Comparison of the predictive accuracy of DNA array-based multigene classifiers across cDNA arrays and Affymetrix GeneChips

Developing a 'personalome' for precision medicine: emerging methods that compute interpretable effect sizes from single-subject transcriptomes

The development of computational methods capable of analyzing -omics data at the individual level is critical for the success of precision medicine. Although unprecedented opportunities now exist to gather data on an individual's -omics profile ('personalome'), interpreting and extracting meaningful information from single-subject -omics remain underdeveloped, particularly for quantitative non-sequence measurements, including complete transcriptome or proteome expression and metabolite abundance. Conventional bioinformatics approaches have largely been designed for making population-level inferences about 'average' disease processes; thus, they may not adequately capture and describe individual variability. Novel approaches intended to exploit a variety of -omics data are required for identifying individualized signals for meaningful interpretation. In this review-intended for biomedical researchers, computational biologists and bioinformaticians-we survey emerging computational and translational informatics methods capable of constructing a single subject's 'personalome' for predicting clinical outcomes or therapeutic responses, with an emphasis on methods that provide interpretable readouts. Key points: (i) the single-subject analytics of the transcriptome shows the greatest development to date and, (ii) the methods were all validated in simulations, cross-validations or independent retrospective data sets. This survey uncovers a growing field that offers numerous opportunities for the development of novel validation methods and opens the door for future studies focusing on the interpretation of comprehensive 'personalomes' through the integration of multiple -omics, providing valuable insights into individual patient outcomes and treatments.

Signatures of breast cancer metastasis at a glance

Gene expression profiling has yielded expression signatures from which prognostic tests can be derived to facilitate clinical decision making in breast cancer patients. Some of these signatures are based on profiling of whole tumor tissue (tissue signatures), which includes all tumor and stromal cells. Prognostic markers have also been derived from the profiling of metastasizing tumor cells, including circulating tumor cells (CTCs) and migratory-disseminating tumor cells within the primary tumor. The metastasis signatures based on CTCs and migratory-disseminating tumor cells have greater potential for unraveling cell biology insights and mechanistic underpinnings of tumor cell dissemination and metastasis. Of clinical interest is the promise that stratification of patients into high or low metastatic risk, as well as assessing the need for cytotoxic therapy, might be improved if prognostics derived from these two types of signatures are used in a combined way. The aim of this Cell Science at a Glance article and accompanying poster is to navigate through both types of signatures and their derived prognostics, as well as to highlight biological insights and clinical applications that could be derived from them, especially when they are used in combination.

Fractal dimension and lacunarity of tumor microscopic images as prognostic indicators of clinical outcome in early breast cancer

AIM

Research in the field of breast cancer outcome prognosis has been focused on molecular biomarkers, while neglecting the discovery of novel tumor histology structural clues. We thus aimed to improve breast cancer prognosis by fractal analysis of tumor histomorphology.

PATIENTS & METHODS

This retrospective study included 92 breast cancer patients without systemic treatment.

RESULTS

Fractal dimension and lacunarity of the breast tumor microscopic histology possess prognostic value comparable to the major clinicopathological prognostic parameters.

CONCLUSION

Fractal analysis was performed for the first time on routinely produced archived pan-tissue stained primary breast tumor sections, indicating its potential for clinical use as a simple and cost-effective prognostic indicator of distant metastasis risk to complement the molecular approaches for cancer risk prognosis.

Comparative study of joint analysis of microarray gene expression data in survival prediction and risk assessment of breast cancer patients

Microarray gene expression data sets are jointly analyzed to increase statistical power. They could either be merged together or analyzed by meta-analysis. For a given ensemble of data sets, it cannot be foreseen which of these paradigms, merging or meta-analysis, works better. In this article, three joint analysis methods, Z -score normalization, ComBat and the inverse normal method (meta-analysis) were selected for survival prognosis and risk assessment of breast cancer patients. The methods were applied to eight microarray gene expression data sets, totaling 1324 patients with two clinical endpoints, overall survival and relapse-free survival. The performance derived from the joint analysis methods was evaluated using Cox regression for survival analysis and independent validation used as bias estimation. Overall, Z -score normalization had a better performance than ComBat and meta-analysis. Higher Area Under the Receiver Operating Characteristic curve and hazard ratio were also obtained when independent validation was used as bias estimation. With a lower time and memory complexity, Z -score normalization is a simple method for joint analysis of microarray gene expression data sets. The derived findings suggest further assessment of this method in future survival prediction and cancer classification applications.

Long non-coding RNA expression profiles predict metastasis in lymph node-negative breast cancer independently of traditional prognostic markers

INTRODUCTION

Patients with clinically and pathologically similar breast tumors often have very different outcomes and treatment responses. Current prognostic markers allocate the majority of breast cancer patients to the high-risk group, yielding high sensitivities in expense of specificities below 20%, leading to considerable overtreatment, especially in lymph node-negative patients. Seventy percent would be cured by surgery and radiotherapy alone in this group. Thus, precise and early indicators of metastasis are highly desirable to reduce overtreatment. Previous prognostic RNA-profiling studies have only focused on the protein-coding part of the genome, however the human genome contains thousands of long non-coding RNAs (lncRNAs) and this unexplored field possesses large potential for identification of novel prognostic markers.

