External Validation of a Digital Pathology-based Multimodal Artificial Intelligence Architecture in the NRG/RTOG 9902 Phase 3 Trial.

BACKGROUND: Accurate risk stratification is critical to guide management decisions in localized prostate cancer (PCa). Previously, we had developed and validated a multimodal artificial intelligence (MMAI) model generated from digital histopathology and clinical features. Here, we externally validate this model on men with high-risk or locally advanced PCa treated and followed as part of a phase 3 randomized control trial. OBJECTIVE: To externally validate the MMAI model on men with high-risk or locally advanced PCa treated and followed as part of a phase 3 randomized control trial. DESIGN, SETTING, AND PARTICIPANTS: Our validation cohort included 318 localized high-risk PCa patients from NRG/RTOG 9902 with available histopathology (337 [85%] of the 397 patients enrolled into the trial had available slides, of which 19 [5.6%] failed due to poor image quality). OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Two previously locked prognostic MMAI models were validated for their intended endpoint: distant metastasis (DM) and PCa-specific mortality (PCSM). Individual clinical factors and the number of National Comprehensive Cancer Network (NCCN) high-risk features served as comparators. Subdistribution hazard ratio (sHR) was reported per standard deviation increase of the score with corresponding 95% confidence interval (CI) using Fine-Gray or Cox proportional hazards models. RESULTS AND LIMITATIONS: The DM and PCSM MMAI algorithms were significantly and independently associated with the risk of DM (sHR [95% CI] = 2.33 [1.60-3.38], p < 0.001) and PCSM, respectively (sHR [95% CI] = 3.54 [2.38-5.28], p < 0.001) when compared against other prognostic clinical factors and NCCN high-risk features. The lower 75% of patients by DM MMAI had estimated 5- and 10-yr DM rates of 4% and 7%, and the highest quartile had average 5- and 10-yr DM rates of 19% and 32%, respectively (p < 0.001). Similar results were observed for the PCSM MMAI algorithm. CONCLUSIONS: We externally validated the prognostic ability of MMAI models previously developed among men with localized high-risk disease. MMAI prognostic models further risk stratify beyond the clinical and pathological variables for DM and PCSM in a population of men already at a high risk for disease progression. This study provides evidence for consistent validation of our deep learning MMAI models to improve prognostication and enable more informed decision-making for patient care. PATIENT SUMMARY: This paper presents a novel approach using images from pathology slides along with clinical variables to validate artificial intelligence (computer-generated) prognostic models. When implemented, clinicians can offer a more personalized and tailored prognostic discussion for men with localized prostate cancer.

Artificial Intelligence Predictive Model for Hormone Therapy Use in Prostate Cancer.

Background: Androgen deprivation therapy (ADT) with radiotherapy can benefit patients with localized prostate cancer. However, ADT can negatively impact quality of life and there remain no validated predictive models to guide its use. Methods: Digital pathology image and clinical data from pre-treatment prostate tissue from 5,727 patients enrolled on five phase III randomized trials treated with radiotherapy +/- ADT were used to develop and validate an artificial intelligence (AI)-derived predictive model to assess ADT benefit with the primary endpoint of distant metastasis. After the model was locked, validation was performed on NRG/RTOG 9408 (n = 1,594) that randomized men to radiotherapy +/- 4 months of ADT. Fine-Gray regression and restricted mean survival times were used to assess the interaction between treatment and predictive model and within predictive model positive and negative subgroup treatment effects. Results: In the NRG/RTOG 9408 validation cohort (14.9 years of median follow-up), ADT significantly improved time to distant metastasis (subdistribution hazard ratio [sHR] = 0.64, 95%CI [0.45-0.90], p = 0.01). The predictive model-treatment interaction was significant (p-interaction = 0.01). In predictive model positive patients (n = 543, 34%), ADT significantly reduced the risk of distant metastasis compared to radiotherapy alone (sHR = 0.34, 95%CI [0.19-0.63], p < 0.001). There were no significant differences between treatment arms in the predictive model negative subgroup (n = 1,051, 66%; sHR = 0.92, 95%CI [0.59-1.43], p = 0.71). Conclusions: Our data, derived and validated from completed randomized phase III trials, show that an AI-based predictive model was able to identify prostate cancer patients, with predominately intermediate-risk disease, who are likely to benefit from short-term ADT.

