Distinct longitudinal patterns of urine tumor DNA in patients undergoing surveillance for bladder cancer.

Cystoscopy is the gold standard for surveillance of non-muscle invasive bladder cancer (NMIBC), but the procedure is invasive and has suboptimal accuracy. The aim of this study was to investigate the potential of analyzing urine samples for the presence of urine tumor DNA (utDNA) to replace cystoscopy for surveillance of bladder cancer recurrence. In this longitudinal, prospective, and observational study, 47 patients were followed for recurrence for 2 years, involving analysis of utDNA using the BladMetrix DNA methylation biomarker test at each cystoscopy. In total, utDNA was detected in 21/23 recurrences (91% sensitivity), including 5/5 T1, T2, and carcinoma in situ (CIS) tumors (100%) and 10/12 Ta tumors (83%), with < 1% false-negative test results. Importantly, utDNA analysis showed the potential to reduce the number of cystoscopies by 55%, benefitting 79% of the patients. Eleven of 23 recurrences (48%) were detected earlier with utDNA than with cystoscopy, and distinct patterns of residual utDNA post-surgery indicated minimal residual disease (MRD) or field effect in 6% and 15% of the patients, respectively. In conclusion, utDNA analysis shows high sensitivity to detect tumor recurrence, potential to reduce the number of cystoscopies, and promise to guide patient-specific surveillance regimens.

Multiregional transcriptomics identifies congruent consensus subtypes with prognostic value beyond tumor heterogeneity of colorectal cancer.

Intra-tumor heterogeneity compromises the clinical value of transcriptomic classifications of colorectal cancer. We investigated the prognostic effect of transcriptomic heterogeneity and the potential for classifications less vulnerable to heterogeneity in a single-hospital series of 1093 tumor samples from 692 patients, including multiregional samples from 98 primary tumors and 35 primary-metastasis sets. We show that intra-tumor heterogeneity of the consensus molecular subtypes (CMS) is frequent and has poor-prognostic associations independently of tumor microenvironment markers. Multiregional transcriptomics uncover cancer cell-intrinsic and low-heterogeneity signals that recapitulate the intrinsic CMSs proposed by single-cell sequencing. Further subclassification identifies congruent CMSs that explain a larger proportion of variation in patient survival than intra-tumor heterogeneity. Plasticity is indicated by discordant intrinsic phenotypes of matched primary and metastatic tumors. We conclude that multiregional sampling reconciles the prognostic power of tumor classifications from single-cell and bulk transcriptomics in the context of intra-tumor heterogeneity, and phenotypic plasticity challenges the reconciliation of primary and metastatic subtypes.

Evaluation of commercial kits for isolation and bisulfite conversion of circulating cell-free tumor DNA from blood.

BACKGROUND: DNA methylation biomarkers in circulating cell-free DNA (cfDNA) have great clinical potential for cancer management. Most methods for DNA methylation analysis require bisulfite conversion, causing DNA degradation and loss. This is particularly challenging for cfDNA, which is naturally fragmented and normally present in low amounts. The aim of the present study was to identify an optimal combination of cfDNA isolation and bisulfite conversion kits for downstream analysis of DNA methylation biomarkers in plasma. RESULTS: Of the five tested bisulfite conversion kits (EpiJET Bisulfite Conversion Kit, EpiTect Plus DNA Bisulfite Kit (EpiTect), EZ DNA Methylation-Direct Kit, Imprint DNA Modification Kit (Imprint) and Premium Bisulfite Kit), the highest and lowest DNA yield and recovery were achieved using the EpiTect kit and the Imprint kit, respectively, with more than double the amount of DNA for the EpiTect kit. Of the three tested cfDNA isolation kits (Maxwell RSC ccfDNA Plasma Kit, QIAamp Circulating Nucleic Acid Kit (CNA) and QIAamp MinElute ccfDNA Mini Kit), the CNA kit yielded around twice as much cfDNA compared to the two others kits, although with more high molecular weight DNA present. When comparing various combinations of cfDNA isolation kits and bisulfite conversion kits, the CNA kit and the EpiTect kit were identified as the best-performing combination, resulting in the highest yield of bisulfite converted cfDNA from normal plasma, as measured by droplet digital PCR (ddPCR). As a proof of principle, this kit combination was used to process plasma samples from 13 colorectal cancer patients for subsequent ddPCR methylation analysis of BCAT1 and IKZF1. Methylation of BCAT1 and/or IKZF1 was identified in 6/10 (60%) stage IV patients and 1/3 (33%) stage III patients. CONCLUSIONS: Based on a thorough evaluation of five bisulfite conversion kits and three cfDNA isolation kits, both individually and in combination, the CNA kit and the EpiTect kit were identified as the best-performing kit combination, with highest DNA yield and recovery across a range of DNA input amounts. The combination was successfully used for detection of clinically relevant DNA methylation biomarkers in plasma from cancer patients.