METHODS

We evaluated lncRNA microarray data from 164 primary breast tumors from adjuvant naïve patients with a mean follow-up of 18 years. Eighty two patients who developed detectable distant metastasis were compared to 82 patients where no metastases were diagnosed. For validation, we determined the prognostic value of the lncRNA profiles by comparing the ability of the profiles to predict metastasis in two additional, previously-published, cohorts.

RESULTS

We showed that lncRNA profiles could distinguish metastatic patients from non-metastatic patients with sensitivities above 90% and specificities of 64-65%. Furthermore; classifications were independent of traditional prognostic markers and time to metastasis.

CONCLUSIONS

To our knowledge, this is the first study investigating the prognostic potential of lncRNA profiles. Our study suggest that lncRNA profiles provide additional prognostic information and may contribute to the identification of early breast cancer patients eligible for adjuvant therapy, as well as early breast cancer patients that could avoid unnecessary systemic adjuvant therapy. This study emphasizes the potential role of lncRNAs in breast cancer prognosis.

Targeting p35/Cdk5 signalling via CIP-peptide promotes angiogenesis in hypoxia

Cyclin-dependent kinase-5 (Cdk5) is over-expressed in both neurons and microvessels in hypoxic regions of stroke tissue and has a significant pathological role following hyper-phosphorylation leading to calpain-induced cell death. Here, we have identified a critical role of Cdk5 in cytoskeleton/focal dynamics, wherein its activator, p35, redistributes along actin microfilaments of spreading cells co-localising with p(Tyr15)Cdk5, talin/integrin beta-1 at the lamellipodia in polarising cells. Cdk5 inhibition (roscovitine) resulted in actin-cytoskeleton disorganisation, prevention of protein co-localization and inhibition of movement. Cells expressing Cdk5 (D144N) kinase mutant, were unable to spread, migrate and form tube-like structures or sprouts, while Cdk5 wild-type over-expression showed enhanced motility and angiogenesis in vitro, which was maintained during hypoxia. Gene microarray studies demonstrated myocyte enhancer factor (MEF2C) as a substrate for Cdk5-mediated angiogenesis in vitro. MEF2C showed nuclear co-immunoprecipitation with Cdk5 and almost complete inhibition of differentiation and sprout formation following siRNA knock-down. In hypoxia, insertion of Cdk5/p25-inhibitory peptide (CIP) vector preserved and enhanced in vitro angiogenesis. These results demonstrate the existence of critical and complementary signalling pathways through Cdk5 and p35, and through which coordination is a required factor for successful angiogenesis in sustained hypoxic condition.

iTRAQ-based quantitative proteomics reveals myoferlin as a novel prognostic predictor in pancreatic adenocarcinoma

Histological differentiation is a major pathological parameter associated with poor prognosis in patients with pancreatic adenocarcinoma (PAC) and the molecular signature underlying PAC differentiation may involve key proteins potentially affecting the malignant characters of PAC. We aimed to identify the proteins which could be implicated in PAC prognosis. We used isobaric tags for relative and absolute quantitation (iTRAQ) coupled with two-dimensional liquid chromatography-tandem mass spectrometry to compare protein expression in PAC tissues with different degrees of histological differentiation. A total of 1623 proteins were repeatedly identified by performing the iTRAQ-based experiments twice. Of these, 15 proteins were differentially expressed according to our defined criteria. Myoferlin (MYOF) was selected to validate the proteomic results by western blotting. Immunohistochemistry in a further 154 PAC cases revealed that myoferlin significantly correlated with the degree of histological differentiation (P=0.004), and univariate and multivariate analyses indicated that MYOF is an independent prognostic factor for survival (hazard ratio, 1.540; 95% confidence interval, 1.061-2.234; P=0.023) of patients with PAC after curative surgery. RNA interference-mediated knockdown of MYOF alleviated malignant phenotypes of both primary and metastatic PAC cell lines in vitro and in vivo. Thus, ITRAQ-based quantitative proteomics revealed the prognostic value of MYOF in PAC.

BIOLOGICAL SIGNIFICANCE

Our results provide the possibility of novel strategies for pancreatic adenocarcinoma management.

Expression of myoferlin in human and murine carcinoma tumors: role in membrane repair, cell proliferation, and tumorigenesis

Cancer cells are often characterized by high proliferation rates, a consequence of increased mitotic signaling coupled with unchecked cellular growth. We recently demonstrated that vascular endothelial cells unexpectedly express ferlins, a family of muscle-specific proteins capable of regulating the fusion of lipid patches to the plasma membrane, and that these highly regulated membrane fusion events are essential to endothelial cell proliferation and homeostasis. Here, we show that human and mouse breast cancer cell lines also express myoferlin at various levels, and that the processes of transformation, epithelial-mesenchymal transition, and metastasis do not appear to have any effect on myoferlin expression in vitro. In vivo, we observed that solid mouse and human carcinoma tissues also express high levels of myoferlin protein. Loss-of-function studies performed in mice revealed that myoferlin gene knockdown can attenuate cancer cell proliferation in vitro and decrease tumor burden, and that accelerated tumor cell growth appears to rely on intact myoferlin-dependent membrane repair and signaling under exponential growth conditions. To our knowledge, these data provide the first evidence of myoferlin expression in solid human and mouse tumors. We have thus identified a novel membrane repair process that likely helps sustain the high growth rates characteristic of tumors, and we suggest that interfering with normal myoferlin expression and/or membrane repair and remodeling may provide therapeutically relevant antiproliferative effects.

