Construction and Evaluation of Robust Interpretation Models for Breast Cancer Metastasis Prediction

Interpretability of machine learning (ML) models represents the extent to which a model's decision-making process can be understood by model developers and/or end users. Transcriptomics-based cancer prognosis models, for example, while achieving good accuracy, are usually hard to interpret, due to the high-dimensional feature space and the complexity of models. As interpretability is critical for the transparency and fairness of ML models, several algorithms have been proposed to improve the interpretability of arbitrary classifiers. However, evaluation of these algorithms often requires substantial domain knowledge. Here, we propose a breast cancer metastasis prediction model using a very small number of biologically interpretable features, and a simple yet novel model interpretation approach that can provide personalized interpretations. In addition, we contributed, to the best of our knowledge, the first method to quantitatively compare different interpretation algorithms. Experimental results show that our model not only achieved competitive prediction accuracy, but also higher inter-classifier interpretation consistency than state-of-the-art interpretation methods. Importantly, our interpretation results can improve the generalizability of the prediction models. Overall, this work provides several novel ideas to construct and evaluate interpretable ML models that can be valuable to both the cancer machine learning community and related application domains.

Phosphorylated MED1 links transcription recycling and cancer growth

Mediator activates RNA polymerase II (Pol II) function during transcription, but it remains unclear whether Mediator is able to travel with Pol II and regulate Pol II transcription beyond the initiation and early elongation steps. By using in vitro and in vivo transcription recycling assays, we find that human Mediator 1 (MED1), when phosphorylated at the mammal-specific threonine 1032 by cyclin-dependent kinase 9 (CDK9), dynamically moves along with Pol II throughout the transcribed genes to drive Pol II recycling after the initial round of transcription. Mechanistically, MED31 mediates the recycling of phosphorylated MED1 and Pol II, enhancing mRNA output during the transcription recycling process. Importantly, MED1 phosphorylation increases during prostate cancer progression to the lethal phase, and pharmacological inhibition of CDK9 decreases prostate tumor growth by decreasing MED1 phosphorylation and Pol II recycling. Our results reveal a novel role of MED1 in Pol II transcription and identify phosphorylated MED1 as a targetable driver of dysregulated Pol II recycling in cancer.

Toward uncharted territory of cellular heterogeneity: advances and applications of single-cell RNA-seq

Among single-cell analysis technologies, single-cell RNA-seq (scRNA-seq) has been one of the front runners in technical inventions. Since its induction, scRNA-seq has been well received and undergone many fast-paced technical improvements in cDNA synthesis and amplification, processing and alignment of next generation sequencing reads, differentially expressed gene calling, cell clustering, subpopulation identification, and developmental trajectory prediction. scRNA-seq has been exponentially applied to study global transcriptional profiles in all cell types in humans and animal models, healthy or with diseases, including cancer. Accumulative novel subtypes and rare subpopulations have been discovered as potential underlying mechanisms of stochasticity, differentiation, proliferation, tumorigenesis, and aging. scRNA-seq has gradually revealed the uncharted territory of cellular heterogeneity in transcriptomes and developed novel therapeutic approaches for biomedical applications. This review of the advancement of scRNA-seq methods provides an exploratory guide of the quickly evolving technical landscape and insights of focused features and strengths in each prominent area of progress.

Comparative evaluation of network features for the prediction of breast cancer metastasis

Background

Discovering a highly accurate and robust gene signature for the prediction of breast cancer metastasis from gene expression profiling of primary tumors is one of the most challenging tasks to reduce the number of deaths in women. Due to the limited success of gene-based features in achieving satisfactory prediction accuracy, many methodologies have been proposed in recent years to develop network-based features by integrating network information with gene expression. However, evaluation results are inconsistent to confirm the effectiveness of network-based features, because of many confounding factors involved in classification model learning process, such as data normalization, dimension reduction, and feature selection. An unbiased comparative evaluation is essential for uncovering the strength of network-based features.

Methods

In this study, we compared several types of network-based features obtained using different mathematical operators (Mean, Maximum, Minimum, Median, Variance) on geneset (i.e., a gene and its’ neighbors in the network) in protein-protein interaction network and gene co-expression network for their ability in predicting breast cancer metastasis using gene expression data from more than 10 patient cohorts.

Results

While network-based features are usually statistically more significant than gene-based feature, a consistent improvement of prediction performance using network-based features requires a substantial number of patients in the dataset. In contrary to many previous reports, no evidence was found to support the robustness of network-based features and we argue some of the robustness may be due to the inherent bias associated with node degree in the network. In addition, different types of network features seem to cover different pathways and are complementary to each other. Consequently, an ensemble classifier combining different network features was proposed and was found to significantly outperform classifiers based on gene-based feature or any single type of network-based features.

