Prognostic Value of Tumor-stroma Ratio in Oral Squamous Cell Carcinoma: Contribution of Cancer Associated Fibroblasts

Background: The aim of this study is to conrm the prognostic value of the tumor–stroma ratio (TSR) in a large cohort of oral squamous cell carcinoma (OSCC) and further demonstrated the cancer associated broblasts (CAFs)-stroma ratio (CSR) served as a critical biomarkers contributed to the prognostic value of TSR Results: The threshold level of TSR value is 50%, which divides patients into high (>50%) and low (<50%) stroma. We examined the TSR on hematoxylin and eosin-stained tissue samples from 581 patients with oral squamous cell carcinoma and 298 cases were included in the high-stroma group. In multivariate analysis, the TSR was identied as an independent prognostic factor for disease-free survival (DFS) (hazard ratio (HR), 2.11; 95% condence interval (CI), 1.56–2.86; p < 0.001) and oral cancer-specic survival (OCSS) (HR, 2.56, 95% CI, 1.78–3.67; p < 0.001). The interaction term reached statistical signicance for histological grade. Multivariate analysis conrmed the discriminative value of the TSR in well differentiated tumors for DFS and OCSS separately (P=0.001, P=0.003). The prognostic value of TSR was not varied by other clinically subgroups. Furthermore, the high-stroma group had a higher Fibroblast Activation Protein (FAP+) CSR and α-Smooth Muscle Actin (α-SMA+) CSR than the low-stroma group (p < 0.001). Conclusion: High-stroma levels indicated a negative consequence and a higher CAFs–stroma ratio than low-stroma levels in OSCC. The TSR is not altered by other clinically elements rendering it a credible histological parameter and informing the rational design of individual cancer management. CSR, Cancer associated broblasts (CAFs)-stroma ratio; DFS, Disease-free survival; OCSS, Oral cancer-specic survival; OS, Overall survival; HR, Hazard ratio; CI, Condence interval; FAP, Fibroblast Activation Protein; α-SMA, α-Smooth Muscle Actin; H&E, Hematoxylin and eosin stained slides; ECM, Extracellular matrix; EMT, Epithelial–mesenchymal transition.

clinical application is limited. Incorporating relevant parameters that translate microenvironment in the routine pathology diagnosis is imperative.
The tumor-stroma ratio (TSR) has a promising prospects, which represents the relative abundance of stroma in cancer cells validated in many solid tumors, including colon, breast and oesophagus cancers [6][7][8]. Additionally, the core competitiveness of TSR evaluation is easy, quick, reproducible, and inexpensive by visually assessing routinely retrieved H&E slides used for pathological cases.
Previously, in early-stage oral tongue cancers Almangush et al. has assessed the prognostic impact of TSR [9]. By comparison, we con rmed the preponderance of the TSR as a prognostic tool in the whole oral cancer, speci cally those with buccal, tongue, and gingival squamous cell carcinoma. Furthermore, we found the prognostic value of TSR is chie y discerning in moderately and highly differentiated tumors. More importantly, we rstly substantiated that a difference in the cancer-associated broblast (CAFs)-stroma ratio(CSR) exists between the high and low-stroma groups and this discrepancy has in uence on the prognostic power of TSR. Our present work clari ed the prognostic value of TSR and highlight the contribution of CAFs-stroma ratio (CSR).

Patients and tumor characteristics
Initially n = 850 patients with oral squamous cell carcinoma at School & Hospital of Stomatology, Wuhan University were incorporated. After telephone follow-up n=632 patients with available follow-up data were included. Finally, we evaluated 581 H&E-stained slides from patients for TSR scoring because of the loss or inferior quality of the excluded slides(n=51). It is worth noting that inferior quality refers to di culty in judging the TSR groups of tumor with little invasiveness (inplementary materias gure 5). Interestingly most of the analyzed area on the slide are in the invasive front. The mean age of the patients was 57.08 years at the time of operation. The mean follow-up period was 30.33 months (range, 1-36 months). The basic clinicopathological characteristics of the patients are shown in Table 1. Most tumors were of oral tongue cancers (56.11%), male (62.13%), no lymph node metastasis (73.67%), <4cm (81.41%), and grade I or II (87.09%).

