Regimens of chronic immunosuppression—eg, ciclo-sporin—have shown promise for attenuation of the loss of insulin secretion in new-onset disease, but have unacceptable toxic effects (potential risk of infections and tumours from continuous immunosuppression and nephrotoxicity). Antigen-specific therapies to restore β-cell tolerance have shown low toxic effects but little efficacy.^5,6 Non-antigen-specific short-course therapies, such as anti-CD3 and anti-CD20, have had more success.^7,8 Of these, anti-CD3 had a durable effect, with efficacy up to 4 years after one 1-week treatment in a pilot study, and longlasting efficacy in non-obese diabetic mice.^9,10

Teplizumab is a humanised, anti-CD3 monoclonal antibody that has been mutated to greatly reduce Fc receptor and complement binding.¹¹ In an early trial of anti-CD3 antibody,¹² 24 patients with recent-onset diabetes were randomised equally to receive open-label teplizumab (34 mg cumulative dose for one 14-day course in a 70 kg individual) or no antibody for 14 days, with daily dose based on previous transplantation trials. At 12 months, C-peptide response to a mixed meal was maintained in 60% of treated patients versus 8% of controls (p<0·03).

In a trial of otelixizumab,¹³ another monoclonal anti-CD3 antibody with reduced binding to the Fc receptor, β-cell function was preserved in patients receiving otelixizumab and their insulin needs were decreased up to 48 months after treatment. Adverse events, including Epstein-Barr virus reactivation, were more frequent than in the teplizumab trial,¹² which is consistent with the higher cumulative dose.¹⁴ A much lower dose of 3·1 mg otelixizumab was subsequently used in a phase 3 trial, but the primary efficacy outcome of change in C-peptide at month 12 was not met.¹⁵

Baseline characteristics were well balanced across treatment groups (table 1), but varied by region (webappendix p 7). Most notably, patients in India had higher HbA_1C and insulin use, lower AUC of C-peptide (characteristic of later disease stage), and lower frequency of anti-ICA512 antibodies than did those in other regions.

HbA_1C and insulin use decreased after patients entered the study, and then increased after day 91 (webappendix pp 8–9). Mean changes in HbA_1C from baseline were essentially identical in the 14-day full-dose and placebo groups at all times. The 14-day full-dose group had numerically larger decreases in mean insulin use relative to the placebo group, and the difference was maintained at all times (webappendix pp 8–9). Differences between the 14-day full-dose and placebo groups were not significant for the primary and secondary outcomes (table 2). The amount of missing data for these outcomes was generally small. Results were largely unchanged in sensitivity analyses in which alternative methods were used for imputation of missing data.

AUC of C-peptide was analysed with non-parametric methods because data were not normally distributed. Median change from baseline to 1 year showed less decline in the 14-day full-dose group than in the placebo group (figure 2A, table 2). The shift between treatment groups is evident from the empirical cumulative distribution curves in the overall population, in which 40% of the 14-day full-dose group had a preservation or increase in AUC of C-peptide relative to baseline compared with 28% of the placebo group (figure 3A).

Change in insulin use from baseline did not differ between the 14-day full-dose group and placebo (table 2). However, at any given HbA_1C threshold, a greater percentage of patients in the 14-day full-dose group achieved the threshold at lower insulin dose cutoffs than in the placebo group (figure 4A, webappendix p 10). This effect was most notable at insulin doses of lower than 0·5 U/kg per day. 5% (19/415) of patients receiving teplizumab were not taking insulin at 1 year compared with no patients in the placebo group (p=0·03), and 15 of these 19 patients had HbA_1C of less than 7%. In the Diabetes Prevention Trial,⁵ insulin was given to the relatives of patients with type 1 diabetes at a dose of 0·25 U/kg per day because this dose did not increase the occurrence of hypoglycaemia.⁵ We therefore used the dose of 0·25 U/kg per day as a cutoff in subsequent analyses. The proportion of patients who achieved HbA_1C of less than 7% and insulin use of less than 0·25 U/kg per day was greater in the 14-day full-dose group than in the placebo group for the study overall (table 2, figure 4A, webappendix p 10) and at each study visit (figure 5). A similar effect was seen at HbA_1C of less than 6·5% and insulin use of less than 0·25 U/kg per day (p=0·02).

The lower reduction in median AUC of C-peptide in the 14-day full-dose group than in the placebo group was evident in three predefined subgroups (table 3): children aged 8–11 years (figure 2B, figure 3B), the US region (figure 2B, figure 3C), and patients randomised at 6 weeks or fewer after diagnosis (webappendix p 11). Because of violations of normality assumptions, inferen tial comparisons were made with non-parametric methods.

