Tumor-infiltrating lymphocytes are present in a variety of tumors and play a central role in antitumor immune responses. Nevertheless, most cancers progress probably because tumors are only weakly immunogenic and develop multiple immunosuppressive mechanisms. In the present study, on head and neck squamous cell carcinoma, we found high intraepithelial infiltration of regulatory FOXP3+ T cells, and relatively high levels of BDCA2+ and FOXP3+ cells in stromal (peripheral) regions of the tumors. Tumor-infiltrating (intraepithelial) FOXP3+ T cells were significantly more frequent in patients with oropharynx and oral cavity squamous cell carcinoma and in patients without lymph node metastasis. Furthermore, arginase-II (ARG2) was expressed by 60%, inducible nitric oxide synthetase by 9%, cyclooxygenase-2 by 43%, and B-cell lymphoma 2 (BCL2) by 26% of tumors. Interestingly, the absence of ARG2 expression, enhanced stromal infiltration of CD11c+ myeloid dendritic cells, and high numbers of FOXP3+ T cells were each significantly associated with prolonged overall survival, and the latter two parameters were also confirmed by multivariate analysis. For disease-free survival, multivariate analysis revealed significant negative correlations with BCL2 and ARG2 expression by tumor cells. These findings shed new light on mechanisms of cancer progression, and provide rationales for therapeutic inhibition of immunosuppressive mechanisms in head and neck squamous cell carcinoma.
Head and neck cancers are among the most common types of human tumors, with an incidence of approximately 600,000 new cases per year worldwide.1 They are mostly squamous cell carcinomas and arise from the mucosa of the upper aerodigestive tract, including the oral cavity, oropharynx, hypopharynx and larynx. Despite recent advances in diagnosis and treatment which improved the patients' quality of life, the overall 5-year survival rate has not changed significantly in the last two decades and remains roughly 50–59%.2 Most patients with disseminated metastatic disease have locoregional recurrence diagnosed initially.3 New treatment modalities such as multiple daily fractionated radiation or docetaxel, cisplatin and 5-fluorouracil-based induction chemotherapy have shown promising results. However, toxicity as well as locoregional failure remains a major problem, despite some survival advantage,4, 5 urging the need for innovative therapies, not only to improve functional outcomes without increasing toxicity but also to impact long-term survival in these patients.
The immune system plays an important role in controlling tumor development. Indeed, the discovery of T lymphocytes able to specifically recognize tumor cells provided strong evidence that natural immune responses are generated during tumor progression. Numerous studies have shown that various human tumors are infiltrated by T cells (tumor-infiltrating lymphocytes, TILs)6 and CD8+ T cells appear to be important in the anticancer immune response. Their infiltration within cancer nests is a reliable prognostic indicator in some human cancers.7, 8
Squamous cell carcinomas of the head and neck are usually heavily infiltrated by leukocytes, predominantly lymphocytes and dendritic cells (DCs). Their presence within the tumor epithelium has been correlated with a good prognosis in patients with oral and esophageal carcinomas.9–11 In addition, the analysis of TILs showed that CD8+ T cells present in the tumor microenvironment were activated with proliferative and functional capabilities10 and were able to kill autologous tumor cells,12 suggesting that immunotherapy may represent a potential approach for the control of head and neck cancer.
Tumor growth, invasion and metastasis are important aspects of tumor progression, which could be linked to a failure of the immune system to eliminate cancer cells. In addition to escaping efficient T-cell recognition,13, 14 head and neck tumors can also attenuate the magnitude of the antitumor response by several mechanisms,15, 16 including transforming growth factor β, prostaglandin-E2, interleukin (IL)-6 or IL-10, which might be responsible for the direct inhibition of the immune system14, 17 and the attraction of immunosuppressive CD34+ progenitor cells and regulatory T cells into the tumor tissue with strong immunosuppressive potential.18, 19
Thus, for any immunotherapeutic approach to fulfill its promise and become a partner in standard of care in head and neck malignant disease, it is necessary to identify the immunosuppressive molecules operating at the tumor sites and understand the various immunologic barriers to optimize and increase in vivo efficacy of antitumor immune responses. To further characterize the head and neck squamous cell carcinoma (HNSCC) microenvironment in an attempt to better understand the mechanisms used by these tumors to escape the immune response, we collected retrospectively tumor samples from a group of patients treated surgically for HNSSC and analyzed, using immunohistochemistry, the intratumoral and stromal distribution of DCs, natural killer cells and regulatory T cells as well as the tumor expression of immunosuppressive and antiapoptotic molecules. We then attempted to correlate these results with histopathological characteristics of tumor and patients' outcome.