Conclusions

Network-based features and their combination show promise for improving the prediction of breast cancer metastasis but may require a large amount of training data. Robustness claim of network-based features needs to be re-examined with network node degree and other confounding factors in consideration.

Deep sequencing of HPV E6/E7 genes reveals loss of genotypic diversity and gain of clonal dominance in high-grade intraepithelial lesions of the cervix

Background

Human papillomavirus (HPV) is the carcinogen of almost all invasive cervical cancer and a major cause of oral and other anogenital malignancies. HPV genotyping by dideoxy (Sanger) sequencing is currently the reference method of choice for clinical diagnostics. However, for samples with multiple HPV infections, genotype identification is singular and occasionally imprecise or indeterminable due to overlapping chromatograms. Our aim was to explore and compare HPV metagenomes in abnormal cervical cytology by deep sequencing for correlation with disease states.

Results

Low- and high-grade intraepithelial lesion (LSIL and HSIL) cytology samples were DNA extracted for PCR-amplification of the HPV E6/E7 genes. HPV+ samples were sequenced by dideoxy and deep methods. Deep sequencing revealed ~60% of all samples (n = 72) were multi-HPV infected. Among LSIL samples (n = 43), 27 different genotypes were found. The 3 dominant (most abundant) genotypes were: HPV-39, 11/43 (26%); -16, 9/43 (21%); and -35, 4/43 (9%). Among HSIL (n = 29), 17 HPV genotypes were identified; the 3 dominant genotypes were: HPV-16, 21/29 (72%); -35, 4/29 (14%); and -39, 3/29 (10%). Phylogenetically, type-specific E6/E7 genetic distances correlated with carcinogenic potential. Species diversity analysis between LSIL and HSIL revealed loss of HPV diversity and domination by HPV-16 in HSIL samples.

Conclusions

Deep sequencing resolves HPV genotype composition within multi-infected cervical cytology. Biodiversity analysis reveals loss of diversity and gain of dominance by carcinogenic genotypes in high-grade cytology. Metagenomic profiles may therefore serve as a biomarker of disease severity and a population surveillance tool for emerging genotypes.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-017-3612-y) contains supplementary material, which is available to authorized users.

Promoter Methylation Modulates Indoleamine 2,3-Dioxygenase 1 Induction by Activated T Cells in Human Breast Cancers

Triple-negative breast cancer (TNBC) cells are modulated in reaction to tumor-infiltrating lymphocytes. However, their specific responses to this immune pressure are unknown. In order to address this question, we first used mRNA sequencing to compare the immunophenotype of the TNBC cell line MDA-MB-231 and the luminal breast cancer cell line MCF7 after both were cocultured with activated human T cells. Despite similarities in the cytokine-induced immune signatures of the two cell lines, MDA-MD-231 cells were able to transcribe more IDO1 than MCF7 cells. The two cell lines had similar upstream JAK/STAT1 signaling and IDO1 mRNA stability. However, using a series of breast cancer cell lines, IFNγ stimulated IDO1 protein expression and enzymatic activity only in ER^-, not ER⁺, cell lines. Treatment with 5-aza-deoxycytidine reversed the suppression of IDO1 expression in MCF7 cells, suggesting that DNA methylation was potentially involved in IDO1 induction. By analyzing several breast cancer datasets, we discovered subtype-specific mRNA and promoter methylation differences in IDO1, with TNBC/basal subtypes exhibiting lower methylation/higher expression and ER⁺/luminal subtypes exhibiting higher methylation/lower expression. We confirmed this trend of IDO1 methylation by bisulfite pyrosequencing breast cancer cell lines and an independent cohort of primary breast tumors. Taken together, these findings suggest that IDO1 promoter methylation regulates anti-immune responses in breast cancer subtypes and could be used as a predictive biomarker for IDO1 inhibitor-based immunotherapy. Cancer Immunol Res; 5(4); 330-44. ©2017 AACR.

Methyl-binding DNA capture Sequencing for Patient Tissues

Methylation is one of the essential epigenetic modifications to the DNA, which is responsible for the precise regulation of genes required for stable development and differentiation of different tissue types. Dysregulation of this process is often the hallmark of various diseases like cancer. Here, we outline one of the recent sequencing techniques, Methyl-Binding DNA Capture sequencing (MBDCap-seq), used to quantify methylation in various normal and disease tissues for large patient cohorts. We describe a detailed protocol of this affinity enrichment approach along with a bioinformatics pipeline to achieve optimal quantification. This technique has been used to sequence hundreds of patients across various cancer types as a part of the 1,000 methylome project (Cancer Methylome System).

