The technology landscape for detection of DNA methylation in cancer liquid biopsies.

Introduction
One of the most studied epigenetic alterations is DNA methylation. It is the addition of a methyl group (-CH3) to the fifth position of a cytosine in a CpG dinucleotide context. CpG dinucleotides only represent about 1% of the mammalian genome but tend to cluster in so-called CpG islands (CGI). CGIs occur in the majority of human protein-coding genes near the transcription start site but can also occur in the gene body [1,2].
Methylation is crucial for normal cell functioning and development, so abnormalities in the process can lead to various diseases. Not only does its absence or presence mediate its biological function but also its location in the genome [3]. Methylation changes occur early in cancer development where, contrary to normal development, promotor regions of genes (e.g., tumor suppressor genes) are hypermethylated, while the gene body and repetitive elements are hypomethylated. This leads to genomic instability and dysregulation of normal cell function [2,4].
Initially, methylation studies focused on the promotor and transcription start site (TSS) hypermethylation of protein-coding genes. However, more recent genome-wide analyses revealed the role of methylation at CpG sites within introns, intergenic sequences and exons, which led to the identification of CpG shores (2kb from CGI), CG shelves (2-4kb from CGI), and open sea regions (rest of the genome) [1]. This shows the enormous potential of the epigenome as a source of methylation biomarkers, which are already used for several diseases, including neurodegenerative disorders, cardiovascular diseases, and cancer [3]. The use of methylation markers for disease detection has several advantages over mutation-based detection. Methylation is an early event in carcinogenesis, making it very interesting for early cancer detection [5,6]. Also, methylation signatures are more universal than mutation markers, which typically vary at a wide range of sites. Given that no prior knowledge is needed on the tumor molecular profile, methylation-based tests can be used off-the-shelf, making them much faster and cheaper to use [7,8]. Lastly, DNA methylation is specific to cell types, and as such it can be used to study the tissue of origin. This is of great importance when biomarkers are detected in tissue or fluids taken from another origin [9]. Methylation could potentially be used to monitor many common diseases with a simple cfDNA blood test by taking advantage of these tissue-specific differences [10]. With the increased interest in detecting methylation, technological advances have been enormous. Multiple methods are now available for methylation studies. However, most of the technologies have important issues such as high cost, inapplicability for some sample types, high DNA input requirements, complex (bioinformatic) analyses, and low sensitivity and/or specificity. These issues are even more pronounced in a liquid biopsy context, where (cf)DNA is often already fragmented and available in limited amounts [11]. Nevertheless, several studies on cfDNA methylation detection have been performed.
One of the main challenges in biomarker research is bringing a new biomarker to the market for use in clinical diagnostics. Since 2022, stricter regulations for In Vitro Diagnostics (IVD) have made this process more complex. Besides FDA-approved or European Certified In Vitro Diagnostic (CE-IVD) tests, laboratory-developed tests (LDTs) are also widely used. Unlike CE-IVD tests, which are commercially available and regulated for specific uses, LDTs are not marketed. Instead, LDTs are used in three different ways: as a test that is developed and manufactured by a specific lab for in-house use, as a CE-IVD test that is used off-label for a different purpose, or as a CE-IVD test that is modified by the lab for a particular use. LDTs are typically limited to the lab where they are developed and are exempt from FDA approval or CE marking. However, if they are later marketed, they must obtain FDA approval or CE marking [12–14]. In this review, we discuss several LDTs and CE-IVD tests that utilize DNA methylation biomarkers (see Table 1).
Over the years, many methylation detection techniques have been developed. This manuscript provides an overview of both classic and more recent techniques along with their (dis)advantages (see Table 1). We restrict this review to the most used technologies (see Figure 1). Furthermore, the current use of methylation biomarkers in liquid biopsies in the clinic is described.

Classic methylation detection techniques
DNA methylation plays a vital role in numerous biological processes and is a key focus in epigenomic research. Unlike genetic analyses that examine native DNA, the detection of DNA methylation typically involves the conversion of methylated cytosines. This is necessary because the polymerases used in PCR strategies cannot differentiate between methylated (5mC) and non-methylated (C) cytosine residues. To accurately distinguish between these two forms, prior modification of 5mc or C is essential, which can be achieved through various techniques. These can be grouped per methylation detection strategy: 1) bisulfite conversion-based methods, 2) restriction enzyme-based assays, and 3) affinity enrichment-based approaches. In recent years, direct detection of methylation has gained a lot of interest (Figure 1). This review gives an overview of classic and novel methylation detection technologies and their implementation in the clinic.

