Structural and functional interaction of fatty acids with human liver fatty acid-binding protein (L-FABP) T94A variant

Authors


Abstract

The human liver fatty acid-binding protein (L-FABP) T94A variant, the most common in the FABP family, has been associated with elevated liver triglyceride levels. How this amino acid substitution elicits these effects is not known. This issue was addressed using human recombinant wild-type (WT) and T94A variant L-FABP proteins as well as cultured primary human hepatocytes expressing the respective proteins (genotyped as TT, TC and CC). The T94A substitution did not alter or only slightly altered L-FABP binding affinities for saturated, monounsaturated or polyunsaturated long chain fatty acids, nor did it change the affinity for intermediates of triglyceride synthesis. Nevertheless, the T94A substitution markedly altered the secondary structural response of L-FABP induced by binding long chain fatty acids or intermediates of triglyceride synthesis. Finally, the T94A substitution markedly decreased the levels of induction of peroxisome proliferator-activated receptor α-regulated proteins such as L-FABP, fatty acid transport protein 5 and peroxisome proliferator-activated receptor α itself meditated by the polyunsaturated fatty acids eicosapentaenoic acid and docosahexaenoic acid in cultured primary human hepatocytes. Thus, although the T94A substitution did not alter the affinity of human L-FABP for long chain fatty acids, it significantly altered human L-FABP structure and stability, as well as the conformational and functional response to these ligands.

Abbreviations
2-OG

2-oleoyl glycerol

AA

arachidonic acid

ANS

1-anilinonaphthalene-8-sulfonic acid

CD

circular dichroism

DHA

cis-4,7,10,13,16,19-docosahexaenoic acid

EPA

cis-5,8,11,14,17-eicosapentaenoic acid

FATP5

fatty acid transport protein 5

LCFA

long chain fatty acid

L-FABP

liver fatty acid-binding protein

LPA

oleoyl-l-α-lysophosphatidic acid sodium salt

PODG

1-palmitoyl-2-oleoyl-sn-glycerol

POPA

1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate monosodium salt

PPAR

peroxisome proliferator-activated receptor

PUFA

polyunsaturated fatty acids

T94A

human L-FABP T94A variant

T94T

wild-type human L-FABP

TG

triglyceride

Introduction

Liver fatty acid-binding protein (L-FABP) comprises 2–7% of cytosolic protein (0.2–0.7 mm), representing 80–90% of cytosolic long chain fatty acids (LCFA) and LCFA CoA binding capacity [1, 2]. L-FABP is also thought to be a potential modifier protein promoting an early adaptive response to hepatocyte stress by which potentially lipotoxic LCFAs are partitioned into stable intracellular triglyceride (TG) stores [3]. There is significant evidence that a LCFA CoA-binding protein such as L-FABP is required for optimal activity of transacylase enzymes in the murine TG synthesis pathway. First, free unbound LCFA CoA but not L-FABP-bound LCFA CoA inhibits transacylase enzymes mediating the first two steps in TG synthesis [4-7]. In fact, L-FABP-bound LCFA CoA stimulates these enzymes [7-10] by removing the inhibitory LCFA CoA and presenting it for enzymatic utilization [4, 5, 7, 11]. Hepatic L-FABP is up-regulated in human non-alcoholic fatty liver disease and animal models of non-alcoholic fatty liver disease [12-15]. In contrast, L-FABP ablation decreases hepatic TG accumulation [16-21]. Second, L-FABP also prevents LCFA lipotoxicity by binding/linking with oxidized and reactive LCFA species [22-29], which depletes the L-FABP pool as non-alcoholic fatty liver disease progresses to non-alcoholic steatohepatitis [14, 23-27]. Third, in the longer term, L-FABP may also prevent/ameliorate lipotoxicity by enhancing LCFA β-oxidation. Murine and/or human L-FABP enhance LCFA uptake [1, 30-33], [33-36], target/co-transport n–3 polyunsaturated fatty acids (PUFAs) such as eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) [37] as well as other LCFAs [35, 38-41] into the nucleus, directly bind peroxisome proliferator-activated receptor α (PPARα) [42-46], and activate LCFA-induced PPARα transcription of LCFA β-oxidative enzymes [37, 47].

A recently identified polymorphism in the human L-FABP gene, resulting in a T94A substitution, occurs with high frequency (26–38% minor allele frequency (MAF); 8.3 ± 1.9% homozygous) in the human population, and represents the most frequent polymorphism in the FABP family (MAF for 1000 genomes in the National Center for Biotechnology Information dbSNP database and the ALFRED database) [48-54]. However, direct structural and functional evidence of the impact of the L-FABP T94A variant is lacking or contradictory. For example, over-expression of human L-FABP T94A in Chang liver cells did not enhance LCFA uptake and decreased TG accumulation [55]. In contrast, murine L-FABP over-expression increased LCFA uptake and increased hepatic TG [30-33], while L-FABP ablation [1, 34-36] or antisense treatment [33] decreased LCFA uptake and decreased hepatic TG [16-20]. These results initially suggested that the L-FABP T94A variant does not bind LCFA and represents a ‘loss-of-function’ variant analogous to the L-FABP null mouse [55]. However, this conclusion is at variance with the finding that expression of the L-FABP T94A variant is associated with elevated plasma TGs [49, 56], increased low-density lipoprotein cholesterol [49, 53], atherothrombotic cerebral infarction [51], and non-alcoholic fatty liver disease in human subjects [53]. It is not clear that the discrepancy is due to Chang liver cells being derived from human cervical cancer cells [57-59].

In summary, a more complete understanding of the structure, LCFA binding properties, and function of the human L-FABP T94A variant in LCFA-induced PPARα transcriptional activity in human hepatocytes is needed. To address these issues, structural and LCFA binding studies were initiated with purified recombinant human WT and T94A variant L-FABP proteins. While the human T94A variant L-FABP did not differ much from the WT in terms of its affinity for LCFA and LCFA metabolites, its structural response to these ligands differed markedly from that of the human WT T94T L-FABP. Furthermore, T94A decreased n–3 PUFA-mediated induction of PPARα transcriptional activity in human hepatocytes.