ARG2: arginase-II; BCL2: B-cell lymphoma 2; COX2: cyclooxygenase-2; HNSCC: head and neck squamous cell carcinoma; iNOS: inducible nitric oxide synthetase
Material and Methods
Patients and specimens
We studied tumor specimens from 35 patients with head and neck cancer who underwent primary surgery between September 2004 and December 2005 in the department of ENT and Head and Neck surgery, at Lausanne University Hospital, Switzerland. Tumors were staged, according to the tumor-node-metastasis (TNM) classification system of the international Union Against Cancer.20
After signature by the patients of an informed consent form approved by the Institutional Review Committee, surgical specimens and peripheral blood were prospectively collected. Each surgical specimen was fixed in 10% formalin and embedded in paraffin wax for routine histopathological assessment and prepared for immunohistochemistry staining. Mononuclear cells were purified by density gradient and immediately frozen.
Immunohistochemical analyses were performed on resected, paraffin-embedded head and neck cancer tissues. Serial 4-μm-thick paraffin sections were cut consecutively from each specimen and mounted on electrostatically precharged slides (Superfrost Plus, Menzel-Gläser, Frankfurt, Germany). The sections were deparaffinized in xylene followed by 100% ethanol and rehydrated with graded ethanol solutions. After endogenous peroxidase quenching (0.3% H2O2 in distilled water for 5 min), antigens were retrieved by boiling the sections in 1 mM ethylenediaminetetraacetic acid solution, pH 9.0, in a microwave oven with the power set at 750 W for 3 min and at 250 W for 15 min. After the completion of 15 min, slides were allowed to cool at room temperature for 30 min.
Immunohistochemistry was performed using N-VISION system. Mouse monoclonal antibodies recognizing human BDCA-2 (1:5 dilution, clone DDX0041, AbCys, Paris, France), CD11c (1:150 dilution, clone NCL-L-CD11c-563, Novocastra, Newcastle, United Kingdom), CD56 (1:50 dilution, clone NCL-CD56-1B6, Newcastle, United Kingdom), FOXP3 (1:5 dilution, clone 150D/E4, kind gift from Alison H. Banham21), cyclooxygenase-2 (COX2) (1:350 dilution, clone 160112, Cayman chemical, Ann Arbor, MI), B-cell lymphoma 2 (BCL2) (1:100 dilution, clone NCL-BCL2, Newcastle, United Kingdom) and rabbit polyclonal antibodies recognizing Arginase-II (ARG2) (1:200 dilution, clone sc-20151, Santa-Cruz Biotechnology, Heidelberg, Germany), inducible nitric oxide synthetase (iNOS) (1:50 dilution, clone RB-9242-P1, Neo Markers, Fremont, CA) were applied as primary antibodies at room temperature for 1 hr or overnight for BCL2. Normal mouse IgG1 and IgG2 (1:5 dilution, clone IM0571 and IM0572, Immunotech, Marseille, France) and rabbit IgG (1:50 dilution, clone sc-2027, Santa-Cruz Biotechnology, Heidelberg, Germany) isotype-matched antibodies were used as negative controls for immunohistochemistry (Supporting Information Fig. 1). The slides were then incubated with the respective anti-mouse or anti-rabbit horseradish peroxidase for 30 min at room temperature. The antigen detection was performed by a color reaction with 3,3′-diaminobenzidine (DAB+ chromogen, DakoCytomation, USA). The sections were counterstained with hematoxylin and mounted with Aquatex.
Evaluation of stained sections
Slides were examined by an experienced pathologist with no knowledge of the corresponding clinicopathological data of the patients. To evaluate BDCA-2, CD11c, CD56 and FOXP3 immunostaining, digital image pictures of the tumor areas in three different fields (size, 1.1 mm2 each) were selected and the absolute number of immune cells in these digital images was determined using Adobe Photoshop CS software. The absolute number of immune cells was counted within the cancer epithelium and within the cancer stroma (defined as a region surrounding the cancer epithelium of a width of 0.3 mm), and expressed as number of cell/mm2 of cancer epithelium or cancer stroma. As the immunohistochemical staining of immune cells has rarely been reported in the previous studies, there are no widely accepted standard cutoff points for correlations between the clinical outcome and the number of identified immune cells. In our study, we selected the median number of counted immune cells as the cutoff value when we evaluated intraepithelial and stromal distribution. For ARG2, iNOS, COX2, and BCL2 expression, we defined a staining as negative when <1% of tumor cells expressed the specific molecule.
