Amino Acids and CART Distinguish A−β+ Ketosis-Prone Diabetes From Type 1 and Type 2 Diabetes During Hyperglycemic Crises

Abstract

Context

When clinically stable, patients with A−β+ ketosis-prone diabetes (KPD) manifest unique markers of amino acid metabolism. Biomarkers differentiating KPD from type 1 (T1D) and type 2 diabetes (T2D) during hyperglycemic crises would accelerate diagnosis and management.

Objective

Compare serum metabolomics of KPD, T1D, and T2D patients during hyperglycemic crises, and utilize classification and regression tree (CART) modeling to distinguish these forms of diabetes.

Methods

At an urban hospital emergency center, adults with KPD, T1D, and T2D during hyperglycemic crises with or without diabetic ketoacidosis (DKA), and healthy controls, underwent comparisons of serum metabolite and hormonal profiles and CART analyses. Group differences in concentrations of amino acids, their metabolites, and relationship to glucose counterregulatory hormones were determined, as well as C-peptide cutoffs and analytes to distinguish KPD, T1D, and T2D.

Results

Concentrations of most amino acids were similar in KPD and T1D and lower compared to T2D (P < .05). Glucagon and cortisol concentrations were correlated with 3-methylhistidine and blood urea nitrogen in KPD but not in T1D. A C-peptide cutoff of 0.496 ng/mL differentiated T1D from KPD during DKA. CART revealed that a regression model based on the concentrations of β-hydroxybutyrate, C-peptide, glucagon, alpha-keto-β-methylvalerate, cystine, and myristoyl-L-carnitine distinguished KPD from T1D and T2D.

Conclusion

During DKA, KPD and T1D patients have similarly altered amino acid profiles that differentiate them from T2D patients. Elevated protein catabolic hormones drive altered amino acid metabolism in KPD, rather than insulin deficiency as with T1D. A combination of 6 analytes differentiates KPD from T1D and T2D during hyperglycemic crises.

Patients with the distinctive A−β+ subtype of ketosis-prone diabetes (KPD), characterized by overweight/obesity, absence of T1D autoantibodies, and substantially preserved beta cell function despite initial presentation with diabetic ketoacidosis (DKA) (1) have evidence of deranged branched-chain amino acid (BCAA) metabolism, as well as altered arginine-citrulline-glutamine metabolism, when they are clinically stable and near-normoglycemic (2–4). These alterations likely contribute to their proclivity to develop unprovoked DKA (3, 4). A−β+ KPD patients have substantial serum insulin and C-peptide concentrations not just when they are clinically stable (1, 5), but, remarkably, even during acute presentation with DKA (2). These findings suggest the hypothesis that enhanced impact of glucose counterregulatory or stress hormones rather than severe hypo-insulinemia may be critical for changes in amino acid and energy metabolism in KPD patients during DKA.

Currently, it is difficult to diagnose KPD patients accurately at the time of their initial presentation with DKA. Diagnostic confusion occurs frequently in the acute, emergent setting because they have clinical features that resemble both type 1 diabetes (T1D) (occurrence of DKA) and type 2 diabetes (T2D) (older age of onset, overweight/obese habitus, markers of metabolic syndrome) (1, 2). A diagnosis of KPD is usually established during outpatient follow-up several weeks to months later, based on absence of T1D-associated islet autoantibodies, presence of beta cell functional reserve, and characteristics of the post-hospitalization clinical course such as the ability to discontinue insulin therapy while maintaining near-normoglycemia (1, 6). Biomarkers that could unequivocally distinguish KPD from T1D and T2D at initial presentation with acute hyperglycemic crises would accelerate accurate diagnosis and targeted management.

A goal of the present study was to determine whether the serum amino acid profile of KPD patients during DKA differs from that of T1D patients during DKA and of T2D patients during acute hyperglycemic crisis, and whether the serum amino acid concentrations of the 3 groups vary in relation to key hormones that regulate protein, lipid, and glucose metabolism. A second goal was to determine if quantitative measurements of specific analytes related to these pathways could describe a decision tree to accurately categorize the patients as KPD, T1D, or T2D at the time of their presentation in acute metabolic crisis. To achieve these goals, we measured the concentrations of all amino acids, 3-methyl histidine (3-MH), ammonia, urea, C-peptide and several protein anabolic (insulin, growth hormone) and catabolic (cortisol, epinephrine, norepinephrine, glucagon) hormones in blood samples obtained from adult patients at the time of presentation with hyperglycemic crisis, both with and without DKA, in the emergency center of Ben Taub General Hospital, Houston, Texas. To distinguish the 3 forms of diabetes, a classification and regression tree (CART) algorithm was used to model specific metabolites and hormones including serum β-hydroxybutyrate (BOHB) and acetyl- and acyl-carnitines derived from BCAAs and fatty acid metabolism. This paper is the second report of a larger study of serum metabolomics in adults with different forms of diabetes (2).