Histone Deacetylase HDAC8 Promotes Insulin Resistance and β-Catenin Activation in NAFLD-Associated Hepatocellular Carcinoma

The growing epidemic of obesity, which causes nonalcoholic fatty liver disease (NAFLD) and the more severe phenotype nonalcoholic steatohepatitis (NASH), has paralleled the increasing incidence of hepatocellular carcinoma (HCC). Accumulating evidence demonstrates that overnutrition and metabolic pathways can trigger modifications of DNA and histones via deregulation of chromatin modifiers, resulting in aberrant transcriptional activity. However, the epigenetic regulation of HCC development in NAFLD remains obscure. Here, we uncover key epigenetic regulators using both dietary and genetic obesity-promoted HCC models through quantitative expression profiling and characterize the oncogenic activities of histone deacetylase HDAC8 in NAFLD-associated hepatocarcinogenesis. HDAC8 is directly upregulated by the lipogenic transcription factor SREBP-1 where they are coexpressed in dietary obesity models of NASH and HCC. Lentiviral-mediated HDAC8 attenuation in vivo reversed insulin resistance and reduced NAFLD-associated tumorigenicity. HDAC8 modulation by genetic and pharmacologic approaches inhibited p53/p21-mediated apoptosis and G2-M phase cell-cycle arrest and stimulated β-catenin-dependent cell proliferation. Mechanistically, HDAC8 physically interacted with the chromatin modifier EZH2 to concordantly repress Wnt antagonists via histone H4 deacetylation and H3 lysine 27 trimethylation. In human NAFLD-associated HCC, levels of SREBP-1, HDAC8, EZH2, H4 deacetylation, H3K27me3, and active β-catenin were all correlated positively in tumors compared with nontumor tissues. Overall, our findings show how HDAC8 drives NAFLD-associated hepatocarcinogenesis, offering a novel epigenetic target to prevent or treat HCC in obese patients.

Genome-wide DNA methylation profiling reveals parity-associated hypermethylation of FOXA1

Early pregnancy in women by the age of 20 is known to have a profound effect on reduction of lifelong breast cancer risk as compared to their nulliparous counterparts. Additional pregnancies further enhance the protection against breast cancer development. Nationwide trend of delayed pregnancy may contribute to the recently reported increase in the incidence of advanced breast cancer among young women in this country. The underlying mechanism for the parity-associated reduction of breast cancer risk is not clearly understood. The purpose of the current study is to use whole-genome DNA methylation profiling to explore a potential association between parity and epigenetic changes in breast tissue from women with early parity and nulliparity. Breast tissue was collected from age-matched cancer-free women with early parity (age < 20; n = 15) or nulliparity (n = 13). The methyl-CpG binding domain-based capture-sequencing technology was used for whole-genome DNA methylation profiling. Potential parity-associated hypermethylated genes were further verified by locus-specific pyrosequencing, using an expanded cohort of parous (n = 19) and nulliparous (n = 16) women that included the initial samples used in the global analysis. Our study identified six genes that are hypermethylated in the parous group (P < 0.05). Pyrosequencing confirmed parity-associated hypermethylation at multiple CpG islands of the FOXA1 gene, which encodes a pioneer factor that facilitates chromatin binding of estrogen receptor α. Our work identifies several potential methylation biomarkers for parity-associated breast cancer risk assessment. In addition, the results are consistent with the notion that parity-associated epigenetic silencing of FOXA1 contributes to long-term attenuation of the estrogenic impact on breast cancer development.

QChIPat: a quantitative method to identify distinct binding patterns for two biological ChIP-seq samples in different experimental conditions

Background

Many computational programs have been developed to identify enriched regions for a single biological ChIP-seq sample. Given that many biological questions are often asked to compare the difference between two different conditions, it is important to develop new programs that address the comparison of two biological ChIP-seq samples. Despite several programs designed to address this question, these programs suffer from some drawbacks, such as inability to distinguish whether the identified differential enriched regions are indeed significantly enriched, lack of distinguishing binding patterns, and neglect of the normalization between samples.

Results

In this study, we developed a novel quantitative method for comparing two biological ChIP-seq samples, called QChIPat. Our method employs a new global normalization method: nonparametric empirical Bayes (NEB) correction normalization, utilizes pre-defined enriched regions identified from single-sample peak calling programs, uses statistical methods to define differential enriched regions, then defines binding (histone modification) pattern information for those differential enriched regions. Our program was tested on a benchmark data: histone modifications data used by ChIPDiffs. It was then applied on two study cases: one to identify differential histone modification sites for ChIP-seq of H3K27me3 and H3K9me2 data in AKT1-transfected MCF10A cells; the other to identify differential binding sites for ChIP-seq of TCF7L2 data in MCF7 and PANC1 cells.

