High proinsulin: C-peptide ratio identifies patients with Stage 2 type 1 diabetes at high risk for progression to clinical diagnosis and responses to teplizumab

Abstract

Aims/hypothesis:

Tractable biomarkers to identify immunotherapy responders are lacking in type 1 diabetes. We hypothesized that proinsulin:C-peptide ratios (PI:C), a readout of beta-cell stress, could provide insight into type 1 diabetes progression and response to immunotherapy.

Methods:

We analyzed intact PI:C and relationships with progression to Stage 3 diabetes in longitudinal serum samples from 63 participants with Stage 2 type 1 diabetes in the TrialNet teplizumab prevention study.

Results:

Elevated baseline PI:C was strongly associated with more rapid diabetes progression in both treatment groups, but teplizumab abrogated the impact of high pre-treatment PI:C on type 1 diabetes progression. The differential responses of drug treatment in those with high and low PI:C ratios were independent of differential treatment effects on the PI:C ratio or relevant immune cells.

Conclusions:

High pre-treatment PI:C identified individuals with Stage 2 type 1 diabetes exhibiting rapid progression to Stage 3 disease that displayed benefit from teplizumab treatment. These data suggest that readouts of active disease such as PI:C ratio could serve to identify optimal candidates or timing for type 1 diabetes disease-modifying therapies.

Introduction

Type 1 diabetes results from progressive autoimmune beta-cell destruction (1). As type 1 diabetes develops, inflammatory mediators and immune cells, inflict direct and metabolic stress on beta cells (2). These stressors may lead to oxidative stress and endoplasmic reticulum (ER) stress via unfolded protein response (UPR) induction and activation of proapoptotic signals (2–6). These pathways are intrinsic to beta cells, but a marker of these events in the peripheral blood would be a valuable tool to track the effects of immune therapies on beta cells or to determine when retreatment is appropriate. Our group and others have shown that prior to the clinical diagnosis of type 1 diabetes, the circulating proinsulin:C-peptide ratio (PI:C ratio) can be used as a biomarker of multiple stress pathways, whereby elevations in PI:C identify disease progression (7–12). However, stage-specific impacts of PI:C ratio on progression and relationships with type 1 diabetes prevention are unknown.

The anti-CD3 mAb teplizumab, was recently approved for treatment to delay type 1 diabetes diagnosis (Stage 3 type 1 diabetes) in persons with multiple islet autoantibodies and dysglycemia (Stage 2 type 1 diabetes). In the TrialNet teplizumab prevention trial, TN10, beta-cell function, measured as the C-peptide response to an oral glucose tolerance test (OGTT), improved over the first year after therapy compared to deterioration in placebo-treated participants (13; 14). Despite the improvement in stimulated C-peptide responses, while on-treatment, cross-sectional PI:C values were similar between placebo and teplizumab groups (14). These findings suggested that in the overall study cohort, beta-cell stress persisted despite improved beta-cell function with teplizumab treatment. Here, to better understand this surprising finding, we took advantage of this unique and invaluable study cohort to define individual changes in PI:C over time, the relationships between PI:C and progression to Stage 3 type 1 diabetes, and relationships to other mechanistic markers of treatment effect in the successful TN10 prevention trial.

Research Design and Methods

Trial design and participants:

The design of the double-blinded randomized TN10 teplizumab prevention study has been reported (13). Briefly, relatives of individuals with type 1 diabetes, ≥8 years of age, with Stage 2 disease received a 14-day intravenous infusion of teplizumab or saline, then were followed longitudinally with OGTTs after 3 months, then every 6 months for diabetes development. Institutional review board approval with consent and assent as appropriate was obtained as described. The trial was registered on ClinicalTrials.gov NCT01030861.

Sample collection and measurements:

Serum samples were available for 63/76 total study participants. Overall, these participants had similar baseline characteristics as participants from whom samples were not available (n=13, Supplemental Table 1).

Pre-treatment samples were obtained from the screening visit (n=58, randomly timed), baseline visit (n=17, fasting) and approximately every 6 months (fasting) after start of treatment. Screening visit samples were utilized for analyses; if a sample at the screening visit was not available, baseline visit samples were used. When available (n=12), randomly collected vs. facting intraindividual PI:C ratios from both visits were highly correlated (r = 0.993, p<0.0001). Thus for participants with only fasting baseline visit samples (n=5), values were transformed using a linear regression model to approximate screening visit values.