Results

Key amino acids differentiating human wild-type (WT) and T94A variant L-FABP from rat L-FABP

The common ribbon structure of both human and murine L-FABP demonstrates the classic β-barrel comprising β-sheets A–J enclosing the ligand-binding pocket (Fig. 1, yellow) plus two α-helices near the opening of the binding pocket (Fig. 1, pink) (Protein Data Bank ID 2LKK) [60]. While the amino acid sequence of human T94T L-FABP is 82.7% identical and 89.8% similar to the rat L-FABP, these proteins differ significantly in terms of charge, aromatic amino acid residues, the volume of the ligand-binding cavity, and, in the case of the human L-FABP T94A variant, alanine (a smaller non-polar amino acid) is substituted for threonine (a larger polar amino acid) [61-64].

Figure 1.

Key amino acids differentiating human WT and T94A variant L-FABP from rat L-FABP. Ribbon structure of L-FABP (Protein Data Bank ID 2LKK) demonstrating the classic β-barrel (yellow β-sheets A–J) and α-helices (pink) common to both human and rat L-FABPs [60]. Shown within the β-barrel are two bound oleic acids [60]. Key amino acids differentiating human and rat L-FABPs are designated as follows: (a) a tyrosine (Y) in β-sheet A is found in both human and rat L-FABPs; (b) tyrosines (Y) in β-sheets C and J are found only in rat L-FABP; (c) three positively charged amino acids (+) in β-sheets B, D and F in rat L-FABP are replaced by neutral amino acids in human L-FABPs; (d) threonine (T), a polar non-charged amino acid, is found at position 94 (*) in β-sheet G of both WT human and rat L-FABP, but is replaced by the smaller non-polar alanine (A) in the human L-FABP T94A variant.

The human L-FABPs have three neutral amino acids where the rat L-FABP has three positively charged amino acids (Fig. 1, + symbol). This results in significantly lower isoelectric point values: pI = 6.60 for human L-FABP T94T and T94A, compared to pI = 7.79 for rat L-FABP. Although neither human nor rat L-FABPs contain tryptophan, human L-FABP contains only a single tyrosine while rat L-FABP has three tyrosines (Fig. 1, Y). The tyrosine residue common to both the human and rat L-FABPs (Fig. 1, Y in β-sheet A) is outside the ligand-binding pocket. In contrast, the two additional tyrosine residues in the rat L-FABP (Fig. 1, Y in β-sheets C and J) are within/near the ligand binding site. Consequently, the tyrosine emission of rat L-FABP but not human L-FABP is more sensitive to occupancy of the ligand-binding pocket (data not shown). The threonine (T) at position 94 (Fig. 1, asterisk) in β-sheet G of both WT human and rat L-FABP is replaced by alanine (A) in the human L-FABP T94A variant. While threonine is a polar non-charged amino acid, alanine is non-polar, uncharged and significantly smaller – occupying 24% less volume and 18% less surface area [65, 66]. Finally, the volume of the ligand-binding cavity of rat L-FABP is 1.3–2.1-fold larger than that of other FABP family members, and that of human WT L-FABP T94T is larger than that of the rat L-FABP [62-64].

The effects of these differences on the structure, stability, specificity for binding LCFAs and intermediates of TG synthesis, conformational responsiveness to ligand binding, and function are addressed in the following sections.

SDS/PAGE and western analysis of rat, T94T and T94A L-FABPs

Purified rat L-FABP as well as human T94T and T94A L-FABP proteins were detected as single bands at ~ 14 kDa on SDS/PAGE gels (Fig. 2A). The results of MALDI-TOF analysis of the His tag-removed T94T and T94A variant were consistent with molecular weights of these two proteins based on amino acid sequence (data not shown). Western analyses showed that, while the antibody against mouse L-FABP reacted equally well with both human and rat L-FABP (Fig. 2B), the antibody against human L-FABP reacted better with human than rat L-FABP (Fig. 2C). These findings suggested that the differences in amino acid sequence/composition between rat and human L-FABP are sufficient to significantly alter the antigenic epitopes. In contrast, the human L-FABP T94A mutation did not elicit any additional differential response to antibodies against mouse or human L-FABP.

Figure 2.

SDS/PAGE and western blot analysis of rat, T94T WT and T94A variant human L-FABPs. (A) SDS/PAGE analysis of rat, T94T WT and T94A variant human L-FABPs (3 μg each lane). Lanes 1 and 5, Precision Plus protein standard (10 μL, Bio-Rad); lane 2, rat L-FABP; lane 3, T94T WT human L-FABP; lane 4, T94A variant human L-FABP. (B) Western blot analysis using antibody against mouse L-FABP. (C) Western blot analysis using antibody against human L-FABP.

Tyrosine fluorescence of rat, T94T and T94A L-FABP under denaturation conditions

To examine the impact of the differences in the respective amino acid sequence on L-FABP stability, tyrosine fluorescence spectra were recorded under increasing denaturation conditions using 8 m urea (Fig. 3A) and 6 m guanidinium chloride (GnHCl, Fig. 3B). While partial unfolding by urea increased the aqueous exposure of tyrosine residues in both rat and human L-FABPs, the tyrosines in rat L-FABP appeared more resistant to aqueous quenching upon urea-induced denaturation than that of human L-FABPs (Fig. 3A,B, solid line versus dotted and dashed lines). Among the human L-FABPs, the T94A variant appeared more resistant to urea-induced unfolding than T94T (Fig. 3A, dashed line versus dotted line). Use of GnHCl, a stronger protein unfolding agent, abolished these differences between the various types of L-FABPs (Fig. 3B).

Figure 3.

L-FABP tyrosine fluorescence under denaturation conditions. (A,B) Tyrosine fluorescence spectra of rat, human T94T WT and T94A variant in 8 m urea (A) and 6 m fuanidinium chloride (B) were recorded as described in 'Experimental procedures' (solid line, rat L-FABP; dotted line, T94T; dashed line, T94A). (C) Normalized difference spectra (normalized rat L-FABP minus normalized T94T) of tyrosine fluorescence emission. Representative spectra are shown.

Interestingly, qualitative comparison of the rat and human L-FABP tyrosine emission spectra suggested a broad shoulder at longer emission wavelength in rat but not human L-FABP (Fig. 3A,B). This was quantitatively confirmed when the tyrosine emission spectra of the human WT L-FABP T94T were normalized to those of the rat L-FABP and subtracted. The rat L-FABP exhibited an additional emission maximum near 330 nm in the folded state (i.e. buffer only), but shifted to 350 nm in the presence of the unfolding agents urea or GnHCl (Fig. 3C).

