ctDNA as a Translational Endpoint in FIH Oncology Trials
ctDNA 作為 FIH 腫瘤試驗轉譯終點
English
First-in-human (FIH) oncology trials were built around a single organizing question: how much drug can a patient safely receive? Safety, pharmacokinetics, and dose-finding dominated the traditional phase 1 agenda, and tumor biopsies — which could have answered “is the drug doing what we think it’s doing?” — were often technically impractical in a heavily pretreated population. Circulating tumor DNA (ctDNA), the fragments of tumor-derived DNA shed into plasma, offers a window into tumor biology that blood can provide without a needle in the lesion. The emerging challenge is translating that window from an exploratory curiosity into a pre-specified, decision-informing endpoint built into the architecture of FIH trials.
The 2026 Cancer Cell commentary by Tan, Siu, Yap, and colleagues crystallized the modern framing: ctDNA is a multi-utility biomarker in early phase trials, not simply a companion diagnostic for patient selection. In one trial it can simultaneously detect targetable alterations that define eligibility, provide an early molecular response signal weeks before the first CT scan, monitor clonal evolution under treatment pressure, and anchor pharmacokinetic/pharmacodynamic (PK/PD) modeling. Each of these applications is distinct — and each demands a different level of analytical and clinical validation before it can legitimately inform decisions.
The 2025 perspective review in the Journal of Experimental & Clinical Cancer Research identified why ctDNA use in phase 1 has lagged behind its adoption in phase 2 and 3 trials: standardization. Assay variability across platforms, differences in tumor shedding by cancer type and anatomical site, the confounding effect of clonal hematopoiesis (benign age-related mutations in blood cells that can mimic tumor-derived signals), and slow turnaround times relative to cohort decision cycles all create friction. The review proposed that phase 1 does not necessarily require full mutational profiling of every variant; low-pass whole-genome sequencing, fragmentomics, and methylation-based methods can track tumor fraction economically and repeatedly — which is precisely what dose-finding requires.
A parallel Expert Review of Molecular Diagnostics 2025 article set the clinical floor: ctDNA decline must not be automatically equated with radiographic response or survival benefit. The journal sets this out not to dismiss ctDNA but to prevent its premature promotion into surrogate endpoints it has not yet been validated to fill. The correct clinical posture is to read ctDNA kinetics alongside PK exposure, imaging, symptom burden, and dose modification data — a multi-axis integration rather than a single metric verdict.
The FDA formalized this caution in its 2024 public workshop on ctDNA as an early endpoint for accelerating drug development. The agency’s questions were pointed: What constitutes a predefined molecular response threshold? How should low tumor fraction at baseline be handled — excluded or modeled? What is the minimum follow-up needed to establish a meaningful relationship between early ctDNA change and clinical outcomes like progression-free or overall survival? These are not rhetorical questions; they are the conditions under which ctDNA can graduate from exploratory translational endpoint to a regulatory-grade signal.
The Liquid Biopsy Response Evaluation Criteria in Solid Tumors (LB-RECIST) framework proposed in Annals of Oncology 2024 represents the most ambitious attempt to create a ctDNA analog of RECIST for systematic response assessment. The pedagogical message is not that clinicians should immediately replace imaging with blood draws, but that the field is assembling the vocabulary — ctDNA clearance, percent decline, molecular progressive disease, time to molecular progression — needed to make ctDNA a consistent language across trials. Without that vocabulary, results from different trials using different assays, different sampling schedules, and different response definitions cannot be aggregated into usable evidence.
A concrete illustration of what ctDNA integration can look like in practice comes from the TRESR camonsertib study (JNCI 2024). Camonsertib is an ATR inhibitor developed for tumors with DNA damage response gene deficiencies. Its principal on-target toxicity — anemia — tends to appear after the conventional 21-day DLT observation window, making traditional dose-limiting toxicity criteria unreliable guides for regimen selection. The investigators compared three dosing schedules and found that an intermittent regimen (160 mg once daily, three days per week, with a one-week rest every three weeks) substantially reduced grade 3 anemia, transfusion requirements, and dose reductions, without sacrificing antitumor activity as measured in part by ctDNA-based molecular response. This is the Project Optimus argument made concrete: the optimized regimen came from schedule modification rather than dose reduction, and ctDNA provided early biological evidence that the modified regimen retained target engagement.
For a clinician reading a phase 1 paper, the practical checklist is four questions. First: at what role is ctDNA positioned — patient eligibility, early molecular response, pharmacodynamic readout, or resistance profiling? Second: is baseline ctDNA detectable at a meaningful tumor fraction in this cancer type and patient population? Third: is the sampling schedule aligned with dose escalation decisions, or is it a retrospective correlative collected at fixed cycle intervals? Fourth: is there a pre-specified molecular response definition, and does ctDNA change actually feed into cohort expansion or escalation decisions, or does it appear only in the exploratory appendix?
The answer to that fourth question distinguishes a trial that treats ctDNA as a decision tool from one that treats it as a publication artifact.