Bisulfite-based methods
Although bisulfite treatment is genome-wide, read-out of distinct methods can be used to determine the methylation status of a single, specific locus or a more general profile in a wider region. Therefore, bisulfite-based techniques are divided into two categories: target or locus-specific methods and genome-wide approaches [15,19,22,23,29,30,32,35,49,56,57]. All bisulfite-based methods share limitations inherent to bisulfite treatment. Bisulfite conversion is a harsh chemical method (high pH and temperature) that degrades and fragments DNA, resulting in a poor-quality product and loss of input DNA. This is mostly particularly a problem in situations where a limited amount of DNA is available [16,19,22,24,29,32,49,57]. Further, significant degradation of the bisulfite-treated DNA can occur upon storage, as the single-stranded DNA is unstable after conversion. Therefore, converted DNA must be analyzed shortly after conversion so as not to impair the sensitivity of downstream applications [29,58]. Furthermore, incomplete conversion of cytosine to uracil, for example, by re-annealing of ssDNA or by the presence of residual RNA or proteins, can introduce artifacts and lead to inconsistencies in results [16,19,22,29,56,58]. Although commercially available kits are highly efficient (>97% to 99%), the conversion rate can vary slightly between one another [22,30,57]. Moreover, bisulfite treatment cannot discriminate between 5mC and 5hmC, which could affect downstream analyses [19,22,30,58].

Target-specific bisulfite-based methods

(Targeted) bisulfite sequencing
Since the development of bisulfite sequencing (BS-seq) in 1992, many efforts have been made to make bisulfite conversion more time- and cost-effective and to lower its required DNA input, which was a major problem in the original protocol [40]. For example, Masser et al. developed the Bisulfite Amplicon Sequencing (BSAS) method. It combines elements such as the bisulfite treatment and region-specific PCR amplification with transposome-mediated next-generation sequencing (NGS) library construction and benchtop NGS. However, it is still bisulfite-based and expensive for multiplexing more than 20 genomic targets [24,59]. Other targeted BS-seq methods achieve enrichment of CpG-rich regions or other specific regions of interest (ROIs) through hybridization with specific oligo probes either before or after bisulfite treatment, followed by NGS [36,58]. Capture of the DNA before bisulfite conversion ensures better enrichment since the genome is not yet converted but requires more DNA. Hybridization of the probes after bisulfite conversion produces more off-target reads but results in higher complexity of the sequencing library [58]. Several kits for specific targets (disease panels) or regulatory regions are commercially available. Examples include the Agilent Sure-Select Methyl-Seq and TruSeq Methyl Capture panels (capture before bisulfite conversion) and the Roche SeqCap Epi (hybridization after bisulfite treatment) [24,36,58].
With the rise of interest in single-cell analysis around 2010, the development of methylation-specific single-cell analysis methods gained interest. Degradation due to bisulfite treatment and limited DNA input capacity is circumvented in single-cell BS-seq with post-bisulfite adaptor tagging (PBAT). This further allows PCR amplification and deep sequencing at the single-cell level, although PBAT comes with biases and the formation of chimeric reads [23,24]. Because of random fragmentation and random priming, a single-cell library covers only around 10–20% of the genome.
An application of BS-seq is the IvyGene Cancer Blood test (Laboratory for Advanced Medicine Inc., Irvine, CA, USA). This test can detect multiple cancer types, including breast, colon, lung, and liver cancer, simultaneously. It is currently only available as an LDT and targets the methylation status of the MYO1G and TNFAIP8L2 genes in cfDNA. As preliminary results show that a positive result can still indicate another cancer type, the current IvyGene test must be used in combination with other diagnostic tests [14].

Methylation-specific (q)PCR and related technologies
Methylation-Specific PCR (MSP) was described in 1999 but became somewhat obsolete with the introduction of qPCR. MSP and related technologies are described in the supplementary table S1.
To obtain quantitative information, real-time MSP (qMSP) protocols were developed [32]. An intercalating dye, such as SYBR green (MethylQuant), or methylation specific TaqMan probes can be used in the so-called MethyLight assay [29,56,57]. Furthermore, qMSP eliminated the use of gels, leading to its implementation in the clinic [22,29,30]. qMSP is 10 times more sensitive than MSP and can detect low frequencies of hypermethylated CpGs (0.01%) [17,19,22]. In both MSP and qMSP (−related) methods, two major disadvantages include the lack of high multiplexing capacity, reduced sensitivity in fragmented samples, and the lack of standardization amongst different labs [36,56]. Only the multiplex MethyLight assay can theoretically analyze multiple genes simultaneously [19], but in practice, no more than three targets are evaluated simultaneously [60–62]. More related technologies are described in Table 1 and Supplementary Table S1. Quite a few applications have been FDA approved, amongst them the Epi proColon 2.0 CE (Epigenomics AG, Berlin Germany). This blood-based test uses a MethyLight assay to detect methylated SEPT9 aiming to improve early diagnosis of colorectal cancer. Across several studies, this test has been shown to discriminate healthy controls and CRC patients with a sensitivity of 75–81% and a specificity of 96–99% [17]. Although it was approved in 2016, European guidelines still do not recommend this test to be used as a first choice in screening programs [14,20]. Cost-effectiveness analyses, further validation, and regulatory approval will be needed to implement the EpiProcolon in programs across Europe.

