In this study, 541 formalin-fixed paraffin-embedded (FFPE) samples were collected from the patients with no special type invasive breast carcinoma at the Department of Pathology, the First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China from 2010 to 2017. All samples were taken by surgical excision before the patients receiving chemotherapy or radiotherapy. This study obtained written informed consent from all the patients and was approved by the Ethics Committee of the First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China.

Surgically resected specimens were fixed in 10% neutral buffered formalin with the cold ischemia time not exceeding 30 min. Then, they were processed according to standard pathologic procedures. In general, after retrieving, the specimens were embedded in paraffin using Leica ASP300S Fully Enclosed Tissue Processor (Heidelberger, Leica Biosystems Nussloch GmbH) according to the protocol recommended by the manufacturer. 4 μm and 2 μm thick sections were cut and routinely stained with hematoxylin-eosin (H&E) and immunohistochemical assay.

Disease stage and tumor grade were recorded according to the American Joint Committee on Cancer (AJCC) TNM staging system of the 8th edition and WHO Classification of Tumors of the Breast (5th edition, 2019) respectively [24, 25]. The Nottingham Prognostic Index (NPI) predicts the prognosis of the breast cancer patients based on tumor size, tumor grade, and lymph node status [26]. Progression free survival (PFS) was defined as the time from surgery date until first recurrence, second primary tumor, last follow-up or death from any cause. Meanwhile, overall survival (OS) was defined as the time from surgery until the last follow-up or date of death. In this study, 320 patients were followed up with the median time of 35 months (range 5 to 97 months).

RCs are similar to lymphatic or blood vessels, surrounding tumor glands or nests, but without lining endothelium. The extent of RCs was determined by evaluating the proportion of clefts that affected the tumor nests in the whole section. For example, tumors with clefts that affected approximately 10% of tumor nests were classified as 10% RCs. Two pathologists independently evaluated H&E sections to estimate the extent of RCs in invasive breast carcinoma and the average score was given to each sample. Lymphatic/vascular invasion and thermal damage to tissues were not considered to be RCs. Invasive micropapillary carcinomas and invasive carcinomas with micropapillary components were excluded as well. To further reduce the evaluation bias, lower than 5% RCs were counted as RC negative in this study. After the evaluation of RCs, the receiver-operating characteristic (ROC) curves were calculated based on the clinicopathological features and prognosis. The factor with the largest area under ROC curve (AUC) was employed to define the predictive cut-off value of the extent of RC which was used in the following analyses.

According to the expression of ER, PR, HER2 and Ki-67, invasive breast carcinoma can be grouped into four molecular subtypes: Luminal A-like, Luminal B-like, Triple-negative, and HER2-enriched [27]. ER, PR, HER2 and Ki-67 were evaluated by immunohistochemistry (IHC). The criterion of hormone receptor positivity is that the positive staining of tumor nucleus is greater than or equal to 1%, while HER-2 positive means 3 + on immunohistochemical staining or in those reported as 2 +, amplification on fluorescence in situ hybridization according to the current WHO guidelines [28].

The ROC curve was generated by GraphPad Prism software v5.01 (GraphPad, San Diego, CA, United States), the corresponding AUC, 95% confidence interval (CI) and p value were calculated together with the sensitivity and specificity table. The Pearson χ² test was used to compare the relations between the extent of RCs and clinicopathological features. The effect of various tumor variables on prognosis was researched by univariate analysis using the Cox proportional hazards model. The Kaplan-Meier survival curves were plotted using GraphPad Prism software v5.01 (GraphPad, San Diego, CA, United States), and the log-rank test was used to determine significant differences. p < 0.05 was considered statistically significant. SPSS software v19.0.0 (IBM, NY, United States) was used for statistical analysis.

Among the 541 breast cancer patients from Central China, RCs were detected in 84 cases (15.5%) (Table 1; Supplementary Table S2). The extent of RCs in these specimens ranged from 5% to 98% with median value of 60% (55.93% ± 23.31%; mean ± SD). Firstly, we determined the best cutoff value of the extent of RCs for our study. As shown in Supplementary Figure, the ROC curves were plotted under various clinicopathological characteristics and prognosis. The areas under the ROC curve (AUC) were between 0.5003−0.5217 (Supplementary Table S1), which showed no diagnostic discriminatory value. Even though, we set the cutoff point as 75% RCs (specificity = 96.83%, CI = 94.41%–98.22% and sensitivity = 8.287%, CI = 5.086%–13.22%) based on the largest AUC from HER2 assay group. Using this cutoff value, the association between RCs and clinicopathological features was analyzed. As shown in Table 1, specimens with >75% RCs were more common in HER2-enriched molecular subtype (p = 0.008), ER negative subgroup (p = 0.024) and HER2 positive subgroup (p = 0.024). In contrast, no statistically significant association was identified between the 75% RCs and patients’ age (p = 0.231), tumor stage (p = 0.659), lymph node status (p = 0.226), tumor size (p = 0.720), PR status (p = 0.366), or Ki67 index (p = 0.996).

As all the AUC values were not clinically useful, we also examined the associations between RCs and clinicopathological features at different classification methods. Some significant relationships were identified in these assays. Such as the presence of RCs showed statistically significant association with molecular subtypes (p = 0.007), which probably was due to the rare RCs in triple-negative breast cancer (Table 1).

