The prognostic value of tumour–stroma ratio in primary breast cancer with special attention to triple-negative tumours: a review

The influence of the tumour-associated stroma on epithelial tumour progression is mostly derived from functional in vitro studies. Similarly, those in vitro studies have demonstrated events in the stromal compartment that occur during carcinogenesis and could contribute to tumour progression. The production of growth factors and proteases by cancer cells initiate changes in the stromal environment [39]. Those alterations lie within remodelling of the matrix, recruitment of fibroblasts, the migration of immune cells and angiogenesis, all contributing to tumour progression [40]. Cancer-associated fibroblasts (CAFs) contribute to carcinogenesis through the development of unique functions, including an amplified extracellular matrix (ECM) production, higher proliferation rate and the secretion of several cytokines, like vascular endothelial growth factor (VEGF), stromal cell-derived factor 1 (SDF1) and platelet-derived growth factor (PDGF), leading to angiogenesis [40]. Transforming growth factor-β (TGF-β) is another factor that is thought to be strongly involved in the tumour-promoting effects of CAFs as described in colon cancer by Hawinkels et al. [41]. Those behavioural modifications lead to an elevated expression of enzymes, like matrix metalloproteinases (MMPs), resulting in remodelling and deposition of the ECM, with concurrently the release of pro-angiogenic factors [42].

The ECM is frequently disorganised in tumours. One of the most important mechanisms in the ECM contributing to tumour progression is collagen crosslinking. Due to crosslinking collagen by lysyl oxidase (LOX), the ECM of the tumour becomes more stiff, leading to increased focal adhesions and enhanced PI3K signalling, thereby indirectly ensure tumour progression [43]. Besides the fact that alterations in the tumour niche lead to progression directly, the tumourigenesis can also be strengthened indirectly due to the aforementioned production of pro-angiogenic factors by CAFs and immune cells. Thus, during the process of tumourigenesis changes occur in the organisation of stromal cells, contributing both directly and indirectly to tumour growth and progression.

Previous studies investigating gene-expression profiles in stromal cells have demonstrated gene signatures related to clinical outcome and response to treatment in breast cancer [44, 45]. Clinical application of these signatures was impractical and a definitive indication was never discovered. These studies did provide a strong indication that valuable clinical information was ignored by solely focusing on the epithelial compartment. As the stromal processes that are reflected by these assays likely have a quantitative relationship with the amounts of stromal tissue within the tumour, quantitative stromal parameters might equally express prognostic information just by morphology alone [45].

In literature, two methods are described for TSR assessment in breast cancer: visual scoring, utilised by Mesker et al.; and automated point counting, a semi-automated approach, utilised by West et al. [10, 18].

The first study on TSR in breast cancer was published by De Kruijf et al. [12]. TSR was estimated by visual eyeballing according to the method described by Mesker et al. [10]. The authors showed that TSR was an independent prognostic parameter in 574 breast cancer patients with invasive breast tumours without distant metastasis (pT1–4, pN0–3, M0). Stroma-high tumours were associated with a worse RFP (Hazard Ratio (HR) 1.97, 95% Confidence Interval (CI) 1.47–2.64, P < 0.001) and overall survival (OS) (HR 1.50, 95% CI 1.18–1.91, P = 0.001) analysed with multivariate Cox regression analysis (Table 1) [12]. Vangangelt et al. analysed the prognostic value of TSR in a subset of the cohort of De Kruijf et al. in combination with immune status of tumours. Determination of classical human leukocyte antigen (HLA) class I, HLA-E, HLA-G, natural killer cells and/or regulatory T cells additional to TSR showed to have an even stronger prognostic effect [16].

Dekker et al. investigated the prognostic value of the amount of stroma determined by visual eyeballing in 403 premenopausal node-negative breast cancer patients (cT1–3) [14]. These patients were selected from the perioperative chemotherapy trial (POP trial, 10854) [48]. This study supported the earlier finding of TSR as an independent prognostic parameter for disease-free survival (DFS) (HR 1.85, 95% CI 1.33–2.59, P < 0.001) in favour of stroma-low tumours, and borderline statistical significance for OS (HR 1.60, 95% CI 1.00–2.57, P = 0.050) [14].

