All-cause mortality in people with type 2 diabetes and lipodystrophy-like metabolic phenotype

Abstract

Aim

We wanted to investigate whether the lipodystrophy (LD)-like phenotype increases mortality in type 2 diabetes. To this end we first developed and confirmed the triglyceride/BMI ratio (TBR index) as an inexpensive and widely available marker of the LD-like metabolic phenotype. The TBR index was then tested for association with all-cause mortality in patients with type 2 diabetes.

Research design and methods

Associations between TBR and mortality were assessed by Cox in 3 type 2 diabetes cohorts (n = 2137, n = 960, and n = 5060). A 20-SNP genotype risk score (GRS), associated with LD-like phenotype was investigated to identify the molecular basis of TBR variability.

Main finding

In all cohorts, the highest TBR decile showed features of LD-like phenotype, including atherogenic dyslipidemia, lower BMI, worse glycemic control, insulin resistance, elevated gamma-glutamyl transferase, and hypoadiponectinemia. In the meta-analysis, decile 10 had higher mortality risk (HR = 1.46, P < .0001). Finally, in a subset of 1 of the 3 cohorts, the LD-like GRS was significantly associated with TBR.

Conclusion

The TBR is a novel, inexpensive and easily available marker that identifies a subgroup of type 2 diabetes patients with a LD-like phenotype, partly explained by genetic factors. Our findings help disentangle the heterogeneity of death risk in diabetes, refine diagnostic and follow-up algorithms and enable specific treatments for improving long-term outcomes in LD-like individuals.

Type 2 diabetes is a complex and heterogeneous disorder that includes distinct subgroups of patients with different pathogenic backgrounds and clinical outcomes (1, 2). A subset of individuals who exhibit metabolic features resembling a mild form of lipodystrophy, commonly referred to as the lipodystrophy-like (LD-like) phenotype, has recently been described (2-6). This phenotype is characterized by insulin resistance/diabetes, hypertriglyceridemia and fatty liver, in the presence of reduced subcutaneous adipose tissue (2-5). Although there is no established formal diagnostic criterion, the gold standard identification of the LD-like phenotype is currently based on the assessment of body fat distribution by means of dual-energy x-ray absorptiometry (DXA), an expensive and technically demanding approach. Unfortunately, no easy-to-use markers are currently available for the LD-like metabolic phenotype, either in the general population or in type 2 diabetes. This phenotype, therefore, often goes unnoticed in clinical practice, both in nonobese individuals in whom triglyceride levels may still be within the normal range, although higher than expected based on body mass index (BMI) (7) and in overweight-obese individuals, in whom the markedly elevated triglyceride levels go overlooked and are believed to be the consequence of increased adiposity.

Patients with type 2 diabetes have an increased risk of all-cause mortality compared to the general population (8, 9). It is unknown whether this premature risk of death is increased and can be predicted in some phenotypic subgroups of patients with type 2 diabetes, opening avenues for precision medicine. We assessed whether the LD-like metabolic phenotype further increases mortality rates.

To this end we first developed and confirmed the triglyceride/BMI ratio (TBR index) as an inexpensive and widely available marker of the LD-like metabolic phenotype across the full range of lipid and adiposity levels. The TBR index was then used to investigate whether the LD-like metabolic phenotype is associated with all-cause mortality in patients with type 2 diabetes.

Research design and methods

Subjects

Patients were from 3 cohorts of type 2 diabetes individuals, the aggregate Gargano Mortality Study (aGMS) from Apulia, Central-Southern Italy; the Pisa Mortality Study (PMS) from Tuscany, Central Italy; and the standard glycemic arm of the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, which served as an independent replication cohort differing in ancestry, geography, recruitment setting and design.

Aggregate Gargano Mortality Study

The aGMS consists of 2137 people with type 2 diabetes recruited and followed up as already described (10, 11), whose clinical features are shown in Table 1, left side. Briefly, patients were enrolled from 2000 to 2008 (first period) and from 2008 to 2011 (second period) at the Endocrinology Unit of Fondazione Istituto di Ricovero e Cura a Carattere Scientifico “Casa Sollievo della Sofferenza,” and followed until February 2023 (10, 11). Presence of type 2 diabetes according to the 2003 American Diabetes Association guideline was the only inclusion criterion, while the presence of poor life expectancy for non-diabetes-related diseases was the only exclusion criterion. Patients were followed until February 2023 (10, 11). The study and the informed consent procedures were approved by the local Institutional Ethic Committee. All participants gave written consent.

