Endogenous Leptin Concentrations Poorly Predict Metreleptin Response in Patients With Partial Lipodystrophy

Abstract

Context

Leptin replacement with metreleptin improves glycemia and hypertriglyceridemia in severely hypoleptinemic patients with generalized lipodystrophy (GLD), but its effects are variable in partially leptin-deficient patients with partial lipodystrophy (PLD).

Objective

Compare 3 leptin assays (Study I); identify diagnostic performance of leptin assays to detect responders to metreleptin for each assay (Study II).

Design

Study I: cross-sectional analysis of average bias between leptin assays. Study II: retrospective analysis of diagnostic accuracy of potential leptin cut points to detect clinical responders to metreleptin.

Setting

National Institutes of Health; University of Michigan.

Participants and Interventions

Study I: Metreleptin-naïve patients with lipodystrophy (GLD, n = 33, PLD, n = 67) and healthy volunteers (n = 239). Study II: GLD (n = 66) and PLD (n = 84) patients treated with metreleptin for 12 months.

Outcome Measures

Leptin concentrations by Millipore radioimmunoassay (RIA), Millipore enzyme-linked immunosorbent assay (MELISA), and R&D Systems enzyme-linked immunosorbent assay (RDELISA). Response to metreleptin therapy was defined as either reduction ≥1.0% in A1c or ≥30% in serum triglycerides.

Results

RDELISA measured 3.0 ± 9.5 ng/mL higher than RIA; MELISA measured 11.0 ± 17.8 and 14.0 ±19.2 less than RIA and RDELISA, respectively. Leptin by RIA, MELISA, and RDELISA modestly predicted metreleptin response in GLD + PLD [receiver operating characteristic (ROC) area under the curve (AUC) 0.74, 0.69, and 0.71, respectively; P < 0.01 for all] with lower predictive power in PLD (ROC AUC 0.63, 0.61 and 0.65, respectively; P > 0.05 for all). The only reproducible cut point identified on sensitivity analyses was RIA leptin 7.2 ng/mL (sensitivity 56%; specificity 78%).

Conclusions

Three common leptin assays are not interchangeable, and a reliable cut point to select responders to metreleptin was not identified.

Leptin is a 146 amino acid peptide primarily produced by adipose tissue and involved in the regulation of energy homeostasis, glucose, and lipid metabolism (1). Low leptin acts as a starvation signal to the central nervous system, resulting in hyperphagia (2,3). Metreleptin is a recombinant analog of human leptin, differing from endogenous human leptin by an additional methionine residue at its amino terminus. Leptin replacement with metreleptin in patient groups with very low endogenous leptin, such as congenital leptin deficiency (4), has dramatic effects to reduce hyperphagia. However, metreleptin has little clinical effect in states of leptin excess, such as obesity (5).

Lipodystrophy syndromes are rare genetic or acquired disorders characterized by deficiency of adipose tissue (6,7), which may lead to leptin deficiency. Generalized lipodystrophy (GLD) is associated with deficiency of adipose tissue in nearly the entire body, resulting in very low serum leptin concentrations. In partial lipodystrophy (PLD) selected fat depots are absent, most commonly in the limbs and buttocks (6), and serum leptin ranges from low to high, in proportion to the amount of residual adipose tissue (8,9). In contrast to congenital leptin deficiency or nonsyndromic obesity, patients with lipodystrophy cannot store caloric surplus in adipose tissue, resulting in accumulation of lipids in ectopic sites, such as the liver and muscle. Ectopic lipid storage leads to severe insulin resistance (1,10), which in turn contributes to diabetes, nonalcoholic fatty liver disease, hypertriglyceridemia, and hypertriglyceridemia-induced acute pancreatitis (10-13).

Metreleptin administration in patients with lipodystrophy can suppress hyperphagia (2,14) and lead to substantial improvement of metabolic abnormalities, including hypertriglyceridemia, hyperglycemia, and insulin resistance (14,15). In GLD, a dramatic decrease in both hypertriglyceridemia and hyperglycemia, as measured by hemoglobin A1c (A1c) after metreleptin treatment has been reported in several studies (16-19). In PLD, treatment with metreleptin has been associated with a smaller reduction in triglycerides, and conflicting findings have been reported regarding its effects on glycemic control (19-22).

A prior study from our group indicated a significant beneficial effect of metreleptin in reducing triglycerides and A1c in patients with PLD, but only in the subgroup with more severe leptin deficiency, who had pretreatment endogenous leptin [measured by Millipore radioimmunoassay (RIA)] <4 ng/mL (9). However, the sensitivity and specificity of this cut point, or other potential cut points, to predict metabolic improvement in patients with PLD after metreleptin administration was not assessed. Furthermore, the generalizability of this cut point to other leptin assays is not known.

The present study had 2 main goals. First, we evaluated the comparability of endogenous leptin levels measured by 3 widely used assays, including the RIA used in the prior study, and 2 commonly used enzyme-linked immunosorbent assays (ELISA) (Study I). Second, we aimed to identify a reliable cut point of serum leptin by each of these assays that was predictive of clinically meaningful metabolic improvement after metreleptin administration in patients with lipodystrophy (Study II).

