2.1 Study design

We consecutively included pre-treatment participants including inpatients with benign breast tumors, inpatients with breast cancer, and normal controls to a pre-treatment cohort from Tongji Hospital of Huazhong University of Science and Technology (Wuhan, Hubei, China), Qilu Hospital of Shandong University (Jinan, Shandong, China), and Hubei Cancer Hospital (Wuhan, Hubei, China) between January 1, 2013, and September 30, 2017 (Figure 1). In this study, pre-treatment patients refer to patients who did not undergo any kind of treatment (i.e. surgery, chemotherapy, radiotherapy and target therapy) prior to the blood sample collection, and non-breast cancer controls refer to patients with benign breast lesions and healthy participants. The pre-treatment cohort was divided into a training cohort, validation cohort 1, and validation cohort 2 to determine the optimal cut-off value of serum SEMA4C for breast cancer diagnosis and performance evaluation. The training cohort comprised of patients admitted in Tongji Hospital and Qilu Hospital of Shandong University between January 2013 and June 2016. The validation cohort 1 comprised patients admitted in Tongji Hospital between July 2016 and September 2017. The validation cohort 2 included patients admitted in Hubei Cancer Hospital between July 2016 and June 2017. We also recruited pretreated participants, including patients with breast cancer, patients with other 14 types of solid tumors, and normal controls, to a pan-cancer cohort from 15 cancer centers in Tongji Hospital and Qilu Hospital between July 1, 2017, and June 30, 2018 (Figure 1).

2.2 Study eligibility

Consecutive inpatients with breast lesions and other 14 types of solid tumors, including cervical cancer, pancreatic cancer, gastric cancer, liver cancer, kidney cancer, ovarian cancer, lung cancer, prostate cancer, thyroid cancer, colorectal cancer, brain cancer, esophageal cancer, bladder cancer, and endometrial cancer, from the Surgery Department of the three hospitals were included. The normal controls were volunteers from the Physical Examination Center who were admitted for routine health examinations during that same period. Eligible individuals in the pre-treatment or pan-cancer cohort did not undergo any treatment prior to the blood sample collection. Patients with more than two of the 10 key clinical features missing (age, pathologic diagnosis, histological type, tumor grade, tumor size, lymph node status, metastasis status, estrogen receptor [ER] expression, progesterone receptor [PR] expression, and human epidermal growth factor receptor 2 [HER2] status) and with active inflammatory, autoimmune, renal or liver diseases were excluded. The eligibility criteria also included participants aged over 18 and not having cancer histories other than the investigated ones. Pregnant or lactating women were also excluded from this current study.

Pathological diagnosis after radical surgeries or mass resections was retrieved from the electronic health records for benign breast tumors, breast cancers and other types of solid tumors. Diagnosis was made based on the World Health Organization Classification of Tumors of the Breast Guidelines [25]. The pathologists had no knowledge of the interpretations of patients’ imaging results or serum SEMA4C concentrations, and the pathologic diagnosis was used as the reference standard for diagnosing the patients’ disease. The benign breast lesions included fibroadenoma, intraductal papilloma, atypical ductal hyperplasia, benign phyllodes tumors, lipomas, and hamartomas. Invasive breast cancer and DCIS were categorized as breast cancers. The breast cancer stage was determined according to the Tumor-Node-Metastasis (TNM) classification of the Union for International Cancer Control (7^th edition) [26], with stage I (T1N0M0) breast cancer considered as the early stage of the disease.

2.3 Data collection

The clinical information of participants was retrospectively collected from electronic health records by trained investigators after the pathologic diagnosis was obtained and stored in the database of the clinical research center in Tongji Hospital. Data collection was undertaken by two researchers independently (LQ and XL) and supervised by a third investigator (YW).

2.4 Sample collection

The pre-treatment intravenous blood sample was collected into an anticoagulant-free tube (vacuum container) within 48 h prior to treatment. Post-treatment intravenous blood samples were collected 2–8 days after performing breast mass excision or modified radical mastectomy. After centrifugation at 3000 rpm for 10 min, the serum sample was stored at −80°C at each study center. Collection and separation were standardized across centers, with agreed-upon study-specific operating procedures. The serum samples were then shipped on dry ice to the central laboratory of the Cancer Biology Research Center of Tongji Hospital for analysis.