Table 1.

Clinical features of the aGMS, PMS, and ACCORD study

	aGMS (n = 2137)	PMS (n = 960)	ACCORD (n = 5060)
Age, yrs	62.0 ± 9.4	59.6 ± 7.1	62.8 ± 6.6
Women, n; %	993; 46.5	390; 40.6	1935; 38.2
Smoking habit, n; % smokers	310; 14.5	201; 20.9	597; 11.8
BMI, kg/m2	31.0 ± 5.8	29.6 ± 5.2	32.2 ± 5.4
Age at diagnosis, yrs	50.8 ± 10.6	49.6 ± 9.6	51.9 ± 9.1
HbA1c mmol/mol	69.2 ± 20.8	59.3 ± 13.2	67.1 ± 10.9
HbA1c, %	8.5 ± 1.9	7.6 ± 1.2	8.3 ± 1.0
LDL, mg/dL	113.5 ± 37.9	130.1 ± 31.5	104.6 ± 32.9
HDL-cholesterol, mg/dL	44.5 ± 12.7	50.5 ± 13.7	41.8 ± 11.1
Triglycerides, mg/dL	128 (91-181)	133 (97-199)	154 (108-226)
Insulin therapy, n; %	892; 41.7	248; 25.8	1818; 35.9
Statin therapy, n; %	951; 44.5	257; 26.8	3235; 63.9
Anti-hypertension therapy, n; %	1416; 66.3	521; 54.3	4374; 86.4
Follow-up, yrs (person/yrs)	13.7 ± 5.8 (29 278)	12.9 ± 2.7 (12 384)	4.9 ± 1.2 (24 960)
IR of death, 100 person-yr	3.18 (2.97; 3.41)	1.59 (1.37; 1.85)	1.30 (1.16; 1.45)

Continuous variables are reported as mean ± SD or median and IQR (25th-75th percentile), and categorical variables as total frequencies and percentages.

Abbreviations: aGMS, aggregate Gargano Mortality Study; BMI, body mass index; HbA1c, glycated hemoglobin; LDL, low-density lipoprotein; HDL, high-density lipoprotein; IR, incidence rate.

Pisa Mortality Study

The PMS comprises 960 patients whose clinical features are shown in Table 1, middle column. Patients were consecutively recruited from 2002 to 2008 according to the same criteria used for the aGMS and followed until December 2017 at the Endocrine Unit of the University of Pisa (Central Italy) (12-14).

For both cohorts, the vital status of participants was verified by interrogating the Italian Health Card Database upon data anonymization (https://sistemats1.sanita.finanze.it/portale/home) (12-14).

ACCORD study

The ACCORD study was a multicenter clinical trial (NCT00000620) aimed at investigating the effect of intensive glycemic control (glycated hemoglobin [HbA1c] < 6.0%), as compared to standard control (HbA1c 7%-7.9%), on cardiovascular health (PMID: 17599422, PMID: 21366473). A total of 10 251 participants with type 2 diabetes who were at high cardiovascular risk were randomized in a 1:1 ratio to receive intensive or standard glycemic therapy at clinical sites in the United States and Canada. The trial also investigated the effect of intensive vs standard blood pressure (BP) control and fibrate vs placebo therapy on cardiovascular events in a double 2-by-2 factorial design. The full trial protocol, including the eligibility criteria, has been previously published (PMID: 17599422). The glycemia trial was stopped after a mean of 3.7 years because of excess mortality in the intensive arm (PMID: 18539917). Participants in the intensive arm were transferred to the standard glucose protocol and monitored for an additional 17 months while the BP and lipid trials were completed. Due to the effect of intensive glycemic control on mortality, the present analysis was limited to participants in the standard glycemic arm (n = 5123), for whom data were available for all variables included in the regression models (n = 5060). Ethics committees at each center approved the protocol, which adhered to the Declaration of Helsinki. All participants provided written, informed consent.

