Nucleated red blood cells impact DNA methylation and expression analyses of cord blood hematopoietic cells
© de Goede et al. 2015
Received: 30 June 2015
Accepted: 31 August 2015
Published: 11 September 2015
Genome-wide DNA methylation (DNAm) studies have proven extremely useful to understand human hematopoiesis. Due to their active DNA content, nucleated red blood cells (nRBCs) contribute to epigenetic and transcriptomic studies derived from whole cord blood. Genomic studies of cord blood hematopoietic cells isolated by fluorescence-activated cell sorting (FACS) may be significantly altered by heterotopic interactions with nRBCs during conventional cell sorting.
We report that cord blood T cells, and to a lesser extent monocytes and B cells, physically engage with nRBCs during FACS. These heterotopic interactions resulted in significant cross-contamination of genome-wide epigenetic and transcriptomic data. Formal exclusion of erythroid lineage-specific markers yielded DNAm profiles (measured by the Illumina 450K array) of cord blood CD4 and CD8 T lymphocytes, B lymphocytes, natural killer (NK) cells, granulocytes, monocytes, and nRBCs that were more consistent with expected hematopoietic lineage relationships. Additionally, we identified eight highly differentially methylated CpG sites in nRBCs (false detection rate <5 %, |Δβ| >0.50) that can be used to detect nRBC contamination of purified hematopoietic cells or to assess the impact of nRBCs on whole cord blood DNAm profiles. Several of these erythroid markers are located in or near genes involved in erythropoiesis (ZFPM1, HDAC4) or immune function (MAP3K14, IFIT1B), reinforcing a possible immune regulatory role for nRBCs in early life.
Heterotopic interactions between erythroid cells and white blood cells can result in contaminated cell populations if not properly excluded during cell sorting. Cord blood nRBCs have a distinct DNAm profile that can significantly skew epigenetic studies. Our findings have major implications for the design and interpretation of genome-wide epigenetic and transcriptomic studies using human cord blood.
With the increased accessibility of high throughput technologies for epigenetic and gene expression studies, genome-wide approaches have gained popularity in studies of hematopoietic cell lineage relationships [1–4]. However, although genome-wide profiling of isolated blood cells can provide a large amount of information, data interpretation is notoriously difficult in mixed cell populations [5–7]. To address this issue, studies can be performed either on homogeneous cell populations or on mixed cell samples with deconvolution algorithms applied to correct for differences in cell composition [8, 9]. One concern with the former approach in blood is that red blood cells (RBCs) have been shown to engage in functional heterotopic interactions with other hematopoietic cells [10–16]. If not formally excluded using lineage markers, these interactions could impact whole-genome studies of hematopoietic cells sorted by fluorescence-activated cell sorting (FACS), particularly in cord blood which has a notable proportion of nucleated RBCs (nRBCs) .
The proportion of nRBCs in cord blood varies considerably between individuals. Typically, these cells represent only a few percent of the total nucleated cell count; however, they can comprise up to 50 % of all nucleated cells in some chronic hypoxic-ischemic-related pregnancy situations [17–19]. For example, higher nRBC counts have been observed in response to prenatal exposure to infection, preeclampsia, maternal obesity, diabetes, and smoking [17–22]. nRBCs are generally resistant to lysis protocols and tend to sediment in the mononuclear cell fraction during purification by density gradient centrifugation, further complicating the isolation of pure hematopoietic cell populations . Depending on their proportion, the presence of nRBCs could complicate both epigenetic and gene expression studies.
Under non-pathological conditions, DNA methylation (DNAm) shows great biological differences with tissue and cell type. Clustering of adult blood cells based on their DNAm profiles is consistent with the classical model of hematopoietic lineage relationships [6, 9, 24]. However, our initial analysis of genome-wide DNAm in cord blood cell populations isolated by FACS suggested significant cross-contamination between cell types. We observed low-incidence white blood cell (WBC) heterotopic interactions with nRBCs that were undetected by traditional singlet FACS gating, due to the small size of nRBCs. Thus, to obtain pure WBC populations, we developed and implemented a stringent sorting protocol that excludes erythroid-specific surface markers. The DNAm profiles of cell populations obtained by our stringent FACS method were used (1) to evaluate the impact of nRBC contamination on the DNAm profiles of T lymphocytes and monocytes and (2) to identify nRBC-distinct DNAm markers to detect erythroid contamination in genome-wide DNAm studies.
