Atherogenic Lipoproteins Associate with Loss of Glycemic Control in Youth-Onset Type 2 Diabetes: Results from the TODAY Study

Abstract

Background:

Type 2 diabetes (T2D) in adolescents is associated with an unfavorable lipid profile, but lipoprotein particle subspecies and branched-chain amino acid (BCAA) data are scarce.

Objective:

To evaluate lipoprotein particle distributions, Lipoprotein Insulin Resistance Index (LP-IR), and BCAA levels longitudinally and their relationships with sex, race/ethnicity, treatment, and loss of glycemic control in adolescents with youth-onset T2D.

Methods:

Participants from the TODAY study (n=348) had samples analyzed yearly for HbA1c and nuclear magnetic resonance lipoprotein and BCAA assessments.

Results:

At baseline, participants with type 2 diabetes were 13.7 years old with T2D, obesity, and from racial and ethnic minority groups (32.2% Non-Hispanic Black (NHB), 43.7% Hispanic). Smaller low-density lipoprotein (LDL) and larger very low-density lipoprotein (VLDL) sizes, higher high-density lipoprotein (HDL) particle number, and increased LP-IR score predicted worsening of glycemic control. LDL, HDL, and VLDL particle numbers increased over three years with weaker trends for decreasing LDL and HDL size. LP-IR and BCAA levels were higher longitudinally in those who lost glycemic control. Females had larger HDL size than males at baseline and throughout. NHBs had the largest LDL and HDL sizes, smaller VLDL size, and lower LP-IR and BCAA.

Conclusion:

These data in youth with T2D demonstrate a progressive atherogenic lipoprotein phenotype over 3 years. Increased LP-IR and BCAA are associated with worsening glycemic control and may be contributing to the premature development of atherosclerosis in youth with T2D.

INTRODUCTION:

Prior studies in adults show that insulin resistance and type 2 diabetes (T2D) are associated with a pattern of increased circulating branched-chain amino acids (BCAAs)¹ and atherogenic lipoprotein abnormalities including increased low-density lipoprotein particle number (LDL-P) relative to levels of LDL cholesterol (LDL-C), smaller average sizes of LDL and high-density lipoprotein (HDL) particles, and increased very low-density lipoprotein (VLDL) size.^2-4 Combining the VLDL, LDL, and HDL subclass and size hallmarks of insulin resistance into the Lipoprotein Insulin Resistance Index (LP-IR), measured clinically by nuclear magnetic resonance (NMR) spectroscopy, provides strong prediction of both incident T2D and premature coronary heart disease (CHD).^5-7 Simultaneously-assessed BCAA levels are also predictive of CHD and T2D.^7-9 Although limited data are available for youth with T2D, the same lipoprotein and amino acid abnormalities have been documented in nondiabetic insulin resistant children and adolescents.^10-13

The TODAY (Treatment Options for type 2 Diabetes in Adolescents and Youth) study was a multicenter, randomized clinical trial designed to compare the effect of three treatment regimens for maintaining glycemic control in youth with recent-onset T2D. Treatment groups (metformin monotherapy, metformin plus rosiglitazone, metformin plus intensive lifestyle intervention) were compared by time to treatment failure (primary outcome), defined as loss of glycemic control (defined as HbA1c ≥8% [64 mmol/mol] for 6 months or inability to wean from temporary insulin therapy within 3 months after metabolic decompensation).¹⁴ Insulin was initiated at the time of primary outcome.¹⁵ Data from adolescents in the TODAY study showed that conventional lipid profiles became more atherogenic during follow-up.¹⁶ Apolipoprotein B (ApoB) levels also increased and these changes were associated with HbA1c.¹⁶

The primary objective of this analysis was to investigate lipoprotein particle distributions, LP-IR, and BCAA levels longitudinally in the 3 treatment groups, and determine how baseline levels and 1-year lipoprotein and BCAA changes related to future loss of glycemic control. We hypothesized that serial assessment of BCAA and lipoprotein particle distributions would demonstrate adverse changes over time associated with worsening of glycemic control.

