C-Peptide Decline in Type 1 Diabetes Has Two Phases: An Initial Exponential Fall and a Subsequent Stable Phase

Type 1 diabetes is a chronic disease characterized by autoimmune destruction of the β-cells in the pancreas. Traditionally, the autoimmunity has been considered as an ongoing destructive process, ultimately leading to absolute insulin deficiency. However, recent studies have challenged this view by revealing that 29–80% of individuals having type 1 diabetes with over 5 years duration still produce some C-peptide (1–5). Importantly, this is responsive to meal stimulation (1), suggesting that at least some of the residual β-cells are functional. These findings are consistent with histological studies of the pancreas in which residual insulin-containing islets have been found in patients with long-standing type 1 diabetes (6–8). The presence of both C-peptide and β-cells in patients with long-standing type 1 diabetes suggests an attenuation in the rate of β-cell loss over time.

Studying the longer-term trajectory of β-cell decline will be a key step to understanding the preservation of C-peptide secretion in type 1 diabetes. Many studies have examined early C-peptide loss, and these have revealed a rapid and continuing decline in the first 5 years after diagnosis (9–14). However, very little attention has been paid to the progression of C-peptide loss in longer durations of type 1 diabetes. For example, it is not known whether the rate of C-peptide loss slows or stabilizes, and if so, whether this is dependent on the duration of diabetes or the age of the patient.

Therefore, we aimed to examine the trajectory of C-peptide levels measured in a large cohort of patients up to 40 years after type 1 diabetes was diagnosed.

We used both cross-sectional and longitudinal data sets to explore the trajectory of C-peptide levels over time in patients with type 1 diabetes. Characteristics of the patients in these cohorts are in Supplementary Table 1.

Initial analysis examined the association between C-peptide and duration of diabetes in a cross-sectional cohort of 1,549 individuals with type 1 diabetes. Patients were recruited from two discrete geographic regions in the U.K. as part of the Using pharmacogeNetics to Improve Treatment in Early-onset Diabetes (UNITED) study that aimed to recruit all patients who had been diagnosed with diabetes at ≤30 years of age in these regions (15,16). For our study, we only examined patients with a clinical diagnosis of type 1 diabetes who were treated with insulin from the time of diagnosis, and to rule out type 2 diabetes, patients were excluded if they had a BMI >30 kg/m² (or were in >80th percentile if <22 years of age at the time of recruitment) unless they were positive for GAD or IA2 autoantibodies (2). As part of the UNITED study, all patients with a urinary C-peptide/creatinine ratio (UCPCR) >0.2 nmol/mmol and negative islet antibody results were tested for 35 known monogenic diabetes subtypes (15,16). Any patients with an identified monogenic cause for their diabetes were excluded from this analysis. All patients had a duration of diabetes of ≤40 years.

Subjects had their endogenous insulin secretion tested by a postmeal UCPCR. This test has been validated against a formal assessment of C-peptide in a mixed-meal tolerance test and shows a very high correlation with the stimulated C-peptide (r = 0.91) (17). UCPCR results below the limit of detection were coded at 0.00072 nmol/mmol (which is the limit of detection for the urinary C-peptide assay [0.03 nmol/L] divided by the maximum urine creatinine level seen in the study [41.6 mmol/L]).

We analyzed changes in C-peptide levels over time using repeat samples from individuals to verify our findings in the cross-sectional data. The patients were recruited from two different population cohorts, both from a single geographic region (Exeter, U.K.), and met the same inclusion and exclusion criteria for type 1 diabetes as the cross-sectional cohort.

A subset of patients who had UCPCR measured as part of the UNITED study (described above) or a UCPCR validation study (17) had repeat postmeal UCPCR samples taken a median of 4.3 years later (interquartile range [IQR] 3.6, 5.1) (n = 221 patients in total, two repeat measurements except for three individuals with three measurements).

Repeat random nonfasting plasma C-peptide measurements were available on 105 patients with type 1 diabetes recruited to the Diabetes Alliance for Research in England (DARE) study. These patients consented for C-peptide to be measured at the same time as routine HbA_1c testing using EDTA plasma. This enabled regular monitoring without specific research visits. C-peptide is stable for 24 h at room temperature on EDTA plasma (18), and a random nonfasting C-peptide level has been shown to be highly correlated to a 90-min C-peptide level in a mixed-meal tolerance test (r = 0.91) (19). All patients with at least three repeat measurements were included in the analysis. Five hundred twenty-nine C-peptide results were available from the 105 patients, with a median of six results available per patient, over a median time of 2.1 years (IQR 1.3, 2.2).

