LINE-1 Retrotransposons as Cell-free DNA Biomarkers for Multi-Cancer Early Detection

Summary:

LINE-1 retrotransposons, comprising 17% of the genome, drive cancer instability through hypomethylation. The DIAMOND assay, targeting LINE-1 hypomethylation with bisulfite sequencing of cell free DNA, achieved AUCs of 88% to 100% across six cancer types, surpassing mutation-based diagnostics and suggesting utility in early cancer detection and management.

Main Text:

In this issue of Clinical Cancer Research, Michel and colleagues introduce the DIAMOND assay (Detection of Long Interspersed Nuclear Element Altered Methylation ON plasma DNA) (1), a novel circulating tumor DNA (ctDNA) detection approach targeting hypomethylation of LINE-1 retrotransposons (Figure 1). This nonlocalized strategy offers a universal ctDNA biomarker with impressive potential for multi-cancer early detection. LINE-1 elements, comprising approximately 17% of the human genome, are repetitive retrotransposons that play critical roles in genome structure and regulation (2). Hypomethylation of LINE-1 elements is a hallmark of genomic instability in cancer, enabling chromosomal rearrangements, generating chimeric transcripts, and disrupting normal gene regulation—all of which contribute to tumorigenesis. Despite their abundance and mechanistic significance, LINE-1 elements have been underutilized as biomarkers due in part to technical challenges in mapping and interpreting these highly repetitive sequences.

Figure 1:

The publication by Michel and colleagues (1) presents the DIAMOND assay, a non-invasive method for multi-cancer detection using LINE-1 hypomethylation of plasma cell-free (cf) DNA as a biomarker. Plasma cell-free DNA is treated with sodium bisulfite, prepared into libraries, and sequenced to analyze (with machine learning) methylation patterns at LINE-1 elements. This targeted approach enables accurate detection of multiple cancers across various tissue types, highlighting its potential as a universal multi-cancer early detection diagnostic tool. Biorender.com were used to create this figure. Created in BioRender. Chauhan, P. (2025) https://BioRender.com/v04a349

Indeed, the detection and interpretation of ctDNA has transformed oncology, providing critical tools for noninvasive cancer diagnosis, prognosis, and treatment monitoring. Advancements have largely been driven by technologies focused on increasing the signal-to-noise of small variants present in ctDNA. However, as ctDNA technology expands into applications such as early detection and minimal residual disease monitoring, even single copies of small variants may not be present in a reasonable volume of plasma. An alternative approach addressing this fundamental limitation is to detect signal broadly across coherent changes dispersed throughout the genome and epigenome of tumor tissue, in line with the seminal work by Michel and colleagues.

Michel et al. (1) designed a bisulfite sequencing assay with PCR primers targeting 30,000–40,000 primate-specific LINE-1 (L1PA) retrotransposons, encompassing approximately 100,000 CpG sites that are commonly hypomethylated in multiple cancer types. To overcome the challenge of aligning reads to inherently repetitive genomics elements, they instead generated de novo a reference of representative LINE-1 sequences from their sequenced amplicons. Cell-free DNA reads were aligned to this LINE-1-specific reference and distinguished using highly distinct 16N random nucleotide barcodes, enabling precise quantification of unique methylated target molecules from 30 high-quality CpG sites across seven probes. By combining possible methylation states across each probe, the authors identified 372 unique methylation haplotypes. These haplotypes served as inputs for statistical models built using a random forest classifier algorithm.

The DIAMOND assay achieved an area under the curve (AUC) of 88% to 100% across six cancer types, with a sensitivity of 70% (at 99% specificity) for stage I-II malignancy. As a standout, non-metastatic ovarian cancer detection reached an AUC of 96%, exemplifying the assay’s clinical promise. DIAMOND outperforms mutation-based diagnostics in cancers such as localized gastric and ovarian, where mutations frequencies are inherently low. Certain cancers, like those driven by LINE-1-mediated mutagenesis (e.g., colorectal and lung cancers), may uniquely benefit from DIAMOND’s approach. Moreover, subtype stratification opportunities, such as hormone receptor (+) versus HER2 (+) versus triple-negative breast cancer delineation, illustrate how methylation differences in cell-free DNA LINE-1 elements could inform therapeutic decision-making.

Previous retrotransposon-focused cancer biomarkers have explored various strategies to detect these repetitive sequences. ORF1p protein assays have focused on detecting retrotransposon-encoded proteins indicative of active retrotransposition (3). Other researchers have employed specific primers and unique molecular identifiers to target short interspersed nuclear elements (SINEs) also leveraging their genome-wide abundance. A de novo k-mer finding method, ARTEMIS (Analysis of RepeaT EleMents in dISease), recently showed quantification of retrotransposon abundance without the need for specific primers (4). DIAMOND uniquely integrates the strengths of reference-free alignment, identification of unique input molecules, and methylation status, all used together to allow broad LINE-1 hypomethylation detection from cell-free DNA for the first time.

The concept of cancer detection through non-mutational, broadly coherent ctDNA signals is no longer novel but has become foundational to clinical assays such as Galleri (5). Where the field continues to advance, however, is in the fascinating diversity of signal types that demonstrate clinical testing potential. In addition to retrotransposon assays like DIAMOND, recent non-mutational innovations include end motif analysis, which examines the sequence context of fragment termini; long-read fragmentomics, uncovering correlations to length changes that are undetectable in routine massively parallel sequencing; and inferred nucleosome positioning, which elucidates tumor chromatin architecture through the relative coverage of cell-free DNA, among many others (6). At another level, ctDNA assays often offer the opportunity for interrogation of multiple feature classes across the same assay platform. For example, DIAMOND incorporates copy number analysis of its LINE-1 targets, a method pioneered previously as mFast-SeqS (7).

This growing complexity of cell-free DNA biomarkers presents a significant challenge: how to integrate the multitude of features from each assay and across feature types. Each provides potentially unique diagnostic insights, yet features can often be confirmatory or redundant, reflecting the common underlying tumor DNA modification and processing. ctDNA features are commonly aggregated through machine learning models; DIAMOND uses combinations of random forest models. However, there are no universal solutions or best practices for selecting, combining, and optimizing these features into robust predictive frameworks. Progress in this area will require careful, data-driven optimization, but also deep knowledge of specific clinical contexts and the underlying genomics and cancer biology, as no one-size-fits-all strategy exists.

Non-mutational methods like DIAMOND are critical to expanding cell-free DNA based diagnostics beyond the limitations of ultra-low-level variant detection. While performance variability exists across cancer types, these methods are invaluable as part of a larger ctDNA detection armamentarium. Future efforts should prioritize integrating complementary modalities, enabling stepwise or comprehensive strategies for early cancer detection and monitoring. As a universal biomarker platform, DIAMOND holds the potential to redefine multi-cancer early detection and monitoring, heralding a new era in precision oncology.