07 — PK/PD Modeling, Human Dose Projection & Toxicity Evaluation

Overview

Quantitative pharmacology — PK/PD modeling, simulation, and dose projection — bridges non-clinical to clinical development for ADCs. Because ADCs generate multiple analytes with distinct PK behaviors and distinct drivers of efficacy vs. toxicity, the modeling framework is substantially more complex than for small molecules or naked antibodies. Regulatory agencies (FDA, EMA) increasingly expect mechanistic or semi-mechanistic PK/PD analyses at IND and NDA/BLA submission.


1. ADC PK Modeling Framework

1a. Multi-Analyte PK Structure

A minimally adequate ADC PK model must account for the distinct disposition of at least three analytes:

Dose (ADC) 
    │
    ▼
[TAb PK model]  ——  (deconjugation rate, kdc) ——→  [naked Ab accumulates]
    │
    ▼
[cAb PK model]  ——  (linker catabolism, kcat)  ——→  [conjugated payload released]
    │
    ▼
[Free payload PK model]  ←— (also from systemic deconjugation in plasma)

Common modeling approach: Fit TAb and cAb separately, with a first-order deconjugation parameter (kdc) relating cAb decline to TAb. Then fit free payload as a metabolite model driven by both (a) systemic ADC deconjugation and (b) intracellular processing with first-pass release.


1b. Two-Compartment Antibody PK Model

For the mAb backbone (TAb), a standard two-compartment model applies:

dA_central/dt = −(CL + Q) × C_central + Q × C_peripheral − kdc × A_central
dA_peripheral/dt = Q × (C_central − C_peripheral)

Where:

  • CL = total clearance (FcRn-mediated recycling + non-specific proteolysis + target-mediated)
  • Q = inter-compartmental clearance
  • kdc = first-order deconjugation rate (constant or DAR-dependent)
  • V_central, V_peripheral = volumes

Typical IgG1 parameters (human):

ParameterValueNotes
CL0.2–0.5 mL/h/kgIncludes FcRn recycling; longer for optimized Fc
V_central40–60 mL/kg~plasma volume (IgG1 poorly distributed)
V_ss80–130 mL/kgLimited tissue distribution due to size
14–21 daysVaries with target expression, ADA, TMDD

1c. DAR Kinetics (Deconjugation Modeling)

DAR declines over time as payload is lost from the antibody in circulation. Several modeling approaches:

Approach 1 — Empirical first-order:

cAb(t) = TAb(t) × DAR₀/DAR_max × exp(−kdc × t)

where kdc is estimated from the ratio of cAb vs. TAb over time. Simplest; adequate for cleavable linkers with relatively rapid deconjugation.

Approach 2 — DAR-distribution model (stochastic conjugation): Stochastic conjugation (lysine or cysteine) produces a mixture of DAR species (DAR0, 2, 4, 6, 8). Each DAR species has a different clearance rate (higher-DAR species often clear faster due to hydrophobicity → FcγR-mediated uptake). A mixture model tracks each DAR species separately:

∑ DAR_i × fraction_i(t) = observed average DAR(t)

where each fraction_i follows independent PK with different clearance rates. More mechanistic; useful when DAR-specific PK is important (e.g., DAR8 species of T-DXd clear faster than DAR4).

Approach 3 — Mechanistic linker stability model: For cleavable linkers, deconjugation is modeled as a rate function of linker hydrolysis and enzymatic cleavage kinetics in plasma:

kdc = k_hydrolysis × f(pH, temperature) + k_enzyme × [enzyme]

Most complex; rarely used in regulatory submissions but useful for linker optimization.


1d. Target-Mediated Drug Disposition (TMDD)

When the ADC target antigen is expressed on non-tumor cells or shed into circulation, TMDD significantly affects PK:

Full TMDD model (Mager-Jusko):

dC/dt = Rate_in − (CL/V) × C − kon × C × R + koff × RC
dR/dt = ksyn − kdeg × R − kon × C × R + koff × RC
dRC/dt = kon × C × R − (koff + kint) × RC

Where R = free receptor/antigen, RC = bound complex, kint = internalization rate.