Intracellular FOXP3 staining
Peripheral blood was obtained from patients upon informed consent. The healthy subjects were blood donors at the Blood transfusion Center in Bern, Switzerland. Mononuclear cells were purified by density gradient and immediately frozen. Cell staining with fluorescent monoclonal antibodies directed against surface molecules (20 min at 4°C) was followed by FOXP3 labeling according to the manufacturer's recommendations. All antibodies were purchased from BD Biosciences (Allschwill, Switzerland), with exception of anti-human FOXP3 (eBioscience, San Diego, USA). Cells were analyzed by flow cytometry on a LSR II.
Statistical analyses were performed using the statistical software Stata 10 (StataCorp LP, TX). The association between the clinicopathological characteristics and the immune cell distribution was tested using the nonparametric Wilcoxon's test. The Fisher's exact test was used to correlate clinicopathological characteristics with ARG2, iNOS, COX2 and BCL2 expression. The paired T-test was used to analyze the local distribution of different immune cell types in both epithelial and stromal localization. The correlation between overall and disease-free survival with either immune cell distribution or immunosuppressive molecule expression was calculated using the Kaplan–Meier method. Analysis was completed with multivariate analysis using Cox regression models. A difference was considered statistically significant when p ≤ 0.05.
The clinical characteristics of the patients are summarized in Table 1. Tumor localizations were divided into four groups namely oral cavity (46%), oropharynx (31%), hypopharynx (17%) and larynx (6%). The T-stage was divided into four groups namely T1 (31%), T2 (43%), T3 (17%) and T4 (9%) and 19 patients (54%) had lymph node metastasis, finally 37% of patients had stage IV disease. Of the 35 patients examined, 18 (51%) were alive at the end of the study. The median follow-up was 49 months (range, 2–83) and the overall survival at 6 years was 50%.
Table 1. Distribution of clinical and pathological variables in HNSCC specimens
Distribution and frequency of immune cells
Immunohistochemistry analysis of paraffin-embedded tumor tissues from head and neck cancer patients showed a median number/mm2 of BDCA2+ plasmacytoid DCs was significantly higher in the stroma than in the tumor epithelium (p < 0.00001). No significant difference was observed in the distribution for CD11c+ myeloid DCs and CD56+ natural killer cells. The median number/mm2 of FOXP3+ regulatory T cells was significantly higher in the stroma of the tumors as compared to the tumor epithelium (p < 0.00001) (Table 2). Representative photomicrographs of immunohistochemical staining for BDCA2+, CD11c+, CD56+ and FOXP3+ cells are shown in Figures 1a–1d.
Table 2. Median numbers of tumor-infiltrating immune cells in the stroma and intraepithelial compartments
There was no statistically significant correlation between the median number of immune cells and the following clinicopathologic variables: tumor site, TNM stage, nodal status, differentiation grade and stage. One exception was the number of CD56+ cells in the intraepithelial tissue, which was higher in earlier stages of the disease, T1–T2 or Stages I–II, compared to advanced T3–T4 or Stages III–IV (p = 0.031 and 0.048, respectively). In addition, the number of FOXP3+ cells was significantly higher in tumors of the oral cavity/oropharynx compared to hypopharynx/larynx, in both intraepithelial and stromal localization (p = 0.031 and 0.009, respectively), and a higher number of FOXP3+ cells in the stroma was significantly correlated with the absence of metastatic lymph node (p = 0.032) (Table 3).