Pyrosequencing
Despite being an older method, pyrosequencing remains an important commercial technology, for which several kits, platforms, and software from different companies are available today [17,30,36]. The technique itself is described in detail elsewhere [17,19,22,29,30,56,57,63,64] and its (dis)advantages are listed in Table 1.
Perhaps one of its main commercial applications are the different kits for determining the methylation status of the MGMT gene, such as the Therascreen MGMT Pyro kit (Qiagen, Hilden, Germany). MGMT is a known biomarker for glioblastoma classification and treatment decision-making. Pyrosequencing exhibited the most reproducible results, hence the choice for this technology in the commercially available kits. All kits are commercialized by Qiagen, but different systems (e.g., the PyroMark Q24 for the Therascreen kit) are used with different kits, and the Therascreen is the only kit that obtained the CE-IVD mark. Using this kit, 4 CpG sites in the MGMT gene are analyzed in DNA obtained from either blood or formalin-fixed paraffin embedded (FFPE) samples [14,20].

Droplet digital PCR
Droplet digital PCR (ddPCR) was developed to increase the analytical sensitivity of PCR for detecting rare events, for example, methylation of a specific allele. Bisulfite-converted DNA is fragmented into thousands of droplets, allowing as many simultaneous PCR reactions. In theory, each droplet contains only one DNA template, which is negative or positive for the rare event. The fraction of positive droplets represents a rare event. The enormous sample partitioning allows quick amplification, and screening of separate DNA molecules with a limit of detection up to 0.001% and absolute quantification. However, as is also the case for qPCR, primer design is laborious and difficult. This is primarily because of the need for high CG content and the reduction in sequence complexity resulting from the conversion, which in turn elevates the risk of false priming [19,22,56]. Furthermore, the multiplex capacities for ddPCR are very limited, with at most five targets reported in one assay [65]. Lastly, it is important to note that, although ddPCR has superior detection sensitivity compared to (q)PCR, the instrument cost is much higher, and therefore it is less available in (smaller) clinical practice(s) than (q)PCR, which makes it less practical to use.
To further increase the sensitivity, Menschikowski et al. developed Optimized Bias-Based Pre-Amplification ddPCR (OBBPA ddPCR) where they reached analytical sensitivities of 0.0007%. However, the procedure cannot be performed in a close-tube system, increasing sample contamination risks. The technology is claimed to be especially suitable in liquid biopsies [27,28]. Although ddPCR is a very promising method, there are currently no ddPCR methylation-based applications FDA-approved for clinical use.

MIMIC
Schwalbe et al. recently developed the Minimal Methylation Classifier (MIMIC) for the assessment of 17 CpG sites simultaneously. Bisulfite converted DNA is subjected to a single base extension of probe oligonucleotides, followed by MALDI-TOF mass spectrometry. Important advantages over other technologies include low input DNA (<2ng) and accurate assessment of methylation level, which is inherent to mass spectrometry. However, this technology is very expensive [26]. More information is also provided in Table 1.

Genome-wide bisulfite-based methods

WGBS
Since the emergence of NGS technologies, combining bisulfite conversion with NGS has become the new gold standard for global methylation analysis. Using Whole Genome Bisulfite Sequencing (WGBS) analysis, information about the methylation status of every cytosine in the methylome can be obtained, including low-density regions. Relatively low coverage (5–10×) is often used for highly reproducible and accurate determination of methylomes. Although it is the most comprehensive methylation profiling technology, there are some disadvantages (see also Table 1). WGBS is relatively expensive and as most of the human methylome is not methylated, a lot of sequencing capacity and money is wasted [15,23,30–32,40,58]. However, sequencing costs have been decreasing over the years, making WGBS economically more feasible [32]. Furthermore, due to the reduction of genome complexity and the loss of sequence diversity after bisulfite conversion, bioinformatic analysis of the WGBS data is difficult [24,31,36,49]. Specific pipelines, such as Bismark, gemBS and Methylpy, have been designed to streamline the pre-processing of this data and to facilitate the highly complex WGBS alignment. Novel ones, such as Methyldackel and MethylStar, are still emerging [66]. However, bioinformatic expertise and computational resources remain necessary [31].