MATERIAL AND METHODS

TODAY Study Participants and Study Design.

The TODAY study was a multicenter randomized controlled clinical trial designed to evaluate the effects of three treatments for T2D in adolescents and youth (metformin alone [M], metformin plus an intensive lifestyle intervention [M+L], and metformin plus rosiglitazone [M+R]) on time to failure to maintain glycemic control and has been described in detail previously.¹⁴ The primary outcome of the trial was time to treatment failure defined as a sustained HbA1c ≥ 8% for 6 months or inability to wean from insulin after metabolic decompensation. After an average of 3.86 years (median 3.93, minimum 2 years) of follow-up, 319 (45.6%) participants reached the primary outcome. Rates of failure were 51.7%, 46.6%, and 38.6% for M, M+L, and M+R, respectfully. Participants in the M+R group demonstrated significantly less treatment failure than participants in the metformin alone group (p=0.0006).¹⁷ When participants reached primary outcome, metformin was continued, rosiglitazone was discontinued in the M+R group, and insulin was added. At the conclusion of TODAY, participants were enrolled in the TODAY follow-up study (TODAY2), where they continued to receive standard of care diabetes treatment from the TODAY study team and received metformin and/or insulin as needed to maintain glycemic control.¹⁸

Lipid Treatment.

During TODAY, lipid lowering medications, primarily atorvastatin, were initiated for persistent LDL-C levels ≥130 mg/dL or triglyceride (TG) levels 300-599 mg/dL after six months of nutrition and diabetes management per algorithm.¹⁴ For TG ≥600 mg/dL, fibrate therapy could be initiated at the discretion of the study physician.

Anthropometrics and Blood Pressure.

Height (cm) and weight (kg) were measured as previously described¹⁹ and used to calculate BMI in kg/m² which was converted to an age and gender specific BMI percentile using CDC 2000 criteria.²⁰ Blood pressure was measured after a 5-minute rest with the participant in a sitting position using a CAS 740 monitor with standardized oscillometric cuff sizes at every visit. Three measurements were taken at 1-minute intervals. The average of the 2nd and 3rd systolic and diastolic measures was calculated to obtain blood pressure at that visit. Mean arterial pressure (MAP) was calculated as (1/3)*[(2*diastolic blood pressure) + systolic blood pressure].

Laboratory Measurements.

Fasting EDTA plasma, serum and whole blood samples were shipped on dry ice to the Northwest Lipid Research Laboratory (NLRL, University of Washington, Seattle, WA) where standard lipid, ApoB, and HbA1c measurements were performed. The ratio of TG to HDL-C was calculated at each visit as a lipid panel indicator of insulin resistance.²¹ The methods and analytical performance for determination of HbA1c, lipids, and ApoB were previously described.¹⁶ Fasting insulin as a surrogate measure of insulin resistance was measured at baseline, 6 months, and 2 and 3 years by a double-antibody radioimmunoassay as previously described.¹⁴ NMR LipoProfile^® testing of serum aliquots from participants who provided consent for use of archived frozen samples was performed at Labcorp (formerly LipoScience, Raleigh, NC) on the 400 MHz Profiler platform using the LP4 deconvolution algorithm.^22-24 VLDL, LDL, and HDL subclass particle concentrations were quantified on the basis of the amplitudes of their spectroscopically-distinct lipid methyl group NMR signals.²² Total particle concentrations of VLDL (VLDL-P), LDL (LDL-P), and HDL (HDL-P) are the sums of the particle concentrations of their respective subclasses. Mean VLDL, LDL, and HDL particle sizes (nm diameter) are weighted averages derived from the sum of the diameters of each subclass multiplied by its relative mass percentage.²² Also calculated is the Lipoprotein Insulin Resistance Index (LP-IR), a multi-marker score ranging from 0 (most insulin sensitive) to 100 (most insulin resistant) derived by combining appropriately-weighted subclass particle concentrations of large VLDL, small LDL, and large HDL plus mean VLDL, LDL, and HDL particle sizes.⁵ Total BCAA concentrations were calculated as the sum of valine, leucine, and isoleucine concentrations measured as previously described.²⁵

Analysis Sample.

