Glycation of human serum albumin impairs binding to the glucagon-like peptide-1 analogue liraglutide

Abstract

The long-acting glucagon-like peptide-1 analogue liraglutide has proven efficiency in the management of type 2 diabetes and also has beneficial effects on cardiovascular diseases. Liraglutide's protracted action highly depends on its capacity to bind to albumin via its palmitic acid part. However, in diabetes, albumin can undergo glycation, resulting in impaired drug binding. Our objective in this study was to assess the impact of human serum albumin (HSA) glycation on liraglutide affinity. Using fluorine labeling of the drug and ¹⁹F NMR, we determined HSA affinity for liraglutide in two glycated albumin models. We either glycated HSA in vitro by incubation with glucose (G25- or G100-HSA) or methylglyoxal (MGO-HSA) or purified in vivo glycated HSA from the plasma of diabetic patients with poor glycemic control. Nonglycated commercial HSA (G0-HSA) and HSA purified from plasma of healthy individuals served as controls. We found that glycation decreases affinity for liraglutide by 7-fold for G100-HSA and by 5-fold for MGO-HSA compared with G0-HSA. A similarly reduced affinity was observed for HSA purified from diabetic individuals compared with HSA from healthy individuals. Our results reveal that glycation significantly impairs HSA affinity to liraglutide and confirm that glycation contributes to liraglutide's variable therapeutic efficiency, depending on diabetes stage. Because diabetes is a progressive disease, the effect of glycated albumin on liraglutide affinity found here is important to consider when diabetes is managed with this drug.

Introduction

Diabetes is a metabolic disease that is widespread throughout the world, affecting more than 415 million adults in 2015 (1). Diabetes mellitus is characterized by chronic hyperglycemia. Various treatments can be administered, including rapid and long-acting insulin analogue injections, which are used to counteract insulin production deficiency. These insulin analogues include lispro, glargine, or detemir. Detemir (Levemir®) is carried in the blood by reversible binding to human serum albumin (HSA)⁴ through its myristic acid grafted to Lys^B29 (2). More recently, another long-acting drug, a glucagon-like peptide 1 analogue called liraglutide, was commercialized as Victoza® for the treatment of type 2 diabetes. Like detemir, liraglutide binds to HSA via a fatty acid chain. Glucagon-like peptide-1 (GLP-1) is a 30-amino acid incretin hormone secreted by intestinal epithelial endocrine L-cells by the differential processing of proglucagon in response to post-prandial hyperglycemia (3).

Liraglutide shares 97% homology with human GLP-1, which binds to and activates the GLP-1 receptor. It differs from endogenous GLP-1 by a one-amino acid substitution (K34R) and by the addition of a palmitic acid grafted to Lys²⁶ via a glutamate linker (Fig. 1a). Its slow absorption is allowed due to its transport by albumin and also due to its auto-association into oligomers (4–6). Moreover, this GLP-1 analogue undergoes much slower degradation by the enzyme dipeptidyl peptidase-4 than endogenous GLP-1, resulting in a long plasma half-life of 11–15 h. Consequently, liraglutide requires only once-daily subcutaneous injection. Like native GLP-1, liraglutide triggers insulin secretion, inhibits postprandial glucagon secretion, retards gastric emptying, and enhances satiety by acting on the brain (7, 8).

Figure 1.

Liraglutide structure (a) and fluorine labeling reaction (b).

Liraglutide is increasingly used for the treatment of type 2 diabetes due to its increased efficiency compared with other treatments in controlling and stabilizing glucose plasma level and because of its capacity to reduce both cardiovascular risk and HbA1c (hemoglobin A1C) level (9–13).

Nonetheless, diabetes mellitus remains a progressive disease with gradual complications; therefore, treatment has to be closely adapted to patient state evolution. The action of liraglutide is highly dependent on its binding capacity to albumin and to its progressive delivery. This dependent therapeutic action of the drug linked to albumin affinity has already been reported for warfarin. Although a genetic factor has been identified in 30% of cases, the other causes of variability in the therapeutic efficiency of warfarin among patients remain unknown (14–16). In the context of diabetes, hyperglycemia induces albumin biochemical integrity alterations caused by a non-enzymatic process called glycation (17). It has been reported that glycated albumin has a decreased affinity for some fatty acids, such as cis-parinaric acid (18). Therefore, in patients with high glycation ratio evaluated according to HbA1c levels (higher than 7%), the action of liraglutide may be compromised.

For diabetic patients, the level of glycated albumin can be 2–3-fold higher compared with in healthy people (19, 20). This increased glycation rate of albumin could explain the difference in efficiency of fatty acid–grafted drugs observed in some patients suffering from advanced forms of diabetes (21). In this work, ¹⁹F NMR transverse relaxation time experiments were used to investigate how albumin glycation influences its binding properties to liraglutide. Fluorine NMR is one of the best suited techniques for probing protein-ligand interactions (22–27). This method requires either the use of a naturally fluorinated ligand or fluorine labeling of the protein (26), ligand, or spy molecules (24). Here, we used a strategy where the peptidic liraglutide ligand was first labeled with a fluorine probe by trifluoroacetylation of the unique N-terminal free amine group using S-ETFA reagent (Fig. 1b).

The affinity of albumin to liraglutide was assayed through both in vitro and in vivo glycation models: methylglyoxal- and glucose-induced glycated commercial human serum albumins were used as in vitro models, and albumins purified from the plasma of healthy and diabetic subjects were used as in vivo models. This study shows for the first time the impact of glycation on albumin-liraglutide binding and thereby raises the issue of treatment adaptation to glycation levels in diabetic patients.

