This was a cross-sectional study conducted at the Copenhagen CF Center. The inclusion criteria were age ≥18 years and a confirmed CF diagnosis. Except from individuals with lung-transplantation and pregnant women, all adult patients without CFRD and patients with CFRD, who had participated in prior research at our centre were invited to participate.

Participants were fasted ≥8 hours before the OGTT and individuals with insulin therapy were requested to withhold short-acting insulin for 12 hours and long-acting insulin for 24 hours before the test. All participants drank 75 grams of glucose diluted in water. Depending on the personal preference, the participant underwent an extended 3-hour OGTT or a standard 2-hour OGTT. In the extended OGTT, a cannula for blood collection was inserted in an antecubital vein and blood was drawn 10 and 1 min prior to and 10, 20, 30, 45, 60, 90, 120, 150, and 180 min after glucose ingestion. In the standard OGTT, blood samples were collected 1 min prior and 30, 60, and 120 min after the glucose ingestion. Samples were stored on ice until centrifugation. Glucose, C-peptide, insulin (serum tubes) as well as intact proinsulin (EDTA tubes) were centrifuged at room temperature. Glucagon (BD™ P800^® tubes) was centrifuged at 4°C. After sample collection, centrifugation occurred within 1 hour and the samples were frozen (-80°C) immediately after. Glucose (702 module of Cobas 8000), C-peptide (801 module of Cobas 8000), insulin (801 module of Cobas 8000), proinsulin (TECOmedical Intact Proinsulin ELISA with Proinsulin Dynex DS2 ELISA analyser) and glucagon (Mercodia Glucagon ELISA with Glucagon Dynex DS2 ELISA analyser) were analysed with quantitative assays. All biochemistry analyses were performed all together at Exeter Clinical Laboratory (accredited by the United Kingdom Accreditation Service).

In line with previous protocols (11), participants were divided into groups with increasing severity of glucose intolerance; normal glucose tolerance (NGT): 1-hour glucose <8.6 mmol/L and 2-hour glucose <7.8 mmol/L, early glucose intolerance (EGI): 1-hour glucose ≥8.6 mmol/L and 2-hour glucose <7.8 mmol/L, impaired glucose tolerance (IGT): 2-hour glucose between 7.8-11.0 mmol/L and CFRD: 2-hour glucose ≥11.1 mmol/L. Homeostasis model assessment (HOMA-IR) and Matsuda index were used as markers of insulin resistance/sensitivity (17). HOMA-IR was calculated as (Insulin_fasting (μU/mL)*Glucose_fasting (mmol/L))/22.5 and Matsuda index as 10,000/(√[(Glucose_fasting (mg/dL)*Insulin_fasting (μU/mL))*(Glucose_mean (mg/dL)*Insulin_mean (μU/mL))]) using values from the standard OGTT. Among participants with NGT or EGI, hypoglycaemia and non-hypoglycaemia sub-groups were determined based on 3-hour glucose <3.9 mmol/L or ≥3.9 mmol/L, respectively. In addition, the hypoglycaemia threshold was reduced to 3.0 mmol/L in a sensitivity analysis (18). IGT and CFRD were excluded a priori in the hypoglycaemia analysis, given low incidence of post-OGTT low glucose in these groups. Demographic data and health status were obtained from the Danish CF Registry (age, sex, mutation class, faecal elastase, weight, height, lung function, modulator treatment and years on insulin treatment). Pancreas insufficiency was defined as faecal elastase ≤200 µg/g.

Insulin secretion rate and proinsulin appearance were estimated in WinSAAM version 3.3.0 (19). Insulin secretion rate was assessed using a modified minimal model by Breda et al. (20). In contrast to the insulin secretion model, which is unaffected by hepatic extraction, a proinsulin secretion model is not justified, as part of the proinsulin is extracted by the liver. Instead, we estimated the proinsulin appearance, which is a measure of the secreted proinsulin that passes the liver. Nevertheless, only a modest proportion of the proinsulin secretion is presumed to be extracted by the liver, as no difference in proinsulin levels in the hepatic and the systemic circulation could be detected in humans (21) and the hepatic extraction fraction for proinsulin was found to be constant and low (10-15 times lower than the hepatic extraction of insulin) (22). Hence, we assumed that most of the proinsulin secretion will pass the liver. Proinsulin was assumed to have a one compartment, first order elimination (23) with a half-life of 19 min (24), which is consistent with human studies (23). With these assumptions, proinsulin appearance could be estimated by deconvolution of proinsulin concentrations in plasma and a known one compartmental model. As proinsulin was only sampled at -1, 30, 60, 120, 180 min, we used interpolation with cubic splines to eliminate artificial spikes in the proinsulin appearance. The ratio of the area under the curve (AUC) of insulin secretion rate and proinsulin appearance was calculated for different phases of the OGTT (-10- (–1) min, 0-60 min, 60-120 min, and 120-180 min). The ratio was calculated for both AUCs above zero (total) and above baseline (incremental). As a surrogate measure for glucagon secretion, we used the glucagon concentrations with cubic spline interpolation, as glucagon has a short half-life (4.5 min) independent of glucose tolerance in non-CF individuals (25) and a stable hepatic extraction fraction around 20% (26). The glucagon secretion was divided into four phases ([fasting: -10, -1 min], [very early: 10 min], [intermediary: 20, 30, 60, 90, 120 min], and [late: 150, 180 min]). Only data from participants with an extended OGTT were included in the insulin secretion model, the proinsulin appearance model, the AUCs calculations and the glucagon model. To approximate maximum and minimum secretion in all participants, we generated estimates of the maximum concentrations (C_Max) of glucose, C-peptide, and proinsulin as well as the nadir (C_Min) of glucagon, respectively, with cubic spline interpolation between the four concentrations measures in the standard OGTT (-1, 30, 60, and 120 min).

