The application of epiphenotyping approaches to DNA methylation array studies of the human placenta

BACKGROUND

Genome-wide DNA methylation (DNAme) profiling of the placenta with Illumina Infinium Methylation bead arrays is often used to explore the connections between in utero exposures, placental pathology, and fetal development. However, many technical and biological factors can lead to signals of DNAme variation between samples and between cohorts, and understanding and accounting for these factors is essential to ensure meaningful and replicable data analysis. Recently, "epiphenotyping" approaches have been developed whereby DNAme data can be used to impute information about phenotypic variables such as gestational age, sex, cell composition, and ancestry. These epiphenotypes offer avenues to compare phenotypic data across cohorts, and to understand how phenotypic variables relate to DNAme variability. However, the relationships between placental epiphenotyping variables and other technical and biological variables, and their application to downstream epigenome analyses, have not been well studied.

RESULTS

Using DNAme data from 204 placentas across three cohorts, we applied the PlaNET R package to estimate epiphenotypes gestational age, ancestry, and cell composition in these samples. PlaNET ancestry estimates were highly correlated with independent polymorphic ancestry-informative markers, and epigenetic gestational age, on average, was estimated within 4 days of reported gestational age, underscoring the accuracy of these tools. Cell composition estimates varied both within and between cohorts, as well as over very long placental processing times. Interestingly, the ratio of cytotrophoblast to syncytiotrophoblast proportion decreased with increasing gestational age, and differed slightly by both maternal ethnicity (lower in white vs. non-white) and genetic ancestry (lower in higher probability European ancestry). The cohort of origin and cytotrophoblast proportion were the largest drivers of DNAme variation in this dataset, based on their associations with the first principal component.

CONCLUSIONS

This work confirms that cohort, array (technical) batch, cell type proportion, self-reported ethnicity, genetic ancestry, and biological sex are important variables to consider in any analyses of Illumina DNAme data. We further demonstrate the specific utility of epiphenotyping tools developed for use with placental DNAme data, and show that these variables (i) provide an independent check of clinically obtained data and (ii) provide a robust approach to compare variables across different datasets. Finally, we present a general framework for the processing and analysis of placental DNAme data, integrating the epiphenotype variables discussed here.

The application of epiphenotyping approaches to DNA methylation array studies of the human placenta

Background : Genome-wide DNA methylation (DNAme) profiling of the placenta with Illumina Infinium Methylation bead arrays is often used to explore the connections between in utero exposures, placental pathology, and fetal development. However, many technical and biological factors can lead to signals of DNAme variation between samples and between cohorts, and understanding and accounting for these factors is essential to ensure meaningful and replicable data analysis. Recently, "epiphenotyping" approaches have been developed whereby DNAme data can be used to impute information about phenotypic variables such as gestational age, sex, cell composition, and ancestry. These epiphenotypes offer avenues to compare phenotypic data across cohorts, and to understand how phenotypic variables relate to DNAme variability. However, the relationships between placental epiphenotyping variables and other technical and biological variables, and their application to downstream epigenome analyses, have not been well studied. Results : Using DNAme data from 204 placentas across three cohorts, we applied the PlaNET R package to estimate epiphenotypes gestational age, ancestry, and cell composition in these samples. PlaNET ancestry estimates were highly correlated with independent polymorphic ancestry informative markers, and epigenetic gestational age, on average, was estimated within 4 days of reported gestational age, underscoring the accuracy of these tools. Cell composition estimates varied both within and between cohorts, but reassuringly were robust to placental processing time. Interestingly, the ratio of cytotrophoblast to syncytiotrophoblast proportion decreased with increasing gestational age, and differed slightly by both maternal ethnicity (lower in white vs. non-white) and genetic ancestry (lower in higher probability European ancestry). The cohort of origin and cytotrophoblast proportion were the largest drivers of DNAme variation in this dataset, based on their associations with the first principal component. Conclusions : This work confirms that cohort, array (technical) batch, cell type proportion, self-reported ethnicity, genetic ancestry, and biological sex are important variables to consider in any analyses of Illumina DNAme data. Further, we demonstrate that estimating epiphenotype variables from the DNAme data itself, when possible, provides both an independent check of clinically-obtained data and can provide a robust approach to compare variables across different datasets.