Quasi-steady-state (QSS) approximation: Used when binding/unbinding is fast relative to PK:

RC = C × R_total / (KSS + C)

where KSS = (koff + kint)/kon

TMDD is clinically important for:

ADCTMDD driverClinical manifestation
Gemtuzumab ozogamicinCD33 on circulating AML blasts and monocytesDose-proportional PK only when blast burden reduced
T-DM1 (Kadcyla)Shed HER2 ECD (sHER2)Elevated sHER2 → faster ADC clearance; non-linear PK at low doses
Sacituzumab govitecanShed TROP-2Less pronounced; TROP-2 shedding rate lower than HER2
Inotuzumab ozogamicinCD22 on circulating B cellsDose-dependent PK; blast burden covariate in popPK

1e. Population PK (popPK) Models for ADCs

PopPK models characterize PK variability across patients and identify covariates explaining that variability. Submitted in NDA/BLA to support labeling and dose adjustment.

Covariates commonly evaluated in ADC popPK:

CovariateAnalyteExpected EffectClinical Guidance
Body weight / BSATAb, cAbPrimary driver of Vd and, to a lesser extent, CLSupports weight-based (mg/kg) vs. flat dosing decision
ADA statusTAb, cAbPositive ADA → elevated CL (immune complex clearance)Flag ADA+ subjects in PK analysis; may exclude from E-R if large impact
Baseline sHER2 (for HER2 ADCs)TAb, cAbHigh sHER2 → ~30–50% faster CL in some analysesCovariate in T-DM1 popPK; used to explain lower exposure in high-sHER2 patients
Hepatic function (ALT, bilirubin)Free payloadPayload CL reduced in hepatic impairment (CYP3A4/5 substrates)Dose reduction guidance for moderate/severe hepatic impairment
Renal function (GFR, CrCl)Free payloadMinor to moderate effect for most payloadsGenerally no dose adjustment needed unless payload is renally cleared
Tumor burden (size, lesion count)cAbHigh tumor burden → faster antigen-mediated clearanceDescriptive covariate; not always included in final model
Albumin (baseline)TAbLow albumin → FcRn competition → slightly altered antibody PKIncluded in some models as exploratory covariate

Software: NONMEM (most common regulatory submission tool), Monolix (SAEM algorithm), Phoenix NLME. Estimation methods: FOCE-I (NONMEM), SAEM. Diagnostics: VPC (Visual Predictive Check), GOF plots, bootstrap CIs.


2. PK/PD Modeling — Efficacy

2a. Tumor Growth Inhibition (TGI) Models

Non-clinical (xenograft): The standard PK/PD model for ADC efficacy in mouse xenograft:

dW/dt = (kg − kd × C_payload) × W

Where:

  • W = tumor volume (mm³)
  • kg = tumor growth rate constant
  • kd = drug-effect (kill) rate constant
  • C_payload = intracellular free payload concentration (estimated from plasma cAb via a cellular disposition submodel)

More mechanistic models include:

  • Two-compartment tumor: growing fraction + resting fraction (Simeoni model)
  • Transit compartment model for drug effect delay
  • Bystander effect: separate kill rate for antigen-negative neighbors driven by diffusible payload

Translational challenge: Mouse tumor xenografts have different antigen density, cathepsin B activity, and tumor architecture than human tumors. Allometric or empirical scaling must be applied when translating TGI parameters to clinical tumor size endpoints.

2b. Indirect Response Models (Clinical PD)

For cytostatic effects or biomarker-based PD:

dR/dt = kin × (1 − Imax × C/(IC₅₀ + C)) − kout × R

Where R = biomarker (tumor size, serum marker), kin/kout = synthesis/degradation, Imax = maximum inhibition, IC₅₀ = half-maximal concentration of the relevant PK driver.

