A DNA methylation-based liquid biopsy for triple-negative breast cancer

Abstract

Here, we present a next-generation sequencing (NGS) methylation-based blood test called methylation DETEction of Circulating Tumour DNA (mDETECT) designed for the optimal detection and monitoring of metastatic triple-negative breast cancer (TNBC). Based on a highly multiplexed targeted sequencing approach, this assay incorporates features that offer superior performance and included 53 amplicons from 47 regions. Analysis of a previously characterised cohort of women with metastatic TNBC with limited quantities of plasma (<2 ml) produced an AUC of 0.92 for detection of a tumour with a sensitivity of 76% for a specificity of 100%. mDETECT_TNBC was quantitative and showed superior performance to an NGS TP53 mutation-based test carried out on the same patients and to the conventional CA15-3 biomarker. mDETECT also functioned well in serum samples from metastatic TNBC patients where it produced an AUC of 0.97 for detection of a tumour with a sensitivity of 93% for a specificity of 100%. An assay for BRCA1 promoter methylation was also incorporated into the mDETECT assay and functioned well but its clinical significance is currently unclear. Clonal Hematopoiesis of Indeterminate Potential was investigated as a source of background in control subjects but was not seen to be significant, though a link to adiposity may be relevant. The mDETECT_TNBC assay is a liquid biopsy able to quantitatively detect all TNBC cancers and has the potential to improve the management of patients with this disease.

Fig. 1. Overview of the mDETECT assay.

a mDETECT relies on the blood-based detection of large numbers of differentially methylated regions that are specific to tumours represented by the green stripes on the grey chromosome. b To identify these regions with a high density of tumour-specific methylation, those CpG dinucleotides found to show high levels of methylation in the tumour populations but low levels in the normal tissues were identified as being differentially hypermethylated. β-values represent methylation levels at a given probe (CpG dinucleotide). c Bisulfite conversion of the DNA samples of these differentially methylated regions allows the multiple CpG residues in each region to be assessed and subsequently allows for differentiation between tumour and normal samples. d Methylation biased primers, as indicated by the arrows, are designed to preferentially but not exclusively amplify methylated DNA. This ensures that small amounts of methylated tumour DNA are amplified even in a high background of normal DNA. These PCR “probes” are designed to amplify products less than 125 bp (<125 bp) due to the highly fragmented nature of ctDNA (e). PCR is carried out by various types of multiplexing. f Next-generation sequencing (NGS) is carried out to identify multiple CpG dinucleotides that were methylated and to allow for high-level multiplexing.

Fig. 2. Methylation in the promoter region of the gene CDKL2 in tumour and normal samples.

Identification of differentially methylated target regions. Illumina 450 K methylarray probe beta values from various TCGA databases in the CDKL2 promoter were averaged (average beta value) for normal prostate (normal prostate), colon (normal colon), lung (normal lung), and breast (normal breast) tissues, and for triple- negative breast cancer (TNBC) from selected TCGA datasets. The methylation values for the each of these samples were plotted relative to genomic position on the x-axis (hg18 nucleotide position on Chromosome 4) to determine the extent of differential methylation in surrounding areas. The first exon of the CDKL2 gene is shown (blue box) along with the identified CpG island (CpG Island in green). The location of the CDKL2 probe is shown (light-blue box) and the expanded region shows the location of CpG residues in relation to the PCR primers (arrows) and product.

Fig. 3. Singleplex sequencing results for the probe NFIC-A in cell lines.

Shown here are the BiQ Analyser generated heatmaps of the NGS reads for the probe NFIC-A as analysed in numerous cell lines. Each column represents one of the 5 CpG dinucleotide in the amplified region while each row indicates an individual sequencing read. Blue fill indicates that the CpG dinucleotide was not methylated in a particular read while red fill indicates that the CpG dinucleotide was methylated in a particular read. The number of reads and the fraction that are methylated in each probe is indicated below each map. As expected, there is low methylation in the immortalised mammary epithelial cell lines 184-hTERT and MCF-10A, as well as the normal buffy coat DNA. Two of three TNBC cell lines (MDA-MB-231, MDA-MB-436, and HCC1937) were methylated demonstrating heterogeneity between individual cancers. Furthermore, MCF7 and SKBR3, an ER + and HER2 + cell line show methylation, indicating that the probe is not exclusively methylated in TNBC but also in other subtypes of cancer. In general, methylation was an all or none phenomena.