As neither rat nor human L-FABP contain a tryptophan residue, the additional emission peak was probably due to tyrosinate-like emission. Tyrosinate-like fluorescence emission occurs in proteins where one or more of the tyrosines are hydrogen-bonded through the phenolic hydroxyls, especially to neighboring basic amino acid residue(s) [glutamic acid (E) or aspartic acid (D)] [67]. The rat and human forms of L-FABP contain a common tyrosine, Y7, but the rat L-FABP also contains two additional tyrosines, Y54 and Y120. As tyrosinate-like emission is not evident in human L-FABP proteins, it appears that these additional residues are the most likely candidates for producing tyrosinate-like emission in the recombinant rat L-FABP. However, examination of the lowest energy-minimized solution-state structures of rat L-FABP revealed that only the hydroxyl of the Y7 tyrosine residue is involved in hydrogen bonding, with no evidence of such hydrogen bonding in the human L-FABP structures. In the rat apo-L-FABP structure (Protein Data Bank ID 2JU3), hydrogen bonding occurs between Y7 and the phenylalanine at position 3 (F3), but, interestingly, in the rat holo-L-FABP structure (Protein Data Bank ID 2JU7), hydrogen bonding was evident between Y7 and an aspartic acid residue, D107 [68]. However, as D107 appeared to be quite flexible and is located in a solvent-exposed region of a β-turn [68], it was unclear why the interaction was only subtly affected by denaturation in 8 m urea and 6 m GnHCl.

Rat, T94T and T94A L-FABP binding to 1-anilinonaphthalene-8-sulfonic acid (ANS)

As the human and rat L-FABPs differ significantly in amino acid sequence, charge and volume of the ligand-binding cavity [61-64], a displacement assay was developed to examine the impact of these variations on the binding affinity and specificity of these proteins for LCFAs and intermediates of TG synthesis. ANS was chosen for this assay because it is a synthetic fluorophore that has been used previously to examine the binding sites in FABPs [45, 69-71].

Although the fluorescence of ANS was very weak in buffer (Fig. 4A, solid line), it differentially increased upon binding to L-FABPs, with the highest emission intensity obtained for rat L-FABP. Emission spectra of ANS bound to human WT L-FABP T94T and the T94A variant were similar to each other but at lower intensity than for rat L-FABP (Fig. 4A). Quantitative analysis of multiple spectra established that the ANS maximal emission intensity was blue-shifted 3 nm and ~ 1.3-fold higher (< 0.05) when bound to rat L-FABP than human L-FABPs (Table 1). This suggests that, when bound to rat L-FABP, ANS was localized in a more hydrophobic environment, and/or more ANS was bound per protein molecule.

Table 1. ANS binding to rat, T94T and T94A L-FABP.
 RatT94TT94A
  1. Values are means SE (= 3–5).

  2. a

    < 0.05, human versus rat L-FABP.

  3. b

    < 0.05, T94A versus T94T.

Fluoresence emission maximum (nm)477480480
A.U./nM when fully bound0.78 ± 0.021.03 ± 0.01a0.98 ± 0.04a
Kdm)2.5 ± 0.22.3 ± 0.12.4 ± 0.1
B max 1.42 ± 0.010.80 ± 0.02a0.86 ± 0.01ab
Figure 4.

ANS binding to rat, T94T and T94A L-FABPs. (A) Representative ANS (35 μm) emission spectra when bound to L-FABPs (500 nm). (B) Representative reverse titration curves. ANS (100 nm) was titrated with increasing amounts of the L-FABPs. (C) Representative forward titration curves. L-FABPs (500 nm) were titrated with increasing amounts of ANS.

To distinguish these possibilities, reverse titrations were performed in order to obtain the fluorescence efficiency (fluorescence intensity/nm ANS) when ANS was fully bound to L-FABP. The fluorescence efficiency was in turn used in forward titrations to determine Kd values as described in 'Experimental procedures'. In reverse titrations, ANS (100 nm) was titrated with increasing amounts of L-FABP proteins, and ANS fluorescence intensity per nm ANS was plotted against L-FABP concentrations (Fig. 4B). Curve fitting of these data showed that the ANS fluorescence intensity/nm ANS fully bound to the L-FABPs was 0.8 for rat L-FABP and ~ 1.0 for both human L-FABPs (Table 1). Next, forward titration was performed by titrating 500 nm L-FABP with increasing amount of ANS. From the ANS fluorescence intensity/nm and total ANS concentration, the fractional saturation and free ANS concentration were calculated and the ANS binding curves were plotted (Fig. 4C).

Quantitative analysis of multiple binding curves showed that the ANS binding affinities (Kd values) of human and rat L-FABPs were similar (Table 1). In contrast, the Bmax values were significantly higher for rat L-FABP (1.4) than human L-FABPs (Bmax = 0.8 and 0.86) (Table 1). This is consistent with rat L-FABP having two binding sites for ANS, while human L-FABPs have only one. As ANS is a negatively charged molecule, its binding to proteins depends on both hydrophobicity and positive charges of the binding sites. As indicated by the ANS spectral shift above, the human L-FABP ligand binding site was a less hydrophobic environment. In addition, human L-FABP has three fewer positively charged amino acid residues than rat L-FABP (Fig. 1, + symbol).

Binding of long chain fatty acid (LCFA) to rat and human L-FABPs: ANS displacement assays

As changes in even a single amino acid may significantly alter the LCFA binding affinity and/or specificity of L-FABP and other FABPs [72-76], the specificity of rat and human L-FABPs for LCFAs was examined.

A variety of LCFAs effectively displaced ANS bound to both rat and human L-FABPs, including saturated palmitic and stearic acids (Fig. 5A,B), monounsaturated oleic acid (Fig. 5C), di-unsaturated linoleic acid (Fig. 5D), n–6 polyunsaturated arachidonic acid (AA) (Fig. 5E) and n–3 polyunsaturated EPA and DHA (Fig. 5F,G). In almost all cases, the LCFAs more completely displaced ANS from the human L-FABPs than from the rat L-FABP, consistent with the almost twofold higher quantity of ANS bound to the latter. The displacement curves of T94T and T94A were essentially superimposable, and indicated displacement of the ANS from its single binding site.

Figure 5.

