A total of 1,129 blood samples (575 breast cancer patients, 454 healthy controls, 100 other cancers) were examined, including an assessment of their association with other types of cancer, and anesthesia. The number of samples in the validation set was calculated by the Medical Research Collaborating Center (MRCC) of Seoul National University Hospital (Seoul, Korea).

All previous experiments were performed using plasma samples, and additional tests were performed with 460 new serum samples (228 breast cancer, 232 healthy controls) not used in the experiments to assess the validity of using serum. At the same time, a total of 64 samples from patients who were prospectively recruited before and after anesthesia were used to confirm associations with anesthesia. The samples were obtained from individuals who visited the Seoul National University Hospital for treatment. Among these patients, those who agreed to have tissues sampled before treatment were approved by the Institutional Review Board (IRB) of Seoul National University Hospital (IRB No. 1603-053-748); the study protocol adhered to the principles of the Declaration of Helsinki. Blood samples from healthy individuals who visited the Seoul National University Hospital Healthcare System Gangnam Center for routine health check-ups were also obtained. The three proteins selected as early breast cancer biomarkers at this institution are presented in Supplementary Table 1 and further described in Supplementary Table 2. These three biomarkers exhibited changes in concentration in those with breast cancer; however, no specific changes were observed in those with other cancers. Previous studies have identified changes in the blood levels of these three markers in other cancers.

Quantitative analysis of the three proteins was performed using commercially available software (Analyst version 1.6, AB SCIEX, Framingham, USA) and reagent solutions (dithioerythritol [DTT], iodoacetamide [IAA], urea, and trypsin). A mass spectrometer (API 5000, AB SCIEX [medical device license No. Seoul, Korea: 10-1245]) was used to perform liquid chromatography with tandem mass spectrometry (LC-MS/MS) in multiple reaction monitoring (MRM) mode.

In a tube with 2 mg of synthesized standard peptide, 1 mL of 20% acetonitrile and 0.1% formic acid solution were added, the sample was vortexed for 5 minutes, and centrifuged for 30 seconds. The stock solution (2 mg/mL) was prepared by sonication in an ultrasonic grinder for 5 minutes and centrifuged at 12,000 rpm for 5 minutes. The working solution was diluted 1:50 from the stock solution and stored at −80°C or lower before use. The sample was lysed in lysis buffer (8M urea, 75 mM NaCl, 50 mM alkylation buffer [ABC]), and protein was quantified using the bicinchoninic acid assay to calculate the volume needed for 200 μg of protein (2.5–4 μL). To reduce cysteine disulfide bonds, 100 mM DTT was added. After the reaction, the sample was alkylated with 200 mM IAA to prevent disulfide bonding. The test sample was diluted 1:10 with 50 mM ABC. Trypsin (20 μg/400 μL) was added at a concentration of 1:50, and a digestion reaction was performed in a 37°C incubator for 16 hours. After trypsinization, the sample was centrifuged at 4°C for 5 minutes, and 10% TFA was added to the sample to prevent further trypsinization. After the sample was cleaned using a spin column (C18), the sample was lyophilized and stored at -80°C or lower before use. To generate a standard calibration curve, 10 pooled samples were mixed with 10 synthetic peptides (0.05–500 ng/mL), 5% acetonitrile (ACN), and 0.1% formic acid (FA). The mixture was pulverized in an ultrasonic mill for 5 min and centrifuged at 12,000 rpm at 4°C for 5 minutes. After transferring 80 μL to a vial, 2 μL of the solution was injected into the LC-MS/MS instrument. Data were analyzed using peak integration in the Analyst program. Calibration curves were generated using spreadsheet software (Excel, Microsoft Corporation, Redmond, USA).

Analytical performance tests were performed using pooled plasma samples before sample analysis. The calibration curve was generated using the pooled plasma samples and linearity was evaluated. Quality control of the sample (low, mid, or high) and the lower limit of quantification (LLOQ) concentrations were then determined. From this concentration value, the specificity, precision, and accuracy were determined. Data were analyzed using peak integration and analyst software (AB SCIEX).

Plasma samples from each group were subjected to a plasma pretreatment method and analyzed using MRM and analyst software. Subsequently, data were evaluated using the breast cancer biomarker algorithm.

The diagnostic accuracy of the proteomic assay used to quantify levels of three proteins (i.e., CAH1, APOC1 and NCHL1) was optimized using plasma samples from 100 breast cancer patients and 100 healthy controls. After optimization, the sensitivity, specificity, and accuracy of the proteomic assay were 67.0% and 82.0%, and the area under the receiver operating characteristic curve (AUC) was 0.771, respectively (Supplementary Table 3). Based on this, the number of samples needed for the final confirmatory experiment was determined. This result from the optimization process was used as a basis for sample calculation in further validation studies.

A total of 305 plasma samples, 183 from breast cancer patients and 122 from healthy controls, was used to validate the optimized algorithm. Using the proteomic assay, 131 of 183 breast cancer patients (71.6%) were classified with cancer, and 104 of 122 healthy controls (85.3%) were classified as normal. The positive predictive value was 82.9% and the negative predictive value was 75.0%. The statistic assumption of sensitivity and specificity for validation was satisfied.

Whether the diagnostic accuracy of the proteomic assay varied according to disease severity was also investigated. Among the 183 breast cancer patients, 45 exhibited axillary lymph node metastasis, all of which were pN1. The sensitivity of the proteomic assay in patients without lymph node metastasis was 71.7% and 73.3% in those with metastasis (Table 1). Regarding the molecular subtypes of breast cancer, the sensitivity for luminal A, luminal B, human epidermal growth factor receptor 2, and triple-negative breast cancers were 72.1%, 57.1%, 60.0%, and 73.9%, respectively (Table 1). In addition, the diagnostic accuracy of the proteomic assay when using serum samples from patients was investigated using an additional cohort of 460 serum samples (cancer 228, normal 232). The diagnostic accuracy of the proteomic assay was maintained, with an AUC of 0.8831.

To determine whether the performance of the proteomic assay described above was specific to breast cancer or applicable to other types of cancer, blood samples from patients who were diagnosed with other types of malignancies were collected. A total of 100 plasma samples from patients with thyroid, lung, colon, pancreatic, or ovarian cancers were analyzed using the proteomic assay. The proteomic assay failed to identify patients with other types of malignancies, suggesting that the accuracy of the assay was organ-specific and unique to patients with breast cancer (Table 2).

Because samples from the breast cancer patients were collected during the perioperative period, it is possible that the induction of general anesthesia may have affected the circulating levels of the three target proteins. To address this, the levels of the three proteins before and after anesthesia were compared. Only CAH1 demonstrated a significant difference according to the use of anesthesia; the levels of the other two proteins (i.e., NCHL1, APOC1) were not significantly altered (Table 3).