Baseline characteristics of the participants were summarized by medians and interquartile range (IQR) or proportions (%). A Chi-squared test was used to test whether females and males had different glucose tolerance distributions in our sample. Glucose concentrations, insulin secretion rate and proinsulin appearance during the OGTT were presented as mean and 95% confidence interval (95% CI) and the glucagon concentrations were plotted as median (IQR), as data were highly skewed. The relationship between proinsulin appearance and insulin secretion rate (time intervals between maximum levels (ΔT_Max) and ratios of AUC), and time to nadir (T_Min) of glucagon were compared in linear regression models across glucose tolerance groups. We assessed the correlation between glucose or insulin secretion rate and glucagon with Spearman’s correlation. Associations between glucose or insulin secretion rate and glucagon during four OGTT phases were assessed in linear mixed models with and without effect modification of glucose tolerance group and with random intercepts for each participant. Linear regression models were used to assess the associations between C_Max of glucose, C-peptide, and proinsulin vs. the logarithmic value of C_Min of glucagon. This model was further adjusted for age, sex, body mass index and exocrine pancreas function (insufficient/sufficient). The prevalence of hypoglycaemia in those with NGT and EGI was compared in a Chi-squared test. Glucose and different pancreas hormone concentrations (logarithmically transformed) in individuals with hypoglycaemia were compared to individuals without hypoglycaemia in linear mixed models. The linear mixed model accounted for individual variation with random intercepts and was also adjusted for age, sex, body mass index and exocrine pancreas function (insufficient/sufficient). All models, including the mixed models, used robust standard errors (27). Robust standard errors reference to the estimation procedure, which provides robust estimates of the variance including standard errors. This was used to correct for small departures of the outcome from normality. Beta coefficients from models with logarithmic transformed outcomes were back transformed into percentage changes. The estimates of C_Max of glucose, C-peptide, and proinsulin and C_Min of glucagon were validated against the modelled estimates from the extended OGTTs with Spearman’s correlation. P-values <0.05 (two-tailed) were considered significant and R x64 (version 4.2.2) and RStudio (version 1.2.5001) were used for the statistical analyses.

Ethical approval (H-19085530) was obtained from the Regional Committee on Health Research Ethics prior to enrolment of participants and all participants signed the consent form before entering the study.

Between August 2020 and January 2021, 93 out of 159 eligible adults with CF participated in the study. Sixty-seven participants (72%) underwent the extended OGTT. Most participants were pancreas insufficient and had severe mutations. Ninety-one (98%) had at least one F508del mutation and 60 (65%) were homozygous. Out of the 93 participants, 21 (23%) presented with NGT, 30 (32%) with EGI, 14 (15%) with IGT and 28 (30%) with CFRD ( Table 1 ). The prevalence of the different glucose tolerance groups was comparable for females and males in our sample (p-value=0.499).

Glucose levels, insulin secretion rate, proinsulin appearance and glucagon concentrations stratified by glucose tolerance during the OGTT are presented in dynamic plots ( Figures 1A–D ). Figure 1 shows that glucose intolerance was characterised by a delayed insulin secretion, but only those with CFRD had a diminished overall insulin secretion ( Figure 1B ). Time to peak for the proinsulin appearance was less affected by glucose tolerance, however proinsulin secretion was highest in those with IGT ( Figure 1C ). The figure also shows comparable peak glucagon concentration over the first 30 minutes post-glucose ingestion ( Figure 1D ), but increasing time to glucagon nadir with worsening glucose tolerance (C_Min) ( Supplementary Figure 1 ).

