Tracking breast cancer progression using Methylscape

Abstract

Cancer progression is driven by epigenetic reprogramming, where promoter hypermethylation of tumour-suppressor genes and global hypomethylation reshape gene regulation and cellular phenotypes, promoting oncogenesis and disease advancement. We previously introduced the Methylscape, a cancer-specific DNA methylation landscape characterized by clustered promoter hypermethylation and gene body hypomethylation that enhances DNA’s physical affinity for gold surfaces. Here, we demonstrate that Methylscape can be leveraged to monitor cancer progression. In a TGF-β-induced breast cancer epithelial-mesenchymal transition (EMT) model, we observe increased Methylscape enrichment of mesenchymal-state DNA, indicating that this method can sensitively detect subtle epigenetic remodelling linked to tumour progression. Using a gold-based DNA desorption enrichment strategy coupled with methylation sequencing and qPCR, we show that hypermethylated regions are preferentially enriched on gold surface. Finally, we developed a low-cost, disposable screen-printed electrode platform for stage-specific breast cancer monitoring. Together, these findings establish Methylscape as a promising biophysical biomarker for non-invasive, real-time monitoring of cancer progression, advancing its potential for clinical translation.

INTRODUCTION

Cancer progression is driven by epigenetic dysregulation, with dynamic changes in DNA methylation emerging as a key mechanism that modulates gene expression and cellular behaviour (1). Aberrant DNA methylation is characterized by both hypermethylation and hypomethylation, causing disruptions of cellular homeostasis and facilitating oncogenic transformation, and ultimately enabling cancer cells to evade normal regulatory mechanisms and drive uncontrolled growth (2, 3). Specifically, hypermethylation tends to cluster at CpG islands within promoter regions, silencing tumour suppressor genes and facilitating oncogenesis. In contrast, global hypomethylation predominantly occurs in intergenic regions, contributing to genomic instability (2–4). Together, this distinct methylation landscape continues to grow during cancer progression that substantially differs from the normal methylome (5, 6).

Current methods for monitoring cancer progression include imaging techniques like computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET) (7–10). While these methods provide anatomical insights, they are limited by their inability to detect molecular changes and early-stage progression. Methylation-based methods, including whole-genome bisulphite sequencing (WGBS), reduced representation bisulphite sequencing (RRBS), and targeted methylation assays, are valuable for capturing cancer-specific epigenetic changes (11, 12). However, these approaches frequently require intricate workflows, substantial expenses, and extensive sample preparation, limiting their clinical utility (13, 14). Overcoming these challenges is essential for developing simplified, cost-effective, and robust tools capable of reliably monitoring cancer progression at the molecular level, ultimately leading to improved patient outcomes.

Previously, we introduced the concept of Methylscape, a phenomenon describing the enrichment of hypermethylated DNA regions through specific methylation changes, including promoter hypermethylation and gene body hypomethylation (15). In our earlier work, we demonstrated that DNA derived from a cancerous context exhibited significantly higher physical adsorption on a gold surface compared to DNA from non-cancerous samples (15). This was hypothesized to result from the selective exposure of hypermethylated regions, driven by global hypomethylation, which reduces the overall hydrophobicity of the DNA polymer and increases the interaction of methylated regions with hydrophobic surfaces. Exploring this unique surface affinity property of methylated DNA, we developed a simple interfacial biosensing assay capable of electrochemically quantifying DNA adsorption levels and identifying cancer specific methylated DNA.

In this study, we leveraged the unique ability of Methylscape to enrich hypermethylated regions in cancer to demonstrate its potential for tracking epigenetic remodelling during cancer progression (Fig. 1). We employed an epithelial-to-mesenchymal transition (EMT) model, a critical process in cancer progression characterised by dynamic DNA methylation changes that promote mesenchymal traits such as motility, invasiveness, and resistance to apoptosis (16, 17). To gain deeper mechanistic insights, a desorption-based approach was developed to extract DNA from the gold surface, followed by qPCR and methylation sequencing. This approach enabled us to determine which specific DNA regions, such as hypermethylated promoters or hypomethylated gene bodies, preferentially interact with the gold surface during EMT progression. This selective isolation of cancer-specific methylated DNA offers a targeted and efficient approach for analysing epigenetic changes in cancer progression. We also evaluated our method on clinical breast cancer samples and successfully differentiated disease stages using a cost-effective, disposable gold-printed electrode system. The clinical study demonstrates that Methylscape has the potential to serve as a versatile tool for enriching cancer-specific methylation signals, enhancing our understanding of cancer-associated epigenetic changes, and supporting the development of robust, clinically applicable assays for monitoring cancer progression.

Fig. 1. Methylscape analysis for cancer progression and clinical application.

(A) Adsorption: monitoring the EMT process using Methylscape. (B) Desorption: DNA desorption from the gold surface, followed by qPCR and WGBS for comprehensive methylation analysis. (C) Clinical application: employing disposable gold screen-printed electrodes to distinguish between different stages of breast cancer. Created in BioRender. Zhang, Z. (2026) https://BioRender.com/0x9b4ok.

RESULTS AND DISCUSSION

Application of Methylscape for monitoring cancer progression via the EMT model

EMT is a critical process in cancer progression that plays a key role in initiating metastasis. During EMT, epithelial cancer cells undergo significant phenotypic and morphological changes, transforming into mesenchymal-like cells. This transition enables the cells to acquire enhanced migratory and invasive properties, leading to the release of circulating tumour cells (CTCs) into the bloodstream, which can subsequently result in cancer metastasis (16, 18, 19). During metastasis, the EMT process enables CTCs to evade immune surveillance, contributing to the development of drug resistance. Interestingly, DNA methylation-mediated dynamic epigenetic reprogramming is a key mechanism that initiates the EMT process (17, 20–23). For instance, during the EMT process, the methylation patterns of promoter regions in key epithelial genes (e.g., CDH1) and mesenchymal genes (e.g., VIM) undergo dynamic changes in response to EMT-inducing stimuli such as TGF-β, hypoxia, and growth factors present within the tumour microenvironment (24). Moreover, recent reports have found that DNA methylation changes underlying EMT are among the most common mechanism leading to the development of therapy resistance (17). Since Methylscape is designed to identify changes in the DNA methylation landscape in cancer, we decided to use a synthetic EMT model to determine whether our method can accurately detect cancer EMT events and trace cancer progression in a breast cancer model.