Participants randomized into TODAY with complete NMR lipoprotein data at baseline and annually for 3 years were included. As previously described, subjects were seen at baseline and follow-up visits for assessments of anthropometric measures (including weight and height), blood pressure, and yearly fasting laboratory studies.¹⁴ Lipoproteins and BCAA were assessed at baseline and yearly thereafter. This resulted in a constant cohort of 348 participants in an effort to maximize sample size and obtain meaningful longitudinal data. Due to the inclusion of 3 years of data, the period of reporting in this paper extends into TODAY2 for some participants.¹⁸

Statistical analysis.

Means (± SD), medians (Q1, Q3), or frequency distributions (for categorical variables) were used to summarize the characteristics of the cohort included in the analytic sample. Statistical differences between those included and not included in the analytic sample were evaluated using t-tests or ANOVA for continuous variables and χ² tests for categorical variables. Baseline and longitudinal associations of each NMR variable with treatment, sex, and race ethnicity were assessed using a generalized linear mixed model with data censored after the initiation of statins or loss of glycemic control. Mean profile plots for the evaluation period were produced for each NMR and chemical variable and trend was assessed for the first year (Y0 to Y1) and for the subsequent two years (Y1 to Y3). Mean profile plots were also produced for each NMR and chemical variable for subgroups of interest (treatment assignment, sex, race-ethnicity, and loss of glycemic control). All figures are presented using the original scale of the NMR and chemical variables. Differences in subgroup variables over time during first year and the subsequent two years of the study were evaluated using a repeated measures mixed model with a visit by subgroup interaction term, adjusted for baseline age, BMI, and MAP. Association of loss of glycemic control with the NMR parameters was assessed using logistic regression models that included baseline values only adjusted for sex, race/ethnicity, treatment assignment, baseline age, MAP, and BMI percentile. The analysis was repeated using baseline NMR values as well as the 1-year change in the value (year 1 value – baseline value) excluding any participants who lost glycemic control in the first year (N=47). The values for LDL-P, BCAA, Triglycerides, ratio of TG to HDL, and insulin were log-transformed to obtain approximate normality for all statistical analyses. Analyses were performed using SAS 9.4 (SAS Institute, Cary, NC).

RESULTS

Demographic and Metabolic Characteristics.

Almost half of the TODAY cohort met the inclusion criteria for the present analysis (lipoprotein characteristics measured yearly from baseline to 3 years). Comparison of participants included in the analysis (N=348) to those excluded (N=351) indicates that those included were slightly younger (0.6 years), had a higher proportion of Hispanics, a lower proportion in the “other” race category, and slightly lower BMI (1.2 kg/m²) and HbA1c (0.2%) at baseline (Supplementary Table 1) with no differences between randomized treatment assignment groups (Supplementary Table 2). Baseline demographic and laboratory characteristics of the study participants are shown in Table 1. Correlations among the lipid, lipoprotein and metabolic variables (Supplementary Table 3) are consistent with those reported in other cohorts with and without T2D.^2,3,5,13 For example, LDL-P is related more strongly to ApoB (r=0.92) than to LDL-C (r=0.79); triglycerides are related inversely to both LDL size (r=−0.39) and HDL size (r=−0.50); and LP-IR is highly correlated with its lipid insulin resistance counterpart, TG:HDL-C ratio (r=0.84), and more weakly with BCAA (r=0.41) and fasting insulin (r=0.34).