Results and discussion
Heterotopic cell interactions impact genome-wide signatures of hematopoietic cells
Revised DNAm profiles of hematopoietic cells obtained by a more stringent cell sorting strategy
Number of cell-specific DM CpG sites (FDR <5 %) following the standard and stringent FACS strategies
nRBCs, standard FACS
nRBCs, stringent FACS
Monocytes, standard FACS
Monocytes, stringent FACS
T cells, standard FACS
T cells, stringent FACS
Top DM sites for each cell type (FDR <5 %, |Δβ| > 0.20) were then compared between the two sorting protocols. For T cells, the majority of DM sites (>98 %) discovered by the standard method overlapped with the DM sites identified by the stringent protocol (Fig. 3c). A notable percentage (47 %) of monocyte DM sites found by the standard protocol were also discovered by the stringent protocol (Fig. 3d). For nRBCs, the DM sites identified by the two protocols showed the least overlap (36 %), with the stringent protocol identifying far more nRBC DM sites than the standard protocol (8982 versus 2338) (Fig. 3e). Of the 8982 stringent nRBC DM sites, six were located in hemoglobin genes we found to be highly expressed in cord blood WBCs sorted by a standard protocol (and thus presumed to be contaminated with RBCs) (Fig. 2; Additional file 1: Table S2). These genes were also found to be highly expressed in publicly available datasets of cord blood WBCs, again indicating widespread erythroid contamination (Additional file 1: Figure S2). The DNAm differences at these loci were striking, with the mean nRBC DNAm up to 43 percentage points less than the mean DNAm for all WBCs. Several of these CpG sites are located in either the body of the associated hemoglobin gene or within 300 bases upstream of its transcriptional start site and may be associated with erythroid-specific gene expression.
Erythroid-specific DM sites
Top eight CpG sites with nRBC-distinct DNAm from white blood cells in cord blood
450K array CpG identifier
CpG location: chromosome, closest gene
Location in gene
Mean nRBC β (min., max.)
Mean non-erythroid cell β (min., max.)
16, SNORA64 & RPS2
Some of these erythroid DNAm markers are associated with genes that have known erythropoietic function, such as ZFPM1 and HDAC4 (Table 2). The zinc finger protein ZFPM1 acts as a cofactor for GATA-1, a key transcription factor in erythroid differentiation [31, 32]. Histone deacetylase 4 (HDAC4) directly associates with GATA-1 and its expression is specifically reduced during erythroid maturation, likely being localized to the nucleus . HDAC4 may be involved in the enucleation process of nRBCs: histone deacetylation by HDACs is essential for heterochromatin formation, and condensed chromatin is a main requirement for enucleation and terminal erythroid differentiation . Interestingly, other erythroid DNAm markers are near genes involved in immune functions, such as MAP3K14 and IFIT1B, consistent with the idea that nRBCs have an immunoregulatory role in early life . MAP3K14 induces NF-kappa-B signaling, a major inflammatory response pathway . IFIT1 is typically silent in most cells, but becomes highly expressed in response to interferons, viral infection, and certain molecular patterns, with IFIT proteins having antiviral effects through binding and modulation of host and viral proteins and RNA . As these erythroid DNAm marker sites are located largely in enhancer regions, reduced DNAm in nRBCs may reflect either specific upregulation of these genes in erythroid cells or a more primitive permissive state that is actively shut off in differentiation of other cell types.
Although these erythroid DNAm markers are the top nRBC DM sites, they display notable inter-individual variability in nRBC DNAm, with β value standard deviations ranging from 0.048 to 0.091. We hypothesize that this variability in DNAm may be related to important inter-individual differences in nRBC function or maturation, based on the negative Pearson correlation we observed between array-wide median nRBC DNAm and nRBC proportion (r = −0.86, p = 0.013) (Additional file 1: Figure S5A). Linear modeling identified 5935 CpG sites significantly associated (FDR <5 %) with nRBC proportion, including three of the eight CpG sites identified as erythroid DNAm markers (Additional file 1: Figure S5B–C). These results suggest that DNAm changes in cord blood nRBCs occur dynamically as a function of nRBC production and maturation, thereby revealing an additional level of functional complexity to consider in whole-genome DNAm analyses of nRBCs.