Results

Fluorine labeling, purification, and characterization of liraglutide

The peptidic drug liraglutide was labeled by a nucleophilic acyl substitution reaction between the unique N-terminal amine group of liraglutide and S-ETFA (Fig. 1b). During this reaction, the pH had to be carefully controlled (between 9.6 and 10.0), as the amine of liraglutide (pK_a ∼9.5) has to be deprotonated for the reaction to occur, whereas a too basic medium led to labeled product hydrolysis and potentially altered the protein structure. After fluorine labeling, the purity and biochemical integrity of liraglutide were checked. Anion-exchange chromatography profiles exhibited only one peak at 50% of the sodium chloride gradient, corresponding to the purified labeled liraglutide form (Fig. 2a). Liraglutide was labeled with an average final yield of 80% of the initial amount. 1D ¹H NMR results confirmed that labeling did not affect liraglutide structure, as demonstrated by the overlaid spectra (Fig. 2b). Furthermore, 1D ¹⁹F NMR analysis showed that liraglutide was correctly labeled with a single peak at −74.333 ppm. Labeled degradation product TFA gave a peak at −74.472 ppm (Fig. 2c). Hydrolysis kinetics of labeled liraglutide was assessed by measuring the ¹⁹F NMR signal for liraglutide and trifluoroacetate over a 60-h duration. Thereby, a time limitation for the experiment was fixed at 6 h, corresponding to the beginning of the significant detectable hydrolysis of labeled liraglutide into liraglutide and trifluoroacetate (Fig. S1). To also confirm that the fluorine label does not alter the binding of liraglutide to albumin, isothermal titration calorimetry experiments were recorded on HSA titrated with either labeled or unlabeled liraglutide. Both labeled and unlabeled peptides yielded similar K_d values, in the micromolar range,⁵ demonstrating that the labeling does not influence the drug binding to albumin.

Figure 2.

Labeled liraglutide analysis. a, chromatography UV profile indicates that protein is present in fractions 36–42. UV, absorbance at 280 nm; Cond, conductivity; Conc, concentration of NaCl from 0 to 1 m in 50 mm borate buffer, pH 7.5. b, identical 1D ¹H spectra for unlabeled and labeled liraglutide show evidence for no significant structural alteration of the protein due to labeling process (top red spectrum, labeled and ultra-filtrated liraglutide; bottom blue spectrum, unlabeled and dialyzed liraglutide). c, 1D ¹⁹F spectrum of purified labeled liraglutide presents a main peak at −74.333 ppm extending between −74.302 and −74.369 ppm and a degradation peak at −74.472 ppm corresponding to TFA.

Glucose and methylglyoxal induced different levels of glycation on albumin

After glycation with glucose (100 mm) or methylglyoxal, the commercial fatty acid-free albumin was analyzed by SDS-PAGE (12% acrylamide) and Coomassie Blue staining. Progressively increased molecular weight bands were observed for glycated G100-HSA and MGO-HSA relative to commercial non-glycated albumin (Fig. 3a). Data from mass spectrometry confirmed this observation quantitatively (Fig. 3b). Albumin molecular weights were increased by glucose glycation and even more by MGO treatment, relative to commercial fatty acid-free albumin (66,615 Da): G100-HSA (67314 Da) and MGO-HSA (68071 Da). This increase in weight with respect to non-glycated samples probably corresponds to the formation of AGE adducts on the native molecule of albumin.

Figure 3.

In vitro glycated albumin characterization. a, SDS-PAGE analysis of commercial (A1887)-HSA and G0-, G100-, and MGO-HSA. b, mass-to-charge ratio (m/z) versus the percentage intensity plot for A1887-, G0-, G100-, and MGO-HSA. c, fructosamine levels of G0-, G25- G100-, and MGO-HSA. All data are expressed as means ± S.D. (error bars) of three independent experiments. ***, unpaired t test compared with G0-HSA. G100-HSA, p < 0.0001; MGO-HSA, p = 0.007. ##, unpaired t test comparison between G100- and MGO-HSA, p = 0.003. d, AGE fluorescence measurement on G0-, G25-, G100-, and MGO-HSA. **, unpaired t test compared with G0-HSA: p < 0.01; ***, unpaired t test compared with G0-HSA: p < 0.001; ####, unpaired t test comparison: p < 0.0001.

These AGE adducts were further identified as described previously by mass shift determination of AGE-modified peptides resulting from the trypsin digestion of each albumin sample (28). Results from this identification are shown in Tables 1 and 2. More modification sites have been reported for G100-HSA (21) and MGO-HSA (26) compared with native albumin G0-HSA (14). Only four additional modification sites were common between both in vitro glycated HSA (Arg⁹⁸, Arg¹⁴⁵, Lys²⁶², and Lys⁵²⁵), eight sites were specifically modified on G100-HSA, and 11 were only found for MGO-HSA. Among the glycated sites observed only on in vitro glycated albumins, four residues (Arg³⁴⁸, Lys⁴⁰², Arg⁴¹⁰, and Lys⁵²⁵) are known to be involved in the binding of palmitic acid (29).

Table 1.

Glycation-modified peptides identified in native HSA (G0-HSA) by MALDI-TOF-MS peptide mapping

Tryptic peptides of G0-HSA were analyzed by MALDI-TOF-MS. By comparing with theoretical digestion of unmodified serum albumin, the presence of glycation adducts on albumin was identified after determining the tryptic peptide ions with mass values, which were only present in the samples. Then modifications present in these peptides were characterized by comparing the mass shift with a list of shift mass deviations caused by different glycation adducts. Modified residues are indicated in bold. Residues located at the N terminus or C terminus of the cleaved peptide in undigested HSA are in parentheses.

Position of sequence	Modified residue	Peptide sequence	Peptide mass	Modification	Measured mass	Modification
1–10	Lys4	(R)DAHKSEVAHR(F)	1,149.58	162.05	1,311.63	Fructosyl-lysine
99–114	Lys106	(R)NECFLQHKDDNPNLPR(L)	1,996.93	72.02	2,068.95	Nϵ-Carboxyethyl-lysine
99–114	Lys106	(R)NECFLQHKDDNPNLPR(L)	1,996.93	126.03	2,122.96	Fructosyl-lysine-2H2O
175–186	Lys181	(K)AACLLPKLDELR(D)	1,398.78	126.03	1,524.81	Fructosyl-lysine-2H2O
198–205	Lys199	(R)LKCASLQK(F)	947.53	108.02	1,055.56	Pyrraline
219–224	Arg222	(R)LSQRFPK(A)	875.51	80.03	955.54	Argpyrimidine
219–224	Arg222	(R)LSQRFPK(A)	875.51	142.03	1,017.54	Imidazolone B
263–276	Lys274	(K)YICENQDSISSKLK(E)	1,684.82	58.01	1,742.83	Nϵ-Carboxymethyl-lysine/Pentosidinea
324–337	Arg336	(K)DVFLGMFLYEYARR(H)	1,779.89	142.03	1,921.92	Imidazolone B
349–359	Lys351	(R)LAKTYETTLEK(C)	1,296.70	72.02	1,368.73	Nϵ-Carboxyethyl-lysine
414–428	Lys414	(K)KVPQVSTPTLVEVSR(N)	1,639.94	58.01	1,697.94	Nϵ-Carboxymethyl-lysine/Pentosidinea
473–484	Lys475	(R)VTKCCTESLVNR(R)	1,466.71	58.01	1,524.71	Nϵ-Carboxymethyl-lysine/Pentosidinea
539–545	Lys541	(K)ATKEQLK(A)	817.48	58.01	875.48	Nϵ-Carboxymethyl-lysine/Pentosidinea
542–557	Lys545	(K)EQLKAVMDDFAAFVEK(C)	1,856.91	58.01	1,914.92	Nϵ-Carboxymethyl-lysine/Pentosidinea
542–557	Lys545	(K)EQLKAVMDDFAAFVEK(C)	1,840.92	126.03	1,966.95	Fructosyl-lysine-2H2O
546–560	Lys557	(K)AVMDDFAAFVEKCCK(A)	1,790.79	108.02	1,898.81	Pyrraline
565–574	Lys573	(K)ETCFAEEGKK(L)	1,198.54	252.11	1,450.65	Crossline

