The target of these assays is DNA methylation, the addition of a methyl group to cytosine bases at CpG sites. Cancer cells carry characteristic methylation patterns that recur across promoters, enhancers, and developmental regulatory regions, and these patterns can even hint at the tissue of origin. Circulating cell-free DNA, or cfDNA, shed into plasma by tumors and healthy tissues alike, therefore offers a rich epigenetic signal that complements mutation-based liquid biopsies. Large sequencing studies have already shown that plasma methylation mapping can support tissue deconvolution and cancer classification, and clinical assays targeting markers such as SEPT9, SHOX2, SDC2, and BCAT1/IKZF1 have entered colorectal screening, lung cancer detection, and liver cancer surveillance.

The central technical obstacle is that CRISPR-associated proteins are programmed almost entirely by sequence and structure, whereas 5-methylcytosine, the methylated form of cytosine, generally preserves normal Watson-Crick base pairing. In other words, a Cas protein cannot inherently tell a methylated cytosine from an unmethylated one. Cas12a’s famous collateral cleavage activity provides enormous signal amplification, but amplification alone does not confer methylation sensitivity. The review organizes the field’s solutions into four discrimination routes: sequence conversion, enzymatic gating, affinity-assisted recognition, and native-state Cas modulation, with an additional layer of engineered molecular transduction that recodes already-discriminated signals into artificial PAMs, toeholds, or single-stranded DNA triggers.

Sequence conversion, the first route, recodes the epigenetic state as a sequence difference. Bisulfite chemistry converts unmethylated cytosines to uracil while methylated cytosines remain read as cytosine, creating a difference that guide RNA design can exploit. Platforms such as HOLMESv2 paired this logic with Cas12b for quantitative readout, while meHOLMES uses TET2 and APOBEC conversion before PCR and Cas12a detection with engineered PAM logic. The catch is that bisulfite treatment degrades DNA and reduces library complexity, a serious problem when starting material is scarce, short, and fragmented. A recent head-to-head comparison of enzymatic and bisulfite conversion kits on healthy-donor cfDNA and colorectal cancer plasma found that enzymatic methods preserved longer fragments but yielded less recoverable DNA, underscoring that the preferred method depends entirely on whether the endpoint is fragment preservation or template yield.

Enzymatic gating, the second route, has the broadest analytical evidence base among the studies reviewed. Here, a methylation-sensitive restriction enzyme or methylation-dependent nuclease converts methylation state into fragment integrity or an amplifiable end, and Cas then recognizes the surviving or newly generated sequence. One system, DESCS, used BstUI and HhaI gating before recombinase polymerase amplification, T7 transcription, and Cas13a readout, reporting a limit of detection of 86.4 attomolar and discrimination of methylation fractions down to 0.01 percent in controlled mixtures. A nanoporous-gold electrochemical assay reported detection limits of 3.4 to 5.2 femtomolar and 1 percent methylation-fraction discrimination in artificial plasma and blinded contrived specimens, though without patient-derived cfDNA validation. These impressive numbers characterize the complete gating-amplification-readout architecture, not any intrinsic methylation selectivity of Cas itself.

The review is blunt about the failure modes of gating designs. Incomplete digestion is the central vulnerability: in protection-based schemes, an unmethylated molecule that escapes its intended nuclease can become a false positive after high-gain amplification, while in methylation-dependent trigger designs, missed cleavage causes false negatives. Dual enzymes, protected primers, and sequential gates can reduce leakage but add components, reaction time, and lot-to-lot variability. Specimen evidence across Route II studies remains limited and heterogeneous, spanning cervical exfoliated-cell DNA, cell-line genomic DNA, serum-like matrices, and only small exploratory patient plasma cohorts. None has reached independent locked-assay validation.

The third route, affinity-assisted recognition, uses methyl-CpG-binding domain proteins or anti-5mC antibodies to capture modified DNA without conversion or cleavage. Non-CRISPR precedents exist, including MBD-enriched plasma cfDNA sequencing that derived a twelve-region colorectal cancer model, and a chemical-labeling affinity Cas12a assay for 5-hydroxymethylcytosine. But direct evidence for an end-to-end 5mC affinity-to-Cas assay in patient cancer cfDNA remains sparse. Capture is also not uniform: immunoprecipitation and MBD capture preferentially recover different CpG contexts, meaning enrichment factors measured on one substrate cannot be transferred to another, and the methylated fraction after capture need not equal the fraction in the starting sample.