Artificial Intelligence Predictive Model for Hormone Therapy Use in Prostate Cancer.

BACKGROUND: Androgen deprivation therapy (ADT) with radiotherapy can benefit patients with localized prostate cancer. However, ADT can negatively impact quality of life, and there remain no validated predictive models to guide its use. METHODS: We used digital pathology images from pretreatment prostate tissue and clinical data from 5727 patients enrolled in five phase 3 randomized trials, in which treatment was radiotherapy with or without ADT, as our data source to develop and validate an artificial intelligence (AI)­derived predictive patient-specific model that would determine which patients would develop the primary end point of distant metastasis. The model used baseline data to provide a binary output that a given patient will likely benefit from ADT or not. After the model was locked, validation was performed using data from NRG Oncology/Radiation Therapy Oncology Group (RTOG) 9408 (n=1594), a trial that randomly assigned men to radiotherapy plus or minus 4 months of ADT. Fine­Gray regression and restricted mean survival times were used to assess the interaction between treatment and the predictive model and within predictive model­positive, i.e., benefited from ADT, and ­negative subgroup treatment effects. RESULTS: Overall, in the NRG/RTOG 9408 validation cohort (14.9 years of median follow-up), ADT significantly improved time to distant metastasis. Of these enrolled patients, 543 (34%) were model positive, and ADT significantly reduced the risk of distant metastasis compared with radiotherapy alone. Of 1051 patients who were model negative, ADT did not provide benefit. CONCLUSIONS: Our AI-based predictive model was able to identify patients with a predominantly intermediate risk for prostate cancer likely to benefit from short-term ADT. (Supported by a grant [U10CA180822] from NRG Oncology Statistical and Data Management Center, a grant [UG1CA189867] from NCI Community Oncology Research Program, a grant [U10CA180868] from NRG Oncology Operations, and a grant [U24CA196067] from NRG Specimen Bank from the National Cancer Institute and by Artera, Inc. ClinicalTrials.gov numbers NCT00767286, NCT00002597, NCT00769548, NCT00005044, and NCT00033631.)

Prostate cancer therapy personalization via multi-modal deep learning on randomized phase III clinical trials.

Prostate cancer is the most frequent cancer in men and a leading cause of cancer death. Determining a patient's optimal therapy is a challenge, where oncologists must select a therapy with the highest likelihood of success and the lowest likelihood of toxicity. International standards for prognostication rely on non-specific and semi-quantitative tools, commonly leading to over- and under-treatment. Tissue-based molecular biomarkers have attempted to address this, but most have limited validation in prospective randomized trials and expensive processing costs, posing substantial barriers to widespread adoption. There remains a significant need for accurate and scalable tools to support therapy personalization. Here we demonstrate prostate cancer therapy personalization by predicting long-term, clinically relevant outcomes using a multimodal deep learning architecture and train models using clinical data and digital histopathology from prostate biopsies. We train and validate models using five phase III randomized trials conducted across hundreds of clinical centers. Histopathological data was available for 5654 of 7764 randomized patients (71%) with a median follow-up of 11.4 years. Compared to the most common risk-stratification tool-risk groups developed by the National Cancer Center Network (NCCN)-our models have superior discriminatory performance across all endpoints, ranging from 9.2% to 14.6% relative improvement in a held-out validation set. This artificial intelligence-based tool improves prognostication over standard tools and allows oncologists to computationally predict the likeliest outcomes of specific patients to determine optimal treatment. Outfitted with digital scanners and internet access, any clinic could offer such capabilities, enabling global access to therapy personalization.