Methods

These studies were retrospective analyses of participants enrolled in several prospective trials or observational studies of patients with lipodystrophy and healthy volunteers. Patients with a diagnosis of genetic or acquired, non-HIV related lipodystrophy were included from 3 studies conducted at the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) and 2 studies conducted at the University of Michigan between 2000 and 2018. All protocols were approved by the appropriate Institutional Review Board. Written informed consent was obtained from patients or their legal guardians. Assent was obtained from participants under 18 years of age. Patients had consented to use of their data and samples for future research. One of the studies conducted at the NIDDK was a prospective observational study evaluating the natural history of insulin resistance (NCT00001987). The additional 2 studies conducted at the NIDDK (NCT00025883, NCT01778556) and those conducted at the University of Michigan (NCT00677313, NCT01679197) were prospective, single-arm open-label studies assessing the efficacy and safety of metreleptin treatment in patients with lipodystrophy. Healthy subjects with a wide range of adiposity (hereafter referred to as “controls”) and no known comorbidities were selected from a prospective observational study conducted at the National Institutes of Health (NIH) and evaluating the phenotype of overweight and obese individuals (NCT00428987). Due to the retrospective nature of the study, the sample size was determined by data and sample availability. All NIH subjects with serum leptin measured with at least 2 different assays were included in Study I.

Patients with lipodystrophy who were treated with metreleptin had a median follow-up of 12 months (range: 5.8-19.7). For all other subjects (controls and non-metreleptin treated patients), data from a single, baseline visit were included. Triglycerides and A1c were collected after an 8- to 12-h fast and measured using standard methodology, both at baseline and at the end of follow-up after treatment with metreleptin.

Measurement of Serum Leptin

Leptin concentrations were measured using commercial RIA and ELISA kits. Serum samples were drawn after an 8- to 12-h fast and stored at −80°C until assays were performed, and the number of freeze/thaw cycles was <5 for all samples. Measurements were performed either as a part of the original study protocol or in a post hoc manner depending on sample availability. For samples collected from patients with lipodystrophy at the NIH, RIA was run in a relatively short time frame (<1 year) after collection, whereas ELISA assays were run up to 15 years after collection. For samples collected from patients with lipodystrophy at the University of Michigan, Millipore ELISA (MELISA) was run in a relatively short time frame (<1 year) after collection, whereas the RIA and R&D Systems ELISA (RDELISA) were run up to 6 years after collection. For the samples collected from control subjects at the NIH, RIA and ELISA assays were run up to 10 years after collection. All samples were run in duplicate, with the mean of replicates used for analysis. The RIA [EMD Millipore, catalog no. HL-81HK, RRID AB_2894698 (https://antibodyregistry.org/search.php?q=AB_2894698) assay sensitivity: 0.78 ng/mL, intra- and interassay coefficients of variability (CV): 9.29% and 9.96%, respectively] (23), MELISA (EMD Millipore, catalog no. EZHL80SK, RRID AB_2894697 (https://antibodyregistry.org/search.php?q=AB_2894697) assay sensitivity: 0.2 ng/mL, intra- and interassay CVs: 3.89% and 4.76%, respectively), and RDELISA (R&D Systems, catalog no. DLP00, RRID AB_2783014 (https://antibodyregistry.org/search?q=AB_2783014), assay sensitivity: 0.015 ng/mL, intra- and interassay CVs: 1.34% and 1.58%, respectively) were performed per manufacturers’ instructions.

Measurement of Percentage Body Fat

Total tissue percentage fat mass (FM%) was assessed using whole body dual energy X-ray absorptiometry (DXA; Hologic QDR 4500, Hologic, Bedford, MA, USA, or Lunar iDXA or Prodigy, GE Healthcare, Madison WI, USA). FM% data by GE Lunar DXA were converted to be equivalent to those measured by Hologic DXA as previously described (FM%_Hologic = 1.943 + 0.894 * FM%_GE) (24).

Statistical Analyses

Effect of sample storage duration

Leptin values were nonnormally distributed and were log-transformed prior to analyses. The association between sample storage duration and the measured leptin was evaluated with simple linear regression. Data from subjects with lipodystrophy and controls were analyzed separately, and data from only 1 site (NIH) was used to avoid confounding effects of differing proportions of subjects with GLD vs PLD at different sites. GraphPad Prism v9.1 (LaJolla, CA, USA) was used for analyses.

Study I: agreement between different leptin assays

We assessed the relationship among serum leptin measured by three assays (RIA, MELISA, and RDELISA) using data from metreleptin-naïve patients with lipodystrophy and nonlipodystrophic controls enrolled in studies at the NIH. Subjects who had serum leptin levels measured using least 2 of the assays of interest were included. Leptin was log-transformed prior to analyses. A linear mixed-effect model was used to compare leptin measured by the three assays. Linear regression was used to analyze the relationship between FM% and leptin in all subjects. Interassay correlations were assessed with Pearson correlation, and Bland-Altman analysis was used to determine the average bias. GraphPad Prism v9.1 was used for analyses of Study I.

