Limits...
Normal cell-type epigenetics and breast cancer classification: a case study of cell mixture-adjusted analysis of DNA methylation data from tumors.

Houseman EA, Ince TA - Cancer Inform (2014)

Bottom Line: Historically, breast cancer classification has relied on prognostic subtypes.This classification scheme has been shown to have relevance to clinical prognosis.Taken together, these findings suggest that the epigenetics of breast tumors is ultimately based on the underlying phylogeny of normal breast tissue.

View Article: PubMed Central - PubMed

Affiliation: School of Biological and Population Health Sciences, College of Public Health and Human Sciences, Oregon State University, Corvallis, OR, USA.

ABSTRACT
Historically, breast cancer classification has relied on prognostic subtypes. Thus, unlike hematopoietic cancers, breast tumor classification lacks phylogenetic rationale. The feasibility of phylogenetic classification of breast tumors has recently been demonstrated based on estrogen receptor (ER), androgen receptor (AR), vitamin D receptor (VDR) and Keratin 5 expression. Four hormonal states (HR0-3) comprising 11 cellular subtypes of breast cells have been proposed. This classification scheme has been shown to have relevance to clinical prognosis. We examine the implications of such phylogenetic classification on DNA methylation of both breast tumors and normal breast tissues by applying recently developed deconvolution algorithms to three DNA methylation data sets archived on Gene Expression Omnibus. We propose that breast tumors arising from a particular cell-of-origin essentially magnify the epigenetic state of their original cell type. We demonstrate that DNA methylation of tumors manifests patterns consistent with cell-specific epigenetic states, that these states correspond roughly to previously posited normal breast cell types, and that estimates of proportions of the underlying cell types are predictive of tumor phenotypes. Taken together, these findings suggest that the epigenetics of breast tumors is ultimately based on the underlying phylogeny of normal breast tissue.

No MeSH data available.


Related in: MedlinePlus

Correspondence of estimated normal cell proportions. Panel (A): left subpanel indicates estimated proportions; right subpanel illustrates variation among bootstrap samples. Panel (B): distributions of cell proportions inferred from tumor data or normal cell data obtained from Santagata et al.10
© Copyright Policy - open-access
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4264613&req=5

f3-cin-suppl.4-2014-053: Correspondence of estimated normal cell proportions. Panel (A): left subpanel indicates estimated proportions; right subpanel illustrates variation among bootstrap samples. Panel (B): distributions of cell proportions inferred from tumor data or normal cell data obtained from Santagata et al.10

Mentions: We next sought to compare the biological significance of Δ-based coefficients with B-based coefficients for normal breast tissue. Table 3 summarizes normal breast tissue available via the three GEO data sets used in this analysis. We first used mRNA expression data for Estrogen Receptor 1 (ESR1), AR, and VDR to assign each “tumor” in GSE20712 to one of eight categories. These categories are mapped to 11 normal cell-type categories, each defined by expression of ER, AR, VDR, and K5, as described previously10 and as shown in Table 4. Additionally, to address potential contamination by blood, we used DNA methylation data from 500 CpGs to infer leukocyte proportions for each tumor, using our reference-based deconvolution algorithm.15 Over the union of 100 leukocyte DMRs with the 5000 CpGs having most significant Δ coefficients (for a total of 5081 CpGs), we estimated the mean methylation matrix MΔ for the eight hormonal cell subgroup categories described in Table 4, subsequently applying our reference-based deconvolution algorithm15 to normal data described in Table 3 to obtain cell proportion estimates for each of 14 cell types (8 breast cells and 6 leukocyte types). We repeated the analysis using top 5000 CpGs having the most significant B coefficients (for a total of 5077 CpGs), obtaining the corresponding cell proportions . Finally, as a basis of comparison, we used data provided in Supplementary Tables S2A–S2D of Santagata et al.10 to infer the proportion of normal cell types in normal breast tissue. Figure 3A illustrates the proportion of normal cell types inferred from the Santagata et al data,10 while Figure 3B illustrates the correspondence between these inferred proportions and proportions and of cell-type obtained from GEO data. Note that bootstrap estimates of cell-type probabilities are also shown in Figure 3A and were used to generate the distribution of parameter estimates shown in gray in Figure 3B. Additionally, Figures S5 and S6 depict the GEO-based estimates in clustering heat map format. The absolute correspondence was far from perfect in either case ; in particular, the GEO analysis suggested elevated proportions of L11 cells, while the Santagata et al data suggest relatively low proportions. However, in all other respects, there was general semiquantitative agreement: in both analyses, proportions of L1–3 and L6–7 were elevated relative to L4, L5, L8, L9, and L10. Note that while imperfect in both cases, the correspondence was worse for (8 degrees-of-freedom χ2 statistic = 1528) than for (8 degree-of-freedom χ2 statistic = 2608). The clustering heat map shown in Figure S6 shows that the leukocyte assignment appears principally in the natural killer (NK) category, ie, for a relatively rare cell type in comparison to granulocytes or CD4+ T cells; such a finding would contradict overall contamination of tumor sample by blood, and instead suggest either biased estimates MB of normal cell-specific epigenetic states or infiltration of tumor via NK cells, in either case failing to reflect normal breast cell activity. Note also that a paired t-test comparing the total proportion assigned leukocyte cells differed significantly (P < 10−12) between and with greater leukocyte proportion assigned for ; this finding, illustrated in Figure S7, provides additional evidence that represents greater misclassification with respect to normal breast cell phylogenetics.


