The protocol for this trial and supporting CONSORT checklist are available as supporting information: see Checklist S1 and Protocol S1.
Clinical Trial Participants
All clinical trial participants were required to be adults, ages 18 to 50 years, with long-term diabetes treated continuously with insulin from the time of diagnosis; have no demonstrable insulin secretion (fasting and glucagon-stimulated C-peptide less than 0.2 pmol/L) as assessed by a standard C-peptide assay by an outside vendor; be pancreatic GAD autoantibody positive; have a normal complete blood count (CBC); and have a negative purified protein derivative (PPD) test. Diabetic patients were excluded if they were pregnant or not using acceptable birth control; had a chronic infectious disease, including human immunodeficiency virus (HIV); had a history of tuberculosis (TB) or current TB infection; were currently receiving treatment with glucocorticoids, chronic immunosuppressive medications or high dose aspirin (>160 mg/day); or were currently living with an immunosuppressed individual. Also excluded were type 1 diabetics with keloid formation or hemoglobin A1C (HbA1C) values greater than 8%.
Non-diabetic Matched Controls
Healthy, non-diabetic control subjects were included if they were 18 to 45 years of age, with no history of autoimmune disease or diabetes, no history of HIV, and no history of autoimmunity in first-degree family members. These participants were paired weekly/bi-weekly to the diabetic patients who were randomized to BCG or placebo.
Reference Groups and Subjects
The study also included several reference groups: a reference group of type 1 diabetic individuals serially monitored for at least 20 weeks (n
57) and a one-time serial studied reference group of type 1 diabetics (n
17) studied for one outcome measure (insulin-autoreactive T cells) and matched in disease duration and age to the diabetic clinical trial subjects. The clinical trial subjects were compared to one or more of these groups, depending on the outcome measure as shown in . The criteria for inclusion and exclusion of diabetic reference subjects were the same as those for the clinical trial subjects as related to age of onset, duration of diabetes and HbA1C values. The reference subjects studied for insulin-autoreactive T cells were also matched for human leukocyte antigen (HLA)-A2 status. The serial study of these reference subjects was performed to expand the database of autoreactive T cell variation and serially studied C-peptide values in single subjects, i.e., these separate and sequential blood draws defined the biological variation in assays in single cohorts and distinguished this biological variation from variation possibly attributable to BCG treatment in the randomized clinical trial subjects also studied in a serial fashion.
CONSORT flow chart (A) and flow diagraph (B) with depicts of treatment concept, outcomes and subject comparison groups for the study.
This study was approved by the Human Studies Committee at Massachusetts General Hospital, Boston, MA and by the FDA. All patients provided written informed consent.
This was a proof-of-principle, double-blinded, placebo-controlled clinical trial that also included a paired healthy control population and reference subjects. All interventions were administered and clinical trial participants seen at one clinical center in North America (Massachusetts General Hospital, Boston, MA, USA) between 2009 and 2011. The FDA approved this protocol in 2007 and when funding was secured, the enrollment was launched in 2009.
Intervention Population and Paired Healthy Controls
For the double-blind, placebo-controlled portion of the study, diabetic subjects were randomly assigned to BCG or placebo (saline) vaccinations according to the randomization scheme prepared by the Massachusetts General Hospital (MGH) research pharmacy. The BCG injection was prepared by the research pharmacy from lyophilized BCG (TheraCys®, Sanofi-Pasteur, Toronto, Ontario, Canada), and all syringes (BCG and saline) were prefilled by the pharmacy. Randomized patients received two 0.1 ml intradermal injections into the deltoid area containing either low-dose BCG (1.6–3.2×106 colony-forming units/injection) or saline placebo, administered four weeks apart (Week 0 and Week 4). Weekly blood sampling was performed until Week 8, followed by bi-weekly blood sampling until Week 12 and then a final visit at Week 20. This frequent blood monitoring was performed to validate outcomes and observe any early effects of therapy. All subjects were seen in the morning and were required to be fasting and normoglycemic prior to having their blood drawn.
All injections were administered in the MGH diabetes clinic. Staff who administered BCG or placebo injections were not the same as those who examined the participants to grade any reactions at the injection site. All blood was processed within two hours of being drawn. All blood samples were blinded and simultaneously sent to the laboratory for monitoring of T cell response and for storage of serum for pancreas response tests (ultrasensitive C-peptide assay and autoantibodies), which were performed by outside vendors at the completion of the trial as described in “Assay methods”.
