Serological markers of exocrine pancreatic function are differentially informative for distinguishing individuals progressing to type 1 diabetes

Abstract

Introduction: Altered serum levels of growth hormones, adipokines, and exocrine pancreas enzymes have been individually linked with type 1 diabetes (T1D). We collectively evaluated seven such biomarkers, combined with islet autoantibodies (AAb) and genetic risk score (GRS2), for their utility in predicting AAb/T1D status.

Research design and methods: Cross-sectional serum samples (n=154 T1D, n=56 1AAb+, n=77 ≥2AAb+, n=256 AAb-) were assessed for IGF1, IGF2, adiponectin, leptin, amylase, lipase, and trypsinogen (n=543, age range 2.7-30.0 years) using random forest modeling.

Results: GRS2, age, lipase, trypsinogen, and AAb against ZnT8, GAD65, and insulin were the most informative markers. Notably, these variables were differentially informative according to AAb/T1D status. Higher GRS2 (p<0.001) and lower lipase levels (p=0.002) favored ≥2AAb+ versus AAb- classification. AAb against ZnT8 (p<0.01), GAD65 (p=0.021), or insulin (p=0.01) each independently favored ≥2AAb+ versus 1AAb+ classification. Reduced trypsinogen (p<0.001) and increased lipase levels (p<0.001) favored recent-onset T1D versus ≥2AAb+ classification.

Conclusions: Among the serological markers tested, lipase and trypsinogen levels were the most informative for differentiating among clinical groups, with the utility of each enzyme varying according to GRS2 and AAb/T1D status. These data support exocrine pancreas enzymes as candidates for longitudinal follow-up.

Keywords: Autoantibodies; Diabetes Mellitus, Type 1; Enzymes; Models, Statistical.

Figure 1. AAb− and 1AAb+ subjects cannot be robustly differentiated according to serological markers or GRS2. The following variables were tested for their ability to distinguish AAb− (n=256) from 1AAb+ study participants (n=56): age, sex, race, ethnicity, fasted state, BMI percentile (BMI), IGF1 percentile (IGF1), IGF2, trypsinogen, lipase, amylase, adiponectin to leptin ratio (ALR), and GRS2. (a) Variables are plotted according to their mean decrease in accuracy (MDA) and Gini index values. (b) Mean minimal depth and its distribution are shown for variables that were included in ≥50% of decision trees. (c) Variables that were included in ≥50% of decision trees were used to fit a logistic regression model. Data are represented as log odds (dots)±95% CIs (error bars) for each variable. BMI, body mass index; GRS2, genetic risk score; IGF, insulin-like growth factor.

Figure 2. Age, lipase level, and GRS2 are most informative for differentiating AAb− versus ≥2AAb+ subjects. The following variables were tested for their ability to distinguish AAb− (n=256) from ≥2AAb+ study participants (n=77): age, sex, race, ethnicity, fasted state, BMI percentile (BMI), IGF1 percentile (IGF1), IGF2, trypsinogen, lipase, amylase, adiponectin to leptin ratio (ALR), and GRS2. (a) Variables are plotted according to their mean decrease in accuracy (MDA) and Gini index values. (b) Mean minimal depth and its distribution are shown for variables that were included in ≥50% of decision trees. (c) Variables that were included in ≥50% of decision trees were used to fit a logistic regression model. Data are represented as log odds (dots)±95% CIs (error bars) for each variable. (d) Study participant status (ie, AAb− vs ≥2AAb+) is shown according to age (years) and lipase levels, with each data point representing one individual. Lipase=12.2 U/L (horizontal dashed line) and age=18 years (vertical dashed line) were most commonly jointly chosen as optimal split values by the random forests. AAb, autoantibody; BMI, body mass index; GRS2, genetic risk score; IGF, insulin-like growth factor.

Figure 3. Most informative markers for classifying AAb− versus ≥2AAb+ study participants vary according to GRS2. (a) Light red and teal histograms represent the distribution of GRS2 for AAb− (n=256) and ≥2AAb+ (n=77) subjects, respectively. The vertical dashed line represents the cut-off threshold of 11.9 for GRS2-low versus GRS2-high designation. The bar graph depicts the frequency of GRS2 values used to split decision tree nodes. (b–e) The following variables were tested for their ability to distinguish AAb− from ≥2AAb+ study participants: age, sex, race, ethnicity, fasted state, BMI percentile, IGF1 percentile, IGF2, trypsinogen, lipase, amylase, adiponectin to leptin ratio (ALR), and GRS2. Among GRS2-low (b) and GRS2-high (c) subsets, variables were tested for their capacity to distinguish AAb− (n=146 (b); n=77 (c)) from ≥2AAb+ (n=17 (b); n=56 (c)) subjects and were plotted according to their mean decrease in accuracy (MDA) and Gini index values. The probability of ≥2AAb+ classification according to GRS2 and lipase levels is shown among GRS2-low (d) and GRS2-high (e) subsets. AAb, autoantibody; BMI, body mass index; GRS2, genetic risk score.

Figure 4. Age, GRS2, and ZnT8A, IAA, and GADA status are most informative for differentiating 1AAb+ versus ≥2AAb+ subjects. (a) AAb specificities among 1AAb+ study participants (n=56) are shown. (b) The combinations of AAb specificities among ≥2AAb+ subjects (n=77) are shown. The inset plot shows the number of subjects who were positive for each AAb. (c–d) The following variables were tested for their ability to distinguish 1AAb+ from ≥2AAb+ study participants: age, sex, race, ethnicity, fasted state, BMI percentile, IGF1 percentile, IGF2, trypsinogen, lipase, amylase, adiponectin to leptin ratio, GRS2, and status of (ie, positive or negative for) GADA, IAA, IA-2A and ZnT8A. (c) Variables were plotted according to their mean decrease in accuracy (MDA) and Gini index values. (d) Mean minimal depth and its distribution are shown for variables that were included in ≥50% of decision trees. (e) Subjects were divided into four groups by AAb frequency (ie, the number of AAbs for which one was positive; red=1AAb+, green=2AAb+, blue=3AAb+, purple=4AAb+) and were further plotted according to lipase and trypsinogen levels. AAb, autoantibody; BMI, body mass index; GADA, AAbs against GAD65; GRS2, genetic risk score; IAA, insulin AAbs; IGF, insulin-like growth factor.

Figure 5. Trypsinogen and lipase levels are most informative for differentiating ≥2AAb+ study participants from individuals with type 1 diabetes (T1D). The following variables were tested for their ability to distinguish 2AAb+ subjects from those with T1D (duration <1 year): age, sex, race, ethnicity, fasted state, IGF2, trypsinogen, lipase, amylase, GRS2, and status of GADA, IAA, IA-2A, and ZnT8A. (a) Variables were plotted according to their mean decrease in accuracy (MDA) and Gini index values. (b) Mean minimal depth and its distribution are shown for variables that were included in ≥50% of decision trees. (c) Variables that were included in ≥50% of decision trees were used to fit a logistic regression model. Data are represented as log odds (dots)±95% CIs (error bars) for each variable. GADA, AAbs against GAD65; GRS2, genetic risk score; IAA, insulin AAbs; IGF, insulin-like growth factor.