Table 3. Wilcoxon's test between clinicopathologic variables and studied biomarkers
Analysis of exvivo FOXP3+ CD4+ T cells from head and neck cancer patients
To further confirm that FOXP3+ cells are highly represented in HNSCC patients, we analyzed directly exvivo by flow cytometry the frequency of FOXP3+ CD4+ T cells in lymphocytes isolated from peripheral blood and two paired tumor samples from patients. We identified significantly higher frequency of circulating FOXP3+ CD4+ T cells in the blood of patients compared to a cohort of healthy subjects (p = 0.033). More strikingly and in line with immunohistochemical data, we detected a highly significant enrichment of FOXP3+ CD4+ T cells in TILs compared to peripheral blood lymphocytes from both patients and healthy donors (p < 0.001). An example of flow cytometry analysis is shown in Figure 2a and the overall results are shown in Figure 2b.
Expression and frequency of immunosuppressive molecules
We performed an immunohistochemistry analysis of ARG2, iNOS, COX2 and anti-apoptotic BCL2 protein expression. Twenty-one (60%) out of the 35 head and neck cancer patients were positive for ARG2, three (9%) for iNOS, 15 (43%) for COX2 and nine (26%) for BCL2 (Supporting Information Table 1). Representative photomicrographs of immunohistochemical staining for ARG2, iNOS, COX2 and BCL2 cells are shown in Figures 1e–1h. No association between the absence and the presence of ARG2, iNOS, COX2 and BCL2 expression and the clinicopathological variables was found (Supporting Information Table 2).
Overall survival and distribution of immune cells and expression of immunosuppressive molecules
Distribution of immune cells both in the intraepithelial and in the stromal localization, as well as expression of immunosuppressive molecules and antiapoptotic protein was correlated with overall survival. CD11c+ count superior to median in the stroma and total FOXP3+ cells (both intraepithelial and stromal localization) were significantly correlated with a better prognosis (p = 0.042 and 0.045, respectively, Figs. 3a and 3b). No significant correlation was observed between survival and plasmacytoid BDCA2+ cells or natural killer CD56+ cells count (Supporting Information Table 3). Patients presenting with ARG2-positive tumors had a significantly poorer outcome compared to the ones with ARG2-negative tumors (p = 0.035, Fig. 3c). Expression of iNOS, COX2 and BCL2 (9, 43 and 26% positive tumors, respectively) had no impact on overall survival (Supporting Information Table 3). Multivariate analysis, using Cox regression model, showed that overall survival was positively correlated with CD11c+ cells in the stroma (hazard ratio, 0.29; 95% confidence interval, 0.09–0.95; p = 0.041), and with total FOXP3+ cells (hazard ratio, 0.33; 95% confidence interval, 0.11–0.99; p = 0.048). In contrast, ARG2 did not reach statistical significance in this multivariate analysis (hazard ratio, 2.90; 95% confidence interval, 0.90–9.32; p = 0.073). Thus, CD11c+ and FOXP3+ cells are independent prognostic indicators for overall survival.
Disease-free survival and distribution of immune cells and expression of immunosuppressive molecules
As expected, nodal status significantly predicted for disease-free survival (p = 0.043, data not shown). Correlation of density of immune cells at the tumor microenvironment and expression by tumor cells of immunosuppressive enzymes and antiapoptotic protein was censored against disease-free survival but no significant correlation was observed (Supporting Information Table 4). Interestingly, patients with high total FOXP3+ cells count showed a trend toward a better outcome without reaching statistical significance (p = 0.056, Fig. 3d), as did patients with BCL2-negative tumors compared to those with BCL2-positive tumors (p = 0.055, Fig. 3e). Multivariate analysis, using Cox regression model, showed that disease-free survival was correlated with lymph node metastasis (hazard ratio, 6.51; 95% confidence interval, 1.23–33.98; p = 0.026), BCL2 (hazard ratio, 4.36; 95% confidence interval, 1.17–16.27; p = 0.028) and ARG2-positive tumors (hazard ratio, 9.70; 95% confidence interval, 2.03–46.26; p = 0.004). In contrast, FOXP3+ cells did not reach statistical significance using multivariate analysis (hazard ratio, 0.70; 95% confidence interval, 0.18–2.63; p = 0.599). Those factors were independent prognostic indicators in head and neck squamous cell carcinoma.
