Immunoscore Combining CD8, FoxP3 and CD68 Expression and Distribution Predicts the Prognosis of Head and Neck Cancer Patients

Purpose: The objective of this study was to assess immune cell inltrates to develop an immunoscore for prognosis and to investigate its correlation with clinical data of patients with head and neck squamous cell carcinoma Methods: CD8, FoxP3 and CD68 were evaluated by immunohistochemistry in 258 carcinoma samples and counted in stromal and intra-tumoral compartments. Optimal cut-offs were assessed to divide population regarding to survival while the prognostic value was established by using Kaplan-Meier curves and Cox regression models for each immune marker alone and in combination. Results: We found with univariate analysis that inltration of immune cells in both compartments was predictive for RFS and OS. Multivariate analysis revealed that CD8+ number inuenced independently patient prognosis. Additionally, the combination of CD8, FoxP3 and CD68 in an immunoscore provided a signicant association with OS (p=0.002, HR=9.87). Such immunoscore stayed signicant (p=0.018, HR=11.17) in a multivariate analysis in comparison to tumour stage and histological grade which had lower prognostic values. Conclusion: Altogether, our analysis indicated that an immunoscore including CD8, FoxP3 and CD68 immunostaining was a strong, independent, and signicant prognostic marker which could be introduced into the landscape of current tools to improve the clinical management of head and neck cancer patients.


Introduction
Head and neck squamous cell carcinomas (HNSCC) are among the most prevalent cancers worldwide, setting them in the 6th place (Ferlay et al. 2019). In Belgium, their incidences are higher and such cancers arise at the 4th position in men (Jerome R. Lechien et al. 2019). Despite advances in therapeutic approaches, the mortality rate has remained relatively constant in recent years, with a 5-year survival rate around 50% and recurrences occurring in 40-60% of treated patients (Windon et al. 2018). This poor response to treatment can be explained in part by late diagnosis and a lack of e cient drugs in the case of tumour recurrence. However, it appears that the cell composition of the tumour microenvironment (TME) is likely to in uence patient outcome (Valkenburg, de Groot, and Pienta 2018). The well-known risk factors of HNSCC are the consumption of alcohol and tobacco, as well as infection with the human papillomavirus (HPV), which is known to be associated with a better prognosis for the youngest patients with oropharyngeal carcinoma (Young et al. 2015). In this context, several studies suggest that HPV + patients with HNSCC have a speci c TME which may in uence the response to treatment (Mirza et  Macrophages and more speci cally tumour-associated macrophages (TAMs) are the most abundant cells within the TME and they are able to stimulate regulatory T lymphocytes (Treg) cells to switch to a pro-tumour environment (Evrard et al. 2019; Jérôme R. Lechien et al. 2020). Regarding macrophages, we previously showed that CD68 + in ltration arises during HNSCC progression in the intra-tumoral (IT) compartment and is associated with the tumour stage. We also highlighted that CD68 + recruitment is higher in HPV + patients than in HPV-ones. Moreover, a high in ltration of CD68 + cells was related to a shorter recurrence-free survival (RFS) as well as a shorter overall survival (OS) (Seminerio et al. 2018).
Immune surveillance is also governed by the tumour-in ltrating T lymphocytes (TILs) (de Ruiter et al. 2017). Among them, the CD8 + T lymphocytes act speci cally on the cancer cells in order to eliminate them (S. M. Y. . In HNSCC, a high density of CD8 + is correlated with a good prognosis (Lechner et al. 2017). Concerning Treg, which are characterized by the transcription factor forkhead box P3 (FoxP3) (Sakaguchi et al. 2010), we previously showed that FoxP3 + Treg in ltration increased during HNSCC progression (from dysplasia to carcinoma) and that tumours with high Treg in ltration were associated with longer RFS and OS Kindt et al. 2017).
In advanced HNSCC, the gold standard treatment remains the concomitant chemoradiotherapy, but the emergence of immunotherapy over the last years has changed the landscape of HNSCC treatment (Lyford-Pike et al. 2013; Gavrielatou et al. 2020). However, using anti-cancer drug like immunotherapy is challenging due to the heterogeneity of the TME composition (Canning et al. 2019). Importantly, HPV status is now included in the tumour-node-metastasis (TNM) staging system (Leemans, Snijders, and Brakenhoff 2018) indicating the importance of additional prognostic information to propose the most appropriate treatment for patients. Currently, there is no immune-based classi cation of head and neck cancer. However, the evaluation of immune cell recruitment to classify HNSCC patients in different immunologic subgroups depending on the TME composition could be helpful to improve patient prognosis. The value of immunoscore has already been highlighted in several cancers, such as cervical In this study, we propose an immune signature based on CD8+, FoxP3 + and CD68 + count in IT and/or stromal (ST) compartments in a large clinical series of 258 patients with HNSCC. The immunoscore is compared to tumour stage and histological grade using multivariate analyses.