FOXP3-positive regulatory T lymphocytes and epithelial FOXP3 expression in synchronous normal, ductal carcinoma in situ, and invasive cancer of the breast.

FOXP3-expressing T regulatory lymphocytes (Tregs) have been described as putative mediators of immune tolerance, and thus facilitators of tumor growth. When found in association with various malignancies, Tregs are generally markers of poor clinical outcome. However, it is unknown whether they are also associated with cancer progression. We evaluated quantitative FOXP3 expression in lymphocytes as well as in epithelial cells in a set of thirty-two breast tumors with synchronous normal epithelium, ductal carcinoma in situ (DCIS), and invasive ductal carcinoma (IDC) components. Tumors were stained for FOXP3 and CD3 expression and Tregs quantified by determining the ratio of colocalized FOXP3 and CD3 relative to 1) total CD3-expressing lymphocytes and 2) to FOXP3-expressing epithelial cells. The median proportion of FOXP3-expressing CD3 cells significantly increased with malignant progression from normal to DCIS to IDC components (0.005, 0.019 and 0.030, respectively; p ≤ 0.0001 for normal vs. IDC and p = 0.004 for DCIS vs. IDC). The median intensity of epithelial FOXP3 expression was also increased with invasive progression and most markedly augmented between normal and DCIS components (0.130 vs. 0.175, p ≤ 0.0001). Both Treg infiltration and epithelial FOXP3 expression were higher in grade 3 vs. grade 1 tumors (p = 0.014 for Tregs, p = 0.038 for epithelial FOXP3), but did not vary significantly with hormone receptor status, size of invasive tumor, lymph node status, or disease stage. Notably, Treg infiltration significantly correlated with epithelial up-regulation of FOXP3 expression (p = 0.013 for normal, p = 0.001 for IDC). These findings implicate both Treg infiltration and up-regulated epithelial FOXP3 expression in breast cancer progression.

Genomic alterations and phenotype of large compared to small high-grade ductal carcinoma in situ.

A clinically distinct subgroup of pure ductal carcinoma in situ presents as an extensive, high-grade lesion, which nevertheless lacks invasion. We sought to evaluate differences between those ductal carcinomas in situ presenting as large versus small lesions while controlling for high-grade, to determine whether there exist phenotypic and genetic differences between the 2 groups. Fifty-two cases of pure high-grade ductal carcinomas in situ were collected retrospectively, consisting of 27 large (>40 mm) and 25 small (<15 mm) cases. The 2 groups were compared based on genomic copy number assessed by array-based comparative genomic hybridization and by phenotype determined by immunohistochemistry for estrogen receptor, progesterone receptor, Ki-67, p53, cyclin D1, p16, cyclooxygenase 2, human epidermal growth factor receptor 2, and CD68. Large lesions presented at a younger age, with lower incidence of comedonecrosis and periductal macrophage response. Larger lesions also had significantly lower estrogen receptor expression, lower cyclin D1 expression, and lower Ki-67 index. The subset of 9 large palpable tumors had significantly lower p16/cyclooxygenase 2 expression and lower Ki-67 index compared to nonpalpable tumors. Genomically, larger lesions had fewer break points, fewer amplifications, and decreased copy number gains involving chromosome 8q and chromosome 20q when compared to the small lesions. Among pure high-grade tumors, small and large groups show specific genomic and phenotypic differences. Interestingly, larger tumors showed some molecular features associated with better prognosis. A more thorough evaluation of these differences could help identify the likelihood of recurrence or progression for in situ lesions.

Similarity in genetic alterations between paired well-differentiated and dedifferentiated components of dedifferentiated liposarcoma.