RRBS
Before the introduction of WGBS, Reduced Representation Bisulfite Sequencing (RRBS) was already used in epigenetic DNA analysis. For this technology, methylated regions are enriched using MspI digestion (at CCGG sites) and bisulfite conversion before NGS. Digestion and size selection (40–220 bp) allow for the assessment of a smaller fraction of the genome while still yielding representative results [24,32,40]. Around 85% of CpG islands are evaluated with this technology, which only represents <3% of the genome [23,24,32,36]. As such, the cost is drastically decreased compared to WGBS, while the sample throughput is increased [29,32,40]. Despite RRBS being more reproducible than affinity-based methods, for example, it still is less reproducible than WGBS and microarrays due to its use of enzymes, targeted nature, and fewer sequencing reads [29,30,67]. Moreover, CpG-rich regions are enriched, while distal regulatory elements and intergenic regions have relatively low coverage in RRBS as the CpG-containing recognition sites for MspI are limited [29,32,58]. To sequence CpG island shores, longer restriction fragments can be sequenced in a technique known as enhanced RRBS [24]. In recent years, single-cell RRBS has been developed to enable the use of this technology in cfDNA. DNA loss is avoided by integrating all key steps in a single-tube reaction [18,32]. However, since adaptor hybridization takes place before bisulfite conversion, a fraction of well-ligated fragments is destroyed by the bisulfite and is therefore unavailable for amplification in library preparation [25]. Both RRBS and WGBS protocols have been adapted for use in single-cell BS [25].

cfRRBS
Recently, RRBS was adapted for use in cfDNA by De Koker et al. [68]. They found an ingenious way to circumvent the problem of cfDNA fragmentation, which hampers RRBS applicability. With their cfRRBS protocol, only MspI generated fragments are amplified, while ‘off-target’ cfDNA fragments are degraded. This is possible due to the use of hairpin adaptors that specifically bind to the phosphorylated 5’-ends that are created at MspI cut sites. These circular fragments withstand the exonuclease treatment. Thereafter, classical RRBS libraries are generated using bisulfite conversion and PCR amplification [68]. Some interesting applications have already been published, where cfRRBS is used to develop a diagnostic test for Cancers of Unknown Primary (CUP) [69] and pediatric cancers [70–72]. It is a cost-effective, scalable method and thanks to the single-tube protocol, it allows the use of as little as 0.4–10 ng DNA input [68,69]. However, despite covering 3 million CpG sites, targets remain limited to MspI restriction sites, which might exclude interesting candidate biomarkers.

Microarrays
Throughout the last decades, Illumina has brought several microarray platforms to the market. The first-generation assays were replaced by the HumanMethylation450 BeadChip in 2011 [73]. It was used for large projects such as The Cancer Genome Atlas (TCGA) and the International Cancer Genome Consortium (ICGC). In 2016, the novel methylation EPIC microarray was released [37]. This assay interrogates over 850,000 preselected CpG sites, covering 90% of 450K sites and more sites within enhancer regions [30,32]. As of 2023, the Infinium Methylation EPIC v2.0 has replaced the former version. Compared to the v1, poor-performing probes have been removed and an additional 186,000 CpG sites have been added. They target enhancers, copy number variation (CNV) detection regions, and additional CTCF-binding sites. Moreover, >450 cancer driver mutations were added. Importantly, the v2.0 has been validated for use in FFPE material, which is often difficult to work with due to the crosslinking and subsequent DNA degradation that occurs during the preservation procedure [33,34]. All microarrays work with similar technology, described elsewhere [30,36]. Methylation microarrays are relatively low-cost techniques, give accurate measurements, and are suitable for large sample numbers [31,37]. However, cross-hybridization of probes, erroneous signals from single nucleotide polymorphisms (SNPs) and probe-specific dye biases lead to some bioinformatic challenges. Moreover, batch effects can occur if not handled properly. Also, a large input amount is still required (500–1 µg) [22,31,58].
Based on the HumanMethylation450 BeadChip, the EPICUP™ test (Ferrer, Spain) was developed for the biological defining of tissue of origin in CUP in FFPE or fresh frozen samples. By comparing methylation patterns of known primary cancers with CUP, the origin can be predicted with high sensitivity and specificity. The 450K chips are no longer manufactured, but recent studies show identical results with the Infinium EPIC microarrays. Despite the importance of a CUP test, the EPICUP™ test cannot be executed in all laboratories due to the technology behind it, which leads to rather long turnaround times (TATs) from test to result of 2 weeks [14,20,21].