The studies were approved by the National Research Ethics Service Committee South West (Exeter and Bristol). All patients gave signed informed consent.

Urinary C-peptide and plasma C-peptide levels were measured by electrochemiluminescence immunoassay (intra-assay coefficient of variation 3.3%; interassay coefficient of variation 4.5%) on a Roche Diagnostics (Mannheim, Germany) E170 Analyzer by the Blood Sciences Department at the Royal Devon and Exeter NHS Foundation Trust, Exeter, U.K. The assay is a two-site immunoassay using monoclonal antibodies against human C-peptide, calibrated to World Health Organization (WHO) International Reference Reagent for C-peptide of humans. Urinary creatinine was analyzed on the Roche P800 platform using creatinine Jaffé reagent (standardized against isotope dilution mass spectrometry) to calculate the UCPCR (in nmol/mmol).

GAD and IA2 antibodies were measured as part of the UNITED study in those who had UCPCRs >0.2 nmol/mmol (15), and as part of the longitudinal plasma C-peptide studies. GAD and IA2 antibody analyses were performed using commercial ELISA assays (RSR Ltd., Cardiff, U.K.) and a Dynex DSX automated ELISA system (Launch Diagnostics, Longfield, U.K.). The laboratory participates in the Diabetes Autoantibody Standardization Program. Patients were considered positive for antibodies if their results were >97.5th percentile (11 WHO units/mL for GAD, 15 WHO units/mL for IA2).

All C-peptide results, in both plasma and urine, were natural log transformed for analysis, in line with previous studies (2,10,20), as the distribution of their values was heavily skewed.

Initial analysis of cross-sectional data used nonlinear regression modeling to examine the association between duration and log UCPCR. Generalized additive models were used to explore the initial shape of the association. This revealed a pattern consistent with two phases that could be modeled with two lines of best fit. Segmented regression was used to determine the optimal breakpoints where the lines of best fit would meet and to enable calculation of the intercept and slope of the two different phases, thereby modeling the starting point and rate of C-peptide decline.

To determine whether the association was similar in both patients who had received a diagnosis in childhood and in those who received a diagnosis in their teenage years/young adulthood, the data set was split by the median age at diagnosis and the analysis repeated in each group.

For the longitudinal analysis, data were split into two groups for the two phases: before and after the optimal breakpoints identified from cross-sectional analysis. The intercept and slopes of the two different phases were determined using mixed-effects models to model C-peptide against duration, with random effects at the patient level to allow each patient to contribute multiple C-peptide values at different time points. We used a random-intercept, random-slope model to allow for variability between individuals in terms of both C-peptide level at diagnosis (the intercept) and in the percentage change in C-peptide over time (the slope).

We repeated the analysis excluding those patients whose first value was below the lower limit of detection of the assay to ensure that the finding did not represent a floor effect (i.e., that the results were not an artifact of those below the lower limit of the assay not being able to fall). Model assumptions were tested by examining the normality of residuals and by plotting associations between residuals and fitted values and the duration of diabetes.

The intercepts were back transformed (using the exponential) to estimate C-peptide levels at diagnosis from the models. Because slopes were on a log scale, they were interpreted in terms of the percentage change per year (calculated from the exponential of the β-coefficient −1). The half-life of C-peptide was calculated from log_e(0.5)/β. The variability of individual slopes in the longitudinal models was determined using the SD range (calculated by back transforming the β-coefficient ± 1 SD of the slopes).

All analysis was carried out in R version 3.3.2, including the mgcv package (for GAM models), the lme4 package (for mixed-effects models), and the segmented package (for segmented regression).

Generalized additive modeling of cross-sectional data was used to explore the initial shape of the association and revealed a nonlinear association between log UCPCR and disease duration (Fig. 1A). This is suggestive of two phases: an initial log-linear (exponential) decline followed by a more stable period where the association flattens out. Characteristics of the 1,549 individuals in the cross-sectional cohort are shown in Supplementary Table 1.

To model the slopes of these two phases, segmented regression was used. Figure 1B shows the two fitted slopes, and the corresponding summary statistics, including the estimated UCPCR as modeled at diagnosis and at the breakpoint, are presented in Table 1. The optimal breakpoint (i.e., the point at which the slope changes) was modeled at 6.9 years from initial diagnosis (95% CI 6.3, 7.5). Over this period, UCPCR declined by 47%/year (95% CI 43, 51) (P < 0.0001), equivalent to a half-life of 1.10 years (95% CI 0.99, 1.25). Beyond 6.9 years, the slope was flat, suggesting a more stable period with no further decline (+0.07%/year [95% CI −1.3, +1.5], P = 0.8).

