk stratification of medulloblastomas, in particular formalin-fixed paraffin embedded tissues, wherein the expression profile of protein-coding genes is often less reliable due to RNA fragmentation.

BCR-ABL1-Like Cases In Pediatric Acute Lymphoblastic Leukemia: A Comparison Between COG/St. Jude and Dutch DCOG Signatures

BackgroundPediatric precursor B-cell acute lymphoblastic leukemia (BCP-ALL) with the BCR-ABL1 translocation forms a small group with a poor prognosis. Approximately 15% of BCP-ALL cases are characterized by a gene expression signature similar to BCR-ABL1-positive disease. This so-called BCR-ABL1-like group shares the poor prognosis of the BCR-ABL1-positive subtype but without presence of a BCR-ABL1 translocation. Using different cohorts and methods, two different BCR-ABL1-like signatures have been identified (Den Boer et al., Lancet Oncol 2009; Mullighan et al., New Engl J Med 2009). Both signatures identify groups of patients with a poor prognosis and with deletions in B-cell development genes (Den Boer et al., Lancet Oncol 2009; Roberts et al., Cancer Cell 2012). The aim of this study was to compare these two signatures and the association to molecular aberrations and clinical outcome in diagnostic samples of the Dutch Childhood Oncology Group (DCOG) and the German Cooperative Study Group for Childhood Acute Lymphoblastic Leukemia (COALL). Patients and methodsThe BCP-ALL study cohort consisted of 452 patients, including 146 patients with B-precursor ALL without known recurrent aberrations (BCR-ABL1 translocation, ETV6-RUNX1 translocation, MLL rearrangements, TCF3 rearrangements, hyperdiploidy). BCR-ABL1-like cases were identified in these 146 B-precursor cases using Prediction Analysis of Microarrays (PAM) with 257 probe sets (Roberts et al., Cancer Cell 2012) and separately by Hierarchical Clustering (HC) with 110 probe sets (Den Boer et al., Lancet Oncol 2009), using Affymetrix U133Plus2 expression data. Frequencies of molecular aberrations detected by multiplex ligation-dependent probe amplification, array-comparative genomic hybridization and Sanger sequencing, including IKZF1 deletions, JAK2 mutations and CRLF2 aberrations, were compared between groups of patient samples classified by the two prediction algorithms using Fisher's Exact test. We analyzed the cumulative incidence of relapse (CIR) with death as a competing event. ResultsIn the DCOG/COALL cohort, which included patients from all ALL risk groups, we identified 33 BCR-ABL1-like cases with the PAM signature, of which 24 (73%) were high-risk according to the NCI-Rome criteria (WBC ≥ 50 cells/nl and/or age ≥ 10 years), and 79 cases with the HC signature, of which 57 (72%) were high-risk. Compared with 59 B-other cases not identified by either of the two BCR-ABL1-like signatures, the most enriched molecular aberrations in the PAM-identified group were IKZF1 deletions (52% vs 19%, p=0.002), high CRLF2 expression (39% vs 8%, p<0.001), intrachromosomal amplification of chromosome 21 (iAMP21; 32% vs 0%, p<0.0001), and RB1 deletions (23% vs 6%, p=0.04). Enriched aberrations in the HC-identified group included IKZF1 deletions (38% vs 19%, p=0.02), dic(9;20) (28% vs 4%, p<0.001), and iAMP21 (15% vs 0%, p=0.002), while BTG1 deletions occurred less frequently (6% vs 17%, p=0.04). Directly comparing PAM and HC-identified BCR-ABL-like cases indicated significantly higher proportions of CRLF2-high (p=0.004) and JAK2 mutations (p=0.02) in the PAM group, and higher dic(9;20) in the HC group (p=0.03). Next, we compared the overlap between cases identified by the two signatures. PAM identified 32% (25/79) of the BCR-ABL1-like cases identified by HC, and vice versa HC identified 76% (25/33) of the cases identified by PAM. The cumulative incidence of relapse at 5-years in the BCR-ABL1-like groups identified by PAM (20%, p=0.04), HC (27%, p<0.0001), or both methods (18%, p=0.15), was worse than in the non-BCR-ABL1-like/BCR-ABL1-negative/MLL wild-type BCP-ALL reference group (8.1%). Cases identified only by HC (31%, p<0.0001) or only by PAM (25%, p=0.05) also had higher 5-year cumulative relapse incidences than the reference group. ConclusionsBoth BCR-ABL1-like expression signatures defined by HC or PAM identified a group of patients with an unfavorable prognosis in the DCOG/COALL study. The two groups largely overlapped, but each signature also identified unique patients not identified by the other signature. Importantly, each of the signatures was associated with poor outcome, while patients identified by both signatures (PAM/HC-double positive) did not have a worse outcome than those identified only by PAM or HC. Since the two signatures are partially complementary, we suggest that both should be used. Disclosures:Evans:St. Jude: In accordance with institutional policy, St. Jude allocates a portion of the income it receives from licensing inventions and tangible research materials to those researchers responsible for creating this intellectual property., In accordance with institutional policy, St. Jude allocates a portion of the income it receives from licensing inventions and tangible research materials to those researchers responsible for creating this intellectual property. Patents & Royalties, Under this policy, I and/or my spouse have in the past received a portion of the income St. Jude receives from licensing patent rights related to TPMT polymorphisms as clinical diagnostics. Other.

