Development of an accurate breast cancer detection classifier based on platelet RNA

Abstract

Platelets possess cancer-induced reprogramming properties, thereby contributing to RNA profile alterations and further cancer progression, while the former is considered a promising biosource for cancer detection. Hence, tumor-educated platelets (TEP) are considered a prospective novel method for early breast cancer (BC) screening. Our study integrated the data from 276 patients with untreated BC, 95 with benign disease controls, 214 healthy controls, and 2 who underwent mastectomy in Chinese and European cohorts to develop a 10-biomarker diagnostic model. The model demonstrated high diagnostic performance for BC in an independent test set (n = 177) with an area under the curve of 0.957. The sensitivity for BC diagnosis was 89.2%, with 100% specificity in asymptomatic controls, while that for the symptomatic group, including benign tumors and inflammatory diseases, was 62.1%. The model demonstrated substantial accuracy for stages 0-III BC (80% for stage 0 [n = 5], 83.3% for stage I [n = 12], 94.6% for stage II [n = 37], and 88.9% for stage III [n = 9]) and precisely helped determine residual cancer in two patients who underwent mastectomy. Moreover, our developed classifiers distinguish different BC subtypes properly. In summary, we created and tested a new TEP-RNA-based BC diagnostic model that was confirmed valid and demonstrated high efficiency in detecting early-stage BC and heterogeneous subtypes, including recurrent tumors. However, these results warrant more validation in larger population-based prospective studies before clinical implementation.

Fig. 1

Study design and patient enrollment. A total of 221 samples were disqualified, and Agilent 2100 Bioanalyzer was used to detect genomic contamination and severe degradation along with 3 samples that failed quality control. Quality control keeps samples that have a total detected RNA count of > 1500 per sample and a correlation of > 0.5 between samples. BC breast cancer.

Fig. 2

Platelet differences in patients with BC and healthy donors. (a) Volcano plot illustrating the 2388 differentially expressed RNAs (P < 0.001, log CPM ≥ 3). A total of 1181 demonstrated significant upregulation (red), and 1207 exhibited downregulation (blue) in the BC group. (b) Heatmap of unsupervised clustering with BC and healthy individuals in the training cohort. We revealed that BC or normal samples could be distinguished through platelet RNA expression patterns.

Fig. 3

Model development and optimization workflow. We identified 1761 differentially expressed genes using the training dataset with patients with BC and non-BC controls (P < 0.001, log CPM ≥ 3). Subsequently, we performed LASSO and single-factor logistic regression (with each marker AUC > 0.8) for feature reduction, leading to the selection of the top four genes. The comparison between patients with BC and symptomatic controls identified 120 differentially expressed genes (P < 0.05, log CPM ≥ 3). After dimensionality reduction (each marker AUC > 0.65), we received the top seven genes. Two separate panels were modeled for SVM. The 4 markers were employed for the initial data classification, and positive predictions (BC) were further refined using the 7 markers to obtain the final diagnostic outcome.

Fig. 4

Performance of TEP-RNA panel in training and validation cohorts for BC diagnostics. ROC curves (a) for BC diagnostics using the 4-BC-TEP-RNA panel in training (black line) and validation (red line) cohorts along with the confusion matrix (b) tailored for the validation cohort. Furthermore, using the 10-BC-TEP-RNA panel to produce ROC curves (c) in both the training and validation cohorts, and presenting the confusion matrix (d), specifically for the latter.

Fig. 5

Performance of TEP-RNA panel in an independent test cohort for BC diagnostics. (a) ROC curves for BC diagnostics using 4-BC-TEP-RNA and 10-BC-TEP-RNA panels on the independent test cohort. We observe a significantly higher AUC value of the 10-BC-TEP-RNA panel than that of the former (p = 1.578e–05). (b) Confusion matrix for BC diagnostics using the 4-BC-TEP-RNA panel in the test cohort. (c) Bar plot of detection accuracy for various stages of BC and control samples. High detection accuracy is observed for all BC stages. Stage na denotes samples without recorded tumor staging information.

Fig. 6

SVM Model for breast cancer companion diagnosis with TEP-derived RNA. Extracting feature genes based on HER2 amplification or HR positivity using 60% of samples for model training and allocating 40% for validation. (a, b) Unsupervised clustering heatmap and classifier ROC curve based on HER2 receptor status. The AUC value in the validation set is 0.882 (95% CI 79.58%–96.92%), indicating a significant improvement over the random classifier (p = 2.238e–05). (c, d) Unsupervised clustering heatmap and classifier ROC curve based on HR receptor status. The AUC value in the validation set is 0.861 (95% CI 76.71%–95.54%), revealing the significant enhancement over the random classifier (p = 0.0063). Selected genes: the number of RNAs included in the training dataset of the SVM algorithm.