^a Both modifications are possible on lysine, and they are indistinguishable by the mass.

Table 2.

Glycation-modified peptides identified in glucose-glycated HSA (G100-HSA) and methylglyoxal-glycated HSA (MGO-HSA) by MALDI-TOF-MS peptide mapping

Tryptic peptides of G100-HSA and MGO-HSA were analyzed by MALDI-TOF-MS. By comparing with theoretical digestion of unmodified serum albumin, the presence of glycation adducts on glycated albumin was identified after determining the tryptic peptide ions with mass values, which were only present in the glycated samples. Then modifications present in these peptides were characterized by comparing the mass shift with a list of shift mass deviations caused by different glycation adducts. Modified residues are indicated in bold. Residues located at the N terminus or C terminus of the cleaved peptide in undigested HSA are in parentheses.

Position of sequence	Modified residue	Peptide sequence	Peptide mass	Modification	Measured mass	Modification
G100-HSA
1–10	Lys4	(R)DAHKSEVAHR(F)	1,149.58	162.05	1,311.63	Fructosyl-lysine
82–98	Lys93	(R)ETYGEMADCCAKQEPER(N)	2,073.83	252.11	2,325.94	Crossline
94–106	Arg98	(K)QEPERNECFLQHK(D)	1,714.80	142.03	1,856.82	Imidazolone B
99–114	Lys106	(R)NECFLQHKDDNPNLPR(L)	1,996.93	144.04	2,140.97	Fructosyl-lysine-1H2O
145–159	Arg145	(R)RHPYFYAPELLFFAK(R)	1,899.00	270.07	2,169.07	1-Alkyl-2-formyl-3,4-glycosyl-pyrrole
175–186	Lys181	(K)AACLLPKLDELR(D)	1,398.78	126.03	1,524.81	Fructosyl-lysine-2H2O
219–224	Arg222	(R)LSQRFPK(A)	875.51	80.03	955.54	Argpyrimidine
219–224	Arg222	(R)LSQRFPK(A)	875.51	142.03	1,017.54	Imidazolone B
258–274	Lys262	(R)ADLAKYICENQDSISSK(L)	1,941.92	126.03	2,067.95	Fructosyl-lysine-2H2O
263–276	Lys274	(K)YICENQDSISSKLK(E)	1,684.82	58.01	1,742.83	Nϵ-Carboxymethyl-lysine/Pentosidine*
318–336	Lys323	(K)NYAEAKDVFLGMFLYEYAR(R)	2,316.10	126.03	2,442.13	Fructosyl-lysine-2H2O
324–337	Arg336	(K)DVFLGMFLYEYARR(H)	1,779.89	142.03	1,921.92	Imidazolone B
337–348	Arg337	(R)RHPDYSVVLLLR(L)	1,467.84	270.07	1,737.92	1-Alkyl-2-formyl-3,4-glycosyl-pyrrole
338–351	Arg348	(R)HPDYSVVLLLRLAK(T)	1,623.96	80.03	1,703.98	Argpyrimidine
349–359	Lys351	(R)LAKTYETTLEK(C)	1,296.70	72.02	1,368.73	Nϵ-Carboxyethyl-lysine
390–410	Lys402	(K)QNCELFEQLGEYKFQNALLVR(Y)	2,599.30	58.01	2,657.30	Nϵ-Carboxymethyl-lysine/Pentosidine*
445–466	Arg445	(K)RMPCAEDYLSVVLNQLCVLHEK(T)	2,690.31	54.01	2,744.32	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yl)ornithine
473–484	Lys475	(R)VTKCCTESLVNR(R)	1,466.71	58.01	1,524.71	Nϵ-Carboxymethyl-lysine/Pentosidine*
501–521	Lys519	(K)EFNAETFTFHADICTLSEKER(Q)	2,545.17	144.04	2,689.21	Fructosyl-lysine-1H2O
525–534	Lys525	(K)KQTALVELVK(H)	1,128.70	252.11	1,380.81	Crossline
539–545	Lys541	(K)ATKEQLK(A)	817.48	58.01	875.48	Nϵ-Carboxymethyl-lysine/Pentosidine*
561–573	Lys564	(K)ADDKETCFAEEGK(K)	1,499.63	162.05	1,661.68	Fructosyl-lysine
MGO HSA
5–12	Arg10	(K)SEVAHRFK(D)	973.52	270.07	1,243.60	1-Alkyl-2-formyl-3,4-glycosyl-pyrrole
94–106	Arg98	(K)QEPERNECFLQHK(D)	1,714.80	39.99	1,754.79	Nϵ-(5-Hydro-4-imidazolon-2-yl)omithine
138–145	Arg144	(K)YLYEIARR(H)	1,083.59	54.01	1,137.61	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yl)ornithine
145–159	Arg145	(R)RHPYFYAPELLFFAK(R)	1,899.00	54.01	1,953.01	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yl)ornithine
146–160	Lys159	(R)HPYFYAPELLFFAKR(Y)	1,899.00	72.02	1,971.02	Nϵ-Carboxyethyl-lysine
175–186	Lys181	(K)AACLLPKLDELR(D)	1,398.78	126.03	1,524.81	Fructosyl-lysine-2H2O
182–190	Arg186	(K)LDELRDEGK(A)	1,074.54	54.01	1,128.55	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yliornithine
198–205	Lys199	(R)LKCASLQK(F)	947.53	108.02	1,055.56	Pyrraline
206–212	Arg209	(K)FGERAFK(A)	854.45	54.01	908.46	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yl)ornithine
206–212	Arg209	(K)FGERAFK(A)	854.45	80.03	934.48	Argpyrimidine
213–222	Arg218	(K)AWAVARLSQR(F)	1,157.65	54.01	1,211.66	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yl)ornithine
219–225	Arg222	(R)LSQRFPK(A)	875.51	80.03	955.54	Argpyrimidine
234–257	Lys240	(K)LVTDLTKVHTECCHGDLLECADDR(A)	2,857.29	126.03	2,983.32	Fructosyl-lysine-2H2O
258–274	Lys262	(R)ADLAKYICENQDSISSK(L)	1,941.92	72.02	2,013.94	Nϵ-Carboxyethyl-lysine
258–274	Lys262	(R)ADLAKYICENQDSISSK(L)	1,941.92	126.03	2,067.95	Fructosyl-lysine-2H2O
263–276	Lys274	(K)YICENQDSISSKLK(E)	1,684.82	58.01	1,742.83	Nϵ-Carboxymethyl-lysine/Pentosidinea
349–359	Lys351	(R)LAKTYETTLEK(C)	1,296.70	72.02	1,368.73	Nϵ-Carboxyethyl-lysine
352–372	Lys359	(K)TYETTLEKCCAAADPHECYAK(V)	2,518.07	108.02	2,626.09	Pyrraline
403–413	Arg410	(K)FQNALLVRYTK(K)	1,352.77	54.01	1,406.78	Nϵ-(5-Hydro-5-methyl-4-imidazolon-2-yl)ornithine
414–428	Lys414	(K)KVPQVSTPTLVEVSR(N)	1,639.94	58.01	1,697.94	Nϵ-Carboxymethyl-lysine/Pentosidinea
437–444	Lys439	(K)CCKHPEAK(R)	1,029.46	108.02	1,137.48	Pyrraline
473–484	Lys475	(R)VTKCCTESLVNR(R)	1,466.71	58.01	1,524.71	Nϵ-Carboxymethyl-lysine/Pentosidinea
525–534	Lys525	(K)KQTALVELVK(H)	1,128.70	252.11	1,380.81	Crossline
535–541	Lys536	(K)HKPKATK(E)	809.50	162.05	971.55	Fructosyl-lysine
539–545	Lys541	(K)ATKEQLK(A)	817.48	58.01	875.48	Nϵ-Carboxymethyl-lysine/Pentosidinea
546–560	Lys557	(K)AVMDDFAAFVEKCCK(A)	1,806.79	162.05	1,968.84	Fructosyl-lysine
565–574	Lys573	(K)ETCFAEEGKK(L)	1,198.54	252.11	1,450.65	Crossline