While nRBCs are generally absent or rare in adult blood, they are commonly present in low proportion in cord blood, with a higher nRBC count associated with a variety of maternal and fetal health factors [17, 19–22]. Our data show that nRBCs have a distinct DNAm profile, with an association between nRBC DNAm and overall nRBC proportion in cord blood (Table 1; Figs. 3a–b and 4; Additional file 1: Figure S5). The complex DNAm profile of nRBCs has implications for epigenetic studies of whole cord blood and mononuclear cells, in which nRBCs have a demonstrable effect on DNAm (M.J.J. et al., manuscript in preparation). Despite the variability in nRBC DNAm at our identified erythroid DNAm markers, we believe that these sites will be sufficient to detect erythroid cells due to the low variation within all WBCs at these sites, as well as the large magnitude of DNAm difference between nRBCs and other cell types.
Heterotopic interactions between erythroid cells and WBCs are likely biologically meaningful events, since RBCs have immune functions that require cell-to-cell contact with WBCs. These include modulation of T lymphocyte and neutrophil survival [10, 11, 14] and immunosuppression in T and B lymphocytes and dendritic cells [12, 13, 16]. Additionally, RBCs adhere to macrophages to form erythroblastic islands during both fetal and adult RBC maturation . Our data highlight the importance of formally excluding these interactions in lineage studies of cord blood hematopoietic cells using flow cytometry. This has major ramifications for the design of epigenetic, transcriptomic, and functional studies of cord blood cells.
Availability of supporting data
The data set supporting the results of this article is available in the NCBI Gene Expression Omnibus repository, GSE68456, [http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE68456].
Ethics, consent and permissions
Ethics approval for this study was obtained from the University of British Columbia Children’s & Women’s Research Ethics Board (certificate numbers H07-02681 and H04-70488). Written, informed parental consent to participate was obtained. Individual patient data is not reported.
Cord blood was collected from neonates delivered by elective cesarean section in the absence of labor at the Children’s & Women’s Health Centre of BC (Vancouver, Canada).
Hematopoietic cell purification and sorting is described in the Additional file 1: Supplemental Methods. Two sorting protocols were compared, which are referred to as the “standard” and “stringent” protocols. The stringent method includes additional negative gating steps, mainly for erythroid lineage-specific cell surface protein markers (Additional file 1: Figure S1; Additional file 1: Table S1). For DNAm studies, cells were sorted from a total of 12 subjects: whole (CD3+) T cells, monocytes, and nRBCs were collected from five individuals by the standard sorting method; B cells, CD4 T cells, CD8 T cells, granulocytes, monocytes, NK cells, and nRBCs were collected from seven individuals by the stringent sorting method. For genome-wide gene expression analysis, naïve CD4 T cells were sorted from 12 additional subjects. Transcriptomic profiling is described in the Additional file 1: Supplemental Methods.
DNAm measurement and data normalization
DNA was extracted from isolated cell populations using standard protocols and purified with the DNeasy Blood and Tissue kit (QIAGEN, MD, USA). DNA was bisulfite-converted using the EZ DNA Methylation Kit (Zymo Research, CA, USA) before amplification and hybridization to the Illumina Infinium HumanMethylation450 BeadChip (450K array) following manufacturer’s protocols (Illumina Inc., CA, USA). With a HiScan reader (Illumina), 450 K chips were scanned.
Raw intensity data (GSE68456) were background normalized in GenomeStudio (Illumina). Quality control was performed using the 835 control probes included in the array. The intensity data were then exported from GenomeStudio and converted into M values using the lumi package  in R software . Sample identity and quality were evaluated as described in the Additional file 1: Supplemental Methods, and one NK cell sample was removed as an outlier. The 450K array targets 485,577 DNAm sites, but probe filtering was performed as described in the Additional file 1: Supplemental Methods to produce a final dataset of 440,315 sites . Red-green color bias was corrected for using the lumi package, and the data were normalized with subset within-array quantile normalization [38, 41].
