Over the past decade, liquid biopsy has become a workhorse of modern clinical research, providing low-risk blood draws that can be repeated, scaled, and paired with outcomes in a way tissue rarely can. Yet across clinical oncology and, even more markedly, in autoimmune and chronic disease research, a gap remains. Researchers still depend on the mechanistic insight that tissue can deliver, but repeated biopsies are neither ethically defensible nor operationally feasible.
This tension has now been explicitly stated in policy. Published in January 2025, the joint draft guidance from the FDA and the Office for Human Research Protections (OHRP), “Considerations for Including Tissue Biopsies in Clinical Trials”, does not ban tissue biopsies but establishes clearer guardrails around their necessity, participant burden, and available alternatives. Sponsors must justify each biopsy, distinguish required from optional procedures, and demonstrate why less invasive alternatives are insufficient, especially when the biopsy serves exploratory goals rather than key endpoints. This mandate forces a new question for the field: if tissue biopsies are limited, can blood-based approaches truly deliver the mechanistic insight researchers need?
Yuval Dor, a leading expert on the applications of cell-free DNA (cfDNA) and methylation, recently published an article that highlights the current skepticism around these approaches. cfDNA offers a powerful window into tissue injury; however, it is a highly filtered signal because only a fraction of cellular turnover is detectable, and this fraction varies by cell type and context. Although the field may seem to have reached a ceiling, this limit is not necessarily biological.
Physical tissue access is invasive, episodic, localized, and biased
Tissue biopsies are often essential for eligibility, target confirmation, or key endpoints in clinical trials. However, tissue sampling is episodic and localized, providing a snapshot that is typically taken at baseline and rarely capturing the heterogeneity, dynamic responses, adaptation, and relapse occurring throughout the organ or joint over time.
In autoimmune diseases, where biology fluctuates and treatment decisions are made over time, the episodic nature of tissue sampling poses a significant scientific limitation. Furthermore, the organs most affected — such as the synovium in rheumatoid arthritis, the kidneys in lupus, and the gut in inflammatory bowel disease — are rarely sampled repeatedly in routine care. Conducting repeated research biopsies adds substantial patient burden and often limits trial enrollment.
The FDA’s guidance does not prohibit tissue collection. Instead, it requires that any planned biopsies be justified as necessary to achieve the trial objectives. The implication for sponsors is immediate: trial designs that depended on serial biopsies for mechanistic story-telling will need a new solution.
Blood is the obvious substitute as it is low risk, repeatable, and scalable. However, moving from tissue to blood introduces new challenges.
cfDNA workflows often obscure mechanistic detail
A major challenge is that current cfDNA isolation workflows often reduce complex biology to generic fragments. These workflows are optimized to recover nucleosomal DNA efficiently and reproducibly, as mononucleosomal cfDNA is abundant, stable, and well-suited to sequencing. However, when the goal is to gain insight into tissue-level mechanisms, these generic fragments can become a bottleneck.
This occurs because the most mechanistically informative regions of chromatin are not evenly represented. DNA wrapped around nucleosomes conveys one type of information, such as nucleosome spacing, chromatin compaction, and tissue-of-origin signals. In contrast, DNA protected by regulatory protein complexes — transcription factors, RNA polymerase, and enhancer machinery — carries insights into regulatory states and pathway activity. Standard isolation and library-preparation methods tend to under-sample, distort, or lose these rarer, more structured fragments.
Dor’s article emphasizes that cfDNA is a filtered readout, with only a fraction of cellular turnover detectable in plasma, and that fraction varies by cell type and context. Workflows that preferentially recover the most abundant, least state-specific fragments are inherently limited when addressing questions about regulatory or functional states.
Methylation is a powerful “who,” but a limited “what”
Another limitation is the one most discussed in the field, known as the methylation ceiling. Methylation-based deconvolution has been transformative because it answers a foundational question about the origin of the DNA in question. It is particularly useful for determining cell identity, as methylation marks encode lineage history and remain relatively stable.
However, in many chronic diseases, the key question is not which tissue contributed DNA, but what the tissue is doing and whether therapy is altering its activity. This is a question of cellular state, with answers often encoded in regulatory occupancy, enhancer activity, and chromatin architecture — features that can change rapidly.
Dor’s critique can be seen as a reminder that each feature class addresses a specific, bounded question. Methylation is not inappropriate but is primarily suited for determining cell identity. It is less effective for capturing pathway-level pharmacodynamics, immune activation states, or, in rheumatology, synovial program modulation — particularly in tissues where only a small fraction of cfDNA is measurable. The FDA guidance sets a clear standard, requiring that blood-based assays provide not only lineage information but also mechanistic insights capable of withstanding regulatory review.
Meeting FDA expectations without losing mechanistic rigor
The regulatory direction is clear, emphasizing the reduction of unnecessary invasive procedural risk while preserving interpretability. The scientific direction is equally clear, focusing on expanding the information obtainable from cfDNA. Recognizing the limitations outlined above points to a path forward.
- Tissue remains necessary, but less frequent. A limited number of carefully justified biopsies can anchor disease biology and establish reference standards while validating blood-based readouts. This aligns with the FDA’s intent to include tissue in trials where it provides unique insight, rather than as a default.
- Biochemistry must preserve informative fragments. Rather than relying on whatever cfDNA isolation yields, workflows should intentionally enrich or retain chromatin features associated with regulatory activity. Because state information is sparse, capturing it requires deliberate, optimized methods.
- Feature sets must go beyond methylation. Methylation measurement remains foundational for tissue mapping, but assessing cellular state requires orthogonal signals, such as chromatin occupancy, enhancer dynamics, and other regulatory signatures that convey what is happening now.
Viewed in this light, Dor’s perspective is not a warning about biology but a reminder about measurement. cfDNA is inherently filtered, and progress depends on selecting the right features and capturing them effectively. This represents an inflection point, with the apparent ceiling being technical rather than biological. Assay limitations can be overcome, and the next advances will come from breaking them.