Profiling placental DNA methylation associated with maternal SSRI treatment during pregnancy

Selective serotonin reuptake inhibitors (SSRIs) for treatment of prenatal maternal depression have been associated with neonatal neurobehavioral disturbances, though the molecular mechanisms remain poorly understood.  In utero exposure to SSRIs may affect DNA methylation (DNAme) in the human placenta, an epigenetic mark that is established during development and is associated with gene expression. Chorionic villus samples from 64 human placentas were profiled with the Illumina MethylationEPIC BeadChip; clinical assessments of maternal mood and SSRI treatment records were collected at multiple time points during pregnancy. Case distribution was 20 SSRI-exposed cases and 44 SSRI non-exposed cases. Maternal depression was defined using a mean maternal Hamilton Depression score > 8 to indicate symptomatic depressed mood ("maternally-depressed"), and we further classified cases into SSRI-exposed, maternally-depressed (n = 14); SSRI-exposed, not maternally-depressed (n = 6); SSRI non-exposed, maternally-depressed (n = 20); and SSRI non-exposed, not maternally-depressed (n = 24). For replication, Illumina 450K DNAme profiles were obtained from 34 additional cases from an independent cohort (n = 17 SSRI-exposed, n = 17 SSRI non-exposed). No CpGs were differentially methylated at FDR < 0.05 comparing SSRI-exposed to non-exposed placentas, in a model adjusted for mean maternal Hamilton Depression score, or in a model restricted to maternally-depressed cases with and without SSRI exposure. However, at a relaxed threshold of FDR < 0.25, five CpGs were differentially methylated (|Δβ| > 0.03) by SSRI exposure status. Four were covered by the replication cohort measured by the 450K array, but none replicated. No CpGs were differentially methylated (FDR < 0.25) comparing maternally depressed to not depressed cases. In sex-stratified analyses for SSRI-exposed versus non-exposed cases (females n = 31; males n = 33), three additional CpGs in females, but none in males, were differentially methylated at the relaxed FDR < 0.25 cut-off. We did not observe large-scale alterations of DNAme in placentas exposed to maternal SSRI treatment, as compared to placentas with no SSRI exposure. We also found no evidence for altered DNAme in maternal depression-exposed versus depression non-exposed placentas. This novel work in a prospectively-recruited cohort with clinician-ascertained SSRI exposure and mood assessments would benefit from future replication.

Genetic variants in genes related to inflammation, apoptosis and autophagy in breast cancer risk

BACKGROUND

Inflammation contributes to breast cancer development through its effects on cell damage. This damage is usually dealt with by key genes involved in apoptosis and autophagy pathways.

METHODS

We tested 206 single nucleotide polymorphisms (SNPs) in 54 genes related to inflammation, apoptosis and autophagy in a population-based breast cancer study of women of European (658 cases and 795 controls) and East Asian (262 cases and 127 controls) descent. Logistic regression was used to estimate odds ratios for breast cancer risk, and case-only analysis to compare breast cancer subtypes (defined by ER/PR/HER2 status), with adjustment for confounders. We assessed statistical interactions between the SNPs and lifestyle factors (smoking status, physical activity and body mass index).

RESULTS AND CONCLUSION

Although no SNP was associated with breast cancer risk among women of European descent, we found evidence for an association among East Asians for rs1800925 (IL-13) and breast cancer risk (OR = 2.08; 95% CI: 1.32-3.28; p = 0.000779), which remained statistically significant after multiple testing correction (padj = 0.0350). This association was replicated in a meta-analysis of 4305 cases and 4194 controls in the Shanghai Breast Cancer Genetics Study (OR 1.12, 95% CI: 1.03-1.21, p = 0.011). Further, we found evidence of an interaction between rs7874234 (TSC1) and physical activity among women of East Asian descent.