Analysis of hematopoietic cell lineage DNAm relationships
Since the stringent FACS strategy was designed based on results from the standard FACS strategy, sample collection and 450K array runs for cells collected by these two protocols were done separately. To avoid confounding by batch effects, DNAm analyses were also performed separately for the data from each FACS protocol. Our analytic approach was to compare cell types sorted by the same FACS protocol to each other and then to evaluate whether a given cell type’s epigenetic relationship with the other cell types changed between FACS methods. To eliminate DNAm differences that can arise due to genetic effects, comparisons were made between cell types derived from the same set of individuals.
For the standard sorting method, DNAm data were available for nRBCs, monocytes, and T cells from five individuals at 440,315 sites after pre-processing. Unsupervised Euclidean clustering of the samples based on DNAm β values was performed as an initial global analysis step. Differential DNAm between each blood cell pairing was tested by linear modeling through the R package limma . Surrogate variable (SV) analysis using the R package sva  was performed to account for unwanted variability in the linear modeling. SVs were used as covariates in the model, with cell type as the main effect. Resulting p values were adjusted for multiple comparisons by the Benjamini and Hochberg  FDR method, and we limited statistically significant sites to those that passed an FDR <5 %. SV-corrected data was used for DNAm-based filtering of the statistically significant sites. At each site, a between-group difference in DNA methylation (Δβ) was calculated by subtracting mean DNAm for one cell type from the other. Differentially methylated (DM) sites were considered as those having both an FDR <5 % and |Δβ| >0.20.
For the stringent sorting method, DNAm data were available for B cells, CD4 T cells, CD8 T cells, granulocytes, monocytes, NK cells, and nRBCs from seven individuals at 440,315 CpG sites after pre-processing. To analyze the data in a comparable way to the standard FACS protocol, only CD4 T cells, monocytes, and nRBCs were considered. The DNAm profiles of these cell populations were analyzed as described for the standard sorting protocol.
To identify DNAm markers specific to nRBCs, data from the stringent sorting method for all seven cell types were used. DM sites between nRBCs and every other cell type were detected by linear modeling with nRBCs as the reference cell type and SVs included as covariates. Significantly, DM sites were defined as those with a FDR <5 % and a |Δβ| >0.50. Finally, to evaluate the relationship between nRBC proportion in whole cord blood and DNAm of nRBCs, the SV-corrected M values for the seven nRBC samples collected by stringent FACS methods were used. Linear modeling was performed with nRBC proportion (as measured by number of nRBCs/100 WBCs in whole blood) as the main effect and no covariates.
- 450K array:
Illumina Infinium HumanMethylation450 BeadChip
fluorescence-activated cell sorting
false detection rate
- NK cells:
natural killer cells
nucleated red blood cell
red blood cell
white blood cell
We thank the BC Children’s & Women’s Hospital staff for their help with subject recruitment; Ruby Jiang, Mihoko Ladd, Paul Villeneuve, Drs. Julie MacIsaac and Maria Peñaherrera for their work in sample processing, and Dr. Lisa Xu for flow cytometer operation. This research was funded by grants from the Canadian Institutes of Health Research (CIHR; MOP-123478 to PML and MOP-49520 to WPR). OMdG is supported by a CIHR Frederick Banting and Charles Best Graduate Scholarship—Master’s Award. HRR is supported by a fellowship from the Mitacs national research organization. EMP is supported by a CIHR Frederick Banting and Charles Best Canada Graduate Scholarship—Doctoral Award. MJJ is supported by a Mining for Miracles fellowship from the Child and Family Research Institute (CFRI). MSK is the Canada Research Chair in Social Epigenetics. PML is supported by Clinician-Scientist Awards from the CFRI and the Michael Smith Foundation for Health Research (MSFHR). WPR is supported by an investigator award from the CFRI.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Chen L, Kostadima M, Martens JH, et al. Transcriptional diversity during lineage commitment of human blood progenitors. Science. 2014;345(6204):1251033.PubMed CentralView ArticlePubMedGoogle Scholar