Prediction of breast cancer metastasis by gene expression profiles: a comparison of metagenes and single genes

Background

The popularity of a large number of microarray applications has in cancer research led to the development of predictive or prognostic gene expression profiles. However, the diversity of microarray platforms has made the full validation of such profiles and their related gene lists across studies difficult and, at the level of classification accuracies, rarely validated in multiple independent datasets. Frequently, while the individual genes between such lists may not match, genes with same function are included across such gene lists. Development of such lists does not take into account the fact that genes can be grouped together as metagenes (MGs) based on common characteristics such as pathways, regulation, or genomic location. Such MGs might be used as features in building a predictive model applicable for classifying independent data. It is, therefore, demanding to systematically compare independent validation of gene lists or classifiers based on metagene or individual gene (SG) features.

Methods

In this study we compared the performance of either metagene-or single gene-based feature sets and classifiers using random forest and two support vector machines for classifier building. The performance within the same dataset, feature set validation performance, and validation performance of entire classifiers in strictly independent datasets were assessed by 10 times repeated 10-fold cross validation, leave-one-out cross validation, and one-fold validation, respectively. To test the significance of the performance difference between MG- and SG-features/classifiers, we used a repeated down-sampled binomial test approach.

Results

MG- and SG-feature sets are transferable and perform well for training and testing prediction of metastasis outcome in strictly independent data sets, both between different and within similar microarray platforms, while classifiers had a poorer performance when validated in strictly independent datasets. The study showed that MG- and SG-feature sets perform equally well in classifying independent data. Furthermore, SG-classifiers significantly outperformed MG-classifier when validation is conducted between datasets using similar platforms, while no significant performance difference was found when validation was performed between different platforms.

Conclusion

Prediction of metastasis outcome in lymph node–negative patients by MG- and SG-classifiers showed that SG-classifiers performed significantly better than MG-classifiers when validated in independent data based on the same microarray platform as used for developing the classifier. However, the MG- and SG-classifiers had similar performance when conducting classifier validation in independent data based on a different microarray platform. The latter was also true when only validating sets of MG- and SG-features in independent datasets, both between and within similar and different platforms.

Gene expression, molecular class changes, and pathway analysis after neoadjuvant systemic therapy for breast cancer

PURPOSE

To examine gene expression differences between pre- and post-neoadjuvant systemic therapy (NST) specimens of breast cancers and identify biologic changers that may lead to new therapeutic insights.

METHODS

Gene expression data from prechemotherapy fine needle aspiration specimens were compared with resected residual cancers in 21 patients after 4 to 6 months of NST. We removed stroma-associated genes to minimize confounding effects. PAM50 was used to assign molecular class. Paired t test and gene set analysis were used to identify differentially expressed genes and pathways.

RESULTS

The ER and HER2 status based on mRNA expression remained stable in all but two cases, and there were no changes in proliferation metrics (Ki67 and proliferating cell nuclear antigen expression). Molecular class changed in 8 cases (33.3%), usually to normal-like class, which was associated with low residual cancer cell cellularity. The expression of 200 to 600 probe sets changed between baseline and post-NST samples. In basal-like cancers, pathways driven by increased expression of phosphoinositide 3-kinase, small G proteins, and calmodulin-dependent protein kinase II and energy metabolism were enriched, whereas immune cell-derived and the sonic hedgehog pathways were depleted in residual cancer. In non-basal-like breast cancers, notch signaling and energy metabolism (e.g., fatty acid synthesis) were enriched and sonic hedgehog signaling and immune-related pathways were depleted in residual cancer. There was no increase in epithelial-mesenchymal transition or cancer stem cell signatures.

CONCLUSIONS

Our data indicate that energy metabolism related processes are upregulated and immune-related signals are depleted in residual cancers. Targeting these biologic processes may represent promising adjuvant treatment strategies for patients with residual cancer.