Tsr Strati ed By Clinically Crucial Subgroups
To nd the prognostic value of TSR in clinically essential subgroups, Cox regression analysis was performed. We introduced the interaction term to assess the clinically discriminative subgroups and found that grade was statistically signi cant for DFS (p < 0.001) and OCSS (p = 0.003) ( Table 3). No statistical differences were observed between the groups when strati ed by gender,T status, N status, and pTNM stages. In the multivariate Cox regression analysis, the prognostic value of the TSR was most discriminative in grade I carcinomas for DFS (HR, 4.38; 95% CI, 2.14-8.95; p < 0.001) and in grade I, II carcinomas for OSCC (HR, 4.75; 95% CI, 1.93-11.68; p = 0.001 and HR, 2.46; 95% CI, 1.53-3.97; p < 0.001, respectively). The Kaplan-Meier analysis for DFS of the TSR combined with grade displayed a statistically signi cant difference among the subgroups (Figure.1C). The CAFs-stroma ratio(CSR) essentially advanced to the prognostic value of TSR To further examine whether the prognosis of the TSR is related to the CAFs-stroma ratio, we selected 100 patients from the low and high-stroma groups to detect C FAP+ SR and C SMA+ SR through immunohistochemistry. The baseline characteristics of the patients and tumors are shown in Supplementary materials Table 2. For negative control, we used PBS instead of antibodies against FAP, α-SMA and vimentin for immunohistochemistry and no positive staining for FAP, α-SMA and vimentin (Supplementary materials Figure 4). Then we stained OSCC slides with FAP and α-SMA, speci c markers for CAFs, and vimentin a speci c marker for stroma. Expression of FAP and α-SMA were separately regarded a CAF subtype to evaluate the CAFs-stroma ratio as low CAFs group, medium CAFs group and high CAFs group (Figure 2). Medium and high expression of CAFs-stroma ratio were observed for FAP (72%, 72/100) and α-SMA (73%, 73/100) respectively (Table 4). What's more, Patients with high stroma had higher C FAP+ SR and C SMA+ SR than those with low stroma (p < 0.001) ( Table 4) via immunohistochemical analysis. The Kaplan-Meier curves are shown in Figure 2, comparing the low, medium, and high CAFs groups for SMA and FAP with statistical signi cance. In the Cox multivariate model, the C FAP+ SR was signi cantly correlated with DFS (p < 0.001) and OCSS (p = 0.008) after correcting for confounders (Table 4). Meanwhile the C SMA+ SR was signi cantly related to the DFS (p < 0.001) and OCSS (p = 0.002) ( Table 4). In addition, it's impressive to nd the high expression of FAP in the tumor cells in the stroma-low group (n = 17) compare to the stroma-high group (n = 4), which may account for the disparity of the prognosis in the stroma-high and stroma-low groups (Supplementary materials Figure 5).