To ensure a well-defined and reproducible EMT model, TGF-β-induced EMT was used in this study, as it represents the canonical signalling pathway governing EMT transcriptional networks (25–28). We treated two breast cancer cell lines with TGF-β to create a synthetic EMT-induced model. The epithelial breast cancer cell line (MCF7) and the mesenchymal breast cancer cell line (MDA-MB-231) were cultured for 12 days: 6 days with TGF-β treatment and 6 days post-treatment. Cells were collected at six time points: Day 0 (pre-treatment), Day 3 and Day 6 (during treatment), and Day 9 and Day 12 (post-treatment). The EMT transition was confirmed by flow cytometry (Fig. 2A and fig. S1A), showing a downregulation of E-cadherin and an upregulation of N-cadherin on Day 3 and Day 6, indicating a shift to a mesenchymal phenotype. After the treatment was withdrawn, both marker expressions returned to a more epithelial-like phenotype, suggesting that the EMT process was reversed after the cessation of treatment.

Fig. 2. Methylscape analysis on drug induced EMT model (MCF7) for investigating cancer progression.

(A) Flow cytometry analysis of E-cadherin and N-cadherin in epithelial breast cancer cell line MCF7 among TGF-β treatment pre-treatment (Day 0), during treatment (Day 3 and Day 6), post-treatment (Day 9 and Day 12). (B) The number of differentially expressed genes (DEGs) (FDR <0.05 and |log2fold change| ≥ 1) in the epithelial cell model (MCF7) during- and post- treatment were compared to the pre-treated samples. (C) Enrichment analysis of differentially expressed transcripts during- and post TGF-β treatment on the epithelial cell line (MCF7). The number of enriched transcripts is shown next to each plot while the y axis displays the proportion of differentially expressed transcripts that are enriched in each pathway when compared with the hallmark gene sets. The number of enriched transcripts is shown next to each plot while the y axis displays the proportion of differentially expressed transcripts that are enriched in each pathway when compared with the hallmark gene sets. (D) Methylscape analysis for adsorbing MCF7 DNA on gold surface among the timepoints. (E) Relative global DNA methylation at Day 0 and Day 6.

To evaluate EMT-associated transcript changes, we conducted RNA sequencing analysis on MCF7 cells at pre-treatment (Day 0), during treatment (Day 6), and post-treatment (Day 12), using EdgeR to identify differentially expressed genes (DEGs). The DEGs were visualized in a volcano plot, with significance thresholds set at a P value ≤0.05 and a log10 fold change of 1. While several EMT-related genes, including CDH1, EPCAM, VIM, and ZEB1, were expressed during TGF-β treatment, only FN1 and VIM were significantly differentially expressed, corresponding to the mesenchymal markers fibronectin and vimentin, respectively (Fig. 2B). After treatment withdrawal, FN1 expression decreased, and CDH1, which was downregulated during treatment, returned to baseline levels, indicating a reversal of the EMT process (Fig. 2C). To further investigate, we analysed the mesenchymal cell line MDA-MB-231 under similar conditions, where TGF-β was used to enhance the EMT transition. Consistent with MCF7, VIM and FN1 were differentially expressed during and after treatment, while CDH1 was downregulated six days post-treatment (fig. S1, B and C). This mirrored the pattern observed in MCF7 cells, supporting the induction of EMT by TGF-β treatment.

To understand the potential functions of the differentially expressed EMT-associated transcripts, we performed a hypergeometric enrichment analysis on genes with significant fold changes during and after TGF-β treatment. In MCF7 cells, the analysis revealed enrichment in the TGF-β signalling pathway during treatment, consistent with the induction of a mesenchymal state (Fig. 2C). However, this enrichment disappeared after treatment was withdrawn, aligning with the observed phenotypic reversal (Fig. 2C). Similarly, in MDA-MB-231 cells, differentially expressed transcripts were enriched in the TGF-β signalling and EMT pathways during treatment, with this enrichment dissipating post-treatment (fig. S1C). These findings highlight the role of TGF-β in driving EMT, as evidenced by the transient activation of these pathways during treatment.

Lastly, after validation of the drug-induced EMT model, we aim to utilize Methylscape for monitoring the EMT process. We optimized the DNA adsorption conditions on the gold surface such as DNA purity, concentration, and solvation parameters. Nucleic acids were extracted using phenol-chloroform extraction. Because Methylscape relies on the direct adsorption of DNA onto a gold surface, the presence of other biomolecules or contaminants co-isolated during nucleic acid extraction could potentially influence assay specificity. To address this concern, we performed spiking experiments with possible contaminants such as ethanol, GlycoBlue, proteins, and RNA. As shown in the fig. S2, these co-contaminants exhibited little to no effect on the Methylscape signal. Additionally, the reproducibility demonstrated in our previous studies (15, 29), as well as in independent reports by other researchers (30, 31), highlights the robustness of the assay. Therefore, we believe that even trace amounts of co-contaminants, if present, do not affect our data quality or the conclusions drawn from this study.

We next evaluated four DNA concentrations (1 ng/μl, 2 ng/μl, 5 ng/μl and 10 ng/μl) under two solvation conditions (5X SSC buffer and H₂O). As shown in fig. S3, we observed increased DNA adsorption on the gold surface with rising DNA concentrations at three treatment points. DNA concentrations of 5 ng/μL and 10 ng/μL resulted in saturated measurements on the gold surface, leading to inflated signals during treatment (Day 6) and post-treatment (Day 12), and reducing the ability to resolve treatment-dependent differences. In contrast, the 1 ng/μl DNA samples produced weaker signals compared to the 2 ng/μl samples. Based on these findings, we selected 2 ng/μl as the standard DNA concentration.

We also observed higher DNA adsorption on the gold surface when DNA was dissolved in 5X SSC buffer compared to pure H₂O. According to a previous study (15), the addition of salt from the 5X SSC buffer can induce charge neutralisation, thereby promoting DNA adsorption on the gold surface, results consistent with our observations. However, monitoring the EMT process requires a more sensitive detection method, and we needed to minimize signal saturation caused by salt-induced charge neutralisation. When we switched the solvent to H₂O, we observed lower signals compared to 5X SSC. For example, in fig. S3, MCF7 DNA in 5X SSC buffer maintained high adsorption at Day 12 post-TGF-β treatment, but when the same DNA sample was dissolved in pure H₂O, DNA adsorption decreased on Day 12 (fig. S3). This result aligns with previous flow cytometry findings, which showed a restoration of the epithelial phenotype on Day 12, consistent with the decreased adsorption observed in H₂O on Day 12. Therefore, to minimize salt-driven DNA-gold interactions and improve signal consistency, all assays were performed in H₂O using an optimized DNA concentration of 2 ng/μl, which provided a favourable signal to noise ratio, and stable, reproducible ir (%) responses within the working range of the Methylscape system.

