Fasting-induced HMGCS2 expression in the kidney does not contribute to circulating ketones

Abstract

Mitochondrial hydroxymethylglutaryl-CoA synthase 2 (HMGCS2) is the rate-limiting enzyme in ketogenesis. The liver expresses high levels of HMGCS2 constitutively as the main ketogenic organ. It has been suggested that the kidney could be ketogenic as HMGCS2 is expressed in the kidney during fasting and diabetic conditions. However, definitive proof of the capacity for the kidney to produce ketones is lacking. We demonstrated that during fasting, HMGCS2 expression is induced in the proximal tubule of the kidney and is peroxisome proliferator activated receptor-α dependent. Mice with kidney-specific Hmgcs2 deletion showed a minor, likely physiologically insignificant, decrease in circulating ketones during fasting. Conversely, liver-specific Hmgcs2 knockout mice exhibited a complete loss of fasting ketosis. Together, these findings indicate that renal HMGCS2 does not significantly contribute to global ketone production and that during fasting, the increase in circulating ketones is solely dependent on hepatic HMGCS2. Proximal tubule HMGCS2 serves functions other than systemic ketone provision.

NEW & NOTEWORTHY The mitochondrial enzyme hydroxymethylglutaryl-CoA synthase 2 (HMGCS2) catalyzes the rate-limiting step of ketogenesis. Although the liver constitutively expresses HMGCS2 and is considered the main ketogenic organ, HMGCS2 is induced in the kidney during fasting, leading to the proposal that the kidney contributes to fasting ketosis. We showed kidney HMGCS2 does not contribute to circulating ketones during fasting and cannot compensate for hepatic ketogenic insufficiency.

INTRODUCTION

Chronic kidney disease is a growing global health problem. Diabetes mellitus remains the leading cause of chronic kidney disease and end-stage kidney disease. The kidneys are often considered bystander “victims” of diabetes. As an organ that reabsorbs and synthesizes glucose, the kidney has an important role in glucose homeostasis and may contribute to systemic diabetic metabolic derangements (1). Evidence for abnormal renal fatty acid oxidation in diabetic and nondiabetic kidney disease suggests that dysregulated metabolism is a key component of kidney disease pathogenesis (2). Maintenance of fatty acid oxidation and related metabolic pathways are suggested to be important for kidney health. One of these metabolic pathways is ketogenesis, during which ketone bodies are produced from fatty acid oxidation. Although the liver is the main ketogenic organ (3), the rate-limiting enzyme for ketogenesis, mitochondrial hydroxymethylglutaryl-CoA synthase 2 (HMGCS2), is induced in the kidney during fasting (4) and in diabetes mellitus (5), suggesting a potential ketogenic role for the kidney. Paradoxically, HMGCS2 expression in the kidney has been proposed to be both potentially pathogenic in diabetic nephropathy (5) and beneficial in mediating the Na⁺-glucose cotransporter-2 inhibitor induced protection against diabetic kidney disease (6). Moreover, whether the kidney supplies ketones into the systemic circulation remains controversial (7). Thus, the function of renal HMGCS2 is unclear.

Previous data suggesting the existence of renal ketogenesis have relied on kidney HMGCS2 expression, indirect HMGCS2 activity measurements, tissue ketone measurements, and arteriovenous ketone concentration differences (4, 5, 8–10). Unfortunately, tissue and arteriovenous ketone measurements cannot distinguish between kidney- or liver-derived ketones. In addition, arteriovenous differences suffer from many limitations, including interconversion of metabolites and differential substrate utilization along the nephron. Even with tracers, high renal blood flow results in relatively small arteriovenous differences, and imprecision in measurements can lead to substantial errors in calculating metabolite flux (1). Classic work by Weidenmann and Krebs (11) observed ketogenesis from fatty acids in ex vivo kidney cortex slices and butyrate in perfused rat kidneys, albeit at much lower rates compared with the liver. One limitation of these classic experiments includes the use of supraphysiological substrate concentrations, which can disrupt the redox balance, resulting in nonphysiological reactions. In addition, butyrate is not a physiological substrate for ketogenesis. Although kidney ketone production has been inferred in the context of diabetes (5) and autophagy-deficient mouse models (4), definitive proof of kidney ketogenesis is lacking (7). Thus, whether renal ketogenesis contributes to the systemic pool of ketones is unclear and remains to be determined.

