Growing evidence highlights the importance of tumor microenvironment in tumor progression [15,16,17]. The tumor microenvironment, especially the tumor stroma, has been recognized as an important driver of tumor progression . However, indicators of tumor stroma have not yet been integrated into routine clinical decision-making. A novel parameter reflecting the content of tumor-associated stroma is TSR, which has been extensively described as a rich source of prognostic information for various solid cancer types. The biological role of TSR is recorded in different cancer types [19,20,21]. Previous research has focused on the prognostic effect of TSR, as a high proportion of tumor stroma is often associated with poor prognosis [22, 23]. However, the results were not always consistent in different molecular subtypes.
In the present study, the prognostic value of TSR was assessed in 240 patients with breast cancer, and the clinically relevant subgroups for breast cancer prognosis were analyzed. The main findings of this study are as follows. First, patients with a stroma-high tumors (low TSR) had worse survival outcomes compared with patients with stroma-low tumors (high TSR). Second, the prognostic value of TSR was not affected by age at onset, histopathological grade, lymph node status, ER status, PR status, HER2 status, menopausal status, or tumor size. Third, subgroup analysis according to the molecular subtypes suggested that the prognostic significance of TSR in the HER2-positive breast cancer, TNBC, and luminal-HER2-negative breast cancer subgroups did not differ from the prognostic value in the total breast cancer series. Fourth, TSR was a continuous variable. Categorical and continuous variable analyses were performed on TSR, and both analyses gave similar results on the performance of TSR for prognosis prediction.
The prognostic value of TSR in breast cancer has been reported in several studies. TSR was first identified as a prognostic factor in breast cancer by Kruijf et al, who demonstrated that patients with stroma-rich tumors, especially TNBC, have a higher risk of recurrence . Dekker et al. assessed the prognostic value of TSR in premenopausal patients with breast cancer with node-negative status, the results indicated that TSR is an independent prognostic parameter for DFS and is independently associated with locoregional recurrence; this finding validated the prognostic value of TSR in breast cancer . The prognostic value of TSR was also confirmed in ER-positive breast cancer  and inflammatory breast cancer . In line with these previous studies, our research revealed that TSR was positively associated with 5-DFS, and patients with stroma-high tumors (low TSR) have worse survival outcomes than patients with stroma-low tumors (high TSR).When TSR was assessed as a categorical variable, univariate and multivariate analyses showed that TSR was associated with 5-DFS. When TSR was analyzed as a continuous variable, univariate and multivariate analyses revealed that TSR was associated with 5-DFS in the total breast cancer series. However, in the subgroup analysis, TSR was not statistically significant in all of the three subtypes in the univariate analyses but was significant in the multivariate analysis. The different outcomes could be because some clinical pathological parameters were not considered in the univariate analysis. The meaningless values may be caused by confounding factors . Multivariate analysis considers the influence of multiple factors and excludes confounding factors, resulting in different results. Moreover, multivariable analysis showed that histopathological grade, nodal status, tumor size, HER-2 status, and TSR were independent prognostic factors of breast cancer.
According to the recommended TSR evaluation method, the tissue section with the most invasive part of the primary tumor should be selected to evaluate TSR [24, 29]. TMAs can quantify features and emphasize the extent of the tumor-stroma interface, it is also suitable for large sample detection, computer recognition, and automatic analysis . In the process of TMA construction, the tumor block representing the deepest tumor infiltration into the wall was selected, and two cores were sampled from each donor tumor to ensure the reproducibility and homogenous staining of the slides. These conditions can make the TSR evaluation results more objective and accurate.
A reliable tumor area assessment method is the basis for exploring the prognostic value of TSR. Visual inspection by experienced pathologists  and computer assessment analysis  are the two main methods for assessing TSR. Currently, TSR is largely assessed in the H&E staining section, which may not accurately identify the boundary of tumor nests because of the low contrast between tumor and stroma in some regions. These factors may affect the reproducibility of results and make accurate identification and analysis difficult. CK IHC staining can label tumor cells and can be used to distinguish tumor cells from stromal cells . Therefore, the IHC staining of CK was used to specifically label tumor cells in this study, and the results showed a strong color contrast, in which tumor cells were marked in brown and tumor stroma were marked in off-white. Digital image quantification analysis was applied to TSR assessment based on the IHC staining of CK. In this study, the optimal cut-off TSR value was 0.335, which corresponds to approximately 66.5% of stroma and 33.5% of tumor. The stromal value cut-point of 66.5% was slightly higher than the predefined cut-point for 50% stroma in other studies. These results are similar to a previous study, which found that in colorectal cancer, the cutoff point of stroma value based on convolutional neural networks is higher than the 50% stroma visual assessment . The differences in cutoff points between visual assessment and computer assessment suggested that there may be a common discrepancy between humans and computers when evaluating tumor pathology images.
