Breast cancer stage at diagnosis has a large impact on the 10-year survival probability. Mammography based screening has enabled early detection of breast cancer. However, mammography has limited sensitivity, and suffers from overdiagnosis and overtreatment. We aim to improve screening by developing a classification algorithm based on mRNA signatures from blood platelets (Tumor Educated Platelets; TEP).
Platelet mRNA was sequenced from 265 women with stage I-IV breast cancer, 216 asymptomatic female controls, 144 women with non-cancerous disease and 321 women with other tumor types. The samples were randomly divided in age-matched training and evaluation series for development of the TEP-based breast cancer classification algorithm. Two optimal classifier thresholds were selected, to attain a test with either a high sensitivity or a high specificity. This enables employment of the algorithm for two clinically relevant purposes. The first application is to rule out breast cancer in women with an abnormal mammography. The second application is to conform, or ‘rule in’, breast cancer in asymptomatic women who are being screened due to increased breast cancer risk. We examined if classifications depended on patient age, tumor stage, subtype, BRCA mutation status, and breast density. In addition, we applied the algorithm to a pan-cancer cohort to test if it was breast cancer specific.
Validation of the algorithm in early-stage breast cancer patients and non-cancer samples, resulted in an AUC of 0.72 (CI95% 0.66 – 0.79, p < 0.001). Setting the threshold according to the rule-out application resulted in a sensitivity of 91% with a specificity of 37%. For the rule-in application the specificity was 92% with a 30% sensitivity. Correct classification by the TEP-based algorithm is independent of tumor stage, clinical subtype, BRCA1/2 status, and breast tissue density. By training a second algorithm, we show that TEP profiles from breast cancer patients can be discriminated from those with other tumor types (Validation n = 245, AUC 0.78, CI95%: 0.73 – 0.84, p < 0.001).
RNA signatures from tumor educated platelets enable detection of early breast cancer, and warrant validation in a confirmatory and screening setting.
VU University Medical Center, Netherlands Cancer Institute.
The European Research Council E8626 and 336540, the Dutch Organisation of Scientific Research 93612003 and 91711366, and 531002002, the Dutch Cancer Society, NCI at NIH CA176359 and CA069246.
T. Würdinger: Funding: Illumina; Shareholder: Grail, Inc. All other authors have declared no conflicts of interest.