ANS displacement from L-FABPs by LCFAs. L-FABP (500 nm) was incubated with ANS (35 μm) for 10 min. ANS displacement was performed by titration with small aliquots of ligands: (A) palmitic acid; (B) stearic acid; (C) oleic acid; (D) linoleic acid; (E) arachidonic acid; (F) eicosapentaenoic acid; (G) docosahexaenoic acid.

Quantitative analysis of multiple ANS displacement curves revealed that rat L-FABP bound LCFAs with the following order of binding affinities (Table 2): saturated (palmitic, stearic acid) and monounsaturated (oleic acid) LCFAs > di-unsaturated (linoleic acid) and polyunsaturated LCFAs (DHA > AA > EPA). Compared to rat L-FABP, human L-FABPs bound some LCFAs (palmitic acid, oleic acid) more strongly, but bound others slightly more weakly (linoleic acid) or with similar affinities (stearic acid, AA, EPA, DHA) (Table 2). For all LCFAs, there were only small differences in binding affinities between human T94T and T94A L-FABPs, with the T94A variant binding palmitic acid and linoleic acid slightly more weakly (Table 2). Taken together, the ANS displacement studies suggest that all three L-FABPs effectively bound LCFAs, with some preference for saturated and monounsaturated LCFAs. However, there was little difference between the human WT L-FABP T94T and T94A variant in terms of affinities for the various LCFAs.

Table 2. Impact of human L-FABP T94A variant on binding of LCFA and intermediates of TG synthesis.
 ANS displacement: Kim)a
LigandRatT94TT94A
  1. a

    Values are means ± SE (= 3–5); ND, no displacement. LPA binding to rat L-FABP was measured by tyrosine quenching rather than ANS displacement.

  2. b

    < 0.05, human versus rat L-FABP.

  3. c

    < 0.05, T94A versus T94T.

Fatty acids
Palmitic acid (16:0)0.048 ± 0.0010.038 ± 0.001b0.043 ± 0.001bc
Stearic acid (18:0)0.037 ± 0.0040.029 ± 0.0020.028 ± 0.001
Oleic acid (18:1n–9)0.043 ± 0.0020.035 ± 0.001b0.039 ± 0.001
Linoleic acid (18:2n–6)0.081 ± 0.0050.113 ± 0.005b0.132 ± 0.003bc
Arachidonic acid (AA, 20:4n–6)0.110 ± 0.0060.113 ± 0.0060.108 ± 0.006
Eicosapentaenoic acid (EPA, 20:5n–3)0.24 ± 0.030.20 ± 0.030.19 ± 0.01
Docosahexaenoic acid (DHA, 22:6n–3)0.081 ± 0.0150.072 ± 0.0050.065 ± 0.009
Intermediates of TG synthesis
Oleoyl CoA (18:1)2.41 ± 0.121.11 ± 0.04b1.03 ± 0.02b
LPA (18:1)99 ± 100.052 ± 0.007b0.036 ± 0.004b
2-OG (18:1)NDNDND
PODG (16:0, 18:1)NDNDND
POPA (16:0, 18:1)9.1 ± 0.82.8 ± 0.3b2.8 ± 0.2b

Binding of intermediates of TG synthesis by rat and human L-FABPs: ANS displacement assays

To avoid the toxicity elicited by accumulation of excess LCFAs and their active metabolites (LCFA CoAs), these lipids are rapidly oxidized or incorporated into TGs for storage or secretion [7, 10, 77, 78]. However, rat and human L-FABPs differ significantly in amino acid sequence [61], and changes in even a single amino acid may significantly alter the affinity and/or specificity of L-FABP and other FABPs for larger ligands (e.g. LCFA CoA, lysophosphatidic acid (LPA), and cholesterol) than LCFA [73, 79-81]. Therefore, the ability of the rat and human L-FABPs to bind intermediates in the synthesis of TGs was examined.

In the ANS displacement assays, only some intermediates in the synthesis of TGs effectively bound to rat and human L-FABPs, including oleoyl CoA (Fig. 6A), LPA (Fig. 6B) and 1-palmitoyl-2-oleoyl-glycero-3-phosphate (POPA) (Fig. 6E), but not mono- (2-OG) or diglycerides (PODG) (Fig. 6C,D). Ligands that did not displace ANS may have much weaker binding affinity than ANS. Quantitative analysis of multiple binding curves showed that these ligands were generally more weakly bound than LCFAs by the three L-FABPs in the order: LCFAs ~ LPA > oleoyl CoA >POPA >>> 2-OG or DOPG (Table 2). Both human L-FABPs bound oleoyl CoA, LPA and POPA more strongly than did the rat L-FABP (Table 2). However, the human L-FABP T94A variant did not significantly differ from the WT L-FABP T94T in terms of affinity for these ligands (Table 2).

Figure 6.

ANS displacement from L-FABP by intermediates of TG synthesis. L-FABP (500 nm) was incubated with ANS (35 μm) for 10 min. ANS displacement was performed by titration with small aliquots of ligands: (A) oleoyl CoA; (B) LPA (for rat L-FABP, the curve was obtained by tyrosine quenching); (C) 2-OG; (D) POPG; (E) POPA.

Effects of ligand binding on rat, human WT T94T and human T94A variant L-FABP protein secondary structure: circular dichroism

Ligand-induced conformational changes in L-FABP may significantly affect the transfer of ligand to and the interaction with target proteins [43-45, 82]. To resolve the impact of ligand binding on the secondary structures of L-FABPs, circular dichroic spectra were obtained for each protein in buffer with or without ligand, followed by secondary structure analysis as described in 'Experimental procedures'.

The secondary structure of the rat L-FABP was sensitive to LCFA binding and highly specific for the type of LCFA bound (Fig. 7A). There was no obvious pattern common to all LCFAs with regard to changes in the total proportions of helix, sheet, turn and unordered secondary structure elicited by LCFA binding to rat L-FABP (Fig. 7A and Table S1). Each was distinctive.

Figure 7.

Changes in the secondary structure of rat and human L-FABP upon interaction with stearic acid, arachidonic acid, EPA or DHA. L-FABP (0.5 μm) was examined by CD spectroscopy, and subsequent secondary structure analysis in the absence or presence of 5 μm ligand was performed as described in 'Experimental procedures'. The data are presented as the percentage change in secondary structure (L-FABP/ligand – L-FABP only) for rat L-FABP (A), T94T (B) and T94A (C) upon interaction with stearic acid (SA), arachidonic acid (AA), EPA and DHA. *< 0.05 for L-FABP/ligand secondary structure versus L-FABP only secondary structure; < 0.05 for T94A/ligand secondary structure versus T94T/ligand secondary structure.

