The Liquid Biopsy Frontier: Unlocking the Promise of cfRNA for Early Cancer Detection

Graphical abstract

The unique allure of cfRNA in oncology

The fundamental premise of liquid biopsy is the notion that tumors shed their molecular essence into the bloodstream. Cell-free RNA (cfRNA), comprising messenger RNAs (mRNAs), microRNAs (miRNAs), long non-coding RNAs (lncRNAs), and circular RNAs (circRNAs), carries rich phenotypic information. As RNA levels dynamically reflect transcriptional activity, cfRNA can reveal active pathological processes—such as angiogenesis, immune evasion, and metabolic reprogramming—often before irreversible genomic mutations accumulate. Furthermore, distinct RNA transcripts and splicing isoforms possess tissue- and cell-type-specific signatures, allowing for the precise deconvolution of the tumor’s cellular origin.

The cfRNA liquid biopsy workflow

A typical cfRNA-based liquid biopsy workflow consists of four interconnected stages, each with critical bottlenecks (Figure 1): (1) Blood collection—peripheral blood is drawn into specialized tubes containing RNA stabilizers to inhibit ribonucleases (RNase) activity; improper handling or delays in processing can introduce substantial pre-analytical variability. (2) Plasma separation and cfRNA extraction—plasma is isolated within 2–4 h of collection, and cfRNA is extracted using commercial kits (column-based or magnetic bead-based) or emerging methods such as metal-organic framework (MOF)-based enrichment; extraction efficiency and RNA integrity vary dramatically across protocols. (3) Library preparation and sequencing—RNA is reverse-transcribed, and sequencing libraries are constructed; traditional ligation-based methods bias against fragments with non-canonical termini, whereas newer methods such as SLiPiR-seq (terminal-modification-independent cfRNA sequencing) circumvent this limitation. (4) Bioinformatic analysis and interpretation—sequencing reads are aligned, normalized, and analyzed using machine learning or artificial intelligence (AI) models (e.g., Orion) to identify tumor-associated signals and deconvolve tissue of origin. Each of these stages introduces potential sources of technical noise that must be rigorously controlled for cfRNA to achieve clinical-grade reproducibility.

The signal-to-noise conundrum: a quantitative perspective

Any analysis of cfRNA as a cancer biomarker must confront the fundamental signal-to-noise problem inherent to circulating nucleic acid detection. In early-stage disease, the tumor-derived fraction (circulating tumor RNA and ctRNA) typically constitutes an extremely small fraction of total cfRNA, presenting a formidable detection challenge. For SLiPiR-seq, the enhanced capture of diverse RNA termini improves the lower limit of detection but does not eliminate the background. Wang et al.[1] demonstrated that their method could reliably detect tumor-associated transcripts when the tumor fraction exceeded approximately 0.05%, although below this threshold, classification accuracy degraded substantially. The MOF-based extraction method developed by Sun et al.[2] enriched low-abundance cfRNA by an estimated 5- to 10-fold compared to commercial kits, effectively reducing the minimum detectable tumor fraction to roughly 0.005%. For AI-based approaches such as the Orion model, the effective detection floor is not defined by a single molecular threshold but rather by the model’s capacity to learn subtle, high-dimensional patterns across thousands of orphan non-coding RNA (oncRNA) features; however, the reported 94% sensitivity in non-small cell lung cancer (NSCLC) was achieved in cohorts with a mean tumor fraction of approximately 0.08%[3]. Across all technologies, the signal-to-noise ratio remains the paramount biological constraint—one that no computational or chemical innovation can fully circumvent without parallel improvements in sample preparation, depth of sequencing, and pre-analytical quality control.

Current applications and recent breakthroughs (2023–2026)

Recent technological advances in 2023 to 2026 have significantly accelerated the translational potential of cfRNA, transitioning it from a mere conceptual biomarker to a viable clinical tool (Table 1).

Pan-cancer early detection and technological innovation

Building on this momentum, Nesselbush et al.[4] introduced RARE-seq (random priming and affinity capture of cfRNA fragments for enrichment analysis), a method approximately 50-fold more sensitive than standard RNA-seq for detecting tumor-derived cfRNA transcripts. Applied to a cohort of 437 plasma samples from 369 subjects across four cancer types (NSCLC, pancreatic, prostate, and liver cancer), RARE-seq achieved a limit of detection of 0.05% tumor fraction. In NSCLC, sensitivity ranged from 30% in stage I to 83% in stage IV at 95% specificity, representing a substantial improvement over conventional cfRNA detection methods. Notably, the study identified platelet contamination as a critical confounder in cfRNA analysis, underscoring the importance of pre-analytical standardization.