Discussion
This study was designed to con rm the prognostic signi cance of TSR in OSCC in the largest cohort so far, thus rendering us to proceed to subgroup analysis in a high degree of con dence. Not only that, we show that the prognostic value of TSR was especially sensitive in tumor with grade I and II by means of introducing interaction terms in COX proportional mode. More importantly, we further ascertained that CAFs differ between high-and low-stroma tumors, uncovering the underlying possibilities of poor outcomes in the high-stroma group in a certain extent.
First, TSR was of prognostic value validated in OSCC, speci cally its independent prognostic value in different locations. The high-stroma group had worse outcomes than the low-stroma group. Second, gender, N status, and differential grade were signi cantly correlated with DFS and OCSS through the multivariate Cox regression analysis. Meanwhile, pTNM stage (p = 0.792) illustrated a poor prognostic power for DFS. Additionally, we introduced the interaction terms in the Cox regression analysis, including TSR*gender, TSR*T status, TSR*N status, TSR*pTNM stages, and TSR*differential grade. The results showed that the use of the TSR was most sensitive in high differentiated tumors for DFS and OCSS. No statistically signi cant differences were observed among the subgroups; however, this does not mean that the prognostic effect of TSR does not differ among the total cohort. Moreover, the concept of interaction term was rst used in the clinical research of the TSR in oral cancer. Finally, immunohistochemical analysis re ected that the high-stroma group has higher C FAP+ SR and C SMA+ SR than the low-stroma group. Additionally, patients with a high C FAP+ SR implied a poor outcome, which may be accounted for the adverse outcomes in the high-stroma group.
Moreover, this is the rst study to digitally assess the TSR in oral cancer using the Mesker method [10].
Concretely, we chose a 3.2-mm2 circle to perform TSR scoring, which was exactly parallel to the *10 objection in light microscopy. A study has indicated that the diameter of a light microscope does not differ in the nal score 11 . All slides were scored three times to avoid intra-observer variability. Finally Kappa coe cient reached perfect agreement though in the beginning only 0.762; most initial discordance were settled. Moreover, the remaining slides from four patients were consulted to the third observer to get consensus. Kappa values will further elevate to a perfect level because of the learning curve. Thus, we should digitally examine enough H&E-stained slides to ensure the reliability of the TSR score, which is also mentioned by Vangangelt et al [12]. Through proper training, the TSR as a prognostic tool will be easy, repeatable, and inexpensive.
The tumor stroma, or other the tumor microenvironment (TME), was one of the three key unsolved issues that impede effective clinical therapy for tumors [13], which were heterogeneous and plastic for including the intracellular environment of tumor bulks and the surrounding stromal cells with abundant protumorigenic factors [14]. Though tremendous advance in elucidating the mechanisms underlying tumor-promoting effects of the TME, tackling the complicated mechanisms still has a long way to go.
Over the last decade, the prognosis in OSCC has been substantially studied, including biomarkers, such as caveolin-1 [15],MAGE-A11 [16], and clinicopathological parameters, such as histological grade [17,18] according with our current study. Although accuracy is limited, Tobias Ettl et al. (2016) have found that positive frozen section margins was associated with recurrence [19]. In contrast, the TSR was more practicable as a prognostic tool.
More interestingly, Amol Ramchandra Gadbail et al. (2017) have discovered that the OSCC has a better prognosis in the background of submucous brosis [20], which reminded the effects of CAFs implicated in the prognosis in OSCC. The term "cancer-associated broblast" were the most abundant stromal component in the TME, originating from at least six cellular categories, including normal broblast, quiescent stellate cell, endothelial cell, epithelial cell and pericyte, smooth muscle cell, and adipocyte [4].
Besides, CAFs can secrete a plentiful growth factors, chemokines, and exosomes regulating the course of cancer and immunosuppression, along with interleukin-6 and tumor growth factor-beta [21][22][23]. Owing to their cellular origins and breadth of functions, CAFs can have either protumorigenic or antitumorigenic effects on different solid tumors [24]. which is the major challenge for CAF-targeting therapies. Preclinical studies have indicated that nonspeci c targeting of CAFs does not achieve the desired results in cancer treatment [25]. In view of the abundance and priority of CAFs in the stroma, we set to estimate the CSR in OSCC to attempt to explain the prognostic value of TSR.
Recently, several studies have mentioned the prognostic value of CAF density in OSCC [26]. Moreover, the biological behaviors of CAF were altered by microRNAs to facilitate invasion of OSCC and CD68(+)CAFS predicted poor prognosis illustrated that CAFs is Superior in Prognosis compare to the Epithelial-Mesenchymal Transition(EMT) Score [27][28][29]. Although molecular studies of CAFs deepen our insight of CAF intratumor heterogeneity and breadth of function, they remain hard to specify for lacking speci c markers [4]. Here we chose to use α-SMA and FAP to represent CAF subtypes and found that the highstroma group has a higher CSR than the low-stroma group. Additionally, the C FAP+ SR and C SMA+SR were related to the prognosis, which might be accounted for the negative consequences in the high-stroma group. However, the underlying mechanism remains to be explored. Daniel Öhlund had observed the expansion of stromal broblast number, the process was called "stroma-genesis" with concomitant oncogenesis [30,31]. A study has found broblasts encircling early lesions, indicating that the early phase of broblasts could be suppressive, and these broblasts evolve to tumor-promoting broblasts through stroma-genesis [25,32]. However, the hypothesis was di cult to testify due to the impractical and longitudinal sample of the same lesion and the conversion of the cell states. In this study, the richstroma group suggested a worse prognosis than the poor-stroma group, and the rich-stroma group had higher C FAP+ SR and C SMA+ SR. Moreover, patients with higher C FAP+ SR and C SMA+ SR had a worse prognosis. Additionally, we found that the low-stroma group has higher positive staining of FAP+ in tumor cells. According to the latest study that overexpression of FAP is related to the epithelial-mesenchymal transition (EMT) in OSCC [33].we inferred that low-stroma tumors might initially be transformed into highstroma tumors through the events where broblasts evolve to protumorigenic broblasts and tumor cells via EMT, though the detailed mechanisms remain to be explored.
First, one of the study strengths is manifesting that the prognostic value of the TSR was most prominent in moderately and highly differentiated tumors, and indicating the unique role of the C FAP+SR and C SMA+ SR in advancing the prognostic value of TSR. Second, this study has the largest oral cancer sample, which enabled us to render the subgroup analysis in high con dence. Third, the method we used to evaluate the TSR con rmed that we can locate high-stroma areas in the entire slides, and images can be stored to solidify the authenticity of the results compared to those obtained using light microscopy, which are disturbed by uncontrollable factors when scoring the TSR. Limitations of our study are retrospective study design with a short follow-up period and all cases came from a single institution; thus, whether the application of the TSR in a prospective study can obtain reproducible results in OSCC remains to be seen.
In conclusion, the prognostic value of TSR was positively validated in oral squamous cell carcinoma particularly in tumors with moderate or high differentiation. and the prognostic effect of TSR didn't differ from the prognostic value of the whole cohort in gender, tumor size, lymph node status subgroups.
Furthermore, the fact that higher C FAP+ SR and C SMA+ SR in tumors with stroma-rich simultaneously suggesting the poor outcome, illustrating C FAP+ SR and C SMA+ SR contributed to the prognostic impact of TSR in OSCC. In a Word, the advantages of TSR were convenient, practicable and lowcost in routine pathological examination and do not need extra staining, which opened a great opportunity for decisionmaking in individual therapy.