Mechanistic modeling of the effects of myoferlin on tumor cell invasion

Myoferlin (MYOF) is a member of the evolutionarily conserved ferlin family of proteins, noted for their role in a variety of membrane processes, including endocytosis, repair, and vesicular transport. Notably, ferlins are implicated in Caenorhabditis elegans sperm motility (Fer-1), mammalian skeletal muscle development and repair (MYOF and dysferlin), and presynaptic transmission in the auditory system (otoferlin). In this paper, we demonstrate that MYOF plays a previously unrecognized role in cancer cell invasion, using a combination of mathematical modeling and in vitro experiments. Using a real-time impedance-based invasion assay (xCELLigence), we have shown that lentiviral-based knockdown of MYOF significantly reduced invasion of MDA-MB-231 breast cancer cells in Matrigel bioassays. Based on these experimental data, we developed a partial differential equation model of MYOF effects on cancer cell invasion, which we used to generate mechanistic hypotheses. The mathematical model predictions revealed that matrix metalloproteinases (MMPs) may play a key role in modulating this invasive property, which was supported by experimental data using quantitative RT-PCR screens. These results suggest that MYOF may be a promising target for biomarkers or drug target for metastatic cancer diagnosis and therapy, perhaps mediated through MMPs.

Effects of progesterone treatment on expression of genes involved in uterine quiescence

An important action of progesterone during pregnancy is to maintain the uterus in a quiescent state and thereby prevent preterm labor. The causes of preterm labor are not well understood, so progesterone action on the myometrium can provide clues about the processes that keep the uterus from contracting prematurely. Accordingly, we have carried out Affymetrix GeneChip analysis of progesterone effects on gene expression in immortalized human myometrial cells cultured from a patient near the end of pregnancy. Progesterone appears to inhibit uterine excitability by a number of mechanisms, including increased expression of calcium and voltage-operated K(+) channels, which dampens the electrical activity of the myometrial cell, downregulation of agents, and receptors involved in myometrial contraction, reduction in cell signal components that lead to increased intracellular Ca(2+) concentrations in response to contractile stimuli, and downregulation of proteins involved in the cross-linking of actin and myosin filaments to produce uterine contractions.

Bayesian approach to transforming public gene expression repositories into disease diagnosis databases

The rapid accumulation of gene expression data has offered unprecedented opportunities to study human diseases. The National Center for Biotechnology Information Gene Expression Omnibus is currently the largest database that systematically documents the genome-wide molecular basis of diseases. However, thus far, this resource has been far from fully utilized. This paper describes the first study to transform public gene expression repositories into an automated disease diagnosis database. Particularly, we have developed a systematic framework, including a two-stage Bayesian learning approach, to achieve the diagnosis of one or multiple diseases for a query expression profile along a hierarchical disease taxonomy. Our approach, including standardizing cross-platform gene expression data and heterogeneous disease annotations, allows analyzing both sources of information in a unified probabilistic system. A high level of overall diagnostic accuracy was shown by cross validation. It was also demonstrated that the power of our method can increase significantly with the continued growth of public gene expression repositories. Finally, we showed how our disease diagnosis system can be used to characterize complex phenotypes and to construct a disease-drug connectivity map.

Effect of training-sample size and classification difficulty on the accuracy of genomic predictors

Introduction

As part of the MicroArray Quality Control (MAQC)-II project, this analysis examines how the choice of univariate feature-selection methods and classification algorithms may influence the performance of genomic predictors under varying degrees of prediction difficulty represented by three clinically relevant endpoints.

Methods

We used gene-expression data from 230 breast cancers (grouped into training and independent validation sets), and we examined 40 predictors (five univariate feature-selection methods combined with eight different classifiers) for each of the three endpoints. Their classification performance was estimated on the training set by using two different resampling methods and compared with the accuracy observed in the independent validation set.

Results

A ranking of the three classification problems was obtained, and the performance of 120 models was estimated and assessed on an independent validation set. The bootstrapping estimates were closer to the validation performance than were the cross-validation estimates. The required sample size for each endpoint was estimated, and both gene-level and pathway-level analyses were performed on the obtained models.

Conclusions

We showed that genomic predictor accuracy is determined largely by an interplay between sample size and classification difficulty. Variations on univariate feature-selection methods and choice of classification algorithm have only a modest impact on predictor performance, and several statistically equally good predictors can be developed for any given classification problem.

Classification across gene expression microarray studies

BACKGROUND

The increasing number of gene expression microarray studies represents an important resource in biomedical research. As a result, gene expression based diagnosis has entered clinical practice for patient stratification in breast cancer. However, the integration and combined analysis of microarray studies remains still a challenge. We assessed the potential benefit of data integration on the classification accuracy and systematically evaluated the generalization performance of selected methods on four breast cancer studies comprising almost 1000 independent samples. To this end, we introduced an evaluation framework which aims to establish good statistical practice and a graphical way to monitor differences. The classification goal was to correctly predict estrogen receptor status (negative/positive) and histological grade (low/high) of each tumor sample in an independent study which was not used for the training. For the classification we chose support vector machines (SVM), predictive analysis of microarrays (PAM), random forest (RF) and k-top scoring pairs (kTSP). Guided by considerations relevant for classification across studies we developed a generalization of kTSP which we evaluated in addition. Our derived version (DV) aims to improve the robustness of the intrinsic invariance of kTSP with respect to technologies and preprocessing.