In parallel, Bao et al.[6] developed cfPeak, a computational framework for cfRNA fragmentomic analysis that characterizes fragmentation patterns of cell-free transcripts. Analogous to cell-free DNA (cfDNA) fragmentomics, cfRNA fragmentation signatures encode tissue-of-origin information that can be leveraged for cancer detection. Applied to colorectal cancer and oral cancer datasets, cfPeak successfully detected low-abundance tissue-derived cfRNA signals at tumor fractions as low as 0.5% (recall = 70%), opening a new analytical dimension previously unexplored in cfRNA liquid biopsy.

Advancing organ-specific diagnostics

Beyond pan-cancer applications, cfRNA has demonstrated exceptional utility in the early detection of specific malignancies:

•Gastric cancer (GC): A comprehensive systematic review and meta-analysis by Zhang et al.[7] evaluated 58 studies on cfRNA biomarkers for GC. The analysis revealed that individual cfRNA markers boasted a median sensitivity of 80% and specificity of 80%, outperforming traditional protein markers such as CA153, CA211, and CA50. Intriguingly, panels of cfRNA markers pushed the sensitivity up to 86%, underscoring the value of multi-analyte approaches in enhancing diagnostic robustness.

•Hepatocellular carcinoma (HCC): The stealthy nature of HCC often renders it undetectable until advanced stages. Sun et al.[2] recently pioneered a groundbreaking method for extracting circulating nucleic acids using an MOF. This MOF-based technique exhibited a vastly superior enrichment efficiency for low-abundance cfRNA compared to conventional commercial kits, effectively preventing RNA degradation. By integrating this extraction method with high-throughput sequencing, the researchers constructed a diagnostic model that achieved an impressive 90% accuracy in non-invasively diagnosing liver cancer.

•Lung cancer: Deep generative AI models are now being leveraged to analyze circulating “orphan” non-coding RNAs (oncRNAs)—a class of cancer-specific, actively secreted RNAs. Recent studies utilizing multi-task generative AI (such as the Orion model) on NSCLC patient cohorts have reported an outstanding overall sensitivity of 94% and specificity of 87% across various cancer stages, vastly outperforming traditional circulating tumor DNA (ctDNA)-based mutation detection in early-stage settings.

•Brain tumors: Extending cfRNA diagnostics to the central nervous system, Huang et al.[8] evaluated seven RNA biotypes (mRNA, lncRNA, miRNA, Piwi-interacting RNA (piRNA), tRNA-derived small RNA (tsRNA), small nuclear RNA (snRNA), and small nucleolar RNA (snoRNA)) in cerebrospinal fluid and plasma from brain tumor patients. Using the SLiPiR-seq platform, they achieved an area under the curve (AUC) of 1.0 for brain tumor detection (validation set) and 0.94 for glioma subtype classification (test set). A cfRNA-based risk score yielded a hazard ratio of 9.9, outperforming traditional clinical risk factors, and demonstrating the translational potential of cfRNA across diverse anatomical sites.

•Colorectal cancer (CRC): Beyond traditional transcriptomic profiling, Ju et al.[5] pioneered LIME-seq (low-input multiple methylation sequencing), a method capable of detecting RNA modifications (N¹-methyladenosine [m¹A], N¹-methylguanosine [m¹G], N³-methylcytidine [m³C], N², N²-dimethylguanosine [m²²G]) in microbial-derived cfRNA from less than 2 nanograms of plasma RNA. Applied to colorectal cancer screening, this microbiome-cfRNA modification strategy achieved robust discrimination between CRC patients and healthy controls, introducing a new layer of epitranscriptomic biomarker discovery.

cfRNA in context: comparison with ctDNA and protein biomarkers

To contextualize cfRNA within the broader liquid biopsy landscape, Table 2 provides a direct comparison of key performance dimensions across cfRNA, ctDNA, and protein biomarker modalities. It is critical to note that direct head-to-head comparisons from identical patient cohorts are exceedingly rare; the values presented below are drawn from representative studies and should be interpreted with caution. This absence of true comparative studies constitutes a critical knowledge gap that the field must urgently address.