METHODS

Mice

All animal experiments were performed in accordance with institutional regulations after protocol review and approval by the Institutional Animal Care and Use Committees of Yale University and the University of Texas Southwestern Medical Center. The Hmgcs2^fl/fl conditional knockout mouse model was generated by CRISPR-Cas9 methodology as previously described (12). In brief, T7-sgRNA templates were prepared by PCR, incorporating the guide sequences from the desired target regions in the mouse Hmgcs2 locus on chromosome 3 (Mus musculus strain C57BL/6J, National Center for Biotechnology Information Gene ID:15360, GRCm39 Reference Annotation Release 109), with a 5′-guide sequence of AGTCACCCAAACCTGACTTG (sense orientation) and a 3′-guide sequence of ACTGGCTAAGATAACCAAGA (sense orientation). The T7-sgRNA PCR templates were used for in vitro transcription (MEGAshortscript T7 Transcription kit, Thermo Fisher Scientific) and purification (MEGAclear Transcription Clean-Up kit, Thermo Fisher Scientific). Cas9 mRNA (CleanCap, 5-methoxyuridine modified) was purchased from TriLink Biotechnologies. The homology-directed repair templates containing the loxP sites were purchased as Ultramer DNA oligonucleotides (Integrated DNA Technologies). C57BL/6N mice were obtained from Charles River Laboratories, and cytoplasmic microinjections of sgRNAs (50 ng/µL each), loxP oligos (100 ng/µL), and Cas9 mRNA (100 ng/µL) into single-cell embryos at 0.5 days postconception were performed by the Molecular Genetics Core of Yale Diabetes Research Center. Hmgcs2^fl/fl mice harbor loxP sites in the first and second introns, at positions 8,543 and 12,102 in the Hmgcs2 gene, respectively. Cre-mediated excision removes the entire second exon of 455 bp, resulting in a translational frameshift. Table 1 shows the genotyping primers.

Table 1.

Genotyping primers

Gene	Name	Primer Sequence (5′-3′)
AlbCreERT2	gABV93	GGAACCCAAACTGATGACCA
	gABT290	ATCATTTCTTTGTTTTCAGG
	gABT294	TTAAACAAGCAAAACCAAAT
Six2TGC Insert	gSix2TGC F1	GCTTCACGCAGGAGCAAGT
	gSix2TGC R1	TGACTTTGCTCTTGTCCAGTC
Hmgcs2	gHmgcs2-3 F2	AGGACCTTGGCCATCTCTGTATTT
	gHmgcs2-3 R2	GATTGATGCTGGCACCCTCCTA
Ppara KO	gPpara F	GAGAAGTTGCAGGAGGGGATTGTG
	gPpara R1	CCCATTTCGGTAGCAGGTAGTCTT
	gPpara R2	GCAATCCATCTTGTTCAATGGC

Hmgcs2, hydroxymethylglutaryl-CoA synthase 2; Ppara KO, peroxisome proliferator activated receptor-α knockout.

Alb-CreERT2 mice (13) were a gift from Dr. Pierre Chambon. Six2-Cre mice [Tg(Six2-EGFP/cre)^1Amc/J, No. 009606] and peroxisome proliferator activated receptor-α (PPAR-α) knockout (KO) (Ppara^−/−; Ppara^tm1Gonz/J, No. 008154) mice were purchased from Jackson Laboratories. Alb-CreERT2;Hmgcs2^fl/fl mice (8–10 wk old) were injected intraperitoneally with tamoxifen in peanut oil (100 mg/kg body wt, Sigma) daily for five doses and then allowed a rest week before experimental use. All mouse strains were maintained on a C57BL/6 background and housed under standard laboratory conditions with a 12:12-h light-dark cycle. Male mice were provided ad libitum access to water and standard chow (No. 2916, Harlan Teklad) or fasted for 24 h.