Liposarcoma represents a unique model insofar as some well-differentiated liposarcomas progress to non-lipogenic, so-called 'dedifferentiated,' forms. The well-differentiated and dedifferentiated family of liposarcomas demonstrates amplification of the chromosome subregion 12q13-q15 with resultant amplification of the MDM2 and CDK4 genes. However, the specific genetic changes that distinguish between well-differentiated and dedifferentiated liposarcomas are less well understood. To study the genetic changes in dedifferentiated liposarcomas, paired well-differentiated and dedifferentiated components of 29 tumors were analyzed separately by array-based comparative genomic hybridization. A bacterial artificial chromosome array at approximately 1-Mb resolution was used. The genetic changes were compared with clinical presentation, grade of the dedifferentiated component and overexpression of MDM2 and CDK4. Most tumors (n=21, 72%) were retroperitoneal, with both components present at initial diagnosis (n=25, 86%). Eight tumors (28%) were classified as low-grade dedifferentiation. In four cases (14%), a well-differentiated liposarcoma preceded the presentation of the dedifferentiated tumor by 1-5 years. 12q13-q15 was amplified in all tumors. Using unsupervised hierarchical clustering of copy-number changes, all but two tumors showed close similarities between well-differentiated and dedifferentiated components, and segregated as pairs. Dedifferentiated components had more total amplifications (P=0.008) and a trend for gain at 19q13.2, but no genetic changes were significant in distinguishing between the two components. High-level amplifications of 1p21-32 (n=7, 24%), 1q21-23 (n=9, 31%), 6q23-24 (n=6, 21%) and 12q24 (n=3, 10%) were common, but none significantly correlated with differentiation. Presentation and grade correlated with the frequency of changes at a number of genetic loci (P<0.001), whereas CDK4 immunostaining showed negative correlation with 12q13.13 amplification. The genotypic similarity, at the limit of the array's resolution, between components implies that most genetic changes precede phenotypic 'progression,' early in tumorigenesis. The relationship between genetic changes and presentation or grade may reflect differences in factors that control genomic instability or the background genotype of the tumor.

Pathologic and biologic response to preoperative endocrine therapy in patients with ER-positive ductal carcinoma in situ.

BACKGROUND: Endocrine therapy is commonly recommended in the adjuvant setting for patients as treatment for ductal carcinoma in situ (DCIS). However, it is unknown whether a neoadjuvant (preoperative) anti-estrogen approach to DCIS results in any biological change. This study was undertaken to investigate the pathologic and biomarker changes in DCIS following neoadjuvant endocrine therapy compared to a group of patients who did not undergo preoperative anti-estrogenic treatment to determine whether such treatment results in detectable histologic alterations. METHODS: Patients (n = 23) diagnosed with ER-positive pure DCIS by stereotactic core biopsy were enrolled in a trial of neoadjuvant anti-estrogen therapy followed by definitive excision. Patients on hormone replacement therapy, with palpable masses, or with histologic or clinical suspicion of invasion were excluded. Premenopausal women were treated with tamoxifen and postmenopausal women were treated with letrozole. Pathologic markers of proliferation, inflammation, and apoptosis were evaluated at baseline and at three months.Biomarker changes were compared to a cohort of patients who had not received preoperative treatment. RESULTS: Median age of the cohort was 53 years (range 38-78); 14 were premenopausal. Following treatment, predominant morphologic changes included increased multinucleated histiocytes and degenerated cells, decreased duct extension, and prominent periductal fibrosis. Two postmenopausal patients had ADH only with no residual DCIS at excision. Postmenopausal women on letrozole had significant reduction of PR, and Ki67 as well as increase in CD68-positive cells. For premenopausal women on tamoxifen treatment, the only significant change was increase in CD68. No change in cleaved caspase 3 was found. Two patients had invasive cancer at surgery. CONCLUSION: Preoperative therapy for DCIS is associated with significant pathologic alterations. These changes may be clinically significant. Further work is needed to identify which women may be the best candidates for such treatment for DCIS, and whether best responders may safely avoid surgical intervention. TRIAL REGISTRATION: ClinicalTrials.gov NCT00290745.