Table 1:

Demographic and laboratory characteristics of study participants at baseline (n=348)

Demographic characteristics
Treatment Assignment (%)
Metformin	33.1
Metformin + Rosiglitazone	33.3
Metformin + Lifestyle	33.6
Female (%)	62.4
Age (years)	13.7 ± 2.0
Duration of T2D (months)	7.9 ± 5.9
Race/ethnicity (%)
Non-Hispanic Black	32.2
Hispanic	43.7
Non-Hispanic White	20.7
Other	3.5
Tanner Stage ≥ 4 (%)	88.8
BMI (kg/m2)	34.3 ± 7.6
BMI Z-Score	2.2 ± 0.5
MAP (mm Hg)	81.7 ± 8.4
NMR variables
LDL-P (nmol/L)	1154 (951, 1368)
LDL size (nm)	20.4 ± 0.5
HDL-P (μmol/L)	18.9 ± 3.0
HDL size (nm)	8.6 ± 0.3
VLDL-P (nmol/L)	104.4 ± 42.2
VLDL size (nm)	47.8 ± 7.4
LP-IR score (0-100)	56.9 ± 21.4
BCAA (μmol/L)	451 (404, 500)
Chemical variables
LDL-C (mg/dL)	81.8 ± 25.0
HDL-C (mg/dL)	38.7 ± 8.5
TG (mg/dL)	90.5 (65.0, 131.5)
ApoB (mg/dL)	74.1 ± 20.8
TG:HDL-C ratio	2.4 (1.5, 3.8)
Fasting insulin (μU/mL)	25.6 (16.2, 36.6)
HbA1c (%)	5.9 ± 0.7

Data are mean ± SD, median (Q1, Q3), or percent. ApoB, apolipoprotein B; BCAA, branched-chain amino acids; HDL-C, HDL cholesterol; HDL-P, HDL particles; LDL-C, LDL cholesterol; LDL-P, LDL particles; LP-IR, lipoprotein insulin resistance index; MAP, mean arterial pressure; TG, triglycerides; VLDL-P, VLDL particles.

Longitudinal trends overall and by treatment group.

Fig. 1 shows mean concentrations of NMR-measured lipoprotein and BCAA during the 3-year follow-up, with separate statistical tests for first-year change (reflecting mainly treatment effects) and changes occurring subsequent to year 1. Accompanying graphs separate these longitudinal data by treatment group assignment. Supplementary Fig. 1 contains the corresponding information for the other lipid and metabolic variables and Supplementary Table 4 contains the p-values for trend Particle numbers of LDL, HDL, and VLDL all increased during follow-up, consistent with previously reported trends for LDL-C, HDL-C, TG, and ApoB.¹⁶ LDL (Fig. 1C) and HDL (Fig. 1G) size increased from baseline to year 1, attributable mainly to those in the M+R treatment group (Fig. 1D and Fig. 1H, respectively), and subsequently trended downward. LP-IR (Fig. 1N) and BCAA (Fig. 1P) decreased during year 1 in the M+R treatment arm, consistent with the insulin sensitizing effect of rosiglitazone.⁴ However, both markers increased during the last 2 years of follow-up, signifying worsening insulin resistance.

Fig. 1:

Trajectories over time of NMR variable mean values for all participants and by treatment subgroups. Asterisks (*) at Y1 and carats (^) at Y3 on the x-axis indicate either a significant change during the first year (Y0-Y1) or subsequent follow-up period (Y1-Y3) for mean change in overall plots or a significant interaction between treatment assignment and time during the first year (Y0-Y1) or subsequent follow-up period (Y1-Y3), respectively. All graphs use raw values, but LDL-P and BCAA were log-transformed for statistical analysis; all models adjusted for sex, race-ethnicity, baseline age, BMI, and MAP.

Longitudinal trends by sex and race/ethnicity.

Baseline and longitudinal trends of the NMR data are compared by sex and race/ethnicity in Supplementary Fig. 2 and Supplementary Fig.3, respectively, and in Table 2. At both baseline and longitudinally, females had larger HDL size and lower BCAA levels, and LP-IR scores were lower (longitudinally only). No race/ethnicity differences between Hispanics (H) and non-Hispanic whites (NHW) were seen for any NMR variable. However, non-Hispanic Blacks (NHB) compared to NHW had significantly lower VLDL particle concentrations and sizes, BCAA levels, and LP-IR scores and higher HDL and LDL particle sizes.

