Associations of mdNLRs with TNBC
The methylation levels of the selected neutrophil- and pan-lymphocyte-specific CpG sites are shown in Figure 2A. The neutrophil-specific sites are methylated in the neutrophils and unmethylated in pan-lymphocytes (main subtypes: NK, TCD4+, TCD8+, and B cells) and monocytes. The pan-lymphocyte-specific sites are methylated in the pan-lymphocytes and unmethylated in neutrophils and monocytes. The characteristics of the CpG sites are provided in Table 1.
Methylation analysis at the selected CpG sites in 231 TNBC cases and 231 controls revealed a significantly higher ratio of neutrophils and lower ratio of pan-lymphocytes in TNBC cases compared with controls (mean ratios: 49% vs. 45% and 36% vs. 40.3%, respectively; conditional logistic regression: all Padj.<1e-04) (Figure 2B). The neutrophil ratios were associated with a higher likelihood of being a TNBC case (OR range (2.07-3.02); conditional logistic regression: all Padj.<1e-04) and the pan-lymphocyte ratios with a lower likelihood (OR range (0.56-0.61); conditional logistic regression: all Padj.<1e-04) (Suppl. Table 3). There were no associations of neutrophil and pan-lymphocyte ratios with histopathological characteristics of TNBCs including grade, size, node status, stage, and OS (size/stage: Jonckheere-Terpstra trend test; Grade/Node status: Mann-Whitney test; OS: Cox regression; all P>0.05). The average means of the mdNLR for cases and controls were 1.5 ± 0.63 and 1.18 ± 0.43, respectively. Logistic regression analysis revealed that all nine mdNLRs were associated with an increased likelihood of being a TNBC case (OR range (2.66-4.29); all Padj.<1e-04) (Figure 3). No associations were observed for any of the nine mdNLRs with tumor histopathological characteristics and OS (size/stage: Jonckheere-Terpstra trend test; Grade/Node status: Mann-Whitney test; OS: Cox regression; all P>0.05).
To validate our mdNLR findings, we used the reference-based method (mdNLRref), which estimates cell proportions. A higher mdNLRref was also observed in TNBC cases compared with controls (2.49 ± 1.53 versus 1.81 ± 0.94) (Suppl. Fig. 1), which was associated with a higher likelihood of being a TNBC case (OR=1.71 [1.38-2.11], conditional logistic regression; Padj.<1e-04) (Suppl. Fig. 2). The mdNLR and mdNLRsref were highly correlated (Suppl. Fig. 3), with the mdNLR cg23954655.cg26942829 ratio showing the highest correlation (Spearman’s rank correlation; r=0.97).
Adjusting additionally for confounding factors (BMI, menopausal status, and smoking status) in a multivariable conditional logistic regression model, the associations of neutrophil and pan-lymphocyte ratios, mdNLRs, and mdNLRref with TNBC did not change fundamentally (Suppl. Tables 3-5).
Associations of Leukocyte Subtype Ratios with TNBC
The methylation levels of the 21 selected ISUS are shown in Figure 4A. All 21 ISUS were hypomethylated in the target cell types but methylated in the other subtypes. The characteristics of each ISUS are provided in Table 2. In order to investigate the association between leukocyte subtypes ratios with TNBC in the retrospective sample set, the immune cell type ratios of TNBC cases were compared with those in controls using beta values of the specific ISUS. Analysis of these proxies showed that six of seven immune cell type ratios were associated with TNBC. Lower ratios of NK, TCD4+, TCD8+ cells, monocytes, and B cells in cases compared with controls were associated with TNBC (Figure 4B), with decreased NK cell ratios showing the strongest association. Further, a higher ratio of neutrophils in TNBC cases compared with controls was associated with TNBC, while no difference between the two groups was found in the Treg cell ratio.
To validate our findings obtained with the immune cell methylation proxies, we applied immune cell proportions estimated by reference-based deconvolution method. For each sample, the estimated proportions of the six immune cell types sum to one. Univariable comparison of the various immune cell types between TNBC cases and controls revealed a statistically significant difference in the proportions of neutrophils, NK, TCD4+, and B cells (Suppl. Fig. 4). No difference was observed in the proportions of TCD8+ cells and monocytes. Logistic regression analysis showed associations of neutrophils, NK, TCD4+, and B cell proportions with TNBC, with a decreased NK cell proportion showing the strongest association signal (Suppl. Fig. 2).