OGTT C-peptide levels (pmol/ml) were measured with the TOSOH C-peptide assay. Proinsulin (pmol/ml) was measured using the Teco intact proinsulin ELISA (performance characteristics externally validated and reported in detail(15)). PI:C values were calculated as an equimolar ratio x 100. Average on-study PI:C was calculated as described (16).

Peripheral blood mononuclear cells (PBMCs) were processed and stored at the NIDDK repository, then analyzed via flow cytometry analysis by the Immune Tolerance Network Core at Benaroya Research Institute as described (16; 17).

Statistical Methods:

In this post-hoc analysis, between-group differences were compared using Wilcoxon rank sum tests or chi-square or Fisher’s exact tests. Correlations were assessed using Spearman rank correlation tests. Cutpoint analyses in relation to time to Stage 3 type 1 diabetes used the aggregate mean and were validated with recursive partitioning analyses (rpart package). Ratio and fold change variables were log₂ transformed for model inclusion, and C-peptide AUC measures were log(X+1) transformed. Time to Stage 3 type 1 diabetes was estimated using the methods of Kaplan and Meier, and Cox regression models were used to assess influence of markers of interest in univariate, multivariable, and time-dependent covariate analyses. Time-dependent receiver operating curve (ROC) and area under the curve (AUC) analyses (timeROC package in R) as well as C-indices (compareC package in R) were utilized to evaluate the discriminative ability of baseline PI:C ratio as a prognostic factor for the time to progression to Stage 3 T1D. Averaged AUC PI:C ratio to assess aggregate beta cell stress across study time points was evaluated using the trapezoidal method divided by number of months. The pretreatment PI:C ratio was highly correlated with the on-study averaged AUC of the PI:C ratio (Pearson r= 0.785, p<0.0001) (Supplemental Figure 1). Therefore, for modeling and assessment of treatment effects, corrections for the baseline PI:C were made. Given the hypothesis-generating nature of these analyses, multiple comparison corrections were not used, and statistical significance was defined as p<0.05. Analyses were conducted using the R statistical package (v.4.0.3, https://www.R-project.org/)).

Results

Evaluable Cohort Characteristics

Demographic features at baseline (n=63) (Supplemental Tables 1–3) were similar across treatment groups. The mean pre-treatment PI:C ratio (Figure 1A) was 0.48 (median: 0.42; range: 0.06–1.42) for teplizumab-treated participants (n=41) and. 0.49 (median: 0.36; range: 0.14–1.60) for placebo (n=22) (p=0.87). Pre-treatment absolute proinsulin values were also similar across treatment groups (Supplemental Table 2, p=0.49). Similar to previous observations (8; 15), age at enrollment and the pre-treatment PI:C ratio were inversely related (Figure 1B; r = −0.36; p=0.004).

Figure 1. In individuals with stage 2 type 1 diabetes PI:C is inversely associated with age, and predicts progression to stage 3 type 1 diabetes.

A. Pre-treatment PI:C values for 63 individuals participating in the TrialNet Teplizumab prevention study. Values were not different between treatment groups (p=0.87). B. Age was inversely related to pre-treatment PI:C ratios (r = −0.36; p=0.004). C. Kaplan-Meier curve showing distribution of the time to progression to Stage 3 diabetes based on high vs. low baseline PI:C ratio of entire cohort, regardless of treatment group, to Stage 3 diabetes, with multivariable cox regression adjusting for age and baseline C-peptide area under the curve.

The PI:C ratio predicts time to diagnosis of Stage 3 type 1 diabetes

Participants were followed for a median of 72 months (95% CI: 61–87 months). Stage 3 type 1 diabetes was diagnosed in 41 participants (23/41 in the teplizumab and 18/22 in the placebo arm). The median time to diagnosis was somewhat longer than reported in the complete study cohort amongst the placebo group (35.35 months vs teplizumab, 59.7 months, log-rank p=0.037). We evaluated the PI-C ratio and its influence on prognosis (time to progression to Stage 2 T1D) as both a continuous and a dichotomized measure (Supplemental Table 3). First, we evaluated the data based on a continuous measure of the PI:C ratios. When both treatment arms were pooled, the pre-treatment PI:C ratio as a continuous measure was significantly associated with the likelihood of progressing to Stage 3 type 1 diabetes (HR=2.29, 95% CI: 1.45 to 3.61; p=0.0004), even after adjusting for age and baseline AUC C-peptide. We also examined the association with the pre-treatment absolute proinsulin levels and time to Stage 3 diabetes, but this relationship was not significant after adjustment for age and baseline C-peptide AUC (p=0.97).

