Rational Identification of Novel Antibody-Drug Conjugate with High Bystander Killing Effect against Heterogeneous Tumors

Abstract

Bystander-killing payloads can significantly overcome the tumor heterogeneity issue and enhance the clinical potential of antibody-drug conjugates (ADC), but the rational design and identification of effective bystander warheads constrain the broader implementation of this strategy. Here, graph attention networks (GAT) are constructed for a rational bystander killing scoring model and ADC construction workflow for the first time. To generate efficient bystander-killing payloads, this model is utilized for score-directed exatecan derivatives design. Among them, Ed9, the most potent payload with satisfactory permeability and bioactivity, is further used to construct ADC. Through linker optimization and conjugation, novel ADCs are constructed that perform excellent anti-tumor efficacy and bystander-killing effect in vivo and in vitro. The optimal conjugate T-VEd9 exhibited therapeutic efficacy superior to DS-8201 against heterogeneous tumors. These results demonstrate that the effective scoring approach can pave the way for the discovery of novel ADC with promising bystander payloads to combat tumor heterogeneity.

Keywords: antibody‐drug conjugates; bystander‐killing effect; camptothecin derivatives; rational design; tumor heterogeneity.

Figure 1

The construction and validation of the GAT‐driven bystander‐killing score model. A) The architecture of the model is divided into two sections, each consisting of a molecular characterization layer, multiple GAT layers, aggregation layers, and a prediction layer. These sections are identical in structure but process different input data. Finally, the scores calculated by these two sections are comprehensively considered and serve as the final B‐K score. B–D) The performance of the B score model on the (B) train, C) validation, and D) test dataset. E) The prediction tests of the B score model for payload currently in clinic or development from ADCdb, value 1.5 in B score model is the dividing line for with and without bystander effect. (F‐H) The performance of the K score model on the F) train, G) validation, and H) test dataset.

Figure 2

Design and evaluation of novel Eds with membrane permeability. A) Schematic illustration of the structure and generation strategy exatecan derivatives (Eds). B) Modification site analysis by determining solvent exposure of DXd in DXd/Topo I/DNA complex. C) Generation and screening process of Eds. D) B score and K score of generated Eds, calibrated with DXd (0,0). E) The heat map of the B score, K score, B.K. score, and pIC₅₀ of Eds and DXd. F) Passive cell membrane diffusion of Eds and DXd by a PAMPA assay with atenolol as a negative control and carbamazepine as a positive control. G) Topo I inhibitory activity of Eds and DXd at 10 µM and 100 µM, with CPT (100 µM) as positive control. H) Proliferation inhibitory activity of Ed9 and DXd against human cancer cell lines with different HER2‐expression. Tumor cells were treated with payloads for 3 days and cell viability (%) was calculated. Data shown are representative of more than two independent duplicates.

Figure 3

Linker optimization, construction and characterization of Ed9 ADCs. A) Schematic illustration of the linker optimization strategy and structure of the Gly‐Gly‐Phe‐Gly and Val‐Ala linker conjugates. B) Proliferation inhibitory activity of ADCs against NCI‐N87 cells. Tumor cells were treated with ADCs for 5 days and relative cell viability (%) was calculated. C) Human cathepsin B‐mediated and D) Human liver S9‐mediated cleavage of NAC‐linker‐payloads (NAC‐L‐P) at 37°C. Cleavage of each probe was monitored by HPLC and LC/ESI‐MS. E) The half‐maximal inhibitory concentration (IC₅₀) of ADCs with different L‐Ps. F) Schematic illustration of the ADC conjugation through the inter‐chain disulfide reduction and classic Michael addition. G) Reversed‐phase high‐performance liquid chromatography (RP‐HPLC) of DAR8 ADCs. Absorbance wavelength was 280 nm. H) SDS‐PAGE analysis of Trastuzumab, T‐DXd, T‐VDXd, and T‐VEd9. I) Hydrophobic interaction chromatography (HIC) of DAR8 ADCs. Absorbance wavelength was 280 nm. J) Saturation‐binding curves obtained by NCI‐N87 (HER2+) cell‐based ELISA. K) Size‐exclusion chromatography (SEC) of DAR8 ADCs. Absorbance wavelength was 280 nm. Isomer ratio was calculated by absorbance area. DAR, drug‐to‐antibody ratio. All data shown are representative of more than two independent duplicates. Error bars represent S.D. Curve fitting and IC₅₀ calculation was performed using GraphPad Prism 9.3 software.

Figure 4

In vitro evaluation of ADCs. Proliferation inhibitory activity of ADCs against A) MDA‐MB‐231/GFP (HER2‐) cells, B) SKBR‐3 (HER2+) cells, and C) NCI‐N87 (HER2+) cells. Tumor cells were treated with ADCs for 5 days and relative cell viability (%) was calculated. D) Fluorescence imaging of NCI‐N87 and MDA‐MB‐231/GFP (Green) co‐culture system. the nuclei were stained with DAPI (Blue). Scale bar: 25 µm. E) Proliferation inhibitory activity of ADCs against MDA‐MB‐231/GFP (HER2‐) cells in the co‐culture system. The relative cell viability (%) was calculated by measuring GFP fluorescence intensity. F) Proliferation inhibitory activity of ADCs against both HER2+ and HER2‐ cells in the co‐culture system, error bars represent S.E.M. G) Proliferation inhibitory activity of ADCs against HER2‐ cells in the co‐culture system, error bars represent S.E.M. H) Proliferation inhibitory activity against HER2‐ cells of HER2+ cells supernatant cultured with ADCs, error bars represent S.E.M. All data shown are representative of more than two independent duplicates. Error bars represent S.D. without additional annotation. Curve fitting and IC₅₀ calculation was performed using GraphPad Prism 9.3 software.

Figure 5

In vivo evaluation of ADCs. A) Anti‐tumor activity in HER2+ NCI‐N87 gastric cancer model following a single intravenous ADC dose of 1 or 3 mg kg⁻¹. B) HER2 expression of remaining tumors after treatment with PBS and T‐DXd (1 mg k⁻¹g). Scale bar: 100 µm. C–F) Bystander killing in co‐inoculation xenograft model following a single intravenous ADC dose of 2 or 5 mg kg⁻¹. Luciferase activity was detected by in vivo imager after intraperitoneal injection of substrate. C) Tumor volume change. D) Luciferase activity. E) Tumor weight. F) Bioluminescence imaging data of luciferase activity. G) Tumors collected at treatment endpoint of PBS control and each ADC. H) HER2 expression on tumors consisting of either HER2‐positive or HER2‐negative cells, co‐inoculation (2: 5 ratio at the time of implantation) and the remaining and regrown tumors after treatment with each ADC. Scale bar: 100 µm. All data shown are representative of more than two independent duplicates. Error bars represent S.D. Curve fitting and p‐value calculation was performed using GraphPad Prism 9.3 software.