Table 2:

Associations between each NMR lipoprotein variable and risk factors of interest at baseline and longitudinally.

	LDL-P (nmol/L)		LDL size (nm)		HDL-P (μmol/L)		HDL size (nm)		VLDL-P (nmol/L)		VLDL size (nm)		LP-IR PI		BCAA (μmol/L)
	β- sign	p- value	β- sign	p- value	β- sign	p- value	β- sign	p- value	β- sign	p- value	β- sign	p- value	β- sign	p- value	β- sign	p- value
Baseline *
Treatment (vs. Metformin)
Metformin + Rosi	↓	0.45	↑	0.85	↑	0.59	↓	0.52	↓	0.43	↓	0.68	↑	0.96	↑	0.40
Metformin + Lifestyle	↓	0.51	↑	0.11	↑	0.33	↓	0.82	↓	0.90	↓	0.44	↓	0.33	↑	0.33
Sex (vs. Male)
Female	↑	0.85	↓	0.33	↑	0.46	↑	<0.01	↑	0.55	↓	0.33	↓	0.17	↓	0.02
Race/Ethnicity (vs. NHW)
Non-Hispanic Black	↑	0.76	↑	<0.01	↓	0.47	↑	<0.01	↓	0.14	↓	<0.01	↓	<0.01	↓	<0.01
Hispanic	↑	0.95	↓	0.44	↓	0.54	↑	0.08	↑	0.65	↓	0.23	↓	0.21	↓	0.15
Glycemic Control (vs. Durable)
Lost	↑	0.35	↓	0.05	↑	0.02	↓	0.31	↑	0.78	↑	0.12	↑	0.17	↑	0.32
Longitudinal **
Treatment (vs. Metformin)
Metformin + Rosi	↓	0.05	↑	<0.01	↑	0.93	↓	0.92	↓	0.52	↓	0.15	↓	0.13	↓	0.53
Metformin + Lifestyle	↓	0.23	↑	0.15	↓	0.96	↓	0.48	↓	0.48	↓	0.30	↓	0.31	↑	0.93
Sex (vs. Male)
Female	↑	0.62	↑	0.82	↑	0.53	↑	<0.01	↑	0.74	↓	0.13	↓	0.03	↓	<0.01
Race/Ethnicity (vs. NHW)
Non-Hispanic Black	↑	0.78	↑	<0.01	↓	0.32	↑	<0.01	↓	0.01	↓	<0.01	↓	<0.01	↓	<0.01
Hispanic	↑	0.91	↓	0.68	↓	0.20	↑	0.20	↑	0.79	↓	0.28	↓	0.44	↓	0.11
Glycemic Control (vs. Durable)
Lost	↑	0.02	↓	<0.01	↑	0.01	↓	0.04	↑	0.42	↑	0.02	↑	<0.01	↑	<0.01

^*Baseline model results obtained from a general linear model include baseline data only for lipoproteins

^**Longitudinal model results obtained using a repeated measures mixed model with time included as a covariate. The models include all postbaseline values for the lipoproteins until year 3 with censoring after initiation of statins or loss of glycemic control.

Temporal patterns in those who lost versus maintained glycemic control.

Baseline and longitudinal differences in the NMR variables among those who lost versus did not lose glycemic control are compared in Fig. 2 and Table 2. At baseline, only HDL-P (higher) and LDL size (lower) differed in those who lost glycemic control compared to participants that maintained glycemic control over 3 years. More differences were seen longitudinally in those who failed treatment: LDL-P, HDL-P, VLDL size, LP-IR, and BCAA were higher and LDL and HDL sizes were smaller. Corresponding results for the other laboratory tests showed longitudinal differences only for insulin and ApoB (higher concentrations in participants with worse glycemic control; Supplementary Fig. 4 and Supplementary Table 5).

Fig. 2:

Trajectories over time of NMR variable mean values in subgroups that lost or maintained glycemic control during follow-up.