- Ji H, Ehrlich LI, Seita J, et al. Comprehensive methylome map of lineage commitment from haematopoietic progenitors. Nature. 2010;467(7313):338–42.PubMed CentralView ArticlePubMedGoogle Scholar
- Laslo P, Pongubala JM, Lancki DW, Singh H. Gene regulatory networks directing myeloid and lymphoid cell fates within the immune system. Semin Immunol. 2008;20(4):228–35.View ArticlePubMedGoogle Scholar
- Simon LM, Edelstein LC, Nagalla S, et al. Human platelet microRNA-mRNA networks associated with age and gender revealed by integrated plateletomics. Blood. 2014;123(16):e37–45.PubMed CentralView ArticlePubMedGoogle Scholar
- Lam LL, Emberly E, Fraser HB, et al. Factors underlying variable DNA methylation in a human community cohort. Proc Natl Acad Sci U S A. 2012;109 Suppl 2:17253–60.PubMed CentralView ArticlePubMedGoogle Scholar
- Reinius LE, Acevedo N, Joerink M, et al. Differential DNA methylation in purified human blood cells: Implications for cell lineage and studies on disease susceptibility. PLoS One. 2012;7(7):e41361.PubMed CentralView ArticlePubMedGoogle Scholar
- Whitney AR, Diehn M, Popper SJ, et al. Individuality and variation in gene expression patterns in human blood. Proc Natl Acad Sci U S A. 2003;100(4):1896–901.PubMed CentralView ArticlePubMedGoogle Scholar
- Houseman EA, Accomando WP, Koestler DC, et al. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics. 2012;13:86. -2105-13-86.PubMed CentralView ArticlePubMedGoogle Scholar
- Houseman EA, Molitor J, Marsit CJ. Reference-free cell mixture adjustments in analysis of DNA methylation data. Bioinformatics. 2014;30(10):1431–9.PubMed CentralView ArticlePubMedGoogle Scholar
- Aoshiba K, Nakajima Y, Yasui S, Tamaoki J, Nagai A. Red blood cells inhibit apoptosis of human neutrophils. Blood. 1999;93(11):4006–10.PubMedGoogle Scholar
- Arosa FA, Pereira CF, Fonseca AM. Red blood cells as modulators of T cell growth and survival. Curr Pharm Des. 2004;10(2):191–201.View ArticlePubMedGoogle Scholar
- Elahi S, Ertelt JM, Kinder JM, et al. Immunosuppressive CD71+ erythroid cells compromise neonatal host defence against infection. Nature. 2013;504(7478):158–62.PubMed CentralView ArticlePubMedGoogle Scholar
- Elahi S. New insight into an old concept: role of immature erythroid cells in immune pathogenesis of neonatal infection. Front Immunol. 2014;5:376.PubMed CentralView ArticlePubMedGoogle Scholar
- Fonseca AM, Porto G, Uchida K, Arosa FA. Red blood cells inhibit activation-induced cell death and oxidative stress in human peripheral blood T lymphocytes. Blood. 2001;97(10):3152–60.View ArticlePubMedGoogle Scholar
- Hunt JS, Beck ML, Hardman JT, Tegtmeier GE, Bayer WL. Characterization of human erythrocyte alloantibodies by IgG subclass and monocyte interaction. Am J Clin Pathol. 1980;74(3):259–64.PubMedGoogle Scholar
- Schakel K, von Kietzell M, Hansel A, et al. Human 6-sulfo LacNAc-expressing dendritic cells are principal producers of early interleukin-12 and are controlled by erythrocytes. Immunity. 2006;24(6):767–77.View ArticlePubMedGoogle Scholar
- Hermansen MC. Nucleated red blood cells in the fetus and newborn. Arch Dis Child Fetal Neonatal Ed. 2001;84(3):F211–5.PubMed CentralView ArticlePubMedGoogle Scholar
- Baschat AA, Gungor S, Kush ML, Berg C, Gembruch U, Harman CR. Nucleated red blood cell counts in the first week of life: a critical appraisal of relationships with perinatal outcome in preterm growth-restricted neonates. Am J Obstet Gynecol. 2007;197(3):286. e1-286.e8.View ArticlePubMedGoogle Scholar
- Aali BS, Malekpour R, Sedig F, Safa A. Comparison of maternal and cord blood nucleated red blood cell count between pre-eclamptic and healthy women. J Obstet Gynaecol Res. 2007;33(3):274–8.View ArticlePubMedGoogle Scholar
- Redline RW. Elevated circulating fetal nucleated red blood cells and placental pathology in term infants who develop cerebral palsy. Hum Pathol. 2008;39(9):1378–84.View ArticlePubMedGoogle Scholar
- Yeruchimovich M, Dollberg S, Green DW, Mimouni FB. Nucleated red blood cells in infants of smoking mothers. Obstet Gynecol. 1999;93(3):403–6.View ArticlePubMedGoogle Scholar
- Yeruchimovich M, Mimouni FB, Green DW, Dollberg S. Nucleated red blood cells in healthy infants of women with gestational diabetes. Obstet Gynecol. 2000;95(1):84–6.View ArticlePubMedGoogle Scholar
- Fuss IJ, Kanof ME, Smith PD, Zola H. Isolation of whole mononuclear cells from peripheral blood and cord blood. Curr Protoc Immunol. 2009; 85:711-18.