In our study, we performed a detailed immunohistochemical evaluation of immune cells distribution, immunosuppressive molecules and antiapoptotic protein expression in HNSCC. We showed that these tumors are weakly infiltrated by CD11c+ myeloid DCs and CD56+ natural killer cells, whereas they are highly infiltrated by BDCA2+ plasmacytoid DCs and FOXP3+ regulatory T cells. Expression of ARG2 was found in 60%, COX2 in 43%, iNOS in 9% and the antiapoptotic BCL2 protein in 26% of HNSCC tumor samples analyzed. High levels of regulatory FOXP3+ T cells in the tumor environment and the absence of ARG2 were found to significantly impact on survival of patients. In addition, the absence of BCL2 expression and high levels of FOXP3+ T cells appear to positively influence locoregional control.
Natural killer cells and DCs are implicated in early phases of antitumor immune responses. We found a low proportion of infiltrating NK cells, as observed in the previous studies.22 Although we found no correlation between the distribution of NK cells and the HNSCC patients' survival, early-stage tumors (T1–T2) had a significantly higher intraepithelial infiltration of NK cells (Table 3) as compared to advanced stages (T3–T4). These observations may indicate that changes in the tumor microenvironment notably with advanced tumor stage may play a role in the infiltration of NK cells, which are yet not clearly understood. Recently, van Herpen et al.23 showed that immunotherapy, using intratumoral administration of human IL-12, could increase the density of NK cells in HNSCC and would then improve patients' overall survival. They concluded that cytokine expression at the tumor site is important for attraction of certain immune cells, such as NK cells. A precise characterization of immunomodulatory molecule profiles in the tumor microenvironment in the various phases of cancer development could allow targeting suppressor molecules to potentiate host antitumor response. Finally, our analysis of NK cells with a single marker (i.e., CD56) did not take into account the fact that small subsets of T cells may also express this marker. We conclude that more studies are needed for the full characterization of the different subpopulations, as well as their phenotypes and functions.
DCs are specialized antigen-presenting cells and are essential mediators of immunity and tolerance.24 A couple of studies described dampening of functional competence of DCs in HNSCC patients.25, 26 Flow cytometry analysis revealed that the percentage of myeloid CD11c+ but not plasmacytoid CD123+ DCs was significantly lower in peripheral blood of cancer patients compared to healthy donors. In addition, when tumors were removed, the percentage of myeloid DCs was increased,27 suggesting that the reduction observed in the blood was owing to reversible redistribution of the cells in the tumor. In our study, we showed that myeloid CD11c+ DCs were reduced at the tumor microenvironment compared to plasmacytoid BDCA-2+ DCs. There was no significant correlation between the distribution of plasmacytoid DCs and the prognosis, but a high count of myeloid DCs in the stroma significantly correlated with improved overall survival (Fig. 3a). It is tempting to speculate that after surgical removal of the tumor, the percentage of myeloid DCs increases and promotes a protective host immune response against remaining microscopic disease, and thus improving overall survival. Following the same idea, it appears clear that decreasing the number of myeloid DCs in the microenvironment is one tumor immune escape mechanism in HNSCC patients.
Numerous studies have documented the recruitment of regulatory T cells (FOXP3+) to human carcinomas. However, their role in the anticancer immune response as well as their influence on patients' prognosis is not yet clear. Our results are in line with multiple reports, showing increased levels of regulatory T cells in peripheral blood of head and neck cancer patients compared to those measured in blood from healthy controls, as well as in tumor tissue and among TILs.28, 29 The exact function of these cells in the modulation of host immune response against cancer remains to be clarified. Strikingly, in ovarian cancer and hepatocarcinoma, they negatively influence patients' prognosis,30, 31 whereas in Hodgkin's disease, follicular lymphoma, colorectal cancer32–34 as well as head and neck cancer, regulatory T-cell infiltration appears to be associated with favorable outcome.13, 35