2.5 Chemical materials for enzyme-linked immunosorbent assay (ELISA)

Monoclonal anti-human SEMA4C antibody (#MAB6125, Research & Diagnostics Systems, Incorporated, R&D Systems Inc., Minneapolis, MN, USA) and anti-human SEMA4C antibody (#AF6125, R&D Systems Inc.) were used for analyses. SEMA4C recombinant protein was provided by R&D Systems Inc. (#6125-S4) and was used as the standard protein. The biotin-labeled detection antibody was prepared using a Biotin Labeling Kit-NH2 (Dojindo Molecular Technologies Inc., Kumamoto, Japan).

2.6 Measurement of serum SEMA4C level

Assays for measuring the serum SEMA4C levels were conducted by two researchers (JY and QSC) at the Cancer Biology Research Center of Tongji Hospital who were blinded to the clinical information and pathologic diagnosis of the participants. The serum SEMA4C levels were measured with a double-antibody sandwich ELISA method using in-house SEMA4C detection kits [23]. The ELISA Kit preparations and assays were performed according to the ELISA Development Guide of the R&D Systems Inc. Briefly, 96-well Nunc-immunomicrotiter plates with MaxiSorp surface (Greiner, Germany) were coated with 100 μL of sheep anti-human SEMA4C antibody (4 μg/mL, #AF6125, R&D Systems Inc.) and incubated at 4°C overnight. The reaction was blocked with 1% bovine serum albumin. Sera were diluted by 10-fold and incubated for 2 h at 37°C. The detection antibody, biotinylated mouse anti-human SEMA4C (0.4 μg/mL, #MAB6125, R&D Systems Inc.), was incubated for 2 h at 37°C, followed by the addition of 100 μL at a 1:200 dilution of streptavidin-horseradish peroxidase for 20 min. Color development was achieved by adding 100 μL per well of 3,3,5,5-tetramethylbenzidine and hydrogen peroxide as a substrate, and sulfuric acid (1 mol/L) was added to stop the reaction. The optical density was measured at 450 nm and referenced to 570 nm on a SpectraMax190 plate reader (Molecular Devices, CA, USA). The SEMA4C levels were obtained using linear regression analysis and then fitted to the standard value and multiplied by the dilution factor. When the serum SEMA4C level was <0.1953 ng/mL (the lowest limit of the standard curve), the value was set to zero. All measurements were performed in duplicate. The standard curve covered a range of 0.1953–50.0000 ng/mL. Our in-house SEMA4C ELISA kit was optimized by evaluating the calibration curve, detection limit, recovery, cross-reactivity, dilution linearity. Variations in intra-assay, inter-assay, and day-to-day precision studies were assessed. All measurements were performed in duplicate, and the average values obtained by the two performers were used in the analysis.

2.7 Immunohistochemistry analysis

Specimens from normal/benign-hyperplasic mammary glands (18 cases) and primary invasive breast carcinoma (22 cases) were acquired during surgery as approved by the Ethical Committee of the Medical Faculty of Tongji Medical College (Wuhan, PR China). Tumor specimens were acquired from patients with cancer who had not undergone preoperative radiotherapy or chemotherapy. Tissue sections were subjected to immunohistochemical analysis using the avidin-biotin complex Vectastain Kit (Zsgb-Bio, Beijing, China) according to the manufacturer's protocol. Anti-human CD4 (ab67001, Abcam), anti-human CD8 (ab93278, Abcam), anti-human CD68 (ab213363, Abcam), and anti-human SEMA4C (AF6125, R&D Systems) antibodies were used as primary antibodies. Fixed positive and negative controls were evaluated in each experiment to control the staining variability among batches of experiments. An immunoreactivity-scoring system (HSCORE) was used for the semiquantitative evaluation of protein levels in tissues. Briefly, staining intensity was graded as follows: 0, absence; 1, weak; 2, moderate; 3, strong. The HSCORE score was calculated using the following formula: HSCORE = ∑Pi × i, where i is the staining intensity of immunocytes and Pi is the percentage of corresponding cells at each level of intensity. An HSCORE score of < 1.5 indicated a low protein level while an HSCORE score of ≥ 1.5 indicated a high protein level. Each data point represents the mean score assigned by two pathologists, who were blinded to all clinicopathological variables.