Role Of Insulin Like Growth Factor mRNA Binding Protein-3 (IGF2BP3) In Mixed Lineage Leukemia (MLL) Positive B-Cell Lymphomas

Oncogenic transformation of early B-cell progenitors leads to the human disease of B-acute lymphoblastic leukemia or B-ALL, which affects both children and adults. Among the different subtypes of B-ALL, defined by particular cytogenetic anomalies, there are two which are difficult to treat and have a dismal prognosis. These are B-ALL with chromosomal translocation t (9; 22)/BCR-ABL and MLL gene rearrangements, which show distinctive gene expression profiles.Gene expression is now known to be significantly regulated by post-transcriptional mechanisms. These involve RNA binding proteins and microRNAs. Deregulation of microRNAs as well as RNA binding protein expression is associated with numerous cancers. Here, we hypothesized that RNA binding proteins may be important in regulating gene expression in MLL rearranged leukemias.To examine this hypothesis, we undertook a microarray study examining the expression of both protein-coding and non-coding genes in B-ALL, including MLL translocations. A total of 44 samples were used for the microarray. Supervised class prediction was carried out using the R library of prediction analysis for microarrays (PAM). One of the most significantly differentially expressed genes was Insulin Like Growth Factor mRNA Binding Protein-3 (IGF2BP3). The expression of IGF2BP3 was highest in the MLL rearranged B-ALL group. IGF2BP3 is an oncofetal protein known to be highly expressed in a number of epithelial malignancies such as glioblastomas. IGF2BP3 has been known to bind to the 5'-UTR and stabilize mRNAs like CD44 the expression of which correlates with epithelial tumors metastasis. IGF2BP3 has been shown to bind to the Insulin like Growth Factor-2 (IGF-2) mRNA and enhance translation in glioblastomas.We confirmed the expression of IGF2BP3 and CD44 in these 44 tumor samples and 90 other B-cell lymphoma samples by RT-qPCR. This corroborated with our previous data showing that the expression of both these genes is significantly higher in the group with MLL translocations. In the MLL rearranged leukemias, there was a significant correlation between the expression of CD44 and IGF2BP3. Interestingly however, there was no significant difference in the expression of IGF2 mRNA between these different subsets, indicating either that IGF2BP3 might be acting on IGF-2 mRNA at the translational level or that IGF-2 regulation may be cell-type specific.To evaluate whether IGF2BP3 affects the growth of B-ALL cells, we used NALM6, a B-ALL cell line which expresses IGF2BP3. We generated microRNA-155 formatted siRNAs against human IGF2BP3 and subcloned them into pHAGE6 based lentiviral vectors. Our preliminary data demonstrates that these vectors are capable of knocking down IGF2BP3 in the NALM6 cell line. In addition, cells with knockdown showed a dramatic decrease in their growth rates, as measured by the MTS assay.The IGF-2 paracrine signaling system is thought to be important in the maintenance of HSCs as well as in lymphocyte development. We separated different precursors of B-cells (Hardy fractions) from murine bone marrow using FACS and measured the mRNA expression of CD44 and IGF2BP3 in these different subsets. The expression of both these genes correlated well with each other and showed a dynamic expression pattern with the highest expression seen in the Hardy Fraction C (late pro-B cells). This indicates that the IGF2BP3/CD44 axis might play a role in regulating normal B-cell development and this may be dysregulated in MLL-translocated B-ALL.To examine whether IGF2BP3 overexpression causes leukemia, we cloned the murine and human IGF2BP3 coding regions in a murine retroviral expression vector, MIG (MSCV-IGF2BP3-IRES-GFP). Retroviral packaging was done using 293T cell line, virus was collected and used to infect 7Oz/3, a murine pre-B ALL cell line. Western blot and qPCR confirmed overexpression of IGF2BP3. We have infected bone marrow cells from CD 45.2 positive wild type donor mice with the virus and transferred them into irradiated CD 45.1 recipient mice. We have confirmed engraftment in these mice using flowcytometry for CD 45.1/2 and are presently following the mice for the development of leukemia.In summary, IGF2BP3 is dysregulated in MLL-rearranged leukemia, and its knockdown can cause decreased growth rates in B-ALL cell lines. The current study explores whether IGF2BP3 is oncogenic and the mechanisms of action of IGF2BP3 in B-cell development and neoplasia. Disclosures:No relevant conflicts of interest to declare.