Methods

Study Participants

The study was approved by the Institutional Review Boards for Human Studies at Baylor College of Medicine and the Harris County Hospital District. Adult patients, sequentially admitted from 2015 to 2018 to the Ben Taub Hospital Emergency Center were recruited if they presented with acute hyperglycemic crisis, defined as an admission plasma glucose concentration ≥250 mg/dL with evidence of clinical decompensation. DKA was diagnosed if the patient also had serum BOHB >1.5 mmol/L, anion gap >12, venous pH ≤7.3 and serum total CO₂ ≤ 15 mEq/L.

Nondiabetic obese controls (N = 17) were recruited through advertisement. Selection criteria included adults aged 18 to 65 years, with fasting plasma glucose <100 mg/dL, glycated hemoglobin (HbA1c) < 5.6%, body mass index (BMI) >25 kg/m², in good health, with no evidence of chronic or acute illness by medical history, physical examination, complete blood count, comprehensive metabolic panel and liver function tests.

Sample Collection and Laboratory Analyses

Potential study participants were approached after diagnosis of acute hyperglycemic crisis in the emergency center, and blood samples were drawn immediately into prechilled tubes from those who provided informed consent. Blood samples from healthy controls were drawn into prechilled tubes containing heparin. Samples were immediately centrifuged at 4 °C and the serum/plasma removed and stored at −70 °C.

Serum insulin (Roche Cat# 12017547, RRID: AB_2756877), C-peptide (Roche Cat# 03184897, RRID: AB_2909476) and cortisol (Roche Cat# 06687733 160, RRID: AB_2802131) concentrations were measured by immunoassay (Cobas e411 analyzer, Roche Diagnostics, Indianapolis, IN, USA) and growth hormone (R and D Systems Cat# DGH00, RRID: AB_2923238), glucagon (R and D Systems Cat# DGCG0, RRID: AB_2893019), epinephrine and norepinephrine (Abnova Cat# KA1877, RRID: AB_3674439) by ELISA (R&D, Minneapolis, MN, and Abnova, Taipei, Taiwan). Serum glucose concentration was measured by the glucose oxidase method (YSI 2300 glucose analyzer, YSI, Yellow Springs, OH, USA). Serum concentrations of amino acids were measured by ultra-high-performance liquid chromatography (Waters Corporation, Milford, MA) using pre-column derivatization with 6-amino-quinolyl-N-hydroxysuccinimidyl carbamate.

Antibodies against the 65 kDa glutamic acid decarboxylase (GAD65Ab) were measured in participant serum samples as previously described (7). Briefly, recombinant GAD65 was labeled with [³⁵S]-methionine (GE Healthcare Life Sciences, Amersham, U.K.) by in vitro-coupled transcription and translation in the TNT SP6 coupled reticulocyte lysate system (Promega, Southampton, UK). Full-length cDNA coding for human GAD65 in the pTNT vector (Promega) (pThGAD65) was used. GAD65Ab was analyzed in a radioligand binding assay in the serum samples incubated with ³⁵S-labeled GAD65 in a final reaction volume of 60 µL. The samples were transferred to filtration plates (Millipore, Solna, Sweden) and free ³⁵S-labeled GAD65 separated from antibody-bound with PAS (Zymed Laboratories, San Francisco, CA). The radioactivity of antibody-bound ³⁵S-labeled GAD65 was counted in a Wallac Microbeta Trilux (PerkinElmer) β counter. GAD65Ab levels were expressed as units per milliliter derived from the World Health Organization standard 97/550. Samples were considered positive if GAD65Ab levels were >50 U/mL The intra-assay coefficient of variation for duplicates in the GAD65Ab assay was 7%.