RESULTS

For each individual study the generalization error was benchmarked via complete cross-validation and was found to be similar for all classification methods. The misclassification rates were substantially higher in classification across studies, when each single study was used as an independent test set while all remaining studies were combined for the training of the classifier. However, with increasing number of independent microarray studies used in the training, the overall classification performance improved. DV performed better than the average and showed slightly less variance. In particular, the better predictive results of DV in across platform classification indicate higher robustness of the classifier when trained on single channel data and applied to gene expression ratios.

CONCLUSIONS

We present a systematic evaluation of strategies for the integration of independent microarray studies in a classification task. Our findings in across studies classification may guide further research aiming on the construction of more robust and reliable methods for stratification and diagnosis in clinical practice.

Can survival prediction be improved by merging gene expression data sets?

BACKGROUND

High-throughput gene expression profiling technologies generating a wealth of data, are increasingly used for characterization of tumor biopsies for clinical trials. By applying machine learning algorithms to such clinically documented data sets, one hopes to improve tumor diagnosis, prognosis, as well as prediction of treatment response. However, the limited number of patients enrolled in a single trial study limits the power of machine learning approaches due to over-fitting. One could partially overcome this limitation by merging data from different studies. Nevertheless, such data sets differ from each other with regard to technical biases, patient selection criteria and follow-up treatment. It is therefore not clear at all whether the advantage of increased sample size outweighs the disadvantage of higher heterogeneity of merged data sets. Here, we present a systematic study to answer this question specifically for breast cancer data sets. We use survival prediction based on Cox regression as an assay to measure the added value of merged data sets.

RESULTS

Using time-dependent Receiver Operating Characteristic-Area Under the Curve (ROC-AUC) and hazard ratio as performance measures, we see in overall no significant improvement or deterioration of survival prediction with merged data sets as compared to individual data sets. This apparently was due to the fact that a few genes with strong prognostic power were not available on all microarray platforms and thus were not retained in the merged data sets. Surprisingly, we found that the overall best performance was achieved with a single-gene predictor consisting of CYB5D1.

CONCLUSIONS

Merging did not deteriorate performance on average despite (a) The diversity of microarray platforms used. (b) The heterogeneity of patients cohorts. (c) The heterogeneity of breast cancer disease. (d) Substantial variation of time to death or relapse. (e) The reduced number of genes in the merged data sets. Predictors derived from the merged data sets were more robust, consistent and reproducible across microarray platforms. Moreover, merging data sets from different studies helps to better understand the biases of individual studies and can lead to the identification of strong survival factors like CYB5D1 expression.

MammaPrint 70-gene signature: another milestone in personalized medical care for breast cancer patients

The MammaPrint assay (Agendia BV, The Netherlands) is the first fully commercialized microarray-based multigene assay designed to individualize treatment for patients with breast cancer. MammaPrint, the first assay to be cleared at the 510(k) level by the US FDA's new in vitro diagnostic multivariate index assay classification, is offered as a prognostic test for women under the age of 61 years with either estrogen receptor-positive or -negative, lymph node-negative breast cancer. Unlike the Oncotype DX assay (Genomic Health, CA, USA), this test requires freshly prepared tissues collected into an RNA preservative solution. The 70 genes that comprise the MammaPrint assay are focused primarily on proliferation with additional genes associated with invasion, metastasis, stromal integrity and angiogenesis. The Microarray In Node-negative Disease may Avoid Chemotherapy (MINDACT) trial, sponsored by the European Organization for Research and Treatment of Cancer, involves the assessment of patients in the adjuvant treatment setting by the standard clinicopathologic prognostic factors included on Adjuvant! Online and by the 70-gene MammaPrint assay. The following article will consider the basic biology, technology, ease of clinical use, level of clinical validation and potential clinical utility of this test.

Multigene classifiers, prognostic factors, and predictors of breast cancer clinical outcome

A series of multigene classifiers, prognostic and predictive tests have recently been introduced as potentially useful adjuncts for the management of recently diagnosed breast cancer patients. These tests have used both slide-based methods including immunohistochemistry and fluorescence in situ hybridization and nonmorphology driven molecular platforms including quantitative multiplex real time polymerase chain reaction and genomic microarray profiling. In this review, a series of partially and completely commercialized multigene assays are compared with the standard breast cancer clinico-pathologic variables and biomarkers and evaluated as to the level of their scientific validation, current clinical utility, regulatory approval status, and estimated cost-benefit. A comparison of the Oncotype Dx and MammaPrint assays indicates that the Oncotype Dx test has the advantages of an earlier commercial launch in the US, wide acceptance for payment by third party payors, the ease of use of formalin fixed paraffin embedded tissues, a recommendation as ready for use by the American Society of Clinical Oncology Breast Cancer Tumor Markers Update Committee, a continuous rather than dichotomous algorithm, inclusion of both estrogen receptor (ER) and human epidermal growth factor receptor 2 in the mRNA profile, an ability to serve as both a prognostic and predictive test for certain hormonal and chemotherapeutic agents, demonstrated cost-effectiveness in 1 published study, and a high accrual rate for the prospective validation clinical trial (Trial Assigning Individualized Options for Treatment Rx). The MammaPrint assay has the advantages of a 510(k) clearance by the US Food and Drug Administration, a larger gene number which may enhance further utility, and the potentially wider patient eligibility including lymph node-positive, ER-negative, and younger patients being accrued into the prospective trial (the Microarray in Node-negative Disease may Avoid ChemoTherapy). A number of other assays have specific predictive goals most often focused on the efficacy of tamoxifen in ER-positive patients such as the Two-gene Ratio test and the Cytochrome P450 CYP2D6 genotyping assay.