Genetic and phenotypic characteristics of pleomorphic lobular carcinoma in situ of the breast.

The clinical, pathologic, and molecular features of pleomorphic lobular carcinoma in situ (PLCIS) and the relationship of PLCIS to classic LCIS (CLCIS) are poorly defined. In this study, we analyzed 31 cases of PLCIS (13 apocrine and 18 nonapocrine subtypes) and compared the clinical, pathologic, immunophenotypic, and genetic characteristics of these cases with those of 24 cases of CLCIS. Biomarker expression was examined using immunostaining for E-cadherin, gross cystic disease fluid protein-15, estrogen, progesterone, androgen receptor, human epidermal growth factor receptor2, CK5/6, and Ki67. Array-based comparative genomic hybridization to assess the genomic alterations was performed using microdissected formalin-fixed paraffin-embedded samples. Patients with PLCIS presented with mammographic abnormalities. Histologically, the tumor cells were dyshesive and showed pleomorphic nuclei, and there was often associated necrosis and microcalcifications. All lesions were E-cadherin negative. Compared with CLCIS, PLCIS showed significantly higher Ki67 index, lower estrogen receptor and progesterone receptor expression, and higher incidence of HER2 gene amplification. The majority of PLCIS and CLCIS demonstrated loss of 16q and gain of 1q. Apocrine PLCIS had significantly more genomic alterations than CLCIS and nonapocrine PLCIS. Although lack of E-cadherin expression and the 16q loss and 1q gain-array-based comparative genomic hybridization pattern support a relationship to CLCIS, PLCIS has clinical, mammographic, histologic, immunophenotypic, and genetic features that distinguish it from CLCIS. The histologic features, biomarker profile, and genomic instability observed in PLCIS suggest a more aggressive phenotype than CLCIS. However, clinical follow-up studies will be required to define the natural history and most appropriate management of these lesions.

Protein acetylation and histone deacetylase expression associated with malignant breast cancer progression.

PURPOSE: Excess histone deacetylase (HDAC) activity can induce hypoacetylation of histone and nonhistone protein substrates, altering gene expression patterns and cell behavior potentially associated with malignant transformation. However, HDAC expression and protein acetylation have not been studied in the context of breast cancer progression. EXPERIMENTAL DESIGN: We assessed expression levels of acetylated histone H4 (ac-H4), ac-H4K12, ac-tubulin, HDAC1, HDAC2, and HDAC6 in 22 reduction mammoplasties and in 58 specimens with synchronous normal epithelium, ductal carcinoma in situ (DCIS), and invasive ductal carcinoma (IDC) components. Differences among groups were tested for significance using nonparametric tests. RESULTS: From normal epithelium to DCIS, there was a marked reduction in histone acetylation (P < 0.0001). Most cases showed similar levels of acetylation in DCIS and IDC, although some showed further reduction of ac-H4 and ac-H4K12 from DCIS to IDC. Expression of HDAC1, HDAC2, and HDAC6 was also significantly reduced but by a smaller magnitude. Greater reductions of H4 acetylation and HDAC1 levels were observed from normal to DCIS in estrogen receptor-negative compared with estrogen receptor-positive, and in high-grade compared with non-high-grade tumors. CONCLUSION: Overall, there was a global pattern of hypoacetylation associated with progression from normal to DCIS to IDC. These findings suggest that the reversal of this hypoacetylation in DCIS and IDC could be an early measure of HDAC inhibitor activity.

Fluorescence in situ hybridization of ductal lavage samples identifies malignant phenotypes from cytologically normal cells in women with breast cancer.