Study II. Identification of serum leptin cut points to predict response to metreleptin

Patients with lipodystrophy who were treated with metreleptin as a part of their respective studies and were usually followed up for either 6 or 12 months. Details of metreleptin dosing and the primary outcomes of the studies are reported by Brown et al (25,26). and Oral et al (27). For this analysis, subjects were categorized as responders or nonresponders based on the achievement of a reduction of either A1c ≥1.0% or triglycerides ≥30% of baseline value at the end of follow-up. Those with at least 6 months of follow-up were included in the analyses. Patients with both A1c < 6.5% and triglycerides < 150 mg/dL prior to metreleptin administration were excluded. Between group differences between responders and nonresponders were assessed using Wilcoxon rank-sum test. The usefulness of RIA, ELISA, and RDELISA assays for discriminating responders from nonresponders was assessed as area under the curve (AUC) using receiver operator characteristic (ROC) curve analyses. The ability of body mass index (BMI) and FM% to predict metreleptin response was assessed in the same manner. Optimal leptin cut points were determined using Youden’s index (28). Multiple cut points are reported in cases where multiple peaks of Youden’s J existed. Cut points were calculated for each assay type for all patients (GLD + PLD), and in the subgroup of only PLD.

To assess the reproducibility of cut points for each assay, sensitivity analyses were conducted after systematic or random removal of data. Systematic removal was performed as follows: for each separate cohort included in this study, we repetitively removed its participants and recalculated the cut point. Random removal was performed as follows: the data were block-randomized into 5 subsets (denoted as random set 1 through 5). For each subset, we repetitively removed the subset and recalculated the cut point. The block-randomization parameters were as follows: lipodystrophy subtype (GLD or PLD), presence of familial PLD type 2 (Dunnigan variety, confirmed by genetic testing for pathogenic LMNA variants), sex, and responder status. We hypothesized that if a biologically valid cut point exists, random removal of data should not change the results.

Post hoc analyses were performed to assess whether the a priori definition of responders had an effect on our results. A1c reduction and triglycerides percent reduction targets were tuned across a wide range of values (between −2.0% to −0.1% with increments of 0.1% for A1c and between −80% to −5% with increments of 5% for triglyceride percent change) to select the responder definition that yields the highest average AUC for all 3 assays. Sensitivity analyses were repeated as appropriate.

Analyses and calculations for study II were performed using R v3.6.1 (Vienna, Austria) (29). Cut points were estimated using OptimalCutpoints package for R v1.1.4 (30). ROC analyses were performed using e1071 package v1.7.3 (31). Block randomization was performed using randomizer package v0.20.0 (32).

Results

Effect of Sample Storage Duration on Leptin Measurements

Samples from patients with lipodystrophy were stored for a median 134 days (range: 18-349), 1229 days (range: 27-2337), and 1260 days (range: 27-2360) for RIA, MELISA, and RDELISA, respectively. Samples from controls were stored for a median 2688 days (range: 401-3582), 2412 days (range: 100-3581), and 2709 days (range: 422-3604) for RIA, MELISA, and RDELISA, respectively. Sample storage duration did not correlate with measured leptin levels by any assay in either cohort (Fig. 1).

Figure 1.

Effect of sample storage duration on the measured leptin levels. Abbreviations: RIA, Millipore radioimmunoassay; MELISA, Millipore enzyme-linked immunosorbent assay; RDELISA, R&D Systems enzyme-linked immunosorbent assay.

Study I: agreement between leptin measured by RIA, MELISA, and RDELISA assays

A total of 100 metreleptin-naïve patients with lipodystrophy, including 33 with GLD (23 congenital, 6 acquired, 4 atypical progeroid; 24 female, 9 male; age median 13, range 0-44 years) and 67 with PLD (59 familial partial, 6 acquired partial; 59 female, 8 male; age median 35, range 10-67 years) were included along with 239 controls (153 female, 86 male; age median 38, range 18-70 years). All 3 assays were performed on all subjects except for 12 who had missing RDELISA measurements due to lack of stored samples. Baseline characteristics of the study cohort are presented in Table 1. Interassay differences were observed among controls (P < 0.001) but not among patients with lipodystrophy (P = 0.11) (Fig. 2A). Serum leptin by all assays correlated with FM% with the strongest association observed between FM% and RIA leptin (Fig. 2B).

Table 1.

Descriptive statistics of subjects included in Study I

	Lipodystrophy	Controls	P-value
N, total	100 (33 GLD, 67 PLD)	239
Sex	83F, 17M	153F, 86M	<0.001a
Age, years	23 (15, 43)	38 (28, 50)	<0.001b
Race
White	77	130
Black or African American	11	89
Asian	7	11	<0.001a
American Indian or Alaskan Native	0	1
Multiple races	5	6
Other or unknown	0	2
Ethnicity
Hispanic or Latino	18	27	0.11a
BMI, kg/m2	23.7 (19.3: 27.4)	34.0 (23.6: 41.3)	<0.001b
DXA total tissue percentage fat, %	22.9 (15.7: 27.9)	40.5 (27.4: 48.0)	<0.001b
RIA leptin, ng/mL	4.5 (1.7: 10.1)	32.2 (10.1: 53.7)	<0.001b
MELISA leptin, ng/mL	4.8 (1.1: 9.8)	17.8 (8.7: 28.2)	<0.001b
RDELISA leptin, ng/mL	6.3 (1.3: 13.1)	31.0 (11.2: 60.5)	<0.001b

Data presented as n or median (interquartile range).