Abstract A022: The use of LN status on developing prognostic gene signatures for ER+ breast cancer

Abstract Introduction: Estrogen Receptor (ER) positive Breast Cancers account for approximately 70% of all breast cancers and have a better prognosis than ER- breast cancer. These patients are amenable to endocrine treatment, including tamoxifen, which eliminates recurrence in a large group of patients, but approximately 30% will relapse within 15 years of diagnosis. The most important predictor of recurrence in ER+ breast cancer is lymph node (LN) status. Patients with LN metastases (LN+) have increased risk of systemic recurrence, compared to ER+ patients without LN metastases (LN-). However, it is difficult for clinicians to determine appropriate treatment for ER+ LN+ breast cancer, so this group is generally treated aggressively. Several commercially available molecular signatures have been developed to predict outcome of early stage breast cancers, but none have been exclusively designed for ER+ breast cancer patients, inclusive of lymph node status. Methods: Here, three publicly available datasets (Gene Expression Omnibus, NCBI), consisting of gene expression profiles from primary ER+ breast cancer tumours were used to develop prognostic gene signatures. Patients from these cohorts were treated exclusively with tamoxifen for 5 years and were followed for at least 10 years past diagnosis. Gene expression significantly related to high risk of distant metastasis free survival (DMFS) of patients from our training cohort, at 10 years, was examined using the Prediction Analysis of Microarray (PAM, Stanford) and used to comprise our novel molecular signatures. Three independent signatures were developed using cohorts of patients with LN- disease exclusively, LN+ disease exclusively, or combined lymph node status. The performance of these signatures was evaluated using an independent cohort of patients with either LN- or LN+ disease. We also examined biologically relevant pathways, using Gene Set Enrichment Analysis (GSEA, Broad Institute), to examine whether the heterogeneous nature of ER+ breast cancers can be related to phenotype or outcome. Results: Gene expression and DMFS data from LN-, LN+, or combined patient samples were evaluated to identify sets of genes that predict patient outcome. The LN- signature could accurately predict DMFS of LN- patients from independent cohorts, but was unable to assign LN+ patients to low and high risk of DMFS groups. Similarly, the LN+ signature could accurately predict outcome of LN+ patients, but not LN- patients. Conversely, the combined signature was able to predict DMFS of all patients, regardless of LN status. We further evaluated gene set enrichment and found differences in gene sets associated with LN- and LN+ disease and with different outcomes. Conclusions: This research demonstrates the importance of considering the lymph node status of patients with both developing and employing prognostic gene signatures to predict outcome of early stage ER+ breast cancer patients. Also, it appears that the development of a signature using an exclusive population (i.e. LN-) of patients is not optimal to predict outcome in patients with different pathological parameters. In the future, using a combined gene signature may help direct treatment decisions for patients with early stage ER+ breast cancer. Further, understanding the biological heterogeneity of this disease, through GSEA, may lead to discovery of appropriate therapeutic targets for patients. Citation Format: Jessica G. Cockburn, Robin M. Hallett, John A. Hassell, Anita Bane. The use of LN status on developing prognostic gene signatures for ER+ breast cancer. [abstract]. In: Proceedings of the AACR Special Conference on Advances in Breast Cancer Research: Genetics, Biology, and Clinical Applications; Oct 3-6, 2013; San Diego, CA. Philadelphia (PA): AACR; Mol Cancer Res 2013;11(10 Suppl):Abstract nr A022.

Abstract 2811: A prostate stroma-derived profile is predictive of early relapse and reflects potential mechanisms of aggressive disease.

Abstract The microenvironment of prostate cancer plays a major role in tumor initiation, growth, and metastatic spread. Using RNA expression data from fresh frozen human prostatectomies for which the clinical outcome was followed for over 5 years, we have developed a general approach to identify gene expression changes of the tumor-adjacent stroma that distinguishes aggressive prostate cancer cases from indolent cases. Highly significant gene associations were defined by correlating expression with Disease-free Survival (DFS) and also, separately, by comparing RNA expression between cases that remained disease free for at least 4 years following surgery with those cases that exhibited relapse within 1 year of surgery. A total of 193 highly significant genes were identified, of which 15 were common to the two approaches. These 15 genes were used to develop a PAM (Prediction Analysis for Microarrays)-based algorithm to classify new cases as indolent or aggressive and to provide a probability for each classification. The algorithm was tested on stroma-enriched samples from 47 new cases with known outcome with an accuracy of 87% [1]. A potential functional basis for some the genes associated with aggressive disease was suggested by the observation that p53 expression is decreased in aggressive disease whereas p53-target genes and cell cycle inhibitors including p21, GADD45A, and GADD45B are increased, which is a pattern others have noted as suggestive of senescence [2]. Increased IGFBP6, Tubulin alpha 1a, and Lamin A/C further suggest a senescence-associated secretory phenotype [3]. A potential secretory factor of known significance in the growth of prostate cancer (IGF-1/Somatomedin-C) has elevated RNA levels in the stroma of the same cases. These features argue that an accurate and general prediction of aggressiveness may be derived from the tumor microenvironment and may be composed of genes that have functional roles in bringing about aggressive disease. [1] Jia et al. 2012 PLoS ONE, 7(8):e41371 [2] Liebermann and Hoffman 2008 Journal of Molecular Signaling, 3:15 [3] Coppe et al. 2008 PLoS Biology, 6(12):e301 Citation Format: Zhenyu Jia, Farahnaz Rahmatpanah, Xin Chen, Michael McClelland, Dan Mercola. A prostate stroma-derived profile is predictive of early relapse and reflects potential mechanisms of aggressive disease. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 2811. doi:10.1158/1538-7445.AM2013-2811 Note: This abstract was not presented at the AACR Annual Meeting 2013 because the presenter was unable to attend.