^a Both modifications are possible on lysine, and they are indistinguishable by the mass.

According to the fructosamine assay, the albumin glycation level was about 3-fold higher for G25-HSA (equivalent of 0.30 ± 0.02 mol of hexose/mol of HSA) and 25-fold higher for G100-HSA (equivalent of 2.13 ± 0.14 mol of hexose/mol of HSA) compared with G0-HSA (0.08 ± 0.01 mol of hexose/mol of HSA). Interestingly, MGO-HSA (1.16 ± 0.16 mol of hexose/mol of HSA) was almost 2-fold less glycated than G100-HSA (Fig. 3c). In contrast, MGO-glycated HSA exhibited 5-fold higher AGE levels (6.2 × 10⁶ cps g⁻¹ liter) compared with G100-HSA (1.2 × 10⁶ cps g⁻¹ liter) (Fig. 3d). The glycated albumin obtained under typical hyperglycemia conditions (G25-HSA) had a lower level of AGE (0.7 × 10⁶ cps g⁻¹ liter) than the other in vitro glycated albumin models, just slightly higher than G0-HSA (0.6 × 10⁶ cps g⁻¹ liter).

In vitro glycated albumin exhibits impaired affinity for liraglutide

The NMR ¹⁹F transverse relaxation time T₂ of labeled liraglutide at variable concentrations was measured at a fixed concentration of albumin, and the plot showing [HSA]₀/ΔR_2obs with respect to trifluorinated liraglutide concentration [Lira¹⁹F] shows a linear curve whose horizontal intercept corresponds to K_d (Fig. 4a). These plots exhibit a good linearity with correlation coefficients (r²) >0.976 (mean 0.979 ± 0.016). The dissociation constants K_d featured in Fig. 4b for albumin samples were calculated using these plots.

Figure 4.

NMR experiments for in vitro glycated albumin interaction with liraglutide. a, [HSA]₀/ΔR_2obs is plotted as a function of liraglutide concentration. A linear regression is calculated, and K_d is obtained from the vertical intercept. G0-HSA, K_d = 35 ± 8 μm, n = 3, r² = 0.989 ± 0.009; G25-HSA, K_d = 40 ± 12 μm, n = 3, r² = 0.955 ± 0.037; G100-HSA, K_d = 240 ± 10 μm, n = 3, r² = 0.974 ± 0.028; MGO-HSA, K_d = 173 ± 7 μm, n = 3, r² = 0.976 ± 0.014. b, dissociation constant comparison for non-glycated and in vitro glycated commercial albumin. All data are expressed as means ± S.D. (error bars) of three independent experiments. ***, unpaired t test compared with G0-HSA; p < 0.0001.

When albumin is glycated with physiopathological glucose concentration (25 mm), no significant alteration of liraglutide binding was observed, as attested by a dissociation constant for G25-HSA (42 ± 12 μm) quite similar to control G0-HSA (35 ± 8 μm). By contrast, albumin that was glycated following glucose incubation in suprapathological conditions (G100-HSA) exhibited an almost 7-fold higher dissociation constant (240 ± 10 μm). Methylglyoxal-glycated albumin affinity for liraglutide was also significantly affected with a K_d of 173 ± 6 μm (i.e. 5-fold higher than G0-HSA) (Fig. 4b). Unexpectedly, MGO-glycated HSA that underwent a more drastic glycation process compared with G100-HSA presented an almost equivalent impairment of its affinity to liraglutide. Indeed, as seen on SDS-PAGE, the MGO-HSA sample exhibited a higher-molecular weight band, confirmed by measurement of its exact molecular mass (Fig. 3a). All of these results, obtained by using in vitro glycation models, demonstrate that glycation at suprapathological conditions significantly impaired the albumin-liraglutide interactions. To ascertain the relevance of these observations, we investigated this albumin-liraglutide interaction using in vivo glycated albumin purified from plasma.