Fig. 4. mDETECT analysis of 15 TNBC tumours.

The shading indicates the fraction of reads exhibiting methylation at all CpG residues in the region between primers compared to all reads, for the 86 probes in each of the 15 TNBC tumours analysed, as well as PBMC DNA from a normal individual. Each column represents an individual probe while each row represents a different tumour sample. Between 18 and 65 probes had a methylation fraction above 0.1 for each sample (average 46) while the PBMC sample had 2 positive probes.

Fig. 5. Multi-Singleplex approach.

A Samples were linear amplified using all 68 negative strand primers for 50 cycles. After purification the sample was split into 68 separate PCR reactions with each probe. The products were then pooled and barcoded. B A heatmap of the fraction methylated DNA for each probe. Two TNBC cell lines (MDA-MB-436 and 231), normal PBMC DNA and PBMC synthetically methylated (mePBMC) were analysed. For each probe the number of reads methylated at all CpGs positions between the primers was divided by the total number of reads to give the fraction of methylation for each with the shading representing this fraction from 0 to 1.

Fig. 6. Test cohort.

Twelve plasma samples from Cohort 03 patients and 20 plasma samples from healthy age-matched volunteers (QBC 002-049) were analysed for mDETECT as described in Fig. 5 and a heatmap of the fraction of methylated DNA for each probe is shown. (Top) The original 68 probes set is shown. (Bottem). Probes were removed based on the presence of 3 or more positive probes in the controls to leave 54 probes.

Fig. 7. mDETECT analysis of Validation Cohort 03 samples and new control plasma DNA.

The DNA from 17 patients not included in the Test Cohort, as well as 31 new plasma samples from healthy age-matched volunteers (QBC 16-109) were analysed using mDETECT for the 54 optimal probes as described in Fig. 6. A heatmap of the fraction of methylated DNA for each probe is shown as described in Fig. 5.

Fig. 8. mDETECT versus ctDNA levels determined by TP53 mutation sequencing.

A next-generation sequencing TP53 mutation-based test was previously used to quantitate ctDNA levels in the plasma from these patients and is expressed on a log scale as ng of ctDNA per ml of plasma (x-axis). The number of mDETECT probes that were positive for methylation for each sample is plotted (right y- axis). For samples that were negative for the TP53 assay (Diamonds) the mDETECT levels are plotted on the left y-axis as zero cannot be plotted on a log scale. Regression analysis shows that the relationship between mDETECT and TP53-based levels is reasonably well correlated (R² = 0.54).

Fig. 9. Analysis of serum samples from metastatic TNBC patients.

Thirty serum samples from metastatic TNBC patients (CRB1–30) and 30 new serum samples from healthy age-matched volunteers (QBC020–106) were analysed using mDETECT for the 54 optimal probes as described in Fig. 6. A heatmap of the fraction of methylated DNA for each probe is shown as described in Fig. 5.

Fig. 10. Comparison of mDETECT and CA15-3 detection.

The levels of mDETECT and CA15-3 were compared for patients in both Cohort 03 and CRB. The normal range of CA15-3 is up to 30 units per ml (red-dashed line) while the threshold for mDETECT is 2.5 positive probes (Solid green line). In the Cohort 03, eight patients that were negative for CA15-3 were positive as determined by mDETECT (orange circles). Similarly, five patients in the CRB cohort that were negative by CA15-3 were positive by mDETECT (orange circles). The mDETECT and CA15-3 levels were somewhat correlated with an R²-value of 0.225 for Cohort 03 and 0.238 for the CRB cohort (blue dashed lines).