Exploring the effects of seasonality and chemical pollution on the hepatopancreas transcriptome of the Manila clam

The assessment of marine environmental health is a complex but fundamental task both for ecosystem conservation and food safety related to the human consumption of marine products. Manila clams inhabiting the Venice Lagoon constitute an excellent case study for evaluating the effects of complex mixtures of industrial and urban effluents on aquatic organisms. Clams were collected in different seasons at four locations within the Venice Lagoon. The sampling sites were characterized by a range of pollutant concentrations and included Porto Marghera, a highly polluted industrial area where clam harvesting for human consumption is strictly forbidden. Pooled soft tissues were subjected to mass spectroscopy analysis to measure the concentrations of PCDDs/PCDFs/PCBs-DL, PCBs, PBDEs, HCB and PAHs, and pooled digestive gland samples were used for gene expression profiling. While seasonal variation was found to be responsible for the largest proportion of transcriptional changes, significance analysis of microarrays quantitative correlation analysis identified 162 transcripts that were correlated with at least one class of chemicals measured in the samples from the four different sampling sites. Prediction Analysis of Microarrays (PAM) identified a minimal set of seven genes that correctly assigned samples collected in the restricted polluted area (Porto Marghera), independent of the season in which they were collected. An integrated approach combining transcriptomics and chemical analyses of the Manila clam provided a global picture of how Manila clams respond to complex mixtures of xenobiotics and their interplay with other biotic and abiotic factors. We were also able to identify gene expression signatures for different classes of chemicals and a set of robust biomarkers of exposure to these chemicals.

TSG: a new algorithm for binary and multi-class cancer classification and informative genes selection

BackgroundOne of the challenges in classification of cancer tissue samples based on gene expression data is to establish an effective method that can select a parsimonious set of informative genes. The Top Scoring Pair (TSP), k-Top Scoring Pairs (k-TSP), Support Vector Machines (SVM), and prediction analysis of microarrays (PAM) are four popular classifiers that have comparable performance on multiple cancer datasets. SVM and PAM tend to use a large number of genes and TSP, k-TSP always use even number of genes. In addition, the selection of distinct gene pairs in k-TSP simply combined the pairs of top ranking genes without considering the fact that the gene set with best discrimination power may not be the combined pairs. The k-TSP algorithm also needs the user to specify an upper bound for the number of gene pairs. Here we introduce a computational algorithm to address the problems. The algorithm is named Chisquare-statistic-based Top Scoring Genes (Chi-TSG) classifier simplified as TSG.ResultsThe TSG classifier starts with the top two genes and sequentially adds additional gene into the candidate gene set to perform informative gene selection. The algorithm automatically reports the total number of informative genes selected with cross validation. We provide the algorithm for both binary and multi-class cancer classification. The algorithm was applied to 9 binary and 10 multi-class gene expression datasets involving human cancers. The TSG classifier outperforms TSP family classifiers by a big margin in most of the 19 datasets. In addition to improved accuracy, our classifier shares all the advantages of the TSP family classifiers including easy interpretation, invariant to monotone transformation, often selects a small number of informative genes allowing follow-up studies, resistant to sampling variations due to within sample operations.ConclusionsRedefining the scores for gene set and the classification rules in TSP family classifiers by incorporating the sample size information can lead to better selection of informative genes and classification accuracy. The resulting TSG classifier offers a useful tool for cancer classification based on numerical molecular data.