Figure 2 shows the patterns of association between disease duration and UCPCR when splitting the data by the median age at diagnosis (10.8 years). In both “age at diagnosis” groups, the pattern of β-cell decline was similar, showing an initial exponential fall followed by a more stable period. There was no significant difference in the slope of the first phase of decline in those who had received a diagnosis of diabetes at ≤10.8 years of age compared with those diagnosed at >10.8 years of age (42% vs. 49% decrease/year, P = 0.13) (Table 2). The association between UCPCR and duration was much flatter in the second phase in both groups (Table 1). Although the initial slopes were similar in each age-group, the absolute UCPCR value was higher across all time points in those who were older at diagnosis: the intercept was higher (indicating the UCPCR at diagnosis was higher) (1.32 vs. 0.27 nmol/mmol, P < 0.0001), as well as the modeled UCPCR at the breakpoint (i.e., the level for the start of the second more stable phase) (0.022 vs. 0.005 nmol/mmol, P < 0.0001). Based on the initial UCPCR and half-life estimated from the slopes, we can calculate that for patients receiving a diagnosis of diabetes at ≤10.8 years of age, it would take 0.6 years to reach the clinically important threshold of absolute insulin deficiency (0.2 nmol/mmol [equivalent to 200 pmol/L]) (21) compared with 2.7 years in the older group (diagnosed at >10.8 years of age).

To validate the existence of two distinct phases, separate models were analyzed using longitudinal data obtained either in the first 7 years (individual patient data rounded to the nearest year) or after 7 years, in line with the estimated inflection point from cross-sectional data. Of 221 patients with repeat UCPCR results, 41 had both initial and repeat results within 7 years of diagnosis and 121 had both initial and repeat results beyond 7 years after diagnosis.

The patterns were similar to those seen in the cross-sectional data, with an initial exponential fall in UCPCR during the first 7 years (48% decrease/year [SD range for variability of individual slopes −67 to −18], half-life 1.1 years) (Table 2) and a more stable phase showing no further decline after that time (0.1% decrease/year [SD range −1.8 to +1.7]) (Table 2). When excluding those patients whose first result was below the lower limit of the assay (to examine only those whose C-peptide levels could fall), there was a slight decline in the second phase, but this was far slower than that seen in the first 7 years (−4.5%/year [SD range −9.0 to +0.3]), half-life 15 years) (Table 2).

In the 26 patients who had a UCPCR >0.2 nmol/mmol (significant endogenous insulin) (21,22) beyond 7 years after diagnosis, there was no decrease in slope on repeat sampling (−0.7%/year [95% CI −4, +3], P = 0.7). Fifteen of twenty-six of these patients were positive for either GAD or IA2 autoantibodies, and given the high positive predictive value of islet autoantibodies in this age-group and our strict inclusion criteria, this reinforces the conclusion that these individuals have type 1 diabetes despite their high C-peptide levels.

Similar patterns were seen in the long-duration patients when assessing the longitudinal plasma C-peptide data. Of the 105 patients with repeat plasma C-peptide results, only 5 had repeat samples in the initial 7 years after diagnosis, so analysis of the first phase was not carried out. Data were available from 93 patients who had initial and all repeat samples beyond 7 years duration, and, again, there was no decline in slope over this second phase (−2.6% decrease/year, P = 0.2 [SD range −12.6 to +8.5]) (Table 2). Results were similar when excluding those whose first measurement was below the lower limit of the assay (Table 2).

We have shown, using both cross-sectional data and longitudinal data, that there are two phases of C-peptide decline in the first 40 years after the diagnosis of type 1 diabetes. These comprise an initial exponential fall over the first 7 years, followed by a more stable period, where C-peptide levels either completely plateau or decline much more slowly.

The decline in C-peptide levels over the first few years after diagnosis has been studied in detail in a number of other studies (10–14,20). The rate of decline we show of ∼47%/year up to 7 years is similar to that reported previously (10,20). Some studies (11,13) have not used log-transformed data for analysis, but, despite this, the patterns reported are consistent with an exponential fall.