To facilitate comparisons between groups, we also dichotomized the pre-treatment PI:C ratio as “high” vs. “low” in approximation to the overall mean and based on partitioning analyses. The cutpoint identified approximated the mean baseline PI:C ratio for the placebo group (i.e. ~0.50) supporting this approach to this exploratory evaluation. High ratios (i.e. ≥0.5%) were present in 7/22 (32%) of placebo-treated participants patients and 16/41 (39%) of teplizumab patients (p=0.77) (Supplemental Table 3). Participants with high ratios were younger compared to those with low ratios (median: 15.3 vs. 20.8 yrs, respectively; p=0.017). Compared to participants with low ratios, those with a high pre-treatment PI:C ratio had similar pre-treatment AUC C-peptide values (p=0.50); rather, a high pre-treatment PI:C was linked to higher absolute proinsulin values (median 5.59 vs. 1.90 pmol/L, p<0.001). Those with high pre-treatment PI:C ratio values were at significantly greater risk of progressing to Stage 3 diabetes (median: 18.8 months) than those with low pre-treatment PI:C ratios (median: 75 months) even after adjusting for age and baseline C-peptide AUC (HR (95% CI): 3.30 (1.63–6.68); p=0.0009); Figure 1C, Supplemental Table 4).

Teplizumab treatment delays progression to Stage 3 diabetes in subjects with high PI:C ratio

We determined whether the pre-treatment PI:C ratio affected the outcome of teplizumab treatment and the risk to progression to Stage 3 type 1 diabetes (Figure 2, Supplemental Table 5). When used as a continuous or dichotomous variable, a high pre-treatment PI:C ratio had a profound impact on the time to Stage 3 type 1 diabetes, as did the treatment arm. To further evaluate the discriminative ability of baseline PI:C ratio as a prognostic factor for the time to progression to Stage 3 diabetes, as shown in Table 1, we used time-dependent ROC and AUC analyses as well as C-indices, including treatment group, age, and baseline PI:C ratio as either a continuous or a dichotomized measure. Overall, models incorporating baseline PI:C ratio performed better with higher AUC across time points (6-yr AUC= 64.1% vs. 80.8% for continuous PI:C or 80.5% for dichotomized PI:C) and lower (better) C-indices (C=0.71 vs. 0.72 for continuous PI:C or 0.276 for dichotomized PI:C).

Figure 2. Relationship between pre-treatment PI:C, type 1 diabetes progression, and teplizumab treatment.

A. Kaplan-Meier Curve showing rates of progression amongst participants from both treatment groups, stratified based on high vs. low pre-treatment PI:C ratio (above or below 0.5%). Cox regression stratified by pre-treatment ratio and treatment group showed that differences in Stage 3 progression were only statistically significant for participants starting the trial with a high pre-treatment PI:C ratio.

Table 1.

Evaluation and comparison of time-dependent ROC, corresponding AUC, and C index measures

Model	C-index	p value vs. Model 1	6-year AUC*	p value vs. Model 1
1. Teplizumab (vs. placebo) + age	0.63	---	64.1%	---
2. PI:C ratio (continuous) + Teplizumab (vs. placebo) + age	0.71	0.028	80.8%	0.008
3. PI:C ratio ≥0.5 (vs. <0.5) + Teplizumab (vs. placebo) + age	0.72	0.027	80.5%	0.019

^*Median follow-up was 72.2 months, and thus for the purposes of this table we provided the time-dependent AUC at 6 years.

We further identified a significant interaction between treatment arm and pre-treatment PI:C ratio on the time to progression to type 1 diabetes (p=0.017 for treatment group as effect modifier). In a subset-stratified analyses (Figure 2), teplizumab mitigated progression to Stage 3 diabetes the treatment effect was most pronounced in those with a high ratio at baseline (p=0.002) vs low ratio (p=0.1)(Supplemental Table 5B).

Drug treatment did not consistently reduce PI:C values in participants with either high or low pre-treatment PI:C

To better understand the potential mechanisms of this effect modification of teplizumab on the influence of high pre-treatment PI:C ratio on risk of Stage 3 type 1 diabetes, we evaluated the study average AUC PI:C ratio, adjusted for age and the baseline value and how these differed between treatment arms. This study-average AUC PI:C ratio did not differ significantly between treatment arms (p=0.36), (Figure 3A–B). We performed an exploratory analysis across all of the time points to determine whether there may be transient differences in the response to drug treatment (Supplemental Table 6A, Supplemental Figure 2). This analysis did not identify a consistent impact of treatment on PI:C, with only a transient suggested difference in the values at month 12 (a greater decline in the teplizumab group). This change was only seen in those with high, but not low pre-treatment PI:C (Supplemental Table 6B).