Conclusions

Several advantages of our program include: 1) it considers a control (or input) experiment; 2) it incorporates a novel global normalization strategy: nonparametric empirical Bayes correction normalization; 3) it provides the binding pattern information among different enriched regions. QChIPat is implemented in R, Perl and C++, and has been tested under Linux. The R package is available at http://motif.bmi.ohio-state.edu/QChIPat.

CMS: a web-based system for visualization and analysis of genome-wide methylation data of human cancers

BACKGROUND

DNA methylation of promoter CpG islands is associated with gene suppression, and its unique genome-wide profiles have been linked to tumor progression. Coupled with high-throughput sequencing technologies, it can now efficiently determine genome-wide methylation profiles in cancer cells. Also, experimental and computational technologies make it possible to find the functional relationship between cancer-specific methylation patterns and their clinicopathological parameters.

METHODOLOGY/PRINCIPAL FINDINGS

Cancer methylome system (CMS) is a web-based database application designed for the visualization, comparison and statistical analysis of human cancer-specific DNA methylation. Methylation intensities were obtained from MBDCap-sequencing, pre-processed and stored in the database. 191 patient samples (169 tumor and 22 normal specimen) and 41 breast cancer cell-lines are deposited in the database, comprising about 6.6 billion uniquely mapped sequence reads. This provides comprehensive and genome-wide epigenetic portraits of human breast cancer and endometrial cancer to date. Two views are proposed for users to better understand methylation structure at the genomic level or systemic methylation alteration at the gene level. In addition, a variety of annotation tracks are provided to cover genomic information. CMS includes important analytic functions for interpretation of methylation data, such as the detection of differentially methylated regions, statistical calculation of global methylation intensities, multiple gene sets of biologically significant categories, interactivity with UCSC via custom-track data. We also present examples of discoveries utilizing the framework.

CONCLUSIONS/SIGNIFICANCE

CMS provides visualization and analytic functions for cancer methylome datasets. A comprehensive collection of datasets, a variety of embedded analytic functions and extensive applications with biological and translational significance make this system powerful and unique in cancer methylation research. CMS is freely accessible at: http://cbbiweb.uthscsa.edu/KMethylomes/.

A modulator based regulatory network for ERα signaling pathway

Background

Estrogens control multiple functions of hormone-responsive breast cancer cells. They regulate diverse physiological processes in various tissues through genomic and non-genomic mechanisms that result in activation or repression of gene expression. Transcription regulation upon estrogen stimulation is a critical biological process underlying the onset and progress of the majority of breast cancer. ERα requires distinct co-regulator or modulators for efficient transcriptional regulation, and they form a regulatory network. Knowing this regulatory network will enable systematic study of the effect of ERα on breast cancer.

Methods

To investigate the regulatory network of ERα and discover novel modulators of ERα functions, we proposed an analytical method based on a linear regression model to identify translational modulators and their network relationships. In the network analysis, a group of specific modulator and target genes were selected according to the functionality of modulator and the ERα binding. Network formed from targets genes with ERα binding was called ERα genomic regulatory network; while network formed from targets genes without ERα binding was called ERα non-genomic regulatory network. Considering the active or repressive function of ERα, active or repressive function of a modulator, and agonist or antagonist effect of a modulator on ERα, the ERα/modulator/target relationships were categorized into 27 classes.

Results

Using the gene expression data and ERα Chip-seq data from the MCF-7 cell line, the ERα genomic/non-genomic regulatory networks were built by merging ERα/ modulator/target triplets (TF, M, T), where TF refers to the ERα, M refers to the modulator, and T refers to the target. Comparing these two networks, ERα non-genomic network has lower FDR than the genomic network. In order to validate these two networks, the same network analysis was performed in the gene expression data from the ZR-75.1 cell. The network overlap analysis between two cancer cells showed 1% overlap for the ERα genomic regulatory network, but 4% overlap for the non-genomic regulatory network.

Conclusions

We proposed a novel approach to infer the ERα/modulator/target relationships, and construct the genomic/non-genomic regulatory networks in two cancer cells. We found that the non-genomic regulatory network is more reliable than the genomic regulatory network.