Polymorphisms of Insulin-Like Growth Factor 1 Pathway Genes and Breast Cancer Risk

Genetic variants of insulin-like growth factor 1 (IGF1) pathway genes have been shown to be associated with breast density and IGF1 levels and, therefore, may also influence breast cancer risk via pro-survival signaling cascades. The aim of this study was to investigate associations between IGF1 pathway single nucleotide polymorphisms (SNPs) and breast cancer risk among European and East Asian women, and potential interactions with menopausal status and breast tumor subtype. Stratified analyses of 1,037 cases and 1,050 controls from a population-based case–control study were conducted to assess associations with breast cancer for 22 SNPs across 5 IGF1 pathway genes in European and East Asian women. Odds ratios were calculated using logistic regression in additive genetic models. Polytomous logistic regression was used to assess heterogeneity by breast tumor subtype. Two SNPs of the IGF1 gene (rs1019731 and rs12821878) were associated with breast cancer risk among European women. Four highly linked IGF1 SNPs (rs2288378, rs17727841, rs7136446, and rs7956547) were modified by menopausal status among East Asian women only and associated with postmenopausal breast cancers. The association between rs2288378 and breast cancer risk was also modified by breast tumor subtype among East Asian women. Several IGF1 polymorphisms were found to be associated with breast cancer risk and some of these associations were modified by menopausal status or breast tumor subtype. Such interactions should be considered when assessing the role of these variants in breast cancer etiology.

DNA repair variants and breast cancer risk

A functional DNA repair system has been identified as important in the prevention of tumour development. Previous studies have hypothesized that common polymorphisms in DNA repair genes could play a role in breast cancer risk and also identified the potential for interactions between these polymorphisms and established breast cancer risk factors such as physical activity. Associations with breast cancer risk for 99 single nucleotide polymorphisms (SNPs) from genes in ten DNA repair pathways were examined in a case-control study including both Europeans (644 cases, 809 controls) and East Asians (299 cases, 160 controls). Odds ratios in both additive and dominant genetic models were calculated separately for participants of European and East Asian ancestry using multivariate logistic regression. The impact of multiple comparisons was assessed by correcting for the false discovery rate within each DNA repair pathway. Interactions between several breast cancer risk factors and DNA repair SNPs were also evaluated. One SNP (rs3213282) in the gene XRCC1 was associated with an increased risk of breast cancer in the dominant model of inheritance following adjustment for the false discovery rate (P < 0.05), although no associations were observed for other DNA repair SNPs. Interactions of six SNPs in multiple DNA repair pathways with physical activity were evident prior to correction for FDR, following which there was support for only one of the interaction terms (P < 0.05). No consistent associations between variants in DNA repair genes and breast cancer risk or their modification by breast cancer risk factors were observed.

Genetic variation in vitamin D-related genes and risk of breast cancer among women of European and East Asian descent

Studies of vitamin D-related genetic variants and breast cancer have been inconsistent. This study aimed to investigate associations of vitamin D-related polymorphisms and breast cancer risk among European and East Asian women and potential interactions with menopausal status and breast tumour subtypes. Data from a case-control study of breast cancer (1037 cases and 1050 controls) were used to assess relationships between 21 polymorphisms in two vitamin D-related genes (GC and VDR) and breast cancer risk. Odds ratios were calculated in stratified analyses of European and East Asian women, using logistic regression in an additive genetic model. An interaction term was used to explore modification by menopausal status. Polytomous regression was used to assess heterogeneity by breast tumour subtype. False discovery rate adjustments were conducted to account for multiple testing. No association was observed between GC or VDR polymorphisms and breast cancer risk. Modification of these relationships by menopausal status was observed for select polymorphisms in both Europeans (VDR rs4328262 and rs11168292) and East Asians (GC rs7041 and VDR rs11168287). Heterogeneity by tumour subtype was seen for three VDR polymorphisms (rs1544410, rs7967152 and rs2239186) among Europeans, in which associations with ER-/PR-/HER2+ tumours, but not with other subtypes, were observed. In conclusion, associations between vitamin D-related genetic variants and breast cancer were not observed overall, although the relationships between vitamin D pathway polymorphisms and breast cancer may be modified by menopausal status and breast tumour subtype.