- Ziller MJ, Gu H, Muller F, et al. Charting a dynamic DNA methylation landscape of the human genome. Nature. 2013;500(7463):477–81.View ArticlePubMedGoogle Scholar
- Accomando WP, Wiencke JK, Houseman EA, Nelson HH, Kelsey KT. Quantitative reconstruction of leukocyte subsets using DNA methylation. Genome Biol. 2014;15(3):R50-2014-15-3-r50.
- Broske AM, Vockentanz L, Kharazi S, et al. DNA methylation protects hematopoietic stem cell multipotency from myeloerythroid restriction. Nat Genet. 2009;41(11):1207-1215.
- Cedar H, Bergman Y. Epigenetics of haematopoietic cell development. Nat Rev Immunol. 2011;11(7):478-488.
- Katsura Y. Redefinition of lymphoid progenitors. Nat Rev Immunol. 2002;2(2):127-132.
- Mansson R, Hultquist A, Luc S, et al. Molecular evidence for hierarchical transcriptional lineage priming in fetal and adult stem cells and multipotent progenitors. Immunity. 2007;26(4):407-419.
- Yuen RK, Avila L, Penaherrera MS, et al. Human placental-specific epipolymorphism and its association with adverse pregnancy outcomes. PLoS One. 2009;4(10):e7389.
- Rodriguez P, Bonte E, Krijgsveld J, et al. GATA-1 forms distinct activating and repressive complexes in erythroid cells. EMBO J. 2005;24(13):2354-2366.
- Tsang AP, Visvader JE, Turner CA, et al. FOG, a multitype zinc finger protein, acts as a cofactor for transcription factor GATA-1 in erythroid and megakaryocytic differentiation. Cell. 1997;90(1):109-119.
- Varricchio L, Dell'Aversana C, Nebbioso A, et al. Identification of NuRSERY, a new functional HDAC complex composed by HDAC5, GATA1, EKLF and pERK present in human erythroid cells. Int J Biochem Cell Biol. 2014;50:112-122.
- Ji P, Yeh V, Ramirez T, Murata-Hori M, Lodish HF. Histone deacetylase 2 is required for chromatin condensation and subsequent enucleation of cultured mouse fetal erythroblasts. Haematologica. 2010;95(12):2013-2021.
- Lawrence T. The nuclear factor NF-kappaB pathway in inflammation. Cold Spring Harb Perspect Biol. 2009;1(6):a001651.
- Fensterl V, Sen GC. Interferon-induced Ifit proteins: Their role in viral pathogenesis. J Virol. 2015;89(5):2462-2468.
- de Back DZ, Kostova EB, van Kraaij M, van den Berg TK, van Bruggen R. Of macrophages and red blood cells; a complex love story. Front Physiol. 2014;5:9.
- Du P, Kibbe WA, Lin SM. lumi: A pipeline for processing Illumina microarray. Bioinformatics. 2008;24(13):1547-1548.
- R Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2014
- Price ME, Cotton AM, Lam LL, et al. Additional annotation enhances potential for biologically-relevant analysis of the Illumina Infinium HumanMethylation450 BeadChip array. Epigenetics Chromatin. 2013;6(1):4-8935-6-4.
- Maksimovic J, Gordon L, Oshlack A. SWAN: subset-quantile within array normalization for Illumina Infinium HumanMethylation450 BeadChips. Genome Biol. 2012;13(6):R44-2012-13-6-r44.
- Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43(7):e47.
- Leek JT, Storey JD. Capturing heterogeneity in gene expression studies by surrogate variable analysis. PLoS Genet. 2007;3(9):1724-1735.
- Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Statist Soc B. 1995;57(1):289.