Whole exome sequencing

In the aGMS, 1 out of 4 patients (53/214) randomly selected from decile 10 of the TBR index, underwent next-generation sequencing. This was an exploratory analysis not designed to estimate the frequency of pathogenic variants. DNA quality and integrity assessment were performed by using genomic DNA ScreenTape assay on Agilent 2200 TapeStation system according to the manufacturer's protocol. Capture-based target enrichment kit (SOPHIA Whole Exome Solution v1) for library preparation followed by sequencing on NovaSeq6000 (Illumina) were assessed. Results were aligned to GRCh37/hg19 as reference genome and variant annotation was performed by means of Sophia DDM solution (pipeline version p.5.5.36 del 25.02.2021). Variants’ interpretation was accomplished according to the American College of Medical Genetics and Genomics guidelines (15). The 3 established genes responsible for familial partial lipodystrophy (FPLD) were analyzed (ie, LMNA [NM_170707.4], PLIN1 [NM_002666.5] and PPARG [NM_138711.6]). Sanger sequencing by Eurofins Genomics service using a BigDye Terminator v3.1 on ABI- 3130 (Applied Biosystems) was performed to confirm variant identified by next-generation sequencing.

Genome-wide genotyping

In 699 out of 2137 individuals of the aGMS, 20 single nucleotide polymorphisms (SNPs), namely, rs6795735 (ADAMTS9), rs11708067 (ADCY5), rs459193 (ANKRD55), rs3843467 (ANKRD55_2), rs702634 (ARL15), rs9891146 (C17orf58), rs11063069 (CCND2), rs2925979 (CMIP), rs17106184 (FAF1), rs13389219 (GRB14), rs2943641 (IRS1), rs2276853 (KIF9), rs3996352 (KLF14), rs10503669 (LPL), rs2820443 (LYPLAL1), rs2296172 (MACF1), rs1106240 (MPHOSPH9), rs3132524 (POU5F1), rs1801282 (PPARG), and rs4457053 (ZBED3), previously reported to associate with an LD cluster (2) were used to create a genotype risk score (GRS) as the sum of variant genotype dosage. Genotypes derive from a genome-wide genotyping carried out at the Wellcome Trust Centre for Human Genetics laboratory (Oxford University, UK), using the Affymetrix UK Biobank Axiom 2.0 array on a Gene Titan platform (Thermo Fisher Scientific), according to vendor's standard protocol. All SNPs were genotyped or imputed with high quality (r² values >0.95) (V.T., manuscript in preparation).

Statistical analysis

Baseline features of the study individuals are shown as mean ± SD or median with interquartile range (IQR, ie, 25th, 75th percentile) for continuous variables and as frequency and percentage for categorical traits.

In all 3 cohorts, follow-up was defined as the time difference between date of the event (ie, all-cause mortality) and date of enrollment. The date of the last available clinical follow-up was considered for subjects who did not experience the event. All-cause mortality incidence rate was expressed as the number of events per 100 person-years, adjusted for age and sex.

The ratio of triglycerides over BMI (TBR) was computed as the natural logarithm (ie, Ln [triglycerides/BMI]). TBR strata were assessed by computing deciles based on its empirical distribution. Briefly, the TBR values were ranked in ascending order, and the population was divided into 10 equally sized groups (deciles), each representing 10% of the study cohort. Individuals in the first decile had the lowest TBR values, while those in the tenth decile had the highest.

TBR was modeled differently according to the analyses. First, TBR was used as a continuous variable for testing the associations (by linear or logistic regression, as appropriate), with baseline metabolic traits related to the LD-like phenotype. Second, after noticing the nonlinear nature of the association with mortality (by comparing minimally adjusted Cox models in which TBR was entered either as a continuous or as a categorical variable; ie, Akaike Information Criterion [AIC]), TBR was used as a dichotomous variable (decile 10 vs deciles 1-9) by Cox regression. The latter association was estimated both in a model that accounted only for general confounders (ie, age, sex, and smoking status) and in a fully adjusted model that also included HbA1c, TyA1c (triglycerides-HbA1c calculated as Ln[(HbA1c × triglycerides/2]), high-density lipoprotein (HDL), gamma-glutamyl transferase (GGT), and insulin therapy. Risks were estimated as hazard ratios (HRs) along with their 95% CI. Third, when baseline clinical features between decile 10 vs deciles 1-9 were compared, linear or logistic regression analyses were used, as appropriate.

For the aGMS, all analyses were also adjusted for the “recruitment period” (ie, 0 or 1 for the first period—2000 to 2008—and the second period—2008 to 2011—respectively). For ACCORD, analyses were additionally adjusted for assignments to the BP or lipid sub-trials and corresponding treatments.