Based on the optimization assay described earlier, we proceeded with 2 ng/μl of DNA dissolved in pure H₂O as our standard condition. Both breast cancer cell lines showed a significant increase in DNA adsorption during the TGF-β treatment period (Day 3 and Day 6) (Fig. 2D and fig. S1D). After the treatment was withdrawn, a decrease in DNA adsorption was observed in both cell lines, which was consistent with the flow cytometry results. These findings demonstrate that Methylscape can monitor the EMT process by tracking DNA adsorption on the gold surface, with results that are further corroborated by transcriptomic changes detected by RNA sequencing and phenotypic changes observed through flow cytometry. In addition, relative global methylation was quantified at Day 0 and Day 6 only, representing the baseline and the major EMT transition stage, respectively. As shown in Fig. 2E, we observed an increase in global methylation from 44.02% at Day 0 to 54.1% at Day 6. This finding is consistent with previous reports demonstrating increased overall methylation in TGF-β-induced EMT samples (20, 32–34). These results suggest that Methylscape could serve as an advanced tool for monitoring cancer progression by tracking changes in its levels.

Various stimuli such as hypoxia and growth factors including EGF and HGF can also trigger EMT and can be used to study cancer progression and metastasis. In this study, however, our goal was to investigate how Methylscape signals change in response to dynamic DNA methylation changes during cancer progression. To achieve this, we required a EMT model that is mechanistically well defined, reproducible, and widely accepted for studying dynamic methylation changes. Among the different EMT-inducing pathways, TGF-β-induced EMT represents the most canonical and extensively characterised model. It robustly activates the core EMT transcriptional program and has been consistently used in previous studies exploring epigenetic and DNA methylation dynamics during EMT progression (26–28). Therefore, we employed TGF-β-induced EMT as our model system, as it provides a consistent and reliable framework for elucidating the mechanistic basis of Methylscape-associated methylation changes. To further clarify this mechanism, we plan to develop an approach to desorb both untreated and TGF-β-treated DNA from the gold substrate, followed by quantitative PCR and whole-genome sequencing analyses, to provide deeper insight into the molecular basis of Methylscape’ s detection capability.

Understanding the mechanism of Methylscape through the desorption model

Clustered hypermethylation at DNA promoter regions is a key factor in cancer progression, as it silences transcription and inactivates crucial genes. This epigenetic reprogramming alters the physicochemical properties of the cancerous genome, significantly impacting its interaction with gold surfaces. To gain deeper mechanistic insights, our study developed a desorption model to investigate the fundamental interactions between DNA and gold surfaces, demonstrating that hypermethylated cancerous DNA exhibits a stronger adsorption preference compared to non-cancerous DNA. Figure S4 represents the scheme of the DNA adsorption, desorption, qPCR detection, and sequencing approaches. Briefly, purified DNA from the MCF7 breast cancer cell line was initially adsorbed onto gold electrode surfaces and relative adsorption was measured electrochemically (fig. S5, A and C). Desorption of DNA was then achieved either electrochemically or via a competitive method using MCH, which has a higher affinity for gold than DNA and effectively displaced the adsorbed DNA. We did not observe any significant change in the DPV signal after DNA desorption when using the MCH-based method (fig. S5C), whereas a sharp increase in DPV current was evident in the case of electrochemical desorption (fig. S5A). This likely occurred because MCH formed a densely packed self-assembled monolayer (SAM) that continued to hinder ferrocyanide access to the electrode surface. This observation aligns with previous studies reporting that MCH SAMs on gold can reorganize into tightly packed and highly passivating structures, which markedly increase charge-transfer resistance and reduce ferro/ferricyanide accessibility (35). To further verify this phenomenon, we conducted control experiments comparing bare gold electrodes with MCH-modified electrodes forming SAMs (fig. S6) and observed that MCH treatment caused a pronounced decrease in redox current. Finally, MCH-based competitive desorption method was selected for subsequent experiments due to its superior yield. The viability of the desorbed DNA for downstream analysis was confirmed by PCR amplification (fig. S5, B and D) leveraging a universal primer-probe sets (n = 4) to detect any human DNA (table S1) (36). Finally, the methylation analysis of the desorbed DNA was performed by qPCR, whole genome sequencing and whole genome bisulphite sequencing (WGBS) approaches.

To obtain greater insight into the desorbed DNA, we designed 8 distinct sets of primers from four different genes (such as ALX1, MACF1, IGSF3, and CLPTM1L) to detect the desorbed DNA more specifically. For each of the genes, 1 primer set (forward and reverse primers) was chosen from the CpG-rich hypermethylated promoter region and another set from the distant gene body (intergenic) regions which are hypomethylated (table S2). These genes were selected based on TCGA (The Cancer Genome Atlas) data, which revealed that they exhibit promoter hypermethylation and gene body hypomethylation in breast cancer. Specifically, the targeted CpG islands of ALX1 and MACF1 each contain a CpG site that is methylated in breast cancer and unmethylated in normal breast tissue, with their corresponding gene-body assays located 20 kb and 148 kb downstream, respectively. In contrast, the targeted CpG islands of IGSF3 and CLPTM1L contain three and four CpG sites, respectively, that are highly methylated in both breast cancer and normal tissue, with their gene-body assays situated 50 kb and 27 kb away from the respective CpG islands (table S2). To validate these selections in our study, we analysed WGBS data by comparing untreated MCF7 samples with normal controls. We observed consistent patterns, ALX1 and MACF1 CpG islands were hypermethylated in MCF7 cells relative to normal tissue, whereas CLPTM1L and IGSF3 showed similar methylation levels between the two groups. All gene-body regions were consistently hypomethylated in MCF7 compared with normal controls (table S3). After that, fig. S7 presents the qPCR results, showing successful amplification of all targeted regions, thereby confirming the adsorption of hypermethylated and gene-body regions of DNA onto the gold electrodes.