PoDCall: positive droplet calling and normalization of droplet digital PCR DNA methylation data.

MOTIVATION: Droplet digital PCR (ddPCR) holds great promises for investigating DNA methylation with high sensitivity. Yet, the lack of methods for analyzing ddPCR DNA methylation data has resulted in users processing the data manually at the expense of standardization. RESULTS: PoDCall is an R package performing automated calling of positive droplets, quantification and normalization of methylation levels in ddPCR experiments. A Shiny application provides users with an intuitive and interactive interface to access PoDCall functionalities. AVAILABILITY AND IMPLEMENTATION: The PoDCall R package is freely available on Bioconductor at https://bioconductor.org/packages/PoDCall/. The Shiny application can be executed from the R console using the wrapper function PoDCall::podcallShiny(). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BladMetrix: a novel urine DNA methylation test with high accuracy for detection of bladder cancer in hematuria patients.

BACKGROUND: Cystoscopy is the gold standard for bladder cancer detection, but is costly, invasive and has imperfect diagnostic accuracy. We aimed to identify novel and accurate DNA methylation biomarkers for non-invasive detection of bladder cancer in urine, with the potential to reduce the number of cystoscopies among hematuria patients. RESULTS: Biomarker candidates (n = 32) were identified from methylome sequencing of urological cancer cell lines (n = 16) and subjected to targeted methylation analysis in tissue samples (n = 60). The most promising biomarkers (n = 8) were combined into a panel named BladMetrix. The performance of BladMetrix in urine was assessed in a discovery series (n = 112), consisting of bladder cancer patients, patients with other urological cancers and healthy individuals, resulting in 95.7% sensitivity and 94.7% specificity. BladMetrix was furthermore evaluated in an independent prospective and blinded series of urine from patients with gross hematuria (n = 273), achieving 92.1% sensitivity, 93.3% specificity and a negative predictive value of 98.1%, with the potential to reduce the number of cystoscopies by 56.4%. CONCLUSIONS: We here present BladMetrix, a novel DNA methylation urine test for non-invasive detection of bladder cancer, with high accuracy across tumor grades and stages, and the ability to spare a significant number of cystoscopies among patients with gross hematuria.

Targeted genetic and epigenetic profiling of esophageal adenocarcinomas and non-dysplastic Barrett's esophagus.

