Myocardial sample identification
RA sample identification
The study was approved by the Institutional Review Board of The Johns Hopkins Hospital, which waived the need for obtaining decedent informed consent for use of archived myocardial samples and records. Myocardial samples acquired at autopsy from individuals with RA were identified using a series of ascertainment and adjudication steps. First, an initial search was conducted of the Johns Hopkins Anatomic Pathology database for mention of the term 'rheumatoid arthritis' within the text of any autopsy report performed between January 1, 1995 and July 1, 2009. The start date was chosen as RA diagnosis and treatment practices would be more consistent with current practices, and preservation and organization of samples would be superior to those stored prior. Also, case adjudication was facilitated using the Electronic Patient Record system of The Johns Hopkins Hospital, with records available from 1995 onward. We created a standard data abstraction tool for each disease group and determined, a priori, the extractable data elements that would constitute definite, possible, and unlikely for each condition. Categorization as RA required at least two outpatient clinic evaluations by a rheumatologist in which RA was identified as the diagnosis with indication of active small joint polyarthritis or clinical joint deformities typical for RA plus treatment with medications typically used for the treatment of RA (that is, biologic and non-biologic disease modifying anti-rheumatic drugs [DMARDS]). Possible and unlikely cases had more limited clinical information available. Data were abstracted by a single study team member (JP and adjudicated by agreement between two study rheumatologists (JTG and JMB). Of the 49 potential cases with 'RA' mentioned in the autopsy report, 20 adjudicated RA cases, five possible, and 24 unlikely cases were identified. Of the RA cases, 17 had available myocardial samples and constituted the RA group used for histopathology.
Control sample identification
Controls without rheumatic disease or myocarditis were selected in a similar manner as described for the RA cases and group matched on age to the RA group. For the scleroderma group, an initial Pathology database query identified 28 potential cases with specimens since 1995. Of these, further review with the scleroderma data abstraction tool yielded 14 adjudicated definite, three possible, and 11 unlikely cases. Among the definite cases, 10 had available myocardial samples. For myocarditis, only lymphocytic and eosinophilic myocarditis cases were included. Of the 56 potential cases identified from the initial Pathology database query, 14 met inclusion criteria for the desired myocarditis characteristics; however, only five had myocardial specimens available for histopathology.
Clinical characteristics
Gender and ethnicity were collected from medical records. Age at death, postmortem interval ([PMI], the time in hours between death and autopsy), and the presence and severity of coronary atherosclerosis were collected from autopsy records. Other characteristics, including cardiac history, were not uniformly available or reliable among all individuals with samples, and were thus not used for analysis beyond confirmation of diagnosis.
Histopathology and immunohistochemistry
Autopsy blocks, of the left ventricular free wall, were obtained from the archives of the Johns Hopkins Department of Pathology. Five-micron sections were cut and placed on charged slides. These were immersed in Trilogy rinse (Cell Marque, Hot Spring, AR, USA) and placed in an electric pressure cooker until reaching 127°C and 17 psi. Slides stained for citrulline were modified in a strong acid solution containing 2.3-butanedoine for 3 h at 37°C. Endogenous peroxidase activity was blocked by incubation in 0.3% H2O2 in methanol for 18 min. Non-specific protein activity was blocked by incubation with a non-serum protein solution (DAKO Corporation, Carpinteria, CA, USA). Slides were incubated with antibodies targeting PAD1 (1:1000 Abcam, Cambridge, MA, USA), PAD2 (1:200, Abcam), PAD3 (1:500, Abcam, PAD4 (1:150, Sigma-Aldrich, St Louis, MO, USA), PAD6 (1:150 Abcam) or citrulline (1:1000, Millipore Corporation, Billerica, MA) overnight at 4°C. For anti-PAD1, slides underwent an additional treatment with HRP polymer detection kit (Invitrogen, Carlsbad, CA, USA). Sections were then incubated with biotinylated donkey anti-rabbit IgG (Jackson ImmunoResearch Laboratories, West Grove, PA, USA).
Immunoperoxidase staining was performed using Vectastain ABC Elite (Vector Labs, Burlingame, CA, USA). The avidin-biotin complex was visualized using 3.3' diaminobenzidine (DAB) peroxidase substrate (Vector Labs). Sections were counterstained in hematoxylin (Richard-Allen, Kalamazoo, MI, USA).
All cardiac tissue scoring of the myocardium was performed by a single-blinded cardiovascular pathologist (MKH). Citrullination staining was based on a 5-point scale (1 to 3 in 0.5 increments) corresponding to minimal, moderate, and marked citrullination staining. Both an overall staining score and a peak staining score were generated for each tissue as there was some heterogeneity in staining intensity across the slides. Cardiac hypertrophy was measured subjectively based upon the surrogate nuclear features of size, hyperchromasia, and irregularity of the nuclear membranes [10]. Myocarditis was initially reported in the autopsy record and was confirmed through the histopathologic findings of lymphocytic infiltration causing myocyte injury. Areas of fibrosis included both interstitial (fibrosis between myocytes) and replacement (fibrotic bands replacing myocytes) types and were scored using the same semi-quantitative scale. Scoring of PAD staining was also performed blinded using the same 1 to 3 scale. Staining was independently evaluated and scored for the myocytes, smooth muscle cells, endothelium, and leukocytes.
Statistical analysis
The distributions of clinical and histologic characteristics were examined and compared between groups using analysis of variance (ANOVA), with Bonferonni correction, for normally distributed continuous variables, the Kruskal-Wallis test for non-normally distributed continuous variables, and the chi-square goodness of fit test or Fisher's exact test, as appropriate, for categorical variables. Generalized linear models were constructed to explore the associations of patient characteristics with average and peak citrullination scores in the RA and control groups, with heterogeneity in the associations between groups tested using analysis of co-variance (ANCOVA). The Shapiro-Wilk test was used to ensure normality of the modeled outcome variables across the extent of independent variables. All calculations were performed using Intercooled Stata 10 (StataCorp, College Station, TX, USA). In all tests, a two-tailed α of 0.05 was utilized.