HbA1c plasma levels of diabetic patients correctly matched with the level of glycation in purified albumin samples

Human albumin samples were isolated from the plasma of both healthy subjects and diabetic patients using a differential ammonium sulfate (AS) method based on the difference in HSA solubility with other plasmatic proteins. As shown in Fig. 5a, the SDS-PAGE pattern showed the major band at around 66 kDa for all purified albumin samples. This pattern attests to the high purity of our human albumin preparations, which were obtained with high yields (above 85%, calculated by densitometry) (Fig. 5a).

Figure 5.

Plasma-extracted albumin analysis. a, SDS-PAGE analysis of albumin sample purified from plasma of patients divided into three groups (ND, D, and D+). b, HbA1c levels (%) of patients divided into three groups (ND, D, and D+). ***, unpaired t test compared with HSA from ND; p < 0.00010 for D and D+. ##, unpaired t test comparison between D and D+; p = 0.0041. c, fructosamine levels of purified albumin from healthy (ND) and diabetic (D and D+) patients. **, unpaired t test compared with HSA from ND; p = 0.0016 and 0.0062, respectively, for D and D+ groups. #, unpaired t test comparison between D and D+ groups; p = 0.0277. d, correlation between percentage HbA1c and albumin fructosamine levels. e, AGE fluorescence measurement on HSA from ND, D, and D+ groups (n = 3); unpaired t test compared with HSA from ND; p = 0.0621 and 0.1282, respectively, for D and D+ groups. f, correlation between percentage HbA1c and albumin-AGE levels. g, levels of FFA bound to purified albumin from healthy (ND) and diabetic (D and D+) patients. h, correlation between levels of FFA bound to purified albumin and albumin fructosamine levels. Error bars, S.D.

Human plasma samples used for this study correspond to diabetic patients with a mean HbA1c level of 13.4 ± 3.3%. As clinical observations of drug inefficiency concerned advanced or poorly controlled diabetic patients, our plasma samples were divided into two groups: moderate uncontrolled diabetic patients named group D (9% < HbA1c < 11% and with a mean of 10.0 ± 0.5% HbA1c) and highly uncontrolled diabetic patients named group D+ (13% < HbA1c < 18% and with a mean of 16.0 ± 1.7% HbA1c) (Fig. 5b). The HbA1c percentage for each diabetic patient and healthy subject is given in Table S1.

The fructosamine assay showed that albumin purified from diabetic groups D and D+ were 1.7- and 3-fold more glycated, respectively, than albumin purified from non-diabetics (ND) (0.19 ± 0.02 and 0.35 ± 0.10 versus 0.11 ± 0.01 mol of hexose/mol of HSA for ND) (Fig. 5c) (the fructosamine levels for each diabetic patient and healthy subject are reported in Fig. S2). In parallel, albumin purified from diabetic patients displayed an increase in fluorescent AGE levels, although it did not reach statistical significance (Fig. 5e; p = 0.06 for D versus ND; p = 0.12 for D+ versus ND).

In addition, albumin samples from diabetic patients and non-diabetic individuals showed a significant and positive correlation between Hb1A_C, fructosamine, and AGE levels. As shown in Fig. 5d, a strong correlation (r = 0.926, p < 0.0001) between Hb1A_C and fructosamine parameters was established. Similarly, a significant correlation (r = 0.762, p = 0.0104) was obtained between Hb1A_C and AGE levels (Fig. 5f).

Total free fatty acid (FFA) levels bound to HSA were also determined by GC/MS (Fig. 5g). Albumin purified from diabetic groups D and D+ displayed a tendency to higher FFA levels than albumin purified from ND (16.15 ± 13.00 and 13.04 ± 5.03 versus 8.38 ± 3.36 nmol/mg HSA for ND) with a large variability between individuals. No statistical difference was observed between both diabetic groups, and no significant correlation could be established (r = 0.115, p = 0.735) between FFA and fructosamine levels (Fig. 5h).

After determination of the glycation level of plasma-extracted albumin samples, their interaction with liraglutide was assessed by the same method used for in vitro glycated commercial albumin.

HSA from highly uncontrolled diabetic patients exhibits an impairment of binding to liraglutide

HSA from non-diabetic subjects displayed a liraglutide dissociation constant of 48 ± 13 μm, which appears to be quite similar to the value obtained for commercial HSA. As seen in Fig. 6, a marked difference was observed between groups of diabetic patients in terms of K_d values (the K_d value of albumin is reported for each diabetic patient and healthy subject in Fig. S3. Indeed, albumin samples from moderately uncontrolled diabetic patients (D group) were found to have an affinity for liraglutide (48 ± 2 μm), similar to control HSA, whereas a significant increase in K_d values was measured for albumin samples of the D+ group (224 ± 60 μm) (Fig. 6). Globally, the affinity of HSA for liraglutide was found to be reduced by almost 5-fold for patients with highly uncontrolled diabetes only. These results are consistent with those obtained with in vitro models of albumin and attest to the impairment of albumin affinity for liraglutide in the situation of glycation. Finally, these results suggest that a high level of glycation is required to significantly impair the binding of albumin to liraglutide.

Figure 6.

NMR results for plasma-extracted albumin interaction with liraglutide. Shown are albumin-liraglutide K_d values for non-diabetics (ND), diabetics below 11% HbA1c (D), and diabetics above 11% HbA1c (D+). **, unpaired t test comparison between ND and D+ groups; p = 0.0017. #, unpaired t test comparison between D and D+ groups; p = 0.0057. Error bars, S.D.

Discussion

Liraglutide is a recent promising treatment for type 2 diabetes mellitus with a number of benefits relative to existing drugs, especially regarding cardiovascular risk improvement, as described earlier (9–13). The main strategy for the protracted action of liraglutide over the day mainly relies on its transport and progressive release by albumin. Unfortunately, some diabetic patients fail to respond correctly to certain treatments. In particular, this was noticed in clinical observations that glycemia was hardly maintained at acceptable levels for patients treated with long-acting albumin-transported drugs, such as insulin detemir (Levemir®).⁶

Our study aimed to better understand the link between plasmatic oxidative damages and alterations in the treatment efficiency, which remains highly dependent on diabetes evolution and therapeutic adaptation. Notably, GLP-1 analogues are recommended in the “therapeutic climbing” generally in association with a biotherapy or insulin, when first-line oral anti-diabetics, such as metformin or sulfonylureas, are not efficient enough to reduce the HbA1c level to the therapeutic target (8).

