Studying blood-based immune biomarkers for cancer and their confounders

Abstract

Background: Cancer is a leading cause of death worldwide, with survival strongly dependent on stage at diagnosis. Current screening approaches for early-stage cancer are largely designed to detect a single cancer type, can be invasive, and often have limited sensitivity and specificity. Blood-based immune biomarkers offer a promising, minimally invasive approach for multi-cancer detection, as tumour development induces measurable systemic alterations in circulating leukocyte composition. However, immune profiles are inherently variable and influenced by biological and environmental factors independent of disease status, including genetic background, sex, age, diet, and exercise. The impact of these confounders on immune-based cancer detection is poorly characterised, representing a gap between preclinical findings and clinical translation.

Hypotheses/Aims: The central premise of this study is that circulating leukocyte profiles differ between healthy and cancer-bearing individuals, but that intrinsic and extrinsic variables influence these profiles and reduce their ability to indicate cancer if not appropriately accounted for. We further hypothesise that predictive models matched to the host background detect cancer more reliably than mismatched models. The aims were therefore to evaluate the utility of blood-based immune profiles for multi-cancer detection, to assess how biological variability influences classification performance, and to determine whether matching the model to the host background improves detection.

Methods: Three syngeneic murine tumour models were used: CT26 (colorectal), 4T1 (mammary) in female BALB/c mice, and RM1 (prostate) in male C57BL/6 mice. Blood was collected at early and late tumour stages, and immune cells were characterised using high-dimensional spectral flow cytometry. Multivariate analyses (PERMANOVA and PCoA) assessed group differences, and Random Forest classification evaluated cancer detection performance, with SHAP-based feature importance identifying key immune predictors. Biological variability was assessed by comparing immune profiles across healthy populations differing by genetic background, sex, age, exercise, diet, and housing environment, and across tumour-bearing populations varying by age and sex within a consistent BALB/c background.

Results: First, late-stage tumours produced distinct, tumour-specific immune signatures reliably detectable from blood, whereas early-stage cancers closely resembled healthy controls, with substantial overlap limiting sensitivity. Second, intrinsic factors, particularly genetic background, produced immune variation exceeding tumour-associated changes in many comparisons. Models trained on mismatched healthy populations showed significantly reduced accuracy, primarily impairing early-stage detection while late-stage disease remained detectable; this was not resolved by increasing training data alone, indicating that biological matching, not sample size, is the primary determinant of performance. Third, classification performance was strongly background-dependent in tumour-bearing populations, with old female BALB/c mice achieving the strongest early-stage detection (92% accuracy, 88% sensitivity), while old male BALB/c mice showed consistently reduced performance.

Conclusion: Blood-based immune profiles can detect cancer under biologically matched conditions, but reliability is substantially reduced by biological variability, particularly for early-stage disease. This study establishes a preclinical framework for how biological heterogeneity shapes immune biomarker performance and highlights that reliable multi-cancer detection requires a framework accounting for the biological context of both healthy and tumour-bearing populations. These findings inform the future development and clinical translation of blood-based immune biomarkers for cancer screening.