To investigate further, we analysed the desorbed DNA by qPCR to calculate copy numbers and concentration of the desorbed DNA sample to reveal the methylated DNA enrichment ratio (i.e., DNA copies from promoter regions versus intergenic regions). Figure 3A and fig. S8 show the computed copy numbers and concentrations of desorbed MCF7 DNA detected by primers targeting the promoter and gene body regions of each gene, respectively. Both copy numbers and concentrations of DNA were higher in promoter regions where the IGSF3 promoter region detected 13 copies/μl, while the gene body region detected 4 copies/μl, yielding a promoter/gene-body enrichment ratio of 8.33. Similar ratios were observed for ALX1, MACF1, and CLPTM1L, at 1.40, 3.06, and 4.39, respectively (Fig. 3B). These findings support the hypothesis that hypermethylated regions preferentially adsorb onto the gold surfaces. Furthermore, Fig. 3, A and B illustrate the copy number and promoter-to-gene body enrichment ratio of desorbed DNA from cancerous (MCF7) and non-cancerous (healthy buffy coat) samples. As anticipated, the healthy samples exhibited lower copy numbers and promoter-to-gene body ratios compared to the MCF7 DNA, reflecting reduced desorption yields due to weaker adsorption onto the gold surface. These findings are consistent with our previous observations that normal genomic DNA, despite its higher global methylation levels (60–80%), shows reduced adsorption to gold. This is likely due to the hydrophobicity induced by widespread methylation, which limits the accessibility of locally clustered hypermethylated regions to interact effectively with the gold surface.

Fig. 3. qPCR and whole-genome and whole genome bisulphite sequencing of desorbed DNA.

(A) qPCR analysis revealed a significantly higher DNA copy number and (B) methylated DNA enrichment ratio in MCF7 desorbed DNA compared to non-cancerous DNA. Significant differences between time points were assessed using a Two-way ANOVA. ***P < 0.001, ****P < 0.0001. (C) A highly significant difference was observed in the number of CpGs within CpG islands (CGIs) between non-desorbed and desorbed DNA. ***P < 0.001. (D) Methylated DNA coverage at promoter CpG islands was substantially higher than unmethylated DNA coverage in both untreated and TGF-β-treated samples. (E) An increasing trend in read coverage per MB correlated with a higher methylation ratio (%) in both untreated and treated desorbed DNA samples at promoter regions. Additionally, methylated read coverage at promoter regions was significantly higher than at intergenic regions for both sample types. (F) Average gene expression showed a decreasing trend with increasing methylation ratio (%) at promoter regions in both untreated and treated RNA samples. (G) Most reads (∼76–79%) aligned uniquely, ensuring high sequencing quality. (H) Majority of CpG islands had high sequencing depth (≥50X), supporting robust methylation analysis. (I) High bisulphite conversion efficiency (∼99.2–99.5%), stable CpG coverage (∼45–48 million CpGs), and adequate sequencing depth (4X-15X) confirm data reliability.

This findings from the qPCR assays prompted us to sequence the desorbed DNA to further investigate the mechanism of Methylscape. We performed whole genome sequencing analysis of two desorbed DNA samples (untreated and TGF-β treated Day 6 MCF7 DNA) and compared read coverage between promoter regions (CpG islands) and intergenic regions (fig. S9). In our whole genome sequencing analysis, we identified a significant difference in the number of CpG sites within CpG islands (CGIs) between non-desorbed (positive control) and desorbed untreated MCF7 DNA samples. The desorbed DNA displayed a substantially higher number of CpG sites in CGIs compared to the non-desorbed sample (Fig. 3C). This is potentially due to the preferential adsorption of hypermethylated CpG-rich DNA on the gold surface. These findings indicate that desorbed DNA provides a selective enrichment of methylated DNA within CpG islands, which could be beneficial for accurately characterising epigenetic modifications in cancer cells. To determine the methylation status of the sequenced reads, we used TCGA methylation data from breast cancer patients and correlated it with our sequenced reads, assuming similar methylation pattern in MCF7 breast cancer cells. Figure 3D shows that read coverage of methylated CpGs was significantly higher than that of unmethylated CpGs in promoter regions for both samples, supporting the hypothesis that methylated DNA has a higher affinity for gold surfaces, consistent with a recent simulation study (37). Interestingly, the read coverage of both methylated and unmethylated CpGs are significantly higher in the case of TGF-β treated Day 6 MCF7 Samples (Mesenchymal states) which is consistent with the higher adsorption of mesenchymal DNA towards the gold surface. This desorption sequencing data supports the higher adsorption of DNA from TGF-β-treated Day 6 MCF7 cells onto the gold surface (Fig. 2D). This further confirms that Methylscape can effectively track the dynamic changes in methylation patterns during EMT and cancer progression, the gradual increase in CpG hypermethylation and intergenic hypomethylation.

To better understand the methylation status of desorbed DNA samples, we created a graph categorizing promoter regions by percentage of methylation (0–10%, 11–20%, 21–30%, etc.) and assessed whether read coverage increased with higher methylation percentages. We also compared methylation rates between promoter and gene body (GB) regions. Figure 3E shows that read coverage increased with higher methylation percentages in promoter regions for both samples, particularly for the TGF-β-treated sample. Additionally, methylated DNA read coverage was significantly higher in promoter regions (CpG islands) compared to intergenic regions (gene bodies). Read coverage was normalized per megabase (MB) due to significant differences in nucleotide base counts among the methylation sub-groups. Moreover, we sequenced RNA from untreated and TGF-β treated MCF7 cells to explore the link between methylation status and gene expression. Figure 3F shows that average gene expression decreased as methylation levels increased in promoter regions. This supports previous findings that hypermethylation can inhibit gene expression and regulate transcription. Methylation recruits’ methyl-CpG-binding domain proteins and histone deacetylases, which prevent RNA polymerase binding and repress gene expression (4, 38).

To directly investigate the epigenetic reprogramming patterns, we performed whole-genome bisulphite sequencing (WGBS) on a set of three desorbed MCF7 DNA samples (untreated/pre-treated Day 0 MCF7, TGF-β-treated Day 6 MCF7, and post-treated Day 12 MCF7). The Bismark alignment score from WGBS were present in Fig. 3G, revealing that approximately 78% of reads are uniquely aligned. This high alignment rate indicates excellent alignment quality and supports the reliability of our sequencing data for accurate methylation analysis. Moreover, our WGBS experiments achieved high CpG island coverage in all samples (positive control, pre-treatment, during treatment, and post-treatment), with over 50x coverage for around 25,000 CpG islands (Fig. 3H). This coverage ensures a robust and confident assessment of methylation patterns within CpG-rich regulatory regions, facilitating precise comparative analyses of methylation changes across different treatment stages. Table 3I evaluates sequencing quality, demonstrating high bisulphite conversion efficiency (99.2–99.5%), a critical measure of reliable unmethylated cytosine deamination.