Non-Hodgkin lymphoma risk and variants in genes controlling lymphocyte development

Non-Hodgkin lymphomas (NHL) are a heterogeneous group of solid tumours of lymphoid cell origin. Three important aspects of lymphocyte development include immunity and inflammation, DNA repair, and programmed cell death. We have used a previously established case-control study of NHL to ask whether genetic variation in genes involved in these three important processes influences risk of this cancer. 118 genes in these three categories were tagged with single nucleotide polymorphisms (SNPs), which were tested for association with NHL and its subtypes. The main analysis used logistic regression (additive model) to estimate odds ratios in European-ancestry cases and controls. 599 SNPs and 1116 samples (569 cases and 547 controls) passed quality control measures and were included in analyses. Following multiple-testing correction, one SNP in MSH3, a mismatch repair gene, showed an association with diffuse large B-cell lymphoma (OR: 1.91; 95% CI: 1.41–2.59; uncorrected p = 0.00003; corrected p = 0.010). This association was not replicated in an independent European-ancestry sample set of 251 diffuse large B-cell lymphoma cases and 737 controls, indicating this result was likely a false positive. It is likely that moderate sample size, inter-subtype and other genetic heterogeneity, and small true effect sizes account for the lack of replicable findings.

Sex- and subtype-specific analysis of H2AFX polymorphisms in non-Hodgkin lymphoma

H2AFX encodes a histone variant involved in signaling sites of DNA damage and recruiting repair factors. Genetic variants in H2AFX may influence risk of non-Hodgkin lymphoma (NHL), a heterogeneous group of lymphoid tumors that are characterized by chromosomal translocations. We previously reported that rs2509049, a common variant in the promoter of H2AFX, was associated with risk for NHL in the British Columbia population. Here we report results for 13 single nucleotide polymorphisms (SNPs) in 100 Kb surrounding H2AFX in an expanded collection of 568 NHL cases and 547 controls. After correction for multiple testing, significant associations were present for mantle cell lymphoma (p=0.007 for rs604714) and all B-cell lymphomas (p=0.046 for rs2509049). Strong linkage disequilibrium in the 5 Kb upstream of H2AFX limited the ability to determine which specific SNP (rs2509049, rs7759, rs8551, rs643788, rs604714, or rs603826), if any, was responsible. There was a significant interaction between sex and rs2509049 in the all B-cell lymphomas group (p=0.002); a sex-stratified analysis revealed that the association was confined to females (p=0.001). Neither the overall nor the female-specific association with rs2509049 was replicated in any of four independent NHL sample sets. Meta-analysis of all five study populations (3,882 B-cell NHL cases and 3,718 controls) supported a weak association with B-cell lymphoma (OR=0.92, 95% CI=0.86-0.99, p=0.034), although this association was not significant after exclusion of the British Columbia data. Further research into the potential sex-specificity of the H2AFX-NHL association may identify a subset of NHL cases that are influenced by genotype at this locus.

Shift work, circadian gene variants and risk of breast cancer

Circadian (clock) genes have been linked with several functions relevant to cancer, and epidemiologic research has suggested relationships with breast cancer risk for variants in NPAS2, CLOCK, CRY2 and TIMELESS. Increased breast cancer risk has also been observed among shift workers, suggesting potential interactions in relationships of circadian genes with breast cancer. Relationships with breast cancer of 100 SNPs in 14 clock-related genes, as well as potential interactions with shift work history, were investigated in a case-control study (1042 cases, 1051 controls). Odds ratios in an additive genetic model for European-ancestry participants (645 cases, 806 controls) were calculated, using a two-step correction for multiple testing: within each gene through permutation testing (10,000 permutations), and correcting for the false discovery rate across genes. Interactions of genotypes with ethnicity and shift work (<2 years vs ≥2 years) were evaluated individually. Following permutation analysis, two SNPs (rs3816360 in ARNTL and rs11113179 in CRY1) displayed significant associations with breast cancer and one SNP (rs3027188 in PER1) was marginally significant; however, none were significant following adjustment for the false discovery rate. No significant interaction with shift work history was detected. If shift work causes circadian disruption, this was not reflected in associations between clock gene variants and breast cancer risk in this study. Larger studies are needed to assess interactions with longer durations (>30 years) of shift work that have been associated with breast cancer.

Absence of NR2E1 mutations in patients with aniridia

PURPOSE

Nuclear receptor 2E1 (NR2E1) is a transcription factor with many roles during eye development and thus may be responsible for the occurrence of certain congenital eye disorders in humans. To test this hypothesis, we screened NR2E1 for candidate mutations in patients with aniridia and other congenital ocular malformations (anterior segment dysgenesis, congenital optic nerve malformation, and microphthalmia).

