Plasma Exosomal RNA Signature for Prostate Cancer Bone Metastasis

Background Prostate cancer (PCa) is a leading cause of cancer-related morbidity worldwide, with bone metastasis being the most frequent and devastating complication. Early and accurate detection of bone metastasis is critical for timely intervention and improved patient outcomes. However, current biomarkers such as PSA lack sufficient sensitivity and specificity for early metastasis detection, and conventional imaging (e.g., bone scan, CT) often identifies metastases only after they become clinically apparent. PSMA PET has emerged as the most sensitive imaging modality for defining metastatic status at baseline, but its high cost, limited availability, and radiation exposure preclude its use as a universal screening tool.

Concurrently, liquid biopsy-particularly the analysis of plasma exosomal RNAs-offers a unique window into tumor biology. Exosomal RNAs are stable in circulation, reflect the molecular characteristics of the primary tumor, and can be detected using sensitive methods such as high-throughput sequencing and digital droplet PCR. Despite their promise, large-scale prospective multicenter studies with rigorous multi-phase validation are lacking.

Need There is a critical unmet need for a non-invasive, robust biomarker that can identify patients at very low risk of bone metastasis, thereby allowing safe deferral of PSMA PET/CT. Existing tools lack adequate negative predictive value to confidently rule out bone metastasis in low-risk populations. A successful rule-out test would reduce unnecessary imaging, lower healthcare costs, and minimize patient radiation exposure. This study implements a four-stage design with phase-specific sample size considerations aligned with contemporary standards for biomarker development.

Study Design and Technical Phases This is a prospective, multicenter, phase-sequential biomarker development and validation study. The gold standard for bone metastasis status is baseline, treatment-naïve PSMA PET/CT. All blood samples are collected prior to any prostate cancer-related treatment and prior to prostate biopsy (if performed) to avoid biopsy-induced contamination. Whole blood (approximately 10 mL in EDTA tubes) is processed within 2 hours to obtain plasma, which is stored at -80°C until analysis.

The four phases are defined as:

Phase 1 (Discovery, n=250): High-throughput RNA sequencing (e.g., small RNA-seq) of plasma exosomes. Differential expression analysis (e.g., DESeq2 or edgeR) identifies candidate RNAs distinguishing patients with versus without bone metastasis. Multiple testing correction (FDR <0.05) is applied.

Phase 2 (Model Development, n=300): An independent cohort enriched for bone metastasis (approximately 200 positive, 100 negative) is used for quantitative measurement of candidate RNAs using ddPCR. A continuous risk-scoring model is constructed using machine learning (e.g., regularized regression, random forests, or gradient boosting). The model is locked after internal cross-validation, before any validation data are examined.

Phase 3 (Internal Validation, n=300): A consecutive cohort reflecting natural disease prevalence (expected bone metastasis rate ~30%) is used to evaluate the locked model. In this phase, a single cut-off value is selected to achieve a sensitivity of ≥95%. The specificity at that cut-off is the primary endpoint. Secondary metrics (NPV, PPV, AUC, calibration, decision curve analysis) are also assessed.

Phase 4 (External Validation, n=150): A geographically distinct, multi-center cohort enriched for bone metastasis is used to assess generalizability, applying the same cut-off determined in Phase 3.

Statistical Considerations Sample Size Justification Sample sizes were chosen based on published recommendations for phased biomarker studies and on formal precision analysis for the primary endpoint (specificity at the ≥95% sensitivity threshold).

Phase 1 (Discovery, n=250): This sample size is typical for high-throughput discovery studies. With an expected bone metastasis prevalence of ~30%, approximately 75 positive and 175 negative cases will be available, providing sufficient power to detect differential expression with FDR <0.05.

Phase 2 (Model Development, n=300): Following the "events per variable" (EPV) rule (≥10 events per candidate predictor) and assuming a parsimonious final model (≤20 candidate RNAs after Phase 1), a minimum of 200 positive events is required. The cohort is enriched for bone metastasis (target ~67% positive), thus a total of 300 patients (≈200 positive, ≈100 negative) is planned.

Phase 3 (Internal Validation, n=300): The primary endpoint is specificity at the cut-off achieving ≥95% sensitivity. Assuming a true specificity of 35% at this threshold, a sample of approximately 210 negative patients (from total n=300 with ~30% bone metastasis prevalence) yields a 95% confidence interval half-width of approximately ±6-7%. This precision is more than sufficient to rule out a clinically useless specificity (e.g., <15%) and allows robust subgroup analyses.

Phase 4 (External Validation, n=150): This multi-center cohort is enriched for bone metastasis (target ~40-50% positive). The sample size of 150 (≈75 positive, ≈75 negative) allows precise estimation of specificity (95% CI half-width ±11% assuming 35% specificity) and sensitivity in an independent setting, confirming generalizability.

All sample sizes may be adjusted modestly based on actual recruitment; any adjustments will be documented.

Statistical Analysis Plan (High-Level) Analyses will be performed using R (version ≥4.2). The final analysis plan will be finalized before locking validation data.

Phase 1: Differential expression analysis (DESeq2/edgeR). Candidate selection based on fold change, adjusted p-value, and abundance.

Phase 2: Machine learning model building using cross-validation. The final continuous model is locked.

Phase 3: Apply the locked model to the internal validation cohort. Determine the cut-off value that achieves sensitivity ≥95%. Report specificity, NPV, PPV, AUC, calibration, and decision curve analysis at this cut-off.

Phase 4: Apply the same cut-off to the external validation cohort and repeat the performance evaluation.

Secondary/exploratory analyses: Correlation, subgroup analyses, mechanistic studies (as detailed in the secondary outcome measures).

Data Management Data will be stored centrally in a REDCap database with 3-step authentication. Data entry will occur approximately every 3-6 months. Patient confidentiality will be maintained.