AAI Internal Wiki

Starting Dose Beyond MABEL: Immune-Activating Drugs and Population-Based Modeling

閱讀時間約 17 分鐘

Starting Dose Beyond MABEL: Immune-Activating Drugs and Population-Based Modeling

超越 MABEL 的起始劑量：免疫活化藥物與族群建模

English

MABEL — minimal anticipated biological effect level — was born in the aftermath of the TGN1412 catastrophe. In 2006, a healthy volunteer trial of a CD28 superagonist monoclonal antibody at 1/500th of the fully active animal dose produced life-threatening cytokine storms in all six recipients within minutes of infusion. The drug was safe in animals because primate CD28 receptors respond very differently to superagonism than human CD28 does in the presence of memory T-cells. The lesson was not that biologics are inherently dangerous, but that immunostimulatory biologics can have a pharmacological cliff between “undetectable in animals” and “catastrophic in humans” — a cliff that toxicological endpoints like NOAEL and HNSTD cannot see.

MABEL was the regulatory response: for any drug where the primary risk is immune activation rather than tissue toxicity, the starting dose should be anchored to the lowest concentration expected to produce detectable biological activity in the most sensitive, human-relevant experimental system. The intent is protective: make sure the first dose is far enough below pharmacological activity that immune-mediated toxicity is improbable, even if the precise shape of the human concentration-response curve is unknown.

But MABEL, implemented mechanically — “find the lowest detectable effect level and choose a dose even further below it” — can produce its own form of harm. For patients with advanced cancer who have exhausted other treatment options and enroll in a phase 1 trial expecting some probabilistic chance of benefit, a starting dose so conservative that it requires 10–12 cohorts to reach a pharmacologically plausible range is not neutral. It means many patients are treated at doses that cannot produce benefit, the trial takes years to generate interpretable data, and the drug’s development is slowed precisely when urgency is highest. The 2026 Regulatory Toxicology and Pharmacology commentary by Matsumoto and colleagues from the HESI Immuno-Safety Technical Committee frames this as a two-sided ethical problem: MABEL protects patients from over-dosing in the first cohort; but over-conservative MABEL creates its own harm through systematic under-dosing of a vulnerable population across many cohorts.

The “beyond MABEL” argument is not a call to abandon caution. It is a call to replace mechanical conservatism with informed conservatism: starting doses that integrate target biology, target expression profile across normal and tumor tissues, pharmacologically relevant in vitro or ex vivo assay systems, population variability in biological response, human PK projection, mechanism-specific risk factors (step-up dosing requirements, CRS risk windows, monitoring intensity), and the benefit-risk context of the specific patient population. When all of those inputs converge on a starting dose that is higher than a naive MABEL calculation would produce — but still clearly below the range of serious immune toxicity — the protocol should explain that reasoning explicitly rather than defaulting to the lowest possible number to avoid regulatory scrutiny.

The most methodologically instructive “beyond MABEL” case study is the bispecific T-cell engager (Bi-TCE) approach described by Liu and colleagues in Clinical Pharmacology & Therapeutics 2025. Traditional MABEL calculations for T-cell engagers have used in vitro assays with effector-to-target cell ratios far above what is physiologically plausible in the tumor microenvironment — typically 10:1 or higher. This produces very low MABEL concentrations, and consequently very low starting doses. Liu’s group modified the approach by using physiologically relevant effector-to-target ratios (closer to those observed in actual tumors) in the in vitro killing and cytokine release assays used to define MABEL. They also integrated target biology, indication-specific factors, nonclinical safety data, in vitro and in vivo pharmacology, and translational PK/PD modeling. The result was a starting dose approximately 10-fold higher than a conventional MABEL calculation would have produced — and that dose was confirmed to be safe and well-tolerated in the FIH study, while also eliminating at least two escalation cohorts that would otherwise have been needed. The efficiency gain translated directly to fewer patients being treated at clearly subtherapeutic doses.

The SNX281 STING agonist case (Yuan et al., British Journal of Clinical Pharmacology 2025) demonstrates how population-level uncertainty can be explicitly incorporated into MABEL derivation. SNX281 is a small-molecule immunostimulant that activates the STING pathway, inducing type I interferon responses and downstream innate immune activation. Because no animal species adequately models the human immune response to STING agonism, the investigators built a population concentration-response model from ex vivo whole-blood cytokine induction assay data across multiple human donors. Rather than using the average donor response — which would underestimate risk in the most sensitive individuals — they used the 10th percentile of the dose-response distribution for maximum interferon-beta induction as the MABEL concentration. This choice explicitly protects the most sensitive patients in the population, not just the average patient. The MABEL concentration was then projected to a human starting dose using cross-species allometric scaling for PK, and an additional safety factor was applied. When first-cohort PK was measured, predicted and observed concentrations agreed within a factor of two, and no dose-limiting toxicities occurred. The starting dose was a hypothesis that turned out to be correct.