To our knowledge, this is the first study examining the continuous pattern of C-peptide concentrations over time in long-duration type 1 diabetes. Analysis of the T1D Exchange cohort investigated the prevalence of detectable C-peptide and found a decrease with increasing duration, but as the outcome was categorical this did not fully capture the changing association (4). The longest previous longitudinal study we have been able to identify modeled C-peptide over the first 7.4 years after diagnosis but used older, less sensitive C-peptide assays, so was unable to evaluate the pattern of decline at lower levels (23). Our data suggest that there is a major change at around this point, with a dramatic decline in C-peptide secretion (half-life of ∼1 year) over the first 7 years after diagnosis, followed by a relatively stable period beyond 7 years, where C-peptide levels either remain fairly constant or decline much more slowly (half-life estimated at 15 years from longitudinal plasma C-peptide data). This change is consistent with data showing HbA_1c “tracks” over time stabilizing after ∼6 years (24).

In our study, the absolute values of C-peptide differed according to age at diagnosis, with younger patients having lower levels on average, but the rate of decline and disease duration before the second stable phase was similar in individuals who received a diagnosis below or above 10.8 years of age (the median age at diagnosis). The finding that younger patients have lower levels of C-peptide at diagnosis (and throughout the disease process) is well established (4,10,13,20) and fits nicely with studies of the pancreas that show fewer residual insulin-containing islets in patients having younger ages at disease onset (8). The similarity in the rate of decline between those patients who received a diagnosis in childhood and in teenage years has also been seen previously (4,10,11). By contrast, the studies that have suggested that the rate of C-peptide decline is faster in patients who received a diagnosis at younger ages have used other outcomes, such as time to a given low C-peptide threshold, to judge the rate and, as such, are not directly comparable with the present data (9,12). Nevertheless, given that we find an exponential loss, and that younger patients with lower C-peptide levels at diagnosis would reach a low threshold more quickly, these previous results are not inconsistent with our data.

The finding of two phases suggests a change in the underlying biological processes leading to β-cell demise at ∼7 years of disease duration. The fact that the pattern and inflection points were similar in those who received a diagnosis at both younger and older ages suggests that this is a feature of disease progression, rather than being determined by the chronological age of the patient. This means that it is more likely to be a manifestation of changes occurring in the disease process in the pancreas rather than differences in puberty or in the maturity of the pancreas.

The nature of the biological changes that result in the stabilization of C-peptide are not revealed by our study, so we see our findings as largely hypothesis generating. The stability of C-peptide after ∼7 years could reflect either the persistence of a less susceptible population of β-cells that remains after the period of exponential decay or a change in the immune-mediated attack at this time. Recent work (25) has described subpopulations of β-cells, including a putative “hub” cell population within islets, and it may be that certain cells are able to escape the immune destruction that affects all other β-cells. A change in the immune response is another possibility, and there is evidence that the immune attack may subside over time, given the recent finding that HLA class I hyperexpression on the residual insulin-containing islets of individuals with type 1 diabetes (which is prominent at diagnosis) declines with disease duration (26). However, altered immune responses could also reflect changes in antigen expression and/or antigen presentation.

Follow-up prospective studies involving repeated C-peptide measurements before and after the 7-year inflection point in larger numbers of people would be of considerable value. These would allow the timing of the change in the rate of C-peptide decline to be examined in individual subjects and combined with simultaneous immune studies. However, the considerable intraindividual variation in both C-peptide estimation and immune cell populations would mean that large numbers of people must be studied.

Understanding the mechanisms that mediate the change in C-peptide decline occurring at the inflection point will not only help to elucidate the underlying biological mechanisms of β-cell destruction over time in type 1 diabetes, but may also lead to improved strategies for β-cell preservation. If the level of C-peptide attained in any given person at 7 years after diagnosis is sustained, then this would also have implications for future intervention trials. The majority of trials currently focus on preserving β-cell function close to the diagnosis of type 1 diabetes, but our new finding offers the potential for therapeutic trials to be undertaken later in the disease process.