Integrative disease classification based on cross-platform microarray data

Background

Disease classification has been an important application of microarray technology. However, most microarray-based classifiers can only handle data generated within the same study, since microarray data generated by different laboratories or with different platforms can not be compared directly due to systematic variations. This issue has severely limited the practical use of microarray-based disease classification.

Results

In this study, we tested the feasibility of disease classification by integrating the large amount of heterogeneous microarray datasets from the public microarray repositories. Cross-platform data compatibility is created by deriving expression log-rank ratios within datasets. One may then compare vectors of log-rank ratios across datasets. In addition, we systematically map textual annotations of datasets to concepts in Unified Medical Language System (UMLS), permitting quantitative analysis of the phenotype "distance" between datasets and automated construction of disease classes. We design a new classification approach named ManiSVM, which integrates Manifold data transformation with SVM learning to exploit the data properties. Using the leave one dataset out cross validation, ManiSVM achieved the overall accuracy of 70.7% (68.6% precision and 76.9% recall) with many disease classes achieving the accuracy higher than 80%.

Conclusion

Our results not only demonstrated the feasibility of the integrated disease classification approach, but also showed that the classification accuracy increases with the number of homogenous training datasets. Thus, the power of the integrative approach will increase with the continuous accumulation of microarray data in public repositories. Our study shows that automated disease diagnosis can be an important and promising application of the enormous amount of costly to generate, yet freely available, public microarray data.

Commercialized multigene predictors of clinical outcome for breast cancer

In the past 5 years, a number of commercialized multigene prognostic and predictive tests have entered the complex and expanding landscape of breast cancer companion diagnostics. These tests have used a variety of formats ranging from the familiar slide-based assays of immunohistochemistry and fluorescence in situ hybridization to the nonmorphology-driven molecular platforms of quantitative multiplex real-time polymerase chain reaction and genomic microarray profiling. In this review, 14 multigene assays are evaluated as to their scientific validation, current clinical utility, regulatory approval status, and estimated cost-benefit ratio. Emphasis is placed on two tests: oncotype DX and MammaPrint. Current evidence indicates that the oncotype DX test has the advantages of earlier commercial launch, wide acceptance for payment by third-party payors in the U.S., ease of use of formalin-fixed paraffin-embedded tissues, recent listing by the American Society of Clinical Oncology Breast Cancer Tumor Markers Update Committee as recommended for use, continuous scoring system algorithm, ability to serve as both a prognostic test and predictive test for certain hormonal and chemotherapeutic agents, demonstrated cost-effectiveness in one published study, and a high accrual rate for the prospective validation clinical trial (Trial Assigning Individualized Options for Treatment). The MammaPrint assay has the advantages of a 510(k) clearance by the U.S. Food and Drug Administration, a larger gene number, which may enhance further utility, and a potentially wider patient eligibility, including lymph node-positive, estrogen receptor (ER)-negative, and younger patients being accrued into the prospective trial (Microarray in Node-Negative Disease May Avoid Chemotherapy). A number of other assays have specific predictive goals that are most often focused on the efficacy of tamoxifen in ER-positive patients, such as the two-gene ratio test and the cytochrome P450 CYP2D6 genotyping assay.

Review of the literature examining the correlation among DNA microarray technologies

DNA microarray technologies are used in a variety of biological disciplines. The diversity of platforms and analytical methods employed has raised concerns over the reliability, reproducibility and correlation of data produced across the different approaches. Initial investigations (years 2000-2003) found discrepancies in the gene expression measures produced by different microarray technologies. Increasing knowledge and control of the factors that result in poor correlation among the technologies has led to much higher levels of correlation among more recent publications (years 2004 to present). Here, we review the studies examining the correlation among microarray technologies. We find that with improvements in the technology (optimization and standardization of methods, including data analysis) and annotation, analysis across platforms yields highly correlated and reproducible results. We suggest several key factors that should be controlled in comparing across technologies, and are good microarray practice in general.

Analytical validation of the Oncotype DX genomic diagnostic test for recurrence prognosis and therapeutic response prediction in node-negative, estrogen receptor-positive breast cancer

BACKGROUND

Oncotype DX is a clinically validated, high-complexity, multianalyte reverse transcription-PCR genomic test that predicts the likelihood of breast cancer recurrence in early-stage, node-negative, estrogen receptor-positive breast cancer. The Recurrence Score (RS) provides a more accurate, reproducible measure of breast cancer aggressiveness and therapeutic responsiveness than standard measures. Individualized patient management requires strict performance criteria for clinical laboratory tests. We therefore investigated the analytical performance of the assay.