Plasma Metabolite Measurements

Blood glucose was determined by whole blood via a tail vein prick and measured via a glucometer (OneTouch). For other tests, retroorbital or submandibular blood was harvested, and plasma was isolated using lithium heparin-coated plasma separator tubes (BD). Total ketones were measured using enzymatic colorimetric assay kits (Wako Diagnostics). Plasma creatinine was assayed using capillary electrophoresis by The George M. O’Brien Kidney Research Core at the University of Texas Southwestern Medical Center.

RNA Extraction and Quantification

For tissue RNA extraction, tissues were harvested into RNA Bee or RNA-STAT RNA isolation reagent (Tel Test) and disrupted by bead homogenization in Fisherbrand Prefilled Bead Mill Tubes using a Fisherbrand Bead Mill 24 Homogenizer. RNA was extracted using Direct-Zol (Zymo Research) kits per the manufacturer’s protocol. cDNA synthesis was performed using Moloney murine leukemia virus reverse transcriptase (Clontech) with oligo(dT) primers. Quantitative RT-PCRs were performed on a QuantStudio7 Flex (Applied Biosystems) using iTaq Universal SYBR Green Supermix (Bio-Rad). Transcript levels were normalized to ribosomal protein L13a (Rpl13a). Primers (Sigma) used for quantitative RT-PCR are shown in Table 2.

Table 2.

Quantitative RT-PCR primers

Gene	Forward (5′-3′)	Reverse (5′-3′)
Acat1	TCTGGAACACGGTCTTGAGC	TCACGGCAGGAACAGGATAC
Bdh1	TGATGCTGGGGTCAAGGAAC	CCAAACGTTGAGATGCCTGC
Hmgcl	TTTCCCGGCATCAACTACCC	CGAGGGCACAGGAGACATAC
Hmgcs1	GCAGAAATCTCTAGCTCGGATG	GACACAAGCAAAGACGCCTT
Hmgcs2	GAGGGCATAGATACCACCAACG	AATGTCACCACAGACCACCAGG
Rpl13a	GAGGTCGGGTGGAAGTACCA	TGCATCTTGGCCTTTTCCTT

Acat1, acetyl-CoA acetyltransferase 1; Bdh1, 3-hydroxybutyrate dehydrogenase 1; Hmgcl, 3-hydroxy-3-methylglutaryl-CoA lyase; Hmgcs1 and Hmgcs2, hydroxymethylglutaryl-CoA synthase 1 and 2, respectively; Rpl13a, ribosomal protein L13a.

Immunoblot Analysis

Harvested tissues were snap frozen in liquid nitrogen and then bead homogenized in RIPA buffer (Teknova) supplemented with HALT protease and phosphatase inhibitors (Thermo Fisher Scientific). Twenty micrograms of protein/sample were loaded into 4%–20% Mini-PROTEAN TGX stain-free polyacrylamide gels (Bio-Rad), transferred onto activated PVDF membranes (Bio-Rad), blocked in 5% milk in Tris-buffered saline with Tween 20 for 30 min, and incubated with primary antibodies (Table 3) overnight at 4°C. Membranes were washed and then incubated with secondary antibodies (Table 3) for 1 h at room temperature. Protein was visualized using enhanced chemiluminescence reagent (Bio-Rad), and densitometry was determined based on total protein in Image Lab (Bio-Rad).

Table 3.

