Next generation immuno-oncology tumor profiling using a rapid, non-invasive, computational biophysics biomarker in early-stage breast cancer

Abstract

Background: Immuno-oncology (IO) therapies targeting the PD-1/PD-L1 axis, such as immune checkpoint inhibitor (ICI) antibodies, have emerged as promising treatments for early-stage breast cancer (ESBC). Despite immunotherapy's clinical significance, the number of benefiting patients remains small, and the therapy can prompt severe immune-related events. Current pathologic and transcriptomic predictions of IO response are limited in terms of accuracy and rely on single-site biopsies, which cannot fully account for tumor heterogeneity. In addition, transcriptomic analyses are costly and time-consuming. We therefore constructed a computational biomarker coupling biophysical simulations and artificial intelligence-based tissue segmentation of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRIs), enabling IO response prediction across the entire tumor.

Methods: By analyzing both single-cell and whole-tissue RNA-seq data from non-IO-treated ESBC patients, we associated gene expression levels of the PD-1/PD-L1 axis with local tumor biology. PD-L1 expression was then linked to biophysical features derived from DCE-MRIs to generate spatially- and temporally-resolved atlases (virtual tumors) of tumor biology, as well as the TumorIO biomarker of IO response. We quantified TumorIO within patient virtual tumors (n = 63) using integrative modeling to train and develop a corresponding TumorIO Score.

Results: We validated the TumorIO biomarker and TumorIO Score in a small, independent cohort of IO-treated patients (n = 17) and correctly predicted pathologic complete response (pCR) in 15/17 individuals (88.2% accuracy), comprising 10/12 in triple negative breast cancer (TNBC) and 5/5 in HR+/HER2- tumors. We applied the TumorIO Score in a virtual clinical trial (n = 292) simulating ICI administration in an IO-naïve cohort that underwent standard chemotherapy. Using this approach, we predicted pCR rates of 67.1% for TNBC and 17.9% for HR+/HER2- tumors with addition of IO therapy; comparing favorably to empiric pCR rates derived from published trials utilizing ICI in both cancer subtypes.

Conclusion: The TumorIO biomarker and TumorIO Score represent a next generation approach using integrative biophysical analysis to assess cancer responsiveness to immunotherapy. This computational biomarker performs as well as PD-L1 transcript levels in identifying a patient's likelihood of pCR following anti-PD-1 IO therapy. The TumorIO biomarker allows for rapid IO profiling of tumors and may confer high clinical decision impact to further enable personalized oncologic care.

Keywords: ESBC; ICI; biophysical simulation; computational biomarker; immune checkpoint inhibitor; immuno-oncology; virtual clinical trial; virtual tumors.

Figure 1

Development of the TumorIO biomarker, as well as the design and implementation of the TumorIO Score predicting response to IO therapy. (A) The TumorIO biomarker was developed using non-IO-treated (and pre-IO-treated) patients from the I-SPY1 and I-SPY2 clinical trials, as well as single cell RNA-sequencing data. Transcriptomic gene expression signatures of cancer hallmarks were used to quantify metabolic activity and angiogenesis within single cells. These signatures were then correlated with PD-L1 expression to quantify the relationship between PD-L1 expression and tumor metabolic activity and angiogenesis. Subsequently, metabolic activity and angiogenesis were quantified within patient tumors based on biophysical modeling of DCE-MRI image series to generate spatially resolved probability maps of PD-L1 expression within individual patient tumors. (B) The TumorIO biomarker was implemented using IO-treated patients from the I-SPY2 clinical trial to develop the TumorIO Score associated with likelihood of pCR in response to IO therapy. The TumorIO Score was then validated in an independent set of patients and through a virtual clinical trial.

Figure 2

Automatic image segmentation of breast DCE-MRIs performed by artificial intelligence. The segmentation model uses a convolutional neural network (CNN) to identify and segment tissues based on multiple DCE-MRI slices (left). The tissue categories of the segmentation are: air (black), chest (yellow), skin (blue), fat (brown), gland (purple), vasculature (red), and tumor (green), shown at center. The median intensity projection across all segmented layers is shown at right.

Figure 3

PD-L1 heterogeneity across and within ESBC tumors. (A) Schematic of experimental approach depicting use of both scRNA-seq and whole-tumor bulk RNA-seq. (B) PD-L1 expression levels in bulk RNA-seq from 972 ESBC patients from the I-SPY2 trial. (C) Fraction of cells expressing PD-L1 in TNBC tumor cells, as judged by scRNA-seq from eight patients. (D) PD-L1 expression levels within single cells across TNBC tumors. (E) Fraction of cells expressing PD-L1 in HR+/HER2- tumors, as judged by scRNA-seq from nine patients. (F) PD-L1 expression levels across HR+/HER2- tumors. In (C–F), individual patients with no representative data indicate that no PD-L1+ cells were detected in these samples (i.e., HR+/HER2- patient 2, TNBC patient 2, and TNBC patient 6).

Figure 4

Gene expression signatures associated with PD-L1 expression in single breast cancer cells. (A) Levels of the angiogenesis signature across 1,886 cells derived from eight TNBC patients vs. PD-L1 expression, showing an association between angiogenesis and PD-L1 levels. (B) Levels of the metabolic activity signature across these TNBC cells vs. PD-L1 expression, showing an association between metabolic activity and PD-L1 levels. (C) Levels of the angiogenesis signature across 1,492 cells derived from 9 HR+/HER2- patients vs. PD-L1 expression. (D) Levels of the metabolic activity signature across these HR+/HER2- cells vs. PD-L1 expression.

Figure 5

TNBC patient tumors from the validation cohort demonstrate the distribution of the TumorIO biomarker and utility of the TumorIO Score in predicting pCR in response to ICI therapy. The general workflow entails analysis of patient DCE-MRIs in order to generate virtual tumors. Spatial maps of TumorIO (reflecting metabolic activity and angiogenesis) show areas of the tumor and TME contributing to the TumorIO Score. The DCE-MRI dataset and resulting virtualized tumor are shown (left and middle-left). The heatmap (middle-right) ranges from red to blue, representing a high to low gradient of the TumorIO biomarker within each tumor. These features are inversely correlated to the IO sensitivity of a tumor. In patient 6, low metabolic activity and areas of high angiogenesis contribute to a high TumorIO Score (right) and a correct prediction of pCR. In contrast, in patient 9, high metabolic activity and relatively few areas of angiogenesis contribute to a low TumorIO Score and a correct prediction of residual disease (right).

Figure 6

TumorIO analysis shows comparable efficacy to PD-L1 transcriptomics. Odds ratios of PD-L1 transcriptional levels, in comparison to the TumorIO Score, when predicting pCR in response to ICI therapy.

Figure 7

Schematic of factors contributing to tumor IO-sensitivity (left) and IO-insensitivity (right) both within tumors and within the local tumor microenvironment. These factors include degrees of nutrient delivery, T cell dormancy, T cell proliferation, and levels of exhaustion in local T cell populations.