Chronic hyperglycemia, which characterizes diabetes, is the main factor of protein alteration by glycation that is known to be implicated in diabetes aggravation, such as renal failure and atherosclerosis (30). Glycation particularly alters structural and functional properties of albumin (31), including the protein affinity for therapeutic drugs (32). For instance, the affinities of many drugs, including warfarin or ketoprofen, were found to be affected by albumin glycation, resulting in the variability of therapeutic efficacy for advanced diabetic patients (for warfarin) (32, 33). Other studies reported an altered albumin affinity for several sulfonylurea drugs commonly used in the treatment of type 2 diabetes (i.e. acetohexamide, tolbutamide, and gliclazide) after glycation (34). In parallel, several studies showed that albumin glycation resulted in an impaired affinity for fatty acids, such as lipoic, oleic, linoleic, lauric, and caproic acids (35, 36).

In the present study, we showed for the first time that glycation alters the affinity of albumin for liraglutide. ¹⁹F NMR is one of the best validated techniques for monitoring ligand-protein interactions (22–27). This method has already been used to study the binding of various drugs to albumin, either using naturally fluorinated drugs or by competition with fluorinated spy molecules (24, 37–42). Although a labeling step was required for liraglutide, the ¹⁹F NMR relaxation time method presents the advantage of specifically targeting the labeled peptide drug signal. Moreover, it requires a small amount of protein, as there is an excess of ligand. The N-terminal amine group was targeted for trifluoroacetylation because it leads to a single substitution on the liraglutide peptide. Moreover, this position is located in an unstructured part of liraglutide far from the palmitoyl fatty acid (d > 40 Å on the X-ray structure: Protein Data Bank entry 4APD) and does not influence the binding to albumin, as confirmed by isothermal titration calorimetry measurements (data not shown). Most previous studies on liraglutide and other fatty acid–grafted GLP1 analogues focused on the measure of their binding to the GLP-1 receptor and rarely on their interaction with albumin (43–45). Recently, a K_d value of about 150 μm for albumin-liraglutide binding was estimated by fitting data from small-angle X-ray scattering and static light scattering methods (46). This latter value is in good agreement with our results obtained by NMR, which present the advantage of giving a direct and accurate measure of the affinity between liraglutide and human serum albumin. Indeed, the measured K_d values obtained for liraglutide in our study (between 35 and 240 μm) indicate a weak binding to human serum albumin. Similarly, for detemir, which also binds to albumin via a fatty acid chain, a K_a value of 2.4 × 10⁵ m⁻¹ was reported, corresponding to a dissociation constant in the micromolar range (47). These values are much lower than those reported for the binding of free palmitate or fatty acid C16 to albumin, which has an approximate K_a of 14.5 × 10⁷ m⁻¹, corresponding to a dissociation constant in the nanomolar range (48). Thus, fatty acid–grafted peptides (detemir and liraglutide) bind weakly to albumin compared with free fatty acids. This may be the consequence of a restricted access to a number of fatty acid–binding sites on albumin due to steric hindrance by the peptide.

In our study, in vitro glucose- or methylglyoxal-glycated commercial albumins appear to mimic the in vivo glycation effects, leading to similar alterations of its binding properties. Indeed, K_d values were comparable between G100-HSA or MGO-HSA and HSA from highly uncontrolled diabetic patients. On the other hand, quite similar K_d values were obtained for G0-HSA, G25-HSA, and non-diabetic HSA or HSA from moderately uncontrolled diabetic patients.

In vitro glycation with 25 mm glucose is commonly used for mimicking diabetes conditions. By contrast, in vitro glycation with 100 mm glucose or methylglyoxal appeared to be more drastic conditions than in vivo glycation; this was confirmed by fructosamine measurements and mass spectrometry analysis. Methylglyoxal is mainly formed as a by-product of glycolysis, and its concentration is higher in diabetic patients, reaching 1 μm in blood or plasma relative to 0.55 μm for controls (49); this plays a role in the exacerbation of protein glycation (50). As confirmed in the present study, MGO-HSA contained more AGE products compared with G100-HSA. Indeed, methylglyoxal is 20,000-fold more reactive than glucose and affects arginine side chains rather than lysines (51), as confirmed by the characterization and location of AGE modifications in albumin molecules. Methylglyoxal is known to induce an important irreversible formation of AGE in glycated proteins (52). Although both in vitro G100- and MGO-glycated HSA models display different biochemical modifications, their binding constants for liraglutide appear quite similar. On the contrary, the G25-HSA model affinity for liraglutide was not affected by the glycation, suggesting that a moderate biochemical modification of the protein is not sufficient to alter its drug-binding capacity.

In the same manner, the glycation level strongly influenced the binding for in vivo HSA models. Indeed, for healthy subjects and diabetic patients with an HbA1c level up to 10%, the affinity for liraglutide was similar. A decreased affinity was measured for highly uncontrolled diabetic patients only (HbA1c > 13%). This observation suggests that albumin glycation should not affect the therapeutic actions of liraglutide until an advanced stage of chronic hyperglycemia, higher than 300 mg/dl. The altered affinity to albumin could lead to higher free active liraglutide in serum, with altered kinetics and duration of action, which could explain a lower efficacy and unsustained metabolic effects over 24 h, but also a greater susceptibility to gastro-intestinal side effects of the drug due to higher peaks of free GLP1 RA concentrations.