A group of paired healthy control participants, receiving neither BCG nor placebo, had blood samples obtained at the same time as diabetic subjects. Their samples were analyzed immediately for T cells in a masked fashion on the same day as the samples from diabetic subjects.
Masking and Unblinding
The MGH research pharmacy performed all masking of BCG and saline vaccinations. All blood samples that were collected were randomly coded prior to blinded submission to the MGH lab or outside vendor lab for processing. Unblinding did not occur until all samples were processed and all data were downloaded into the central computers.
Primary Outcome Measures
We monitored the safety of BCG in advanced type 1 diabetes and its action on immune and pancreas outcomes, including levels of insulin-autoreactive T cells, Treg cells, autoantibodies (including GAD), and C-peptide, an indicator of endogenous insulin secretion.
T Cell Assay Methods
The two cell-based assays (Treg cells and autoreactive T cells) were performed through Week 12.
CD4 and CD8 T cells were isolated from fresh human blood within 2 hours of venipuncture using Invitrogen™ Dynal® CD4 positive isolation kit and Dynal® CD8 positive isolation kit (Life Technologies Corporation, Carlsbad, CA, USA). This method is unique in yielding cells both free of magnetic particles and free of an attached positive selection antibody to either the CD4 protein or the CD8 protein. The blood was drawn into BD Vacutainer® tubes (BD, Franklin Lakes, NJ, USA) containing acid citrate and dextrose or ethylenediaminetetraacetic acid (EDTA). The CD8 or CD4 cells extracted for these studies were selected from fresh blood and were required, for standardization purposes, to be greater than 98% pure, 95% viable, and 85% yield for the validated T cell assays 
Use of controls
For all T cell assays in this study, a diabetic blood sample was always drawn at the same time as blood from a paired healthy control to allow assay standardization.
Detection of autoreactive CD8 T cells to a fragment of insulin
Insulin-autoreactive T cells were assayed by flow cytometry after fresh blood cell separations 
to obtain high-yield and highly pure and viable CD8 T cells for tetramer staining. Tetramers are T cell detection reagents composed of the binding region of specific HLA class I proteins with loaded peptides in the exterior binding grooves. The tetramers, which are made fluorescent, bind to specific T cells with specific reactivity to the presented peptide fragment, thereby allowing for cell identification. To detect autoreactive T cells to insulin, we used tetramers to HLA-A2 *0210 insulin beta 10–18 with a fragment of HLVEALYLV (Beckman Coulter #T02001) 
. To further confirm the specificity of insulin-autoreactive T cell detection, cell samples were examined simultaneously with T cell reagents to detect oncogene-specific human epidermal growth factor receptor-2 (HER-2) or Epstein-Barr virus (EBV)-specific T cells of acute infection. For simultaneously studied healthy controls, the following tetramer reagents were used: HLA*0201 Her-2/neu with a sequence to KIFGSLAFL (Beckman Coulter #T02001), a breast cancer peptide; HLA*0201 null without a non-specific peptide fragment (Beckman Coulter #T01044); or an EBV tetramer reagent HLA-A*0201 EBV with sequence of GLCTLVAML (Beckman Coulter #T01010).
Tetramer reagent staining was conducted on the highly pure CD8 T cells after 12 hours of culture at 26°C followed by 6 hours at 37°C and/or 1 hour rest at 26°C followed by 12 hours at 37°C. Cells were then stained with phycoerythrin-labeled class I tetramers (Beckman Coulter, Fullerton, CA) and SYTOX green dye (MBL International, Woburn, MA) and/or CD8 antibodies (BD Biosciences, San Jose, CA). All CD8 T cells were stained at 4°C in the dark for 30 minutes and then washed twice in Hanks balanced salt solution with 2% heat inactivated bovine serum. On average, 100,000 highly pure CD8 T cells were analyzed to ensure optimized data points on the Becton Dickinson FACSCalibur using the Cell Quest acquisition program and allow the detection of rare autoreactive T cells. All cells were analyzed while fresh to prevent fixation artifacts and enable quantification of dead versus viable cells. Prior to tetramer staining, cells were neither frozen nor expanded. Calculations of insulin-positive T cells were reported as the percentage of insulin-autoreactive T cells to the total numbers of isolated pure CD8 T lymphocytes.