Antibody/staining reagents

Target	Vendor	Catalog Number/RRID	Dilution
Anti-HMGCS2	Abcam	ab137043/AB_2749817	1:3,000 for Western blots; 1:150 for immunohistochemistry
Donkey anti-rabbit IgG Alexa Fluor 594	Invitrogen	A21207/AB_141637	1:200
Lotus tetragonolobus lectin-fluorescein	Vector Laboratories	FL-1321-2/NA	1:100
Anti-rabbit horseradish peroxidase	Jackson ImmunoResearch	111-035-003/AB_2313567	1:5,000
DAPI	Thermo Fisher Scientific	62248/NA	1:10,000

HMGCS2, hydroxymethylglutaryl-CoA synthase 2; NA, not applicable.

Immunostaining

Tissues were harvested and fresh frozen in OCT for immunofluorescent staining. Kidney sections (6 µm) were fixed with 4% paraformaldehyde for 30 min and blocked in 5% normal donkey serum for 1 h at room temperature. Primary antibodies (Table 3) were incubated overnight at 4°C, and secondary antibodies were incubated for 1 h at room temperature. Images were taken with a Nikon Eclipse 80i microscope using a Nikon DS-Fi3 camera and processed in ImageJ.

Statistics

Statistical analyses were performed using Prism 9.0 (GraphPad). Student’s unpaired two-way t test and two-way ANOVA with Tukey’s multiple comparison analysis were used when appropriate. Data are expressed as means ± SD. P values of ≤0.05 were considered statistically significant.

RESULTS

Kidney HMGCS2 Is Induced in Fasting

In the fed state, HMGCS2 was constitutively expressed in the liver and colon but was not detectable in the kidney (Fig. 1A). After a 24-h fast, blood glucose decreased and circulating ketones were elevated (Fig. 1, B and C). HMGCS2 expression was slightly increased in the liver and colon but was significantly induced in the kidney (Fig. 1A). Transcript levels of other enzymes in the ketogenic pathway (Fig. 1D) did not change significantly in the kidney during fasting (Fig. 1E).

Figure 1.

Kidney hydroxymethylglutaryl-CoA synthase 2 (HMGCS2) is induced during fasting. A: whole tissue protein lysates immunoblotted for HMGCS2 protein. Representative blots with densitometry are shown as fold changes relative to the respective fed controls. n = 8 mice/group. B: plasma ketones. n = 8 mice/group. C: plasma glucose. n = 8–16 mice/group. D: ketogenic pathway highlighting key ketogenic enzymes. E: whole tissue mRNA expression of ketogenic genes. n = 5 mice/group. F: representative images of kidney and liver sections immunostained for HMGCS2 (red), DAPI (blue), and Lotus tetragonolobus lectin (LTL; green). Scale bars = 100 µm for ×10 images and 50 µm for ×20 images. Statistics were determined by an unpaired two-sided t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Fast, fasted; Rpl13a, ribosomal protein L13a. ACAT1, acetyl-CoA acetyltransferase 1; HMGCL, 3-hydroxy-3-methylglutaryl-CoA lyase; BDH1, 3-hydroxybutyrate dehydrogenase 1; BHB, β-hydroxybutyrate.

Because the kidney is a heterogenous organ and each kidney cell type has a unique function, we next determined which tubule segments express HMGSC2. In the fasted state, HMGCS2 was expressed in the renal cortex and exclusively in Lotus tetragonolobus lectin-positive proximal tubular cells (Fig. 1F). Conversely, liver HMGCS2 expression was homogeneous.

Kidney HMGCS2 Expression Is PPAR-α Dependent

PPAR-α is a known transcriptional regulator of hepatic Hmgcs2 (14). As expected, fasting ketosis was significantly blunted in Ppara^−/− mice (Fig. 2A). To determine whether PPAR-α also regulates renal Hmgcs2, we examined Hmgcs2 mRNA and HMGCS2 protein expression in Ppara^−/− mice. Fasting Hmgcs2 mRNA expression was suppressed in both the liver and kidney (Fig. 2C). Interestingly, fasting-induced HMGCS2 protein expression in the kidney was completely dependent on PPAR-α, whereas hepatic HMGCS2 protein expression was not significantly changed in Ppara^−/− mice (Fig. 3D).

Figure 2.