Patients and clinical data
A total of 258 patients presenting HNSCC were enrolled on our study. The Table 1 describes the clinicopathological characteristics, treatment, and follow-up data. Formalin-xed para n-embedded specimens obtained after surgical resection at Saint-Pierre Hospital (Brussels, Belgium), Jules Bordet Institute (Brussels, Belgium), EpiCURA Baudour Hospital (Baudour, Belgium) and CHU Sart-Tilman (Liège, Belgium) between 2002 and 2019 were used for immunohistochemical labelling. This retrospective study has been approved by the Institutional Review Board (Jules Bordet Institute, number CE2319).

Immunohistochemistry
The 5 µm thick slices of HNSCC were depara nized in toluene and rehydrated in a graded series of alcohols, then peroxidase was blocked using H 2 O 2 and nally slices were rinsed with water for 7 min.
Antigen retrieval was processed by immersing the samples in 10% EDTA/H 2 O or in 10% citrate/H 2 O followed by heating in a pressure cooker or in a microwave (buffer and timing are dependent on antibodies, see Supplementary Table 1). Non-speci c sites were blocked with 0.5% caseine for 15 min.
Slices were incubated with primary antibody (anti-human CD68 monoclonal mouse, dilution 1:200, and anti-human CD8 monoclonal mouse, dilution 1:200, both from Dako) for 1 hour at room temperature or overnight at 4°C. Kit PowerVision Poly-HRP IgG were used for the second antibody. For FoxP3 immunostaining, the anti-human FoxP3 monoclonal mouse (dilution 1:200, from Invitrogen) was used and the detection of this primary antibody was performed with the CSAII kit (Dako). For each immunohistochemistry, tonsil tissue was used as positive (and negative (no primary antibody)) controls.

Calculation of an immunoscore
The number of each immune cell type was counted in 5 elds in the IT and ST compartments with an Axio-Cam MRC5 optical microscope (Zeiss, Hallbergmoos, Germany) at 400x magni cation by two investigators (S.F. and G.D.). The mean of each counting was calculated for each patient. For each marker in IT and ST, the cut-off value giving the best separation between two groups (HR and p for OS) was evaluated using the RStudio software. Then, if the mean number of the 5 elds was greater than the cut-off, the case was considered as "high" and if it was lower, the case was considered as "low". Based on such cut-offs, the prognostic value of each immune marker was examined regarding RFS and OS. From these analyses, an immunoscore was de ned combining the most signi cant immune markers.

Statistical analyses
The optimal cut-off points of the population for each immuno-biomarker were calculated by using RStudio software (package from Cutoff Finder web application). Collected data were analysed with IBM SPSS software (version 23) (IBM, Ehningen, Germany). RFS and OS analyses were performed using Kaplan-Meier curves. Univariate and multivariate Cox regression models were applied to calculate the hazard ratio (HR), 95% con dence interval and signi cance. p-values < 0.05 were considered as signi cant.

Correlations between clinical characteristics and RFS or OS
Our clinical series included a total of 258 HNSCC patients, among which 177 (68.6%) were men and 81 (31.4%) were women, with a median age of 61 years old (range, 29-90). Among these patients, 104 patients presented tumour recurrence and 102 died. The clinicopathologic characteristics are provided in the Table 1.
We evaluated the association between tumour stage, histological grade, tumour invasion or risk factors with RFS or OS. Cox regression models highlighted that among such parameters only tumour stage and histological grade correlated with OS. Evaluating Kaplan-Meier survival curves, patients with tumour stage I-II were associated with longer OS (p = 0.005) compared to patients with tumour stage III-IV. Moreover, well differentiated tumours were also associated with longer OS (p = 0.029) compared to poorly differentiated ones (Fig. 1).

Immune cell number and patient survival
In the HNSCC surgical specimens, immune cells were detected by using speci c antibodies against CD8, FoxP3 and CD68 within the ST and the IT compartments ( Fig. 2a-c). Lymphocytes T (CD8+), Treg (FoxP3+) and macrophages (CD68+) were counted in 5 random elds (magni cation 400x) in both ST and IT compartments ( Fig. 2d-i).
Cut-offs were calculated using RStudio software regarding optimal HR and p-values for OS and patients were classi ed as expressing a low or high number of immune cells. The cut-off values were 55. for the three immuno-markers CD8, FoxP3 and CD68 in the two compartments for RFS and OS (Table 2). Multivariate analysis showed that the CD8 + cell number was a strong and independent prognostic marker.  Using these cut-offs, Kaplan-Meier curves were established for each immune cell in each compartment for RFS and OS. Regarding the ST compartment, longer RFS was signi cantly associated with a high FoxP3 + cell number, while longer OS correlated with low CD8 + and high FoxP3 + cell numbers (Fig. 3). In ST, the CD68 + cell number did not correlate with RFS or OS. In the IT compartment, low CD8+, high FoxP3 + and low CD68 + cell numbers were signi cantly linked to a longer RFS as well as a longer OS (Fig. 4).