Normal cell-type epigenetics and breast cancer classification: a case study of cell mixture-adjusted analysis of DNA methylation data from tumors.

Houseman EA, Ince TA - Cancer Inform (2014)

Correspondence of estimated normal cell proportions. Panel (A): left subpanel indicates estimated proportions; right subpanel illustrates variation among bootstrap samples. Panel (B): distributions of cell proportions inferred from tumor data or normal cell data obtained from Santagata et al.10
© Copyright Policy - open-access
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4264613&req=5

f3-cin-suppl.4-2014-053: Correspondence of estimated normal cell proportions. Panel (A): left subpanel indicates estimated proportions; right subpanel illustrates variation among bootstrap samples. Panel (B): distributions of cell proportions inferred from tumor data or normal cell data obtained from Santagata et al.10
Mentions: We next sought to compare the biological significance of Δ-based coefficients with B-based coefficients for normal breast tissue. Table 3 summarizes normal breast tissue available via the three GEO data sets used in this analysis. We first used mRNA expression data for Estrogen Receptor 1 (ESR1), AR, and VDR to assign each “tumor” in GSE20712 to one of eight categories. These categories are mapped to 11 normal cell-type categories, each defined by expression of ER, AR, VDR, and K5, as described previously10 and as shown in Table 4. Additionally, to address potential contamination by blood, we used DNA methylation data from 500 CpGs to infer leukocyte proportions for each tumor, using our reference-based deconvolution algorithm.15 Over the union of 100 leukocyte DMRs with the 5000 CpGs having most significant Δ coefficients (for a total of 5081 CpGs), we estimated the mean methylation matrix MΔ for the eight hormonal cell subgroup categories described in Table 4, subsequently applying our reference-based deconvolution algorithm15 to normal data described in Table 3 to obtain cell proportion estimates for each of 14 cell types (8 breast cells and 6 leukocyte types). We repeated the analysis using top 5000 CpGs having the most significant B coefficients (for a total of 5077 CpGs), obtaining the corresponding cell proportions . Finally, as a basis of comparison, we used data provided in Supplementary Tables S2A–S2D of Santagata et al.10 to infer the proportion of normal cell types in normal breast tissue. Figure 3A illustrates the proportion of normal cell types inferred from the Santagata et al data,10 while Figure 3B illustrates the correspondence between these inferred proportions and proportions and of cell-type obtained from GEO data. Note that bootstrap estimates of cell-type probabilities are also shown in Figure 3A and were used to generate the distribution of parameter estimates shown in gray in Figure 3B. Additionally, Figures S5 and S6 depict the GEO-based estimates in clustering heat map format. The absolute correspondence was far from perfect in either case ; in particular, the GEO analysis suggested elevated proportions of L11 cells, while the Santagata et al data suggest relatively low proportions. However, in all other respects, there was general semiquantitative agreement: in both analyses, proportions of L1–3 and L6–7 were elevated relative to L4, L5, L8, L9, and L10. Note that while imperfect in both cases, the correspondence was worse for (8 degrees-of-freedom χ2 statistic = 1528) than for (8 degree-of-freedom χ2 statistic = 2608). The clustering heat map shown in Figure S6 shows that the leukocyte assignment appears principally in the natural killer (NK) category, ie, for a relatively rare cell type in comparison to granulocytes or CD4+ T cells; such a finding would contradict overall contamination of tumor sample by blood, and instead suggest either biased estimates MB of normal cell-specific epigenetic states or infiltration of tumor via NK cells, in either case failing to reflect normal breast cell activity. Note also that a paired t-test comparing the total proportion assigned leukocyte cells differed significantly (P < 10−12) between and with greater leukocyte proportion assigned for ; this finding, illustrated in Figure S7, provides additional evidence that represents greater misclassification with respect to normal breast cell phylogenetics.

Bottom Line: Historically, breast cancer classification has relied on prognostic subtypes.This classification scheme has been shown to have relevance to clinical prognosis.Taken together, these findings suggest that the epigenetics of breast tumors is ultimately based on the underlying phylogeny of normal breast tissue.

View Article: PubMed Central - PubMed

Affiliation: School of Biological and Population Health Sciences, College of Public Health and Human Sciences, Oregon State University, Corvallis, OR, USA.

ABSTRACT
Historically, breast cancer classification has relied on prognostic subtypes. Thus, unlike hematopoietic cancers, breast tumor classification lacks phylogenetic rationale. The feasibility of phylogenetic classification of breast tumors has recently been demonstrated based on estrogen receptor (ER), androgen receptor (AR), vitamin D receptor (VDR) and Keratin 5 expression. Four hormonal states (HR0-3) comprising 11 cellular subtypes of breast cells have been proposed. This classification scheme has been shown to have relevance to clinical prognosis. We examine the implications of such phylogenetic classification on DNA methylation of both breast tumors and normal breast tissues by applying recently developed deconvolution algorithms to three DNA methylation data sets archived on Gene Expression Omnibus. We propose that breast tumors arising from a particular cell-of-origin essentially magnify the epigenetic state of their original cell type. We demonstrate that DNA methylation of tumors manifests patterns consistent with cell-specific epigenetic states, that these states correspond roughly to previously posited normal breast cell types, and that estimates of proportions of the underlying cell types are predictive of tumor phenotypes. Taken together, these findings suggest that the epigenetics of breast tumors is ultimately based on the underlying phylogeny of normal breast tissue.

No MeSH data available.


Related in: MedlinePlus