Abbreviations: BMI, body mass index; DXA, dual energy X-ray absorptiometry; GLD, generalized lipodystrophy; MELISA, Millipore enzyme-linked immunosorbent assay; PLD, partial lipodystrophy; RDELISA, R&D Systems enzyme-linked immunosorbent assay; RIA, radioimmunoassay.

^aFisher’s exact test.

^bWilcoxon rank sum test.

Figure 2.

Study I: Agreement between commercial leptin assays. (A) Serum leptin measured by Millipore radioimmunoassay, Millipore enzyme-linked immunosorbent assay, and R&D Systems enzyme-linked immunosorbent assay in patients with lipodystrophy (red circles, generalized lipodystrophy; blue circles, partial lipodystrophy) and controls without lipodystrophy (gray circles) and (B) their correlation with tissue %fat measured by dual energy X-ray absorptiometry. (C-E) Scatterplots comparing the 3 assays of interest (insets show the subset of data in subjects with leptin levels <20 ng/mL) and (F-H) Bland-Altman plots showing the average bias between the assays. Abbreviations: DXA, dual-energy X-ray absorptiometry; LD, lipodystrophy; MELISA, Millipore enzyme-linked immunosorbent assay; RIA, Millipore radioimmunoassay; RDELISA, R&D Systems enzyme-linked immunosorbent assay.

Leptin measurements by the 3 assays were all highly correlated [R² for all subjects (lipodystrophy + controls): RIA vs MELISA, 0.76; RIA vs RDELISA, 0.77; MELISA vs RDELISA, 0.83; P < 0.001 for all]. Slightly weaker correlations were observed when the analysis was restricted to only the subjects with RIA < 20 ng/mL [R² for lipodystrophy + controls: RIA vs MELISA, 0.59 (Fig. 2C); RIA vs RDELISA, 0.56 (Fig. 2D); MELISA vs RDELISA, 0.83 (Fig. 2E); P < 0.001 for all]. Bland-Altman analysis showed that, on average, MELISA measured 11.0 ± 17.8 less than RIA (Fig. 2F), RDELISA measured 3.0 ± 9.5 ng/mL more than RIA (Fig. 2G), and MELISA measured 14.0 ± 19.2 ng/mL less than RIA (Fig. 2H). The observed bias was larger for higher leptin levels and smaller for lower leptin levels.

Study II: identification of endogenous leptin cut points to predict response to metreleptin

Of the 190 patients with lipodystrophy available, 150 (66 GLD, 84 PLD) were treated with metreleptin and had the required follow-up data. Of these, 7 (4 GLD, 3 PLD) were excluded from analyses due to normal baseline metabolic parameters (A1c < 6.5%; triglycerides <150 mg/dL). Responders to metreleptin had higher baseline A1c and triglycerides compared to nonresponders (Table 2). When the entire lipodystrophy population was considered (GLD + PLD), nonresponders had a significantly higher baseline leptin compared to responders (RIA leptin: 8.4 vs 1.5 ng/mL, P < 0.001; MELISA leptin: 12.1 vs 5.0 ng/mL, P = 0.001; RDELISA leptin: 13.9 vs 5.6 ng/mL, P = 0.003 for nonresponders vs responders, respectively) (Fig. 3A-C). However, this pattern was driven by the GLD population and was not statistically significant in the PLD subgroup alone (RIA leptin: 12.3 vs 6.7 ng/mL, P = 0.08; MELISA leptin: 12.9 vs 9.7 ng/mL, P = 0.12; RDELISA leptin: 16.0 vs 9.6 ng/mL, P = 0.06; for nonresponders vs responders with PLD, respectively) (Fig. 3D-F). ROC analyses showed that serum leptin was a statistically significant predictor of clinical response for GLD + PLD [AUCs for RIA leptin 0.74 (95% CI 0.64-0.84) (Fig. 3G), MELISA leptin 0.69 (95% CI 0.58-0.80) (Fig. 3H), RDELISA leptin 0.71 (95% CI 0.57-0.84) (Fig. 3I)]. However, the same was not true in the PLD subgroup only [AUCs for RIA leptin 0.63 (95% CI 0.49-0.77) (Fig. 3G), MELISA leptin 0.61 (95% CI 0.48-0.74) (Fig. 3H), and RDELISA leptin 0.65 (95% CI 0.49-0.80) (Fig. 3I)]. BMI and FM% had comparable performance to serum leptin in predicting response to metreleptin (Figs. 3J and 3K).

Table 2.