Statistical Analysis

Group comparisons of serum amino acid and hormonal profiles

For all normally distributed data, between-group comparisons were made using one-way ANOVA with the Tukey post hoc test. For data that were not normally distributed, nonparametric one-way ANOVA (Kruskal-Wallis test) was performed with post hoc Dunn's multiple comparison test. Patient demographics and characteristics were summarized by mean with standard error and compared between the diabetes groups (KPD, T1D, T2D) by Kruskal-Wallis test and Chi-square test. Multivariable linear regression was used to test if diabetes groups were significantly associated with clinical characteristics and metabolites and hormones after adjusting for age, gender, BMI, and new-onset diabetes. Associations between serum hormones with blood urea nitrogen (BUN) and 3-MH concentrations were assessed in the KPD and T1D groups using Spearman correlation. P < .05 was considered significant for all outcome measures. All analyses were performed with GraphPad Prism version 9 (GraphPad Software, San Diego, CA, USA).

CART and receiver operating characteristic curve analysis

CART with a complexity parameter of 0.01 was used to determine how the diabetes groups were classified based on the characteristics of interest. CART is a machine learning method that can identify which factors are important in a model or relationship in terms of explanatory power and variance. In this analysis, all demographics and biomarkers were included in the model, but only those considered important based on explanatory power were retained in the final parsimonious tree. The CART analysis was performed using the R package “rpart” (8).

Receiver operating characteristic (ROC) area under the curve (AUC) analysis was conducted for C-peptide concentrations to distinguish KPD patients from T1D patients with 95% CI. The optimal C-peptide cutoff value was determined using Youden's index with sensitivity and specificity. The analysis was done in Rstudio (8).

Results

Group Comparisons of Serum Amino Acid and Hormonal Profiles

Patient characteristics and hormonal profiles

A total of 102 patients presenting with acute hyperglycemic crises to the emergency center were recruited and provided informed consent. Among these patients, 74 were diagnosed with DKA and 28 did not have DKA. Of the 74 DKA patients, 53 subsequently met criteria for A−β+ KPD based on the absence of GAD65Ab, as well as significantly higher mean C-peptide, BMI, and age compared to those with DKA who were GAD65Ab-positive. Twenty-one DKA patients were GAD65Ab-positive and were denoted as having T1D. Of the 28 hyperglycemic patients without DKA, 21 were GAD65Ab-negative and denoted T2D. The remaining 7 hyperglycemic patients without DKA were GAD65Ab-positive and removed from the analysis because of the small sample size.

Demographic, anthropometric, and biochemical characteristics and hormonal profiles of the entire study cohort have been reported previously (2) and are presented in a supplementary appendix with permission from the journal. KPD patients had significantly higher BMI and age than the T1D patients (Supplementary Table S1) (9). Fifteen of the patients with DKA (12 KPD, 3 T1D) also had new-onset diabetes. Overweight/obesity was present in 100% of controls, 91% of T2D, 55% of KPD, and 33% of T1D patients. The patients with T2D were significantly older than KPD, T1D, and control participants. There were more male than female participants, and more African American and Hispanic than White participants, in all groups.

During treatment in the emergency center, 17 of the T2D patients, 38 of the KPD patients and 13 of the T1D patients received a bolus injection of insulin with or without an insulin infusion before a blood sample could be drawn for the study (median delay = 113 minutes). Except for serum lactate concentration, there was no statistically significant difference in any biochemical characteristic between patients who received or did not receive insulin before the blood sample was drawn. Glucose and lactate concentrations were significantly higher in all diabetes groups compared with controls and both were significantly lower in KPD compared with T2D. Venous pH and total CO₂ were significantly lower, while BOHB and anion gap were significantly higher, in the KPD and T1D groups compared with T2D. There was no difference in serum creatinine between the groups. Hemoglobin was significantly lower in T2D compared with KPD, albumin was significantly lower in T2D and KPD compared with controls, and triglycerides were significantly higher in T2D compared with controls.

Hormonal profiles of the participants are presented in Supplementary Table S2 (9). Of the 2 protein anabolic hormones measured, insulin concentration (in patients who did not receive therapeutic insulin) was significantly lower in KPD and T1D but not in T2D patients compared with controls. Insulin concentration of the T1D group (but not that of the KPD group) was also significantly lower than the corresponding value of the T2D group. Serum C-peptide concentrations of KPD patients were significantly higher than those of T1D patients. Only 13 of the 53 KPD patients (25%) had serum C-peptide concentrations below the normal range. There was no difference in mean C-peptide concentration between controls and T2D patients regardless of whether they received insulin therapy, and C-peptide concentrations in these 2 groups were higher than those of both KPD and T1D patients, regardless of whether the latter received insulin therapy. Serum growth hormone (GH) concentrations were significantly higher in all 3 diabetes groups compared with controls. It was also significantly higher in all T1D patients compared with the corresponding values of the T2D group.