Predictors of treatment failure.

Table 3 presents associations of the NMR and chemical variables with the primary endpoint of loss of glycemic control from adjusted models considering baseline values alone and those including both baseline levels and 1-year changes. Higher baseline HDL-P, VLDL size, and LP-IR score and smaller LDL size all predicted loss of glycemic control over three years, whereas none of the chemical measures did, including fasting insulin. To help account for treatment-related changes during the follow-up period, models that included both baseline and 1-year changes were analyzed and showed particularly strong associations for LP-IR score (both p<0.01) and VLDL size (baseline p=0.03; 1-year change p<0.01). Among chemical variables, 1-year changes of HDL-C (p=0.01), triglycerides (p=0.01), and TG:HDL-C ratio (p<0.01) predicted the primary outcome.

Table 3:

Associations of baseline values and 1-year changes of lipoprotein and lipid variables with loss of glycemic control during follow-up.

		Baseline Only Adjusted Model (N=348)*		Baseline and Year 1 Change Adjusted Model (N=301)**
		Odds Ratio (95% CI)	p-value	Odds Ratio (95% CI)	p-value
NMR Variables
LDL-P#	(BL)	1.18 (0.93, 1.51)	0.17	1.07 (0.80, 1.44)	0.64
	(ΔY1)	-	-	1.26 (0.94, 1.68)	0.12
LDL Size	(BL)	0.71 (0.56, 0.91)	<0.01	0.82 (0.62, 1.09)	0.18
	(ΔY1)	-	-	0.72 (0.53, 0.96)	0.03
HDL-P	(BL)	1.36 (1.07, 1.71)	0.01	1.08 (0.80, 1.45)	0.62
	(ΔY1)	-	-	0.91 (0.67, 1.23)	0.53
HDL Size	(BL)	0.80 (0.62, 1.04)	0.09	0.72 (0.52, 1.0)	0.05
	(ΔY1)	-	-	0.77 (0.56, 1.05)	0.09
VLDL-P	(BL)	1.07 (0.85, 1.34)	0.57	1.04 (0.78, 1.38)	0.80
	(ΔY1)	-	-	1.35 (1.00, 1.82)	0.05
VLDL Size	(BL)	1.36 (1.06, 1.75)	0.02	1.44 (1.04, 2.00)	0.03
	(ΔY1)	-	-	1.54 (1.15, 2.06)	<0.01
LP-IR Score	(BL)	1.41 (1.08, 1.84)	0.01	1.65 (1.17, 2.33)	<0.01
	(ΔY1)	-	-	1.92 (1.38, 2.68)	<0.01
BCAA#	(BL)	1.21 (0.95, 1.55)	0.11	1.32 (0.92, 1.87)	0.13
	(ΔY1)	-	-	1.14 (0.82, 1.59)	0.45
Chemical Variables
LDL-C	(BL)	0.92 (0.73, 1.16)	0.47	0.95 (0.71, 1.27)	0.74
	(ΔY1)	-	-	1.07 (0.80, 1.42)	0.67
HDL-C	(BL)	1.08 (0.86, 1.36)	0.49	0.93 (0.70, 1.23)	0.60
	(ΔY1)	-	-	0.63 (0.44, 0.90)	0.01
ApoB	(BL)	1.07 (0.84, 1.36)	0.58	0.98 (0.74, 1.31)	0.91
	(ΔY1)	-	-	1.31 (0.99, 1.74)	0.06
Triglycerides#	(BL)	1.27 (0.99, 1.62)	0.06	1.49 (0.80, 2.76)	0.21
	(ΔY1)	-	-	1.50 (1.10, 2.05)	0.01
TG:HDL-C#	(BL)	1.19 (0.93, 1.52)	0.17	1.17 (0.87, 1.59)	0.30
	(ΔY1)	-	-	1.61 (1.18, 2.18)	<0.01
Fasting Insulin#	(BL)	1.14 (0.88, 1.49)	0.30	3.96 (0.88, 17.79)	0.07
	(ΔY1)	-	-	3.05 (0.68, 13.55)	0.14