Immunoscore and patient survival
The Fig. 5a describes how we calculated our immunoscore. Each tumour of the patients was categorized into low (Lo) or high (Hi) density for each immune cell in each tumour region according to the cut-off values. Depending on the type of immune cells and the tumour compartment, the Lo and Hi classes were associated to the blue and red groups corresponding to the 0 and 1 scores, respectively. The immunoscore was created by adding the individual score (0/1) of each marker, which was signi cant for OS. Based on univariate analyses (Table 1), we included CD8 ST/IT, FoxP3 ST/IT and CD68 IT in the immunoscore. CD68 + cell number in the ST did not correlate with RFS or OS and will therefore not included in the immunoscore. Thus, the scoring system ranges from 0 to 5. Kaplan-Meier curves were drawn for each value of the immunoscore and a cut-off discriminating good and poor patient prognosis was chosen at 3 ( Supplementary Fig. 1). Thus, each patient was classi ed in the blue group (good prognosis, low immune score < 3, n = 23) or in the red group (poor prognosis, high immune score > 3, n = 97).
Very signi cant differences were observed between the blue and red immunoscore. Kaplan-Meier curves using our immunoscore showed a signi cant correlation for RFS (p = 0.007) and OS (p = 0.002) (Figs. 5b and c respectively) with high immunoscore (> 3) being associated with shorter RFS and OS, compared with low immunoscore (< 3) which was associated with better prognosis for RFS and OS.
Finally, we performed univariate and multivariate Cox regression analyses to compare our immunoscore with the conventional tumour stage and histological grade. Our immunoscore correlated more signi cantly, and with a greater separation of the two groups, regarding OS (p = 0.002, HR = 9.87) compared to tumour stage (p = 0.005, HR = 1.91) and histological grade (p = 0.029, HR = 0.62) ( Table 2). Multivariate analyses revealed that the immunoscore was the only parameter associated with a strong and independent prognosis value.

Discussion
To the best of our knowledge, this study assessed, for the rst time, the abundance and distribution of innate and adaptive cellular elements according to CD8 T cells, FoxP3 Treg and CD68 macrophages in a series of 258 patients with HNSCC in order to de ne a more global immune contexture. Then, we investigated their potential prognostic value separately and in combination to stratify patients using a low immunoscore corresponding to a longer RFS and OS, whereas a high immunoscore was related to a poorer prognosis. Our results con rm that the establishment of an immunoscore has a higher prognostic value than that of TNM staging system and histological grade.
For many years, clinical research around head and neck cancers has been constantly asking for new prognostic biomarkers to better guide patient management. Given the complexity of the interactions between immune in ltrates and the TME, the tumour must no longer be considered as a single entity but must be studied in relation to its microenvironment and the host immune response in order to bring clinical relevance and value in determining the tumour progression and the patient prognosis. As such, many clinicians and researchers have been interested in the in ltration of immune cells in head and neck cancers. In most cases, massive in ltration of CD8 + T lymphocytes in HPV-as well as in HPV + oropharyngeal carcinomas correlates positively with patient prognosis (Nordfors et  Our recent study has shown that a high recruitment of CD68 + macrophages in a population of 110 HNSCC was correlated with shorter patient RFS and OS. Moreover, the analysis of the M1/M2 ratio in the TME, with a double staining using anti-CD68/anti-CD163 antibodies, revealed that 80% of the macrophage population had an M2 phenotype (Seminerio et al. 2018). Interestingly, it appears that TAMs can secrete IL-10 in order to induce the differentiation of T lymphocytes into Treg (Murai et al. 2009) and thus participate in immune cell evasion.
The existence of complex regulatory loops between these three major mediators of the immune system has led us to quantify their recruitment in ST and IT localizations in a large cohort of head and neck tumours. Indeed, we hypothesized that analysing each tumour compartment (ST versus IT) may provide distinct and complementary prognostic information. This was also supported in the context of rectal cancer where the location of CD8 + T cells and FoxP3 + Treg cells in distinct compartments (epithelium versus stroma) result in different prognostic responses (Posselt et al. 2016). Combining the three markers in an immunoscore signature, we found that a low immunoscore was signi cantly associated with a longer RFS and a prolonged OS. Based on the calculated optimal cut-offs, the IT immune in ltrations associated with a better prognosis correspond to a high number of CD8 and FoxP3 and a low number of CD68 macrophages. Conversely, in the ST, a better prognosis was observed in patients with a low CD8 in ltration but always a high number of FoxP3. Because of the tumour lysis capacity of CD8 cells, these anticancer actors are strong allies for cancer patients. On the other hand, despite their anti-tumour response suppressor characteristics, Tregs in ltrates have been found to be associated with a favourable outcome, which may be partially attributed to a downregulation of the in ammatory process (Shang et al. 2015; Badoual et al. 2006). A correlation had also been found between a higher number of Treg in the stroma and an absence of metastatic lymph nodes which means that Treg could generate proin ammatory processes in the tumour microenvironment favouring a delay in the tumour evolution and consequently would generate a better prognosis of the patients (Bron et al. 2013

Declarations
Funding: This study was funded by the University of Mons, ProtherWal and the Walloon Region and FRMH. S.F. is funded by the University of Mons and the Epicura Hospital and G.D. is funded by the Walloon Region via the ProtherWal society.

Con icts of interest:
The authors declare that they have no con ict of interest.
Availability of data and material: The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Consent to participate: Informed consent was obtained from all individual participants included in the study.

Consent for publication:
Not applicable