Descriptive statistics of subjects included in Study II

	GLD + PLD			PLD
	Nonresponder	responder	P	Nonresponder	responder	P
n	45	98		39	42
Sex	41F, 4M	82F, 16M	0.30a	36F, 3M	39F, 3M	1.00a
Age, years	34 (18, 47)	20 (15, 41)	0.067b	38 (28, 54)	41 (28, 52)	0.59b
Race
American Indian or Alaskan Native	0	1		0	0
Asian	2	6		1	1
Black or African American	4	13	0.81a	2	0	0.41a
White	35	73		32	39
Multiple, other or unknown	4	5		4	2
Ethnicity
Hispanic or Latino	2	24	0.14a	2	6	0.26a
Study
Michigan, FHA101	11	7		11	7
Michigan, NASH4	10	9	<0.01	10	9	0.33
NIH, generalized	6	56		0	0
NIH, partial	18	26		18	26
Follow-up duration, months	12 (12, 12)	12 (12, 13)	0.78b	12 (12, 12)	12 (12, 13)	0.025b
BMI, kg/m2)	27.6 (21.9, 30.5)	23.0 (19.3, 25.4)	<0.001b	27.8 (23.8, 31)	25.1 (23.3, 30.5)	0.16b
DXA total body fat, %	28.0 (22.8, 32.9)	16.3 (12.9, 25.1)	<0.001b	28.6 (23.5, 33)	25.3 (21.6, 33.5)	0.095b
Hemoglobin A1c, %	7.0 (5.8, 8.4)	9.2 (8.0, 10.4)	<0.001b	7.3 (6.0, 8.5)	9.1 (8.0, 10.3)	<0.001b
Δ Hemoglobin A1c, %	0.1 (−0.2, 0.4)	−2.0 (−3.2, −0.6)	<0.001b	0.1 (−0.2, 0.4)	−1.0 (−2.3, −0.2)	<0.001b
Triglycerides, mg/dL	261 (208, 402)	513 (274, 1347)	<0.001b	263 (218, 403)	591 (269, 1769)	0.001b
Δ Triglycerides, mg/dL	0 (−38, 190)	−283 (−1143, −100)	<0.001b	−10 (−50, 191)	−301 (−1371, −100)	<0.001b
Fasting glucose, mg/dL	107 (86, 140)	185 (137, 245)	<0.001b	112 (96, 158)	177 (147, 242)	<0.001b
Insulin, mcU/mL	41.0 (12.1, 73.0)	43.0 (22.1, 98.2)	0.18b	22.9 (11.6, 64.9)	34.0 (14.1, 52.5)	0.69b
C-peptide, ng/mL	3.8 (2.0, 5.9)	4.2 (3.0, 5.8)	0.46b	3.6 (1.7, 4.6)	4.2 (2.6, 5.1)	0.39b
RIA leptin, ng/mL	8.4 (3.3, 26.8)	1.5 (1.0, 4.5)	<0.001b	12.3 (7.3, 34.6)	6.7 (3.9, 17.9)	0.084b
MELISA leptin, ng/mL	12.1 (6.8, 22.5)	5.0 (1.1, 11.2)	0.001b	12.9 (7.2, 23)	9.7 (5.5, 15.8)	0.11b
RDELISA leptin, ng/mL	13.9 (6.5, 28.5)	5.6 (1.1, 10.5)	0.003b	16.0 (7.6, 29.8)	9.6 (6.3, 15.1)	0.06b

Data presented as either n or median (interquartile range).

Abbreviations: BMI, body mass index; DXA, dual energy X-ray absorptiometry; GLD, generalized lipodystrophy; MELISA, Millipore enzyme-linked immunosorbent assay; PLD, partial lipodystrophy; RDELISA, R&D Systems enzyme-linked immunosorbent assay; RIA, radioimmunoassay.

^aFisher’s exact test.

^bWilcoxon rank sum test.

Figure 3.

Study II: Performance of baseline serum leptin levels as predictors of treatment response in patients with lipodystrophy treated with metreleptin. Baseline serum leptin differed between responders and nonresponders in the entire lipodystrophy population [generalized lipodystrophy (GLD) + partial lipodystrophy population (PLD)], measured by (A) Millipore radioimmunoassay (RIA), (B) Millipore enzyme-linked immunosorbent assay (MELISA), and (C) R&D Systems enzyme-linked immunosorbent assay (RDELISA). Red circles: GLD; blue circles: PLD. Baseline serum leptin did not differ between responders and nonresponders in the PLD only, measured by (D) RIA, (E) MELISA, and (F) RDELISA. Receiver operating characteristic (ROC) curves for baseline leptin measured by (G) RIA (H) MELISA, and (I) RDELISA. (J) Body mass index and (K) total tissue fat mass% ROC curves were comparable to baseline serum leptin. (L) Baseline leptin by RIA correlated with absolute reduction in hemoglobin A1c in the entire lipodystrophy population (GLD + PLD, black line), but this association was not present for either GLD or PLD alone (red and blue lines, respectively). Dashed black line in panels (A) and (D) show the RIA cut point of 7.2 ng/mL, which had the highest sensitivity and specificity. Abbreviations: AUC, area under the curve; BMI, body mass index; FM%, total body tissue %fat measured by dual-energy X-ray absorptiometry; LD, lipodystrophy; MELISA, Millipore enzyme-linked immunosorbent assay; NR, nonresponder; RDELISA, R&D Systems enzyme-linked immunosorbent assay; R, responder; RIA, Millipore radioimmunoassay; ROC, receiving operator characteristic.