Odds ratios are per 1 SD increment. Models adjusted for sex, race, randomized treatment assignment, baseline BMI percentile, baseline age, and baseline MAP

^*Model includes the baseline value of each lipoprotein or lipid, as well as all covariates, to assess association with loss of glycemic control

^**Model includes the baseline value and the change in the value at year 1 from baseline of each lipoprotein or lipid, as well as all covariates, to assess association with loss of glycemic control. The analysis excludes all participants who lost glycemic control between baseline and 1 year (N=47).

^#Analysis conducted using the log transformed version of the variables.

DISCUSSION

In both adults and children with T2D, the combination of obesity, insulin resistance, and relative insulin deficiency is associated with an atherogenic dyslipidemic phenotype of elevated serum triglycerides, decreased HDL cholesterol, and higher numbers of ApoB-containing lipoprotein particles.^2,3,26 We previously reported that the TODAY cohort had a very high prevalence of this dyslipidemia at baseline which worsened during the 3-year follow-up period despite diabetes treatment.¹⁶ The current analysis extends these observations by examining NMR-measured lipoprotein particle concentration and size changes over the same period, plus baseline and longitudinal levels of LP-IR and BCAA, insulin-independent surrogate markers of insulin resistance that have been little studied in youth with T2D. In general, the lipoprotein particle changes paralleled the deleterious lipid changes, most notably the continuous rise in LDL-P (mirroring the ApoB increase) and the post 1-year trend of smaller LDL and HDL particle sizes. Surprisingly, there was a first-year increase in HDL-P (mirroring HDL-C), but no further change longitudinally.

Dyslipidemia, particularly elevated triglycerides, predicts the future onset of type 2 diabetes. Herein, using longitudinal data, we show that lipoprotein characteristics at baseline and one year are strong predictors of loss of diabetes control. On a conventional lipid profile, elevated triglycerides have a moderate association with future worsening of glucose control. Among conventional lipid measures, TG/HDL-C was the most similar in performance to BCAA. A much stronger relationship with the primary outcome exists for lipoprotein characteristics, particularly smaller LDL particle size and larger VLDL particle size. In older adults, the association of higher ApoB levels, higher LDL particle number, and smaller LDL size are often better predictors of future risk of atherosclerotic disease than LDL-C levels.²⁷ The discordance between NHB and NHW with regard to lipid parameters and insulin resistance measures is similar to relationships identified in children without diabetes.¹⁰ In those who lost glycemic control, higher apoB levels reflect increased LDL and VLDL particle numbers.

Of the three diabetes interventions tested in TODAY, only M+R demonstrated superior durability of glycemic control compared with M.¹⁷ It was hypothesized that the 6-month increase in insulin sensitivity achieved uniquely with M+R would help explain the rosiglitazone benefit, but assessment by oral glucose tolerance testing indicated that greater preservation of β-cell function was primarily responsible.¹⁵ In that study, insulin sensitivity assessed by fasting insulin levels were not different over time between those maintained versus failed to maintain glycemic control.¹⁵ Our findings using LP-IR score and BCAA levels as surrogate markers of insulin sensitivity differed, indicating that lipoprotein and amino acid abnormalities do indeed presage the loss of glycemic control (the former more strongly than the latter). Not only did M+R compared to M or M+L lower LP-IR and BCAA (Fig. 1), but T2D youth who ultimately failed to maintain glycemic control had significantly higher longitudinal levels of both insulin resistance markers (Fig. 2). Baseline LP-IR in adjusted logistic regression models predicted glycemic failure, whereas fasting insulin and TG:HDL-C ratio did not (Table 3). The strongest prediction of treatment failure was found using a model that included both baseline and 1-year change values of LP-IR (OR [95% CI]: 1.65 [1.17, 2.33] and 1.92 [1.38, 2.68], respectively; p<0.01 for both). It should be noted that while this analysis excluded participants who initiated insulin therapy between baseline and year 1 for the models that included the 1-year change (N=47), the reported trajectories for insulin values in those who did and did not lose glycemic control (Supplementary Fig. 4 and Supplementary Table 5) included subjects receiving exogenous insulin. As such, higher insulin levels in those who lost control may reflect greater insulin resistance, but also the contribution of exogenous insulin.