BACKGROUND: Ductal lavage (DL) does not routinely identify cytologically malignant cells. For this study, the authors asked whether molecular analyses of DL specimens from women with cancer would identify abnormal cells, even if they appeared cytologically normal. METHODS: DL was performed and yielded fluid in 29 of 45 consenting women who were undergoing breast cancer surgery. Array comparative genomic hybridization (CGH) was performed on the corresponding tumor tissue from 14 women. There was no single, common alteration; thus, bacterial artificial chromosome-specific fluorescence in situ hybridization (FISH) probes were selected based on CGH alterations. RESULTS: FISH copy number changes were detected in tumor sections in 9 women. In the corresponding 9 DL samples, 1 sample was clearly malignant on cytology, 1 showed marked atypia, 1 showed mild atypia, and the rest were benign. Five of the 9 DL samples had epithelial cells that showed genetic changes identical to those observed in the tumor by FISH. The remaining 4 of 9 DL samples that did not show molecular changes were probably (N = 1) or possibly (N = 3) from the same duct as the tumor. CONCLUSIONS: Although only 11% of the DL samples were identified as malignant cytologically, 55% showed molecular changes that were identical to those observed in the tumor. FISH was more sensitive for finding tumor in DL specimens than cytology. However, the ductal system in which the tumor was located did not always yield fluid, limiting the sensitivity of DL. The results from this study showed that genetic changes can be detected in the absence of morphologic changes in cytologically benign cells, but the application will be limited without a better approach for acquiring cells and a common set of probes for detecting molecular abnormalities that are found in breast malignancies.

Identification of a robust gene signature that predicts breast cancer outcome in independent data sets.

BACKGROUND: Breast cancer is a heterogeneous disease, presenting with a wide range of histologic, clinical, and genetic features. Microarray technology has shown promise in predicting outcome in these patients. METHODS: We profiled 162 breast tumors using expression microarrays to stratify tumors based on gene expression. A subset of 55 tumors with extensive follow-up was used to identify gene sets that predicted outcome. The predictive gene set was further tested in previously published data sets. RESULTS: We used different statistical methods to identify three gene sets associated with disease free survival. A fourth gene set, consisting of 21 genes in common to all three sets, also had the ability to predict patient outcome. To validate the predictive utility of this derived gene set, it was tested in two published data sets from other groups. This gene set resulted in significant separation of patients on the basis of survival in these data sets, correctly predicting outcome in 62-65% of patients. By comparing outcome prediction within subgroups based on ER status, grade, and nodal status, we found that our gene set was most effective in predicting outcome in ER positive and node negative tumors. CONCLUSION: This robust gene selection with extensive validation has identified a predictive gene set that may have clinical utility for outcome prediction in breast cancer patients.

A collection of breast cancer cell lines for the study of functionally distinct cancer subtypes.

Recent studies suggest that thousands of genes may contribute to breast cancer pathophysiologies when deregulated by genomic or epigenomic events. Here, we describe a model "system" to appraise the functional contributions of these genes to breast cancer subsets. In general, the recurrent genomic and transcriptional characteristics of 51 breast cancer cell lines mirror those of 145 primary breast tumors, although some significant differences are documented. The cell lines that comprise the system also exhibit the substantial genomic, transcriptional, and biological heterogeneity found in primary tumors. We show, using Trastuzumab (Herceptin) monotherapy as an example, that the system can be used to identify molecular features that predict or indicate response to targeted therapies or other physiological perturbations.

Genomic and transcriptional aberrations linked to breast cancer pathophysiologies.

This study explores the roles of genome copy number abnormalities (CNAs) in breast cancer pathophysiology by identifying associations between recurrent CNAs, gene expression, and clinical outcome in a set of aggressively treated early-stage breast tumors. It shows that the recurrent CNAs differ between tumor subtypes defined by expression pattern and that stratification of patients according to outcome can be improved by measuring both expression and copy number, especially high-level amplification. Sixty-six genes deregulated by the high-level amplifications are potential therapeutic targets. Nine of these (FGFR1, IKBKB, ERBB2, PROCC, ADAM9, FNTA, ACACA, PNMT, and NR1D1) are considered druggable. Low-level CNAs appear to contribute to cancer progression by altering RNA and cellular metabolism.