Next, we performed methylation Analysis of our WGBS data and identified differentially methylated regions (DMRs) and differentially methylated cytosines between the three desorbed DNA conditions: untreated, treated, and post-treated samples. Figure 4, A and B and Table 4G shows that there are higher number of hypermethylated regions and methylated cytosines in the desorbed TGF-β-treated Day-6 MCF7 DNA samples in comparison to the desorbed untreated Day 0 DNA samples. The WGBS data from the desorbed DNA further validates our earlier qPCR and whole-genome sequencing findings, reinforcing that hypermethylated regions exhibit a stronger affinity for the gold surface and as cancer progresses, the adsorption of hypermethylated DNA increases, a trend that Methylscape can effectively identify. Figure 4, A to F present the DMRs and DMCs observed under additional experimental conditions.

Fig. 4. The WGBS data reveal treatment-induced DNA methylation changes across three conditions: untreated, treated, and post-treated samples.

Differentially methylated regions found in (A) Untreated vs Treated samples, (B) Treated vs Post-treated and (C) Untreated vs Post-treated samples and differentially methylated cytosines found in (D) untreated vs treated samples, (E) Treated vs Post-treated and (F) Untreated vs Post-treated samples. Table (G) showing the actual number of DMCs and DMRs were detected across treatment conditions.

Clinical translation - Methylscape diagnostic assay using disposable electrodes to detect breast cancer progression

After validating Methylscape for monitoring the EMT process and investigating its underlying mechanisms using a desorption enrichment method combined with sequencing, we aimed to assess the translational potential of the Methylscape platform in tracking cancer progression. Leveraging the unique physicochemical properties of cancer DNA, which preferentially adsorbs to a bare gold surface, we evaluated epigenome remodelling in a cohort of breast cancer patients across different stages (I-III).

Considering that clinical pathology prefers single-use detection test to minimise sample contamination and operator variability, one potential avenue for electrochemical biosensing-based diagnostic application is screen-printed electrodes (SPE) as they offer the advantages of low-cost production and practical feasibility. The presented interfacial biosensing methodology was thus developed on commercially available gold SPE by comparing the relative DNA adsorption to each electrode surface during resistive measurement using DNA derived from a breast cancer model cell line (MCF7) to simulate tumour sample and buffy coat DNA from a pool of blood donors to simulate a control, healthy sample. All genomic DNA samples used in this exploratory study were purified using a gold-standard phenol-chloroform extraction process and for each electrode testing, 100 ng (i.e., 10 μl at 10 ng/μl) of DNA was incubated on the gold surface for 10 minutes before differential pulse voltammetry (DPV) experiment.

Results in Fig. 5A shows the ability of the developed SPE biosensing approach to differentiate between control buffy coat and MCF cancerous samples with a difference in gold adsorption of almost 50%, a nearly four-fold increase compared to conventional re-usable lab-based electrode system.

Fig. 5. Clinical utility of the screen-printed gold electrode (SPE) platform to stratify stages of breast tumours using Methylscape.

(A) Differences in biosensing detection performance between single-use (SPE) and traditional re-usable gold electrodes. Mean with SD. (B) Graphs showing the relative gold adsorption (%Ir) of pooled buffy coat DNA samples spiked with MCF7 cancer DNA at different ratios (from 0% to 100%). (C) Clinical analysis of DNA purified from 6 different normal adjacent breast tissues (blue) and 24 different breast tumour tissue clinically classified as stage 1 (pink, n = 8), stage 2 (red, n = 8) and stage 3 (purple, n = 8). Statistical significance between groups (unpaired t-test) is shown for normal adjacent vs. stage 1 (P = 0.387, ns), vs. stage 2 (P = 0.046, **) and vs. stage 3 (P < 0001, ****) respectively. The horizontal dotted line shows an arbitrary threshold value of %Ir = 40%. Per cent values (%) indicate the proportion of patient samples with a mean %Ir value being below the %Ir = 40% threshold for normal adjacent and stage 1 or being above the threshold for stages 2 and 3. Mean with SD. (D to I) Receiver operative characteristics curves were computed using the data plotted in (C).

Given that many tumour samples contain significant proportions of normal adjacent tissue, the sensitivity of the proposed electrochemical detection system to contaminating “normal” DNA was evaluated by spiking buffy coat DNA with MCF7 breast cancer line DNA at different ratios. Results in Fig. 5B indicate that the SPE platform can detect as low as 5% of cancer DNA in a matrix of 95% of buffy coat DNA when analysing 100 ng of samples.

Finally, to demonstrate the clinical utility of the methodology to detect global epigenetic remodelling associated with cancer progression, DNA purified from breast cancer tumour samples across three different cancer stages (i.e., stages 1, 2 and 3) was profiled and compared to DNA isolated from normal adjacent breast tissues. Eight different patients were selected for each stage for a total of 24 breast cancer samples, together with six normal adjacent breast tissue controls (see table S4 for patient details). Each tumour DNA samples were analysed in electrode triplicates while normal adjacent samples in duplicates. The resulting data show a global trend with the relative DNA gold adsorption increasing with tumour stages, where stage 2 and 3 samples tend to sit above an arbitrary chosen threshold value of %Ir = 40% as opposed to stage 1 and normal adjacent samples (Fig. 5C). More precisely, when considering the mean %Ir value of each patient sample and comparing each different diseased group, 100% of stage 3 samples (8/8) and 62.5% of stage 2 samples (5/8) were above the 40% threshold, while 87.5% of stage 1 samples (7/8) and 100% of normal adjacent samples (6/6) were below. In addition, statistical significances (unpaired t-test) were observed when comparing the patient’s data from the normal adjacent group to the ones from stage 2 (P value = 0.0046) and stage 3 (P value <0.0001) respectively. However, no significant statistical difference was observed between the normal adjacent and the stage 1 samples (P = 0.387). Together, these data translate to an area under the receiving operating curve (AUROC) of 0.65, 0.92, and 1 when comparing normal samples vs stage I, normal vs stage 2, and normal vs stage 3 samples respectively (Fig. 5, D to F), and to an AUROC of 0.92, 0.95, and 0.70 when comparing stage 1 vs2, stage 1 vs 3, and stage 2 vs 3 (Fig. 5, G to I).