Cut points were calculated to detect responders to metreleptin. RIA leptin = 7.2 ng/mL was identified as a cut point for both GLD+PLD and the PLD subgroups, with accuracy ranging between 77% and 78% for GLD + PLD and 64% and 66% for PLD. Although this cut point had a rather low accuracy, especially for the PLD subgroup (sensitivity = 56%, specificity = 78%), it remained relatively stable in sensitivity analyses (Table 3). The 2 ELISA assays both failed to converge at a specific cut point, and their accuracies ranged between 61% and 73% for GLD + PLD and 55 to 69% for PLD. A BMI cut point of 26.4 kg/m² was identified in both GLD + PLD and PLD subgroups. FM% produced a cut point of 20.2% for GLD + PLD and 26.9% for PLD. Accuracies were similar for both BMI and FM% and ranged from 70% to 75% for GLD + PLD and 5% to 65% for PLD.

Table 3.

Sensitivity analyses of leptin cut points to identify metreleptin responders after systematic (study names) or random removal of data

	Generalized + Partial lipodystrophy				Partial lipodystrophy only
Excluded set	n	Cut point	AUC	Acc (Se/Sp), %	n	Cut point	AUC	Acc (Se/Sp), %
RIA leptin, ng/mL
None	123	7.2	0.74	78 (83/64)	61	7.2	0.63	66 (56/78)
FHA study	NAa				NAa
NASH study	106	7.2	0.72	78 (83/64)	44	7.2	0.64	66 (56/78)
NIH, GLD	61	7.2	0.63	78 (83/64)	NAa
NIH, PLD	79	12.2	0.72	77 (89/45)	17	24.6	0.51	64 (85/37)
Random set 1	101	7.2	0.78	78 (83/64)	51	7.2	0.69	66 (56/78)
Random set 2	98	7.2	0.73	78 (83/64)	49	7.2	0.58	66 (56/78)
Random set 3	99	7.2	0.71	78 (83/64)	49	7.2	0.62	66 (56/78)
Random set 4	100	7.4	0.71	77 (83/61)	50	7.4	0.59	64 (56/74)
Random set 5	94	7.2	0.77	78 (83/64)	45	7.2	0.66	66 (56/78)
MELISA leptin, ng/mL
None	96	5.1	0.69	64 (50/84)	74	20.8	0.61	62 (89/33)
FHA study	78	11.8	0.70	67 (76/53)	56	11.8	0.64	59 (63/56)
NASH study	77	5.1	0.67	64 (50/84)	55	5.1	0.56	55 (24/89)
NIH, GLD	74	20.8	0.61	69 (93/32)	NAa
NIH, PLD	59	4.4	0.76	61 (45/87)	37	23.8	0.60	59 (92/25)
Random set 1	78	5.1	0.69	64 (50/84)	60	20.8	0.61	62 (89/33)
Random set 2	72	17.9	0.65	68 (88/37)	58	17.9	0.56	61 (82/39)
Random set 3	78	6.3	0.71	64 (52/82)	59	11.8	0.63	59 (63/56)
Random set 4	79	5.1	0.66	64 (50/84)	61	11.8	0.58	59 (63/56)
Random set 5	77	5.1	0.74	64 (50/84)	58	20.8	0.65	62 (89/33)
RDELISA leptin, ng/mL
None	75	11.0	0.71	73 (80/62)	54	11 or 17.9b	0.65	67 (67/67) or 69 (83/50)
FHA study	NAa				NAa
NASH study	58	6.4	0.66	64 (57/77)	37	17.9	0.59	69 (83/50)
NIH, GLD	54	11.0 or 17.9b	0.65	73 (80/62) or 75 (90/46)	NAa
NIH, PLD	38	13.7	0.78	72 (82/54)	17	26.6	0.79	65 (87/38)
Random set 1	63	11.0	0.70	73 (80/62)	45	11.0	0.64	67 (67/67)
Random set 2	55	17.9	0.68	75 (90/46)	42	17.9	0.61	69 (83/50)
Random set 3	63	10.5	0.70	71 (73/65)	45	10.5	0.66	63 (57/71)
Random set 4	61	12.7	0.66	73 (82/58)	44	17.9	0.61	69 (83/50)
Random set 5	58	11.0	0.79	73 (80/62)	40	24.4	0.73	69 (87/46)
BMI, kg/m2c
None	143	26.4	0.68	75 (82/60)	81	26.4	0.59	65 (62/69)
Fat mass %c
None	123	20.2	0.78	71 (60/95)	77	26.9	0.61	61 (59/64)

Random sets 1 through 5 represent one fifth of the data, block randomized according to sex, GLD or PLD subtype, presence of FPLD2 (Dunnigan variety) and responder status (the variability in n in each row is due to missing samples). Optimal cut points were calculated with Youden’s J and are reported as baseline serum leptin (ng/mL).

Abbreviations: AUC, area under the curve of receiver operating characteristic curve; BMI, body mass index; FHA study, partial lipodystrophy patients from the FHA101 study at Michigan; GLD, generalized lipodystrophy; MELISA, leptin measured by Millipore enzyme-linked immunosorbent assay; NA, not applicable; NASH study, partial lipodystrophy patients from the NASH4 study at Michigan; NIH, National Institutes of Health; PL, partial lipodystrophy; RDELISA, leptin measured by R&D Systems enzyme-linked immunosorbent assay; RIA, leptin measured by radioimmunoassay (Millipore); Se, sensitivity; Sp, specificity.