The influence of cis-regulatory elements on DNA methylation fidelity

It is now established that, as compared to normal cells, the cancer cell genome has an overall inverse distribution of DNA methylation (“methylome”), i.e., predominant hypomethylation and localized hypermethylation, within “CpG islands” (CGIs). Moreover, although cancer cells have reduced methylation “fidelity” and genomic instability, accurate maintenance of aberrant methylomes that underlie malignant phenotypes remains necessary. However, the mechanism(s) of cancer methylome maintenance remains largely unknown. Here, we assessed CGI methylation patterns propagated over 1, 3, and 5 divisions of A2780 ovarian cancer cells, concurrent with exposure to the DNA cross-linking chemotherapeutic cisplatin, and observed cell generation-successive increases in total hyper- and hypo-methylated CGIs. Empirical Bayesian modeling revealed five distinct modes of methylation propagation: (1) heritable (i.e., unchanged) high- methylation (1186 probe loci in CGI microarray); (2) heritable (i.e., unchanged) low-methylation (286 loci); (3) stochastic hypermethylation (i.e., progressively increased, 243 loci); (4) stochastic hypomethylation (i.e., progressively decreased, 247 loci); and (5) considerable “random” methylation (582 loci). These results support a “stochastic model” of DNA methylation equilibrium deriving from the efficiency of two distinct processes, methylation maintenance and de novo methylation. A role for cis-regulatory elements in methylation fidelity was also demonstrated by highly significant (p<2.2×10⁻⁵) enrichment of transcription factor binding sites in CGI probe loci showing heritably high (118 elements) and low (47 elements) methylation, and also in loci demonstrating stochastic hyper-(30 elements) and hypo-(31 elements) methylation. Notably, loci having “random” methylation heritability displayed nearly no enrichment. These results demonstrate an influence of cis-regulatory elements on the nonrandom propagation of both strictly heritable and stochastically heritable CGIs.

Identifying hypermethylated CpG islands using a quantile regression model

BACKGROUND

DNA methylation has been shown to play an important role in the silencing of tumor suppressor genes in various tumor types. In order to have a system-wide understanding of the methylation changes that occur in tumors, we have developed a differential methylation hybridization (DMH) protocol that can simultaneously assay the methylation status of all known CpG islands (CGIs) using microarray technologies. A large percentage of signals obtained from microarrays can be attributed to various measurable and unmeasurable confounding factors unrelated to the biological question at hand. In order to correct the bias due to noise, we first implemented a quantile regression model, with a quantile level equal to 75%, to identify hypermethylated CGIs in an earlier work. As a proof of concept, we applied this model to methylation microarray data generated from breast cancer cell lines. However, we were unsure whether 75% was the best quantile level for identifying hypermethylated CGIs. In this paper, we attempt to determine which quantile level should be used to identify hypermethylated CGIs and their associated genes.

RESULTS

We introduce three statistical measurements to compare the performance of the proposed quantile regression model at different quantile levels (95%, 90%, 85%, 80%, 75%, 70%, 65%, 60%), using known methylated genes and unmethylated housekeeping genes reported in breast cancer cell lines and ovarian cancer patients. Our results show that the quantile levels ranging from 80% to 90% are better at identifying known methylated and unmethylated genes.

CONCLUSIONS

In this paper, we propose to use a quantile regression model to identify hypermethylated CGIs by incorporating probe effects to account for noise due to unmeasurable factors. Our model can efficiently identify hypermethylated CGIs in both breast and ovarian cancer data.

Multivalent epigenetic marks confer microenvironment-responsive epigenetic plasticity to ovarian cancer cells

"Epigenetic plasticity" refers to the capability of mammalian cells to alter their differentiation status via chromatin remodeling-associated alterations in gene expression. While epigenetic plasticity has been best associated with lineage commitment of embryonic stem cells, recent studies have demonstrated chromatin remodeling even in terminally differentiated normal cells, and advanced-stage melanoma and breast cancer cells, in context-dependent responses to alterations in their microenvironment. In the current study, we extend this attribute of epigenetic plasticity to aggressive ovarian cancer cells, by using an integrative approach to associate cellular phenotypes with chromatin modifications ("ChIP-chip") and mRNA and microRNA expression. While we identified numerous gene promoters possessing the well-known "bivalent mark" of H3K27me3/H3K4me2, we also report 14 distinct, lesser-known bi-, tri-, and tetravalent combinations of activating and repressive chromatin modifications, in platinum-resistant CP70 ovarian cancer cells. The vast majority (>90%) of all the histone marks studied localized to regions within 2000 bp of transcription start sites, supporting a role in gene regulation. Upon a simple alteration in the microenvironment, transition from two- to three-dimensional culture, an increase (17% to 38%) in repressive-only marked promoters was observed, concomitant with a decrease (31% to 21%) in multivalent (i.e., juxtaposed permissive and repressive histone marked) promoters. Like embryonic/tissue stem and other (non-ovarian) carcinoma cells, ovarian cancer cell epigenetic plasticity reflects an inherent transcriptional flexibility for context-responsive alterations in phenotype. It is possible that this plasticity could be therapeutically exploited for the management of this lethal gynecologic malignancy.

