Case Studies

ADCs
ADCs, IHC

AI-augmented Scoring of the Bystander Effect for Effective ADC Patient Selection

Challenge

In a recently published study, the teams at AstraZeneca and Daiichi Sankyo found that the Spatial Proximity Score (SPS), an in situ measurement of the bystander effect, was a better predictor of response to Enhertu (trastuzumab deruxtecan) than just antigen expression (HER2) alone1. They found that the HER2-low or -negative subgroup could be further segmented into ‘SPS-high’ and ‘SPS-low’ scores. The SPS-high group benefited from the drug even when expression of the drug target (HER2) was low or negative (Figure 1), primarily due to the ADC-specific mechanism-of-action (MOA) known as the bystander effect.

To enable ADC developers to improve their patient selection strategies and derisk late-stage trials, we have developed a similar SPS-based analysis, to establish a scalable, AI-augmented model for scoring the bystander effect and accounting for its impact on biomarker development.

Figure 1. Kaplan-Meier Analyses from the AstraZeneca-Daiichi study showed that Spatial Proximity Scoring could predict progression-free survival across multiple patient sub-groups, including HER2 negative patients [IHC 1+, IHC 2+/ISH- and IHC 0] 1.

Solution

We have developed a framework to score the bystander antitumor activity of ADCs, which affects both antigen-positive tumor cells and antigen-negative tumor cells within a specific radius from an antigen-positive cell (Figure 2). Based on this definition of an ‘affected cell’, the proximity score reflects the proportion of affected cells as a percentage of all tumor cells. AI-augmented SPS allows quantitative scoring of tissue structure, accounting for tissue heterogeneity, tumor architecture, and staining intensity and variability.  

Figure 2. Schematic illustrating Spatial Proximity Scoring (SPS) based on the ADC-specific MOA known as the bystander effect. The dashed lines reflect one of the tunable parameters for downstream analysis and building a strategy for patient selection.

Using this framework, we have created a tunable scoring model that can be calibrated to patient outcomes across different patient cohorts. These scoring parameters included:

  • Distance (radius) from the negative cell
  • Number of positive neighbor cells
  • Antigen expression level of positive cells

Reference:
Kapil, Ansh et al. “HER2 quantitative continuous scoring for accurate patient selection in HER2 negative trastuzumab deruxtecan treated breast cancer.” Scientific reports vol. 14,1 12129. 27 May. 2024, doi:10.1038/s41598-024-61957-9
Link to paper >