Of the protein catabolic hormones, serum glucagon concentration was significantly higher in all T2D patients (including those who received insulin) compared to controls. There was high variability in individual glucagon concentrations among KPD and T1D patients, hence the values were not significantly different from that of the controls. Compared with control values, serum cortisol, epinephrine, and norepinephrine concentrations were significantly higher in both KPD and T1D but not in T2D patients.

Serum amino acid concentrations

Because there was no statistical difference in any amino acid measurement based on whether the patients within a group received insulin or not prior to blood sampling, their data were combined into a single group as presented in Table 1. Among dietary essential amino acids, serum threonine and tryptophan concentrations were significantly lower in each of the 3 groups of diabetes patients compared to the control values (P < .05 for all comparisons); tryptophan concentrations were also significantly lower in KPD and T1D patients compared with T2D patients (P = .006 and 0.04, respectively). Serum histidine concentration was significantly lower in the T2D group compared to controls and the KPD group (P = .002 for both comparisons), and there was no difference among groups in lysine, methionine, and phenylalanine concentrations. Also, as we reported previously, concentrations of the BCAAs were significantly higher in KPD and T1D patients compared with controls and T2D (2).

Table 1.

Serum concentrations of amino acids of nondiabetic controls, type 2 diabetes patients with hyperglycemia, ketosis-prone diabetes patients with DKA, and type 1 diabetes patients (T1D) with DKA^a

Amino acids (µmol/L)	Control	T2D with hyperglycemia	KPD with DKA	T1D with DKA
N	17	21	53	21
Dietary essential
Threonine	132 ± 8.28	72.7 ± 5.69b	86.1 ± 5.00b	73.9 ± 6.00b
Tryptophan	52.3 ± 2.15	39.9 ± 2.73b	30.1 ± 0.937b,c	30.0 ± 1.64b,c
Histidine	74.7 ± 2.52	53.0 ± 2.53b	72.4 ± 4.51c	67.5 ± 4.10
Lysine	161 ± 10.2	168 ± 9.78	152 ± 6.94	143 ± 13.0
Methionine	24.4 ± 0.819	22.7 ± 1.67	24.5 ± 1.40	21.7 ± 2.15
Phenylalanine	56.4 ± 2.15	66.4 ± 5.21	61.8 ± 2.52	56.8 ± 2.69
Dietary nonessential
Glutamine	445 ± 12.1	613 ± 18.7b	594 ± 23.0b	626 ± 35.1b
Glutamate	56.6 ± 4.35	45.2 ± 5.09	22.8 ± 1.53b,c	19.5 ± 1.85b,c
Asparagine	57.7 ± 5.72	43.2 ± 2.70	49.7 ± 2.95	43.6 ± 2.90
Aspartate	2.87 ± 0.223	8.95 ± 1.65b	7.38 ± 0.865b	7.19 ± 1.16b
Arginine	85.2 ± 3.25	94.0 ± 5.64	84.5 ± 3.52	76.0 ± 4.38
Ornithine	41.4 ± 2.26	42.3 ± 3.83	36.8 ± 2.54	32.0 ± 4.03
Citrulline	29.9 ± 0.857	24.7 ± 2.29	22.4 ± 1.22b	27.6 ± 2.85
Proline	182 ± 12.1	211 ± 13.7	153 ± 8.26c	144 ± 15.2c
Alanine	307 ± 13.7	425 ± 29.6	257 ± 13.7c	238 ± 22.9c
Glycine	201 ± 7.13	165 ± 10.9	203 ± 11.5	189 ± 18.6
Serine	91.4 ± 3.93	79.9 ± 5.56	80.0 ± 3.77	69.8 ± 5.19b
Tyrosine	75.5 ± 3.77	58.9 ± 5.18b	54.6 ± 2.26b	53.4 ± 3.41b
Cystine	50.6 ± 2.11	65.2 ± 4.39	50.5 ± 3.45c	46.5 ± 4.69c
AA metabolites
3-methyl-histidine	6.14 ± 0.326	5.17 ± 0.853	3.67 ± 0.582b,c	3.58 ± 0.555b
NH3	88.1 ± 3.40	60.1 ± 2.14b	67.8 ± 2.67b	64.0 ± 2.68b
BUN (mmol/L)	4.24 ± 0.210	7.69 ± 1.38b	8.14 ± 1.03b	8.18 ± 0.808b

Abbreviations: AA, amino acid; BUN, blood urea nitrogen; DKA, diabetic ketoacidosis; KPD, ketosis-prone diabetes; T1D, type 1 diabetes; T2D, type 2 diabetes.