Breast tumor copy number aberration phenotypes and genomic instability.

BACKGROUND: Genomic DNA copy number aberrations are frequent in solid tumors, although the underlying causes of chromosomal instability in tumors remain obscure. Genes likely to have genomic instability phenotypes when mutated (e.g. those involved in mitosis, replication, repair, and telomeres) are rarely mutated in chromosomally unstable sporadic tumors, even though such mutations are associated with some heritable cancer prone syndromes. METHODS: We applied array comparative genomic hybridization (CGH) to the analysis of breast tumors. The variation in the levels of genomic instability amongst tumors prompted us to investigate whether alterations in processes/genes involved in maintenance and/or manipulation of the genome were associated with particular types of genomic instability. RESULTS: We discriminated three breast tumor subtypes based on genomic DNA copy number alterations. The subtypes varied with respect to level of genomic instability. We find that shorter telomeres and altered telomere related gene expression are associated with amplification, implicating telomere attrition as a promoter of this type of aberration in breast cancer. On the other hand, the numbers of chromosomal alterations, particularly low level changes, are associated with altered expression of genes in other functional classes (mitosis, cell cycle, DNA replication and repair). Further, although loss of function instability phenotypes have been demonstrated for many of the genes in model systems, we observed enhanced expression of most genes in tumors, indicating that over expression, rather than deficiency underlies instability. CONCLUSION: Many of the genes associated with higher frequency of copy number aberrations are direct targets of E2F, supporting the hypothesis that deregulation of the Rb pathway is a major contributor to chromosomal instability in breast tumors. These observations are consistent with failure to find mutations in sporadic tumors in genes that have roles in maintenance or manipulation of the genome.

Bladder cancer stage and outcome by array-based comparative genomic hybridization.

PURPOSE: Bladder carcinogenesis is believed to follow alternative pathways of disease progression driven by an accumulation of genetic alterations. The purpose of this study was to evaluate associations between measures of genomic instability and bladder cancer clinical phenotype. EXPERIMENTAL DESIGN: Genome-wide copy number profiles were obtained for 98 bladder tumors of diverse stages (29 pT(a), 14 pT1, 55 pT(2-4)) and grades (21 low-grade and 8 high-grade superficial tumors) by array-based comparative genomic hybridization (CGH). Each array contained 2,464 bacterial artificial chromosome and P1 clones, providing an average resolution of 1.5 Mb across the genome. A total of 54 muscle-invasive cases had follow-up information available. Overall outcome analysis was done for patients with muscle-invasive tumors having "good" (alive >2 years) versus "bad" (dead in <2 years) prognosis. RESULTS: Array CGH analysis showed significant increases in copy number alterations and genomic instability with increasing stage and with outcome. The fraction of genome altered (FGA) was significantly different between tumors of different stages (pT(a) versus pT1, P = 0.0003; pT(a) versus pT(2-4), P = 0.02; and pT1 versus pT(2-4), P = 0.03). Individual clones that differed significantly between different tumor stages were identified after adjustment for multiple comparisons (false discovery rate < 0.05). For muscle-invasive tumors, the FGA was associated with patient outcome (bad versus good prognosis patients, P = 0.002) and was identified as the only independent predictor of overall outcome based on a multivariate Cox proportional hazards method. Unsupervised hierarchical clustering separated "good" and "bad" prognosis muscle-invasive tumors into clusters that showed significant association with FGA and survival (Kaplan-Meier, P = 0.019). Supervised tumor classification (prediction analysis for microarrays) had a 71% classification success rate based on 102 unique clones. CONCLUSIONS: Array-based CGH identified quantitative and qualitative differences in DNA copy number alterations at high resolution according to tumor stage and grade. Fraction genome altered was associated with worse outcome in muscle-invasive tumors, independent of other clinicopathologic parameters. Measures of genomic instability add independent power to outcome prediction of bladder tumors.

Bladder cancer outcome and subtype classification by gene expression.