2.8 Statistical analysis

Continuous variables are presented as mean ± standard deviation (SD) or median and interquartile range (IQR), while categorical variables are expressed as absolute frequencies and percentages. The sample size was calculated based on a sensitivity of 80% and a specificity of 90% using the diagnostic test as the present study was not a randomized clinical trial and did not compare other risk factors. The differences in serum SEMA4C values and other continuous variables between the two independent groups were tested using the Mann-Whitney U test. Pearson's chi-square test or Fisher's exact test was used to analyze the association between pre-treatment serum SEMA4C levels and clinical or pathological factors. A receiver operating characteristic (ROC) curve was developed to determine the area under the curve (AUC). The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated with their corresponding 95% confidence intervals (CIs). PPV is a diagnostic statistic that detects the presence of diseases and is calculated as follows: PPV = true positive cases/(true positive cases + false-positive cases). In contrast, NPV is a diagnostic statistic that detects the absence of diseases and is calculated as follows: NPV = true negative cases/(true negative cases + false-negative cases). The optimal cut-off value for the training cohort was obtained by maximizing the sum of sensitivity and specificity, minimizing the overall error [square root of the sum of (1−sensitivity)² + (1–specificity)²], and minimizing the distance of the cut-off value to the top-left corner of the ROC curve [27]. All statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS v23.0, IBM Corp., NY, USA) and R software (v3.4.1; www.r-project.org). Statistical significance was set at P < 0.05.

3.1 Baseline characteristics of the study participants

Overall, 6213 participants were consecutively included (Figure 1): 1215 pre-treatment patients were assigned to the training cohort, 685 pre-treatment patients were assigned to the validation cohort 1, 536 pre-treatment patients were assigned to the validation cohort 2, and 3777 pretreatment cases were assigned to the pan-cancer cohort, which included healthy participants, patients with breast cancer, and those with other 14 types of solid tumors.

The demographic and clinicopathological characteristics of the training cohort, validation cohort 1, and validation cohort 2 are summarized in Table 1. The overall frequencies at diagnosis of stages I, II, and III were 34.49%, 47.20%, and 9.38% in the training cohort; 22.59%, 38.86%, and 10.84% in validation cohort 1; and 24.58%, 50.83%, and 11.67% in validation cohort 2, respectively. Only four patients diagnosed with stage IV disease were included, two from the training cohort and two from the validation cohort 1. The training cohort and validation cohorts were matched for the mean age between breast cancer patients and normal controls; patients with benign breast tumors were younger than those with cancerous tumors as benign breast tumors were observed to develop at a younger age (P < 0.05).

3.2 Diagnostic value of serum SEMA4C in breast cancer

To accurately measure the serum SEMA4C, we optimized the in-house SEMA4C ELISA detection system by evaluating the calibration curve, detection limit, recovery, cross-reactivity, and dilution linearity (Supplementary Figure S1). The variation in intra-assay, inter-assay, and day-to-day precision studies was limited to less than 10% (Supplementary Table S1).

In the training cohort, the serum SEMA4C levels were significantly higher in patients with breast cancer (7.46 ± 2.69 ng/mL) than in those with benign breast tumors (3.51 ± 1.60 ng/mL) and normal controls (3.19 ± 0.96 ng/mL; both P < 0.001) (Figure 2A, Supplementary Table S2). Meanwhile, the serum SEMA4C levels were not significantly different between patients with benign breast lesions and normal controls (P < 0.05). The optimal cut-off value of serum SEMA4C for diagnosing breast cancer was 5.00 ng/mL when an adjacent integer was chosen from the optimal cut-off values of 5.125 ng/mL in Tongji Hospital and 4.754 ng/mL in Qilu Hospital (Supplementary Table S3). The AUC was 0.938 (95% CI: 0.924–0.951), with a sensitivity of 84.4% and a specificity of 89.9% (Figure 2B and Table 2).

Similarly, the serum SEMA4C levels in the two validation cohorts were significantly higher in patients with breast cancer than in those with benign breast tumors and normal controls (both P < 0.001; Figure 2C and D and Supplementary Table S2), and the difference between benign tumors and normal controls was not significant (P > 0.05). Based on the determined optimal threshold in the training cohort, serum SEMA4C showed an AUC value of 0.920 (95% CI: 0.900–0.941), a sensitivity of 82.8%, and a specificity of 87.5% for the detection of breast cancer in the validation cohort 1 (Figure 2E and Table 2). For validation cohort 2, the AUC was 0.932 (95% CI: 0.911–0.953), with a sensitivity of 86.7% and a specificity of 87.8% (Figure 2F and Table 2).

In addition, the pre-treatment serum SEMA4C levels were not associated with most clinicopathological features in the training and validation cohorts, including tumor size, tumor grade, lymph node status, and ER/PR/HER2 status (Supplementary Table S4).

