Analysis of Intrinsic Breast Cancer Subtypes: The Clinical Utility of Epigenetic Biomarkers and TP53 Mutation Status in Triple-Negative Cases

This study aimed at analyzing the DNA methylation pattern and TP53 mutation status of intrinsic breast cancer (BC) subtypes for improved characterization and survival prediction. DNA methylation of 17 genes was tested by methylation-specific PCR in 116 non-familial BRCA mutation-negative BC and 29 control noncancerous cases. At least one gene methylation was detected in all BC specimens and a 10-gene panel statistically significantly separated tumors from noncancerous breast tissues. Methylation of FILIP1L and MT1E was predominant in triple-negative (TN) BC, while other BC subtypes were characterized by RASSF1, PRKCB, MT1G, APC, and RUNX3 hypermethylation. TP53 mutation (TP53-mut) was found in 38% of sequenced samples and mainly affected TN BC cases (87%). Cox analysis revealed that TN status, age at diagnosis, and RUNX3 methylation are independent prognostic factors for overall survival (OS) in BC. The combinations of methylated biomarkers, RUNX3 with MT1E or FILIP1L, were also predictive for shorter OS, whereas methylated FILIP1L was predictive of a poor outcome in the TP53-mut subgroup. Therefore, DNA methylation patterns of specific genes significantly separate BC from noncancerous breast tissues and distinguishes TN cases from non-TN BC, whereas the combination of two-to-three epigenetic biomarkers can be an informative tool for BC outcome predictions.


Introduction
In 2020, breast cancer (BC) was the leading cause of women's cancer worldwide, while the mortality from BC was in fifth place [1]. Triple-negative (TN) BC accounts for approximately 10-15% of all diagnosed BC and, in comparison to estrogen-and progesterone receptor-positive (i.e., hormone receptor-positive, ER+ and PR+, respectively) BC cases have the most aggressive course of the disease and the worst prognosis [2]. In contrast to the hormonal or HER2 (human epidermal growth factor receptor) positive BC, the specific molecular pathophysiology of TN BC remains poorly understood, resulting in a lack of efficient target therapies [3].
Nowadays, immunohistochemistry is mainly used for BC subtype classification but the addition of genetic and epigenetic biomarkers could increase the sensitivity and specificity of disease diagnosis, prognosis, and prediction of treatment outcome. Because BC is a multiform disease and there is no particular cause or gene mutation to lead the breast cell to cancer, it is important to find an informative biomarker system to identify and predict the disease development and response to treatment.

TP53 Mutation Spectrum
In total, 86 tumors were analyzed for TP53 mutations (TP53-mut), out of which sequence alterations were identified in 33 cases (38%,), while 53 BC had wild-type TP53 (62%, TP53-wt). After 84 samples were analyzed by using the single-strand conformation polymorphism (SSCP) method, 29 TP53-mut cases were detected and further validated by Sanger sequencing (SS). Thirty-eight samples were selected for more detailed analysis using next-generation sequencing (NGS), out of which two samples were not previously analyzed either by SSCP or by SS (Figure 2A,B). Three samples, previously determined as negative by SSCP and SS methods were identified as TP53-mut positive by NGS analysis.

TP53 Mutation Spectrum
In total, 86 tumors were analyzed for TP53 mutations (TP53-mut), out of which sequence alterations were identified in 33 cases (38%,), while 53 BC had wild-type TP53 (62%, TP53-wt). After 84 samples were analyzed by using the single-strand conformation polymorphism (SSCP) method, 29 TP53-mut cases were detected and further validated by Sanger sequencing (SS). Thirty-eight samples were selected for more detailed analysis using next-generation sequencing (NGS), out of which two samples were not previously analyzed either by SSCP or by SS (Figure 2A,B). Three samples, previously determined as negative by SSCP and SS methods were identified as TP53-mut positive by NGS analysis. Out of 33 TP53-mut BC cases, pathogenic TP53 sequence alterations were detected in 27 BC, 78% (21/27) of which occurred in the DNA binding domain, and 22% (6/27) in introns. According to the mutation type, about a half (52%, 14/27) of pathogenic mutations were missense, 19% (5/27) splice, 11% (3/27) frameshift, 11% (3/27) nonsense, and 4% (1/27) intronic. Out of pathogenic mutations: 26% were AT:GC; 19% were GC:AT at CpG sites, and 15% were not at CpG sites; 11% were GC:TA. Deletions in the studied TP53 gene region were quite common and accounted for 15% of all alterations (range 1-23 nt). The largest, 23-nucleotide (nt) deletion g.7578546_7578568del (23 nt deleted) was found in the fourth intron by the NGS method and affected the splicing site. The detailed TP53 mutation data are provided in Supplementary Table S2.