^a Data are expressed as means ± SE. Because data were not normally distributed, nonparametric one-way ANOVA (Kruskal-Wallis test) was performed with post hoc Dunn's multiple comparison.

^b Value significantly different from control value, (P < .05).

^c Value significantly different from T2D group value (P < .05).

Among dietary nonessential amino acids, serum alanine concentrations of both the KPD and T1D groups were significantly lower than the corresponding value of the T2D group (P < .0001 for both comparisons) but not compared to the control value. There was no difference between groups in serum glycine concentrations, but serum serine concentration was significantly lower in the T1D group compared to the control value (P = .008). Serum glutamine concentrations were significantly higher in all 3 diabetes groups than the control group (P < .001 for all comparisons), and while there was no difference among groups in asparagine concentrations, aspartate concentrations were significantly higher in all 3 diabetes groups than the control group (P < .01 for all comparisons). There were no differences among the groups in serum arginine and ornithine concentrations, but compared to the control value, serum citrulline concentration was significantly lower in the KPD group (P = .004). Serum proline concentrations of both KPD and T1D groups were significantly lower than the corresponding value of the T2D group (P = .002 and .005, respectively) but not compared to the control value. Serum tyrosine concentrations were significantly lower in all 3 diabetes groups than the control group (P < .01 for all comparisons) and cystine concentrations of the KPD and T1D groups were significantly lower compared with the T2D group (P = .003 and .03, respectively).

Compared to the control value, serum ammonia concentrations were lower (P < .001 for all comparisons) whereas BUN concentrations were higher (P < .05 for all comparisons) in each diabetes group compared with the control group. Finally, as previously reported (2), serum glutamate concentrations were lower in KPD and T1D patients compared with controls (P < .0001 for both comparisons) and T2D (P < .001 for both comparisons).

Correlations between C-peptide, GH, and the protein catabolic hormones (cortisol, epinephrine, norepinephrine, glucagon) with 3-MH and BUN are presented for the T1D and KPD groups in Table 2 and in Figs. 1 and 2. In the T1D group there were no significant correlations between any hormone with 3-MH and BUN. However, the KPD group showed significant positive correlations of the catabolic hormones glucagon and cortisol with 3-MH and BUN, a significant negative correlation of the anabolic hormone GH with 3-MH, and a trend toward a significant positive correlation (P = .08) of the catabolic hormone norepinephrine with BUN.

Table 2.

Correlations of serum hormones with 3-methyl-histidine and blood urea nitrogen concentrations in ketosis-prone diabetes patients and type 1 diabetes patients during DKA

Variable 1	Variable 2	Spearman r	P value (2-tailed)
T1D patients
C-peptide	3-methyl-histidine	−0.2485	.2774
	BUN	−0.2595	.2693
Growth hormone	3-methyl-histidine	−0.05564	.8158
	BUN	−0.01318	.9573
Glucagon	3-methyl-histidine	0.002597	.9911
	BUN	−0.1236	.6035
Cortisol	3-methyl-histidine	−0.07532	.7456
	BUN	0.05428	.8202
Epinephrine	3-methyl-histidine	−0.2584	.2580
	BUN	−0.03995	.8672
Norepinephrine	3-methyl-histidine	0.1623	.4820
	BUN	−0.05277	.8251
KPD patients
C-peptide	3-methyl-histidine	0.05506	.6954
	BUN	−0.01537	.9148
Growth hormone	3-methyl-histidine	−0.2982	.0318
	BUN	−0.2301	.1079
Glucagon	3-methyl-histidine	0.4232	.0016
	BUN	0.2980	.0337
Cortisol	3-methyl-histidine	0.2811	.0415
	BUN	0.4986	.0002
Epinephrine	3-methyl-histidine	0.1852	.1887
	BUN	0.1448	.3158
Norepinephrine	3-methyl-histidine	0.06762	.2498
	BUN	0.2955	.0802

All P values <.05 are bolded.

Abbreviations: BUN, blood urea nitrogen; DKA, diabetic ketoacidosis; KPD, ketosis-prone diabetes; T1D, type 1 diabetes.

Figure 1.