METHODS

The NR2E1 coding region, 5' and 3' untranslated regions (UTRs), exon flanking regions including consensus splice sites, and six evolutionarily conserved non-coding candidate regulatory regions were analyzed by sequencing 58 probands with aniridia of whom 42 were negative for PAX6 mutations. Nineteen probands with anterior segment dysgenesis, one proband with optic nerve malformation, and two probands with microphthalmia were also sequenced. The control population comprised 376 healthy individuals. All sequences were analyzed against the GenBank sequence AL078596.8 for NR2E1. In addition, the coding region and flanking intronic sequences of FOXE3, FOXC1, PITX2, CYP1B1, PAX6, and B3GALTL were sequenced in one patient and his relatives.

RESULTS

Sequencing analysis showed 17 NR2E1 variants including two novel rare non-coding variants (g.-1507G>A, g.14258C>T), and one novel rare coding variant (p.Arg274Gly). The latter was present in a male diagnosed with Peters' anomaly who subsequently was found to have a known causative mutation for Peters' plus syndrome in B3GALTL (c.660+1G>A). In addition, the NR2E1 novel rare variant Arg274Gly was present in the unaffected mother of the patient but absent in 746 control chromosomes.

CONCLUSIONS

We eliminated a major role for NR2E1 regulatory and coding mutations in aniridia and found a novel rare coding variant in NR2E1. In addition, we found no coding region variation in the control population for NR2E1, which further supports its previously reported high level of conservation and low genetic diversity. Future NR2E1 studies in ocular disease groups such as those involving retinal and optic nerve abnormalities should be undertaken to determine whether NR2E1 plays a role in these conditions.

Catenin family genes are not commonly mutated in hereditary diffuse gastric cancer

BACKGROUND

Approximately one third of the hereditary diffuse gastric cancer (HDGC) families carry germline mutations in the E-cadherin gene (CDH1). Risk prediction in members of families with this rare but deadly cancer could be improved by the identification of additional HDGC genes in non-CDH1 families.

METHODS

Affected individuals from 22 CDH1 mutation-negative families were screened for germline mutations in four catenin genes: CTNNA1, CTNNB1, JUP, and CTNND1. Catenins interact closely with E-cadherin molecules in cells, and are therefore logical candidate genes for mutation screening in HDGC families.

RESULTS

No nonsynonymous variants were seen in CTNNA1, CTNNB1, or CTNND1; only JUP contained nonsynonymous variants, of which only two rare variants were predicted to be deleterious.

CONCLUSION

Catenin genes are not commonly mutated in non-CDH1 HDGC families.

IMPACT

Germline mutations in CTNNA1, CTNNB1, JUP, or CTNND1 are unlikely to play a major role in HDGC.

Genetic variation in cell death genes and risk of non-Hodgkin lymphoma

Background

Non-Hodgkin lymphomas are a heterogeneous group of solid tumours that constitute the 5^th highest cause of cancer mortality in the United States and Canada. Poor control of cell death in lymphocytes can lead to autoimmune disease or cancer, making genes involved in programmed cell death of lymphocytes logical candidate genes for lymphoma susceptibility.

Materials and Methods

We tested for genetic association with NHL and NHL subtypes, of SNPs in lymphocyte cell death genes using an established population-based study. 17 candidate genes were chosen based on biological function, with 123 SNPs tested. These included tagSNPs from HapMap and novel SNPs discovered by re-sequencing 47 cases in genes for which SNP representation was judged to be low. The main analysis, which estimated odds ratios by fitting data to an additive logistic regression model, used European ancestry samples that passed quality control measures (569 cases and 547 controls). A two-tiered approach for multiple testing correction was used: correction for number of tests within each gene by permutation-based methodology, followed by correction for the number of genes tested using the false discovery rate.

Results

Variant rs928883, near miR-155, showed an association (OR per A-allele: 2.80 [95% CI: 1.63–4.82]; p_F = 0.027) with marginal zone lymphoma that is significant after correction for multiple testing.

Conclusions

This is the first reported association between a germline polymorphism at a miRNA locus and lymphoma.