BACKGROUND: Despite the efforts to describe the molecular landscape of esophageal adenocarcinoma (EAC) and its precursor lesion Barrett's esophagus (BE), discrepant findings are reported. Here, we investigated the prevalence of selected genetic (TP53 mutations and microsatellite instability (MSI) status) and epigenetic (DNA promoter hypermethylation of APC, CDKN2A, MGMT, TIMP3 and MLH1) modifications in a series of 19 non-dysplastic BE and 145 EAC samples. Additional biopsies from adjacent normal tissue were also evaluated. State-of-the-art methodologies and well-defined scoring criteria were applied in all molecular analyses. RESULTS: Overall, we confirmed frequent TP53 mutations among EAC (28%) in contrast to BE, which harbored no mutations. We demonstrated that MSI and MLH1 promoter hypermethylation are rare events, both in EAC and in BE. Our findings further support that APC, CDKN2A, MGMT and TIMP3 promoter hypermethylation is frequently seen in both lesions (21-89%), as well as in a subset of adjacent normal samples (up to 12%). CONCLUSIONS: Our study further enlightens the molecular background of BE and EAC. To the best of our knowledge, this is one of the largest studies addressing a targeted analysis of genetic and epigenetic modifications simultaneously across a combined series of non-dysplastic BE and EAC samples.

Spatial analysis and CD25-expression identify regulatory T cells as predictors of a poor prognosis in colorectal cancer.

Regulatory T cells (Tregs) are a heterogeneous cell population that can either suppress or stimulate immune responses. Tumor-infiltrating Tregs are associated with an adverse outcome from most cancer types, but have generally been found to be associated with a good prognosis in colorectal cancer (CRC). We investigated the prognostic heterogeneity of Tregs in CRC by co-expression patterns and spatial analyses with diverse T cell markers, using multiplex fluorescence immunohistochemistry and digital image analysis in two consecutive series of primary CRCs (total n = 1720). Treg infiltration in tumors, scored as FOXP3+ or CD4+/CD25+/FOXP3+ (triple-positive) cells, was strongly correlated to the overall amount of CD3+ and CD8+ T cells, and consequently associated with a favorable 5-year relapse-free survival rate among patients with stage I-III CRC who underwent complete tumor resection. However, high relative expression of the activation marker CD25 in triple-positive Tregs was independently associated with an adverse outcome in a multivariable model incorporating clinicopathological and known molecular prognostic markers (hazard ratio = 1.35, p = 0.028). Furthermore, spatial marker analysis based on Voronoi diagrams and permutation testing of cellular neighborhoods revealed a statistically significant proximity between Tregs and CD8+-cells in 18% of patients, and this was independently associated with a poor survival (multivariable hazard ratio = 1.36, p = 0.017). These results show prognostic heterogeneity of different Treg populations in primary CRC, and highlight the importance of multi-marker and spatial analyses for accurate immunophenotyping of tumors in relation to patient outcome.

Early and accurate detection of cholangiocarcinoma in patients with primary sclerosing cholangitis by methylation markers in bile.

BACKGROUND AND AIMS: Primary sclerosing cholangitis (PSC) is associated with increased risk of cholangiocarcinoma (CCA). Early and accurate CCA detection represents an unmet clinical need as the majority of patients with PSC are diagnosed at an advanced stage of malignancy. In the present study, we aimed at establishing robust DNA methylation biomarkers in bile for early and accurate diagnosis of CCA in PSC. APPROACH AND RESULTS: Droplet digital PCR (ddPCR) was used to analyze 344 bile samples from 273 patients with sporadic and PSC-associated CCA, PSC, and other nonmalignant liver diseases for promoter methylation of cysteine dioxygenase type 1, cannabinoid receptor interacting protein 1, septin 9, and vimentin. Receiver operating characteristic (ROC) curve analyses revealed high AUCs for all four markers (0.77-0.87) for CCA detection among patients with PSC. Including only samples from patients with PSC diagnosed with CCA ≤ 12 months following bile collection increased the accuracy for cancer detection, with a combined sensitivity of 100% (28/28) and a specificity of 90% (20/203). The specificity increased to 93% when only including patients with PSC with longtime follow-up (> 36 months) as controls, and remained high (83%) when only including patients with PSC and dysplasia as controls (n = 23). Importantly, the bile samples from the CCA-PSC ≤ 12 patients, all positive for the biomarkers, included both early-stage and late-stage CCA, different tumor growth patterns, anatomical locations, and carbohydrate antigen 19-9 levels. CONCLUSIONS: Using highly sensitive ddPCR to analyze robust epigenetic biomarkers, CCA in PSC was accurately detected in bile, irrespective of clinical and molecular features, up to 12 months before CCA diagnosis. The findings suggest a potential for these biomarkers to complement current detection and screening methods for CCA in patients with PSC.