Cohort-specific estimates were meta-analyzed to quantify the association between the TBR risk for all-cause mortality (16, 17). Pooled HRs were estimated using fixed effects model with inverse variant method, as no significant heterogeneity was detected among the 3 studies. Statistical heterogeneity was assessed using the I² statistic, with values > 50% indicating substantial heterogeneity, and Cochran's Q test, with a P value <.10 considered significant.

The association between the TBR and the GRS z-score (ie, the number of SDs each value deviates from the mean) was estimated by linear regression in the subset of aGMS participants with genome-wide genotyping (n = 699). Robustness of the estimate was assessed by nonparametric bootstrap resampling (n = 1000 resamples with replacement).

A two-sided P value <.05 was considered as statistically significant. All analyses were performed using SPSS statistics version 25 and SAS 9.4 TS.

Results

Phenotype studies

The clinical features of aGMS are shown in Table 1, left. Overall, the mean age at baseline was 62.0 ± 9.4 years, 46.5% of participants were women, BMI was 31.0 ± 5.8 kg/m², HbA1c was 69 ± 0.01 mmol/mol (8.5% ± 1.9%), and diabetes duration was approximately 11 years. A total of 892 (41.7%), 951 (44.5%), and 1416 (66.3%) patients were on insulin, statins, and BP treatment, respectively. During a mean follow-up of 13.7 ± 5.8 years (ie, 29 277 person-years), 1064 (49.8%) died. The age and gender adjusted incidence rate of all-cause mortality was 3.18 (2.97-3.41) per 100 person-years.

The LD-like metabolic phenotype was investigated by the TBR index (ie, Ln [triglycerides/BMI]; median and IQR = 1.43, 1.10-1.77), whose distribution is shown in Fig. S1 (18).

As shown in Table 2, in addition to higher triglyceride levels and lower BMI, an expected finding, given how this index is computed, a higher TBR value was associated with several other baseline clinical features: specifically, higher levels of HbA1c, TyA1c (an established index of insulin resistance) (19-21), GGT (a reliable marker of metabolic-associated fatty liver disease), as well as lower levels of HDL-cholesterol and total adiponectin. The clustering of these abnormalities in the presence of reduced adiposity is compatible with the LD-like metabolic phenotype (5).

Table 2.

Associations between TBR and clinical traits related to the LD-like metabolic phenotype in the aGMS

	Β + SE	P value
BMI, kg/m2a	−2.1 ± 0.24	<1.0 × 10−4
HbA1c, %	0.7 ± 0.08	<1.0  ×  10−4
HDL-cholesterol, mg/dL	−8.0 ± 0.5	<1.0  ×  10−4
Triglycerides, mg/dLa	193.0 ± 3.8	<1.0  ×  10−4
TyA1c insulin resistance index, unit	1.01 ± 0.01	<1.0  ×  10−4
Total APM, µg/mLb	−0.22 ± 0.03	<1.0  ×  10−4
GGT, U/L	7.71 ± 2.58	<1.0  ×  10−4

The association between TBR and clinical features was evaluated by univariate or logistic regression, as appropriate after adjusting for the recruitment period. Β values represent the effect size for 1-unit increment of TBR.

Abbreviations: aGMS, aggregate Gargano Mortality Study; BMI, body mass index; HbA1c, glycated hemoglobin; HDL, high-density lipoprotein; TyA1c, triglycerides-HbA1c calculated as Ln[(HbA1c × triglycerides/2)]; APM, adiponectin; GGT, gamma-glutamyl transferase.

^a These 2 variables are used to generate the TBR index. Their associations were, therefore, expected by construction.

^b APM measurements were available only in 1058 individuals.

In addition, the TBR was higher in men vs women (1.51 ± 0.53 vs 1.41 ± 0.50, respectively; P < .0001) and in smokers vs nonsmokers (1.58 ± 0.53 vs 1.44 ± 0.52, respectively; P < .0001) and negatively correlated with age (β = −1.17 ± 0.39, P = .003).