Kidney hydroxymethylglutaryl-CoA synthase 2 (HMGCS2) is peroxisome proliferator-activated receptor-α dependent. A: plasma ketones. n = 6–10 mice/group. B: blood glucose. n = 10–18 mice/group. C: whole tissue Hmgcs2 mRNA expression. n = 5–9 mice/group. D: whole tissue protein lysates immunoblotted for HMGCS2 protein. Representative blots with densitometry are shown as fold changes relative to wild-type (WT) fed controls. n = 6–8 mice/group. Statistics were determined by two-way ANOVA with a multiple comparisons test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Fast, fasted; PPAR-KO, peroxisome proliferator activated receptor-α knockout; Rpl13a, ribosomal protein L13a.

Figure 3.

Kidney hydroxymethylglutaryl-CoA synthase 2 (HMGCS2) does not contribute to circulating ketones. A: whole kidney mRNA expression of Hmgcs2. n = 3–9 mice/group. B: baseline plasma creatinine. n = 14–15 mice/group. C: representative blots of whole tissue protein lysates immunoblotted for HMGCS2 protein. D: densitometry of the results in C, shown as fold changes relative to respective wild-type (WT) fasted (Fast) controls. White bars indicate the WT group; gray bars indicate the kidney-specific knockout (K-KO) group. n = 5 mice/group. E: plasma ketones. n = 9–15 mice/group. F: blood glucose. n = 9 mice/group. G: whole liver mRNA expression of Hmgcs2. n = 6–9 mice/group. H: representative blot of whole tissue protein lysates immunoblotted for HMGCS2 protein. I: densitometry of the results in H, shown as fold changes relative to the respective WT fed controls. White bars indicate the WT group; gray bars indicate the liver-specific knockout (L-KO) group. n = 3–9 mice/group. J: plasma ketones. n = 6–8 mice/group. K: blood glucose. n = 8–9 mice/group. Statistics were determined by two-way ANOVA with a multiple comparisons test (A, E–G, and I–K) or an unpaired two-sided t test (B and D). *P < 0.05, **P < 0.01, ****P < 0.0001. L-KO, liver-specific knockout; Rpl13a, ribosomal protein L13a.

Kidney HMGCS2 Does Not Contribute to Circulating Ketones

We sought to determine whether renal HMGCS2 enables the kidney to contribute to the pool of circulating ketone bodies. We generated mice with an Hmgcs2 floxed allele (Hmgcs2^fl/fl; see methods). To determine the role of kidney HMGCS2 in fasting ketosis, we generated a kidney-specific Hmgcs2 KO mouse (Six2-Cre;Hmgcs2^fl/fl; referred to as K-KO), which deletes Hmgcs2 in nephron progenitors (Fig. 3, A, C, and D). K-KO mice were fertile and did not exhibit any overt developmental abnormalities or defects in baseline renal function (Fig. 3B). Kidney-specific deletion of Hmgcs2 led to a minor decrease in circulating ketone levels during fasting (Fig. 3E). To further determine whether kidney HMGCS2 compensates for the loss of hepatic ketogenesis, we generated a liver-specific Hmgcs2 KO mouse. Hmgcs2 knockdown by systemic antisense oligonucleotides leads to steatosis (15), so we developed an inducible hepatocyte Hmgcs2 KO mouse model (Alb-CreERT2;Hmgcs2^fl/fl; referred to as L-KO) to avoid this phenomenon (Fig. 3, G–I). Hepatic Hmgcs2 deletion resulted in a complete loss of fasting ketosis, as circulating ketone levels in L-KO mice after a 24-h fast were no different than in the fed state (Fig. 3J). These data suggest that the circulating ketones induced during fasting are completely dependent on hepatic HMGCS2. Thus, renal HMGCS2 does not compensate for impaired hepatic ketogenesis.