These promising results reveal the potential of the presented Methylscape to stratify stages of the disease by tracking evolving epigenomes associated with cancer progression. As opposed to other existing approaches in the field, our unique methodology focuses on the analysis of the physical consequences induced by epigenetic rearrangements of DNA molecules. Such features are difficult to detect by sequencing analysis only and could provide complementary information to conventional cancer diagnostic technologies for a more comprehensive diagnosis. Given the clinical potential and the unprecedented simplicity of the presented SPE platform (e.g., label-free and functionalisation-free sensors, no DNA treatment nor amplification), it is believed to find great opportunities as a cost-effective and rapid DNA methylation-based cancer detection assay to screen population in low resource settings.

In conclusion, our study demonstrates that Methylscape offers a simple, rapid, and cost-effective strategy for monitoring cancer progression by quantifying DNA adsorption onto a gold surface. Using an EMT-based cell line model, we first validated the concept, showing significantly higher Methylscape signals in DNA from EMT-transitioned cells. To gain mechanistic insights, we desorbed the pre-adsorbed DNA from the gold surface and performed qPCR, whole-genome sequencing, and whole-genome bisulphite sequencing. The results revealed preferential adsorption of clustered hypermethylated DNA regions onto gold, supporting our hypothesis that hypermethylated domains exhibit stronger gold affinity than hypo- or unmethylated regions. To facilitate clinical translation, we employed a disposable screen-printed electrode as a low-cost detection platform. Together, these findings highlight Methylscape’s potential as a robust tool for tracking cancer progression, enriching cancer-specific methylated DNA, and enabling the development of clinically applicable assays for disease monitoring.

MATERIALS AND METHODS

Materials

The epithelial breast cancer cell line MCF7 (HTB-22) and the mesenchymal cell line MDA-MB-231 (HTB-26) were obtained from ATCC. Cell culture reagents including RPMI1640 medium, fetal bovine serum (FBS), Glutamax, Ultrapure water, penicillin-streptomycin were purchased from Thermo Fisher Scientific (USA/Australia). Qubit RNA HS Assay Kit were purchased from Thermo Fisher Scientific (USA/Australia), and DNA/RNA concentrations were quantified using the Qubit 4 Fluorometer (Invitrogen, USA). Recombinant TGF-β was from Abcam (UK), and phosphate-buffered saline (PBS; 10 mM, pH 7.4) was purchased from Sigma-Aldrich. Proteinase K (New England BioLabs, p8107s), Buffer AL (QIAGEN, 19075), and the DNeasy Blood and Tissue Kit (QIAGEN, 69504) were used for DNA extraction. For bisulphite conversion, the EZ DNA Methylation-Gold Kit was used (Zymo Research). RNA extraction was conducted using the RNeasy Plus Mini Kit (Qiagen GmbH, Hilden, Germany).

For DNA fragmentation, a Covaris S2 sonicator was used (Covaris), and library preparation for both WGS and WGBS was carried out using the NEBNext Ultra II DNA Library Prep Kit and NEB E7760 or E7595s RNA library kits (New England BioLabs). Sequencing was performed using Illumina platforms: the NextSeq 500 and NovaSeq 6000 systems. Quality control and size distribution analysis were conducted using the Agilent Bioanalyzer (Agilent). Electrochemical measurements were carried out using the CH1040C potentiostat (CH Instruments, USA), and electrodes were fabricated using AZnLOF 2070 photoresist (MicroChem, Newton, MA) and processed with a Temescal FC-2000 e-beam evaporator. Software tools used for primer design and analysis included PrimerROC, PrimerSuite, Trim Galore, and Bismark, along with a custom Python script.

Culture of target cell lines

The selected epithelial breast cancer cell line MCF7 (ATCCHTB-22) and mesenchymal MDA-MB-231 (ATCCHTB-26) were cultured in RF10 comprising of RPMI1640 culture media (Thermo Fisher Scientific, US) with 10% (v/v) FBS, (Thermo Fisher Scientific, US), 2 mM Glutamax (Thermo Fisher Scientific, US), and 1% penicillin-streptomycin (Thermo Fisher Scientific, US). The cell line was grown in T75 culture flasks in a humidified incubator in 5% CO2 at 37°C. During Day 0–6, cells were maintained in the low serum culture medium that contained 2% (v/v) FBS. Cells were treated with an additional 10 ng·ml − 1 TGF-β (Abcam, UK) during the treatment. Cells were passaged every 3 days to maintain consistent confluency (60–70%). The cell line was tested to be mycoplasma-negative before and after treatment. The collected cells from the culture flask were washed and diluted with 1X PBS (10 mm, pH 7.4, Sigma-Aldrich) and nucleic acids were isolated.

Sample preparation and nucleic acid extraction

All samples were purified using a gold-standard organic solvent extraction protocol (phenol:chloroform:isoamyl alcohol, 50:49:1) followed by isopropanol/ethanol precipitation. Briefly, cultured cells and clinical samples were suspended in 200 μl of 1× PBS, 20 μl of proteinase K (BioLabs, p8107s) and 200 μl Buffer AL (QIAGEN, 19075). The samples were vortexed and incubated at 56°C for 10 min. Add an equal volume of phenol:chloroform:isoamyl alcohol to the tube and vortexed the solution until the phases are mixed. Centrifuged for 2–3 min at 20238 g to separate phases. Transferred the aqueous phase, added 1/10 of the above volume of 3 M Sodium Acetate (pH 5.2), 1 volume of 100% isopropanol. Centrifuged the mixture at 13000 g for 15 min at 4°C. Removed the isopropanol and added 1 ml of cold 70% ethanol and mixed the solution by gently inverting it 10 times. Centrifuged the mixture at 13000 g for 10 min at 4°C. Removed ethanol and dried the nucleic acid pellet. Added Ultrapure water (Thermo Fisher Scientific, USA) or Low TE buffer to dissolve the DNA fully. No RNase treatment was included during nucleic acid extraction. The concentration of DNA measured by Qubit (Qubit 4 Fluorometer, Invitrogen, USA) followed the manufacturer’s instruction. The purity of sample was assessed by measuring the 260/280 and 260/230 ratios using a NanoDrop Spectrophotometer (ND-1000, version 3.8.1).