Identifying differentially methylated genes using mixed effect and generalized least square models

Background

DNA methylation plays an important role in the process of tumorigenesis. Identifying differentially methylated genes or CpG islands (CGIs) associated with genes between two tumor subtypes is thus an important biological question. The methylation status of all CGIs in the whole genome can be assayed with differential methylation hybridization (DMH) microarrays. However, patient samples or cell lines are heterogeneous, so their methylation pattern may be very different. In addition, neighboring probes at each CGI are correlated. How these factors affect the analysis of DMH data is unknown.

Results

We propose a new method for identifying differentially methylated (DM) genes by identifying the associated DM CGI(s). At each CGI, we implement four different mixed effect and generalized least square models to identify DM genes between two groups. We compare four models with a simple least square regression model to study the impact of incorporating random effects and correlations.

Conclusions

We demonstrate that the inclusion (or exclusion) of random effects and the choice of correlation structures can significantly affect the results of the data analysis. We also assess the false discovery rate of different models using CGIs associated with housekeeping genes.

An integrative ChIP-chip and gene expression profiling to model SMAD regulatory modules

Background

The TGF-β/SMAD pathway is part of a broader signaling network in which crosstalk between pathways occurs. While the molecular mechanisms of TGF-β/SMAD signaling pathway have been studied in detail, the global networks downstream of SMAD remain largely unknown. The regulatory effect of SMAD complex likely depends on transcriptional modules, in which the SMAD binding elements and partner transcription factor binding sites (SMAD modules) are present in specific context.

Results

To address this question and develop a computational model for SMAD modules, we simultaneously performed chromatin immunoprecipitation followed by microarray analysis (ChIP-chip) and mRNA expression profiling to identify TGF-β/SMAD regulated and synchronously coexpressed gene sets in ovarian surface epithelium. Intersecting the ChIP-chip and gene expression data yielded 150 direct targets, of which 141 were grouped into 3 co-expressed gene sets (sustained up-regulated, transient up-regulated and down-regulated), based on their temporal changes in expression after TGF-β activation. We developed a data-mining method driven by the Random Forest algorithm to model SMAD transcriptional modules in the target sequences. The predicted SMAD modules contain SMAD binding element and up to 2 of 7 other transcription factor binding sites (E2F, P53, LEF1, ELK1, COUPTF, PAX4 and DR1).

Conclusion

Together, the computational results further the understanding of the interactions between SMAD and other transcription factors at specific target promoters, and provide the basis for more targeted experimental verification of the co-regulatory modules.

Integrated analysis of DNA methylation and gene expression reveals specific signaling pathways associated with platinum resistance in ovarian cancer

Background

Cisplatin and carboplatin are the primary first-line therapies for the treatment of ovarian cancer. However, resistance to these platinum-based drugs occurs in the large majority of initially responsive tumors, resulting in fully chemoresistant, fatal disease. Although the precise mechanism(s) underlying the development of platinum resistance in late-stage ovarian cancer patients currently remains unknown, CpG-island (CGI) methylation, a phenomenon strongly associated with aberrant gene silencing and ovarian tumorigenesis, may contribute to this devastating condition.

Methods

To model the onset of drug resistance, and investigate DNA methylation and gene expression alterations associated with platinum resistance, we treated clonally derived, drug-sensitive A2780 epithelial ovarian cancer cells with increasing concentrations of cisplatin. After several cycles of drug selection, the isogenic drug-sensitive and -resistant pairs were subjected to global CGI methylation and mRNA expression microarray analyses. To identify chemoresistance-associated, biological pathways likely impacted by DNA methylation, promoter CGI methylation and mRNA expression profiles were integrated and subjected to pathway enrichment analysis.

Results

Promoter CGI methylation revealed a positive association (Spearman correlation of 0.99) between the total number of hypermethylated CGIs and GI₅₀ values (i.e., increased drug resistance) following successive cisplatin treatment cycles. In accord with that result, chemoresistance was reversible by DNA methylation inhibitors. Pathway enrichment analysis revealed hypermethylation-mediated repression of cell adhesion and tight junction pathways and hypomethylation-mediated activation of the cell growth-promoting pathways PI3K/Akt, TGF-beta, and cell cycle progression, which may contribute to the onset of chemoresistance in ovarian cancer cells.

Conclusion

Selective epigenetic disruption of distinct biological pathways was observed during development of platinum resistance in ovarian cancer. Integrated analysis of DNA methylation and gene expression may allow for the identification of new therapeutic targets and/or biomarkers prognostic of disease response. Finally, our results suggest that epigenetic therapies may facilitate the prevention or reversal of transcriptional repression responsible for chemoresistance and the restoration of sensitivity to platinum-based chemotherapeutics.