Abstract P4-08-09: Targeting Thyroid Receptor b in Estrogen Receptor Negative Breast Cancer

Abstract Background: The treatment of estrogen receptor (ER)-negative breast cancer (BC) is a major clinical problem due to the lack of useful therapeutic targets. Nuclear receptors (NRs) are potential targets in these patients because they regulate global transcriptional events and many already have agonists/antagonists available. Material and Methods: We used microarray analysis of 227 ER-negative tumors to identify NR targets, and performed hierarchical clustering using 41 NRs. Expressed receptors were scored using prediction analysis of microarrays (PAM) across clustered groups. Cell lines were matched to subtypes using previously described data (Neve et al. 2006). Candidate gene expression levels were confirmed by qRT-PCR using TaqMan probes. pGIPZ lentiviral vectors encoding shRNA were used to knockdownselected candidates. MTT and soft agar assays were used to measure chemosensitivity and growth following treatment with Docetaxel (Doc), Doxorubicin (Dox), or Cisplatin (Cis). Statistical analysis was performed using Red-R. Results: The 41 NRs clustered tumors into 5 groups. For each group we selected genes representing the highest ranked discriminators, and examined their effects in cell lines matching each groups' gene signature. Thyroid hormone receptor b (THRβ) was selected from group V. The expression levels of this receptor were confirmed by qRT-PCR and Western blot analysis. Knockdown of THRβ in ER-negative HCC2185 cells rendered cells more resistant to all chemotherapeutics by using MTT assay. Similar results were confirmed in ER-negative MDA-MB-453 and HCC202 cells. Knockdown of THRβ enhanced colony forming potential in anchorage-independent soft agar assays in MDA-MB-453 and HCC202 cells. Statistical analysis using clinical data from Sabatier et al. (BCRT 2011) showed that patients with low THRb have a worse clinical outcome. In order to translate these findings into the clinic, we treated cells with a specific THRβ agonists, GC-1 and KB-141. GC-1 inhibited cell growth in growth assays, and synergistic effects were observed when cells were treated with GC-1 and Docetaxel in combination. Re-expression of ERα protein was observed in ER-nagative cells lines after treatment with GC-1 and KB141, suggesting that modulation of THRβ may also extend hormonal therapy to this hormonally insensitive group of patients. Conclusion: Clinical targeting of NRs in ER-negative BCs is a novel strategy since receptors can be specifically targeted with ligands. Our data suggest that chemotherapy response in ER-negative patients overexpressing THRβ could be enhanced with a THRβ agonist. Similarly, functional re-activation of ERα by activating THRβ might extend hormonal therapies to these patients as well. Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P4-08-09.

Abstract P2-10-03: A cross-platform comparison of genomic signatures and OncotypeDx score to discover potential prognostic/predictive genes and pathways

Abstract Background: Microarray assessment of breast cancer demonstrates disease subsets among the major breast cancer biologic categories (ER, PR, HER-2) with likely additional prognostic and treatment (predictive) implications. Additional prognostic and predictive biomarkers and optimization of existing genomic platforms are needed to improve personalized breast cancer care. Methods: 76 early-stage ER+ patients with adequate RNA from fresh frozen tumor specimens and a concurrent 21-gene OncotypeDX recurrence score (RS) were identified across independent Duke studies linked to routine prospective breast biospecimen collection. Expression estimates for Affymetrix H133 Plus 2.0 microarrays were reviewed for quality control (Owzar 2008) and normalized across 4 batches with ComBat. Correlations of genomic prognostic signatures to true RS were assessed using Spearman coefficients. Discovery of new gene-level and pathway-level associations to breast cancer prognosis and prediction, based on RS, were corrected for multiple comparisons using the Bonferonni for the family-wise error rate (FWER) or Benjamini-Hochberg methods for the false discovery rate (FDR). Results: A total of 73 samples passed Affymetrix quality control: 32 “low risk”, 32 “intermediate risk”, and 9 “high risk” using standard recurrence score thresholds of &amp;lt;18 and &amp;gt;30. Median patient age is 55 (range 35–86). Two published algorithms to impute the RS from the Affymetrix model by Fan and Haibe-Keins were highly concordant (94%) with each other and showed strong correlation to actual RS (rho = 0.62, p = 5e−9). However, when the RS thresholds were applied to the microarray-based scores, lower agreement was observed (misclassification rate of 57%). Strong correlation was also observed with other breast signatures and RS, including an imputed 70-gene signature of MammaPrint (rho = 0.59, p = 3e−8) and the 50-gene PAM50 risk-of-relapse score (rho = 0.54, p = 8e−7), while poor correlation was seen for the GENIUS prognostic model (rho = 0.06, p = 0.6). Discovery of gene-level associations to RS identified 28 genes, including known cancer-associated genes such as BRCA2 (FWER adj p = 0.002), Cyclin E1 (FWER adj p = 0.015), and CDCA5 (FWER adj p = 0.048). Pathway-level association in KEGG and GO Biologic Processes, identified 24 and 104 categories respectively, including Cell Cycle pathways KEGG:04110 (FDR adj p = 0.002) and GO:0000085 (FDR adj p = 0.0007). Among 22 in-vitro derived oncogenic pathway signatures, significant negative correlation is seen with p53 (rho = −0.56, adj p = 4e−6) and positive correlation with beta-catenin (rho = 0.38, adj p = 0.015). A multi-gene model of association to OncotypeDX RS classification is being developed using Prediction Analysis of Microarrays (PAM). Conclusions: Microarray-based prognostic breast cancer signatures are generally concordant. However, their application across platforms is currently sub-optimal. In the context of OncotypeDX RS, additional key cancer-associated genes and pathways are found to be associated with breast cancer prognosis, potentially providing insight into treatment opportunities for ER+ breast cancer. Validation efforts of these findings in an independent patient cohort are underway. Funding: NCI: RC2CA14041-01 and W81XWH-07-1-0394 Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P2-10-03.