^aNothing to exclude as there were no relevant measurements in the excluded set.

^bMultiple cut points are reported where Youden’s J had multiple peaks.

^cSensitivity analyses were omitted for brevity purposes but showed stable results.

Baseline leptin did correlate with absolute reduction in A1c in the GLD + PLD cohort (R² = 0.18, 0.23, and 0.21 for RIA, MELISA, and RDELISA, respectively; P < 0.001 for all) (Fig. 3L), but this correlation was weaker in the PLD population alone and significance varied by leptin assay (R² = 0.05, 0.06, 0.01 and P = 0.08, 0.035, and 0.46 for RIA, MELISA, and RDELISA, respectively). Absolute or percentage change in triglycerides did not correlate with baseline leptin (P > 0.05 for all).

Post hoc tuning of responder definitions identified ΔA1c ≥ -0.4% and triglyceride percent change ≥−50% as the thresholds that yielded the highest average AUC for identifying responders with PLD. The tuned thresholds identified 42 responders vs 39 nonresponders compared to the original 39 responders vs 42 nonresponders. After tuning, the AUCs increased from 0.63 to 0.70 for RIA, 0.61 to 0.72 for MELISA, and 0.65 to 0.73 for RDELISA. The cut points identified with the tuned responder definition were 19.6 ng/mL for RIA, 11.8 ng/mL for MELISA, and 17.9 ng/mL for RDELISA, none of which were found to be stable after repeated sensitivity analyses. Accuracy ranged between 57% and 72%.

Discussion

Treatment with metreleptin has been shown to improve metabolic complications of lipodystrophy in patients with GLD, as well as a subgroup of patients with PLD with hypoleptinemia and severe metabolic derangements (9). Based on these findings, a recommendation to consider experimental treatment with metreleptin in patients with PLD who have hypoleptinemia (<4 ng/mL) and severe metabolic derangements (A1c > 8% and/or triglycerides > 500 mg/dL) was endorsed in the Multi-Society Practice Guideline on the diagnosis and management of lipodystrophy syndromes (Class IIb, Level B) (11). Identification of a cut point of endogenous leptin that accurately identifies patients who are likely to respond to metreleptin has the potential to aid in precision medicine, resulting in prevention and/or significant improvement of severe complications, reduction of mortality, and better quality of life.

Identification of a cut point with high sensitivity and specificity to predict metreleptin response first requires an understanding of differences among leptin assays, which have not been standardized among laboratories. In the past, RIA was the most common method employed to measure blood leptin concentrations, including in the studies that led to the development of the practice guideline (9), but ELISA assays are now more commonly used. The RIA method employs radioactively labeled leptin molecules, which compete for leptin-specific antibodies with the leptin molecules in the substrate. The leptin level in the substrate is estimated from a standard curve based on the signal from the gamma counter (33). Both ELISA methods are direct ELISAs in which the leptin molecule is sandwiched between an antileptin antibody coating the sample well and another antileptin antibody with enzymatic activity (34,35). In the present study, we demonstrated that three commonly used leptin assays do differ, and these differences are magnified at higher leptin concentrations. The observed difference in absolute leptin levels suggest that different cut points predictive of response to metreleptin may be required for each assay, and clinicians and investigators alike must practice caution when comparing results from different assays.

Attempts to identify cut points predictive for metreleptin response using ROC analyses yielded low sensitivity and specificity. Although the low AUCs obtained with ROC analyses suggested that any cut points thus identified would have low sensitivity and specificity, we proceeded to calculate cut points and used sensitivity analyses to demonstrate that the cut points determined for ELISA assays were highly sensitive to small changes in the study population. The potential cut point of 7.2 ng/mL identified for RIA remained stable across sensitivity analyses but still demonstrated poor sensitivity and specificity to predict response to metreleptin. Although the existence of such a cut point may hint at the existence of a biological leptin threshold that limits the efficacy of leptin replacement therapy, the clinical applicability of such a cut point remains questionable given the poor performance in selecting responders, especially in the PLD subpopulation, the group in which a leptin cut point is most needed. Any near-useful result obtained in these analyses were driven by the GLD subpopulation, almost all of whom responded to metreleptin therapy. The 6 nonresponders with GLD had relatively low baseline A1c (range: 4.5%-7.2%) and variable triglycerides (range: 186-1296 mg/dL) and had serum leptin concentrations in the typical range for GLD (range 0.8-2.4 per RIA). Thus, there is no obvious biological reason why these 6 subjects did not respond to metreleptin. Within the PLD subpopulation alone, where the discovery of an effective cut point may have had clinical utility, any attempt to identify such a cut point proved futile.