LCFA-induced changes in WT human T94T secondary structure were overall somewhat smaller and differed from those induced by the same LCFA in rat L-FABP (Fig. 7B versus Fig. 7A). Each specific LCFA elicited a different pattern of secondary structure change on binding to human WT L-FABP T94T (Fig. 7B and Table S1) than to rat L-FABP (Fig. 7A and Table S1).

In marked contrast to the rat and human WT L-FABPs, the secondary structure of the human L-FABP T94A variant was relatively insensitive to LCFA binding. The secondary structure changes upon binding arachidonic acid, for example, were all < 8%, with most being only 1–5% (Fig. 7C, AA, and Table S1). Similar results were obtained for all other LCFAs examined. Further, human L-FABP T94A variant structural changes upon binding stearic acid, AA, EPA and DHA were significantly different from those upon binding to human WT L-FABP T94T (Fig. 7C, dagger symbol).

Intermediates of TG synthesis elicited much larger changes than LCFAs in terms of the secondary structure of all three L-FABPs (Fig. 8 versus Fig. 7, Table S2 versus Table S1). Furthermore, the conformational changes induced by binding of TG synthesis intermediates to the L-FABPs were in the opposite order: human L-FABP T94A variant > human WT L-FABP T94T > rat L-FABP. Ligands that did not bind or for which a Kd values could not accurately be resolved [2-oleoyl glycerol (2-OG) and 1-palmitoyl-2-oleoyl-sn-glycerol (PODG)] elicited no or only small changes in L-FABP secondary structure.

Figure 8.

Changes in the secondary structure of rat and human L-FABPs upon interaction with oleoyl CoA, LPA, POPA, 2-OG or PODG. L-FABP (0.5 μm) was examined by CD spectroscopy, and subsequent secondary structure analysis in the absence or presence of 5 μm ligand was performed as described in 'Experimental procedures'. The data are presented as the percentage change in secondary structure (L-FABP/ligand minus L-FABP only) for rat L-FABP (A), T94T (B) and T94A (C) upon interaction with oleoyl CoA, LPA, POPA, 2-OG and PODG. *< 0.05 for L-FABP/ligand secondary structure versus L-FABP only secondary structure; < 0.05 for T94A/ligand secondary structure versus T94T/ligand secondary structure.

With regard to the pattern of secondary structure changes upon binding these ligands, the rat and human L-FABPs differed markedly. The pattern of secondary structure changes induced by ligands [especially oleoyl CoA, oleoyl-l-α-lysophosphatidic acid sodium salt (LPA) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate monosodium salt (POPA)] in rat L-FABP was again somewhat dependent on the type of ligand (Fig. 8A and Table S2). In contrast, both human L-FABPs showed an overall similar qualitative pattern of secondary structure changes upon binding these ligands (Fig. 8B,C and Table S2). However, the secondary structural responses of human WT L-FABP T94T and T94A variant upon binding the respective ligands were significantly different (Fig. 8C and Table S2, dagger symbols).

Functional significance of the human L-FABP T94A mutation: impact on LCFA-mediated transcription of PPARα-regulated proteins in cultured primary human hepatocytes

Both murine and human WT L-FABP T94T directly interact with the respective PPARα [43-46, 82] to facilitate ligand induction of PPARα transcription of genes in LCFA metabolism [42, 47, 82-84]. While human WT L-FABP T94T and the T94A variant did not differ in terms of binding affinity for LCFAs such as EPA and DHA (Table 2), they differed markedly in secondary structural response to these ligands (Fig. 7). As both exogenous n–3 polyunsaturated LCFAs (EPA and DHA) as well as de novo LCFAs synthesized from glucose activate PPARα [47, 85], the functional significance of the T94A substitution on the ability of L-FABP to induce ligand-mediated PPARα transcriptional activity was examined in cultured primary human hepatocytes expressing WT L-FABP T94T or heterozygous or homozygous with respect to the L-FABP T94A variant, as described in 'Experimental procedures'.

In hepatocytes expressing the WT L-FABP T94T (genotyped as TT, Fig. 9A,B, black bars), both EPA and DHA induced transcription of all PPARα-regulated genes examined: PPARα itself, L-FABP, the key protein in cytoplasmic LCFA transport and nuclear targeting, and FATP5, the key plasma membrane fatty acid translocase in LCFA uptake. In contrast, expression of the human L-FABP T94A variant (genotyped as TC and CC), especially in hepatocytes homozygous for the T94A variant (genotyped as CC, Fig. 9A,B, open bars), significantly decreased or tended to impair the ability of EPA and DHA to induce PPARα transcription of PPARα, L-FABP and FATP5.

Figure 9.

Effect of the human L-FABP T94A mutation on ligand-induced transcription of PPARα-regulated proteins. Primary human hepatocytes were cultured overnight, and then incubated for 24 h with fatty acid-free BSA, EPA/BSA or DHA/BSA (200 μm EPA or DHA) in 6 mm glucose-containing medium as described in 'Experimental procedures'. RT-PCR was used to determine human PPARα, L-FABP and FATP5 mRNA levels normalized to an internal control (18S RNA). Values are fold changes induced by BSA/EPA or BSA/DHA complex relative to BSA only (means ± SEM, n = 8–10). *< 0.05, homozygous T94A (CC) and heterozygous (TC) variants compared to wild-type (TT); #< 0.05, homozygous T94A (CC) compared to heterozygous (TC) variants.

Thus, the L-FABP T94A variant impaired signaling to PPARα mediated by LCFAs (i.e. EPA, DHA, de novo synthesized LCFAs). The altered L-FABP T94A function correlated with the diminished ability of LCFA binding to alter the secondary structure of the L-FABP protein, rather than any alteration in LCFA binding affinity.

Discussion

The human L-FABP T94A variant is the most common coding polymorphism in the entire FABP family, occurring with 26–38% minor allele frequency and 8.3 ± 1.9% homozygosity in the human population (MAF for 1000 genomes in National Center for Biotechnology Information dbSNP database and the ALFRED database) [48-54]. It has been associated with lipid dysregulation, as evidenced by elevated plasma TGs [49, 56] and low-density lipoprotein cholesterol [49, 53], atherothrombotic cerebral infarction [51] and non-alcoholic fatty liver disease [53]. However, little is known about the mechanism(s) by which expression of the L-FABP T94A variant elicits its effects. To our knowledge, there have been no reports examining the effect of the T94A substitution on the ability of human L-FABP to bind LCFAs and/or intermediates of TG synthesis, structurally respond to binding of these ligands, or functionally respond to LCFA in primary hepatocytes. The studies described herein provide new insights into these issues.