Drug binding to albumin has been extensively characterized. The two most common drug-binding sites of albumin are located in the hydrophobic cavities of subdomains IIA and IIIA, known as Sudlow's sites I and II, respectively (53). Non-esterified medium- and long-chain fatty acids, such as myristic and palmitic acid, can bind to seven sites on HSA, three of which are located on Sudlow's drug-binding sites I and II (54, 55). In particular, the albumin region from residues 377–582, including part of domain II and the entire domain III, was found to be the primary binding site for long-chain fatty acids (56), and, more precisely, the key roles of Arg⁴¹⁰ (Sudlow's site II) and Lys⁵²⁵ (domain IIIB) for fatty acid binding have been reported (57). In our study, both residues were found to be glycated in MGO-HSA, whereas only Lys⁵²⁵ was modified in G100-HSA (Table 2). The effect of glycation on affinity depends on the extent of glycation and is site-dependent (31). Lys⁵²⁵ is the main site for both in vivo (58) and in vitro albumin glycation (18). These results suggest that the modification of this residue may explain the observed reduced affinity of in vitro glycated HSA for liraglutide. Additionally, Arg³⁴⁸ (Sudlow's site II) and Lys⁴⁰² (domain IIIB) residues, which are involved in fatty acid binding, were also glycated in vitro for G100-HSA and may contribute to a reduction in the affinity of albumin for the drug. These data indicate that glycation reduces albumin affinity for fatty acid–grafted drugs.

Another factor that could influence binding in the in vivo albumin models and contribute to the observed variability between individuals is competition between fatty acids and liraglutide for albumin binding. Such effects at physiological concentrations have been demonstrated on non-glycated and glycated HSA interactions with sulfonylurea drugs (31).

Indeed, diabetic patients have higher plasma fatty acid concentration than healthy subjects. This is especially observed in obese patients with increased insulin resistance (59). In our study, total FFA levels bound to HSA appeared to be higher in diabetic patients but without reaching statistical significance. Furthermore, no relevant difference was observed between D and D+ groups, suggesting that FFA levels do not significantly contribute to the impairment of liraglutide binding to HSA.

In advanced diabetic cases, a 50% higher drug dosage is usually required (60), with poorly known effects on health. Dose-dependent detrimental effects of liraglutide have been reported, such as the induction of a high risk of pancreatic and thyroid cancer in rats (61–63). Our present work showing impaired liraglutide binding in diabetics indicates the need for further research to prevent glycation-mediated alterations of albumin-drug binding capacity. For example, amino acid or fatty acid length modifications in the liraglutide molecule could modulate its affinity for albumin and afford resilience to glycation-induced modifications at specific binding sites. Indeed, this strategy has been explored successfully by researchers from Novo Nordisk, who recently developed a GLP-1 analogue called semaglutide (64). This peptide contains a substitution of an alanine residue (Ala⁸) by a 2-aminoisobutyric acid for stabilization against dipeptidyl peptidase-4 degradation and a stearic acid instead of a palmitic acid on the same lysine residue (Lys²⁶). Its longer-chain fatty acid confers a higher affinity for albumin to semaglutide. It would be interesting to test the hypothesis that the higher affinity of semaglutide for albumin could be less susceptible to alterations by glycation for highly uncontrolled diabetic patients. Optimization of GLP-1 analogues under glycation conditions is an interesting alternative to potentially dangerous dose escalation.

Experimental procedures

Liraglutide fluorine labeling and purification

For liraglutide fluorine labeling, the pH of 1.5 ml of a 6 mg/ml commercial liraglutide pen (Victoza®, Novo Nordisk A/S) was adjusted from the initial value of about 8.3 to 9.6 using 0.01 m NaOH. Every hour, 10 μl of 97% S-ETFA (Aldrich, 177474) was added to liraglutide for 5 h at room temperature. A total of 50 μl of S-ETFA was used for each labeling, corresponding to a substrate/protein ratio of 110. The pH was maintained between 9.6 and 10 throughout the reaction to allow the proper reaction to occur. After the labeling step, the reaction mixture was first purified by triple dialysis (Spectra/Por® 2,000 D MWCO cut-off) against 50 mm borate buffer, pH 7.5. Then labeled liraglutide was centrifuged at 12,000 × g for 15 min and purified by anion-exchange chromatography on a 5-ml Hi-Trap Q column (GE Healthcare) coupled with AKTA Primeplus apparatus (1-ml/min flow). A 100-ml NaCl gradient from 0.01 to 1 m was then applied to elute the proteins. Elution fractions were further washed and concentrated using a centrifugal filter device (20 mm Vivaspin, molecular weight cut-off 3,000). Purified liraglutide concentration was assessed by measuring UV absorbance at 280 nm using the Beer–Lambert law with the molar extinction coefficient ϵ: 1.99 liter g⁻¹ cm⁻¹ (5). Purified labeled liraglutide was aliquoted and frozen in liquid nitrogen and then stored at −80 °C.

In vitro glycation of albumin

Albumin glycation was performed as described previously (65). Fatty acid–free human serum albumin (Sigma, A1887) solutions were prepared in PBS (pH 7.4) at a concentration of 40 g/liter. HSA solution was incubated at 37 °C with either 10 mm methylglyoxal for 48 h (MGO-HSA) or glucose (25 mm or 100 mm) for 3 weeks (G25-HSA and G100-HSA) or without glucose (G0-HSA).

Two of our in vitro glycation preparations constitute common glycated albumin models and use suprapathological concentrations of glucose (100 mm) or methylglyoxal (10 mm). The third model, which uses a pathological concentration of glucose (25 mm), mimics diabetic conditions. The solutions were then dialyzed four times against PBS, filtered through a 0.22-μm Millipore filter, and concentrated using Amicon Ultra-0.5 centrifugal filter devices (3K, Millipore). The final concentration of albumin samples was determined using UV absorbance (ϵ = 34,445 m⁻¹ cm⁻¹) and a Bradford assay. Glycated albumin samples were stored at −20 °C. Commercial delipidated (or fatty acid–free) HSA was used to circumvent eventual competitive effect due to fatty acids that could remain bound to albumin after extraction from human serum.

Albumin extraction from human sera by ammonium sulfate precipitation

Human plasma samples from diabetic patients were obtained from the Centre Hospitalier Gabriel Martin (Saint-Paul, La Réunion). The procedure and the collection of human materials were approved by the local governmental French Ethical Committee and conformed to the standards set by the Declaration of Helsinki. Control blood samples from healthy individuals were taken from non-diabetic volunteers. An average of about 2 ml of plasma was retrieved from a whole-blood sample of 5 ml after two successive centrifugations (2,000 × g for 10 min and then 2,500 g for 15 min). The supernatant was collected and stored at 80 °C.