Multiregional assessment of CIMP in primary colorectal cancers: Phenotype concordance but marker variability.

Intratumor heterogeneity of colorectal cancers (CRCs) is manifested both at the genomic and epigenomic levels. Early genetic aberrations in carcinogenesis are clonal and present throughout the tumors, but less is known about the heterogeneity of the epigenetic CpG island methylator phenotype (CIMP). CIMP characterizes a subgroup of CRCs thought to originate from specific precursor lesions, and it is defined by widespread DNA methylation within promoter regions. In this work, we investigated CIMP in two to four multiregional samples from 30 primary tumors (n = 86 samples) using the consensus Weisenberger gene panel (CACNA1G, IGF2, NEUROG1, RUNX3 and SOCS1). Twenty-nine of 30 tumors (97%) showed concordant CIMP status in all samples, and percent methylated reference (PMR) values of all five markers had higher intertumor than intratumor variation (P value = 1.5e-09). However, a third of the CIMP+ tumors exhibited discrepancies in methylation status in at least one of the five gene markers. To conclude, CIMP status was consistent within primary CRCs, and it is likely a clonal phenotype. However, spatial discordances of the individual genes suggest that large-scale analysis of multiregional samples could be of interest for identifying CIMP markers that are robust to intratumor heterogeneity.

Improved prognostication of glioblastoma beyond molecular subtyping by transcriptional profiling of the tumor microenvironment.

Glioblastoma (GBM), the most aggressive form of brain cancer, is characterized by a high level of molecular heterogeneity, and infiltration by various immune and stromal cell populations. Important advances have been made in deciphering the microenvironment of GBMs, but its association with existing molecular subtypes and its potential prognostic role remain elusive. We have investigated the abundance of infiltrating immune and stromal cells in silico, from gene expression profiles. Two cohorts, including in-house normal brain and glioma samples (n = 70) and a large sample set from TCGA (n = 393), were combined into a single exploratory dataset. A third independent cohort (n = 124) was used for validation. Tumors were clustered based on their microenvironment infiltration profiles, and associations with known GBM molecular subtypes and patient outcome were tested a posteriori in a multivariable setting. We identified a subset of GBM samples with significantly higher abundances of most immune and stromal cell populations. This subset showed increased expression of both immune suppressor and immune effector genes compared to other GBMs and was enriched for the mesenchymal molecular subtype. Survival analyses suggested that tumor microenvironment infiltration pattern was an independent prognostic factor for GBM patients. Among all, patients with the mesenchymal subtype with low immune and stromal infiltration had the poorest survival. By combining molecular subtyping with gene expression measures of tumor infiltration, the present work contributes with improving prognostic models in GBM.

Microbe-host interplay in atopic dermatitis and psoriasis.

Despite recent advances in understanding microbial diversity in skin homeostasis, the relevance of microbial dysbiosis in inflammatory disease is poorly understood. Here we perform a comparative analysis of skin microbial communities coupled to global patterns of cutaneous gene expression in patients with atopic dermatitis or psoriasis. The skin microbiota is analysed by 16S amplicon or whole genome sequencing and the skin transcriptome by microarrays, followed by integration of the data layers. We find that atopic dermatitis and psoriasis can be classified by distinct microbes, which differ from healthy volunteers microbiome composition. Atopic dermatitis is dominated by a single microbe (Staphylococcus aureus), and associated with a disease relevant host transcriptomic signature enriched for skin barrier function, tryptophan metabolism and immune activation. In contrast, psoriasis is characterized by co-occurring communities of microbes with weak associations with disease related gene expression. Our work provides a basis for biomarker discovery and targeted therapies in skin dysbiosis.