ELSA-seq
Liang et al. recently described the ‘Enhanced Linear-Splinter Amplification Sequencing (ELSA-seq),’ an improved bisulfite-based method for cfDNA. With ELSA-seq, detection power is on the one hand improved by increasing the DNA template that can be effectively used. WGBS libraries are generated, and a dual-index system is used for multiplexing samples. On the other hand, detection power is increased by the use of machine learning. This high-resolution technology allows for very low input DNA (500 pg) and has very effective noise suppression, but the adaptor tagging step may lead to incomplete duplicate removal. This method has only been validated in lung cancer up until now, but it is already a promising technology for cfDNA applications [38]. The OverC™ test for multi-cancer detection was developed based on this technology. It received the FDA breakthrough device designation, but it is not described as an LDT nor is there FDA approval or CE-IVD for this test [21].

Hammer-seq
Among the most recent bisulfite-based methods, Ming et al. introduced ‘Hairpin-Assisted Mapping of Methylation of Replicated DNA Sequencing,’ hammer-seq in short in 2021. This technology combines 5-ethynyl-2-deoxyuridine (EdU labeling) of replicated DNA [74], biotin-streptavidin-based purification and whole-genome hairpin BS-seq technologies. Their method’s most unique feature is the simultaneous measurement of methylation on both parent and daughter strands. The technique is ideal for determining maintenance kinetics and de novo methylation events occurring during methylation maintenance. However, the technique only measures the difference between daughter and parent strands. Moreover, it requires around 100 µg genomic DNA, which renders it unfeasible for applications with lower input DNA for the moment [39].

Enzyme-based methods

Restriction enzymes
The use of restriction enzymes was the first approach to assess locus-specific methylation [3,15,36]. Two types of enzymes are mostly used. Methylation-Sensitive Restriction Enzymes (MSREs, e.g., HpaII) are inhibited by the presence of methylation at their recognition site, reflecting the distinct methylation status at a specific locus [15,32,40]. Methylation-Insensitive Restriction Enzymes (e.g., MspI) cleave the DNA regardless of the methylation status at the recognition sites [18,32]. To distinguish methylated and unmethylated CpGs, pairs of both enzymes with the same restriction site but different sensitivity to methylation status (isoschizomers) can be used [19,35]. Depending on the application and technology, MSREs with recognition sites between 4 to 8 bp have been used [75]. A great advantage is that, as opposed to bisulfite, these enzymes do not degrade DNA, allowing the use of lower DNA input amounts. Furthermore, the primer design for enzymatic methods is much easier primer [29,57]. It has also been described that restriction-based methods are superior to other enrichment technologies as a conversion method prior to NGS [36]. They are fast, specific, and easy to use [57]. However, only loci with the restriction site(s) of the enzyme(s) can be investigated, and upon incomplete digestion, false-positive results are a possibility [29,56] (see also Table 1).

MSRE-(q)PCR
MSRE-based methylation assays were traditionally used in combination with Southern blotting but then switched for PCR and later qPCR. With this, DNA input drastically decreased as well as the cost and simplicity.
Combining MSREs with qPCR was first done in 2005. Primers flanking the region of interest are used to analyze methylation. Methylation percentages are counted from the cycle threshold (Ct) values (= number of cycles that is needed to amplify enough DNA to be detected) that are measured for both digested and undigested control samples. Commercial kits have been developed to target multiple sites (e.g., OneStep qMethyl kit from Zymo Research). What complicates this approach is that at least two restriction sites are required to be inside the amplicon, so it is not possible to investigate one particular CpG site [29,36,56,57].
Based on MSRE-qPCR technology, the Bladder EpiCheck (Nucleix Ltd) is a CE-marked test available for the detection of non-muscle invasive bladder cancer (NMBIC) recurrence. Fifteen methylation biomarkers are investigated in DNA extracted from urine. Studies have shown that the Bladder EpiCheck can be combined with cytology to reduce the invasiveness of NMBIC follow-up [14,21].

DREAM
Digital Restriction Enzyme Analysis of Methylation, DREAM in short, makes use of SmaI and XmaI to create specific signatures from cleavage at (un)methylated CpG sites. SmaI is used for digesting unmethylated CpG sites. XmaI has the same recognition site but cuts only at methylated sites. The two enzymes are used sequentially, so XmaI digests the remaining sites that have been protected from SmaI. Only fragments with these distinct signatures are sequenced, and methylation levels are thus calculated. According to the authors, a total of 50,000 unique CpG sites are yielded with high coverage when sequencing 25 million reads per human DNA library. The background is reported to be less than 1%, making the technology suitable for low methylation level detection. However, 1 µg of high-quality DNA is needed, hampering its applicability in liquid biopsy research [41,42].