After adjustment for confounding factors, the associations of the leukocyte subtype ratios and proportions remained statistically significant except the association with cg07499259, which was no longer statistically significant (Suppl. Tables 5 and 6).
Correlation of Leukocyte Subtype Ratios with Clinical, Epidemiological, and Histopathological Parameters of the Participants of the Retrospective Study
Correlation analysis of the immune cell type ratios with selected clinical and epidemiological characteristics of the study participants (age, menopausal status, smoking status, body mass index) revealed correlations with smoking status and age. The neutrophil and B cell ratios correlated with smoking status in controls. Current smokers had a lower neutrophil and a higher B cell ratio compared with non-smokers (Mann-Whitney test; Padj<0.05). Two immune cell type ratios correlated with age. The TCD8+ cell ratio showed an inverse correlation with age in both cases and controls and the NK cell ratio a positive correlation in controls (Spearman’s rank correlation; Padj<0.05). No other correlations were observed. There were no correlations of the immune cell type ratios with selected histopathological tumor characteristics (grade, size, node status, stage).
Diagnostic and Prognostic Performance of ISUS
AUC analysis showed that NK cells and neutrophils had the highest discriminative capability among all immune cell types (Figure 4B). The estimated NK cell-to-neutrophil ratio was higher in controls compared with cases and slightly improved the discrimination performance between cases and controls with AUC values in the range (0.67-0.71) (Suppl. Fig. 5) relative to the values from individual CpGs in the range (0.63-0.67) (Figure 4B). The NK cell-to-neutrophil ratios were associated with TNBC: a higher NK cell-to-neutrophil ratio was associated with a lower likelihood of being a TNBC cases (OR range (0.52-0-70); conditional logistic regression; all Padj.<1e-04) (Suppl. Table 7). Next, a diagnostic model was developed by fitting a multivariable logistic regression model based on all 21 ISUS and applying backward variable selection. A bootstrap-adjusted AUC was computed to account for overfitting. The final model contained four ISUS that discriminated cases from controls with an AUC of 72%. Of these four ISUS, two were specific for NK cells, one for monocytes, and one for TCD8+ cells (Table 3).
The prognostic performance using log-rank test showed that two probes, cg00219921 and cg08326410, which are specific for TCD8+ and NK cells, were associated with survival when using a median split. Higher ratios of TCD8+ and NK cells were associated with a better patient OS (Suppl. Fig. 6). Using a multivariable Cox regression model including age, tumor grade, stage, tumor size, and lymph node status, only the association with cg00219921 remained statistically significant (Cox regression; P=0.04). However, after adjustment for multiple testing, the association lost statistical significance.
Association of the NK Cell Ratio with TNBC in Participants of the Prospective Case-Control Study
Since NK cells were the most pronounced immune cell type associated with TNBC in the retrospective sample set, we investigated whether the observed association could be detected in pre-diagnostic DNA samples of TNBC cases compared with controls. In this respect, the NK cell ratio was measured in 146 TNBC cases and 146 controls using a ddPCR TaqMan assay specific for one NK cell-specific unmethylated site (cg23060465). The obtained “unmethylation“ levels from ddPCR were Logit-transformed and the Wilcoxon signed rank test was used to compare the obtained NK cell ratios between cases and controls. A lower NK cell ratio was observed in TNBC cases compared with controls (Wilcoxon signed rank test, P=0.019) (Figure 5). Using conditional logistic regression, a higher NK cell ratio was associated with a reduced TNBC risk at the margin of statistical significance (OR=0.76, 95% CI [0.58-1.00], conditional logistic regression; P=0.052). Heterogeneity due to age at blood draw, age of diagnosis, and interval time between blood draw and reference date was tested. No heterogeneity/subgroup effect was observed (Interaction test based on conditional logistic regression; all P>0.05).