In this retrospective study, we selected all consecutive patients who had a total hysterectomy for type 1 EC between 1 January 2000 and 4 November 2013 in the Department of Gynecology and Obstetrics, Creteil University Hospital, France. Type 1 EC is defined by to the European Society of Medical Oncology (ESMO) group as endometrioid histological subtype. Exclusion criteria were type 2 EC, unavailable pathology specimens or absence of cancerous cells on hysterectomy pathology specimens. The flow chart of the patients included in the study is presented in Figure 4
. Clinical data have been reported on an Excel database, including age, BMI, gestity, parity, menopausal status, hormonal treatments, history of diabetes, hypertension, dyslipidemia, familial history of cancer, radiological data (computed tomography, sonography, and magnetic resonance imaging), pathological data, FIGO stage, adjuvant treatment, progression free survival, and overall survival data. We used the classification from the International Federation of Gynecology and Obstetrics (FIGO) 2009. Patients were clinically followed every 4 months during the first three years then every six months up to five years then every year according to our national guidelines. The mean follow up was 41 months. Mean progression free survival and overall survival were 36.9 and 40.7 months, respectively. The study protocol was approved by the Ethics Committee of Paris X, France (14 January 2015; No. 2015-01-03).
We conducted IHC analysis from hysterectomy specimen of 136 patients with type 1 endometrial cancer. We used archive paraffin-embedded blocks of formalin-fixed hysterectomy pathology specimens processed by routine pathology. Tissues were fixed in formalin (10%) and then processed as paraffin blocks. Four micron-thick sections of formalin-fixed tissues were deparaffinized in a xylene substitute (EZ prep®) and rehydrated at 75 °C. The sections were immunostained using the Ventana Benchmark GX® automated immunohistochemistry system (Optiview™ and Ultraview™, Universal DAB-Ventana®).
For IHC, the pathologists (IA and DM) chose the slides after H&E staining according to the following criteria: (i) significant tumor area; (ii) without necrosis; (iii) without artefacts of formalin-fixation; and (iv) without calcospherites. Indeed, we used one slide per antibody per patient. All slides came from the same block for one patient and were consecutive to analyze the same area.
4.3. Adipokines and Hormonal Receptors Immunostaining
We used a rabbit monoclonal antibodies directed against ER (1/100, SP1, MMFrance®, Brignais, France), PR (1/100, SP2, MMFrance®, Brignais, France) and SPARC (1/50, D10F10, 8725S, Ozyme®, Montigny-Le-Bretonneux, France). We used a mouse monoclonal antibodies for adiponectin (1/100, 1B2, TA503801, Origene®, Rockville, MD, USA), RBP4 (1/100, 4D9, LS-B6142-50, LifeSpan Bioscience®, Seattle, WA, USA), IL-6 (1/100, ab9324, Abcam®, Cambridge, UK), TNF α (1/100, 6C10, LS-B7086-50, LifeSpan Bioscience®, Seattle, WA, USA). An antigen retrieval procedure was run including an incubation at 95 or 100 °C with CC1 Ventana® for 32 min for adiponectin, RBP4, IL-6 and SPARC or 60 min for ER, PR and TNF α. This automated procedure is based on an indirect biotin-avidin system. A universal biotinylated immunoglobulin was used as a secondary antibody, 3,3-diaminobenzidine as the substrate and hematoxylin as the counterstain. Positive controls were sections of human breast tissue for ER and PR, thyroid carcinoma for adiponectin (according to the manufacturer’s instructions), Langerhans islets for RBP4 (according to the manufacturer’s instructions), muscularis mucosae of vessels for IL-6 (according to the manufacturer’s instructions), endothelium for SPARC (according to the manufacturer’s instructions), lymphocytes, and neutrophils for TNF α (according to the manufacturer’s instructions). Signal was amplified for TNF α (Amplification kit Ventana®). Positive controls were breast adenocarcinoma for IL-6, thyroid carcinoma for adiponectine, pancreas for RBP4, liver for TNF α, endothelium for SPARC, and breast cancer for ER and PR.
4.5. Selection of the Groups
Expression of adipokines and hormonal receptors was evaluated in the general population (136 patients). We then compared the obese group (55 patients, 40.4%) and non-obese group (81 patients, 59.6%), and finally, we compared their expression in three prognosis groups in obese population.
The first prognostic group included the patients with recurrence (vaginal, pelvic, lymph nodes, or general) in comparison with a no recurrence group. The second prognostic group included the patients with positive lymph nodes (LNs) (from pelvic and/or para aortic lymphadenectomy) in comparison with a negative LNs group. The third prognostic group included the patients with high risk stage I tumor defined by FIGO stage IB and histological grade 3 or lymphovascular space involvement (LVSI) in comparison with a low risk stage I group defined by stage IA and histological grade 1, 2, or 3, or stage IB and histological grade 1 or 2 without LVSI. The low-risk stage I group encompassed the low and intermediate risk stage I defined by the European Society for Medical Oncology (ESMO) 2013 guidelines [35
], which we added patients without LVSI, in order to simplify the statistical analysis.
4.7. Supervised Clustering
The algorithm previously described by Dettling et al. was used for the supervised clustering [36
]. The aim of this supervised algorithm is to identify protein clusters that are strongly associated with a supervised categorical response y (obese and non-obese) i.e., whose average expression profile has great potential for explaining the response (patients groups discrimination), given a small number of sample tissues with the expression activities of multiple proteins. The difficulty was that we knew neither the cluster size, nor the number of clusters (q
). This method uses a clustering criterion, S
, corresponding to a (penalized) goodness-of-fit measure from a penalized logistic regression analysis. First, the expression value of every protein is standardized to zero mean and unit variance. Variable selection and grouping are done by a stepwise forward search i.e., by trying each protein and increasing the group by the protein that optimizes the criterion S
. After each forward search, it continues by means of a backward pruning step to root out proteins that have been incorrectly added to the group at the earlier forward stages. When the cluster can no longer be improved, a new cluster is started. The grouping procedure is supervised since all the decisions are based on optimizing the criterion S
that measures the ability of the groups to explain the variable response y. By computing the grouping criterion directly from multiple groups instead of single groups only, we could obtain the best interacting protein groups that explain the response y as an ensemble. This method is advantageous in that it allows overlapping groups and that the groups together contribute most in predicting the response y
. Cluster centroids (i.e., mean expression of the co-expressed proteins within a cluster) can be interpreted as a protein signature that is strongly differentially expressed and carries substantial information about predicting y
. We used a bootstrap method with 1000 replication to specify the confidence interval of misclassification rate.
This statistical method has been used to distinguish atypical hyperplasia and grade 1 endometrioid endometrial cancer [37
] or atypical and non-atypical endometrial hyperplasia [38
] based on immunohistochemical markers of endometrial tissue samples.
Data were managed with an Excel database (Microsoft, Redmond, WA, USA) and analyzed using R 2.15 software with the Supclust library, available online (R: A Language and Environment for Statistical Computing, R Development Core Team, 2.15, 2014, Available online: https://www.r-project.org/