The mRNA-4359 case (Greene et al., CPT: Pharmacometrics & Systems Pharmacology 2026) illustrates what happens when no adequate animal model exists at all. mRNA-4359 encodes IDO and PD-L1 antigens delivered in lipid nanoparticles as an mRNA cancer vaccine. The immunological mechanisms are too human-specific for animal NOAEL to be meaningful. The solution was to use published clinical immunogenicity data from IDO/PD-L1 peptide vaccine trials in melanoma patients to calibrate an immunodynamic model — effectively using existing human trial data as the “animal model.” The model predicted that 180 μg mRNA-4359 might produce T-cell responses comparable to 200 μg peptide vaccine, and factoring in mRNA delivery efficiency uncertainty, 100 μg was selected as the FIH starting dose. This approach turns the starting dose into a model-informed clinical hypothesis rather than a toxicological rule.

The Lefèvre et al. work on Roche CD3 T-cell bispecifics (AAPS Journal 2025) adds an important safety anchor for T-cell engager development: the EC50 from the most sensitive tumor cell killing in vitro assay. When the predicted clinical Cmax is less than approximately 1/25th of that EC50 — far enough below meaningful tumor killing activity to make cytokine release from T-cell activation unlikely — the first dose can be considered safely below the CRS risk zone. This is a pharmacologically anchored MABEL, grounded in the drug’s actual mechanism rather than in conservative safety factors applied to a number derived from a non-physiological assay condition. It also illustrates a general principle: the goal of MABEL is not to avoid all biological activity; it is to avoid uncontrolled biological activity that cannot be managed by the monitoring and intervention protocols in place.

For clinicians reviewing a FIH protocol for an immune-activating drug, the practical checklist should include: Is the MABEL based on a human-relevant assay (not a purely preclinical one)? Does the assay use physiologically appropriate conditions (effector-to-target ratios, cytokine milieu, protein binding)? Was population variability across donors characterized and used in the derivation? Is there a population PK/PD model linking the MABEL concentration to a predicted human dose range? Is the starting dose accompanied by step-up dosing provisions, cytokine monitoring, hospitalization for the first dose, and pre-defined stopping and review criteria? And — the crucial question from a Project Optimus perspective — does the escalation design allow the trial to reach pharmacologically plausible doses within a reasonable number of cohorts, or does the conservatism in the starting dose create a structural barrier to learning?

中文

MABEL——預期最低生物效應劑量——誕生於 TGN1412 災難之後。2006 年，一項 CD28 超激動劑單株抗體的健康志願者試驗，在完全活性動物劑量的 1/500 下，在輸注後幾分鐘內就在六位受試者中全部產生危及生命的細胞激素風暴。這個藥物在動物身上是安全的，因為靈長類 CD28 受體對超激動劑的反應與人類記憶 T 細胞存在的情況下的 CD28 反應非常不同。這個教訓不是生物製劑本質上危險，而是免疫刺激性生物製劑在「動物無法偵測」和「人類災難性」之間可能存在藥理學懸崖——而 NOAEL 和 HNSTD 等毒理終點無法看到這個懸崖。

MABEL 是監管回應：對於任何主要風險是免疫活化而非組織毒性的藥物，起始劑量應錨定到預期在最敏感的人體相關實驗系統中產生可偵測生物活性的最低濃度。其意圖是保護性的：確保第一劑遠低於藥理活性，使免疫介導毒性的可能性很低，即使人類濃度—反應曲線的精確形狀尚不清楚。

但機械地實施 MABEL——「找到最低可偵測效應水平，然後選擇更低的劑量」——可能產生其自身的危害。對於已耗盡其他治療選項、帶著對某種受益概率的期望加入第一期試驗的晚期癌症病人而言，需要 10–12 個世代才能達到藥理上合理範圍的保守起始劑量並不中立。這意味著許多病人在無法產生受益的劑量下接受治療，試驗需要數年才能生成可解讀的資料，而藥物開發在緊迫性最高時恰恰被拖慢。Matsumoto 等人 2026 年來自 HESI 免疫安全技術委員會的監管毒理與藥理評論，將此框架為雙面倫理問題：MABEL 保護病人免於第一個世代的過度給藥；但過度保守的 MABEL 通過在多個世代中對脆弱族群的系統性低劑量給藥，製造了其自身的危害。

「超越 MABEL」的論點不是呼籲放棄謹慎。它是呼籲用知情的謹慎取代機械的謹慎：整合靶點生物學、靶點在正常組織和腫瘤組織中的表現特性、藥理相關的體外或離體分析系統、生物反應的族群變異性、人體 PK 預測、機轉特異性風險因素（逐步升量要求、CRS 風險窗口、監測強度），以及特定病人族群的利益風險背景的起始劑量。當所有這些輸入匯聚於一個比樸素 MABEL 計算會產生的更高、但仍明顯低於嚴重免疫毒性範圍的起始劑量時，試驗方案應明確解釋這個推理，而不是默認選最低可能數字以避免監管審查。