Prediction of Overall Survival
Univariate and multivariate Cox proportional hazards regression analyses were performed to analyze the associations between the biomarkers and the overall survival (OS) of BC cases. In univariate analysis, older age and TN subtype were significantly associated with shorter OS (p < 0.05). Out of the analyzed genetic biomarkers, only the hypermethylation of FILIPL1 tended to be associated with OS (HR = 3.3, 95% CI 0.9-12.0, p = 0.067; Table 2; presented are only the genes demonstrating HR > 1.0).  Table 2) are independent prognostic factors for OS.
In the Kaplan-Meier survival analysis, various combinations of biomarkers were predictive for the outcome: RUNX3 combinations with MT1E or FILIP1L (p = 0.045 and p = 0.039, respectively; Figure 4A,B) or all three biomarkers also significantly predicted the poor outcome (p = 0.031; Figure 4C). In addition, FILIP1L methylation was predictive of poor outcomes in the TP53-mut subgroup (p = 0.045; Figure 4D).

Discussion
The heterogeneity of BC is reflected by gene expression patterns known as intrinsic BC subtypes, which nowadays are classified according to IHC biomarkers; however, these subtypes further vary in the abundance of genetic mutations and epigenetic alterations. The luminal and HER2 receptor-expressing BC can be treated using modern targeted therapy, unlike the TN subtype, which is the most heterogeneous group of BC lacking efficient diagnostic and treatment modalities [9]. Early BC diagnostics, especially the TN subtype, could improve BC survival rates; therefore, the traditional IHC-based diagnostic methods may benefit from supplementing by genetic and epigenetic biomarkers. DNA methylation changes of selected genes were identified in all BC, and the methylation pattern of 10 out of 17 tested genes statistically significantly separated tumors from noncancerous breast tissues. DNA methylation in some of the genes was predominant in less aggressive G1 (RASSF1 and ADAMTS12) or Ki-67 negative (PRKCB and APC) tumors, indicating an early occurrence of epigenetic events. Moreover, specific DNA methylation patterns were characteristic of intrinsic BC subtypes in our and other studies [10,11]. For instance, luminal subtypes harbor subtype-specific methylation biomarkers like RASSF1, GSTP1, APC, ADAMTS12, and PRKCB [10,12]. In our study, the biomarker set of PRKCB, RASSF1, and APC was found to be hypermethylated in a majority of BC samples with the