Interestingly, our data show a significant correlation between high numbers of total FOXP3+ regulatory T cells infiltrating the tumor and overall survival (Fig. 3b), in agreement with those studies suggesting a protective role of FOXP3+ regulatory T cells. In addition, it is noteworthy that the number of FOXP3+ T cells exhibited a significant correlation with the primary site: oral cavity and oropharyngeal cancer showed a significantly higher level of FOXP3+ T cells than hypopharynx and larynx primaries (Table 3). High infiltration of these cells in the stroma seems to correlate with the absence of locoregional metastases (p = 0.032). Infiltration of the tumor by FOXP3+ T cells could thus represent a predictive factor for lymph node metastases, notably in oral cavity and oropharynx cancers. These results could also mean that infiltration of the tumor microenvironment by FOXP3+ T cells may play a role in preventing tumor cell invasion and metastasis. It could thus be extrapolated that a rapid infiltration of tumor sites by immune cells prevents early lymph node metastasis and thus impacts on patient prognosis. The role of FOXP3+ regulatory T cells could depend on their origin. Possibly, FOXP3+ regulatory T cells may be recruited from the periphery and accumulate in tumor tissue-favoring tumor control and elimination. Alternatively, these cells may derive from CD4+CD25−FOXP3− T cells present in the tumor36 and constitute immunosuppressive cells enhancing tumor progression and invasion. Thus, prognosis could depend on the origin, type and functional attributes of tumor-infiltrating regulatory T cells, all of which remain to be determined in more detail. Alternatively, regulatory T cells could inhibit inflammatory processes in the tumor microenvironment favouring tumor progression. Indeed, head and neck cancer is considered as an inflammatory tumor37 and Erdman et al. showed in a preclinical model that transfer of regulatory T cells could delay the onset of cancer linked to inflammation.37, 38
Head and neck squamous cell carcinomas are considered strongly immunosuppressive and develop different mechanisms to evade and inhibit the host antitumor immune response. In addition to the presence of immune suppressive cells in the tumor microenvironment, head and neck tumor cells can produce soluble mediators, which can interfere with immune reactivity to cancer.39, 40 In the search for new targets to enhance immunotherapy, attention has been brought to metabolic pathways involving arachidonic acid and amino acids which have been associated with dysfunction of both adaptive and innate immune cells.41 Various enzymes implicated in these metabolic pathways have been considered as potential targets for modulation of host antitumor immune responses.
The two enzymes that are responsible for arginine catabolism are arginase and NOS, which along with products of both pathways have been implicated in the modulation of tumor growth.42 iNOS expression was significantly correlated with clinical TN stages in head and neck squamous cell carcinomas43 and with poor disease-free survival in laryngeal cancer patients.44 In our study, no correlation could be identified between iNOS expression and clinicopathological features or prognosis survival. This could be owing to a relatively small proportion of tumors in our set found positive (9%). In contrast, ARG2 expression that we identified in 60% of head and neck cancer patients was significantly associated with a poor overall survival (Fig. 3c). To our knowledge, this is the first report showing a prognostic value of ARG2 expression in head and neck cancer. These results are of particular interest as ARG2 silencing was recently shown to induce major modification on tumor properties of thyroid cancer cells. It apparently promotes apoptosis and reduces expression of cell proliferation markers,45 which, in turn, could restore T-cell activation and function at the tumor microenvironment. Some recent results reported a coexpression in prostate tumor cells of ARG2 and iNOS with a potential to induce local immunosuppression.46 In our study, ARG2 and iNOS were never found coexpressed, suggesting that in head and neck cancer ARG2 itself could be responsible for immunosuppression. Further studies should certainly aim at outlining the mechanisms by which ARG2 expression induces immunosuppression. Nonetheless, it seems clear that modulation of ARG2 expression could be relevant to enhance immune response against HNSCC.
A panel of genes is involved in the control of apoptosis, among which different members of the BCL2 family that may promote or inhibit apoptosis by synthesizing anti- and proapoptotic proteins. The antiapoptotic protein BCL2 is often overexpressed in tumors. Interestingly, patients with oral squamous cell carcinoma seem to have a higher survival rate when BCL2 expression is low.47 In our group of HNSCC, we show that 26% expressed BCL2, but no significant correlation could be found with overall survival (Supporting Information Table 3). However, the absence of BCL2 expression tended to be associated with increased disease-free survival (Fig. 3f), suggesting a potential role of BCL2 on locoregional control.
Immunosuppressive mechanisms generated by HNSCC to evade antitumor immune responses are multiple. Tumor cells are able to produce a large panel of cytokines and express molecules implicated both in the inhibition of antitumor-specific T cells and in the attraction or induction of regulatory T cells. In addition, it seems that immunosuppressive molecules may promote tumor growth and metastasis along with the inhibition of host antitumor immune response. New therapeutic strategies are being developed to control the various immunosuppressive molecules and stimulate specific cytotoxic T-cell responses. In particular, our findings suggest that ARG2 silencing could be promising for HNSCC patients.
The authors thank Nicole Prod'Hom and Véronique Noguet Brechbuehl for their precious technical assistance.