Note that all diabetic treated patients in the randomized portion of the study were HLA-A2+ except for diabetic #iv. Although diabetic #iv was HLA-A2 negative, the formal binding site for the HLA-A2 insulin-autoreactive T cell reagent was HLA-A6802. HLA-A6802 is a subtype of the HLA-A2 family and has an identical binding cleft to HLA-A2 and other common subtypes within the HLA-A2 family. Therefore, if diabetic subject #iv were to have detectable insulin-autoreactive T cells, those cells would stain positive for the insulin-autoreactive T cell reagent. Three healthy controls in this study (Control #ii, Control #iii and Control #v) were also HLA-A2+.
Reference diabetics were monitored over a three-year period for the presence or absence of insulin-autoreactive T cells and compared to their paired healthy reference controls.
Detection of Treg CD4+ cells
Treg cells were assayed by flow cytometry after fresh blood cell separations as described above and by Burger et al 
. Two different methods of cell detection were employed. Treg cells were detected as either CD4, CD25bright
with Foxp3 staining, or with CD4, CD25bright
antibody staining. Intracellular staining of Foxp3 was performed with Human Treg Flow™ Kit (Biolegend, San Diego, CA, USA), according to the manufacturer’s instructions. Isolated CD4 positive cells were incubated briefly with CD4-PE-Cy5 (clone RPA-T4) and CD25-PE (clone BC96) antibodies for 20 minutes at room temperature. After washing, cells were fixed with Foxp3 Fix/Perm solution (Biolegend) for 20 minutes at room temperature. Cells were washed again and permeabilized with Foxp3 Perm Buffer (Biolegend) for 15 minutes at room temperature. Cells were then stained with Foxp3 Alexa Fluor® 488 antibody (clone 259D, Biolegend) for 30 minutes. Isotype controls were done for each sample prior to flow cytometric analysis. For detection of Treg cells, staining was performed with a CD4 antibody (clone RPA-T4, BD Biosciences, San Jose, CA, USA), a CD25 antibody (clone 4E3, Miltenyi Biotech, Auburn, CA, USA) and an anti-human CD127 antibody (clone hIL-7R-M21, BD Biosciences).
Flow cytometry for T cell assays
For the flow cytometry studies, the flow gates were set “open” for inclusion of CD8 or CD4 T cells of all sizes, but exclusion of the following: cell debris, red blood cells, fragmented cells, and apoptotic bodies. The “open gate” was chosen for the purified CD8 or CD4 T cells because T cells undergoing cell death, especially by apoptosis, can display changes in light scattering properties. The goal was to ensure accuracy by analyzing high numbers of cells per sample and to capture dying cells of all shapes. Cell viability was quantified by either of two stains that fluorescently labeled dead cells, i.e., Sytox (MBL international Co., Woburn, MA, USA) or propidium iodine (PI). Purified CD8 cells form distinct scatter pictures on forward versus side scatter highlighted the shrunken size of dead versus viable cells.
With open gating and inclusion of all purified CD8 T cells in each sample, some reference diabetics consistently displayed insulin-autoreactive T cells. In contrast, some reference diabetics consistently had undetectable insulin-autoreactive T cells compared to healthy reference controls, which were simultaneously studied at each monitoring time. The data were collected over the multi-year time span. The signal for insulin-autoreactive T cells was in the range of 0.06–0.09%. The healthy control background signal is in the range of 0.04–0.05% 
. The reverse was also true: diabetics who initially lacked insulin-autoreactive T cells, on repeat sampling, continued to lack those cells.
Serum Assay Methods for Pancreas Monitoring
GAD autoantibody and fasting C-peptide levels were assayed by radio-binding and ELISA assays in diabetic subjects to assess whether the subjects had a pancreas response to the BCG injection. For these serum assays, fresh human blood was collected by venipuncture into red top tubes and allowed to clot. The serum was then separated by centrifugation within 2 hours of venipuncture. Serum was stored at −80°C until analysis. The C-peptide assay was undertaken through week 20.