BCL2 mutations in diffuse large B-cell lymphoma

BCL2 is deregulated in diffuse large B-cell lymphoma (DLBCL) by the t(14;18) translocation, gene amplification and/or nuclear factor-κB signaling. RNA-seq data have recently shown that BCL2 is the most highly mutated gene in germinal center B-cell (GCB) DLBCL. We have sequenced BCL2 in 298 primary DLBCL biopsies, 131 additional non-Hodgkin lymphoma biopsies, 24 DLBCL cell lines and 51 germline DNAs. We found frequent BCL2 mutations in follicular lymphoma (FL) and GCB DLBCL, but low levels of BCL2 mutations in activated B-cell DLBCL, mantle cell lymphoma, small lymphocytic leukemia and peripheral T-cell lymphoma. We found no BCL2 mutations in GC centroblasts. Many mutations were non-synonymous; they were preferentially located in the flexible loop domain, with few in BCL2-homology domains. An elevated transition/transversions ratio supports that the mutations result from somatic hypermutation. BCL2 translocations correlate with, and are likely important in acquisition of, additional BCL2 mutations in GCB DLBCL and FL. DLBCL mutations were not independently associated with survival. Although previous studies of BCL2 mutations in FL have reported mutations to result in pseudo-negative BCL2 protein expression, we find this rare in de-novo DLBCL.

Cost-effective prediction of gender-labeling errors and estimation of gender-labeling error rates in candidate-gene association studies

We describe a statistical approach to predict gender-labeling errors in candidate-gene association studies, when Y-chromosome markers have not been included in the genotyping set. The approach adds value to methods that consider only the heterozygosity of X-chromosome SNPs, by incorporating available information about the intensity of X-chromosome SNPs in candidate genes relative to autosomal SNPs from the same individual. To our knowledge, no published methods formalize a framework in which heterozygosity and relative intensity are simultaneously taken into account. Our method offers the advantage that, in the genotyping set, no additional space is required beyond that already assigned to X-chromosome SNPs in the candidate genes. We also show how the predictions can be used in a two-phase sampling design to estimate the gender-labeling error rates for an entire study, at a fraction of the cost of a conventional design.

Genetic variation in the NBS1, MRE11, RAD50 and BLM genes and susceptibility to non-Hodgkin lymphoma

BACKGROUND

Translocations are hallmarks of non-Hodgkin lymphoma (NHL) genomes. Because lymphoid cell development processes require the creation and repair of double stranded breaks, it is not surprising that disruption of this type of DNA repair can cause cancer. The members of the MRE11-RAD50-NBS1 (MRN) complex and BLM have central roles in maintenance of DNA integrity. Severe mutations in any of these genes cause genetic disorders, some of which are characterized by increased risk of lymphoma.

METHODS

We surveyed the genetic variation in these genes in constitutional DNA of NHL patients by means of gene re-sequencing, then conducted genetic association tests for susceptibility to NHL in a population-based collection of 797 NHL cases and 793 controls.

RESULTS

114 SNPs were discovered in our sequenced samples, 61% of which were novel and not previously reported in dbSNP. Although four variants, two in RAD50 and two in NBS1, showed association results suggestive of an effect on NHL, they were not significant after correction for multiple tests.

CONCLUSION

These results suggest an influence of RAD50 and NBS1 on susceptibility to diffuse large B-cell lymphoma and marginal zone lymphoma. Larger association and functional studies could confirm such a role.

A survey of genomic properties for the detection of regulatory polymorphisms

Advances in the computational identification of functional noncoding polymorphisms will aid in cataloging novel determinants of health and identifying genetic variants that explain human evolution. To date, however, the development and evaluation of such techniques has been limited by the availability of known regulatory polymorphisms. We have attempted to address this by assembling, from the literature, a computationally tractable set of regulatory polymorphisms within the ORegAnno database (http://www.oreganno.org). We have further used 104 regulatory single-nucleotide polymorphisms from this set and 951 polymorphisms of unknown function, from 2-kb and 152-bp noncoding upstream regions of genes, to investigate the discriminatory potential of 23 properties related to gene regulation and population genetics. Among the most important properties detected in this region are distance to transcription start site, local repetitive content, sequence conservation, minor and derived allele frequencies, and presence of a CpG island. We further used the entire set of properties to evaluate their collective performance in detecting regulatory polymorphisms. Using a 10-fold cross-validation approach, we were able to achieve a sensitivity and specificity of 0.82 and 0.71, respectively, and we show that this performance is strongly influenced by the distance to the transcription start site.