A robust internal control for high-precision DNA methylation analyses by droplet digital PCR.

Background: Droplet digital PCR (ddPCR) allows absolute quantification of nucleic acids and has potential for improved non-invasive detection of DNA methylation. For increased precision of the methylation analysis, we aimed to develop a robust internal control for use in methylation-specific ddPCR. Methods: Two control design approaches were tested: (a) targeting a genomic region shared across members of a gene family and (b) combining multiple assays targeting different pericentromeric loci on different chromosomes. Through analyses of 34 colorectal cancer cell lines, the performance of the control assay candidates was optimized and evaluated, both individually and in various combinations, using the QX200™ droplet digital PCR platform (Bio-Rad). The best-performing control was tested in combination with assays targeting methylated CDO1, SEPT9, and VIM. Results: A 4Plex panel consisting of EPHA3, KBTBD4, PLEKHF1, and SYT10 was identified as the best-performing control. The use of the 4Plex for normalization reduced the variability in methylation values, corrected for differences in template amount, and diminished the effect of chromosomal aberrations. Positive Droplet Calling (PoDCall), an R-based algorithm for standardized threshold determination, was developed, ensuring consistency of the ddPCR results. Conclusion: Implementation of a robust internal control, i.e., the 4Plex, and an algorithm for automated threshold determination, PoDCall, in methylation-specific ddPCR increase the precision of DNA methylation analysis.

Prognostic relevance of an epigenetic biomarker panel in sentinel lymph nodes from colon cancer patients.

BACKGROUND: Patients with early colorectal cancer (stages I-II) generally have a good prognosis, but a subgroup of 15-20% experiences relapse and eventually die of disease. Occult metastases have been suggested as a marker for increased risk of recurrence in patients with node-negative disease. Using a previously identified, highly accurate epigenetic biomarker panel for early detection of colorectal tumors, we aimed at evaluating the prognostic value of occult metastases in sentinel lymph nodes of colon cancer patients. RESULTS: The biomarker panel was analyzed by quantitative methylation-specific PCR in primary tumors and 783 sentinel lymph nodes from 201 patients. The panel status in sentinel lymph nodes showed a strong association with lymph node stage (P = 8.2E-17). Compared with routine lymph node diagnostics, the biomarker panel had a sensitivity of 79% (31/39). Interestingly, among 162 patients with negative lymph nodes from routine diagnostics, 13 (8%) were positive for the biomarker panel. Colon cancer patients with high sentinel lymph node methylation had an inferior prognosis (5-year overall survival P = 3.0E-4; time to recurrence P = 3.1E-4), although not significant. The same trend was observed in multivariate analyses (P = 1.4E-1 and P = 6.7E-2, respectively). Occult sentinel lymph node metastases were not detected in early stage (I-II) colon cancer patients who experienced relapse. CONCLUSIONS: Colon cancer patients with high sentinel lymph node methylation of the analyzed epigenetic biomarker panel had an inferior prognosis, although not significant in multivariate analyses. Occult metastases in TNM stage II patients that experienced relapse were not detected.

CpG island methylator phenotype identifies high risk patients among microsatellite stable BRAF mutated colorectal cancers.