中文
首次人體試驗（FIH）從誕生起就圍繞一個核心問題：病人能承受多少劑量？安全性、藥物動力學（PK）與劑量探索主導了傳統第一期試驗的設計，而腫瘤組織切片——本可回答「藥物是否真的在做我們預期的事」——在重度預治療的病人身上常難以反覆執行。循環腫瘤 DNA（ctDNA），即腫瘤細胞釋放到血漿中的 DNA 片段，提供了一扇可用血液觀察腫瘤生物學的窗口。現在的挑戰是：如何把這扇窗口從「探索性好奇心」轉化為「預先設計、可影響決策」的終點，真正嵌進 FIH 試驗的骨架中。
2026 年 Tan、Siu、Yap 等人在 Cancer Cell 的評論文章確立了現代框架：ctDNA 是早期試驗的多功能生物標記，不只是伴隨診斷工具。在同一個試驗中，ctDNA 可以同時偵測決定收案資格的可治療基因改變、在第一次電腦斷層幾週前提供早期分子反應訊號、監測腫瘤在治療壓力下的克隆演化，並為 PK/PD 建模提供錨點。每一種用途都不同，每一種也需要不同程度的分析與臨床驗證，才能合理地影響決策。
2025 年 Journal of Experimental & Clinical Cancer Research 的前瞻性評論解釋了 ctDNA 在第一期試驗落後於第二、三期的原因：標準化問題。平台間的分析差異、不同癌種與解剖部位的 DNA 釋放差異、克隆性造血（血液細胞中的良性年齡相關突變，可能混淆腫瘤訊號），以及相對於世代決策週期而言偏慢的報告速度，都構成阻礙。評論指出，第一期試驗未必需要對每個變異進行完整突變分析；低覆蓋率全基因體定序、片段體學與甲基化方法可以低成本、反覆追蹤腫瘤分率——而這正是劑量探索所需要的。
2025 年 Expert Review of Molecular Diagnostics 設定了臨床底線：ctDNA 下降絕不能自動等同於影像反應或存活利益。這不是要否定 ctDNA，而是防止它過早被提升為尚未驗證的替代終點。正確的臨床姿態是將 ctDNA 動態與 PK 暴露、影像、症狀負荷和劑量調整資料一起解讀——多軸整合，而非單一指標裁決。
FDA 在 2024 年 ctDNA 作為早期終點的公開研討會中正式了這個謹慎態度。監管機構提出的問題很具體：什麼構成預先定義的分子反應門檻？基線腫瘤分率很低時如何處理？早期 ctDNA 變化與臨床結果建立有意義關聯所需的最短追蹤期是多少？這些不是修辭問題，而是 ctDNA 從探索性終點畢業成監管認可訊號的條件。
2024 年 Annals of Oncology 提出的液態切片實體腫瘤反應評估準則（LB-RECIST）是最具野心的嘗試，要為 ctDNA 建立類似 RECIST 的系統性反應評估框架。其教學意義不是要臨床醫師立刻以抽血取代影像，而是提示這個領域正在建立一套詞彙——ctDNA 清除、下降百分比、分子進展性疾病、分子惡化時間——讓 ctDNA 成為不同試驗間可共通的語言。沒有這套語言，使用不同平台、不同採血時程、不同反應定義的試驗結果就無法整合成可用的證據。
TRESR camonsertib 研究（JNCI 2024）提供了 ctDNA 整合的具體範例。Camonsertib 是針對 DNA 損傷修復基因缺陷腫瘤開發的 ATR 抑制劑，其主要靶點毒性——貧血——常在傳統 21 天劑量限制毒性（DLT）觀察窗結束後才出現。研究者比較三種給藥方案，發現間歇性方案（每天 160 mg、每週三天、每三週中休一週）大幅降低了 3 度以上貧血、輸血需求和劑量減量，且不犧牲以 ctDNA 分子反應評估的抗腫瘤活性。這就是 Project Optimus 論點的具體化：最佳化方案來自時程調整而非單純降劑量，ctDNA 則提供了早期生物學證據，確認調整後方案仍保留靶點接合（target engagement）能力。
對閱讀第一期試驗論文的臨床醫師，實用清單包含四個問題。第一：ctDNA 被放在哪個角色——病人收案資格、早期分子反應、藥效讀出，還是抗藥性分析？第二：在這個癌種與病人族群中，基線 ctDNA 是否在有意義的腫瘤分率下可偵測？第三：採血時程是否與劑量升量決策對齊，還是只在固定週期收樣？第四：是否有預先定義的分子反應標準，且 ctDNA 變化是否真的影響世代擴增或升量決策，還是只出現在事後的探索性附錄中？
第四個問題的答案，區分了把 ctDNA 當成決策工具的試驗，與把它當成論文美化素材的試驗。
Key Concepts | 核心概念
- ctDNA (circulating tumor DNA): fragments of tumor-derived DNA detectable in plasma via liquid biopsy. Shedding varies by tumor type, anatomical site, and disease burden.
- Tumor fraction: the proportion of cell-free DNA in plasma that is tumor-derived. Low tumor fraction limits ctDNA assay interpretability, especially in brain metastases, peritoneal disease, and low-volume tumors.
- Molecular response: a pre-specified decline in ctDNA level (e.g., >50% decrease from baseline) used as an early pharmacodynamic readout. Must be distinguished from radiographic response.
- LB-RECIST (Liquid Biopsy RECIST): a proposed standardized framework for ctDNA-based response evaluation analogous to imaging RECIST criteria.
- Proof-of-mechanism: demonstration that the drug is engaging its intended biological target in patients. ctDNA kinetics linked to driver mutation allele fraction is one approach.
- Multi-utility biomarker: a single measurement modality (here, ctDNA) that can serve distinct functions — eligibility, early response, PD readout, resistance monitoring — each requiring separate validation.
Related Pages | 相關頁面
- ctdna-informed-dose-finding — How ctDNA dynamics can enter escalation decision frameworks
- ctdna-participant-selection-prognosis — ctDNA as a prognostic tool for patient fitness in FIH trials
- sy-5007-ret-inhibitor-case-study — A concrete FIH example where ctDNA tracked both response and resistance
- fih-starting-dose-noael-mabel-framework — The broader context of biological evidence in FIH dose rationale
- immune-engager-fih-rp2d-evidence-chain — How receptor occupancy and PD biopsies complement ctDNA when MTD is not reached