Probe signal correction for differential methylation hybridization experiments

BACKGROUND

Non-biological signal (or noise) has been the bane of microarray analysis. Hybridization effects related to probe-sequence composition and DNA dye-probe interactions have been observed in differential methylation hybridization (DMH) microarray experiments as well as other effects inherent to the DMH protocol.

RESULTS

We suggest two models to correct for non-biologically relevant probe signal with an overarching focus on probe-sequence composition. The estimated effects are evaluated and the strengths of the models are considered in the context of DMH analyses.

CONCLUSION

The majority of estimated parameters were statistically significant in all considered models. Model selection for signal correction is based on interpretation of the estimated values and their biological significance.

A Poisson mixture model to identify changes in RNA polymerase II binding quantity using high-throughput sequencing technology

We present a mixture model-based analysis for identifying differences in the distribution of RNA polymerase II (Pol II) in transcribed regions, measured using ChIP-seq (chromatin immunoprecipitation following massively parallel sequencing technology). The statistical model assumes that the number of Pol II-targeted sequences contained within each genomic region follows a Poisson distribution. A Poisson mixture model was then developed to distinguish Pol II binding changes in transcribed region using an empirical approach and an expectation-maximization (EM) algorithm developed for estimation and inference. In order to achieve a global maximum in the M-step, a particle swarm optimization (PSO) was implemented. We applied this model to Pol II binding data generated from hormone-dependent MCF7 breast cancer cells and antiestrogen-resistant MCF7 breast cancer cells before and after treatment with 17β-estradiol (E2). We determined that in the hormone-dependent cells, ~9.9% (2527) genes showed significant changes in Pol II binding after E2 treatment. However, only ~0.7% (172) genes displayed significant Pol II binding changes in E2-treated antiestrogen-resistant cells. These results show that a Poisson mixture model can be used to analyze ChIP-seq data.

Methylation Linear Discriminant Analysis (MLDA) for identifying differentially methylated CpG islands

Background

Hypermethylation of promoter CpG islands is strongly correlated to transcriptional gene silencing and epigenetic maintenance of the silenced state. As well as its role in tumor development, CpG island methylation contributes to the acquisition of resistance to chemotherapy. Differential Methylation Hybridisation (DMH) is one technique used for genome-wide DNA methylation analysis. The study of such microarray data sets should ideally account for the specific biological features of DNA methylation and the non-symmetrical distribution of the ratios of unmethylated and methylated sequences hybridised on the array. We have therefore developed a novel algorithm tailored to this type of data, Methylation Linear Discriminant Analysis (MLDA).

Results

MLDA was programmed in R (version 2.7.0) and the package is available at CRAN [1]. This approach utilizes linear regression models of non-normalised hybridisation data to define methylation status. Log-transformed signal intensities of unmethylated controls on the microarray are used as a reference. The signal intensities of DNA samples digested with methylation sensitive restriction enzymes and mock digested are then transformed to the likelihood of a locus being methylated using this reference. We tested the ability of MLDA to identify loci differentially methylated as analysed by DMH between cisplatin sensitive and resistant ovarian cancer cell lines. MLDA identified 115 differentially methylated loci and 23 out of 26 of these loci have been independently validated by Methylation Specific PCR and/or bisulphite pyrosequencing.

Conclusion

MLDA has advantages for analyzing methylation data from CpG island microarrays, since there is a clear rational for the definition of methylation status, it uses DMH data without between-group normalisation and is less influenced by cross-hybridisation of loci. The MLDA algorithm successfully identified differentially methylated loci between two classes of samples analysed by DMH using CpG island microarrays.

Genome-wide analysis of alternative promoters of human genes using a custom promoter tiling array

Background

Independent lines of evidence suggested that a large fraction of human genes possess multiple promoters driving gene expression from distinct transcription start sites. Understanding which promoter is employed in which cellular context is required to unravel gene regulatory networks within the cell.

Results

We have developed a custom microarray platform that tiles roughly 35,000 alternative putative promoters from nearly 7,000 genes in the human genome. To demonstrate the utility of this array platform, we have analyzed the patterns of promoter usage in 17β-estradiol (E2)-treated and untreated MCF7 cells and show widespread usage of alternative promoters. Most intriguingly, we show that the downstream promoter in E2-sensitive multiple promoter genes tends to be very close to the 3'-terminus of the gene, suggesting exotic mechanisms of expression regulation in these genes.

Conclusion

The usage of alternative promoters greatly multiplies the transcriptional complexity available within the human genome. The fact that many of these promoters are incapable of driving the synthesis of a meaningful protein-encoding transcript further complicates the story.