Navigating the technical and pre-analytical bottlenecks

Despite these exhilarating advancements, the road to clinical implementation is obstructed by several deeply entrenched challenges. cfRNA is notoriously fragile; its widespread application is currently stifled by pre-analytical variabilities, technical limitations in extraction and sequencing, and a glaring lack of standardization.

The fragility and degradation dilemma

cfRNA exists in an extremely hostile environment. Blood is replete with ubiquitous RNases that rapidly degrade RNA molecules, meaning that cfRNA in circulation is highly fragmented. Compounding this issue, the concentration of tumor-derived cfRNA in early-stage patients is extraordinarily low, often drowning in a sea of background RNA released from normal physiological processes (e.g., hematopoiesis and platelet turnover). Hemolysis during blood collection, for instance, can artificially skew miRNA profiles, leading to false-positive signals that reflect blood cell disruption rather than tumor biology.

Extraction biases and sequencing blind spots

There is currently no “gold standard” for cfRNA isolation. Different commercial kits employ varying chemistries that preferentially capture certain RNA populations (e.g., small RNAs vs. long RNAs) while neglecting others. Consequently, a significant proportion of the transcriptome remains systematically overlooked. Furthermore, traditional RNA-seq library preparation relies heavily on ligation reactions that require specific RNA termini (e.g., 5’ monophosphate). Because cfRNA undergoes extensive fragmentation and harbors diverse, often non-canonical terminal modifications, a vast majority of these molecules are rendered “invisible” to standard sequencing pipelines. While newer methods such as SLiPiR-seq and MOF-based extractions are beginning to address these biases, they are yet to be universally adopted.

The reproducibility crisis and lack of standardization

Perhaps the most formidable bottleneck is the alarming lack of reproducibility across studies. External factors such as age, sex, inflammatory states, and even nutritional status can dramatically alter an individual’s baseline cfRNA profile. Moreover, the lack of standardized operating procedures (SOPs) for plasma processing, storage temperatures, and computational pipelines means that technical noise often swamps the true biological signal. For example, a rigorous multi-phase investigation into CRC biomarkers by Northrop-Albrecht et al.[9] found that while hundreds of differentially expressed exons could be identified in a discovery cohort, almost none held up under stringent targeted validation in independent patient sets. Such failures highlight the acute vulnerability of cfRNA biomarkers to overfitting and the dire need for multi-center, prospective validations. Importantly, the vast majority of cfRNA-based diagnostic signatures reported to date have not undergone independent multi-center validation, and those that have—such as the CRC exon panel examined by Northrop-Albrecht et al.—have largely failed to replicate. This systemic reproducibility deficit represents perhaps the single greatest barrier to clinical adoption.

Critically, the exRNAQC Consortium[10] conducted the largest systematic evaluation of pre-analytical variables for cfRNA profiling to date, analyzing 456 extracellular transcriptomes from 20 healthy donors. Their findings revealed that all five commercially available cfRNA preservation tubes failed to stabilize extracellular RNA, with significant concentration and transcriptome composition changes over time. In contrast, traditional ethylenediaminetetraacetic acid (EDTA) and citrate collection tubes provided superior cfRNA stability, provided that plasma separation was completed within 4 h of collection. The consortium also demonstrated that RNA purification method and collection tube type exhibit significant interaction effects, necessitating validated combinatorial protocols rather than independent optimization of individual variables. These evidence-based recommendations directly address the standardization gap and provide a practical framework for multi-center cfRNA studies.

This position is supported by emerging consensus guidelines: the Minimum Information for Studies of Extracellular RNA (MISE) standards (2023) establish essential reporting elements (Table 3); the National Cancer Institute (NCI) symposium on liquid biopsy harmonization (2024) explicitly calls for reference materials and benchmark datasets; and the First International cfRNA workshop consensus statement (2024) prioritized standardization of pre-analytical variables over algorithmic novelty. These convergent signals indicate that the next phase of the field will be defined by rigor, not scale.

Conclusion

cfRNA represents one of the most dynamic and information-rich biomarkers in the modern oncologist’s toolkit. The recent breakthroughs in extraction chemistries, sequencing protocols, and AI-driven analytics in 2023–2026 have firmly established cfRNA’s capability to detect cancers at incipient stages with high fidelity. However, the transition from promising academic research to routine clinical practice demands rigorous standardization of pre-analytical workflows, the establishment of universal reference materials, and large-scale prospective clinical trials.

