Detection of early-stage cancers using circulating orphan noncoding RNAs in blood

Background

• Small noncoding RNAs (sncRNAs) have established roles as post-transcriptional regulators of cancer pathogenesis.
• We previously reported a novel and unannotated class of sncRNAs that were found in breast cancer tissue but not in normal tissue adjacent to the tumor, which we termed orphan noncoding RNAs (oncRNAs).¹ Since then, we have identified and validated novel oncRNAs in multiple cancer tissues, using data from The Cancer Genome Atlas (TCGA) and other independent cohorts.²
• We recently showed that these oncRNAs can also be detected in sera and demonstrated prognostic value for treatment response and event-free survival among breast cancer patients.³
• Early detection of breast cancer is crucial for optimal patient outcomes but cannot always be accomplished based on symptoms or mammography.
• We hypothesize that oncRNAs can be used as biomarkers in a liquid biopsy strategy to detect breast cancer across a range of cancer stages and tumor sizes.

Goals

• Develop and validate a methodology using machine learning to accurately predict breast cancer status based on oncRNA profiles detected in patient sera.

Samples

• The study cohort includes clinically diagnosed female breast cancer patients (N=96) and age- and sex-matched individuals from the general population with no known diagnosis of cancer (N=95).
• Breast cancer patients were treatment-naive at sample collection and were selected for this study to represent all stages of breast cancer (I–IV) and a broad range of ages, including patients <45 years old.
• Samples were acquired from two commercial biobanks and processed for small RNA sequencing. Dates of blood draw for serum collection ranged from 2010 to 2022.
• Patients had provided informed consent and contributing centers had obtained IRB approval.

Methods

• RNA was extracted from 191 frozen serum samples of 1.0ml volume and prepared for sequencing. Sample libraries were sequenced to an average depth of 17.7 million 50 bp single-end reads per sample.
• Previously, 261,504 oncRNAs were found to be significantly associated with cancer across multiple tissues, using data from The Cancer Genome Atlas (TCGA) as a discovery cohort through a pan-cancer study.² To refine our TCGA library of tissue-derived oncRNAs for applications in serum samples, all oncRNA sequences detected in >1 serum sample of an independent cohort of 31 cancer-free controls were removed, yielding a library of 250,332 oncRNAs.
• This filtered library of 250,332 oncRNAs was used as a reference database to generate oncRNA expression profiles in our study cohort (N=191). Of these, 171,981 were detected in at least one individual serum sample.
• oncRNA expression profiles were used to build an ensemble of logistic regression models to make predictions of breast cancer vs. control. The ensemble model was trained and evaluated using a 5-fold cross-validation setup. Within each training fold only oncRNAs observed in >3% of samples and yielding an odds ratio for breast cancer >1 were used to train and validate the model.

Study Cohort

oncRNA Library Creation and Profiling

Result 1: oncRNA Content Differentiates Cancer Status

Figure 1. oncRNA Content in Control and Breast Cancer Serum Samples

• Of the 250,332 oncRNAs in the filtered TCGA multicancer library, 171,981 (68.7%) were detected in the study cohort (N=191) with 33,043 observed in >3% of all study samples.
• Total oncRNA content in a sample -- the aggregate count of those 33,043 frequently observed oncRNAs, normalized by sequencing depth -- was significantly greater among cancer samples (one-sided Mann-Whitney U test, P=4.19x10^-16

Result 2: Ability of oncRNA-Based Model to Discriminate Between Breast Cancer Patients & Cancer-Free Controls in Serum

Figure 2. (A) ROC Curve, (B) Sensitivity by Cancer Stage, and (C) Sensitivity by Tumor T Category

• An ensemble of logistic regression models trained on serum oncRNA measurements with 5-fold cross validation (see Methods) discriminated effectively between samples from patients with breast cancer vs. cancer-free controls (A-C).
• On average, 10,416 [range: 9,769–11,312] oncRNAs were used as features within each training fold.
• The ROC curve demonstrated an AUC of 0.94 (95% CI: 0.88–0.98), with a sensitivity of 78.1% (68.5%–85.9%) at 95% specificity (A).
• Sensitivities at 95% specificity were highest for early cancer stages (B) and small tumor sizes (C) and tended to be similar across these categories. 95% CI were calculated using the Clopper-Pearson method.

Figure 3. Model Sensitivity by Breast Cancer Molecular Subtypes

Figure 3. Model Sensitivity by Breast Cancer Molecular Subtypes

• Breast cancer samples (N=96) were assigned molecular subtypes based on ER, PR, HER2 expression through immunohistochemistry (Top).
• Each subtype was subdivided into early (I/II) and late (III/IV) stage at diagnosis (Bottom).
• Sensitivities at 95% specificity were uniformly high, at least 87% for each molecular subtype except Luminal A (Top) and were often greater among stage I/II than stage III/IV (Bottom).

Conclusions

• Analyzing oncRNA data with machine learning models accurately predicted breast cancer across all cancer stages (I–IV) and tumor categories (T1–T4).
• Early stage tumors and small tumors were detected with high sensitivity: 82% for stage I and 90% for T1a/b, respectively (at 95% specificity).
• This oncRNA-based liquid biopsy technology is compatible with standard blood sample requirements enabling integration into conventional clinical workflows.
• The results will be validated prospectively in further population studies.