The time of peak and magnitude of insulin secretion rate and proinsulin appearance were plotted for each glucose tolerance group ( Figures 1E–H ). We assessed the difference in the time interval between maximum insulin secretion rate and proinsulin appearance to indicate whether glucose stimulated proinsulin-to-insulin secretion ratio differed between groups. We found that the time interval between maximum insulin secretion rate and proinsulin appearance was shorter with increasing impairment in glucose tolerance (p-value for trend=0.009) ( Figure 2A ). We also found that the fasting proinsulin appearance/insulin secretion rate ratio (PA/ISR) increased across the glucose tolerance groups from NGT to CFRD (p-value for trend=0.013) ( Figure 2B ). However, during the OGTT the incremental PA/ISR did not vary across glucose tolerance groups ( Figure 2C ). Nevertheless, the incremental PA/ISR was lower during 0-60 min compared to 60-120 min among those with NGT and EGI, but not among those with IGT and CFRD.

NGT and CFRD, but not EGI and IGT, presented with an overall negative correlation between glucose and glucagon. EGI, IGT, and CFRD, but not NGT, had an overall negative correlation between insulin secretion rate and glucagon ( Figures 3A, B ). Further investigation showed that the association between glucose and glucagon and between insulin secretion rate and glucagon were positive at 10 minutes, but negative between 20-180 min ( Figures 3C, D ). The associations did not differ between glucose tolerance groups ( Figures 3E, F ).

Among all participants, the maximum level (C_Max) of glucose and C-peptide were not associated with the minimum level (C_Min) of glucagon. However, the maximum level of proinsulin (C_Max) was positively associated with the minimum level (C_Min) of glucagon ( Table 2 ). The coefficient of determination (R²) between modelled estimates from the extended OGTTs and C_Max and C_Min from the standard OGTTs were between 0.72-0.99 ( Supplementary Figure 2 ).

Among those with either NGT or EGI, 37 (73%) had a 3-hour blood sample. Three (17.6%) out of 17 with NGT had hypoglycaemia, while 12 (60%) out of 20 with EGI had hypoglycaemia. The prevalence of hypoglycaemia differed significantly between NGT and EGI (p=0.023). At 60 minutes, participants with hypoglycaemia had 27.8% higher glucose levels (95% CI [11.3; 46.7%], p<0.001), but similar C-peptide levels, compared to non-hypoglycaemia. At 120 minutes, glucose levels were comparable, but participants with hypoglycaemia had higher C-peptide (42.9%, 95% CI [5.9; 92.8%], p=0.019) and proinsulin levels (32.5%, 95% CI [3.8; 69.2%], p=0.024). Throughout the OGTT, hypoglycaemia was not associated with the total PA/ISR. Glucagon levels were elevated after 180 min in those with hypoglycaemia compared to those without hypoglycaemia (80.3%, 95% CI [14.9; 182.9%], p=0.010) ( Table 3 ). When the threshold for hypoglycaemia was reduced to 3.0 mmol/L, only four were categorized as having hypoglycaemia and they were all in the group with EGI. The estimates were similar with the reduced threshold ( Supplementary Table 1 ).

To elucidate islet pathophysiology underlying the development of CFRD, we studied beta and alpha cell function across the spectrum of glucose tolerance in CF. Our results demonstrate that those with EGI already manifest impaired early-phase insulin secretion, confirming prior investigation (11). We further found increased PA/ISR at fasting in those with CFRD, but during the glucose challenge the ratio was less impacted. Moreover, glucagon secretion was negatively associated with insulin in the late phase of the OGTT, which might explain the delayed suppression of glucagon in those with glucose intolerance. However, neither glucose nor insulin levels predicted the glucagon nadir, while the peak concentration of proinsulin was positively associated with the nadir level of glucagon, suggesting a possible relationship between beta and alpha cell dysfunction in the CF islet. Hypoglycaemia was more common in those with EGI compared to NGT where a delay in early-phase insulin secretion was associated with an increased insulin secretion in the late phase of the OGTT as suggested by a prior report (28).

The proinsulin-to-insulin ratio seems to be a good general indicator of beta cell stress, as previous studies reported increased proinsulin-to-insulin ratios in CF and non-CF subjects with insufficient insulin secretion (9–11, 29, 30). Consistently, we found a reduction in the time interval between the peaks of the insulin secretion rate and proinsulin appearance with severity in glucose intolerance, which suggests that the glucose stimulated proinsulin-to-insulin secretion ratio changes with the progression towards CFRD. In addition, we found an elevated fasting PA/ISR in CFRD, which might indicate changes in constitutive insulin secretion from immature granules (31). However, our results indicate that the difference in PA/ISR between groups was less evident during the OGTT, as the secretion ratio above baseline showed no significant variation between the glucose tolerance groups. Still, these findings support that in addition to absolute beta cell loss, dysfunctional beta cells also play a role in the progression of abnormal glucose tolerance towards diabetes in CF. It is possible that the dysfunctional phenotype is an end-stage consequence of other pancreas pathology, i.e., inflammation or peri-islet fibrosis affecting blood flow and resulting in a delayed response of the beta cells in sensing fluctuations in glucose concentrations. Nevertheless, interventions that could relieve some of the beta cell stress may be of clinical relevance in CFRD.