First, the human L-FABP T94A variant was more resistant to unfolding by urea than WT human L-FABP T94T. The functional significance of the T94A resistance of T94A to unfolding may relate to its ability to interact with target proteins such as PPARα in the nucleus [42-46], glycerol-3-phosphate acyltransferase in the endoplasmic reticulum [4, 5, 7], carnitine palmitoyltransferase 1A in mitochondria [86], and FATP5 at the plasma membrane [36]. It was previously shown that WT L-FABP was more resistant to temperature unfolding [87]. The effect of chemical unfolding and thermal unfolding may be different. For example, a study with human serum albumin showed that guanidine and urea interacted with the protein by electrostatic forces, resulting in a random coiled conformation, while thermal denaturation produced a molten globe state and protein aggregation [88]. In a more recent study, it was reported that, during chemical denaturation, the protein conformations of the transition state were more extended than at high temperature, and the folding routes were different from thermal denaturation [89]. Therefore, chemical and thermal denaturation measure different aspects of protein stability. It was shown that thermal denaturation affected the L-FABP α-helical structure more than the β-sheet structure [87]. Interestingly, the human L-FABPs differed significantly from the rat L-FABP in terms of antigenicity, spectral properties and detergent stability. These species-dependent differences were attributed to the fact that almost 20% of the human WT L-FABP amino acid sequence is non-identical to rat L-FABP, and almost half of the substitutions are non-conservative [61].

Second, the T94A substitution did not alter or only slightly altered the affinity but not the specificity of ligand binding. While the human L-FABP T94A variant differed slightly from the WT L-FABP T94T in terms of affinity for a few LCFAs (palmitic acid and linoleic acid), it did not differ significantly in affinity for other LCFAs or the more complex intermediates of TG synthesis. Our previous report also found that T94T and T94A have the same (or similar) Ki values for phytanic acid, fenofibrate and fenofibric acid [87]. Our fluorescence binding data showed that rat L-FABP binds more than one molecule of ANS (Bmax = 1.4), in agreement a previous report [70]. However, our human L-FABP data showed that it only binds one ANS molecule (Bmax = 0.8). This is in agreement with a previous fluorescence binding assay in which human L-FABP was shown to bind one ANS molecule with a Kd value of 2.0 μm [90]. In contrast, an NMR study showed that human L-FABP bound two molecules of ANS [91]. However, the experiments performed therein utilized a different technique and under very different conditions than those described here. The concentrations of human L-FABP and ANS used for NMR experiments were much higher than those for the fluorescence binding assay, and the NMR experiments were performed at pH 5.5 while the fluorescence binding assay was performed at pH 7.4. As ANS has a negatively charged sulfonate group, electrostatic interactions influenced by pH play very important roles in ANS binding to proteins. For example, ESI-MS of ANS binding to various proteins determined that different amounts of ANS bound at different pH [92]. For apomyoglobin, the maximal ANS binding was observed at pH 4.0, at which each protein molecule contained one to six molecules of bound ANS, while at neutral pH, only a single molecule of ANS was bound [92]. As L-FABP has more protonated amino acids at pH 5.5 than at pH 7.4, this may account for the additional ANS binding at pH 5.5 [92], In addition, the second ANS bound by L-FABP may be in a more aqueous environment that is detectable by NMR but not fluorescence.

Third, the human and rat L-FABPs differed significantly in binding affinity, but not specificity for LCFAs. The human L-FABP exhibited small, but significant, differences in binding of the more saturated LCFAs (16:0, 18:1 and 18:2), but its affinity for the polyunsaturated LCFAs (20:4, 20:5 and 22:6) did not differ from that of rat L-FABP. Although two previous studies separately examined the LCFA binding profiles of rat [7, 9, 93] and human [94] WT L-FABPs, direct comparisons are difficult as different displacement assays were used in each of these studies, and both differed from that presented here. Both human and rat WT L-FABPs bound the key ligand substrates for TG synthesis, i.e. LCFA CoA (consistent with previous studies [7, 9, 93, 94]), lysophosphatidic acid (consistent with previous studies [93, 94]), and phosphatidic acid. L-FABP binding to monoacylglycerol is controversial depending on the technique used. Gel filtration chromatography of liver cytosol, solution NMR, Lipidex assays and a binding assay using the fluorescent monoacylglyceride (MG) analog 12-(9-anthroyloxy)oleoyl-sn-1-glycerol (MG12AO) showed that L-FABP binds MG [95, 96]. However, fluorescence displacement assays such as the ANS displacement assay used here and the 11-(dansylamino) undecanoic acid (DAUDA) displacement assay described previously [75] found little displacement. The exact reason for these differences is not clear. It is important to note that, while the key enzyme in TG synthesis from monoacylglycerides (i.e. monoacylglycerol acyltransferase) was first thought to be exclusively localized in human and rodent intestine, subsequent studies showed that monoacylglycerol acyltransferase was also significantly expressed in human (but not rodent) liver [97]. With respect to the other major intermediate in TG synthesis, neither human nor rat L-FABPs bound diglycerides (PODG). The present data demonstrate for the first time that human L-FABP bound several of the LCFA-derived intermediates of TG synthesis with two- to threefold higher binding affinity than did rat L-FABP.

Fourth, the L-FABP T94A variant exhibited a diminished secondary structural response to LCFA binding [56]. LCFAs altered the L-FABP secondary structure in the order: rat L-FABP > human WT L-FABP T94T >> human L-FABP T94A variant. Ligand-induced conformational changes in L-FABP are thought to facilitate ligand transfer from L-FABP to bound PPARα [43-45]. A recent NMR study showed that human WT T94T L-FABP binding of GW7647, another PPARα-selective drug, altered the ligand-binding cavity and the conformation of its portal region to stabilize/optimize ligand entry/exit from the β-barrel [45]. Indeed, L-FABP T94A variant-expressing human subjects are less responsive to reduction of elevated TG to basal levels by fenofibrate [56].

