Selection of participants and collection of baseline characteristics
Patients and HC were prospectively recruited between January and October 2012 in two French HHT centers (Montpellier and Lyon). Subjects were excluded if they were under 18 years of age or had any active or recent (< 3-month) conditions known to alter the immune system (infection, surgery, pregnancy, solid cancer or lymphoma, autoimmune disease, immunosuppressant or systemic corticosteroid therapy).
All patients included in the study fulfilled 3 or 4 of the Curaçao criteria and had undergone genetic analysis. A thoracic computed tomography scan and an echographic liver assessment were systematically proposed to all patients, according to the French guidelines for HHT diagnosis and treatment. Cerebral or gastrointestinal tract investigations were proposed only according to the clinical context, including symptoms, clinical signs and personal or familial histories.
Patients with a scheduled routine visit in the HHT centers were screened for inclusion during the medical consultation. They were considered to have a HSI if the infectious episode required at least two days of hospitalization (appendicitis excluded). For each patient included with at least one HSI, a HHT patient without any HSI and a HC were included, respecting a matching in sex and age (+/- 2 years).
Data regarding HHT symptoms, complete medical history, treatments and clinical status at the time of the study were prospectively collected. Concerning pulmonary AVM, only those large enough to be considered for treatment were taken into account. Blood samples were collected after the medical assessment. Standard biological measurements (blood cell count, C-reactive protein, ferritin, creatinine, immunoglobulins G-A-M) were made in each center according to their usual procedures.
CXCR4 and CD26 surface expressions on lymphocytes subsets
The phenotypic characterization of lymphocyte subsets was performed using whole blood and standard immunofluorescence / flow cytometry technology. All the analyses were performed on EDTA-collected blood samples, in the department of immunology of the Saint Eloi University Hospital (Montpellier - France).
The following monoclonal antibodies were used for staining: CD3-Krome Orange, CD4-PC7, CD8-APC-Alexa Fluor 700, CD56-PC5.5, CD45RA-ECD, HLA-DR-Pacific Blue (Beckman-Coulter), CXCR4-APC and CD26-FITC (BD biosciences), IgG1-APC and IgG1-FITC (Beckman-Coulter).
Analyses were realized on whole blood. Erythrocyte lysis was realized with the Immunoprep solution on TQ-prep automat (Beckman-Coulter), after antibody fixation. Cells were analyzed by a Navios flow cytometer with the Kaluza software (Beckman Coulter).
The gating strategy is detailed in Additional file 1. Total lymphocytes were identified based on morphological properties. T, T-helper, T-cytotoxic and natural killer (NK) lymphocytes were defined according to the following phenotypes: CD3+, CD3 + CD4+, CD3 + CD8 + and CD56 + CD3-. Naive T-helper and T-cytotoxic lymphocytes were defined as CD45RA + CD4 + and CD45RA + CD8+, respectively. CD4 + HLA-DR + and CD8 + HLA-DR + cells were considered as activated T-helper and T-cytotoxic lymphocytes.
The CXCR4 and CD26 expressions were measured on each subset by calculating the ratio of their MFI to the MFI of their isotype counterparts and subtracting the nonspecific antibody labelling and the cell’s auto-fluorescence.
The B lymphocyte population was simply estimated by subtracting T and NK numbers from the total lymphocytes count.
Clinical and biological characteristics are presented in tables. Immunological results are presented in bar charts with median and interquartile ranges or as individual values with Spearman r coefficient.
Comparisons between groups were made using the chi-squared, the Kruskal-Wallis or the Mann–Whitney tests, as appropriate. Correlations between variables were tested with the Spearman’s rank test.
All the univariate analyses were done with PRISM version 6.01 (GraphPad Software).
For the multivariate analysis, we used the website https://www.pvalue.io  to perform a linear regression. As the distribution of residuals did not follow a normal distribution, confidence intervals and p-values were calculated by bootstrap (1000 iterations).
A p-value < 0.05 was considered as statistically significant.