Spearman correlations between serum concentrations of glucagon and 3-methylhistidine (3-MH) (A, B) and blood urea nitrogen (BUN) (C, D) in ketosis-prone diabetes patients (KPD) and type 1 diabetes patients (T1D) during DKA.

Figure 2.

Spearman correlations between serum concentrations of cortisol and 3-methylhistidine (3-MH) (A, B) and blood urea nitrogen (BUN) (C, D) in ketosis-prone diabetes patients (KPD) and type 1 diabetes patients (T1D) during DKA.

ROC Curve and CART Analyses

ROC curve analysis showed that the optimal cutoff for C-peptide to distinguish T1D from KPD was 0.496 ng/mL, with area under the ROC curve of 77.54% (95% CI, 64.95%-90.13%). This cutoff has a sensitivity of 0.75 and specificity of 0.81, with Youden's index = 0.56 (Fig. 3).

Figure 3.

Receiver operating characteristic curve for C-peptide levels in KPD compared with T1D patients, in serum samples taken at the time of acute presentation with DKA. The horizontal dashed line indicates the sensitivity (0.76) and the vertical dashed line indicates 1 − specificity (0.19) of the optimal cutoff value for C-peptide by Youden's index. See text for details.

Combinations of serum concentrations of 6 metabolites—BOHB (a ketone body), α-keto-β-methylvalerate (KMV, a catalytic product of the BCAA isoleucine), cystine, C-peptide, glucagon and myristoyl-L-carnitine (C14, a long-chain acylcarnitine)—were significantly associated with classification of the patients as KPD, T1D, or T2D in the CART model (Fig. 4). With BOHB <1.775 mmol/L, patients were more likely to be classified as T2D (95% T2D, 5% T1D). With BOHB ≥1.775 mmol/L and KMV ≥38.32 µmol/L, patients were more likely to be classified as T1D (100% T1D). With BOHB ≥1.775 mmol/L, KMV <38.32 µmol/L, and cystine <21.13 µmol/L, patients were more likely to be classified as T1D (100% T1D). With BOHB ≥1775 mmol/L, KMV <38.32 µmol/L, cystine ≥21.13 µmol/L, C-peptide <0.4935 ng/mL, glucagon <74.67 pg/mL, and C14 ≥ 0.04 µmol/L, patients were more likely to be classified as T1D (100% T1D). With BOHB ≥1.775 mmol/L, KMV <38.32 µmol/L, cystine ≥21.13 µmol/L, C-peptide <0.4935 ng/mL, glucagon <74.67 pg/mL, and C14 < 0.04 µmol/L, patients were more likely to be classified as KPD (86% KPD, 14% T1D). With BOHB ≥1.775 mmol/L, KMV <38.32 µmol/L, cystine ≥21.13 µmol/L, C-peptide <0.4935 ng/mL, and glucagon ≥74.67 pg/mL, patients were more likely to be classified as KPD (100% KPD). With BOHB ≥1.775 mmol/L, KMV <38.32 µmol/L, cystine ≥21.13 µmol/L, and C-peptide ≥0.4935 ng/mL, patients were more likely to be classified as KPD (96% KPD, 2% T1D, 2% T2D).

Figure 4.

Classification and regression tree for diagnosis of KPD, T1D, and T2D at presentation with hyperglycemic crisis. (Numbers at the bottom of the colored boxes at the termini of the branches represent, from left to right, numbers of patients with KPD, T1D, and T2D respectively in that box.) See text for details.

Discussion

These results indicate that during an acute episode of DKA, the serum concentrations of amino acids are similar in patients subsequently classified as T1D or KPD. When compared to the control and T2D groups, their amino acid profiles show similar differences with only 3 exceptions: histidine is lower in T2D compared with KPD but not with T1D; compared to the control group, serine is lower in T1D but not in KPD; and citrulline is lower in KPD but not in T1D. However, the relationships of serum concentrations of protein catabolic hormones with 3-MH and BUN differ significantly between KPD and T1D, suggesting that the protein and amino acid metabolic alterations in KPD patients during DKA are primarily mediated by a heightened effect of the catabolic hormones rather than severe insulin deficiency.