To investigate the nature of the relationship between the TBR and mortality, deciles were analyzed. As compared to decile 1, only decile 10 was significantly associated with all-cause mortality (HR = 1.60, 95% CI: 1.2-2.1, after adjusting for general confounders, including age, sex, and smoking habit (Fig. S2 (18)), with a nonlinear association confirmed by ΔAIC = 4.78. The risk of death of the last decile was increased by more than 50% compared to deciles 1-9 (adjusted HR = 1.56, 95%: 1.29-1.88; P < .0001). This association did not change much when all the variables associated with TBR (ie, HbA1c, TyA1c, HDL, GGT, and insulin therapy) were added to the model (HR = 1.50, 95%: 1.14-1.98; P < .004). Very similar results were obtained when the 2 models were investigated in the subgroup of 1058 individuals for whom serum adiponectin was available (HR moving from 1.95 [95% CI 1.56-2.44] to 1.88 [95% CI: 1.35-2.62], respectively).

The clinical features of patients of decile 10 as compared to deciles 1-9 are shown in Table 3, left. Individuals in decile 10, in addition to higher triglyceride levels and lower BMI, a predictable finding, given how this index is calculated, had insulin resistance (as indicated by higher TyA1c index) (20), higher serum levels of GGT, and lower adiponectin serum levels (data of the latter available in 50% of the whole sample), all features clearly resembling those of a LD-like metabolic phenotype (7, 22). Decile 10 patients were also younger, less likely to be women, and more likely to be treated with insulin (Table 3, left).

Table 3.

Clinical features according to TBR deciles (decile 10 vs 1-9) in the aGMS, PMS and ACCORD study

	aGMS (n = 2137)			PMS (n = 960)			ACCORD (n = 5006)
	10th decile	Lower deciles		10th decile	Lower deciles		10th decile	Lower deciles
	(n = 214; 10.0%)	(n = 1923; 90.0%)	P value	(n = 97; 10.1%)	(n = 863; (89.9%)	P value	(n = 506; 10%)	(n = 4554; 90%)	P value
Age, yrs	60.1 ± 9.7	62.2 ± 9.4	.002	57.4 ± 7.5	59.9 ± 7.1	.001	61.7 ± 6.5	62.9 ± 6.6	1 × 10−4
Women, n; %	73; 34.1	920; 47.8	<1.0  ×  10−4	26; 26.8	499; 57.8	.004	182; 36.0	1753; 38.5	.27
Smokers, n; %	42; 19.6	268; 13.9	.025	27; 27.8	174; 20.2	.080	71; 14.0	526; 11.5	.10
BMI, kg/m2	29.9 ± 4.4	31.1 ± 5.9	.003	28.8 ± 3.9	29.7 ± 5.3	.043	30.9 ± 4.9	32.4 ± 5.4	<5  ×  10−4
Age at diagnosis, yrs	49.3 ± 10.5	51.0 ± 10.6	.028	48.4 ± 8.7	49.7 ± 9.7	.21	52.3 ± 7.9	51.9 ± 9.2	.37
HbA1c, %	9.3 ± 2.1	8.4 ± 1.9	<1.0  ×  10−4	7.8 ± 1.3	7.5 ± 1.2	.006	8.4 ± 1.0	8.3 ± 1.0	.0005
LDL cholesterol, mg/dL	112.6 ± 47.3	113.5 ± 36.7	.42	135.1 ± 34.8	129.5 ± 31.1	.098	103.6 ± 37.9	104.7 ± 32.4	.52
HDL-cholesterol, mg/dL	38.4 ± 10.8	45.2 ± 12.8	<1.0  ×  10−4	41.1 ± 8.3	51.5 ± 13.8	<1.0  ×  10−4	34.1 ± 8.4	42.7 ± 11.0	<5 × 10−4
Triglycerides, mg/dL	328 (274-396)	121.0 (88-160)	<1.0 ×10−4	324 (287-399)	125 (94-170)	<1.0  ×  10−4	422 (349-523)	144 (102-198)	<5  ×  10−4
TyA1c insulin resistance index	7.4 ± 0.5	6.2 ± 0.5	<1.0  ×  10−4	7.2 ± 0.3	6.1 ± 0.5	<1.0  ×  10−4	7.4 ± 0.3	7.2 ± 0.4	<1  ×  10−4
Total APM, µg/mL	4.7 ± 2.7 (n = 131)	5.8 ± 3.4 (n = 927)	<1.0  ×  10−4	NA	NA	NA	N.A.	N.A.
GGT, U/L	30.5 (21.8-52.3) (n = 162)	23 (16-36) (n = 1404)	<1.0  ×  10−4	34 (22-61.5)	25 (17-39)	<1.0  ×  10−4	N.A.	N.A.
GPT, U/L	NA	NA		28 (20.5-36.5)	23 (17-31)	.007	27 (20-36)	24 (18-32)	<1  ×  10−4
GOT, U/L	NA	NA		20 (16-24)	19 (16-23)	.326	N.A.	N.A.
Insulin therapy, n; %	108; 50.5	784; 40.8	.006	27; 27.8	221; 25.6	.63	120; 23.7	1698; 37.3	<1  ×  10−4
Anti-hypertension therapy, n; %	140; 65.4	1276; 66.4	.78	50; 51.5	471; 54.6	.57	276; 54.6	2959; 65.0	<1  ×  10−4
Statin therapy, n; %	96; 44.9	855; 44.5	.91	28; 28.9	229; 26.5	.62	430; 85.0	3944; 86.6	.31