Discussion
The heterogeneity of BC is reflected by gene expression patterns known as intrinsic BC subtypes, which nowadays are classified according to IHC biomarkers; however, these subtypes further vary in the abundance of genetic mutations and epigenetic alterations. The luminal and HER2 receptor-expressing BC can be treated using modern targeted therapy, unlike the TN subtype, which is the most heterogeneous group of BC lacking efficient diagnostic and treatment modalities [9]. Early BC diagnostics, especially the TN subtype, could improve BC survival rates; therefore, the traditional IHC-based diagnostic methods may benefit from supplementing by genetic and epigenetic biomarkers. DNA methylation changes of selected genes were identified in all BC, and the methylation pattern of 10 out of 17 tested genes statistically significantly separated tumors from noncancerous breast tissues. DNA methylation in some of the genes was predominant in less aggressive G1 (RASSF1 and ADAMTS12) or Ki-67 negative (PRKCB and APC) tumors, indicating an early occurrence of epigenetic events. Moreover, specific DNA methylation patterns were characteristic of intrinsic BC subtypes in our and other studies [10,11]. For instance, luminal subtypes harbor subtype-specific methylation biomarkers like RASSF1, GSTP1, APC, ADAMTS12, and PRKCB [10,12]. In our study, the biomarker set of PRKCB, RASSF1, and APC was found to be hypermethylated in a majority of BC samples with the highest specificity to BC and was specific to hormonal and HER2+ BC subtypes as well, significantly distinguishing from the TN BC subtype. On the contrary, studies show that TN tumors have fewer DNA methylation changes than non-TN [13] but TN BC is the most heterogenous BC intrinsic subtype, molecularly subcategorized into smaller subgroups [14]. TN BC is diverse and difficult to study and therefore there is only a handful of studies assigning specific biomarkers to TN BC [15,16]. In our study, the increased DNA methylation rate of MT1E and FILIP1L significantly distinguished the TN BC subtype from luminal and HER2+ BC subtypes with twice higher hypermethylation frequency, and both showed an association with Ki-67 expression. Differences in DNA methylation patterns between and even within BC intrinsic subtypes demonstrate high biological variability of these tumors and show the need for further subclassification of BC for cost-effective treatment personalization.
Because of the high heterogeneity and limited treatment options, the survival of TN BC is well known to be the lowest; however, the complete picture of molecular pathways affected in TN BC remains unclear and targets for efficient treatment are yet to be found. Some of the studies reveal the significance of epigenetic factors in TN BC pathogenesis [17]. It has been demonstrated [18] that the hypomethylated profile TN has a better survival than hypermethylated, however, survival of TN BC cases of the medium methylated cluster was shown to be the worst. Genes hypermethylated in TN BC are involved in various cellular pathways and could be used to predict survival outcomes and response to treatment [15,16]. In our study, despite relatively low DNA methylation frequencies detected in TN BC, the hypermethylation of RUNX3, MT1E, and FILIP1L was highly specific to this subtype and associated with a shorter OS when analyzed alone (RUNX3) or in combinations (RUNX3, FILIP1L, and MT1E). RUNX3 encodes a tumor suppressor which regulates cell growth, survival, differentiation, angiogenesis, and invasion [19]; FILIP1L is a protein that inhibits metastases and chemoresistance [20]; MT1E is a cytoskeleton-modifying protein, involved in cell migration and invasion [21]. All these newly identified biomarkers of TN BC demonstrate the potential to accompany classic diagnostic methods and become a part of companion diagnostics for novel therapies, including combined treatment schemes that involve epigenetic drugs.
The TN BC subtype differs from luminal and HER2+ subtypes in a genetic and epigenetic manner. TP53 mutation is found in approximately 80% of TN BC cases [4,22] and is associated with poor prognosis [6]. Similarly, in our study, TP53 mutation predominantly occurred in TN BC (87%) but was rarely observed in other BC subtypes. In addition, our research showed that in TN BC, more than two-thirds of TP53 mutations occurred among poorly differentiated tumors and were associated with higher Ki-67 expression. In our study, more than half of TP53 alterations were missense mutations, which, according to Sousse and colleagues [23], result in a stable p53 protein that lacks its specific DNA-binding activity, accumulates in the cellular nucleus where, by interacting with oncogenes, causes cell transformations [24]. Although TP53 mutations are predominant in the TN subtype, they can also be associated with ER+ patients' survival, affecting their response to endocrine therapy [25]. Taken together, TP53 is an important player in breast carcinogenesis and a significant target for specific treatment development.
Despite this study being performed by investigating both genetic and epigenetic alterations of BC, several shortcomings can be discussed. The study cohort included all BC subtypes and the TN BC part comprised only 14% of all cases; therefore, further analysis of TN BC-specific biomarkers should be extended to a larger independent TN BC cohort. As this study was started some years ago, more extensive use of the NGS method now is possible and looks more informative for TP53 mutations analysis. While different studies show that TP53 mutations could be associated with a poor, good, or neutral outcome, mainly, TP53-mut tumors are associated with worse OS [26]; however, in the current study, a TP53-mut association with worse OS was not demonstrated. Additionally, follow-up data were missing for some patients, which could have affected the OS statistics.