MED-seq
Methylated DNA sequencing, MED-seq in short, was published by Boers et al. in 2018. In this technique, the methylation-dependent restriction enzyme LpnPI is used. LpnPI cuts 16 bp downstream from (hydroxy)methylated CpGs, leading to fragments of ≥32 bp. This allows accurate identification of methylation. Although this method is relatively low-cost, simple and does not need much DNA input (10 ng), only about 50% of methylated CpGs genome-wide can be detected. Unmethylated regions cannot be identified. Therefore, there is no possibility to completely quantify the methylation levels by a ratio or percentage, making this method only semi-quantitative [43]. Importantly, the technology has been successfully used in cfDNA samples, but only when 1) digestion with LpnPI was complete, and when 2) there was sufficient library-prepped DNA available. Starting concentrations were therefore 10 ng. The MED-seq technology is described to be compatible with vacuum concentration, different blood collection tubes and cfDNA isolation methods [44].

EpiGScar
In 2021, EpiGScar was published by Niemöller et al. EpiGScar stands for Epigenomics and Genomics of Single cells analyzed by estriction and allows simultaneous analysis of methylation and genetic variants of the same cell at base pair resolution. Because of its single-tube workflow, contamination risks and, more importantly, DNA loss are reduced. The method is described at length in the paper of Niemöller et al. They use the commonly described MSRE HhaI in combination with other enzymes, resulting in intact (methylated) or scar-tagged HhaI (unmethylated) sites. The resulting PCR amplified library is sequenced using NGS. Despite its good genome coverage (1.69 million RE sites), this technique is quite expensive and laborious [45].

Improved methylation profiling using restriction enzymes and SmMIP sequencing (IMPRESS)
Very recently, a novel methylation detection method, called Improved Methylation Profiling using Restriction Enzymes and SmMIP Sequencing (IMPRESS) was developed [46]. For this technology, a specific combination of MSREs and single molecule Molecular Inversion Probes (smMIPs, explained in [76]), followed by NGS, was derived. Four MSREs (HinP1I, AciI, HpaII, and HpyCH4IV) are used to cover around 40% of the epigenome. The advantages of these MSREs include the reliability, simplicity, and cost-effectiveness of the digestion reaction, while the smMIPs allow for highly accurate capturing of the DNA sequence of interest [46]. However, IMPRESS cannot detect both unmethylated and methylated CpGs. Furthermore, it covers 40% of the (epi)genome, limiting the biomarkers that can be investigated using this technology. Due to its recent addition to the methylation detection field, its applicability in the clinic remains to be proven.

Other enzymes

TAPS
In the past decade, novel enzymes for methylation detection have gained attention. For example, Liu et al. described the novel TAPS, short for TET-Assisted Pyridine borane Sequencing in 2019 [47]. The Tet methylcytosine dioxygenase 1 (TET1) enzyme oxidizes (hydroxy)methylated cytosines (5(h)mC) to carboxyl-cytosines (5caC) or formyl-cytosines (5fC). The novelty of TAPS lies in the subsequent 5caC/5fC-to-T transition chemistry. Pyridine borane is used to reduce 5caC to dihydrouracil (DHU), which is in turn converted to thymine by PCR. As such, cytosine modifications can be detected. An advantage of the TAPS technology is the possibility to also detect hydroxy-methylated cytosines. For this, 5hmC is glucosylated to 5gmC before the TET1 oxidation [47–49]. Compared to WGBS, TAPS has the advantage that the four-base genome is preserved, which allows efficient alignment and primer design. However, borane-mediated conversion requires long incubation times under acidic conditions, although it is less destructive than bisulfite deamination [48].

EM-seq
New England Biolabs commercialized Enzymatic Methyl sequencing (EM-seq) in 2020 [50]. The complete method is purely based on enzymatic conversion. In a first conversion step, two sets of enzymes are used. DNA is treated with TET2 and/or T4-BGT. T4-BGT protects 5hmC, TET2 protects 5mC. Subsequently, APOBEC3A is used for the deamination of cytosines, but not the protected ones. PCR amplification allows distinction between unprotected C’s, that are converted to T’s, and protected C-derivates (read as C). EM-seq combines the oxidation and deamination reactions with NEBNext library preparation. It has been shown that inputs down to 10 ng can be effectively used. Furthermore, EM-seq does not cause DNA damage or DNA fragmentation. EM-seq can also be combined with long-read platforms such as Nanopore. However, EM-seq is still quite expensive [48–51]. It has also been demonstrated that the TET2-enzyme favors certain DNA motifs (e.g., A upstream of CG) due to its intrinsic sequence specificity. The CG flanking sequences affect the TET enzyme conformation, which influences the TET function [77,78]. This is an important issue that must be carefully evaluated, as it might hamper EM-seq’s clinical applications.