Identification of Candidate Transcriptional Biomarkers Associated with Chronic Graft-Versus-Host Disease Following Allogeneic Hematopoietic Cell Transplantation

Abstract 4190 Background:Chronic graft-versus-host disease (cGvHD) and its treatment result in significant morbidity and mortality, limiting the clinical benefit of allogeneic hematopoietic cell transplantation (HCT). The diagnosis of cGvHD can be challenging especially when diagnostic features are absent or when the clinical features are confined to internal organs (i.e., lungs). A noninvasive, diagnostic and prognostic test is needed to determine which pts have or will develop cGvHD. Objective:To identify candidate blood biomarkers using gene expression profiling of the circulating donor immune cell repertoire following allogeneic HCT and develop a classification rule to distinguish cGvHD from non-GvHD pts. Methods:In this cross sectional discovery phase study we identified 63 pts (median age 50 yrs; range 19–64) who were beyond 100 days from transplantation (median 424 days) with complete (100%) donor CD3+ T cell chimerism. Pts were divided into cGvHD and non-cGvHD groups based on conventional criteria. To adjust for the influence of immune suppression (IS) medication a prospective cohort of 7 new onset cGvHD pts who were free of IS drugs at the time of onset of cGvHD and at the time of blood sample acquisition. Peripheral blood mononuclear cells were obtained from the 70 pts on IRB approval protocols. The RNA was processed on the Agilent whole human genome 44k microarray platform. Bioinformatics programs including AILUN, GeneSpring, SAM, and PAM. Results:Three data sets were constructed to address the comparison of cGvHD pts on IS medication to drug free non-GvHD pts. First a ‘screening set', compared gene expression levels in pts with cGvHD on IS (n=14) to those with cGvHD not yet on IS (n=7) using a two-sample t-test and eliminated more than 27k genes whose t-test statistics were beyond the interval of [-1,1] as differences in expression were possibly related to the presence or absence of these medication. We next applied the Prediction Analysis of Microarray (PAM) method to a ‘training set' to discriminate 21 GvHD pts from 21 non-GvHD using the 16,103 selected probes from the screening set and found a minimum misclassification error of 6 samples was achieved with a probe set that contained 10 genes. The candidate genes represented a compensatory response to cGvHD rather than a pro-inflammatory gene signature. We next evaluated the performance of the classification rule from the ‘training set' to differentiate cGvHD pts from non-GvHD pts in the 14 pt ‘test set' and correctly predicted 3 new cGvHD and all 7 non-GvHD controls. We repeated PAM and adjusted for possible confounding variables, that included length in days of sample acquisition date relative to the transplant date, recipient age, donor gender, disease histology (myeloid or lymphoid malignancy), related or unrelated donor to recipient status, and type of conditioning regimen (full dose or reduced intensity conditioning) and observed a result similar to that from the unadjusted analysis. Specifically, adjusted PAM achieved a minimum cross-validated misclassification error rate of 9 among the entire cohort of 70 samples. Finally, the entire analysis was bootstrapped and cross-validated with similar results obtained; the genes with the top frequencies for discriminating cGvHD were over-expression of IL1R2, ADAMTS2 and TS3, TPST1, SESN1, AREG, IRS2, GPER, BCAT1 and CxCR7. IL1R2 is the decoy receptor for the pro-inflammatory IL1 cytokine, ADAMTS2 and TS3 are pro-collagen modifying enzymes that regulate the clipping of collagen molecules to impart structure to connective tissue, AREG is a key gene that underlies the development of oral lichen planus, and pathway analysis revealed that the combination of the other over-expressed genes converge through IL-4, Il-6 and IL-10 signaling pathways which are considered GvHD blocking signals and controllers of inflammation. Conclusion:In the current discovery phase study we used expression profiling of peripheral blood mononuclear cells combined with biostatical analyses and identified several candidate biomarkers that associated with chronic GvHD. The genes and signaling pathways highlighted in the current analysis suggest that compensatory responses that control inflammation or are involved with profibrotic matrix remodeling represent a dominant GvHD “gene footprint”. The identified genes have biological relevance to chronic GvHD and may further the understanding of its pathogenesis. Disclosures:No relevant conflicts of interest to declare.

CI-FISH, GEP, and SNPs Correlate Genomic Categories with Risk Stratification in Children with T-ALL.