As non-CF studies have shown that insulin secretion suppresses glucagon secretion through intra-islet signalling (13), we did not expect to find an increased glucagon secretion and positive associations with both glucose levels and insulin secretion after 10 min of the OGTT. Though an early peak of glucagon has been seen following oral glucose challenge in non-CF studies (32–34), we are not aware of other studies evaluating this in CF. Nevertheless, a very early positive association between insulin and glucagon has previously been observed in non-CF individuals with abnormal glucose tolerance (35). A possible explanation is that early aberrant glucagon secretion is explained completely or partly by synthesis and secretion of glucagon by intestinal endocrine cells (36). Indeed, pancreatectomized patients exhibit paradoxically increased glucagon secretion in response to oral glucose (37) and glucagon has insulinotropic effects, similar to the incretin effect of glucagon-like-peptide-1 (38). However, whether enteric-derived glucagon might also support islet beta cell function in CF is still unknown.

A priori, we hypothesized that CFRD would have higher glucagon secretion during the OGTT compared to NGT, in line with the insufficient intra-islet suppression theory and findings from non-CF studies (39) and CF studies (12). Intriguingly, the nadir glucagon levels were not higher in those with low insulin secretion or high glucose levels, but the lowest glucagon levels were higher in those with high proinsulin levels. The elevation in nadir glucagon with increasing proinsulin secretion supports an intra-islet pancreatic defect affecting both beta and alpha cell functional phenotype. A post mortem tissue study in CF pancreases, has shown that peri-islet fibrosis was associated with having an increased number of alpha cells with an altered phenotype (anomalously expressing the mesenchymal marker, vimentin) (40). This suggests that changes in the islet environment may lead to endocrine cell dysfunction affecting both beta and alpha cells. As this explanation is speculative, more research is needed to prove any direct correlation between intrinsic beta and alpha cell impairment in CF.

The gradual progression to CFRD makes it possible to study early islet impairment and related glucose intolerance in stages before CFRD. The current study demonstrated that reactive hypoglycaemia following an oral glucose challenge was linked to early impairment of glucose tolerance. Our study agrees with previous work (15, 28) showing that high levels of glucose at 60 min related to impaired early-phase insulin secretion and consequently high levels of insulin at 120 min were seen in those with hypoglycaemia. Hence, a shift in the dynamics of insulin secretion seems to best explain the occurrence of late post-prandial or oral glucose load hypoglycaemia in CF. Furthermore, we found compensatory increased glucagon secretion in those with hypoglycaemia, in contrast to what has been reported previously (16, 28). Both Aithken et al. (16) and Kilberg et al. (28) showed no change in the 180 min glucagon response after an OGTT when comparing those with and without hypoglycaemia. However, it should be noted that these studies also included participants with IGT and higher insulin secretion from 120-180 min (28), which might have suppressed the glucagon response despite the concurrent hypoglycaemia. In our study, those with EGI had a low insulin secretion in the late phase of the OGTT, which may have allowed a glucagon response. Although we cannot reject that the hypoglycaemic glucagon response was delayed and/or suboptimal in CF, lack of glucagon response does not seem to explain the reactive hypoglycaemia seen in those with NGT and EGI. In summary, this study holds little evidence that an abnormal glucagon response has a clinical impact in CFRD, although the function of the alpha cells seems to be affected by the pathophysiology of CFRD.

A strength of this study is the inclusion of a relatively large sample size of well-characterised individuals with CF who were clinically stable at the time of metabolic testing. Nevertheless, the absolute numbers in each group were small, which increases the risk for type 1 and type 2 errors. The models developed to estimate proinsulin appearance and glucagon secretion independent of other hormones are novel, with no other published models available for comparison. Therefore, we used simple assumptions to derive markers of proinsulin and glucagon secretion, but some errors related to the hepatic extraction and variation in clearance cannot be disregarded completely. This study was also limited by the use of two different study protocols (i.e., standard and extended) according to participant preference. However, all models took this into account by either excluding the standard OGTTs or excluding samples not in the standard OGTT. Finally, this cross-sectional study was not designed to identify causes of CFRD and hypoglycaemia.