Plasma-derived exosomal analysis and deconvolution enables prediction and tracking of melanoma checkpoint blockade response

Alvin Shi, Gyulnara G. Kasumova, William A. Michaud, Jessica Cintolo-Gonzales, Marta Díaz Martínez, Jacqueline Ohmura, Arnav Mehta, Isabel Chien, Dennie T. Frederick, Sonia Cohen, Deborah Plana, Douglas B. Johnson, Keith T. Flaherty, Ryan J. Sullivan, Manolis Kellis, Genevieve M. BolandbioRxiv 2019 · DOI

Introduction

Immune checkpoint inhibitors (ICI) have fundamentally changed the treatment landscape for metastatic melanoma. However, despite their success, a significant majority of patients do not achieve a durable response. Predicting which patients will benefit remains a major clinical challenge. Current biomarkers, such as cell-free DNA (cfDNA), primarily focus on tumor-intrinsic properties. Yet, the success of ICI therapy depends on a complex "molecular tug-of-war" between the tumor and the host’s immune system.

This research presents a non-invasive "liquid biopsy" approach that leverages plasma-derived exosomes to simultaneously monitor both tumor and immune status. By combining transcriptomic profiling with a novel Bayesian deconvolution model, the authors demonstrate that exosomal RNA can predict ICI response with high accuracy and track the dynamic biological changes occurring during treatment.

The Exosomal Advantage

Exosomes are small extracellular vesicles secreted by cells into the bloodstream. Unlike cfDNA, which is often a byproduct of cell death, exosomes are actively secreted and carry a "snapshot" of the parent cell's transcriptome. This makes them ideal messengers for real-time monitoring. Crucially, because exosomes in the blood originate from both the tumor and various immune cell types, they provide a dual-view of the systemic response to therapy.

Methodology: Unmixing the Signal

The primary technical challenge in analyzing bulk plasma exosomes is the heterogeneity of their origin. To address this, the researchers developed a custom computational framework:

  1. Transcriptomic Profiling: The team analyzed exosomal RNA from a discovery cohort (50 patients) and a validation cohort (30 patients) using microarray and RNA-seq technologies.
  2. Bayesian Deconvolution Model: This core innovation allows for the "unmixing" of the bulk signal. By utilizing data from melanoma cell lines and healthy controls, the model estimates a "packaging coefficient"—the efficiency with which specific RNA transcripts are loaded into exosomes. This allows the researchers to partition transcripts into tumor-specific and non-tumor (immune) components.
  3. Machine Learning and Mutational Analysis: A Random Forest classifier was trained on differentially expressed genes (DEGs) to predict clinical outcomes. Additionally, the team utilized RNA-seq data to identify somatic driver mutations, calculating a "driver mutational load" as a liquid biopsy proxy for tumor mutational burden (TMB).

Key Insights and Results

High-Fidelity Proxies

The study established a strong correlation ($R^2 \approx 0.82$) between the RNA profiles found in a patient’s tumor tissue and their plasma exosomes. This confirms that exosomal cargo is a reliable reflection of the underlying tumor biology.

Predictive Power

The predictive models demonstrated robust performance across cohorts. The Random Forest model achieved an Area Under the Curve (AUC) of 0.805 in the discovery cohort and improved to 0.84 in the independent validation cohort. Furthermore, specific genes such as IGFL1, TFF2, and MAP2K4 were identified as significant predictors of progression-free and overall survival.

Biological Drivers of Resistance

The deconvolution model revealed that resistance to ICI is not solely a tumor-intrinsic property. For instance, the resistance marker KLF10 was found to be primarily derived from the non-tumor (immune) component. In non-responders, the researchers also observed a "noticeable degradation" of T-cell receptor (TCR) pathway activity over the course of treatment, suggesting a failure of the immune system to maintain an effective attack against the tumor.

Mutational Signatures

Responders to ICI therapy tended to have a significantly higher fraction of "driver" somatic mutations in their exosomal RNA compared to non-responders. This suggests that exosomal RNA-seq can be used to identify specific driver mutations previously only accessible via invasive tissue biopsies.

Discussion and Future Directions

While the results are promising, the authors note several areas for future development. The current deconvolution model assumes a linear relationship for RNA packaging and may not fully capture the extreme heterogeneity of individual tumors. Additionally, while the findings were validated across two cohorts, larger-scale clinical trials are necessary for standardizing the platform.

Future work may involve cell-specific exosome selection—such as isolating only T-cell-derived exosomes—to provide even higher resolution into the immune system's behavior. Nevertheless, this study establishes plasma-derived exosomes as a powerful, non-invasive tool for the joint monitoring of tumor-intrinsic and immune-derived signatures, offering a new pathway for personalized immunotherapy management.

Summary updated

Abstract

Abstract Purpose Immune checkpoint inhibitors (ICI) have demonstrated promising therapeutic benefit although a majority will not respond. Here we identify and validate predictive biomarkers from plasma-derived exosomes that allow non-invasive monitoring of tumor intrinsic and host immune status and prediction of ICI success. Experimental Design Transcriptomic profiling of peripheral blood bulk exosomes and tumors from a discovery cohort of 50 patients with metastatic melanoma treated with ICI was undertaken; a further validation cohort of 30 patients was utilized to validate findings from the discovery cohort. We designed a Bayesian probabilistic model to partition bulk exosomes into tumor-specific and non-tumor-specific proportions. Results Exosomal RNA signatures exhibit significant correlations with tumor transcriptomes. Exosomal profiles reflect several key biological drivers of ICI resistance or melanoma progression, exhibit significantly differentially expressed genes and pathways, and correlate with and are predictive of clinical response to therapy. Our deconvolution model estimates contributions from tumor and non-tumor sources, enabling more precise interpretation of differentially-expressed genes and pathways. Exosomal RNA-seq mutational information can be used to segregate responders and non-responders. Conclusions Peripheral blood-derived exosomes can serve as a non-invasive biomarker to jointly probe tumor-intrinsic and immune changes to ICI, and can potentially function as predictive markers of ICI responsiveness and a monitoring tool for tumor persistence and immune activation. Statement of Significance We use transcriptomic analysis of bulk, non-selected, peripheral blood derived exosomes to reveal both tumor-intrinsic and immune-derived signatures predictive of early response to immune checkpoint inhibitor therapy. We develop a novel computational model to classify exosomal transcripts into tumor and non-tumor components and establish relevance in immune checkpoint blockade therapy. We show that tumor driver load from RNA-seq mutational calls are significantly different between responders and non-responders.
BibTeX
@misc{shi2019plasma,
  author = {Alvin Shi and Gyulnara G. Kasumova and William A. Michaud and Jessica Cintolo-Gonzales and Marta Díaz Martínez and Jacqueline Ohmura and Arnav Mehta and Isabel Chien and Dennie T. Frederick and Sonia Cohen and Deborah Plana and Douglas B. Johnson and Keith T. Flaherty and Ryan J. Sullivan and Manolis Kellis and Genevieve M. Boland},
  title = {{Plasma-derived exosomal analysis and deconvolution enables prediction and tracking of melanoma checkpoint blockade response}},
  year = {2019},
  doi = {10.1101/809699},
  howpublished = {bioRxiv},
}
Figure