To ascertain the prevalence of CDH1 methylation in ILCs characterized by both the absence of CDH1 mutation in the exonic region and E-cadherin expression, we conducted quantitative pyrosequencing on a cohort of 22 ILC cases that had undergone comprehensive massive parallel sequencing, revealing a lack of CDH1 mutations and complete E-cadherin expression absence [11, 13, 14]. For comparative purposes, we analyzed 15 ILC cases with CDH1 mutations and full E-cadherin expression loss, along with 19 IBC-NSTs marked by preserved E-cadherin expression and no CDH1 mutation and five cases of usual ductal hyperplasia (UDH). The main clinicopathological and molecular data of the patients are presented in Supplementary Table 2. In summary, tumors were diagnosed in patients aged between 36 and 92 years old (median = 63 years old). Concerning the histological type, 30% of the ILC tumors exhibited pleomorphic lobular characteristics. In terms of molecular subtype, 86% of the tumors were luminal (32 ILC and 16 IBC-NST), 12% were triple-negative (4 ILC and 3 IBC-NST), and 2% were luminal HER2+ (1 ILC).

Pyrosequencing is a high-resolution method for the detection of DNA methylation and provides quantitative information for each CpG site under study, allowing for the control of bisulfite conversion efficiency. Pyrosequencing is the technique with the best reproducibility (even higher than methylation array) and can work well even on minute amounts of highly fragmented DNA [15].

Within this cohort of 56 primary tumors, we comprehensively examined the methylation status of 18 CpG dinucleotides situated in the CpG island (103 bp) of the CDH1 gene and in the Northern and Southern shore (N-shore, S-shore) regions. Analyzing the differences in methylation frequency according to molecular phenotype (triple-negative vs. luminal), we did not observe statistically significant differences, neither in general nor in any of the methylation zones (N-shore, island, and S-shore) (p-value > 0.05). The median methylation frequency in ILCs was 12%, while in IBC-NSTs, it was 15% (p-value > 0.05). The CpG island region encompassed the majority of sites explored in prior MSP studies as well as four CpGs as scrutinized by Ciriello et al. [9] and five CpGs by Alexander et al. [10] through methylation arrays (Supplementary Tables 3, 4, 5). Notably, the observed methylation values in the CpG island were generally modest (ranging between 3 and 18%) (Fig. 1a). CpG methylation values statistically differ among the studied groups for CpG sites at positions 68737141, 68737278, 68737296, and 68737299 located in the CpG island region. Interestingly, these CpGs exhibited slightly heightened methylation levels in IBC-NSTs (p-value < 0.05) (Supplementary Fig. 1 B-E). Furthermore, there were significant differences in methylation levels in the whole island region between the group of mutated ILCs and IBC-NSTs, the latter being higher (Fig. 1b).

a DNA methylation status of sequenced CpGs sites aligning with the CDH1 gene. Grey boxes = data not available. b Violin plots depicting methylation data across each tumor subgroup within the respective region

Tissue- and cancer-specific differentially methylated regions can occur not only within CpG islands themselves but also within CpG island shores, regions of relatively low CpG density, situated proximal to conventional promoter CpGs (up to 2 kb distant). This suggests the potential involvement of shore methylation in tissue differentiation, epigenetic reprogramming, and cancer [16]. Intriguingly, the analysis of CDH1 shore methylation has not been analyzed in MSP studies (Supplementary Table 3). Therefore, we extended our primer design to CpGs located in both N-shore and S-shore (Supplementary Table 3). Methylation levels in these regions were, in general, higher than in the CpG island (ranging from 4 to 35% and 14 to 64%, respectively) (Fig. 1a). There were significant differences between the studied groups for CpG site at position 68737077 in terms of CpG site-specific comparisons (Supplementary Fig. 1A), but there were no significant differences for whole region assessments (Fig. 1b).

We further examined these CpG sites in the five non-tumoral tissue samples, revealing lower methylation percentages compared to tumor samples across all regions (Fig. 1a).

In an effort to corroborate our findings, we compare our results with those reported by Ciriello et al. [9] and Alexander et al. [10]. Unfortunately, the available datasets from Ciriello et al. lack explicit specification of methylation beta values corresponding to the individual probes, offering a graphical overview instead. Since they did not make a differential analysis between the methylation status of CDH1-mutated and non-mutated cases, we compared the methylation frequencies at each CpG site for both ILC groups combined (with and without CDH1 mutation). Conversely, the dataset provided by Alexander et al. [10] allowed us to compare methylation levels in ILCs according to CDH1 mutation status, although the small number of cases lacking CDH1 mutation (n = 4) was a significant limitation of data reproducibility. In general, we observed similarity in methylation levels when compared to those outlined by Ciriello et al. [9], while we demonstrated lower methylation levels in contrast to those observed in the study by Alexander et al. [10] (Supplementary Tables 4 and 5). Furthermore, Fridrichova et al. [17] reported CDH1 methylation levels assessed by pyrosequencing across seven identical CpGs situated within the CpG island among 24 ILC cases, 178 invasive ductal carcinoma, and four other breast cancer patients. Although the mutational status of ILC cases was not assessed in this study, consistent with our current results, there were no disparities in DNA methylation across these groups, and the average value in tumors and paired lymph node metastasis remained below 10.5% [17].

While we did not observe substantial differences in CDH1 methylation across diverse tumor subtypes, noteworthy instances of elevated methylation were noted in selected tumors, such as cases 2, 8, 11, 50, or 56, among others. The relevant aspect to be considered is that CDH1 methylation can occur in TILs, thereby introducing a confounding element that can lead to false positive outcomes, particularly when using MSP [5]. To confirm this hypothesis, we conducted a correlation analysis between TILs and methylation levels across different CpGs, unveiling a modest yet statistically significant correlation between TILs and methylation levels across all examined regions (p-value < 0.05) (Fig. 2). In our observations, the percentage of TILs was found to be slightly higher in the IBC-NST group (median = 30%), exhibiting a statistically significant difference compared to the mutated-ILC group (median = 5%) (p-value < 0.05). This disparity could influence the higher methylation percentage observed in the IBC-NST group across all studied regions (Fig. 1; Supplementary Fig. 1). Additionally, UDH displayed low methylation percentages (median 11%, 6%, and 16% methylation in N-shore, Island, and S-shore regions, respectively) (Fig. 1a), alongside a considerably low TILs percentage ranging from 0 to 10% (median = 4%), which could influence the low methylation levels.

a, b, c Correlation between CDH1 methylation frequency and percentage of TILs across N-shore, island, and S-shore regions. d Hematoxylin-eosin staining displaying the percentage of TILs (40%) in non-mutated ILC case number 11

In conclusion, our findings, facilitated by high-resolution quantitative detection methodology, indicated that the frequency and extent of CDH1 gene methylation in ILCs are not higher than those observed in IBC-NSTs. This result held true irrespective of the presence or absence of CDH1 mutations, thereby challenging the notion of CDH1 methylation as a pervasive mechanism for CDH1 gene inactivation. Moreover, our analysis suggested the potential impact of TIL abundance on CDH1 methylation analysis. Importantly, the conspicuous loss of E-cadherin in the non-mutated ILC subgroup might be driven by mechanisms beyond DNA methylation. The intricate interplay of additional genetic and epigenetic mechanisms, along with non-genetic determinants such as cellular signaling pathways, environmental factors, and cellular context, holds promise in shedding light on alternative mechanisms to the loss of CDH1 for the lobular phenotype [18].