Fifth, binding of intermediates of TG synthesis to the human L-FABP T94A variant altered the secondary structure more than with the WT L-FABP. Intermediates of TG synthesis that were bound by L-FABP altered L-FABP secondary structure in the reverse order compared with LCFA: human L-FABP T94A variant > human WT L-FABP T94T >> rat L-FABP. L-FABP is known to facilitate glycerol-3-phosphate acyltransferase-mediated incorporation of LCFA CoA into lysophosphatidic acid, the key rate-limiting step in de novo phosphatidic acid and TG synthesis [4-7, 11]. Furthermore, the L-FABP conformation significantly determines the ability of glycerol-3-phosphate acyltransferase to facilitate incorporation of LCFA CoA into lysophosphatidic acid [6, 7]. Thus, the greater conformational change of the human L-FABP T94A variant in response to LCFA CoA binding suggests that this may facilitate synthesis of LPA (and thus phosphatidic acid and TG), accounting for the increased incidence of non-alcoholic fatty liver disease in human subjects expressing the L-FABP T94A variant [53].

Sixth, human L-FABP T94A variant-expressing human primary hepatocytes exhibited diminished LCFA-mediated transcription of PPARα-regulated proteins involved in LCFA uptake (FATP5), intracellular transport (L-FABP) and PPARα itself. L-FABP directly interacts with FATP5 at the mouse hepatocyte plasma membrane [36] and PPARα in the nucleus [42-46, 98]. These interactions enhance LCFA uptake [1, 34, 36, 99], LCFA β-oxidation [34, 36, 78, 86] and PPARα transcriptional activity [47, 83, 84]. Indeed, over-expression of human WT L-FABP T94T, but not the T94A variant, enhanced LCFA uptake in cultured transformed Chang liver cells [55]. Furthermore, fenofibrate (like EPA and DHA) was less effective in inducing PPARα transcription of LCFA β-oxidative enzymes (Fig. 9) and less effective in lowering elevated plasma TG to basal levels in L-FABP T94A variant human subjects [56]. Consequently, the T94A substitution elicits TG accumulation in human subjects expressing the T94A variant [53].

In summary, the T94A substitution in the human T94A variant L-FABP significantly altered the secondary structure and conformational response to LCFA binding. This in turn diminished the ability of the human L-FABP to facilitate LCFA induction of PPARα transcriptional activity in cultured primary human hepatocytes. WT L-FABP is known to mediate LCFA uptake [1, 34, 36, 99], transport through the cytoplasm [10, 32, 100], and targeting/co-transport into nuclei [35, 37-40]. Conversely, binding of EPA, DHA or xenobiotic ligands induces L-FABP translocation into the nucleus [47, 83, 84]. Inside nuclei, L-FABP directly interacts with PPARα [43-45] to facilitate LCFA [45] for induction of PPARα transcriptional activity [33, 35, 42, 47, 83, 84]. The sensitivity of L-FABP activity to subtle conformational differences elicited by single amino acid substitutions or conformers is illustrated by studies with the murine L-FABP [5-7, 9, 86]. The clinical significance of these findings is supported by the reduced effectiveness of fenofibrate in lowering elevated serum TGs to basal levels in human subjects expressing the L-FABP T94A variant [56]. Finally, the finding of similar or equal binding affinities for LCFAs demonstrates for the first time that the human L-FABP T94A variant is an altered-function rather than a functionless mutation whose effect is analogous to L-FABP ablation. Loss of L-FABP reduces LCFA and LCFA CoA binding by cytosolic proteins by 80–90%, as well as abolishing L-FABP facilitation of induction of PPARα transcriptional activity mediated by ligands such as PUFAs, fibrate, 5-(tetradecyloxy)-2-furoic acid (TOFA) or C57 in murine hepatocytes [47, 83, 84]. Expression of the T94A variant diminished the ability of PUFAs (EPA and DHA) to induce transcription of a variety of proteins involved in LCFA uptake and metabolism in cultured primary human hepatocytes.

Experimental procedures

Materials

Antibody against human L-FABP (H-120, a rabbit polyclonal antibody raised against amino acids 7–126 mapping within an internal region of L-FABP of human origin) was purchased from Santa Cruz Biotechnology (Dallas, TX, USA). Antibody against mouse L-FABP was raised in our laboratory as described previously [101, 102]. Any kD Mini-PROTEAN TGX pre-cast polyacrylamide gels and Precision Plus Dual Xtra protein standards were purchased from Bio-Rad (Hercules, CA, USA). SimplyBlue SafeStain was obtained from Invitrogen (Carlsbad, CA, USA). ANS was purchased from Life Technologies (Grand Island, NY, USA). Stearic acid, palmitic acid, oleic acid, linoleic acid, arachidonic acid (AA), cis-5,8,11,14,17-eicosapentaenoic acid (EPA), cis-4,7,10,13,16,19-docosahexaenoic acid (DHA), oleoyl CoA and oleoyl-l-α-lysophosphatidic acid sodium salt (LPA) were purchased from Sigma (St Louis, MO). 2-oleoyl glycerol (2-OG), 1-palmitoyl-2-oleoyl-sn-glycerol (PODG) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate monosodium salt (POPA) were obtained from Avanti Polar Lipids (Alabaster, AL, USA). All reagents and solvents used were of the highest grade available.

Proteins

Recombinant rat, T94T WT human and T94A mutant human L-FABPs were isolated, purified and delipidated as described previously [16, 87, 103]. Recombinant rat L-FABP and human WT T94T and T94A mutant L-FABP protein concentrations were analyzed by amino acid analysis, and molecular weights were confirmed by MALDI-TOF mass spectrometry (Protein Chemistry Laboratory, Texas A&M University, College Station, TX, USA). The delipidated, non-His-tagged recombinant proteins were shown to be > 98% pure by SDS/PAGE (3 μg protein/lane) utilizing an Any kD Mini-PROTEAN TGX pre-cast polyacrylamide gel followed by gel staining/destaining with SimplyBlue SafeStain according to the manufacturer's instructions.

Western blotting

To determine whether rat L-FABP, human WT L-FABP T94T and the human L-FABP T94A variant reacted equally well with rabbit polyclonal antisera against WT mouse or human L-FABP, western blotting was performed using the respective recombinant proteins and antibodies as described previously [34, 35, 43].