Affinity-based methods
Affinity-based methods make use of immunoprecipitation with antibodies specific for 5mC or affinity purification with methyl-CpG binding domain (MBD) proteins to enrich methylated regions for further analysis [29,32,56] (Figure 1, Table 1). Previously, combinations with microarray hybridization were mostly used, but a shift towards NGS is now observed [15]. Compared to bisulfite-based methods, the biggest advantages are the elimination of bisulfite treatment and the possibility to discriminate 5-hydroxymethylation from 5mC. Affinity-based methods do not require high-quality DNA [36], but their accuracy suffers compared to RRBS and WGBS. Their resolution is only around 100–300 bp, which does not allow for single CpG site studies, and is biased towards hypermethylated regions [18,32]. Moreover, absolute quantification is not possible using an affinity-based method [31]. Furthermore, the standard protocols require a large amount of DNA input, so optimization for cfDNA is necessary [18,32] (Table 1).

(cf)MeDIP
Methylated DNA Immunoprecipitation or MeDIP uses, antibodies against 5mC to isolate the DNA fragment containing this modification, independently of the surrounding DNA sequence [40,79]. After shearing and immunoprecipitation, the isolated regions are further investigated by qPCR, microarray, or NGS [40,54]. MeDIP works on single-stranded DNA (ssDNA), allowing the profiling of hemi-methylated sites [31]. In contrast to BS-seq, it is biased towards low-density CpG sites [79]. One of the most important limitations of this technology was the high DNA amount required. Recently, the protocol has been adapted by Shen et al. [52] for use in cfDNA. This cfMeDIP-seq technology makes use of exogenous lambda DNA as filler DNA to increase the initial DNA input. As such, cfDNA amounts of as little as 1 ng can be used. However, the technology has not yet been validated with a sufficient number of independent clinical samples.

Direct detection methods
Around 2010, Third-Generation Sequencing (TGS) and Multi-Walled Carbon Nano Tubes (MWCNTs) entered the market. These methods allow for native DNA sequencing based on electrochemical signals. TGS focuses on the current-based signals and kinetics [24], while MWCNTs detect the DNA nucleotides based on their unique oxidation signals [35,80]. These electrochemical technologies are base-specific and can be used for rapid detection of bases without using prior conversion steps [24,80]. The focus in recent research has mostly been on TGS, which will be discussed further below.

Third-generation/long-read sequencing
A few years after the introduction of TGS, also called long-read sequencing, applications for methylation detection with these platforms arose. Currently, two technologies are available: Oxford Nanopore Technologies (ONT) and Single Molecule Real-Time (SMRT) sequencing from Pacific Biosciences (PacBio). Although the first TGS analyses for methylation detection were performed after bisulfite conversion, the strength of these platforms lies in sequencing native DNA. Base modifications are derived based on raw signals that are generated. Despite the advantage of eliminating the conversion step, some limitations still need to be overcome before these technologies can be applied in the clinic. For example, DNA is not amplified, and thus typically requires a large input amount (1 µg for ONT, 5 µg for PacBio). Such amounts of DNA are often not available, namely in FFPE or cfDNA samples. However, one study does use ONT for DNA methylation detection in cfDNA extracted from 4 ml of plasma, indicating potential future use in liquid biopsies [81]. Furthermore, single-base pair calling is not very accurate in long-read sequencing. Also, long-read sequencing is even more expensive than NGS. Lastly, the cumbersome data output (‘big data’) of these technologies brings along bioinformatic challenges [24,54,82].

ONT sequencing
ONT-sequencing is based on measuring the ionic current changes of native single-stranded DNA that is passed through a nanopore. Every base and its modification give a unique signal that is analyzed using a trained artificial neural network for base calling, such as the main base callers Guppy and Bonito [31,82]. However, there is no standardized base calling pipeline and the error rate varies from 5% to 20%, depending on the sequencing context [29,31]. Nevertheless, the field is constantly evolving and newly developed base callers such as CATcaller, CausalCall, and Halcyon have improved performances compared to the original base callers [83]. In all, software is quickly evolving and continuously updated and with time, the error rate will improve.

