High-sensitivity CRP results
Results of hsCRP analysis are summarised in Fig. 1 and detailed in Table 2. In each centre and irrespective of the assay used, hsCRP levels were significantly lower in individuals with HNF1A-MODY than in those with other diabetes aetiologies. The hsCRP levels associated with the respective diabetes subtypes varied between centres. This partly reflects the specific hsCRP assay used (for example the median values of the HNF1A-MODY cases cluster at or near the assay’s lower reporting limit). Meta-analysis, excluding the previously reported UK samples, confirmed that hsCRP levels were lower in HNF1A-MODY than in other diabetes subtypes across the other six centres (z value −20.5, p < 5 × 10−93) . Meta-analysis across all samples concurred with this result (z value −21.8, p < 5 × 10−105).
Clinical characteristics and hsCRP levels
The clinical characteristics of the participants are shown in ESM Table 1. Clinical variables significantly differed between the aetiological groups in line with previous studies . Fasting plasma glucose (r
2 = 0.04), age of diabetes diagnosis (r
2 = 0.08) and BMI (r
2 = 0.23) were significantly correlated with hsCRP levels (p < 0.05). Adjustment for these variables did not affect the magnitude or significance of hsCRP differences observed between the groups; this is important, as any adjustment of the hsCRP value for use as a screening biomarker would be impractical. The effect of statin and/or aspirin therapy on hsCRP levels was examined. In HNF1A-MODY, HNF4A-MODY and young adult-onset type 2 diabetes cases the hsCRP levels were similar between statin and/or aspirin users and non-users (p > 0.05). In comparison, participants with GCK-MODY taking statins and/or aspirin had higher hsCRP levels than non-users (1.78 vs 1.22 mg/l, p = 0.03). Thus, as reported previously, it seems unlikely that the use of these drugs in patients with other forms of diabetes would lower hsCRP towards the levels seen in HNF1A-MODY patients .
High-sensitivity CRP is a clinically valid biomarker for HNF1A-MODY
The differences in hsCRP values between HNF1A-MODY patients and other diabetes subtypes are statistically convincing, but this does not necessarily translate into a clinically valid biomarker. The ROC curve-derived C-statistic (measures discriminative accuracy) was calculated for unadjusted hsCRP levels. The C-statistic to distinguish HNF1A-MODY from other diabetes subtypes ranged from 0.72 to 0.95 across the centres and from 0.76 to 0.86 across the three hsCRP assays (Table 3). The C-statistic for HNF1A-MODY vs young adult-onset type 2 diabetes ranged from 0.79 to 0.97 across the centres and from 0.79 to 0.91 across the three hsCRP assays (Table 3, Fig. 2a). This indicates that hsCRP can reliably distinguish HNF1A-MODY from young adult-onset type 2 diabetes across European populations, even when different hsCRP assays are used. The hsCRP M2 and hsCRP M3 assays have lower discriminative accuracy, which is likely to reflect their less precise lower detection limits (0.25 and 0.16 mg/l, respectively, compared with 0.03 mg/l for hsCRP M1).
In samples from Denmark, Norway, Poland, Slovakia and the UK, a diagnostic threshold of 0.25 mg/l conferred 83% sensitivity and 86% specificity for distinguishing HNF1A-MODY from young adult-onset type 2 diabetes. This is revised to 81% sensitivity and 88% specificity if hsCRP values >10 mg/l are not excluded. Different hsCRP assays, i.e. those used to analyse the French and Finnish/Swedish samples, were associated with different thresholds for optimum sensitivity and specificity; thus a cut-off of 0.5 mg/l conferred 76% sensitivity and 81% specificity in French samples and 74% sensitivity and 68% specificity in Finnish/Swedish samples for distinguishing HNF1A-MODY from young adult-onset type 2 diabetes. It is likely that the higher hsCRP cut-off in the French and Finnish/Swedish samples reflects the different reporting ranges of the hsCRP assays used to assess these samples. The summary ROC curve, which was based on combining all unadjusted hsCRP values ≤10 mg/l irrespective of the assay used, indicated high sensitivity (78%) and specificity (80%) for distinguishing HNF1A-MODY from young adult-onset type 2 diabetes (Fig. 2b).