Patients and Samples
In total, 116 BC patients and 29 control cases with fibroadenoma (all white Caucasian race females) treated at the National Cancer Institute of Lithuania enrolled in the study in [2007][2008][2009]. The Bioethics Committee approved the study (2007-08-03 No. 33) and informed consent was obtained from every case before entering the study. All investigated BC cases were BRCA-negative non-familial cases. The mean age of BC patients was 57 years (range 27-84 yrs.), and 42 years for controls (range 20-62 yrs.); p < 0.05. All patients were diagnosed with invasive BC of early stages T1 (n = 63) and T2 (n = 53). The analyzed BC types were ductal (n = 101), lobular (n = 13) and apocrine (n = 2) breast carcinomas. The intrinsic subtypes of BC were identified based on the IHC status of pathology biomarkers: estrogen (ER) and progesterone (PR) receptors, human epidermal growth factor receptor-2 (HER2), and marker of tumor proliferation (Ki-67). Ki-67 cut-off in our study was 15%; therefore, >15% was considered as Ki-67 positive and, on the contrary, <15% of Ki-67 was considered as Ki-67 negative. In addition, 47% were of luminal A (LA, n = 55), 21% of luminal B (LB, n = 24), 18% were HER2+ (n = 21, out of which 16 and 5 cases were LBHER2 and HER2, respectively), and 14% were triple-negative (TN, n = 16) BC cases. Follow-up data were available for 78 of 116 (67%) BC cases and the average follow-up time was 91 (range 3-113) months. Out of 78 patients whose outcomes were known, 21 cases were deceased, 1 relapsed, and 56 were in remission. Detailed information on demographic and clinical-pathological variables according to intrinsic BC subtypes is provided in Table 3. Table 3. Demographic and clinical-pathological characteristics of breast cancer (BC) patients distributed by breast cancer (BC) subtypes.

DNA Extraction
DNA was extracted by the standard phenol-chloroform purification and ethanol precipitation method and using ZR Viral DNA/RNA Kit™ (Zymo Research, Irvine, CA, USA) from fresh-frozen and ground tumor tissue specimens (n = 79) after digestion with proteinase K and from formalin-fixed paraffin-embedded (FFPE) tissues (n = 45) after the deparaffinization. DNA concentration and quality parameters were evaluated spectropho-tometrically by using NanoDrop TM 2000 (Thermo Scientific, Thermo Fisher Scientific (TFS), Waltham, MA, USA).