To define the populations of metreleptin responders vs nonresponders in this study, we chose thresholds of ≥1 percentage point A1c reduction or ≥30% reduction in triglycerides. A1c reductions of 0.5 percentage points are generally considered clinically meaningful in both the overall and PLD population (27,36). There is little evidence regarding what change in triglycerides should be considered clinically relevant, but an 18.3% reduction in triglycerides was associated with decreased cardiovascular events in a trial of purified eicosopentanoic acid (37). Higher thresholds to define response to metreleptin were chosen for treatment of an orphan condition in light of the availability of approved drugs for diabetes and hypertriglyceridemia in our a priori defined analyses. We further assessed a broad range of responder definitions to determine if a different definition would improve the diagnostic accuracy of endogenous leptin to predict metreleptin response. Although the AUCs did improve slightly with thresholds of A1c = −0.4% and triglyceride % change = −50%, better cut points were not produced and the accuracy measures remained poor. Overall, these findings did not support clinical use of an endogenous leptin cut point to predict metreleptin response in patients with lipodystrophy.

The absence of a leptin cut point to identify clinical responders to metreleptin does not negate the idea that lower leptin levels are associated with greater clinical response. It is noteworthy that all patients with GLD had low baseline leptin, and almost all (90% CI: 80%-96%) responded to metreleptin (Figs. 3A-3C). Furthermore, in the entire cohort with GLD + PLD, lower baseline serum leptin was significantly associated with greater reduction in A1c after metreleptin (Fig. 1L). However, baseline serum leptin accounted for only ~20% of the variability in A1c reduction. The remaining variability in metreleptin response may be explained by biological factors not related to leptin, such as variable adherence to diet or concomitant medications.

A limitation of this study was the use of a single, fasting value of serum leptin to predict metreleptin response. Poor power of a single day’s measurement of serum leptin to predict response to metreleptin may also be caused by intrasubject variability in serum leptin. Serum leptin is largely regulated by fat mass (38,39), with modest diurnal variation (40,41). As all assays were performed on fasting morning samples, this diurnal variability is not likely a factor in the current analyses. However, in addition to reflecting long-term energy stores, circulating leptin concentration is affected over a period of days by acute energy balance, with overfeeding leading to higher leptin (42), and fasting leading to lower leptin (43), out of proportion to changes in adipose tissue mass. Other factors that impact circulating leptin levels include dietary macrocomposition with some fatty acids and ketones causing an acute increase in circulating leptin and high carbohydrate load reducing circulating leptin. Iron stores and acute changes in physical activity can also modify circulating leptin levels (44). Finally, factors that impact soluble leptin receptor levels (eg, sex, adiposity, sex steroid administration, and albumin) can modify the total and free leptin levels (45). As we could not control for all of these factors at the time leptin levels were obtained, we cannot exclude that acute positive or negative energy balance or some of the additional factors might have contributed to the poor predictive power of serum leptin for clinical response to metreleptin. Repeated leptin levels obtained over the course of several days or weeks during known energy balance with simultaneously collected levels of circulating free fatty acid species and other metabolites might allow greater predictive power for metreleptin response. Another limitation is that some patients had reductions in concomitant medications that may have resulted in smaller reductions in A1c or triglycerides and, hence, led to misclassification as a nonresponder, instead of a responder. However, this is less likely as the majority of patients with reduction in medication use had GLD, and almost all GLD patients were classified as responders (25).

Conclusion

In conclusion, our data do not support a specific threshold of leptin in patients with PLD that is a good predictor of clinical response to metreleptin therapy. There is clear biologic evidence that patients with very low leptin (congenital leptin deficiency and GLD) do exhibit biological responses to metreleptin, while those with very high leptin (obesity) do not. Predicting whether patients with intermediate levels of leptin (such as PLD) will respond to metreleptin remains a clinical challenge. The inability to determine useful cut points to select patients with PLD who will respond to metreleptin might be attributable to factors other than metreleptin directly driving the changes in A1c and triglycerides in the open-label studies examined here. For example, the Hawthorne effect (ie, diet alone, better adherence to standard therapy during the trial, etc) might have been a major driver of clinical improvements. Alternative mechanisms such as metreleptin competing for endogenous ligand and working as a partial antagonist or its ability to modify the soluble leptin receptor thus changing free leptin exposure may also be considered. Lastly, there may be effects of both endogenous and exogenous leptin in residual adipose tissue that have not been thoroughly investigated that may shed light on the clinical effects observed in some patients. To prove benefits associated with metreleptin therapy in patients with PLD, randomized clinical trials that adequately address these alternative mechanistic explanations may be necessary.

Additional Information

Disclosure Summary: Metreleptin for the clinical trials conducted at National Institutes of Health was provided by Amgen Pharmaceuticals, Amylin Pharmaceuticals, Bristol Myers Squibb/Astra Zeneca, and Aegerion Pharmaceuticals under a collaborative research agreement with the National Institute of Diabetes and Digestive and Kidney Diseases and also under an investigator-initiated study MTA to the University of Michigan. E.A.O. has received grant support from these entities and has consulted for them or served on Advisory Boards. E.A.O. is also an inventor on the patent that is for method of use of this drug or other leptin analogs in lipodystrophy and in nonalcoholic steatohepatitis.

Data Availability

Datasets analyzed and R code used to run the analyses during the current study are not publicly available but are available from the corresponding author upon reasonable request.