Future perspectives: Toward a multi-omic era

To unlock the full clinical potential of cfRNA, the field must pivot toward integration and standardization. Relying on cfRNA in isolation may not be sufficient; however, combining cfRNA signatures with ctDNA mutational profiles and epigenetic markers creates a powerful synergistic effect. This multi-analyte, multi-omic approach can compensate for the weaknesses of individual biomarkers, drastically improving both sensitivity and specificity.

Furthermore, the integration of AI and machine learning models, such as the Orion AI platform, offers a potent solution to the high dimensionality and complexity of transcriptomic data. As cloud-computing pipelines become more sophisticated and standardized, they will enable the robust deconvolution of cellular origins and the identification of subtle tumor-specific signals amidst overwhelming background noise.

Emerging sequencing platforms are also expanding the cfRNA analytical landscape. Peddu et al.[11] demonstrated the feasibility of long-read nanopore sequencing for full-length cfRNA profiling in esophageal adenocarcinoma, discovering over 270,000 novel intergenic cell-free RNAs and identifying metabolic, signaling, and immune checkpoint pathways upregulated in both precancerous Barrett’s esophagus and frank carcinoma. This approach circumvents the fragmentation bias of short-read sequencing and opens new avenues for discovering intact transcript variants in circulation. Additionally, Karimzadeh et al.[12] introduced Exai-1, a multimodal transformer-based foundation model for cfRNA liquid biopsy, demonstrating strong performance in cancer detection through integrated sequence and expression feature learning.

While these computational advances continue to mature, a prospective multi-center observational study (NCT05833360) has been enrolling patients since July 2023 to validate oncRNA-based liquid biopsy across diverse cancer types and stages, representing a critical step toward clinical translation.

References

[1] Wang J, Huang J, Hu Y, et al. Terminal modifications independent cell-free RNA sequencing enables sensitive early cancer detection and classification. Nat Commun. 2024;15(1):156.

[2] Sun Y, Yu HX, Han SQ, et al. Method for the extraction of circulating nucleic acids based on MOF reveals cell-free RNA signatures in liver cancer. Natl Sci Rev. 2024;11:nwae022.

[3] Karimzadeh M, Momen-Roknabadi A, Cavazos TB, et al. Deep generative AI models analyzing circulating orphan non-coding RNAs enable detection of early-stage lung cancer. Nat Commun. 2024;15(1):10090.

[4] Nesselbush MC, Luca BA, Jeon YJ, et al. An ultrasensitive method for detection of cell-free RNA. Nature. 2025;641:759-768.

[5] Ju CW, Lyu R, Li H, et al. Modifications of microbiome-derived cell-free RNA in plasma discriminates colorectal cancer samples. Nat Biotechnol. 2026;44(5):752-758.

[6] Bao P, Wang T, Liu X, et al. cfPeak: a computational framework for cell-free RNA fragmentomic analysis. Genome Biol. 2025;26:119.

[7] Zhang Q, Du Z, Wang X, et al. Cell-free nucleic acid as promising diagnostic biomarkers for gastric cancer: a systematic review. J Cancer. 2024;15(10):2900-2912.

[8] Huang J, Zhou J, Wang J, et al. Diagnostic and prognostic potential of cell-free RNAs in cerebrospinal fluid and plasma for brain tumors. npj Precis Oncol. 2025;9(1):123.

[9] Northrop-Albrecht EJ, Wu CW, Berger CK, et al. An investigation of plasma cell-free RNA for the detection of colorectal cancer: from transcriptome marker selection to targeted validation. PLoS One. 2024;19(8):e0308711.

[10] exRNAQC Consortium. Blood collection tube and RNA purification method recommendations for extracellular RNA transcriptome profiling. Nat Commun. 2025;16:4513.

[11] Peddu V, Hill A, Velandi Maroli SL, et al. RNA liquid biopsy via nanopore sequencing for novel biomarker discovery and cancer early detection. bioRxiv. Preprint posted online July 2, 2025.

[12] Karimzadeh M, Sababi AM, Momen-Roknabadi A, et al. A multimodal cell-free RNA language model for liquid biopsy applications. Nat Mach Intell. 2025;7:1927-1938.

Rights & permissions

The Author(s) 2026. This article is published by Higher Education Press at journal.hep.com.cn.