DNA Methylation Assay
Isolated DNA (400 ng) was first modified with sodium bisulfite using EZ DNA Methy-lation™ Kit (Zymo Research, Irvine, CA, USA) according to the manufacturer's recommendations. For DNA methylation assessment, the pairs of primers specific to methylated (M) and unmethylated (U) sequences within the 5'region of P14, P16, MGMT, RARB, RASSF1, DAPK1, GSTP1, ESR1 (two 5 regions of the ESR1 gene, one in promoter region and one intragenic sequence, were included into this study and marked as ESR1-1 and ESR1-4, respectively), PRKCB, MT1E, MT1F, MT1G, APC, ADAMTS12, and RUNX3. Genes were designed using Methyl Primer Express v1.0 software (Applied Biosystems (ABI), TFS) or selected based on BC specificity and diagnostic and/or outcome prediction capabilities from our previous studies [27][28][29] (see Supplementary Table S3). Methylation-specific PCR (MSP) mix of the final volume of 25 µL contained 10 ng of bisulfite-modified DNA template, PCR buffer, 1.6 mM of each dNTP, 2.5 mM of MgCl2, 1 µM of each primer, enhancer, and 0.5 U of Gold polymerase (ABI, TFS). PCR was performed in a thermocycler at the conditions provided in Supplementary Table S4. Each PCR run was performed by using two kinds of DNA methylation controls, methylated and unmethylated; in both cases, leukocyte DNA from healthy donors was used and, respectively, treated or untreated with CpG Methylase SssI (New England BioLabs) before the bisulfite modification. In addition, a non-template control (NTC), a reaction with water instead of a DNA template, was performed alongside each PCR run. Reaction products were analyzed electrophoretically in 3% agarose gel, stained with ethidium bromide, and visualized under UV illumination (GelDoc-It ® 310 Imaging system, Fisher Scientific, TFS) using visualization and analysis software VisionWorks ® LS (UVP, Upland, CA, USA).
Sanger sequencing (SS) was used to confirm mutations detected by SSCP. Analyzed 5-9th TP53 exons of SSCP-positive samples were first amplified by PCR, consisting of 200 ng DNA templates and the same reaction components as were used for SSCP analysis, which is described above. The sequencing reaction (20 µL), contained 5 µL PCR product, BigDye Terminator v3.1 Ready Reaction mix (ABI, TFS), 5× Sequencing Buffer, sense and antisense primers, and H 2 O. Sequencing reactions were carried out on ABI Prism 3130 ® Genetic Analyzer and analyzed with SeqScape TM software (ABI, TFS). Results were compared with reference TP53 sequence from GenBank ® database. SSCP and SS methods were adapted from Holmila and Husgafvel-Pursiainen [32].
Next-generation sequencing (NGS) was performed using GS Junior 454 Sequencer (Roche Diagnostics by 454 Life science corp. Branford, CT, USA). A healthy female leukocyte DNA was used as a reference. All fragments were sequenced in both directions. DNA was amplified in 24 µL reaction mix, which contained 1x Phusion HF buffer, 0.2 mM of each dNTP, 0.3 µM of each primer, 0.5 U/µL HiFi Phusion polymerase, and 25 ng of DNA template (see details in Supplementary Table S5) Amplicons were purified with AMPure XP magnetic beads (TFS). Reaction products were fluorometrically analyzed using the Quant-It TM PicoGreen dsDNA Assay kit (TFS) and the QuantiFluor system ® (Promega, Madison, WI, USA). Standard curve value was not less than R2 > 0.98. Emulsion PCR was performed using the emPCR Kit according to manufacturer's instructions. Amplicons were mixed with capture beads using 10 uL of DNA library (at 1.33 molecules per bead concentration) for each forward and reverse strand amplification by emPCR and collected with the GS Junior Titanium emPCR Oil and Breaking Kit. For the sequencing procedure, The GS Titanium Sequencing Kit and GS Junior Titanium series protocol were followed (Roche). Sequencing data analysis was performed using GS Amplicon Variant Analyzer (AVA) (Roche). TP53 sequence NC_000017.10 (NCBI37/hg19; Chr17:7571720 . . . 7590868) was used as the reference sequence (corresponding transcript and protein IDs are ENST00000269305.4 and P04637, respectively).

Statistical Analysis
The two-sided Fisher's exact test was used for analysis of gene methylation status and other categorical clinical variables (for patients' age, two groups of < 50 and > 50 yrs. were compared). Mann-Whitney testing was applied to continuous data. Cox proportional hazards regression (with backward variable selection) and Kaplan-Meier analysis (with multiple testing correction (Bonferroni) and additionally corrected p-values) were used to assess the associations between clinical parameters and survival. Calculations were performed by using GraphPad Prism 8.01 (GraphPad Software, Inc., San Diego, CA, USA) and MedCalc 12.7.0.0 (MedCalc Software Ltd., Ostend, Belgium). In all cases, p ≤ 0.05 was considered statistically significant.

Conclusions
DNA methylation of RASSF1, PRKCB, APC, and RUNX3 significantly separates BC from noncancerous specimens and also is more frequently found in non-TN BC cases, while higher methylation frequency of MT1E and FILIP1L is associated with TN BC. The combination of two-to-three epigenetic biomarkers (FILIP1L, RUNX3, and MT1E) is an informative tool for BC-outcome predictions. Further investigations of these DNA methylation biomarkers are needed, especially for improved characterization of the TN BC subtype.