In the subgroup of the US region, the proportion of patients who achieved an HbA_1c of less than 7% at each low dose insulin cutoff was higher in the 14-day full-dose group than in the placebo group (figure 4B). A similar pattern was seen in children aged 8–11 years and patients randomised at no more than 6 weeks after diagnosis, but data are not shown because sample sizes were small. Furthermore, in the subgroups of children aged 8–11 years, the US region, and patients randomised at 6 weeks or fewer after diagnosis, the proportion of patients who achieved HbA_1C of less than 7% and insulin use of less than 0·25 U/kg per day was greater in the 14-day full-dose group than in the placebo group at all timepoints (webappendix pp 12–14).

In the safety analyses, the proportion of patients who had adverse events (512/516 [99%]) and serious adverse events (51/516 [10%]) was similar across the four study groups (table 4). The proportion of patients with infection did not differ between the four treatment groups (table 4). Although the study population had a high occurrence of anti-EBV IgG at baseline (441/513 [86%]), 22 of 415 (5%) patients receiving teplizumab became positive for anti-EBV IgG or IgM during the study compared with seven of 98 (7%) receiving placebo, and only one patient (teplizumab group) had a small transient increase in EBV viral load. 77% (318/415) of patients treated with teplizumab (150/207 [72%] in the 14-day full-dose group, 78/102 [76%] in the 14-day low dose group, and 90/106 [85%] in the 6-day full-dose group) and 13% (13/98) of those receiving placebo developed anti-drug antibodies, defined as a titre of more than 1:100 when samples were obtained at 28 or 56 days.

Exploratory analyses in all patients who received treatment and in subgroups showed effects similar to those seen in previous, smaller studies of anti-CD3 antibody: C-peptide secretion was preserved, allowing glycaemic control to be achived at a lower insulin dose in the teplizumab groups than in the placebo group. A greater proportion of patients in the teplizumab groups was able to discontinue or use very low doses of insulin than in the placebo group. Investigators were instructed to maintain an insulin dose of more than 0·25 U/kg per day, which might have diminished the effect of teplizumab on insulin use. Although discontinuation of insulin was never a study objective and was driven entirely by glycaemic status, reduced insulin requirement with maintenance of glycaemic control supports a biological effect of teplizumab. Studies of other immune modulators have not consistently achieved this outcome.⁸ During the course of Protégé, accumulating evidence suggested that C-peptide concentration might be a more reliable outcome because C-peptide is a more direct indicator of endogenous insulin secretion than is HbA_1c or insulin use, or both, and it is independent of manipulation by patients or physicians. Our results support this view, showing a treatment effect on C-peptide changes in exploratory analyses.

The effect of teplizumab was only seen with the highest dose (two courses of 17 mg, 6 months apart, in a 70 kg individual on the basis of a cumulative dose of 90343μg/m² and a body surface area of 1·92 m²), yielding a cumulative dose per cycle that was half that in the previous study of teplizumab¹² (34 mg cumulative dose for one 14-day course in a 70 kg individual). Otelixizumab has a similar specificity to the ε-subunit of CD3, but is produced by different methods, and has different affinity and structural features from teplizumab.²⁰ The teplizumab dose seems to be lower than the dose used in the phase 2 trial of otelixizumab (48 mg cumulative dose),¹³ but higher than the dose in the phase 3 trial (3·1 mg cumulative dose).²¹ The dose regimen we used was not associated with severe reactions or biological signs of EBV reactivation, which was a problem in the otelixizumab trial, and, thus, had an acceptable safety profile in view of the seriousness of type 1 diabetes.¹⁴ More research is needed to establish whether higher teplizumab doses (eg, closer to that in the previous study¹²) might improve efficacy without impairing safety, and whether repeated dosing improves efficacy or is countered by the development of teplizumab antibodies and late timing after diagnosis.

Limitations of this study include the failure to achieve the primary outcome, necessitating the use of post-hoc (hypothesis-generating) analyses, and a need for additional follow-up to examine long-term safety and efficacy; 2-year results are forthcoming soon. Another limitation was the lack of extensive data for hypoglycaemia, which was not captured on diary cards. Future studies could benefit from documenting hypoglycaemic events more thoroughly, particularly because hypoglycaemia is an important clinical problem.

The study also identified subgroups that might be more likely to respond to treatment. Larger treatment effects were associated with younger age, recruitment from the USA, and earlier treatment. The reasons for these differences might be as much a reflection of the behaviour of the study subpopulations as the effect of the drug, but this possibility does not detract from their potential importance. This finding underscores the heterogeneity in type 1 diabetes and could have implications for showing efficacy of immune interventions in future trials. For example, the rate of β-cell decline is faster in children than in adults. Immunotherapy might, therefore, be particularly effective for improvement of clinical management of the disease in children—an important finding because the management of diabetes in children is particularly challenging, with higher rates of both hyperglycaemia and hypoglycaemia.^4,7 Also, US patients generally had higher C-peptide concentrations, lower insulin use, and lower HbA_1C at baseline than did patients in other regions, suggesting a more advanced stage of disease progression in these other regions, especially India. The increased response in patients from the USA and those treated within 6 weeks of diagnosis suggests that there is an advantage to treating early in the course of disease. Future studies might benefit from use of metabolic entry criteria that indicate a fairly early stage of disease.