Multimodal approach to capture spatially resolved single-cell tumor heterogeneity in breast cancer

Abstract

Tumor heterogeneity is known to play a key role in drug response and resistance. However, knowledge of tumor heterogeneity is largely limited to genetics and biopsy samples representing one region of the tumor. The study of spatial tumor heterogeneity is gaining more interest as researchers and clinicians realize that biopsies might miss biomarker information. Although several approaches like single-cell RNA seq and multiplex IHC/IF are used to study spatial heterogeneity, these methods tend to be costly and time consuming. Additionally, there is no consensus on how to interpret these results for optimizing therapy, making it harder to implement these approaches in the clinic.

Here we present a state-of-the-art technique to characterize tumor heterogeneity within individual patient tumors by integrating single cell RNA-seq data with biophysical modeling, using SimBioSys’ TumorScope (TS) – a biophysical modeling platform. As a proof of concept, we compared intra-tumoral heterogeneity from experimental single cell RNA-seq studies from TNBC patients (n=8) to to simulated intra-tumoral heterogeneity within TNBC patients from the ISPY-1 trial (n = 10). We also compared simulated vs. experimentally measured bulk transcriptional profiles within individual patients.

We found that the correlation between individual single-cell transcriptomes to biopsy transcriptomes was low (median Spearman correlation ~ 0.5). However, using TS to predict single-cell scale transcriptomics within a tumor based on biophysical characteristics of the tumor resulted in correlations of ~ 0.86, between simulated and experimentally measured transcriptomes (range = 0.85 – 0.87).

TS allowed the simulated single-cell RNA-seq data to create spatially resolved single-cell maps that quantify the spatial expression of key tumor biomarkers across the whole tumor including Ki67, HER2 and PDL1. The spatial distribution of these biomarkers can have important implications for therapy and patient prognosis. A more comprehensive understanding of the distribution of these molecular markers across the tumor could inform if some tumors deserved a different Ki67 score, if some patients are not receiving HER2-targeted therapy because of HER2 uneven expression across the tumor, and if PD-L1 could improve as a biomarker by considering its spatial expression.

In summary, we present a novel approach to characterize spatial tumor heterogeneity directly from patient MRIs using the TS biophysical modeling platform. This technology is less time consuming and less costly than other approaches, making it accessible for use in hospitals at a larger scale.

View the full publication here.