Based on previous research in T1D patients, it was proposed that marked insulin deficiency in the face of elevated glucagon is the primary underlying mediator of deranged protein and amino acid metabolism during DKA (10–13). However, we showed in a previous analysis of the present cohort that KPD patients have substantial beta cell functional reserve during DKA (Supplementary Table S2) (9), suggesting that severe insulin deficiency was not the critical mediator of their altered amino acid metabolism (2). Although insulin concentration in patients (who did not receive therapeutic insulin) was significantly lower in both KPD and T1D compared with controls, it was 222% higher in KPD compared to T1D. The mean serum C-peptide concentration of KPD patients was significantly higher than that of T1D patients and 75% of the KPD patients had serum C-peptide concentrations within the normal range (Supplementary Table S2) (9). Our current findings of significant associations of glucagon and cortisol with 3-MH and BUN (markers of skeletal muscle protein breakdown (14, 15) and whole body protein catabolism, respectively) in the KPD group but not in the T1D group suggest that increased effect of these 2 catabolic hormones, and likely of norepinephrine as well, may be the predominant contributor to the changes in protein and amino acid metabolism in KPD patients.

While all the hormones we measured have regulatory effects on the rates of protein synthesis and breakdown, there is a complex interaction between them in catabolic states such as severe injury, sepsis, and DKA. In good health, insulin and GH are the 2 primary anabolic hormones maintaining whole body protein homeostasis by both inhibiting protein breakdown and stimulating protein synthesis (16). In the insulin-deficient state of untreated T1D, most studies have reported increased breakdown and oxidation of proteins (17, 18). In the current study, the lack of an association between insulin (or C-peptide) with 3-MH and BUN during DKA is likely due to its extremely low and variable concentrations in the T1D patients. GH exerts its anabolic effect both directly and indirectly via IGF-I and insulin (which are both increased after GH administration) (16). GH suppression of protein breakdown is mediated through insulin's anti-proteolytic effects (19). This property is lost in T1D patients during DKA because of extreme hypo-insulinemia, but not in the KPD patients, supported by our observation of a negative association between C-peptide and 3-MH in the KPD patients but not in the T1D patients.

Whereas our findings regarding the serum concentrations of dietary essential amino acids in both T1D and KPD groups were consistent with previous reports in T1D patients during DKA, this was not the case for the dietary nonessential amino acids. All of the nonessential amino acids are glucogenic, with alanine, glutamine, and to a lesser extent proline, serine, and glycine being the principal contributors to gluconeogenesis (20–22). The lower concentrations of alanine, proline, tyrosine, and cystine in both DKA groups compared to either the hyperglycemic T2D group or the control group suggest that increased use of these amino acids to fuel accelerated gluconeogenesis during DKA outstrips their rates of production. Similarly, the lower serine concentration in the T1D group suggests that its increased use for gluconeogenesis and possibly glycine synthesis is greater than its rate of production. Also, because synthesis of alanine and serine requires glutamate as the nitrogen donor and proline is synthesized de novo from glutamate, the lower serum glutamate observed in both groups of DKA patients (2) suggests decreased availability to sustain higher rates of synthesis of these amino acids.

A role for glucagon as a mobilizer of amino acids in conditions characterized by elevated glucagon levels has been debated, since the reverse effect can also occur physiologically, that is, amino acids can stimulate the release of glucagon. Using simultaneous infusions of somatostatin and glucagon compared with normal saline infusions in healthy humans, James et al found that glucagon in the setting of insulin deficiency increased exchange of both glucogenic and essential amino acids across the muscle and splanchnic beds. This effect of glucagon appeared to be due to decreased skeletal muscle protein synthesis rather than increased muscle or splanchnic protein degradation (23). Glucagon in excess can also exert a direct catabolic effect, as shown in patients with glucagonoma syndrome where elevated glucagon secreted by the tumor accelerates proteolysis and protein oxidation (24, 25).

A clinically useful outcome of this comparative metabolomics analysis is the identification of biomarkers that singly and in combination differentiate KPD from T1D and T2D during acute hyperglycemic crises. An elevated level of BOHB, the predominant ketone body of DKA, with a quantitative cutoff of 1.775 mmol/L, clearly differentiated both KPD and T1D patients from T2D patients. Serum C-peptide levels, even in the acute setting of DKA at the time of presentation in the emergency department, can distinguish T1D (lower) from KPD (higher) (2), and ROC curve analysis determined a highly predictive C-peptide cutoff of 0.496 ng/mL. CART analysis revealed a combination of analytes that accurately differentiates the 3 forms of diabetes, deriving cutoffs for BOHB, C-peptide, glucagon, KMV, cystine, and C14. Utilization of these measurements at the time of acute presentation in hyperglycemic crises, or even of a smaller combination of analytes such as BOHB, C-peptide, and glucagon that are readily available through clinical laboratories, could distinguish the 3 forms of diabetes. This would permit early and accurate diagnosis of KPD (which is usually delayed for 2 to 6 months following the index episode of DKA on the basis of subsequent clinical behavior) and thereby accelerate appropriate targeted management which is quite different for each form of diabetes (1, 6, 26).