3.3 Diagnostic value of serum SEMA4C in early-stage disease and DCIS

Late-stage breast cancer is still prevalent in many countries. We examined whether increased serum SEMA4C levels could be used to identify early-stage breast cancer, considering that there is no available effective blood-based biomarker for the early detection of this disease. Table 2 shows that early-stage breast cancer could be accurately distinguished from non-breast cancer controls with high AUCs (training cohort, 0.936 [95% CI: 0.916–0.957]; validation cohort 1, 0.937 [95% CI: 0.913–0.960]; validation cohort 2, 0.929 [95% CI: 0.899–0.958]). DCIS accounts for 15% of all breast cancer types; however, no serum protein marker has been validated for DCIS diagnosis or screening [28]. In this present study, DCIS was diagnosed in 85 (6.89%) inpatients with breast cancer. When differentiating DCIS patients from non-breast cancer controls; serum SEMA4C yielded an AUC of 0.879 (95% CI: 0.832–0.925), sensitivity of 74.1%, and specificity of 89.7% (Supplementary Figure S2, Table 2, and Supplementary Table S5). Briefly, increased serum SEMA4C levels exhibited high accuracy in discriminating early-stage breast cancers/DCIS from non-breast cancer controls.

3.4 Measurements of serum SEMA4C in patients with breast cancer before and after surgery

To investigate whether serum SEMA4C could help measure the response to surgery in patients with breast cancer, we compared the SEMA4C levels before treatment with those after breast mass excision or modified radical mastectomy in paired samples. For the 140 patients who underwent modified radical mastectomy, the post-surgery serum SEMA4C level was significantly lower than the pre-treatment level (P < 0.001; Figure 3A, Supplementary Tables S6 and S7). They had undergone breast mass excision prior to modified radical mastectomy as they were misdiagnosed with benign breast lesions pre- or intraoperatively. Notably, the SEMA4C level after breast mass excision of these patients (6.48 ± 3.11 ng/mL) was significantly higher than that after modified radical mastectomy (5.26 ± 3.00 ng/mL) (P < 0.001; Figure 3A and Supplementary Table S7).

Meanwhile, 151 serum samples were collected from patients with breast cancer after undergoing modified radical mastectomy: 91 samples at day-2, 22 at day-5, and 38 at day-8. The mean serum SEMA4C level before surgery was 7.75 ± 3.25 ng/mL, and the values declined afterward (4.80 ± 1.85 ng/mL at day-2; 5.01 ± 2.19 ng/mL at day-5; and 4.36 ± 2.34 ng/mL at day-8; Figure 3B, Supplementary Table S7).

Generally, decreased serum SEMA4C levels were observed after modified radical mastectomy or breast mass excision, showing the potential of serum SEMA4C in assessing the response to surgery in breast cancer patients.

3.5 Measurements of serum SEMA4C in patients with solid tumors of 15 origins and healthy controls

To further explore whether SEMA4C could be used as a biomarker specific to breast cancer among common solid tumor patients, a pan-cancer cohort was established. The results showed that serum SEMA4C levels were significantly higher in patients with breast cancer than in those with other 14 types of solid tumors and normal controls (P < 0.001; Figure 3C and Supplementary Tables S8–S10). Compared with the normal controls, the serum SEMA4C levels were slightly higher in patients with cervical, pancreatic, gastric, liver, kidney, and ovarian cancers (P > 0.05), and slightly lower in patients with lung, prostate, thyroid, colorectal, brain, esophageal, bladder or endometrial cancers (P > 0.05). Based on a cut-off value of 5.00 ng/mL for serum SEMA4C levels, 84.77% (707/834), 15.75% (144/914), 19.44% (21/108), 12.75% (13/102), 12.61% (14/111), 12.62% (13/103), 9.09% (10/110), 9.65% (11/114), 20.75% (22/106), 7.89% (12/152), 8.08% (11/136), 8.65% 9(/104), 8.41% (9/107), 8.26% (9/109), and 7.84% (8/102) of the patients tested positive for breast, cervical, pancreatic, gastric, liver, kidney, ovarian, lung, prostate, thyroid, colorectal, brain, esophageal, bladder, and endometrial cancers, respectively. Generally, despite the differences in age distribution among female patients, high serum SEMA4C levels were specifically observed among patients with breast cancer (Supplementary Table S11).