As shown in Supplementary Table S2 Assay I, significant associations were found between the extent of RCs and molecular subtypes (p = 0.024) and ER-status (p = 0.005). It seemed that ≤25% and 50%–75% RC groups had more luminal A-like samples, and >75% RC group had more HER2-enriched and ER-negative samples. However, no particular pattern was found in this assay. While the extents of RCs in tumor nests increased, the numbers of the luminal A-like, HER2-enriched and ER-negative samples fluctuated dramatically with no positive correlations. When stratifying the samples by 25% RCs in Assay II, significant correlation between RCs and molecular subtypes (p = 0.038) was detected (Supplementary Table S2). However, all the HER2-enriched samples were present in the >25% group, whereas ≤25% group had no HER2-enriched samples, which may be the main cause of the positive correlation. The same result was obtained in Assay Ⅲ as in Assay II, demonstrating that HER2-enriched subtype of breast cancers appeared only in tumors with a greater extent of RCs (>30%) (Supplementary Table S2). Similarly, tumors larger than 5 cm were exclusively located in the >25% and >30% groups in Assay II and Assay III, respectively. Interestingly, according to Assay III, it seemed to be a trend of increasing tumor size along with the increasing extent of RCs despite the lack of statistically significant difference between groups (p = 0.064). Besides, significant association between the extent of RCs (≤40% group vs. >40% group) and tumor sizes was identified in Assay IV, while ≥5 cm tumors only appeared in >40% RC group (Supplementary Table S2).

Taken together, despite statistical significances were identified in some minor points, no particular pattern of clinicopathological features was observed when classifying the extent of RCs with different settings. Therefore, no constant conclusion could be made based on the results of the above assays concerning the correlation between the extent of RCs and clinicopathological factors.

More directly, as shown in Figure 1, two patients showed similar clinicopathological characteristics (ER, PR, HER2 and Ki-67 staining) while they had entirely different extent of RCs. Therefore, we further examined all the samples and found that only 12% (8/68) of them had coincident RCs and clinicopathological phenotypes. However, even these 8 patients could be divided into 4 groups based on those phenotypes. Thus, these results demonstrated that no convincing data could support the pathological value of RCs in breast cancer.

NPI is widely used to predict the prognosis of breast cancer patients, which is divided into good (2–3.4), moderate (3.4–5.4) and poor (>5.4) classes [26]. As listed in Table 1, there was no significant association between the presence of RCs and prognosis. Similar results were obtained in the six assays of patient groups with different extent of RCs (Table 1 and Supplementary Table S2).

Furthermore, prognosis analysis was performed on the progression free survival (PFS) and overall survival (OS) time of 320 patients with a median follow-up period of 35 months. The Cox proportional hazards model and Kaplan-Meier survival curve were employed to evaluate the correlation between the extent of RCs and PFS or OS of breast cancer patients.

In univariate Cox regression analysis, tumor stage Ⅰ and Ⅱ (p = 0.048) correlated with a better OS. However, the presence of RCs showed no association with PFS (HR [95% CI] = 1.356 [0.792–2.322], p = 0.267) and OS (HR [95% CI] = 1.459 [0.769–2.766], p = 0.247) Supplementary Table S3 and Figures 2A,B). Considering that the most dramatic significances were identified in Assay IV when analyzing the association between the extent of RCs and standard clinical, pathological or biological features, the samples were stratified by 40% RCs and the univariate analysis was applied. However, no statistical significance was detected (Supplementary Table S3). In addition, the cutoff point of RCs (75%) was also analyzed, which indicated no significant relationship with PFS (HR [95% CI] = 0.828 [0.358-0.917], p = 0.660) and OS (HR [95% CI] = 0.779 [0.279–2.179], p = 0.635) as well (Supplementary Table S3 and Figures 2A,B).

Multivariate Cox analysis was also employed to assay the correlation between clinicopathological parameters and PFS/OS in this study. As shown in Supplementary Table S4, Figures 2C,D, only tumor stage correlated with both PFS (p = 0.009) and OS (p = 0.015) which indicated better prognosis in I and II tumor stage. However, 75% RCs was not a correlated factor for prognosis (PFS: HR [95% CI] = 1.215 [0.451–3.275], p = 0.700 and OS: HR [95% CI] = 0.838 [0.233–3.017], p = 0.787).

Considering the limited number of patients in >75% RCs group and the possible real effect of the RCs to prognosis under each clinicopathological factors, patients were divided into positive and negative groups according to clinicopathological parameters, and Cox analysis was performed to determine the correlations between prognosis and the presence of RCs or the extent of RCs (40% and 75% RCs). A significant difference in PFS was observed between lymph node positive and negative subgroups (p = 0.044), suggesting a correlation between prognosis and the presence of RCs (Supplementary Table S5). Meanwhile, worse PFS and OS was exhibited in >75% RCs group in the patients younger than 45 years when their tumors were diagnosed (p = 0.055 and 0.006, respectively) (Supplementary Table S5). Except these, no significance was identified in patients of different clinical, pathological and molecular subtypes.

Using the Kaplan-Meier method and log-rank test, the PFS and OS were examined and no difference was found between the RC’s presence group and RC’s absence group (p = 0.267 and 0.247, respectively) (Figures 3A,B). Interestingly, the presence of RCs significantly improved the PFS of the patients with lymph nodes metastasis (p = 0.044), while it seemed to benefit the OS of these patients although without statistical significance (p = 0.072) (Figures 3C,D). When examined by 75% RCs, the younger patients (≤45 years old) with higher RCs (>75%) had worse PFS and OS (p = 0.055 and 0.006, respectively) (Figures 3G,H). Meanwhile, no significantly different PFS or OS was identified when separated the patients by 75% RCs (p = 0.660 and 0.635, respectively) (Figures 3E,F). Therefore, there was no convincing evidence that the extent of RCs could predict the prognosis of breast cancer patients.