Continuous variables are reported as mean ± SD or median and IQR (25th-75th percentile), and categorical variables as total frequencies and percentages.

Differences between the 2 groups of patients were evaluated by either univariate or logistic regression or Kolmogorov-Smirnov, as appropriate. For aGMS, the analyses were adjusted for the recruitment period. For 207 ACCORD, analyses were additionally adjusted for assignments to the BP or lipid sub-trials 208 and corresponding treatment. P values < .05 are in italic.

Abbreviations: aGMS, aggregate Gargano Mortality Study; BMI, body mass index; HbA1c, glycated hemoglobin A1c; LDL, low-density lipoprotein; HDL, high-density lipoprotein; TyA1c, triglycerides-HbA1c calculated as Ln[(HbA1c × triglycerides/2]; APM, adiponectin; GGT, gamma-glutamyl transferase; GPT, glutamate pyruvate transaminase; GOT, glutamate oxaloacetate transaminase. NA, not available.

The risk of death in decile 10 patients as compared to deciles 1-9 was replicated in the PMS (n = 960 individuals with type 2 diabetes, whose clinical features are shown in Table 1, middle, TBR median IQR = 1.52, 1.23-1.89, with adjusted HR for all-cause mortality being 1.79 (95% CI: 1.21-2.64; P = .0036). When clinical features in decile 10 patients were compared to decile 1-9, similar differences to those observed in the aGMS were seen (Table 3, middle).

In the ACCORD study, whose clinical features are shown in Table 1, right (TBR median, IQR = 1.58, 1.21-1.96), the direction and magnitude of the association with all-cause mortality were consistent with those observed in the 2 Italian cohorts (P for HR heterogeneity being >.3 for both the aGMS and PMS), although not reaching formal statistical significance (adjusted HR = 1.26 95% CI: 0.90-1.78; P = .18). When clinical features in decile 10 patients were compared to decile 1-9, similar differences to those observed in the aGMS and in the PMS were seen (Table 3, right).

In a combined analysis of the 3 studies, the TBR index was strongly associated to all-cause mortality: adjusted HR = 1.53 (95% CI: 1.31-1.78; P < .0001, Fig. 1 upper panel). No major changes were observed when other possible covariates, including HbA1c, and ongoing treatments were added to the minimally adjusted model (Fig. 1 lower panel).

Figure 1.

Meta-analysis performed using fixed effect model with inverse variance method to compare the association between TBR and all-cause mortality in aGMS, PMS, and ACCORD study. TBR was analyzed as a dichotomous variable, comparing individuals in the decile 10 with those in deciles 1-9 (reference group). Panel A shows the summarized HR adjusted for age, sex, and smoking habit (HR =1.53 [95% CI: 1.31; 1.78]; P < 1 × 10⁻⁴). Panel B shows the summarized HR adjusted for age, sex, smoking habit, HbA1c and ongoing treatments (HR = 1.46 [95% CI: 1.26-1.70]; P < 1 × 10⁻⁴). Abbreviations: aGMS, aggregate Gargano Mortality Study.