Measurement of T1D-associated islet autoantibodies (together with clinical features such as age of diabetes onset, BMI and acanthosis nigricans) can also distinguish T1D from KPD patients at the time of acute presentation with DKA, since these autoantibodies are absent in A−β+ KPD patients (1). However, islet autoantibody measurements are not rapidly available in the urgent setting of an emergency center visit. Furthermore, reliance on this marker alone could result in both false-positive and false-negative diagnoses, as it is estimated that approximately 15% of T1D patients may be autoantibody-negative (27); conversely, a separate subgroup of KPD patients (A+β+ KPD) resemble A−β+ KPD patients clinically but are autoantibody-positive and have a different natural history (1, 28). In the past, the authors have demonstrated accuracy and predictive value in the diagnosis, subclassification and management of KPD patients by measuring serum C-peptide levels (fasting basal and glucagon-stimulated) together with T1D-associated islet autoantibodies; however, that experience is based on measurements made 2 to 6 months following the index episode of DKA, when the patients are clinically stable (1, 6, 29).

A limitation of these conclusions is that they are based on static metabolomics measurements. In previous studies of clinically stable KPD patients, we investigated aberrant pathways inferred from plasma metabolomics analysis using kinetic measurements of metabolite fluxes to validate and strengthen the mechanistic implications (3, 4). However, it was not possible to conduct complex isotope infusion studies in acutely ill participants upon presentation at the emergency department, especially prior to treating them with exogenous insulin. Another limitation is the sample size which was relatively small (albeit the product of 2 years of challenging recruitment of acutely ill patients in a busy emergency center). As a result, cross-validation of the conclusions of the CART analysis within the existing dataset could not be performed, as the size of each of the groups, if the data were split, would be too small to be informative. Hence, future investigations should include validation of the CART results in a separate dataset.

In summary, KPD patients experiencing DKA have an altered amino acid profile that resembles that of T1D patients. However, the different pattern of significant relationships between the serum concentrations of their catabolic hormones with those of markers of muscle and whole body protein breakdown (3-MH and BUN) suggests that increased effects of protein catabolic hormones rather than severe insulin deficiency are likely to mediate the changes in protein/amino acid metabolism in KPD patients during DKA. Serum BOHB and C-peptide levels obtained at presentation with hyperglycemic crisis can effectively differentiate KPD, T1D, and T2D patients, but a decision tree that includes a hierarchy of 6 metabolites—BOHB, KMV, cystine, C-peptide, glucagon, and C14—may refine with high accuracy the distinction between the 3 forms of diabetes during acute hyperglycemic crises.

Abbreviations

3-MH

3-methyl histidine

BCAA

branched chain amino acids

BMI

body mass index

BOHB

β-hydroxybutyrate

BUN

blood urea nitrogen

C14

myristoyl-L-carnitine

CART

classification and regression tree

DKA

diabetic ketoacidosis

GAD65

glutamic acid decarboxylase

GH

growth hormone

KMV

α-keto-β-methylvalerate

KPD

ketosis-prone diabetes

ROC

receiver operating characteristic

T1D

type 1 diabetes

T2D

type 2 diabetes

Funding

This work was supported by National Institutes of Health grant RO1 DK101411 (F.J., A.B.); and funds from the Agricultural Research Service, U.S. Department of Agriculture, under Cooperative Agreement no. 58-6250-6001 (F.J.).

Author Contributions

F.J. and A.B. designed the studies, analyzed data, and drafted the manuscript. J.W.H. performed mass spectrometric and biochemical analyses and analyzed data. K.R.K. and W.F.P. recruited the participants, implemented the clinical protocols, and developed the database. R.B. and A.L. performed islet autoantibody analyses and reviewed the manuscript. X.H., D.G., and J.B. performed the CART and ROC curve analyses. M.T. contributed to data analysis and manuscript drafting. A.B. and F.J. obtained funding for the study and supervised the protocols. All authors listed made a substantial, direct, and intellectual contribution to the work and approved it for publication.

Disclosures

No author has an outside interest to disclose.

Data Availability

Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.