DNA adsorption on the gold electrode and DPV current measurements

Added readout DNA solution to cover the full electrode gold surface. Measured DPV as baseline under the standard parameters by CH1040C poteniostat (CH Instruments, USA) as the ‘working electrode’. Differential pulse voltametric (DPV) experiments were conducted in 10 mM PBS solution containing 2.5 mM [K₃Fe(CN)₆] and 2.5 mM [K₄Fe(CN)⁶] electrolyte solution. DPV signals were obtained with a potential step of 5 mV, pulse amplitude of 50 mV, pulse width of 50 ms, and pulse period of 100 ms. DPV signals of clean electrodes were measured to get the baseline current and then, added certain concentration of DNA diluted by Ultrapure water (Thermo Fisher Scientific, USA) or 5 X SSC on each electrode and incubated the chip for 10 min at room temperature with 250 rpm vortex. Rinsed the electrodes and dried by nitrogen air. The adsorption competence was measured using the [Fe(CN)₆]^4−/3- redox system. Upon DNA adsorption, the coulombic repulsion between negatively charged ferrocyanide ions in the buffer and negatively charged DNA phosphate groups on the electrode surface partially hinders the diffusion of ferrocyanide ions to the electrode surface. This generates a Faradaic current signal, which is proportionally lower than the bare electrode signals as increasing numbers of DNA molecules become adsorbed onto the surface. The relative adsorption currents (%ir) was measured by the equation

$$\text{Adsorption current} (\%{\mathrm{i}}_{\mathrm{r}})=\left[\frac{({\mathrm{i}}_{\text{baseline}}-{\mathrm{i}}_{\text{sample}})}{{\mathrm{i}}_{\text{baseline}}}\right]\times 100$$

DNA desorption

Following adsorption, the electrodes were washed with 1X PBS buffer three times. To desorb the pre-adsorbed DNA from the gold electrode surface, Mercaptohexanol (MCH) solution was added upon the dried electrode surfaces and incubated for 30 minutes. Since MCH has a stronger affinity for gold than DNA, it effectively displaces the pre-adsorbed DNA molecules from the gold surface upon application. After desorbing DNA from the gold surface, the collected DNA solution was purified using the QIAquick PCR Purification Kit (Cat. No. 28106, Qiagen) to remove MCH, following the manufacturer’s protocol, prior to qPCR and sequencing experiments.

Electrochemical desorption

Electrochemical desorption was performed by applying a constant negative potential of −2.5 V for 5 min to the DNA-modified gold working electrode in a standard three-electrode configuration (Ag/AgCl reference and Pt counter electrodes). The electrode assembly was immersed in 1 ml of PBS (1x) instead of the [Fe(CN)₆]^4−/3- electrolyte during the desorption process, where the strongly negative bias induces electrostatic repulsion and releases surface-bound DNA into the buffer. Immediately after desorption, the electrodes were transferred into the [Fe(CN)₆]^4−/3- electrolyte for DPV analysis. DPV measurements were acquired using the following parameters: initial potential −0.5 V, final potential +0.5 V, increment 0.005 V, pulse amplitude 0.05 V, pulse width 0.05 s, pulse period 0.1 s, quiet time 2 s, and sensitivity 1 x 10–⁵ A/V.

Detection of desorbed DNA by real-time polymerase chain reaction (qPCR)

Initially, four sets of universal primer-probes of human DNA detection were used for the qPCR assay (36). The information regarding the universal qPCR assays is summarized in table S1 (36). In the final phase of the study, 8 distinct sets of primers from four different breast cancer genes (such as ALX1, MACF1, IGSF1, and CLPTM1L) were designed to detect the desorbed DNA. Among the primer sets for each gene, 1 primer set (forward and reverse primers) was chosen from the CpG-rich hypermethylated region and another set from the distant gene body region. The sequences of these primers have been included in table S2. qPCR assays were designed using the PrimerROC software and PrimerSuite software packages (39, 40), together with a custom python script that also predicted the best oligonucleotide to use for fluorescent Taqman assays, targeting an annealing temperature of 60°C.

Sequencing process

Whole genome sequencing

Whole genome sequencing libraries were constructed by first sonicating the purified desorbed DNA with a Covaris S2 sonicator to achieve a fragment size of approximately 200 bp. The sheared DNA fragments were then subjected to end-repair, A-tailing, and ligation of sequencing adaptors using the NEBNext Ultra II DNA Library Prep Kit (New England BioLabs), following the manufacturer’s recommended protocols. Libraries were then quantified using a Qubit fluorometer (Thermo Fisher Scientific) and diluted to the required concentration. Sequencing was conducted on an Illumina NextSeq 500 platform using a 150 MId kit, with a paired end 2x71 bp read configuration and dual index adaptors to enable accurate demultiplexing of samples. Quality control of sequencing reads was performed post-run to assess read quality.

Whole genome bisulphite sequencing

Whole-genome bisulphite sequencing libraries were prepared to assess genome-wide DNA methylation patterns. Purified desorbed DNA was fragmented to an average size of ∼200 bp using a Covaris S2 sonicator, followed by end-repair, A-tailing, and adaptor ligation using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England BioLabs), according to the manufacturer’s protocols. Adaptor-ligated DNA was treated with the EZ DNA Methylation-Gold Kit (Zymo Research) to convert unmethylated cytosines to uracils, while methylated cytosines remained unchanged. The bisulphite-treated DNA was then amplified with high-fidelity, low-bias PCR to create the final library. Quality and size distribution of the libraries were verified using an Agilent Bioanalyzer, and libraries were quantified via qPCR. Sequencing was carried out on an Illumina NovaSeq 6000 system in paired-end 150 bp mode. Raw sequence reads were processed using Trim Galore to remove adaptor sequences and low-quality bases, followed by alignment to a bisulphite-converted reference genome using Bismark. Bisulphite conversion efficiency was assessed using spike-in unmethylated lambda DNA.