It is likely that a biomarker such as hsCRP would be most effective as a screening tool for HNF1A-MODY when used in conjunction with clinical features. In our previous report, the proposed strategy for selection of young adult-onset type 2 diabetes cases for genetic testing was either hsCRP ≤0.2 mg/l or fulfilment of traditional criteria for MODY testing (i.e. age of diabetes diagnosis ≤25 years plus two consecutive generations of diabetes) . This resulted in ~80% sensitivity and specificity. In the current dataset, use of combined criteria, i.e. hsCRP ≤0.25 mg/l (hsCRP M1) or hsCRP ≤0.5 mg/l (hsCRP M2/ M3) or age of diagnosis ≤25 years, resulted in improved sensitivity for detection of HNF1A-MODY (~90%) without much loss in specificity compared with the sensitivity and specificity quoted above (ranged from 65% to 81% across assays).
We evaluated hsCRP as a biomarker to differentiate between MODY subtypes. The levels of hsCRP were significantly different between HNF1A-MODY and HNF4A-MODY cases, and between HNF1A-MODY and GCK-MODY cases (p < 5 × 10−4 and p < 0.005, respectively, for all pairwise comparisons across assays). The C-statistic for HNF1A-MODY vs HNF4A-MODY cases ranged from 0.77 to 0.93 across centres, indicating that hsCRP can discriminate well between these MODY subtypes. The C-statistic for HNF1A-MODY vs GCK-MODY cases ranged from 0.70 to 0.79, indicating that hsCRP confers less good discrimination between these MODY subtypes.
Analyses excluded samples measured with hsCRP M2 and M3 due to the insufficiently precise hsCRP values for the majority of HNF1A-MODY individuals tested with these assays. ESM Table 2 shows the median hsCRP values reported according to mutation position. Analysis of mutation position by functional domain indicated that the hsCRP levels associated with mutations affecting the dimerisation/DNA-binding domains were significantly lower than those affecting the transactivation domain (0.03 vs 0.10 mg/l, p = 4.4 × 10−6). The effect of functional domain on hsCRP level was restricted to missense mutations only (0.03 vs 0.16 mg/l, p = 0.001), with no significant correlation noted for truncating mutations (p = 0.83). Analysis by mutation type showed median hsCRP was lower in HNF1A-MODY cases with missense mutations than in those with truncating mutations (0.03 vs 0.08, p = 1.5 × 10−5). The isomer affected had no effect on hsCRP (p = 0.25).
High-sensitivity CRP assay comparison
The comparison of hsCRP assays showed good agreement between hsCRP M1 and hsCRP M2 (r
2 = 0.91, p < 9 × 10−5), with no significant difference between values obtained from the two assays (p = 0.20) (Fig. 3a). Although there was good correlation between hsCRP M1 and hsCRP M3 (r
2 = 0.99, p < 8 × 10−26), there were significant differences between the values obtained via these two assays (p = 0.001). Passing–Bablock analysis indicated a constant bias (intercept 0.32, 95% CI 0.03, 0.15) combined with proportional bias (slope 0.70, 95% CI 0.68, 0.75) between hsCRP M1 and hsCRP M3 (ESM Fig. 1). This means that the hsCRP values derived from hsCRP M1 were generally higher, except for values below 1 mg/l where hsCRP M3 generated markedly higher values (Fig. 3b).
Repeat of previously elevated CRP results
Of the participants who had hsCRP >10 mg/l in the previous study, three HNF1A-MODY patients and 19 of the 42 with young adult-onset type 2 diabetes agreed to a repeat test after a minimum interval of 1 week . ESM Fig. 2 shows the initial and repeat hsCRP results. One HNF1A-MODY patient and ten of 19 young adult-onset type 2 diabetes patients had persistently elevated hsCRP >10 mg/l. The repeat hsCRP values in two of the HNF1A-MODY patients remained well above the proposed threshold of 0.25 mg/l, and therefore these individuals would not have been offered genetic testing on the basis of hsCRP alone. Interestingly, only seven of 457 (1.5%) HNF1A-MODY patients in the current study had an hsCRP value >10 mg/l, which suggests that very few MODY cases would be missed even if those with initially high hsCRP values were not re-tested.