In our study, the optimal cut-off TSR value determined by maximally selected rank statistics, which was used to distinguish the stroma-low and stroma-high groups. However, it should be emphasized that the proportion of tumor stromal content is a continuous variable, which can reach any proportion from 0% to nearly 100%. This suggests that TSR is a continuous parameter of tumor cell interaction with stromal components rather than a marker of a specific tumor subtype. In other words, TSR could be viewed as an indirect measure of the stroma’s contribution to malignant progression, which might be similar to the proliferation marker Ki67 to some extent. The results of the continuous variable analysis of TSR were consistent with a previous study, showing that in endometrial carcinoma, TSR is associated with poorer survival outcomes when used as a continuous variable . The continuous analysis of tumor stromal content is more relevant from a statistical point of view and provides a more accurate description of tumor biological behavior. In future research, if TSR can be automatically quantified from H&E sections, it will give a more accurate description of tumor biology, which will lead to more accurate individualized treatment and prognosis prediction for patients. Nonetheless, dividing TSR into stroma-high and stroma-low groups might be more practical for future clinical applications, because clinical trials are easier to implement in groups of patients based on categorical variables. Therefore, in the present study, continuous and categorical variables were analysed, and both analyses gave similar prognostic prediction results.
Another important question is how the tumor stroma contributes to the prognosis of breast cancer. It is known that TSR is composed of fibroblasts, immune cells, endothelial cells, and other supporting cells. These cells could be recruited by cancer cells from nearby endogenous host stroma, which could in turn promote tumor angiogenesis, proliferation, metastasis, and invasion . Previous studies have identified events that occur in the stromal compartment during carcinogenesis, including fibroblast recruitment, stroma remodeling, immune cells migration, and angiogenesis, which may influence tumor progression [36, 37]. To date, studies on the effect of tumor-associated stroma on epithelial tumor progression have largely focused on functional in vitro studies. Cancer-associated fibroblasts is one of the factors critically involved in cancer progression. They regulate the biological function of stromal and tumor cells via intercellular contact, synthesize and remodel the extracellular matrix, elevate the proliferation rate, release numerous cytokines (such as vascular endothelial growth factor and stromal cell-derived factor 1) that lead to angiogenesis, and thus promote cancer initiation and development [37, 38]. These tumor-associated stromal cells also secrete a number of pro-tumorigenic factors, such as stromal-derived factor-1α, IL-6, IL-8, vascular endothelial growth factor, matrix metalloproteinases, and tenascin-C. These factors recruit additional tumor and pro-tumorigenic cells into the developing microenvironment, which may in turn contribute to tumor progression .
However, several limitations should be recognized in the present study. First, the present cohort was retrospectively assembled. The number of patients included in this study was relatively small, especially the number of patients with TNBC and HER2-positive breast cancer. Second, a period of 10 or 15 years is commonly used in breast cancer research. The follow-up period of this study was not long enough owing to the lack of available follow-up data. Previous study found that the high-risk period of breast cancer recurrence is 1–2 years after surgery, and the risk of recurrence decreases rapidly within 2–5 years and returns to a stable period within 5–12 years . These findings indicated that 5-DFS might reflect the primary endpoint to some extent. Third, although the most invasive parts of the primary tumors were selected to construct the TMAs, it might not be representative for the amount of stroma in the entire resected tissue specimen, as different sampling approaches would remarkably impact TSR. Future whole-slide H&E-stained image analysis could reduce or eliminate variations in the TSR assessment results. Therefore, a large, updated retrospective study, which takes into account an appropriate follow-up period, and/or a prospective cohort study with TSR values evaluated on whole-slide H&E-stained images should be implemented to validate the prognostic value of TSR in the next clinical implementation.