Region-level methylation analysis from WGBS

Preprocessing

Validation of the four candidate genes (ALX1, MACF1, IGSF3, CLPTM1L) was carried out through a region-level methylation analysis spanning the primers listed in table S2. Forward and reverse primers for CpG islands and gene bodies were mapped onto the hg38 genome using UCSC In silico PCR (41), and converted into a BED file delineating the amplicon genomic coordinates for methylation analysis. LiftOver (42) was used to convert hg38-aligned .bed files to hg19, to ensure consistency among regions tested. Three Luminal breast tissue healthy controls were downloaded from the Human methylome atlas (43) GEO Series GSE186458 accession numbers: GSM5652347, GSM5652348 and GSM5652349. The obtained Bigwig files were converted to per-CpG β values, and a pseudo-coverage value of 1 was assigned to each CpG. Lastly, Bismark coverage files for the two tumour sample replicates were merged into one technical sample by summing methylated and total read counts per CpG site and recalculating β. Normal and tumour samples were further restricted to amplicon regions for the four candidate genes for region-level methylation analysis.

Analysis

Region-level methylation analysis was carried out in R v.4.4.2 using GenomicRanges (44), rtracklayer (45) and tidyverse (46), on the BED file genomic regions defined by the amplicons. Comparisons were carried out in a one-vs-mean approach, comparing the tumour technical sample to the mean of normal controls. In each case, the β-value was calculated as a simple unweighted mean, for comparison. For healthy controls, simple β values are identical due to the assigned pseudo-coverage. In each case, changes in methylation patterns were computed as Δβ between tumour and normal control across the amplicons.

RNA extraction

High-quality genomic RNA was extracted from cultured breast cancer cells using Qiagen RNeasy Plus Mini Kit (Qiagen GmbH, Hilden, Germany) in accordance with the manufacturer’s protocols. Moreover, we used the Qubit 4.0 Fluorometer (ThermoFisher, Australia) to measure the concentration and quality of the extracted RNA using the Qubit RNA HS Assay Kit (ThermoFisher, Australia). Depending on sample volume, the assay kit provides an accurate measurement for initial RNA sample concentrations of 0.2 to 200 ng/μl, providing a detection range of 4–200 ng.

RNA sequencing

The RNA library was prepared with NEB E7760 kit for rRNA depleted FFPE RNA and NEBNext Ultra II (E7595s) Directional RNA library prep kit for Illumina. RNA Sequencing was performed on the NextSeq 500 Sequencer using default parameters.

RNA-seq and enrichment analysis

The Tuxedo suite (47) was used to perform transcript-level expression analysis of RNA-seq experiments with hist2, stringtie, and ballgown. Briefly, hisat2 (2.2.1) was used to map fastq files that were previously trimmed with [trimmomatic] were mapped to the hg38 genome (options: hisat2 -p 6 --dta --sensitive --no-soft -x < hisat2.index.prefix> − q − 1 sample.R1.fastq −2 sample.R2.fastq -S sample.sam). Transcripts were assembled based on mapping to genome u sing the stringtie (2.1.6) package (option: --merge -p 6 -G reference.gtf -o merged_gtf_file merged_file). gffcompare (0.112) was used to compare the assembled transcripts to known transcripts. Stringtie (2.1.6) was used to estimate abundance of assembled transcripts (options: -e -B -p 6 -G merged_gft_file -o ballgown_gtf_file sample.sortedbam). Transcript annotation and gene level expression analysis was conducted using the R package (ballgown) and visualized using ggplot2. Differential expression analysis was performed using the R package EnhancedVolcano using standard protocols outlined in Blighe (2018), where P ≤ 0.05, and fold change was set to 1. Enrichment analysis was performed in R where transcripts that displayed significant fold change (FDR < = 0.01, and log2 fold change > = 1) were selected for enrichmenta. (All scripts for differential analysis and enrichment analysis are available on request).

Global DNA methylation analysis

Global DNA methylation levels in MCF7 control and TGF-β-treated samples were quantified using the Methylated DNA Quantification Kit (Colorimetric) (Abcam, ab117128) following the manufacturer’s instructions. Briefly, 1 μl each of the negative and positive controls, along with 100 ng of sample DNA, were added to individual wells containing 80 μl of Binding Solution. Plates were incubated at 37°C for 90 minutes. After incubation, the Binding Solution was removed and wells were washed three times with 150 μl of 1× Wash Buffer. Next, 50 μl of diluted Capture Antibody (1:1000) was added to each well and incubated at room temperature for 60 minutes. Wells were washed three times with diluted Wash Buffer, followed by the addition of 50 μl of diluted Detection Antibody (1:2000) and incubation for 30 minutes at room temperature. After another wash, 50 μl of diluted Enhancer Solution (1:5000) was added and incubated for 30 minutes. Wells were washed again and 100 μl of Developer Solution was added, followed by incubation at room temperature for 10 minutes. The reaction was stopped with 50 μl of Stop Solution. Absorbance was measured at 450 nm using a Tecan Infinite M200 Pro microplate reader. Global methylation levels were calculated using the following equation:

$$\begin{array}{c}\text{Relative Global Methylation level} (\%)=\hfill \\ \left[\frac{({\mathrm{A}}_{450}\text{Sample}-{\mathrm{A}}_{450}\text{Negative control})}{({\mathrm{A}}_{450}\text{Positive control}-{\mathrm{A}}_{450}\text{Negative control})}\right]\times 100\hfill \end{array}$$

Clinical samples

Human blood and human breast tumour samples used in this study received ethical approval from the Bellberry Human Research Ethics Committee (HREC) under the study titled “Enabling Clinical Epigenetic Diagnostics: The Next Generation of Personalized Breast Cancer Care” (Episensors01; reference number 2015–12-817-PRE-7). Breast tumour tissues (block stabilized by PAXgene agent) were first sectioned into 10 μm slices using a microtome (Leica) and then dewaxed by incubation in mineral oil at 80°C for 2 min. A digestion buffer containing proteinase K was then added to the tissue sections and incubated overnight at 50°C with shaking (300 RPM). The digested samples were then centrifuged at 10,000 g for 15 sec and the lower aqueous phase containing the DNA was transferred into a new tube for organic extraction as described above.

Statistical analysis

All statistical analyses were performed using GraphPad Prism (GraphPad Software v.10.4.1) and values are given as mean with SD as indicated. When two groups were compared, significance was determined using an unpaired two-tailed t-test. One-way ANOVA, or Two-way ANOVA were used for multiple comparisons, and P values adjusted using Šidák for multiple comparisons where it was appropriate. P value threshold of 0.05 was considered statistically significant.

Supplementary Materials

This PDF file includes:

Figs. S1 to S9

Tables S1 to S4