Genetic studies

We then explored whether rare variants in genes associated with FPLD (ie, LMNA, PLIN1, and PPARG) were present among patients in TBR decile 10 (n = 53, 1 out of 4). Three individuals were carriers of heterozygous variants in the 3 tested genes (see Table S1 (18)). According to the American College of Medical Genetics and Genomics (ACMG) guidelines (15), 2 (1 in PLIN1 and 1 in PPARG) were classified as variants of uncertain significance and were therefore not further considered. The variant in LMNA (NM_170707.4: c.1445G>A, rs11575937; p.Arg482Gln, which is present in 1 out of 196 673 European individuals in the Genome Aggregation Database; gnomAD v4.0.0) was classified as pathogenic (as indicated by human and animal studies (23-27)), consistent with a causal role in FPLD2. Indeed, the clinical features of our carrier were compatible with the diagnosis of FPLD2. The patient was a 61-year-old female with a BMI of 24.7 kg/m². Type 2 diabetes was diagnosed at the age of 57 years. At the same age, the patient had an acute myocardial infarction and was diagnosed with hepatic steatosis at ultrasonography. Her HbA1c was 121 mmol/mol (13.2%) while on treatment with insulin. Triglycerides, total cholesterol, and HDL-cholesterol levels at study entry were 287, 189 mg/dL and 44 mg/dL, respectively. Her BP was 145/85 mm Hg while on anti-hypertension treatment. The TBR index was 2.45, which is close to the median value of decile 10 (ie, 2.37; IQR = 2.25-2.57). Despite the exploratory nature of this finding, it is worth noting that, given its very low frequency, the probability of observing the LMNA p.Arg482Gln variant among 53 individuals drawn at random was very unlikely, equal to approximatively 1 in 3700 attempts.

Next, we wanted to study the putative role of a polygenic risk score associated with LD-like metabolic phenotype (2). To this end, a 20-SNP GRS, previously reported to associate with several variables clustering in a LD phenotype (2) was investigated in 699 individuals from the aGMS. The clinical features of these individuals are representative of the whole cohort (see Table S2 (18)). We found a positive association between the GRS and the TBR index (β coefficient = 0.042 ± 0.02 SE, P = .028 by bootstrap in a minimally adjusted model, including age, sex, and smoking habit) which explained approximately 3% of the TBR variance (adjusted R² = 0.028). In addition, the GRS was significantly higher in decile 10 individuals than in individuals from deciles 1-9 (standardized GRS = 0.20 ± 0.97 vs −0.03 ± 1.0, P = .048).

Discussion

Patients with type 2 diabetes show a decreased life expectancy of 10 to 14 years compared to the nondiabetic population (8, 9). The present study aimed to assess whether the LD-like metabolic phenotype, which characterizes a certain number of patients with type 2 diabetes (2, 3, 28), further increases this risk. To answer this question, we first developed a novel, inexpensive, widely available and easy-to-use proxy of the LD-like metabolic phenotype, based on the evidence that high triglycerides levels relative to the degree of adiposity is the hallmark of this subgroup of individuals (2, 3). Indeed, we confirmed that the TBR index is associated with all clinical and metabolic features known to cluster in the LD-like phenotype (2, 3, 29).

In addition, in a meta-analysis of 3 independent cohorts, individuals in the highest decile of the TBR index have a more than 40% higher risk of all-cause mortality compared to their counterparts. Such an increased risk was independent of several clinical features known to shape the rate of mortality, including age, sex, cigarette smoking, HbA1c, and ongoing treatments of hyperglycemia, hypertension, and dyslipidemia. It is important to emphasize that in all 3 cohorts, we observed a direct significant association between BMI and all-cause mortality (data not shown). This therefore makes it implausible that the association between the TBR index and death we here report is secondary to the “obesity paradox” (30-32) that links low BMI to the increased risk of death in frail patients, such as those with heart and kidney failure.

As compared to participants in TBR deciles 1 to 9, individuals in decile 10 were characterized by atherogenic dyslipidemia, poor glycemic control, insulin resistance, and elevated serum levels of GGT and glutamate pyruvate transaminase (GPT), 2 reliable markers of metabolic-associated fatty liver disease (33), which are all established predictors of increased mortality rate (33-35). Importantly, the clinical features of individuals in TBR decile 10 closely resemble those with LD-like metabolic phenotype, previously identified either by genetic tools (2) or measurements of body fat composition (5).

Although our sequencing was limited to decile 10 patients and, therefore, only exploratory in nature, at least one of such individuals was affected by FPLD2. Whether this indicates that FPLD2 is more common in type 2 diabetes than currently believed (36, 37) is a possibility that deserves further studies, also in light of the fact that FPLD patients can be effectively treated with metraleptin, a leptin analog (38, 39).

In addition, our data are also consistent with a polygenic control explained by a 20-SNP GRS, previously linked to an LD cluster (2), which explains a portion of the TBR variance similar to that controlled by common SNPs in many complex traits (40).