最具方法論啟發性的「超越 MABEL」案例研究，是 Liu 等人在 Clinical Pharmacology & Therapeutics 2025 中描述的雙特異性 T 細胞接合器（Bi-TCE）方法。T 細胞接合器傳統 MABEL 計算使用了效應細胞與靶細胞比遠高於腫瘤微環境生理上合理水平的體外分析——通常是 10:1 或更高。這產生了非常低的 MABEL 濃度，以及相應非常低的起始劑量。Liu 的團隊通過使用生理上相關的效應細胞與靶細胞比（更接近實際腫瘤中觀察到的）修改了這個方法，用於定義 MABEL 的體外殺傷和細胞激素釋放分析。他們還整合了靶點生物學、適應症特異性因素、非臨床安全資料、體外和體內藥理學，以及轉譯 PK/PD 建模。結果是一個約比傳統 MABEL 計算高 10 倍的起始劑量——而在 FIH 研究中確認這個劑量是安全且耐受良好的，同時消除了否則需要的至少兩個升量世代。效率收益直接轉化為更少的病人在明顯低於治療劑量下接受治療。

SNX281 STING 促效劑案例（Yuan 等人，British Journal of Clinical Pharmacology 2025）演示了如何在 MABEL 推導中明確納入族群水平的不確定性。SNX281 是激活 STING 路徑的小分子免疫刺激劑，誘導 I 型干擾素反應和下游先天免疫活化。因為沒有動物物種能充分模擬人類對 STING 促激的免疫反應，研究者從多個人類供體的體外全血細胞激素誘導分析資料中建立了族群濃度—反應模型。他們使用最大干擾素-β 誘導的劑量—反應分布第 10 百分位數作為 MABEL 濃度，而不是使用平均供體反應——這明確保護人群中最敏感的病人，而不只是平均病人。MABEL 濃度然後使用跨物種 PK 異速生長縮放投射為人體起始劑量，並額外套用安全係數。當第一個世代的 PK 被測量時，預測與觀察到的濃度在兩倍以內吻合，且沒有發生劑量限制毒性。起始劑量是一個被證明是正確的假說。

mRNA-4359 案例（Greene 等人，CPT: Pharmacometrics & Systems Pharmacology 2026）說明了當根本沒有足夠動物模型時會發生什麼。解決方案是使用已發表的 IDO/PD-L1 胜肽疫苗試驗臨床免疫原性資料來校準免疫動力學模型——有效地使用現有人體試驗資料作為「動物模型」。模型預測 180 μg mRNA-4359 可能產生與 200 μg 胜肽疫苗相當的 T 細胞反應，考慮到 mRNA 遞送效率不確定性，選擇 100 μg 作為 FIH 起始劑量。這個方法將起始劑量轉化為模型告知的臨床假說，而非毒理規則。

對於審查免疫活化藥物 FIH 試驗方案的臨床醫師，實用清單應包括：MABEL 是否基於人體相關分析？分析是否使用生理上適當的條件？供體間的族群變異性是否被描述並用於推導？是否有族群 PK/PD 模型將 MABEL 濃度連結到預測的人體劑量範圍？起始劑量是否伴隨逐步升量條款、細胞激素監測、首劑住院，以及預先定義的停止和審查標準？以及——從 Project Optimus 角度的關鍵問題——升量設計是否允許試驗在合理的世代數內達到藥理上合理的劑量，還是起始劑量的保守性在學習上製造了結構性障礙？

Key Concepts | 核心概念

  • MABEL (Minimal Anticipated Biological Effect Level): starting dose anchored to the lowest concentration producing detectable biological effect in the most sensitive human-relevant assay. Originally developed post-TGN1412 for immunostimulatory biologics.
  • Population MABEL: MABEL derived using a concentration-response model fitted to multi-donor ex vivo data, allowing explicit characterization of inter-individual variability. Using the 10th percentile protects the most sensitive patients.
  • Physiologically relevant E:T ratio: effector-to-target cell ratio in in vitro assays that matches tumor microenvironment conditions (~0.1:1 to 1:1) rather than artificial high ratios (10:1), producing more accurate MABEL concentrations for T-cell engagers.
  • Immunodynamic modeling: use of immune activation kinetics data to inform starting dose when no adequate animal pharmacology model exists (e.g., mRNA vaccines with human-specific immune mechanisms).
  • Ex vivo whole-blood assay: cytokine induction measured in undiluted human blood, preserving plasma protein binding and cellular composition — more physiologically relevant than isolated cell assays for highly protein-bound drugs.
  • Step-up dosing: a protocol design in which the first dose is deliberately below the target dose, with subsequent intra-patient escalation across 1–3 cycles before reaching target dose. Used for T-cell engagers to habituate the immune response before full pharmacological engagement.
  • TGN1412: CD28 superagonist that caused severe CRS in all 6 healthy volunteers in 2006 at 1/500th of the no-effect animal dose; the event that drove modern MABEL-based starting dose frameworks.

Related Pages | 相關頁面

關係圖譜

反向連結