The prognostic value of CpG island methylator phenotype (CIMP) in colorectal cancer remains unsettled. We aimed to assess the prognostic value of this phenotype analyzing a total of 1126 tumor samples obtained from two Norwegian consecutive colorectal cancer series. CIMP status was determined by analyzing the 5-markers CAGNA1G, IGF2, NEUROG1, RUNX3 and SOCS1 by quantitative methylation specific PCR (qMSP). The effect of CIMP on time to recurrence (TTR) and overall survival (OS) were determined by uni- and multivariate analyses. Subgroup analyses were conducted according to MSI and BRAF mutation status, disease stage, and also age at time of diagnosis (<60, 60-74, ≥75 years). Patients with CIMP positive tumors demonstrated significantly shorter TTR and worse OS compared to those with CIMP negative tumors (multivariate hazard ratio [95% CI] 1.86 [1.31-2.63] and 1.89 [1.34-2.65], respectively). In stratified analyses, CIMP tumors showed significantly worse outcome among patients with microsatellite stable (MSS, P < 0.001), and MSS BRAF mutated tumors (P < 0.001), a finding that persisted in patients with stage II, III or IV disease, and that remained significant in multivariate analysis (P < 0.01). Consistent results were found for all three age groups. To conclude, CIMP is significantly associated with inferior outcome for colorectal cancer patients, and can stratify the poor prognostic patients with MSS BRAF mutated tumors.

Experimental factors affecting the robustness of DNA methylation analysis.

Diverging methylation frequencies are often reported for the same locus in the same disease, underscoring the need for limiting technical variability in DNA methylation analyses. We have investigated seven likely sources of variability at different steps of bisulfite PCR-based DNA methylation analyses using a fully automated quantitative methylation-specific PCR setup of six gene promoters across 20 colon cancer cell lines. Based on >15,000 individual PCRs, all tested parameters affected the normalized percent of methylated reference (PMR) differences, with a fourfold varying magnitude. Additionally, large variations were observed across the six genes analyzed. The highest variation was seen using single-copy genes as reference for normalization, followed by different amounts of template in the PCR, different amounts of DNA in the bisulfite reaction, and storage of bisulfite converted samples. Finally, when a highly standardized pipeline was repeated, the difference in PMR value for the same assay in the same cell line was on average limited to five (on a 0-100 scale). In conclusion, a standardized pipeline is essential for consistent methylation results, where parameters are kept constant for all samples. Nevertheless, a certain level of variation in methylation values must be expected, underscoring the need for careful interpretation of data.

Breast Cancer Cell-Derived GM-CSF Licenses Regulatory Th2 Induction by Plasmacytoid Predendritic Cells in Aggressive Disease Subtypes.

Reciprocal interactions between tumor cells and their microenvironment vitally impact tumor progression. In this study, we show that GM-CSF produced by primary breast tumor cells induced the activation of plasmacytoid predendritic cells (pDC), a cell type critical to anti-viral immunity. pDC that expressed the GM-CSF receptor were increased in breast tumors compared with noninvolved adjacent breast tissue. Tumor-activated pDC acquired naïve CD4(+) T-cell stimulatory capacity and promoted a regulatory Th2 response. Finally, the concomitant increase of GM-CSF and pDC was significantly associated with relatively more aggressive breast cancer subtypes. Our results characterize the first tumor-derived factor that can activate pDC to promote a regulatory Th2 response, with implications for therapeutic targeting of a tumor-immune axis of growing recognition in its significance to cancer.

Human blood BDCA-1 dendritic cells differentiate into Langerhans-like cells with thymic stromal lymphopoietin and TGF-ß.

The ontogeny of human Langerhans cells (LCs) remains poorly characterized, in particular the nature of LC precursors and the factors that may drive LC differentiation. Here we report that thymic stromal lymphopoietin (TSLP), a keratinocyte-derived cytokine involved in epithelial inflammation, cooperates with transforming growth factor (TGF)-ß for the generation of LCs. We show that primary human blood BDCA-1(+), but not BDCA-3(+), dendritic cells (DCs) stimulated with TSLP and TGF-ß harbor a typical CD1a(+)Langerin(+) LC phenotype. Electron microscopy established the presence of Birbeck granules, an intracellular organelle specific to LCs. LC differentiation was not observed from tonsil BDCA-1(+) and BDCA-3(+) subsets. TSLP + TGF-ß LCs had a mature phenotype with high surface levels of CD80, CD86, and CD40. They induced a potent CD4(+) T-helper (Th) cell expansion and differentiation into Th2 cells with increased production of tumor necrosis factor-α and interleukin-6 compared with CD34-derived LCs. Our findings establish a novel LC differentiation pathway from BDCA-1(+) blood DCs with potential implications in epithelial inflammation. Therapeutic targeting of TSLP may interfere with tissue LC repopulation from circulating precursors.