Detection of C-peptide secretion
Measurement of connecting peptide (C-peptide) co-secreted with insulin permits direct estimation of any remaining insulin from the pancreas in contrast with endogenous sources. The first, performed by the Mayo Clinic (Rochester, MN, USA) utilizing the Roche Cobas C-peptide assay (Roche Diagnostics, Indianapolis, IN, USA) for clinically detectable C-peptide, was used for eligibility and had a lower limit of detection of 330–470 pmol/L. This insensitive but standard assay was applied to fasting and glucagon-stimulated blood samples. After screening negative for enrollment purposes, subjects’ serum was stored and freezer-banked. For subsequent samples (baseline through Week 20), the saved serum was sent to Sweden for analysis of serial C-peptide levels by an ultrasensitive C-peptide assay with a lower level of detection of 1.5 pmol/L and an assay range up to 285 pmol/L (Mercodia AB, Uppsala, Sweden). For C-peptide values of 1.5–37 pmol/L, the within-assay coefficient of variation was 3.8%; for values of 38–60 pmol/L, it was 2.6%; and for values of 143–285 pmol/L, it was 2.5%. The Mercodia Ultrasensitive C-peptide ELISA kit, which is an FDA-listed reagent and has a filed document registration, has been evaluated for accuracy and is classified in the United States as a class one device for ultrasensitive detection of C-peptide levels. This assay is calibrated against the International Reference Reagent for C-peptide, IRR C-peptide 84/510. All statistics on C-peptide levels were performed using the lower level of detection of the assay, i.e., 1.5 pmol/L.
Detection of GAD autoantibodies
GAD autoantibodies provide evidence of diabetic autoimmunity since GAD proteins are intracellular proteins specific to insulin secreting cells and are released from T cell mediated beta cell destruction. The release of intracellular GAD results in the immune response of autoantibodies. Enrolled patients were required to be GAD autoantibody positive. Prior to enrollment, a single serum sample for GAD autoantibody was sent either to the Joslin Clinic in Boston, MA, USA (Subject #vi, Subject #i, Subject #ii, Subject #iv) or to Quest Diagnostics (Cambridge, MA, USA) (Subject #iii, Subject #iv). After the first BCG or placebo injection, serum samples collected from baseline to Week 20 were sent to Germany for diabetic autoantibody panels 
at the laboratories of Drs. Ezio Bonifacio and Peter Achenbach of the Diabetes Research Institute in Munich, Germany. The autoantibodies studied were GAD, IA-2A (islet-specific protein tyrosine phosphatase), and ZnT8Carg-A (pancreatic beta cell-specific zinc transporter) 
. The GAD assay sensitivity is 86%, specificity is 100%, and inter-assay variation is 18%. For the IA-2 autoantibody assay, the sensitivity is 72%, the specificity is 100%, and the inter-assay variation is 16%. For the ZnT8Carg-A assay, the sensitivity is 72%, the specificity is 99%, and the inter-assay variation is 17%.
Sample size for the randomized population was determined in conjunction with the FDA and with the intense use of serial biomarker studies as outlined by the Institute of Medicine guidelines for clinical trials 
. A sample size of 6 randomized patients was determined as appropriate for the intense serial blood monitoring required in this proof-of-concept trial for the placebo or BCG interventional limbs and an expanded population of diabetics and non-diabetic controls for assay validation that is referred to as reference subjects.
Randomized participants were compared to self, healthy paired controls, or reference subjects with or without type 1 diabetes, depending on the outcome measure, according to the schema depicted in . None of the analyses compared the BCG-treated to placebo-treated clinical trial subjects.
For each randomized patient, a linear regression model with auto-correlated errors was used for statistical comparisons between baseline and post-treatment periods in autoantibody levels over the course of the study. This was the appropriate test for this comparison because any change in autoantibodies should be sustained over the monitoring period of this trial, i.e., the t 1/2 of B cells that produce antibodies exceeds 60 days. P-values compared the values of each person to their post-baseline values by two-sided test based on a regression model with auto-correlated errors. For C-peptide assays, a cut-off value of 1.5 pmol/L was used since this value is the lower limit of detection of the ultra-sensitive assay used in this study. C-peptide assays were performed by the outside vendors in duplicate; figures are therefore presented as the means +/− the SE. For the comparison of EBV-infected or BCG-injected patients to the long-term diabetic reference samples, the Kolmogorov-Smirnov two-sample test was used to compare the distribution of each patient with the reference samples. We applied this method in a conservative fashion by overestimating the variability of the clinical trial sample, as a more exact comparison is difficult to obtain due to the low sampling frequency and small numbers of measurements per patient in the reference group. P-values of <0.05 were considered statistically significant. SAS® version 9.2 was used for the statistical analysis.
For serum samples sent out to commercial sources for assay performance, both published inter-assay and intra-assay variability was considered for the statistical analysis of the clinical trial samples. We also verified that the inter-assay variability was consistent in the plate for the clinical samples by comparing the pre-treatment values with all post-treatment values of the same patient to self in the same plate. This self-comparison analysis was performed for serum assays such as C-peptide or autoantibodies. The area under the curve (AUC) was calculated for all treatment and control groups, although the control group varied according to the assay.