Identification of novel DNA methylation markers in cervical cancer

Testing for DNA methylation has potential in cancer screening. Most previous studies of DNA methylation in cervical cancer used a candidate gene approach. The aim our study was to identify novel genes that are methylated in cervical cancers and to test their potential in clinical applications. We did a differential methylation hybridization using a CpG island (CGI) microarray containing 8640 CGI tags to uncover methylated genes in squamous cell carcinomas (SCC) of the uterine cervix. Pooled DNA from cancer tissues and normal cervical swabs were used for comparison. Methylation-specific polymerase chain reaction, bisulfite sequencing and reverse transcription polymerase chain reaction were used to confirm the methylation status in cell lines, normal cervices (n = 45), low-grade lesions (n = 45), high-grade lesions (HSIL; n = 58) and invasive squamous cell carcinomas (SCC; n = 22 from swabs and n = 109 from tissues). Human papillomavirus (HPV) was detected using reverse line blots. We reported 6 genes (SOX1, PAX1, LMX1A, NKX6-1, WT1 and ONECUT1) more frequently methylated in SCC tissues (81.5, 94.4, 89.9, 80.4, 77.8 and 20.4%, respectively) than in their normal controls (2.2, 0, 6.7, 11.9, 11.1 and 0%, respectively; p < 0.0001). Parallel testing of HPV and PAX1 methylation in cervical swabs confers an improved sensitivity than HPV testing alone (80% vs. 66%) without compromising specificity (63% vs. 64%) for HSIL/SCC. Testing PAX1 methylation marker alone, the specificity for HSIL/SCC is 99%. The analysis of these novel DNA methylations may be a promising approach for the screening of cervical cancers.

Dominant-negative histone H3 lysine 27 mutant derepresses silenced tumor suppressor genes and reverses the drug-resistant phenotype in cancer cells

Histone modifications and DNA methylation are epigenetic phenomena that play a critical role in many neoplastic processes, including silencing of tumor suppressor genes. One such histone modification, particularly at H3 and H4, is methylation at specific lysine (K) residues. Whereas histone methylation of H3-K9 has been linked to DNA methylation and aberrant gene silencing in cancer cells, no such studies of H3-K27 have been reported. Here, we generated a stable cell line overexpressing a dominant-negative point mutant, H3-K27R, to examine the role of that specific lysine in ovarian cancer. Expression of this construct resulted in loss of methylation at H3-K27, global reduction of DNA methylation, and increased expression of tumor suppressor genes. One of the affected genes, RASSF1, was shown to be a direct target of H3-K27 methylation-mediated silencing. By increasing DNA-platinum adduct formation, indicating increased access of the drug to target DNA sequences, removal of H3-K27 methylation resensitized drug-resistant ovarian cancer cells to the chemotherapeutic agent cisplatin. This increased platinum-DNA access was likely due to relaxation of condensed chromatin. Our results show that overexpression of mutant H3-K27 in mammalian cells represents a novel tool for studying epigenetic mechanisms and the Histone Code Hypothesis in human cancer. Such findings show the significance of H3-K27 methylation as a promising target for epigenetic-based cancer therapies.

Uterine responsiveness to estradiol and DNA methylation are altered by fetal exposure to diethylstilbestrol and methoxychlor in CD-1 mice: effects of low versus high doses

We examined the effects on female CD-1 mice of fetal exposure to low doses of the drug diethylstilbestrol (DES) (0.1 microg/kg/day) and the insecticide methoxychlor (MXC) (10 microg/kg/day) as well as 1000-fold higher doses: 100 microg/kg/day DES and 10,000 microg/kg/day MXC. Pregnant females were administered these chemicals on gestation days 12-18. At 7-8 months of age, female offspring were ovariectomized and implanted for 7 days with a Silastic capsule containing estradiol. Relative to controls, females exposed to the 0.1 microg DES dose showed significantly heavier uteri, while females exposed to the 100 microg DES dose showed significantly lighter uteri. Females exposed prenatally to the 10 microg/kg dose of MXC had significantly heavier uteri relative to females exposed to the 10,000 microg/kg dose of MXC, but neither group differed significantly from controls. Liver weight for females exposed to both doses of DES was significantly greater than controls. Using a microarray approach to analyze DNA methylation, an increase in ribosomal DNA (rDNA) methylation was observed. Sequence data and Southern analysis indicate an increase in 18S rDNA and 45S pre-rDNA methylation in uterine samples exposed prenatally to low and high doses of DES. We thus found opposite effects of fetal exposure to a low and a high dose of DES on the uterine response to estradiol (inverted-U dose-response relationship). In contrast, there was a monotonic dose-response relationship found for prenatal DES exposure on both liver weight and ribosomal DNA hypermethylation.