AbstractAbstract 2485T-cell acute lymphoblastic leukemia (T-ALL) is caused by multiple lesions affecting genes involved in cell cycle control, proliferation, survival and differentiation. Type A abnormalities are driving events in the leukemogenic process; they are mutually exclusive genomic rearrangements, mainly translocations, which delineate specific T-ALL subgroups such as the TAL/LMO, TLX1, TLX3, and HOXA (MLL, CALM-AF10, and SET-NUP214). Type B abnormalities, on the other hand, are much greater in number and cooperate with different molecular lesions, as they are widespread in diverse genetic subgroups. They include genomic imbalances, chromosome translocations and mutations.To confront the heterogeneity of T-ALL associated genomic lesions we used CI-FISH, SNP and GEP, explored their utility as potential diagnostic tools in pediatric T-ALL and correlated the genetic lesions at diagnosis with the Minimal Residual Disease (MRD) clinical stratification. We examined 51 children with T-ALL enrolled in Italy into the AIEOP-BFM ALL2000 protocol; MRD was performed by RQ-PCR monitoring of IG and TCR clonal rearrangements at days 33 and 78. Based on tumor load levels at both time points, patients were classified as MRD standard (MRD-SR), intermediate (MRD-IR) or high risk (MRD-HR)(Schrappe M, et al. Blood 2011;118:2077-84). Combined Interphase FISH (CI-FISH) was done using a panel of probes for 42 candidate genes. Cases with TCR-rearrangements were investigated for partners; aneuploidies were confirmed with probes for the centromeric alpha satellite regions. Single Nucleotide Polymorphism (SNP) arrays were done with Whole-Genome Cytogenetic 2.7M array (Affymetrix, Santa Clara, CA, USA) following the manufacturer’s protocol. Data were analyzed by Chromosome Analysis Suite (ChAS) Software (Affymetrix) with hg19 genome built (http://genome.ucsc.edu/) as reference. Copy number (CN) filters were set at ≥ 20 markers and ≥ 30kb size for losses, ≥50kb for gains, and ≥ 2Mb for CNN-LOH. Gene expression profiles (GEP) (Affymetrix HU-133 Plus 2.0 arrays) were obtained following previously described protocols. Data were normalized with Robust Multi-array Avarage algorithm (RMA). Prediction was performed using the method Prediction Analysis of Microarrays (PAM): a threshold for the selection of predictive probe sets was established using cross validation. This procedure was used to predict and compare classification of 34 cases with CI-FISH and 6 samples for which CI-FISH was not available.CI-FISH grouped 85% of patients according to Type A mutations with distribution in each category reflecting the estimated incidence in childhood T-ALL: 37.5% of cases in the TAL/LMO subgroup, 22.5% in the HOXA, 20% in the TLX3, and 5% in the TLX1. GEP class prediction was concordant with the CI-FISH results. PAM analysis identified 15 probe sets, 11 of which are part of the signature used by Homminga et al. (Cancer Cell 2011;19:484–497) to distinguish the three categories LMOTAL, TLX3 and HOX. SNPs complemented the findings by detecting Type B abnormalities in all cases, thus linking Type A and B lesions. CDKN2A/B abnormalities, as previously observed, were prevalently associated with TLX1 and TLX3(∼90% of TLX1/TLX3 positive cases vs 68% of TAL/LMO and 55.5% of HOXA); PTEN/10q23 deletions occurred in the TAL/LMO subgroup; monoallelic RB1 deletions were found only in the TLX3 subgroup. Interestingly, the TLX3 subgroup was also significantly associated with trisomies and/or tetrasomies of chromosomes 1, 7, 8, and 18 (Fisher exact test; p=0.01). Patients’ stratification according to MRD showed most cases with TAL/LMO (15/18, 83.3%) and HOXA (6/9, 66.6%) fell within MRD-HR while TLX3 positive cases were mainly in MRD-SR (4/8) and MRD-IR (3/8). The 2 TLX1 positive cases were respectively assigned to MRD-SR and MRD-IR.CI-FISH and SNP arrays provided us with an accurate and exhaustive genomic diagnosis of pediatric T-ALL. Based on CI-FISH results a successful predictive GEP test was elaborated. In a prospective design, dissection of type A/B abnormalities within MRD categories might represent a refined prognostic stratification at diagnosis for the majority of patients. Disclosures:No relevant conflicts of interest to declare.