Gujam et al. investigated patients with invasive carcinoma of no special type (NST) (T1–3, N0–>3, grade I-III,) and subsequently found a correlation between stroma-high tumours and a poor 15-year cancer-specific survival (CSS) (HR 2.12, 95% CI 1.37–3.29, P = 0.001) in multivariate survival analysis after visual TSR assessment of H&E slides of 361 patients [15].

Downey et al. dispute this finding in his work by analysing the stromal content with semi-automated point counting [46]. They found that high tumour–stroma content in 118 women with ER+ grade 1–3 invasive breast tumours was independently associated with a better overall survival (OS) and relapse-free survival (RFS) (95% CI 0.2–0.7, P = 0.008 and 95% CI 0.1–0.6, P = 0.006, respectively) [46].

Following on their first study, Downey and colleagues investigated the stromal content in 45 patients with inflammatory breast cancer, a rare and aggressive form of breast cancer, using the semi-automated point counting method [47, 49]. However, no statistical significant difference was observed for this series (OS P = 0.53, DFS P = 0.66) [47].

Roeke et al. (T1–3, N0–2, grade I–III general BC) validated, by visual TSR assessment, that a high stromal content was a prognostic factor for worse OS (HR 1.56, 95% CI 1.18–2.05, P = 0.002), distant metastasis-free survival (DMFS) (HR 1.52, 95% CI 1.12–2.06, P = 0.008) and RFS (HR 1.35, 95% CI 1.01–1.81, P = 0.046), in their study of 737 patients with primary operable invasive breast cancer [17]. Unlike the work of Downey et al., patients with ER + stroma-high tumours were associated with a worse OS (HR 1.43, 95% CI 1.04–1.99, P = 0.030) [17].

De Kruijf et al., Gujam et al. and Roeke et al. described the role of TSR in other subgroups. The results of De Kruijf et al. showed an independent prognostic value of TSR in patients who received only local therapy (P < 0.001), only adjuvant chemotherapy (P = 0.038) and only adjuvant endocrine therapy (P = 0.024) [12]. The latter was confirmed by Roeke et al. (P = 0.001) [17]. The same results were seen in patients with a TN tumour who received only local therapy (P = 0.006).

In non-TNBC patients (P = 0.013), ER-positive patients (P = 0.030) and HER2-negative tumours, the TSR was also of independent prognostic value [12, 17]. This was not the case for ER- and PR-negative breast tumours [17]. The last subgroup in which TSR was evaluated and proved to be statistical significant was node-negative tumours (P = 0.002 and P = 0.003) [15, 17]. A summary of these results is presented in Table 2.

Extensive research has been performed to determine prognostic biomarkers for patient prognosis. Molecular tests, as MammaPrint and Oncotype DX, have seemed to be valuable for the improvement of clinical decision making in early-stage breast cancer [7, 8]. These tests will possibly be endorsed into routine clinical practice in order to reduce the administration of adjuvant chemotherapy and prevent overtreatment [9]. However, the disadvantages of the aforementioned molecular testing are the relatively high cost and the far mostly unknown influence of tumour heterogeneity. More specifically, intermingled non-tumour tissue may have a profound influence on the test results [50].

TSR has shown to be of prognostic value in addition to the traditional prognostic markers which are implemented in standard clinical care, for example, TNM stage, receptor status and HER2 expression, in breast cancer with a robust inter-observer variability. In Supplementary Tables 1 and 2, the effect of TSR in addition to the most important traditional prognostic markers is shown for the entire study population and triple-negative tumours, respectively.

So far, seven studies regarding TSR have been performed in the field of breast cancer, of which five have shown a significant association between high tumour–stroma content and a poor prognosis [12‐15, 17]. However, the results of both studies of Downey and colleagues were not in line with the other five [46, 47]. As Downey et al. have determined the TSR in both studies with semi-automated point counting instead of visual eyeballing and have utilised different cut-off values, it may be concluded that a standardised estimation of TSR is essential for a robust method, applicable for patient management. The method of determining TSR differed considerably, resulting in underestimating the heterogeneity [51]. In contrast with previous studies, where the final TSR category is based on the highest stroma rate in the sample, Downey and colleagues scored only an area of 9 mm² at the edge of the tumour [10, 46, 51, 52].