SMRT sequencing
Single molecule real-time or SMRT sequencing from PacBio detects methylation by monitoring the polymerase kinetics, at high (150–250×) coverage [55,84]. The kinetics of the polymerase are followed as it synthesizes circular DNA double strands using different fluorescently labeled nucleotides. As such, both the nucleotide sequence and the major modifications such as 5(h)mC can be detected simultaneously [23,24,29,31]. Currently, base calling is done in CCS [85]. However, 5mC signals are relatively subtle, leading to lower detection sensitivities compared to, e.g., 6 mA modifications of bacterial genomes. Therefore, SMRT has only been used extensively to study bacterial genomes and bacterial methylations [23,24,29,31].

Future perspectives
The methylation detection field is rapidly evolving. Besides new methods, specifically designed for methylation analysis, some new applications have recently gained interest in epigenetic research after decades of genetic research. Well-established technologies such as liquid chromatography and mass-spectrometry, for example, can be used for methylation detection, but still require DNA conversion with one of the above-described methods, depending on downstream applications and the nature of the available sample [36,86–89] (Figure 1 and Table 1). A newer example is the DNA biosensor, the surface of which contains specific complementary probes that capture DNA. Signal transducers are then used to convert the recognition event into either an optical or an electrochemical signal. For the detection of methylated DNA, a conversion method (discussed above) is used first. A biosensor system provides high sensitivity and specificity, and low cost [35,90]. Despite their advantages, biosensor application on clinical samples is currently lacking, making it difficult to evaluate their implementation prospects in the clinic [29].
Methylation detection strategies vary given the research question. For now, bisulfite treatment remains the most used conversion method in methylation detection, either in a targeted (such as ddPCR) or a whole-genome method (including WGBS and microarrays). However, a shift towards enzymatic methods has been observed in recent years. For example, MED-seq, TAPS, and EM-seq have been developed, and the latter was even commercialized. We believe enzymatic conversion for methylation detection will become more important in the future, as it is especially useful for analyzing liquid biopsies that have been gaining more attention. Due to their relatively low cost and ease of implementation, these methods can become useful in clinical diagnostics. One example is the newly developed and validated IMPRESS assay which targets multiple sites and is an inexpensive technique when combining many samples but is pending further validation before being implemented in the clinic. Furthermore, direct detection methods such as ONT and SMRT-sequencing are relatively underexplored methods that could gain more interest in the future. Despite their great advantage of sequencing native DNA, improving error rates and potential for cfDNA sequencing, these technologies will remain challenging to implement in the clinic due to cost and logistics. Lastly, combining different ‘omics into one assay, for example, genomics and epigenomics, has been gaining attention in recent years. These multi-omics strategies require their own methods and will become important for future applications.

List of Abbreviations

BSASBisulfite Amplicon Sequencing
BS-seqBisulfite sequencing
CE-IVDEuropean Certified In Vitro Diagnostics
cfDNAcell-free DNA
(cf)RRBS(cell-free) Reduced Representation Bisulfite Sequencing
CGICpG island
CNVCopy number variation
CUPCancer of Unknown Primary
DHUDihydrouracil
DREAMDigital Restriction Enzyme Analysis of Methylation
EdU5-ethynyl-2-deoxyuridine
ELSA-seqEnhanced Linear Splinter Amplification sequencing
EM-seqEnzymatic Methyl sequencing
EpiGScarEpigenomics and Genomics of Single-cells analyzed by Restriction
FFPEFormalin-fixed parafin embedded
Hammer-seqHairpin-Assisted Mapping of Methylation of Replicated DNA Sequencing
ICGCInternational Cancer Genome Consortium
IMPRESSImproved Methylation Profiling using Restriction Enzymes and smMIP Sequencing
LDTLaboratory-developed test
MBDMethyl-CpG binding domain
MeDIPMethylated DNA Immunoprecipitation
MED-seqMethylated DNA sequencing
MIMICMinimal Methylation Classifier
MSPMethylation-Specific PCR
MSREMethylation-Sensitive Restriction Enzyme
MWCNTsMulti-Walled Carbon Nanotubes
NGSNext-Generation Sequencing
NMBICNon-muscle invasive bladder cancer
OBBPA ddPCROptimized Bias-Based Pre-Amplification ddPCR
ONTOxford Nanopore Technologies
PBATPost bisulfite adaptor tagging
ROIRegion of interest
smMIPSingle Molecule Molecular Inversion Probe
SMRTSingle mMolecule Real-Time
ssDNAsingle-stranded DNA
TAPSTET-Assisted Pyridine Borane Sequencing
TATTurnaround time
TCGAThe Cancer Genome Atlas
TETTet methylcytosine dioxygenase
TGSThird Generation Sequencing
TSSTranscription start site
WGBSWhole Genome Bisulfite Sequencing

Supplementary Material

Suppl table 1.docx