METHODS

Assays used a pooled RNA sample from fixed paraffin-embedded tissues to evaluate the analytical performance of a 21-gene panel with respect to amplification efficiency, precision, linearity, and dynamic range, as well as limits of detection and quantification. Performance variables were estimated from assays carried out with sample dilutions. In addition, individual patient samples were used to test the optimized assay for reproducibility and sources of imprecision.

RESULTS

Assay results defined acceptable operational performance ranges, including an estimated maximum deviation from linearity of <1 cycle threshold (C(T)) units over a > or =2000-fold range of RNA concentrations, with a mean quantification bias of 0.3% and CVs of 3.2%-5.7%. An analysis of study design showed that assay imprecision contributed by instrument, operator, reagent, and day-to-day baseline variation was low, with SDs of <0.5 C(T).

CONCLUSION

The analytical and operational performance specifications defined for the Oncotype DX assay allow the reporting of quantitative RS values for individual patients with an SD within 2 RS units on a 100-unit scale.

Predicting response to primary chemotherapy: gene expression profiling of paraffin-embedded core biopsy tissue

PURPOSE

Primary chemotherapy provides an ideal opportunity to correlate gene expression with response to treatment. We used paraffin-embedded core biopsies from a completed phase II trial to identify genes that correlate with response to primary chemotherapy.

PATIENTS AND METHODS

Patients with newly diagnosed stage II or III breast cancer were treated with sequential doxorubicin 75 mg/M2 q2 wks x 3 and docetaxel 40 mg/M2 weekly x 6; treatment order was randomly assigned. Pretreatment core biopsy samples were interrogated for genes that might correlate with pathologic complete response (pCR). In addition to the individual genes, the correlation of the Oncotype DX Recurrence Score with pCR was examined.

RESULTS

Of 70 patients enrolled in the parent trial, core biopsies samples with sufficient RNA for gene analyses were available from 45 patients; 9 (20%) had inflammatory breast cancer (IBC). Six (14%) patients achieved a pCR. Twenty-two of the 274 candidate genes assessed correlated with pCR (p < 0.05). Genes correlating with pCR could be grouped into three large clusters: angiogenesis-related genes, proliferation related genes, and invasion-related genes. Expression of estrogen receptor (ER)-related genes and Recurrence Score did not correlate with pCR. In an exploratory analysis we compared gene expression in IBC to non-inflammatory breast cancer; twenty-four (9%) of the genes were differentially expressed (p < 0.05), 5 were upregulated and 19 were downregulated in IBC.

CONCLUSION

Gene expression analysis on core biopsy samples is feasible and identifies candidate genes that correlate with pCR to primary chemotherapy. Gene expression in IBC differs significantly from noninflammatory breast cancer.

DNA arrays as predictors of efficacy of adjuvant/neoadjuvant chemotherapy in breast cancer patients: current data and issues on study design

Chemotherapy provides variable benefit to patients with breast cancer, with usually modest but occasionally severe side effects. Hence, there is a need to identify predictive biomarkers for its efficacy. DNA arrays have been used in this setting as potential novel predictive diagnostic tools. Several gene signatures and single gene markers were proposed to predict response to chemotherapy. Although this technology offers interesting perspectives through large-scale analysis of the transcriptome, its ability to identify clinically relevant predictors is highly dependent on study design. In the present manuscript, we will review currently available results of breast cancer pharmacogenomics and focus on aspects of study design that are critical to reliably identify predictive biomarkers using DNA array technology. We will discuss whether studies should be done in the overall, unselected breast cancer population or in specific homogeneous molecular subclasses. Next, we will compare advantages and limitations of cohort-based and case-control studies. The choice of end-point to discriminate between sensitive and resistant patients will also be examined.

Pharmacogenomic predictor of sensitivity to preoperative chemotherapy with paclitaxel and fluorouracil, doxorubicin, and cyclophosphamide in breast cancer

PURPOSE

We developed a multigene predictor of pathologic complete response (pCR) to preoperative weekly paclitaxel and fluorouracil-doxorubicin-cyclophosphamide (T/FAC) chemotherapy and assessed its predictive accuracy on independent cases.

PATIENTS AND METHODS

One hundred thirty-three patients with stage I-III breast cancer were included. Pretreatment gene expression profiling was performed with oligonecleotide microarrays on fine-needle aspiration specimens. We developed predictors of pCR from 82 cases and assessed accuracy on 51 independent cases.

RESULTS

Overall pCR rate was 26% in both cohorts. In the training set, 56 probes were identified as differentially expressed between pCR versus residual disease, at a false discovery rate of 1%. We examined the performance of 780 distinct classifiers (set of genes + prediction algorithm) in full cross-validation. Many predictors performed equally well. A nominally best 30-probe set Diagonal Linear Discriminant Analysis classifier was selected for independent validation. It showed significantly higher sensitivity (92% v 61%) than a clinical predictor including age, grade, and estrogen receptor status. The negative predictive value (96% v 86%) and area under the curve (0.877 v 0.811) were nominally better but not statistically significant. The combination of genomic and clinical information yielded a predictor not significantly different from the genomic predictor alone. In 31 samples, RNA was hybridized in replicate with resulting predictions that were 97% concordant.