Conclusions
We show that tumor-stroma ratio (TSR) as a prognostic tool in the whole OSCC especially in highly differentiated tumors. There is also a difference in the cancer-associated broblast (CAFs)-stroma ratio(CSR) exists between the high and low-stroma groups and its in uence on the prognostic power of TSR, which informed the rational design of individual cancer management especially for pathological diagnosis.

Study patients
Data of patients with OSCC were obtained from the database of School & Hospital of Stomatology, Wuhan University. We incorporated every patient with primary OSCC who had surgery for the rst time from January 1, 2012 to December 31, 2016. All cases were staged based on the seventh edition of the American Joint Committee on Cancer Staging Manual [34] because the eighth edition of the manual was used from January 2018. Patients treated with chemotherapy and/or radiotherapy before the study, died within a month, and those who those with non-primary tumor were excluded from the study. The basic demographic data and detailed information were collected from the Medical Records Room and the Department of Oral and Maxillofacial and Head and Neck Oncology. Every pathological report and H&E section were retrieved from the Pathology Department.

Assessment Of The Tsr
We visually evaluated the TSR using 3-µm H&E slides of primary OSCC using NDP.view2 (Hamamatsu Photonics, Hamamatsu City, Japan), a software for the digital assessment of microscopic images. The original H&E slides were scanned using NanoZoomer (S210 C13239-01,NanoZoomer) to make highresolution digital images for subsequently analysis.
First, all H&E-stained slides were visually assessed to nd areas with the most abundant stroma. Then, we used a 10× objective and ensured that tumor cells exist in all borders. Almost all chosen elds were near the site of the invasive front. The cutoff TSR value was determined as 50% according to the study by Mesker [10]., and we divided into two groups according to the cutoff TSR value: low stroma (<50% stroma) and high-stroma (>50% stroma) groups. A stroma proportion of approximately 50% was considered stromal-rich. Two blinded observers recorded the slides three times to decrease intra-observer variability. Disagreements between the two observers were resolved by a third observer to get the nal consensus.

Immunohistochemistry And Evaluation Of Immunoreactivity
We chose 100 patients from the aforementioned groups (50 patients from the high and low-stroma groups, respectively). And we could not adopt completely random way due to the para n block available The immunohistochemically stained slides were scanned, and images were analyzed using NDP.view2. We examined the CAFs-stroma ratio throughout the slide based on the distribution of CAFs and separately counted the FAP + broblast and α-SMA + broblast subtypes. To determine the stroma, we referred to the expression of vimentin and original H&E-stained slides.