Figure 3. Average study AUC PI:C ratios were not different between treatment groups.

Average study AUC PI:C ratio corrected for time in study were calculated and adjusted for pre-treatment ratio and age. A: Average study AUC PI:C ratio by treatment arm. B. Average study AUC PI:C ratio by treatment arm and progression to Stage 3 type 1 diabetes. No significant differences were present for any comparisons. n= 22 for placebo group and 41 for teplizumab treated group; 23/41 participants in the teplizumab and 18/22 participants in the placebo arm developed stage 3 type 1 diabetes.

Relationships between MHC, changes in immune cells with teplizumab and PI:C ratios

In the original study report we found that individuals who were HLA-DR4+ or HLADR3- had more robust responses to teplizumab than those who were DR4- or DR3+ (13). Therefore, we evaluated the influence of HLA-DR3 and HLA-DR4 status in these multivariable models to assess if PI:C ratio is still influential. In the model using continuous pre-treatment PI:C ratio (log2 transformed), the PI:C ratio maintained its significance while DR3 and/or DR4 presence was not significant (data not shown).

There may have been immune cell differences among those with high and low pre-treatment PI:C that could account for the effects of teplizumab in the high-ratio group. At 3 and 6 months, teplizumab has been shown to increase partially exhausted CD8+ T cells, defined by expression of EOMES and co-expression of KLRG1 and TIGIT. When the groups, defined by treatment and PI:C ratio were compared using a mixed model, there was an effect of drug treatment on the CD8+CD45RO+CCR7+EOMES+ and the CD8+CCR7+KLRG1+TIGIT+ T cells (p=0.002 and 0.0595, respectively) but we did not identify a significant effect of the PI:C ratio on the frequency of these cells or an interaction between drug treatment and either of the T cell subsets (Supplemental Figure 3).

Discussion

We found that in participants with Stage 2 type 1 diabetes, progression to Stage 3 disease is more rapid in those with high (≥0.5%) vs low (<0.5%) PI:C ratios. Teplizumab treatment was most effective in attenuating Stage 3 progression in those with high pre-treatment ratios. Here, placebo-treated individuals with high pre-treatment PI:C exhibited rapid development of Stage 3 type 1 diabetes, while this relationship was much less pronounced in those receiving teplizumab treatment. The differences in the responses that we found to teplizumab treatment between the individuals with high and low PI:C rations could not be explained by a differential effect on immune cells. In aggregate, these data suggest that the drug activity may be greatest in the group identified with this biomarker of aggressive disease and that the differential response to teplizumab is due to differences in active beta cell stress rather than differences in the response of immune cells to treatment.

Our observations that teplizumab’s effectiveness was best when there is active islet stress are consistent with preclinical studies of anti-CD3 in nonobese diabetic mice, human studies in recent onset vs longer duration type 1 diabetes, and the TN10 trial in which we found that the improvement with treatment was greatest in those in the lower half of C-peptide AUC at entry (13; 18; 19). Of note, high pre-treatment PI:C was strongly predictive of diabetes progression even with adjustment for pre-treatment C-peptide AUC, suggesting that our findings do not reflect impacts of C-peptide alone. Consistent with prior reports (8; 15), age was an important determinant of the PI:C ratio in this Stage 2 population. However, although our sample size is insufficient to analyze the effects between age categories, age in isolation did not significantly impact treatment response (13) and analyses here were adjusted for age.

Our exploratory analysis did not find consistent drug effects on the PI:C ratio in the overall study population over time. Moreover, there was variability in the PI:C response between individuals as well as heterogeneity in the clinical responses to teplizumab. It is possible that, in the context of Stage 2 type 1 diabetes, elevated PI:C ratios may not reflect acute measures of beta cell stressors (such as clinical measures like C-reactive protein) but rather reflect enduring damage to beta cells that is not immediately reversible with immunotherapy and may lead to lasting dysfunction, especially in the context of increasing insulin secretion. In this regard, the anti-TNF mAb golimumab that attenuated beta cell decline in Stage 3 type 1 diabetes did not reduce PI:C ratios, but rather prevented the increase that occurred in placebo-treated participants (20). Therefore, it remains to be established whether a reduction in the PI:C ratio can be used as a marker of improved beta cell stress with interventions that modulate type 1 diabetes progression, particularly those that directly target beta cell health.