CONCLUSION

A 30-probe set pharmacogenomic predictor predicted pCR to T/FAC chemotherapy with high sensitivity and negative predictive value. This test correctly identified all but one of the patients who achieved pCR (12 of 13 patients) and all but one of those who were predicted to have residual disease had residual cancer (27 of 28 patients).

Identical probes on different high-density oligonucleotide microarrays can produce different measurements of gene expression

Background

There are many potential sources of variability in a microarray experiment. Variation can arise from many aspects of the collection and processing of samples for gene expression analysis. Oligonucleotide-based arrays are thought to minimize one source of variability as identical oligonucleotides are expected to recognize the same transcripts during hybridization.

Results

We demonstrate that although the probes on the U133A GeneChip arrays are identical in sequence to probes designed for the U133 Plus 2.0 arrays the values obtained from an experimental hybridization can be quite different. Nearly half of the probesets in common between the two array types can produce slightly different values from the same sample. Nearly 70% of the individual probes in these probesets produced array specific differences.

Conclusion

The context of the probe may also contribute some bias to the final measured value of gene expression. At a minimum, this should add an extra level of caution when considering the direct comparison of experiments performed in two microarray formats. More importantly, this suggests that it may not be possible to know which value is the most accurate representation of a biological sample when comparing two formats.

Reproducibility of Gene Expression Signature–Based Predictions in Replicate Experiments

PURPOSE

The goals of this analysis were to (a) determine concordance of gene expression results from replicate experiments, (b) examine prediction agreement of multigene predictors on replicate data, and (c) assess the robustness of prediction results in the face of noise.

PATIENTS AND METHODS

Affymetrix U133A gene chips were used for gene expression profiling of 97 fine-needle aspiration biopsies from breast cancer. Thirty-five cases were profiled in replicates: 17 within the same laboratory, 11 in two different laboratories, and 15 to assess manual and robotic labeling. We used data from 62 cases to develop 111 distinct pharmacogenomic predictors of response to therapy. These were tested on cases profiled in duplicates to determine prediction agreement and accuracy. To evaluate the robustness of the pharmacogenomic predictors, we also introduced random noise into the informative genes in one half of the replicates.

RESULTS

The average concordance correlation coefficient was 0.978 (range, 0.96-0.99) for intralaboratory replicates, 0.962 (range, 0.94-0.98) for between-laboratory replicates, and 0.971 (range, 0.93-0.99) for manual versus robotic labeling. The mean % prediction agreement on replicate data was 97% (95% CI, 0.96-0.98; SD, 0.006), 92% (95% CI, 0.90-0.93; SD, 0.009), and 94% (95% CI, 0.92-0.95; SD, 0.008) for support vector machines, diagonal linear discriminant analysis, and k-nearest neighbor prediction methods, respectively. Mean accuracy in the test set was 77% (95% CI, 0.74-0.79; SD, 0.014), 66% (95% CI, 0.63-0.73; SD, 0.015), and 64% (95% CI, 0.60-0.67; SD, 0.016), respectively.

CONCLUSION

Gene expression results obtained with Affymetrix U133A chips are highly reproducible within and across two high-volume laboratories. Pharmacogenomic predictions yielded >90% agreement in replicate data.

Breast Cancer Molecular Subtypes Respond Differently to Preoperative Chemotherapy

PURPOSE

Molecular classification of breast cancer has been proposed based on gene expression profiles of human tumors. Luminal, basal-like, normal-like, and erbB2+ subgroups were identified and were shown to have different prognoses. The goal of this research was to determine if these different molecular subtypes of breast cancer also respond differently to preoperative chemotherapy.

EXPERIMENTAL DESIGN

Fine needle aspirations of 82 breast cancers were obtained before starting preoperative paclitaxel followed by 5-fluorouracil, doxorubicin, and cyclophosphamide chemotherapy. Gene expression profiling was done with Affymetrix U133A microarrays and the previously reported "breast intrinsic" gene set was used for hierarchical clustering and multidimensional scaling to assign molecular class.

RESULTS

The basal-like and erbB2+ subgroups were associated with the highest rates of pathologic complete response (CR), 45% [95% confidence interval (95% CI), 24-68] and 45% (95% CI, 23-68), respectively, whereas the luminal tumors had a pathologic CR rate of 6% (95% CI, 1-21). No pathologic CR was observed among the normal-like cancers (95% CI, 0-31). Molecular class was not independent of conventional cliniocopathologic predictors of response such as estrogen receptor status and nuclear grade. None of the 61 genes associated with pathologic CR in the basal-like group were associated with pathologic CR in the erbB2+ group, suggesting that the molecular mechanisms of chemotherapy sensitivity may vary between these two estrogen receptor-negative subtypes.

CONCLUSIONS

The basal-like and erbB2+ subtypes of breast cancer are more sensitive to paclitaxel- and doxorubicin-containing preoperative chemotherapy than the luminal and normal-like cancers.