Previous studies in adults have indicated that the LP-IR score, independent of glycemia, is more strongly associated with future T2D than either its lipid panel analog, TG:HDL-C ratio, or fasting insulin.^6,28-30 A likely reason is that LP-IR is a composite metabolomic biomarker that reflects the complex, multidimensional influence insulin resistance exerts on lipoprotein metabolism.^5,6 LP-IR can be favorably modified by both diet³¹ and exercise³², has good biological stability and the practical clinical advantage compared to insulin-based assessment methods of being derived from a low-cost, high-throughput “single-scan” metabolomic measurement that simultaneously provides a traditional lipid panel, lipoprotein particle numbers, ApoB, BCAA levels, and the GlycA inflammation marker.^33,34 LP-IR is less well described in pediatric populations, but was shown to be associated with both fitness and BMI category in the middle school-based HEALTHY study of nondiabetic adolescents.³⁶ LP-IR (and other NMR metabolomic biomarkers) was also measured in youth the same age as those in TODAY and found to be strongly associated with obesity, HbA1c, and insulin resistance assessed by fasting insulin and glucose.¹³ LP-IR scores were highest in the obese T2D subgroup, similar to levels in our TODAY cohort (mean ~57). What was not known until the present study was that relatively small differences in LP-IR within this high range in youth with T2D are predictive of loss of glycemic control.

Interest in BCAA levels as an alternative or complementary indicator of insulin resistance and glycemic control is growing due to metabolomic profiling studies of young and older adults revealing that elevated levels are related to obesity and insulin resistance and predict future development of T2D.^1,8,9,37 BCAA levels appear to be dynamic and modifiable, as shown by interventions that improve metabolic health, such as gastric bypass surgery.³⁸ Studies in pediatric populations, though fewer, are consistent with results in adults showing positive associations of BCAA with obesity, HbA1c, and insulin-measured insulin resistance.^12,13 However, in agreement with our findings in TODAY, BCAA levels appear to be related less strongly than LP-IR to the degree of glycemic control.¹³

This study has a number of strengths, including a diverse national cohort, racial, ethnic, and sex diversity, careful phenotyping, high quality laboratory determinations, repeat measurements over time, and a relatively large sample size without missing data. A limitation is the inability to assess compliance with statin therapy. A sensitivity analysis was performed where participants had data censored after statins were begun. All findings from the sensitivity analysis were consistent with the findings from the full analysis (data not shown). We suspect that the lack of a robust response to statins may be secondary to non-adherence.

CONCLUSIONS

In the TODAY study of newly diagnosed youth-onset T2D, participants demonstrated worsening dyslipidemia with increasing atherogenic lipoproteins over time, raising concern for premature development of atherosclerosis. We found that amino acid and lipoprotein changes reflecting increased insulin resistance were associated with loss of glycemic control, more strongly than our prior findings with regard to a conventional lipid profile.¹⁶ Clinicians following adolescents and young adults with T2D can use either a conventional lipid profile or NMR spectroscopy to both assess cardiovascular risk and potentially predict those likely to lose glycemic control in the near future. The challenges in achieving optimal glycemic control in adolescents with T2D highlight the critical need to promote a healthy lifestyle and to conduct clinical trials to improve glycemic and lipid control to prevent or postpone cardiovascular comorbidities.

HIGHLIGHTS.

• Atherogenic dyslipidemia worsens over 3 years in adolescents with type 2 diabetes.

• Smaller LDL, larger VLDL, and more HDL particles predict worse glycemic control.

• Worse glucose control was better predicted by high LP-IR score than fasting insulin.

• Branched-chain amino acids were higher in participants who lost glycemic control.