Fluorescence spectra and denaturation of recombinant L-FABP proteins

As rat and human L-FABP contain three and one tyrosine residues, respectively, but no tryptophan, L-FABP was excited at 280 nm, and fluorescence emission spectra of tyrosine residues were recorded between 295 and 420 nm using a Varian Cary Eclipse fluorescence spectrophotometer (Varian Inc., Palo Alto, CA, USA). Fluorescence emission intensities at equivalent quantities of tyrosine were obtained using 200 nm rat L-FABP, 600 nm WT human L-FABP T94T, and 600 nm human T94A variant. The temperature was maintained at 24 °C using a circulating water bath. The denaturation experiments were performed by recording L-FABP tyrosine fluorescence spectra in 8 m urea or 6 m guanidinium chloride at room temperature.

ANS binding to rat and human L-FABPs

ANS is essentially non-fluorescent in buffer, but its fluorescence increases dramatically upon binding to L-FABP. ANS fluorescence emission spectra were obtained by scanning from 410 to 600 nm with 380 nm excitation. In order to determine the Kd values of ANS binding to L-FABP, forward titration (500 nm L-FABP titrated with 0–48 μm ANS) and reverse titration (100 nm ANS titrated with 0–4 μm L-FABP) were performed. From curve fitting of the reverse titration, the fluorescence intensity of ANS (per nm) when fully bound to L-FABP was calculated. This parameter was then used to calculate the fractional saturation and free ANS concentration in the forward titration. Binding curves were constructed by plotting fractional saturation versus free ANS concentration, from which Kd and Bmax were determined by curve fitting.

Binding of LCFA and LCFA metabolites to rat and human L-FABPs

A solution of L-FABP (500 nm) and ANS (35 μm) was equilibrated and titrated with increasing amounts of LCFA, LCFA CoA or intermediates of TG synthesis. Displacement curves were constructed by plotting the percentage ANS fluorescence remaining versus ligand concentration. EC50 was obtained from the displacement curve. Ki values were calculated from the Kd values for ANS determined above, and from the EC50 according to the following equation: EC50/[ANS] = Ki/Kd.

For LPA binding to rat L-FABP, tyrosine quenching was used to construct the binding curve. Tyrosine emission spectra of rat L-FABP (200 nm) were recorded as described above with increasing amounts of LPA. The Kd value was calculated by fitting the curve for 100 minus percentage fluorescence remaining versus LPA concentration to a hyperbolic equation.

Circular dichroism spectroscopy

Circular dichroism (CD) spectroscopy measurements were performed utilizing a JASCO J-815 CD spectrometer (JASCO, Easton, MD, USA) equipped with a PFD-425S Peltier-type fluorescence detected circular dichroism attachment for temperature regulation. All CD experiments were performed at 0.5 μm protein (determined by amino acid analysis as above) in a buffer containing 10 mm potassium phosphate (pH 7.4) with or without 1% ethanol. The 1% ethanol had no effect on protein CD spectra or secondary structure analyses (data not shown). Prior to CD scanning from 185 to 250 nm, samples were incubated with stirring (250 rpm) at 25 °C for 10 min. The final CD spectrum, representing an average of ten scans, was background-subtracted and mathematically smoothed by the means-movement method using a convolution width of 5 using software provided by the manufacturer (JASCO, Easton, MD, USA). Analysis software provided by the manufacturer of the CD spectrometer was used to perform secondary structure analysis, using SDP (soluble and denatured protein) 48 as the reference set.

Ligand interaction CD studies of rat and human L-FABPs (0.5 μm protein in 10 mm potassium phosphate, pH 7.4) were performed as follows. LCFAs or intermediates of TG synthesis (5 μm) were added from stock solutions (500 μm in ethanol) such that the final ethanol concentration was 1%. Each sample was incubated with stirring at 25 °C for 10 min in the fluorescence detected circular dichroism attachment prior to obtaining the CD spectrum. The final CD spectrum (average of ten scans) was again background-subtracted (buffer/ligand/ethanol), mathematically smoothed, and secondary structure was analyzed as described above. The percentage change in secondary structure between samples was calculated using the following formula: [(% secondary structuresample 2 –% secondary structuresample 1) ÷% secondary structuresample 1] × 100%.

Quantitative RT-PCR for determination of PUFA-mediated induction of transcription of PPARα-regulated proteins in cultured primary human hepatocytes

Commercially obtained (Life Technologies) cryopreserved primary human hepatocytes from female Caucasian donors (50 ± 3 years old) were genotyped to determine WT T94T (TT), heterozygous (TC) or T94A (CC) variant expression as described previously [49, 50]. The hepatocytes were thawed, plated and cultured overnight according to the manufacturers' instructions (Life Technologies), and then incubated with 40 μm BSA (fatty acid-free) or BSA/DHA or BSA/EPA complexes (1:5 ratio of BSA to fatty acid) for 24 h in glucose-free William's E medium (US Biological, Salem, MA, USA), to which 6 mm glucose, 100 nm insulin and 10 nm dexamethasone had been added. Total mRNA was isolated from hepatocytes using an RNeasy kit (Qiagen, Valencia, CA, USA) and RNase-free DNase (Qiagen). All human mRNA analysis reagents (One-Step RT-PCR Master Mix, TaqMan gene expression assays and TaqMan probe) were obtained from Applied Biosystems (Grand Island, NY, USA). Human L-FABP, FATP5 and PPARα mRNA levels were determined according to the procedures provided by the manufacturer of the RT-PCR kits.

Statistics

One-way analysis of variance (ANOVA) combined with the Newman–Keuls multiple-comparisons post hoc test (graphpad prism version 3.03, San Diego, CA, USA) was used for all statistical analyses. Values are means ± SE of the mean (= 4–6). SigmaPlot 2002 for windows version 8.02 (SPSS, Chicago, IL, USA) was used for graphical analysis.

Acknowledgements

This work was supported in part by US Public Health Service/National Institutes of Health grants DK41402 (F.S. and A.B.K.) and DK70965 (B.P.A.).

Author contributions

Plasmids for producing the His-tagged L-FABP T94T and T94A proteins were created and provided by Shipra Gupta and Barbara Atshaves. Gregory Martin and Huan Huang purified the proteins. Huan Huang, Gregory Martin, Avery McIntosh, Kerstin Landrock and Danilo Landrock each performed assays under the direction of Ann Kier and Friedhelm Schroeder. Huan Huang and Friedhelm Schroeder wrote the paper. All authors have read and made corrections.

Ancillary