The current analysis is the first to examine PI:C values specifically in a group of individuals with Stage 2 type 1 diabetes. Our findings suggest that elevations in PI:C are strongly associated with progression to diabetes in individuals of all ages meeting stage 2 criteria. Thus PI:C could provide useful information to counsel at-risk patients considering immunotherapies on expected rate of development of Stage 3 type 1 diabetes. Consistent with prior reported increases in proinsulin area in islets from autoantibody positive individuals (21), in individuals with Stage 2 disease from this study, a high PI:C ratio was linked to increases in absolute proinsulin rather than decreased C-peptide values. In contrast, elevations of PI:C at the time of type 1 diabetes diagnosis are more strongly linked to reductions in C-peptide (22). One explanation for this interesting finding could be that at earlier stages on the spectrum of type 1 diabetes development, increases in PI:C ratio are more a reflection of intrinsic beta cell stress due to autoimmune attack, and less determined by impacts of chronic hyperglycemia and secretory granule depletion. In this regard, the ratio may better define the risk for progression in general, but also the “window” of treatment opportunity with the drug i.e.those with the highest ratio showed the greatest drug effect.

There are limitations to this exploratory analysis. The original trial was not powered to formally address this post-hoc analysis. Therefore there may be significant differences that are not identified because of the small sample size. As an example, the treatment curves in the individuals with low PI:C ratio do ultimately diverge but the differences in rates of progression may not have achieved statistical significance. Thus, rather than true treatment “responders”, high PI:C may identify a window for a more immediate and clear benefit of drug treatment. We were missing data from 5 of the original trial participants but we addressed this limitation using imputation and the results remained remarkably consistent. For example, a multivariable model for continuous pre-treatment PI:C ratio (log2 transformed) in the model for time to Stage 3 T1D has a corresponding hazard ratio of 2.53 and a p-value of 0.00037 when missing subjects are excluded vs. 2.47 vs. a p-value of 0.0003 when they are included (both models adjusting for baseline mean AUC C-peptide, treatment arm, and age). Because of sample availability, pre-treatment PI:C values were randomly timed; however for participants with fasting and random pre-treatment PI:C, these values were highly correlated. Our designation of low and high baseline ratios corresponded to an approximation of the mean pre-treatment levels and was validated with recursive partioning, but with a larger sample size, a more conventional approach to division could be utilized. Longitudinal metabolic data were impacted by study participant dropout from OGTT testing due to type 1 diabetes. This disproportionately impacted the placebo group, likely resulting in a progressively increasing bias towards more metabolically “normal” PI:C values in the placebo group over time. We found variability in the measurements of the PI:C ratio which may reflect true biologic variance but the timing of the sampling may have contributed. Finally, our analysis of changes in immune cells and the PI:C ratio are correlative and we do not have direct evidence linking T cell frequencies and beta cell stress. Despite these limitations, given the unique nature and value of this dataset in type 1 diabetes prevention, these results provide valuable hypothesis generating data to be prospectively tested as part of future prevention efforts with disease-modifying therapies.

Our findings have important implications for diabetes prediction as well as diabetes prevention. For individuals with Stage 2 type 1 diabetes, an elevated PI:C is highly indicative for impending progression to Stage 3. Drug treatment delays the progression and our data would suggest that teplizumab is most effective if individuals have high PI:C ratios. Thus the PI:C ratio may be used as a biomarker to optimize selection of study participants or timing of dosing for type 1 diabetes prevention trials.

Research in Context Summary.

What is already known about this subject?

1. Elevated relative levels of proinsulin are found in patients with Stage 3 Type 1 diabetes and in prediabetes.

2. Teplizumab delays the progression from Stage 2 to 3 T1D and has received regulatory approval. It also attenuates the loss of C-peptide in patients with Stage 3 T1D.

3. Teplizumab modulates T cells but effects on beta cells are unknown

What is the key question?

• We hypothesized that proinsulin:C-peptide ratios (PI:C), a readout of beta-cell stress, would provide insight into Stage 2 T1D progression and response to teplizumab.

What are the new findings?

• Individuals with Stage 2 type 1 diabetes and high-pre-treatment PI:C displayed higher absolute proinsulin values and rapid progression to Stage 3 disease.

• Patients with high pre-treatment PI:C ratios show a greater response to teplizumab compared to those with a low pre-treatment ratio.

• The drug treatment does not change the PI:C ratio in spite of the delay in time to clinical diabetes, and the effects of drug on relevant immune cells are similar in those with high and low PI:C ratios.

Impact on clinical practice in the foreseeable future?

• Readouts of active disease such as PI:C ratio could identify optimal candidates or timing for type 1 diabetes disease-modifying therapies.

Data availability statement:

Data are available upon reasonable request to the authors or via submission request via www.trialnet.org.