A differential ammonium sulfate precipitation two-step protocol described by Mandia's team (66) and modified in our laboratory (67) was used to extract HSA from the plasma. The first step consisted of adding 54% AS to a plasma sample using a saturated 100% AS solution (540 g/liter; pH adjusted to 7.4 with NH₄OH). After stirring for 1 h at 4 °C, the whole solution was centrifuged (5,000 × g for 10 min), and the precipitate was discarded. Saturated AS solution was further added to the supernatant under stirring to reach 70% of AS where HSA precipitates. The pellet was collected after centrifugation (10,000 × g for 30 min) and resuspended in 0.1 m sodium phosphate. HSA solution was then washed three times and concentrated by ultrafiltration (Amicon 3K, Millipore). The final concentration of purified albumin was determined using UV absorbance (ϵ = 34,445 m⁻¹ cm⁻¹), and samples were stored at −80 °C.

Fructosamine assay on human serum albumin

To assess the albumin glycation rate, the fructosamine assay developed by Johnson et al. (68) was performed on HSA samples. 1-Deoxy-1-morpholino-d-fructose was used as a fructosamine reference for the standard range. Then 100 μl of nitro blue tetrazolium was added to 100 μl of sample or commercial glycated albumin range (Sigma A8301, mean of 2.2 mol of hexose/mol of HSA) on 96-well plates. Samples were incubated at 37 °C, and absorbance was measured over time at 540 nm using a FLUOstar Omega microplate reader (BMG LABTECH). Kinetic curves were plotted, and slopes (arbitrary units/min) were calculated using the Omega software. Fructosamine concentrations in samples were expressed as mol of hexose eq/mol of albumin.

AGE fluorescence measurement

To assess AGE-peptide levels in glycated commercial albumin, fluorescence measurements were made in a 96-well plate with excitation at 370 nm and emission at 440 nm measured on 100-μl samples using a FLUOstar Omega microplate reader. Fluorescent AGE levels were expressed in arbitrary units.

Free fatty acid level determination on purified human serum albumin

Free fatty acid level was determined on purified albumin isolated from diabetic and non-diabetic plasma by capillary GC-MS on the ICANanalytics lipidomic platform (Paris, France). Purified albumin samples (40 μl) were preliminarily supplemented with an internal standard mixture of three deuterated fatty acids in 500 μl of methanol and 275 μl of 0.1 n HCl. FFAs were successively extracted twice using 1.5 ml of isooctane and derivatized for GC/MS analysis using 50 μl of pentafluorobenzyl bromide (10% in acetonitrile) and 50 μl of 10% diisopropylamine (10% in acetonitrile). Analysis of FFAs in HSA was performed using an ISQ^TM LT single quadrupole GC-MS system (Thermo Fisher Scientific). Separation of FFAs was performed on a FAST 10 m × 0.1 mm, 0.2-μm BPX70 capillary column. The injector temperature was 250 °C, and the samples were injected using the splitless injection mode. Quantification was performed on a single quadripole using chemical ionization with methane.

Mass spectroscopy and peptide characterization

The glycation of serum albumin was analyzed by MALDI-TOF-MS for mass shift determination and for AGE-modified peptide exploration, as described previously (69).

NMR analysis of purified ¹⁹F-labeled liraglutide with albumin

Purified protein solution was analyzed by 1D ¹H and ¹⁹F NMR to check for the presence of correctly folded and ¹⁹F-labeled liraglutide. A final volume of 500 μl of liraglutide in 50 mm borate buffer, pH 7.5, was used, to which 50 μl of deuterium oxide (D₂O) was added. No ¹H or ¹⁹F reference was used in order to avoid possible influence on the binding. 1D ¹H experiments were performed using water suppression with gradients (128 scans, relaxation delay of 2 s). Fluorinated liraglutide hydrolysis kinetics leading to unlabeled liraglutide and TFA was followed by recording 1D ¹⁹F NMR spectra at 25 °C on a 100 μm sample in 50 mm borate buffer, pH 7.5, for 60 h (512 scans, relaxation delay of 7 s, and a transmitter frequency of 564.690823 MHz). All NMR experiments were recorded on a 600-MHz Bruker Avance III spectrometer equipped with a triple resonance ¹H/¹⁹F-¹⁵N-¹³C TCI cryoprobe. NMR spectra were processed and analyzed with Topspin version 2.1.6 software (Bruker).

Transversal relaxation rate analysis for albumin-liraglutide binding K_d determination

To study liraglutide-albumin interaction, transversal relaxation time T₂ (or corresponding rate R₂ = 1/T₂) was measured by a pseudo-2D ¹⁹F CPMG (Carr–Purcell–Meiboom–Gill) pulse program consisting of acquiring signal with increasing relaxation delays (1, 5 10, 20, 30, 50, 70, 100, 150, and 200 ms). The parameters used for the ¹⁹F CPMG experiments were the following: a temperature of 25 °C, 512 scans, a relaxation delay of 3.5 s, and a transmitter frequency of 564.690823 MHz. A specific concentration of HSA (10, 20, or 30 μm) was added to a range of labeled liraglutide from 50 to 300 μm. For each titration point, the transversal relaxation time T₂ (T₂ = 1/R₂) was calculated with the T₂ determination menu from Topspin. From these T₂ values, dissociation constants were calculated by the method developed by the team of J.-P. Girault (70); [P]₀/ΔR_2obs values were plotted as a function of liraglutide concentration to obtain a K_d value from the horizontal intercept. [P]₀ corresponds to total albumin concentration ([HSA]₀ and ΔR_2obs is the difference between R₂ values, in the presence and in the absence of albumin (1/T_2obs − 1/T_2free). Higher K_d values reflect a lower affinity. All titrations were recorded in triplicate.

Statistical analysis

The data are expressed as the means ± S.D. from a minimum of three experiments. Statistical significances were determined using one-way analysis of variance (followed by Student's t test) for multiple comparisons; a p value of <0.05 was considered as significant.

Author contributions

A.G. researched data, wrote, and reviewed the manuscript. P.G. researched data and reviewed the manuscript. A.C. researched data and reviewed the manuscript. S.A.K. researched data. A.G.D. contributed discussion and reviewed the manuscript. X.D. had the original idea, contributed discussion, and reviewed the manuscript. N.L.M. reviewed the manuscript. E.B. reviewed the manuscript. S.B. researched data and reviewed the manuscript. B.P.D. researched data and reviewed the manuscript. E.A. researched data, contributed discussion, and reviewed the manuscript. O.M. reviewed the manuscript. P.R. coordinated the study; researched data; and wrote, reviewed, and edited the manuscript. J.C. had the original idea; conceived and coordinated the study; researched data; and wrote, reviewed, and edited the manuscript.