2c. Exposure–Response (E-R) for Efficacy

Clinical E-R is used to:

  1. Confirm dose selection (is the chosen dose in the efficacious exposure range?)
  2. Support alternative dosing regimens (Q2W vs. Q3W; flat dose vs. weight-based)
  3. Characterize special populations

Typical E-R metrics for ADC efficacy:

Exposure MetricEfficacy EndpointADC Example
cAb AUC₀–τ (cycle 1)ORR, PFS, tumor shrinkageT-DM1, T-DXd
cAb CtroughDCR, depth of responseBrentuximab vedotin
Cumulative cAb AUCOS in some models

3. PK/PD Modeling — Toxicity

3a. Free Payload as Toxicity Driver

Off-target toxicity is primarily driven by free payload systemic exposure (not ADC exposure):

ToxicityPayloadExposure MetricEvidence
Peripheral sensory neuropathyMMAECumulative AUC (free MMAE)E-R in vedotin programs; threshold ~AUC > X µg·h/mL cumulative
Myelosuppression (neutropenia)MMAE, DXd, SN-38Cmax (free payload)Direct myeloid progenitor cytotoxicity
Diarrhea/GI toxicitySN-38AUC (free SN-38)Sacituzumab govitecan; SN-38 AUC in GI epithelium
ILD/pneumonitisDXdUnclear PK-driver; possibly lung tissue DXd AUCT-DXd clinical data; E-R not clearly established for Cmax or AUC
Hepatotoxicity (VOD/SOS)CalicheamicinCmax (free calicheamicin)Mylotarg; rapid DSB in hepatocytes after Cmax peak
Ocular toxicity (microcyst)MMAFCumulative dose/AUCBlenrep; MMAF accumulation in corneal epithelium
ThrombocytopeniaDM1 (catabolite)AUC (Lys-SMCC-DM1)T-DM1 mechanism: DM1 in MK lineage impairs platelet production

3b. Semi-Mechanistic Myelosuppression Models

The Friberg model (transit compartment model for neutrophil dynamics) is commonly applied to ADC-induced neutropenia:

dProl/dt = kin × (1 − E(t)) × Prol × (Circ₀/Circ)^γ − k_tr × Prol
dT₁/dt = k_tr × Prol − k_tr × T₁
dT₂/dt = k_tr × T₁ − k_tr × T₂
dT₃/dt = k_tr × T₂ − k_tr × T₃
dCirc/dt = k_tr × T₃ − k_circ × Circ

E(t) = Slope × C_payload(t)   [linear E-R for neutrophil inhibition]

Where Prol = proliferating cells, T₁–T₃ = transit/maturation compartments, Circ = circulating neutrophils, Circ₀ = baseline ANC, γ = feedback parameter, E = drug effect.

This model has been applied to:

  • Brentuximab vedotin (MMAE-driven neutropenia)
  • Sacituzumab govitecan (SN-38-driven neutropenia)
  • T-DXd (DXd-driven neutropenia)

3c. Platelet Model for T-DM1

T-DM1 causes thrombocytopenia by a unique mechanism: DM1 catabolite (Lys-SMCC-DM1) inhibits megakaryocyte (MK) proplatelet formation, not by killing MKs but by disrupting tubulin dynamics in MK proplatelet extensions. A modified transit model with an MK compartment has been published for T-DM1.


4. Human Dose Projection

4a. NOAEL → First-in-Human (FIH) Starting Dose

Standard regulatory approach (FDA Estimating the Maximum Safe Starting Dose guidance, 2005; EMEA MABEL guidance, 2007):

Step 1 — Identify NOAEL in most sensitive species (usually monkey for mAb/ADC):

  • NOAEL from GLP 4-week repeat-dose toxicology study in cynomolgus monkey
  • Units: mg/kg

Step 2 — Human Equivalent Dose (HED):

HED = NOAEL (animal) × (BW_animal / BW_human)^(1 − 0.75)
    = NOAEL × (BW_animal / BW_human)^0.25

For monkey-to-human: approximately divide NOAEL by ~1.8 (based on 5 kg monkey, 60 kg human on a BSA-normalized basis) — but for mAbs/ADCs, mg/kg scaling is preferred (not BSA scaling) because FcRn-mediated clearance scales more closely with body weight than BSA.