We would like to highlight some important aspects of our results. First, the association between the LD-like metabolic phenotype and all-cause mortality has never been reported until now, which makes our finding novel. Second, this association was statistically significant in 2 independent samples and in a meta-analysis including also a third independent sample, thus making the risk of a false-positive finding unlikely. Third, the TBR index is inexpensive, widely available, and easy to use and therefore paves the way for precision medicine approaches, including stratified follow-up and targeted treatments, even in low- and middle-income settings where laborious and expensive models are difficult to apply and health disparities are exacerbated (41).

In conclusion, our finding, obtained using a novel, inexpensive and widely available marker, indicate that a nontrivial proportion of individuals with type 2 diabetes characterized by a LD-like metabolic phenotype have a more severe form of the disease and are at increased risk of death. Integrating this new piece of knowledge into clinical practice may help refine diagnostic and follow-up algorithms and enable specific treatments aimed at improving long-term outcomes in LD-like individuals.

Abbreviations

ACCORD

Action to Control Cardiovascular Risk in Diabetes

aGMS

aggregated Gargano Mortality Study

BMI

body mass index

BP

blood pressure

DXA

dual-energy x-ray absorptiometry

FPLD

familial partial lipodystrophy

GGT

gamma-glutamyl transferase

GRS

genotype risk score

HbA1c

glycated hemoglobin

HDL

high-density lipoprotein

HR

hazard ratio

IQR

interquartile range

LD

lipodystrophy

PMS

Pisa Mortality Study

SNP

single nucleotide polymorphism

TBR

triglyceride/BMI ratio

TyA1c

triglycerides-HbA1c calculated as Ln[(HbA1c × triglycerides/2]

Funding

This study was supported by Ministero dell'Istruzione, dell'Università e della Ricerca PNRR M4C2I1.3 Heal Italia project PE00000019 grants (to V.T.) and by Ministero della Salute grant RF-2013-02356459 (to C.M.); grant PNRR-MAD-2022-12375970 (CUP I23C22000720007 to S.D.C.) funded by the European Union—Next Generation EU—NRRP M6C2—Investment 2.1 Enhancement and strengthening of biomedical research in the NHS; grant PNRR-MAD-2022-12376543 (CUP I23C22000750001 to: R.D.P.) funded by the European Union—Next Generation EU—NRRP M6C2—Investment 2.1 Enhancement and strengthening of biomedical research in the NHS; Ministero della Salute, grant Ricerca corrente 2022-2024 (R.D.P., C.M. S.D.C, and V.T.); Ministero della Salute, grant Ricerca corrente 2025-2027 (R.D.P. and C.M.) and European Health Data & Evidence Network (EHDEN project, grant no. 806968) funded by Innovative Medicines Initiative 2 Joint Undertaking to C.M. Financial resources for whole-exome sequencing were provided by an unrestricted grant from Chiesi Italia SPA (donation to Fondazione Casa Sollievo della Sofferenza, V.T.). The ACCORD clinical trial was funded to A.D. by the National Heart, Lung, and Blood Institute (N01-HC-95178, N01-HC-95179, N01-HC-95180, N01-HC-95181, N01-HC-95182, N01-HC-95183, N01-HC-95184, IAA-Y1-HC-9035, and IAA-Y1-HC-1010); the National Institute of Diabetes and Digestive and Kidney Diseases, the National Institute on Aging, and the National Eye Institute; by the Centers for Disease Control and Prevention; and by General Clinical Research Centers. The following companies provided study medications, equipment, or supplies to A.D.: Abbott Laboratories, Amylin Pharmaceutical, AstraZeneca, Bayer HealthCare, Closer Healthcare, GlaxoSmithKline, King Pharmaceuticals, Merck, Novartis, Novo Nordisk, Omron Healthcare, Sanofi-Aventis, and Schering-Plough.

Author contributions

R.D.P. and V.T. conceived the study, analyzed the data, interpreted the analyses, and wrote the manuscript. C.M., M.M., M.G., D.L., G.P., and S.D.C. contributed to data collection and interpretation of analysis results. P.F. interpreted the analyses and contributed to the manuscript writing. L.S. performed laboratory testing. A.D. contributed to data collection, analyzed the data, interpreted the results, and contributed to manuscript writing. All authors critically revised the paper and approved its final version. R.D.P. and V.T. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Disclosures

The authors have no conflicts of 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.