Our study has strengths and weaknesses. The major strengths are 1) the large numbers people studied (>1,500); 2) that we combined both cross-sectional and longitudinal studies, measuring both urine and plasma C-peptide, which show consistent support for both the stabilization of C-peptide levels at 7 years and the rate of exponential deterioration before that; and 3) that we examined a large range of disease durations (up to 40 years). The major weaknesses are that the initial analysis was based on cross-sectional data, and the longitudinal studies were based on repeat samples collected over a relatively short period (2–4 years) with the number of measurements limited to two for most individuals. Therefore, larger, longer, and more frequently sampled longitudinal replication would have considerable value, particularly prospective studies crossing the 7-year time point. Without a longer follow-up time, we cannot determine the extent to which our results reflect the pattern of C-peptide loss in all patients. However, given that the second phase, as modeled, is flat, this pattern would not occur if some patients were still declining at this point, without an equivalent number increasing their C-peptide levels to balance this. The fact that the second-phase slope still remains relatively flat even when removing those individuals whose measured C-peptide level is below the limit of the assay suggests that this is not an artifact caused by the inclusion of people with unrecordable values. Moreover, we used strict inclusion criteria to ensure that potential cases of type 2 or monogenic diabetes were excluded. Given the rarity of other causes in those individuals who received a diagnosis at the youngest ages and the high proportion of positive islet autoantibodies in those with high C-peptide levels, we feel that it is unlikely that the individuals studied had a form of diabetes other than type 1, and that any potential misclassification will be minimal. It should also be emphasized that we have used home postmeal UCPCR and random nonfasting plasma C-peptide results rather than results from a gold standard mixed-meal tolerance test for the present analysis. However, both of these C-peptide measurements have been validated against the mixed-meal tolerance test and shown to be highly correlated (17,19). Although, the measurements we used are potentially more prone to noise, we have used large sample sizes, and, importantly, the results were remarkably consistent in both plasma and urine. Finally, it is important to note that this study was carried out on predominantly white cohorts. Further work is needed to determine whether the pattern is generalizable to other racial groups.

In conclusion, we have shown that there are two phases of C-peptide decline in individuals with type 1 diabetes. The stabilization of C-peptide levels at ∼7 years after diagnosis suggests that there are important and previously unrecognized changes in immune function and/or β-cell viability around this time that may have critical implications for future pharmaceutical interventions.

Acknowledgments. The authors thank Bart Roep and Tim Tree of the TIGI steering group for their useful contributions to discussion around the manuscript and Rachel Besser and Angus Jones of the University of Exeter for the collection of samples for the UCPCR validation cohort used in the longitudinal analysis.

Funding. This work was principally supported by JDRF (grant 3-SRA-2014-314-M-R), with additional funding from the Department of Health and a Wellcome Trust Health Innovation Challenge Award (HICF-1009-041, WT-091985). B.M.S. is a core member of the National Institute for Health Research (NIHR) Exeter Clinical Research Facility. T.J.M. is funded by an NIHR Clinical Senior Lecturer Fellowship. R.O. holds a Diabetes UK Harry Keen Fellowship award. E.R.P. holds a Wellcome Trust New Investigator award (102820/Z/13/Z). A.T.H. is a NIHR Senior Investigator and Wellcome Trust Senior Investigator. The authors acknowledge further funding from the JDRF Career Development award to S.J.R. (5-CDA-2014-221-A-N) and from Diabetes UK (project grant 16/0005480) to S.J.R. and N.G.M. This article presents independent research supported by JDRF, Diabetes UK, the Wellcome Trust, and the NIHR Exeter Clinical Research Facility.

The views expressed are those of the authors and not necessarily those of JDRF, Diabetes UK, the Wellcome Trust, the National Health Service, the National Institute for Health Research, or the Department of Health.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. B.M.S. carried out all statistical analysis, conceived the statistical aspects of study design, and wrote the manuscript. T.J.M. led the biochemical analysis, designed the serum follow-up study, contributed to the study design and discussion, and reviewed/edited the manuscript. R.O. is Principal Investigator for the TIGI study (repeat urine C-peptide results), contributed to the study design and discussion, and reviewed/edited the manuscript. A.H. is data manager for the longitudinal C-peptide studies and reviewed/edited the manuscript. M.H. is project manager for the UNITED study and reviewed/edited the manuscript. P.L. contributed to discussion and reviewed/edited the manuscript. E.R.P. was joint lead for the UNITED study and reviewed/edited the manuscript. S.J.R. contributed to discussion and reviewed/edited the manuscript. N.G.M. contributed to discussion and reviewed/edited the manuscript. A.T.H. conceived of the idea, is joint lead author for the UNITED study and the Chief Investigator for the Diabetes Alliance for Research in England (DARE) study, contributed to the study design and discussion, and reviewed/edited the manuscript. B.M.S. and A.T.H. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the Diabetes UK Annual Professional Conference 2017, Manchester, U.K., 8–10 March 2017, and at the 2017 JDRF nPOD 9th Annual Scientific Meeting, Fort Lauderdale, FL, 19–22 February 2017.