Step 3 — Apply safety factor:

  • Typical: HED / 10 = starting dose
  • For ADCs with novel or highly toxic payloads, safety factor may be 1/30 to 1/50

Step 4 — Confirm starting dose against MABEL (for immunomodulatory or high-potency agents):

  • MABEL = minimum dose expected to produce a measurable biological effect
  • For ADCs: MABEL often estimated from receptor occupancy at target IC₅₀ or from minimal TGI in xenograft models
  • Starting dose = lower of NOAEL/10 and MABEL-based dose

4b. Allometric Scaling of PK Parameters

For non-mAb components (free payload PK in non-clinical):

Simple allometry (power function):

CL = a × BW^0.75
Vd = b × BW^1.0

Where a and b are fitted from mouse, rat, monkey data. Projects human CL and Vd.

Fixed exponents (for mAb components):

  • Antibody CL: exponent 0.85–0.9 (empirically derived for IgG)
  • Antibody Vd: exponent ~1.0
  • Because FcRn-mediated recycling dominates mAb CL, neonatal FcRn expression level must be confirmed as a scaling anchor

Two-species approach (Obach & Reed-Hagen method): Use monkey and rat for two-point allometry when mouse/rat data are unreliable (due to anti-mouse cross-reactivity or TMDD in rodents from cross-reactive antigen).

4c. PK/PD-Driven Dose Projection

More sophisticated approach integrating non-clinical TGI data:

  1. Fit non-clinical TGI model (xenograft) → estimate Cdrug_threshold (minimum cAb or payload concentration for tumor stasis/regression)
  2. Scale non-clinical PK to human (allometry)
  3. Simulate human concentration-time profiles at various doses
  4. Identify dose that achieves human C_threshold ≥ threshold for ≥X% of dosing interval
  5. Cross-check against safety margin from NOAEL

This approach has been used for T-DXd dose selection (8 mg/kg Q3W selected in part because simulations showed PK above in vivo effective exposure for >21 days in most patients).


5. Toxicity Evaluation Framework

5a. Non-Clinical Safety Studies Supporting IND (ICH M3(R2) / ICH S9)

For anticancer ADCs (ICH S9 scope — serious/life-threatening indication):

StudySpeciesGLPKey Readouts
Single-dose (dose range finding)Rat + monkeyNoMTD, NOAEL, clinical signs, body weight
4-week repeat-dose + 4-week recoveryCynomolgus monkeyYes (pivotal)NOAEL, reversibility, TK (PK at tox doses)
4-week repeat-doseRat (if rat pharmacologically relevant)YesSpecies comparison; may be replaced by monkey if not relevant
Genetic toxicologyIn vitro ± in vivoYesPayload-driven genotoxicity (most cytotoxic payloads are clastogenic)
Safety pharmacologyCV, CNS, respiratoryYesmAb backbone safety; payload CNS effects (MMAE: neuropathy)

ICH S9 flexibility for oncology: Carcinogenicity and reproductive/developmental toxicology studies not required for advanced-cancer indications at IND stage (evaluated post-approval for adjuvant/curative settings per ICH S9).

5b. Toxicokinetics (TK)

TK samples collected alongside toxicology studies to characterize systemic exposure at each dose level:

  • Minimum TK design: Day 1 and last dosing day of each study phase; 5–8 time points per profile
  • Analytes: TAb (minimum), cAb, free payload
  • TK parameters: AUC₀–τ, Cmax, t½ (from terminal phase if sufficient points)
  • TK/PD correlation: NOAEL dose → TAb/cAb AUC at NOAEL = human exposure target for safety margin calculation

Safety margin calculation:

Safety margin = AUC (NOAEL in monkey) / AUC (human at proposed clinical dose)

Typically ≥3–10× margin required for oncology ADCs (lower acceptable margin than non-oncology drugs per ICH S9).