Models of bladder tumor progression have suggested that genetic alterations may determine both phenotype and clinical course. We have applied expression microarray analysis to a divergent set of bladder tumors to further elucidate the course of disease progression and to classify tumors into more homogeneous and clinically relevant subgroups. cDNA microarrays containing 10,368 human gene elements were used to characterize the global gene expression patterns in 80 bladder tumors, 9 bladder cancer cell lines, and 3 normal bladder samples. Robust statistical approaches accounting for the multiple testing problem were used to identify differentially expressed genes. Unsupervised hierarchical clustering successfully separated the samples into two subgroups containing superficial (pT(a) and pT(1)) versus muscle-invasive (pT(2)-pT(4)) tumors. Supervised classification had a 90.5% success rate separating superficial from muscle-invasive tumors based on a limited subset of genes. Tumors could also be classified into transitional versus squamous subtypes (89% success rate) and good versus bad prognosis (78% success rate). The performance of our stage classifiers was confirmed in silico using data from an independent tumor set. Validation of differential expression was done using immunohistochemistry on tissue microarrays for cathepsin E, cyclin A2, and parathyroid hormone-related protein. Genes driving the separation between tumor subsets may prove to be important biomarkers for bladder cancer development and progression and eventually candidates for therapeutic targeting.

Clinical application of array-based comparative genomic hybridization to define the relationship between multiple synchronous tumors.

Array-based comparative genomic hybridization (CGH) is a technique that allows genome wide screening of gains and losses in DNA copy number. In cases where multiple tumors are encountered, this genetic technique may prove useful in differentiating new primary tumors from recurrences. In this case report, we used array-based CGH to examine the genomic relationships among two leiomyosarcomas and two breast cancers in the same patient, three of which were diagnosed synchronously. Array-based CGH was performed on the four tumor samples using random prime amplified microdissected DNA. Samples were hybridized onto bacterial artificial chromosome arrays composed of approximately 2400 clones. Patterns of alterations within the tumors were compared and genetic alterations among the leiomyosarcomas and breast lesions were found. Overall, three distinct genetic profiles were observed. While the two leiomyosarcomas shared a similar pattern of genetic alterations, the two invasive breast lesions did not. The nearly identical pattern of genetic alterations belonging to the two metachronous leiomyosarcomas confirmed metastatic recurrence while the two different genetic profiles of the invasive ductal carcinomas suggest that the two lesions represented two distinct foci of multifocal disease rather than clonal extension of the primary tumor. We conclude that genetic analysis by array-based CGH can clearly elucidate the relationships between multiple tumors and may potentially serve as an important clinical tool.

Array-based comparative genomic hybridization from formalin-fixed, paraffin-embedded breast tumors.

Identification of prognostic and predictive genomic markers requires long-term clinical follow-up of patients. Extraction of high-quality DNA from archived formalin-fixed, paraffin-embedded material is essential for such studies. Of particular importance is a robust reproducible method of whole genome amplification for small tissue samples. This is especially true for high-resolution analytical approaches because different genomic regions and sequences may amplify differentially. We have tested a number of protocols for DNA amplification for array-based comparative genomic hybridization (CGH), in which relative copy number of the entire genome is measured at 1 to 2 mb resolution. Both random-primed amplification and degenerate oligonucleotide-primed amplification approaches were tested using varying amounts of fresh and paraffin-extracted normal and breast tumor input DNAs. We found that random-primed amplification was clearly superior to degenerate oligonucleotide-primed amplification for array-based CGH. The best quality and reproducibility strongly depended on accurate determination of the amount of input DNA using a quantitative polymerase chain reaction-based method. Reproducible and high-quality results were attained using 50 ng of input DNA, and some samples yielded quality results with as little as 5 ng input DNA. We conclude that random-primed amplification of DNA isolated from paraffin sections is a robust and reproducible approach for array-based CGH analysis of archival tumor samples.