The tumor microenvironment (TME) significantly influences tumor growth and progression. Tumor cells secrete various signals that enter the bloodstream, providing potential avenues for cancer biomarker development. Current methods, such as analyzing cytokines, growth factors, exosomes, and circulating tumor cells, are often resource-intensive and limited to late-stage diagnosis. Epigenetic chromatin remodeling plays a crucial role in gene expression regulation, and recent studies suggest that specific chromosome conformations in PBMCs could serve as diagnostic tools for cancers. PBMC activation involves chromatin remodeling and DNA methylation changes, reflecting the signals sensed by immune cells. This study proposes a novel approach using chromatin imaging of PBMCs combined with machine learning to identify biomarkers for cancer diagnosis and treatment efficacy evaluation. The researchers aimed to develop an imaging and AI-based pipeline to characterize the 3D chromatin organization in PBMCs from patients undergoing proton therapy, evaluating its ability to discriminate between healthy and cancer patients and different tumor groups, and to assess the treatment efficacy of proton therapy.

Numerous studies have explored the use of various biomarkers in liquid biopsies for cancer diagnosis and treatment evaluation. These include transcriptomic and proteomic analyses of blood cells, which have shown promise but often require resource-intensive methods. The role of epigenetic chromatin remodeling in gene expression regulation has also been highlighted, suggesting that chromatin organization could provide a valuable source of biomarkers. Several studies have demonstrated the potential of chromatin conformation changes in PBMCs as diagnostic tools for different cancers. The use of machine learning algorithms combined with fluorescent imaging has opened new possibilities for detecting subtle changes in chromatin structure. However, a comprehensive pipeline integrating chromatin imaging and AI for cancer diagnosis and therapy evaluation using liquid biopsies was lacking, prompting this study.

The study employed a multi-step pipeline involving PBMC isolation, immunostaining, confocal microscopy, and computational image analysis. Blood samples were collected from 10 healthy controls, 10 pan-tumor patients, and 30 patients with either meningioma, glioma, or head and neck tumors undergoing proton therapy. PBMCs were isolated via density gradient centrifugation and immunofluorescently stained with DAPI (chromatin), γH2AX (DNA damage marker), Lamin A/C (nuclear structural protein), and cell surface markers (CD3, CD4, CD8). Confocal microscopy images were acquired, followed by image processing and feature extraction using a custom-developed Python pipeline. This pipeline included nuclei and cell segmentation using Otsu's method and Chan-Vese algorithm, and feature extraction using an adaptation of the chrometric package. Features included nuclear morphology (volume, shape, concavity, curvature), chromatin organization (heterochromatin content, texture features), and protein expression levels (γH2AX, Lamin A/C). Machine learning, specifically random forest classifiers (RFCs), were used to classify PBMCs based on the extracted features, distinguishing between healthy and tumor patients, different tumor types, and treatment time points. Leave-one-patient-out cross-validation was employed to assess the robustness of the classifiers. Statistical tests (Welch's t-test, Wilcoxon rank-sum test) were used to compare feature expression levels between different groups. The pipeline also assessed the abundance of different PBMC subsets (CD4+, CD8+, CD3+) to investigate potential cell type-specific effects.

The study found that the 3D chromatin organization of PBMCs effectively distinguished between healthy and tumor patients. A random forest classifier achieved an average accuracy of 77% in classifying PBMCs from these two groups at the single-cell level and perfect classification at the patient level. Key features differentiating the groups included nuclear volume, shape, and heterochromatin content. The pipeline also successfully differentiated between PBMCs from patients with three different tumor types (meningioma, glioma, head and neck) with an average accuracy of 78%, reaching 89% accuracy for head and neck tumors. Head and neck tumor PBMCs exhibited significantly increased nuclear and heterochromatin volume. Analysis across different time points during proton therapy revealed accurate classification (82-83% for meningioma and glioma, 64% for head and neck) indicating successful monitoring of treatment effects. The chromatin features, Lamin A/C expression, and DNA damage content (γH2AX) showed changes during proton therapy, moving toward levels observed in healthy donors. Furthermore, the tumor-specific signature in PBMC chromatin organization was sequentially lost during the course of proton therapy. Analysis of PBMC subpopulations (CD3+, CD3−) showed distinct chromatin states but did not alter the main findings regarding treatment response, suggesting that the observed changes were not solely due to shifts in cell type proportions.

This study demonstrates the potential of using chromatin organization of PBMCs as a novel biomarker for cancer diagnosis and therapy evaluation. The developed imaging and AI-based pipeline offers a resource-efficient alternative to existing methods, providing a high-dimensional read-out of the integrative functionality of chromatin. The ability to distinguish between different tumor types and monitor treatment responses highlights the potential for personalized medicine. The findings align with previous research showing relationships between nuclear morphology, chromatin organization, and immune cell activation. The different treatment responses observed across tumor groups warrant further investigation in larger clinical trials to optimize therapeutic strategies. The sequential loss of the tumor-specific signature during proton therapy further supports the efficacy of the treatment.

The study successfully developed an imaging and AI-based pipeline for identifying chromatin biomarkers in PBMCs from liquid biopsies. The identified biomarkers accurately distinguished between healthy and tumor patients, different tumor types, and treatment time points during proton therapy. The approach offers a promising tool for cancer diagnostics and personalized treatment evaluation. Future research could focus on expanding the study population, including more tumor types and treatment modalities, and refining the analysis of PBMC subpopulations to improve the accuracy and sensitivity of the biomarkers.

The study involved a relatively small number of patients, potentially limiting the generalizability of the findings. The study focused primarily on proton therapy; further research is needed to evaluate the applicability of this method to other treatment modalities. While the analysis considered some PBMC subpopulations, a more detailed investigation of cell-type specific changes in chromatin organization could potentially yield even better results. The robustness of feature extraction to variations in sample preparation and imaging settings needs further investigation for wider applicability.