5c. Off-Target Tissue Distribution (Tissue Cross-Reactivity, TCR)

TCR studies use the clinical ADC antibody component (or the naked antibody) on a panel of human normal tissues (37 tissue panel per FDA) using IHC:

  • Identifies unexpected antigen expression in normal tissues
  • Guides monitoring for specific adverse events
  • Not predictive of all toxicity (Fc-mediated mechanisms not captured)

6. Regulatory Guidance Reference

6a. FDA Guidances

DocumentApplicabilityKey ADC-Specific Content
FDA Guidance: Clinical Pharmacology Considerations for Antibody-Drug Conjugates (Draft, 2022)Primary ADC-specific clinical pharm guidanceAnalyte selection, E-R recommendations, dose optimization, special populations
FDA Guidance: Population PK (2019)popPK methodologyModel building, covariate selection, VPC requirements for submissions
FDA Guidance: Estimating the Maximum Safe Starting Dose (2005)FIH starting doseNOAEL → HED conversion, safety factor selection
FDA Guidance: M3(R2) (2010)Non-clinical studies for INDStudy timing, GLP requirements
FDA Guidance: S9 Nonclinical Evaluation for Anticancer Pharmaceuticals (2010)Oncology NCI frameworkAbbreviated tox for advanced cancer; reproductive tox timing
FDA BioA Guidance (2018)BMV for all methodsLLOQ, accuracy, precision, stability — full validation
FDA Guidance: Immunogenicity Assessment for Therapeutic Protein Products (2019)ADA testingTiered testing, drug tolerance, clinical impact assessment

6b. EMA Guidances

DocumentApplicability
EMA Guideline on the Clinical Investigation of the Pharmacokinetics of Therapeutic Proteins (2007)Antibody PK design; TAb/cAb analyte selection
EMA Reflection Paper on ADCs (Draft, 2020; Final expected 2024–25)ADC-specific: analyte panel, linker stability studies, immunogenicity
EMA Guideline on BMV (2012 + ICH M10 implementation)Method validation harmonized with ICH M10
CHMP SWP/28367/07 — MABEL Guidance (2007)FIH starting dose using MABEL for high-risk biologics
EMA ICH S9 Implementation (EMA/CHMP/ICH/646107/2008)Non-clinical for oncology

6c. ICH Guidances

ICH DocumentTopic
ICH M10 (2022, Step 4)Bioanalytical Method Validation — harmonized global standard; LBA chapter
ICH M3(R2) (2009)Nonclinical Safety Studies Timing
ICH S9 (2009)Nonclinical Evaluation for Anticancer Drugs
ICH E8(R1) (2021)General Considerations for Clinical Studies
ICH E9(R1) (2019)Statistical Principles; estimands framework (relevant for OS/PFS endpoints)
ICH S6(R1) (2011)Preclinical Safety Evaluation of Biotechnology-Derived Pharmaceuticals
ICH Q8/Q9/Q10Pharmaceutical development, risk management, quality systems (manufacturing/CMC)

6d. Industry Consensus White Papers (Referenced in Regulatory Submissions)

DocumentContent
Kaur et al. (2013) AAPS J (AAPS ADC Working Group)Recommended analytes, assay platforms, validation approach
Gorovits et al. (2013) BioanalysisHybrid LBA-LC/MS/MS methodology consensus
AAPS/FDA ADC Bioanalysis Workshop Report (2020)Updated consensus: free payload stability, hybrid assay formats
Shankar et al. (2014) AAPS JADA testing strategy for ADCs — tiered algorithm
EBF (European Bioanalysis Forum) ADC recommendations (2019, 2022)European bioanalytical practice consensus

Key Papers

  • Gibiansky & Gibiansky (2014) J Pharmacokinet Pharmacodyn — ADC PK/PD modeling framework
  • Singh & Shah (2017) Drug Metab Dispos — allometric scaling of mAb PK
  • Scheuher et al. (2022) Clin Pharmacol Ther — T-DXd exposure-response in DESTINY-Breast trials
  • Friberg et al. (2002) J Clin Oncol — transit model for myelosuppression
  • FDA Clinical Pharmacology Considerations for ADCs (Draft Guidance, 2022)