A comprehensive evaluation of normalization methods for Illumina high-throughput RNA sequencing data analysis.

During the last 3 years, a number of approaches for the normalization of RNA sequencing data have emerged in the literature, differing both in the type of bias adjustment and in the statistical strategy adopted. However, as data continue to accumulate, there has been no clear consensus on the appropriate normalization method to be used or the impact of a chosen method on the downstream analysis. In this work, we focus on a comprehensive comparison of seven recently proposed normalization methods for the differential analysis of RNA-seq data, with an emphasis on the use of varied real and simulated datasets involving different species and experimental designs to represent data characteristics commonly observed in practice. Based on this comparison study, we propose practical recommendations on the appropriate normalization method to be used and its impact on the differential analysis of RNA-seq data.

Multiple testing in large-scale genetic studies.

Recent advances in Molecular Biology and improvements in microarray and sequencing technologies have led biologists toward high-throughput genomic studies. These studies aim at finding associations between genetic markers and a phenotype and involve conducting many statistical tests on these markers. Such Please confirm the changes in the sentence "Such a wide..." a wide investigation of the genome not only renders genomic studies quite attractive but also lead to a major shortcoming. That is, among the markers detected as associated with the phenotype, a nonnegligible proportion is not in reality (false-positives) and also true associations can be missed (false-negatives). A main cause of these spurious associations is due to the multiple-testing problem, inherent to conducting numerous statistical tests. Several approaches exist to work around this issue. These multiple-testing adjustments aim at defining new statistical confidence measures that are controlled to guarantee that the outcomes of the tests are pertinent.The most natural correction was introduced by Bonferroni and aims at controlling the family-wise error-rate (FWER) that is the probability of having at least one false-positive. Another approach is based on the false-discovery-rate (FDR) and considers the proportion of significant results that are expected to be false-positives. Finally, the local-FDR focuses on the actual probability for a marker of being associated or not with the phenotype. These strategies are widely used but one has to be careful about when and how to apply them. We propose in this chapter a discussion on the multiple-testing issue and on the main approaches to take it into account. We aim at providing a theoretical and intuitive definition of these concepts along with practical advises to guide researchers in choosing the more appropriate multiple-testing procedure corresponding to the purposes of their studies.

Should we abandon the t-test in the analysis of gene expression microarray data: a comparison of variance modeling strategies.

High-throughput post-genomic studies are now routinely and promisingly investigated in biological and biomedical research. The main statistical approach to select genes differentially expressed between two groups is to apply a t-test, which is subject of criticism in the literature. Numerous alternatives have been developed based on different and innovative variance modeling strategies. However, a critical issue is that selecting a different test usually leads to a different gene list. In this context and given the current tendency to apply the t-test, identifying the most efficient approach in practice remains crucial. To provide elements to answer, we conduct a comparison of eight tests representative of variance modeling strategies in gene expression data: Welch's t-test, ANOVA [1], Wilcoxon's test, SAM [2], RVM [3], limma [4], VarMixt [5] and SMVar [6]. Our comparison process relies on four steps (gene list analysis, simulations, spike-in data and re-sampling) to formulate comprehensive and robust conclusions about test performance, in terms of statistical power, false-positive rate, execution time and ease of use. Our results raise concerns about the ability of some methods to control the expected number of false positives at a desirable level. Besides, two tests (limma and VarMixt) show significant improvement compared to the t-test, in particular to deal with small sample sizes. In addition limma presents several practical advantages, so we advocate its application to analyze gene expression data.