Transcriptomic markers meet the real world: finding diagnostic signatures of corticosteroid treatment in commercial beef samples

BackgroundThe use of growth-promoters in beef cattle, despite the EU ban, remains a frequent practice. The use of transcriptomic markers has already proposed to identify indirect evidence of anabolic hormone treatment. So far, such approach has been tested in experimentally treated animals. Here, for the first time commercial samples were analyzed.ResultsQuantitative determination of Dexamethasone (DEX) residues in the urine collected at the slaughterhouse was performed by Liquid Chromatography-Mass Spectrometry (LC-MS). DNA-microarray technology was used to obtain transcriptomic profiles of skeletal muscle in commercial samples and negative controls. LC-MS confirmed the presence of low level of DEX residues in the urine of the commercial samples suspect for histological classification. Principal Component Analysis (PCA) on microarray data identified two clusters of samples. One cluster included negative controls and a subset of commercial samples, while a second cluster included part of the specimens collected at the slaughterhouse together with positives for corticosteroid treatment based on thymus histology and LC-MS. Functional analysis of the differentially expressed genes (3961) between the two groups provided further evidence that animals clustering with positive samples might have been treated with corticosteroids. These suspect samples could be reliably classified with a specific classification tool (Prediction Analysis of Microarray) using just two genes.ConclusionsDespite broad variation observed in gene expression profiles, the present study showed that DNA-microarrays can be used to find transcriptomic signatures of putative anabolic treatments and that gene expression markers could represent a useful screening tool.

352. Gene and MicroRNA Expression Predictive of Tumour Response in Patients Treated with Preoperative Chemoradiotherapy for Rectal Cancer

Background Preoperative chemoradiotherapy (pCRT) followed by surgery is the standard treatment for locally advanced rectal cancer (LARC). However, the response to pCRT is not uniform, and there is no effective method to predict tumour response to pCRT. Identification of patients not responsive to pCRT could avoid useless exposure to radiation or chemotherapy which is associated with adverse effects. Moreover, the identification of pathological complete response could select patients candidated to a more preserving surgery. The aim of this study is to investigate whether gene and micro-RNA (miRNA) expression profiling is associated with rectal cancer response to pCRT. Materials and methods Tissue biopsies were obtained from patients with mid-low LARC, before pCRT. All the patients underwent standard pCRT followed by resection. All surgical specimens underwent standardized histopathological examination. The biopsies with ≥50% of cancer tissue were considered for the experiment. Gene and miRNA expression was analyzed using one color microarrays technique (Agilent®), after RNA isolation. The data were normalized intra- and inter-array, filtered and then clustered. Gene and miRNA expression was compared between responders (R) and non responders (NR) as measured by histopathological tumour regression grade (TRG). Validation of microarrays data was made by quantitative PCR (qPCR). Results Thirty-eight patients, 16 (42%) R (TRG1-2) and 22 (58%) NR (TRG3-5), were considered. Using SAM (Significance Analysis of Microarrays) two class, 256 genes were found differentially expressed between NR and R (188 over- and 68 down-expressed). Performing PAM (Prediction Analysis for Microarray), 12 genes were strongly predictive of tumour response. Using SAM two class, 30 miRNAs were found differentially expressed between NR and R (24 over- and 6 down-expressed). Anti-correlation analyses, using MAGIA (miRNA and genes integrated analysis), revealed the same 8 miRNAs both in NR and R group, except for miR-630, over-expressed only in NR group. ABCC2 gene, miR-7, miR-182, miR-200a, miR-630, miR-638, and miR-1300 were validated by qPCR, confirming the data obtained by microarray analysis. Conclusions Pre-treatment gene and miRNA expression profiling may be helpful to predict response to pCRT in LARC. Further analyses to confirm these findings are required.

Gene expression profiles predict emergence of psychiatric adverse events in HIV/HCV-coinfected patients on interferon-based HCV therapy.

The efficacy of pegylated interferon-α and ribavirin (pegIFN/RBV) in the treatment of Hepatitis C infection is limited by psychiatric adverse effects (IFN-PE). Our study examined the ability of differential gene expression patterns before therapy to predict emergent IFN-PE among 28 HIV/HCV-coinfected patients treated with pegIFN-α2b/RBV. Patients dually infected with HIV and HCV were evaluated at baseline and during treatment by board-certified psychiatrists who classified patients into 2 groups: those who developed IFN-PE and those who did not (IFN-NPE). Gene expression analysis (Affymetrix HG-U133A) was performed using peripheral blood mononuclear cells before and after initiation of treatment. Analysis of Variance, post hoc analysis based on pair-wise comparisons, and functional annotation analysis identified differentially expressed genes within and between groups. Prediction analysis for microarrays was used to test the predictive ability of selected genes. Twenty-four genes (16 upregulated and 8 downregulated) that were differentially expressed at baseline in patients who subsequently developed IFN-PE compared with the IFN-NPE group showed the ability to predict IFN-PE with an accuracy of 82%. In 16 patients with IFN-PE, 135 genes (117 upregulated; 18 downregulated) were significantly modulated after treatment. Of these, 10 genes have already been shown to be associated with neuropsychiatric illnesses and were significantly modulated only in patients who experienced IFN-PE. We describe a novel molecular diagnostic biomarker panel to predict emergent IFN-PE in HIV/HCV-coinfected patients undergoing pegIFN/RBV treatment, which may improve the identification of patients at greatest risk for IFN-PE and suggest candidate therapeutic targets for preventing or treating IFN-PE.