Although the difference in results can possibly be attributed to this inconsistency, the different breast cancer subgroups regarding basic characteristics must be taken in consideration as well. The applicability in the subtypes, namely TNBC, ER+ and inflammatory breast cancer, may differ and subsequently the individual relevance of the TSR has to be determined in different breast cancer subgroups, as is previously performed by Roeke and colleagues [17]. For example, in lobular carcinomas, the question is raised on how to determine which part is tumour-induced stroma or supportive stroma. This should be further determined in larger cohorts. With respect to TNBC, five studies have investigated this subgroup, of which three studies have shown significant results [12‐15, 17]. The results of these three studies are rather promising regarding the prognostic effect of TSR [12‐14]. However, two other studies have not validated this prognostic effect despite the favourable results earlier showed. As mentioned by Roeke et al., this discordance could possibly be contributed to the relatively low amount of stroma-high tumours in the TNBC subgroup [17]. The similar reason could be the cause for the effect of TSR in TNBC patients in the study of Gujam et al. [15]. Another explanation could be that the histological type of TNBC plays a role.

Although different studies researched the prognostic value of TSR, little is known about the composition of the stroma. Even when using conventional light microscopy, vast differences in stromal morphology can be appreciated which are surely reflective of enormous differences in stromal functionality. Molecular analyses have identified multiple molecular markers that are associated with varying degrees of stromal activation [53‐55]. These findings might allow us to distinguish activated, highly tumour-promoting stromal tissues from non-activated or only mildly active stromal tissues. Future studies investigating stromal activation might therefore solely focus on specific highly active subsets of stromal tissues as opposed to counting all stromal tissues equally, thereby further refining this parameter: For instance, by identifying PA28 as a marker of stromal activation in a previous publication [53].

Similarly, Ahn et al. investigated the stromal composition of breast cancer tissue. Besides TSR, the dominant histological stroma type (collagen, fibroblast or lymphocytes) offers additional prognostic information. 5- and 10-year RFS rates were most favourable in the lymphocytic stroma type, followed by the fibroblast and collagen type. The latter was associated with the most aggressive tumour and consequently poorest prognosis [56]. Interestingly, Ahn et al. observed a trend between TNBC and a predominantly lymphocytic stroma type, with 56.1% of the samples being classified as ‘lymphocytic’. Considering the fact that TNBC has a relatively poor prognosis, the observed trend between TNBC and a predominantly lymphocytic stroma type, with a favourable prognosis, is striking. Leon-Ferre and colleagues showed similar results in early-stage TNBC in which the presence of low tumour-infiltrating lymphocytes (TILs) contributes to a poor prognosis [57].

Considering the aforementioned generally promising prognostic effect in TNBC, this subgroup is the most obvious candidate for further exploration of TSR. Currently, adjuvant systemic chemotherapy is advocated for all patients that present with operable TNBC due to the aggressive nature of this tumour subgroup. Regarding TNBC, unlike other molecular subtypes, there is no FDA approved targeted therapy yet. Forasmuch as both the aggressive nature of the subtype as the devoid of therapeutic options, supplementary research is necessary. In order to develop curative therapeutics in TNBC, stromal targets have to be determined. Given the fact that TNBC predominantly consists of lymphocytic stroma, according to Ahn and colleagues, the possible target might lie within this stroma. The quantity of programmed death ligand 1 (PD-L1), expressed on tumour cells, could be prognostic as well. Tomioka et al. have shown that low TILs in combination with high PD-L1 expression predict an unfavorable prognosis. Within the abundant lymphocytic stroma in TNBC, PD-L1 could possibly operate as a target for therapeutic options [58]. Thus, in further research, in addition to a standardised estimation of TSR, the biology or quality of the stroma should be taken into account as well, in both general breast cancer and especially in TNBC patients in order to clarify the paradox and subsequently to lay a foundation regarding targeted therapy.

Lastly, it should be noted that although previous studies demonstrated prognostic value in the past, these studies have always been performed as part of retrospective studies by researchers and pathologists with a specific interest in stromal tissues. Breast cancer is a heterogeneous disease and for this reason additional larger retrospective studies could add valuable information about the prognostic value of TSR in specific subgroups as well. Moreover, no prospective feasibility studies have been performed and as such, it remains to be seen whether broad application of this parameter would lead to reproducible test results. Current research efforts in this direction are however ongoing.