The most mechanistically direct route, native-state Cas modulation, is also the least mature. In specific configurations, preserved 5mC itself alters Cas recognition or cleavage. The CRISPR-MeDNA platform reported site-, strand-, density-, and spacing-dependent inhibition of Cas12a, with effects varying among Cas12a orthologs, and ThermoCas9 experiments showed methylation-sensitive PAM recognition. In an exploratory patient comparison, CRISPR-MeDNA analyzed cfDNA from four milliliters of plasma, roughly two nanograms per reaction, across fifteen healthy donors, nine lung cancer patients, and ten pancreatic cancer patients, and observed group-level fluorescence differences after thirty minutes without nucleic-acid preamplification. Yet the study established no prespecified clinical threshold, no receiver-operating-characteristic performance, no independent validation, and no benign or inflammatory control cohort. Its reported one percent methylated-fraction result, obtained at one nanomolar total target DNA, must not be misread as a one percent clinical ctDNA detection limit.

Underlying all four routes is a finite molecular budget that the review formalizes with a simple Poisson bookkeeping model. The expected number of informative tumor-derived molecules in an aliquot depends on plasma volume, tumor fraction, the methylation state’s prevalence in tumor and background, the probability that a fragment retains all required CpGs, guides, PAMs, and enzyme sites intact, and end-to-end recovery. When the expected informative input approaches one molecule, Poisson sampling and upstream recovery dominate over reporter brightness. The probability of capturing at least one informative molecule is 63 percent at an expected count of one and 95 percent at three, and these are sampling probabilities, not diagnostic sensitivities. Most plasma cfDNA in healthy individuals derives from hematopoietic lineages, so even perfect molecular discrimination cannot rescue a biomarker frequently methylated in normal leukocytes or inflamed tissue.

The review’s translational verdict is measured. Analytical limits of detection established on synthetic targets or spiked matrices cannot be extrapolated to clinical sensitivity without accounting for input, recovery, fragmentation, background methylation, and decision thresholds. Prospective SEPT9 screening data showed substantially lower sensitivity, especially for stage I disease, than retrospective case-control studies had suggested, a cautionary benchmark for early CRISPR reports. The authors propose a two-axis evidence framework, running from synthetic targets to patient-derived cfDNA on one axis and from exploratory association to real-world implementation on the other, and find that no representative CRISPR methylation assay has yet reached independent locked-assay validation. Near-term priorities include harmonized reporting of limits of detection, methylation fractions, input volumes, and recovery; fragment-matched reference materials; and small, well-controlled feasibility studies with orthogonal comparators. Compact CRISPR panels may ultimately complement high-dimensional methylome sequencing for screening, residual disease monitoring, and high-risk surveillance, but only if they preserve enough clinically informative molecules to make the promise real.

Subject of Research: CRISPR-enabled biosensing of cancer-associated DNA methylation in circulating cell-free DNA

Article Title: CRISPR-enabled biosensing of cancer-associated DNA methylation in cell-free DNA: molecular architectures, molecular sampling constraints, and translational biosensor engineering

Article References: Luo, H., Huang, X., Yu, Q., Li, Q., & Zha, H. (2026). CRISPR-enabled biosensing of cancer-associated DNA methylation in cell-free DNA: molecular architectures, molecular sampling constraints, and translational biosensor engineering. Frontiers in Bioengineering and Biotechnology, 14, Article 1986098. https://doi.org/10.3389/fbioe.2026.1986098

Image Credits: AI Generated

DOI: 10.3389/fbioe.2026.1986098

Keywords: CRISPR, DNA methylation, cell-free DNA, liquid biopsy, biosensors, Cas12a, cancer diagnostics, molecular residual disease, 5-methylcytosine, limit of detection, clinical validation, point-of-care