DISCUSSION

Maintaining metabolic homeostasis is crucial when nutrient availability is limited, such as during fasting, as well as for the prevention of metabolic disorders in states of nutrient excess, such as obesity and diabetes. Metabolic homeostasis requires the communication and cooperation of multiple organs. The liver contributes to metabolic homeostasis by regulating the balance among global lipid, glucose, and ketone metabolism. Liver ketogenesis is not only an important metabolic response to fasting but also plays a role in preventing diet-induced fatty liver disease (15). Insufficient ketogenic capacity can lead to altered tricarboxylic acid cycle function, resulting in hepatic injury, inflammation, and fibrosis (16, 17).

The kidney is another organ involved in metabolic maintenance; however, its role in whole body metabolism is incompletely understood. The kidney is involved in the regulation of glucose metabolism via glucose reabsorption and gluconeogenesis (18). Evidence suggests that the kidney can compensate for hepatic defects in gluconeogenesis. Mice with liver-specific deletion of either phosphoenolpyruvate carboxykinase 1 (Pck1) or glucose 6-phosphatase (G6pc) gluconeogenic genes can maintain blood glucose levels during fed and fasted states, suggesting extrahepatic gluconeogenesis in the absence of liver glucose production (19–21). Isotope tracing further supports renal gluconeogenesis in liver specific Pck1 KO mice as the primary mechanism of maintaining blood glucose (21). Previous studies have suggested that the kidney can contribute to circulating ketones (4, 9, 10). However, these studies used indirect measurements and models with confounding variables. Here, we present novel, tissue-specific deletion models directly targeting Hmgcs2. Although kidney-specific deletion of Hmgcs2 leads to a very minor decrease in circulating ketones, it is unclear if this observed decrease is physiologically relevant or significant. However, when Hmgcs2 is deleted in the liver, we observed a near complete loss of circulating ketones, suggesting the liver is indeed the main and likely only source of circulating ketones. Despite higher renal HMGCS2 protein levels during fasting in the absence of liver Hmgcs2, the kidney does not contribute to the pool of circulating ketones. The significance of and the mechanisms by which kidney HMGCS2 expression is higher in L-KO mice are unknown.

We next investigated whether regulation of Hmgcs2 expression differed between the kidney and liver to better understand its possible function. We found that both Hmgcs2 transcript and HMGCS2 protein induced in the kidney during fasting were eliminated in Ppara^−/− mice. Although fasting liver Hmgcs2 mRNA expression was attenuated in Ppara^−/− mice, liver HMGCS2 protein was not altered. It is worth noting the Ppara^−/− mice exhibited a slight reduction in circulating ketones during fasting, unlike the complete loss observed in L-KO mice. The lack of change in hepatic HMGCS2 protein expression in Ppara^−/− mice may suggest that the decrease of circulating ketones is directly related to the loss of renal HMGCS2 protein. However, K-KO mice do not exhibit the same difference in fasting-induced ketones as in Ppara^−/− mice. As the K-KO model uses the Six2-Cre driver that is activated during development, the congenital loss of renal HMGCS2 could result in hepatic compensation in the liver. If the loss of kidney HMGCS2 is the cause of the lower ketone levels in Ppara^−/− mice, one would expect L-KO mice to have plasma ketones levels approximating the deficit observed in Ppara^−/− mice. However, this was not observed. The data suggest that hepatic HMGCS2 protein expression is only partially regulated by PPAR-α. Posttranslational modifications affecting HMGCS2 activity could explain these observations, as sirtuin 3 (Sirt3), a downstream PPAR-α target gene (22), deacetylates HMGCS2, increasing its activity (23).

In conclusion, fasting induces ketogenesis in the liver and HMGCS2 expression in the kidney. Our data support the notion that the liver is the main ketogenic organ supplying the periphery with ketones during fasting metabolic adaptation. Although the kidney does not contribute to circulating ketones during fasting, we are currently investigating other possible functions of kidney HMGCS2, such as local ketone production. Both the colon and retina exhibit local ketone production (24, 25). Fasting and a ketogenic diet can induce HMGCS2 expression in the heart without evidence of de novo ketogenesis (26). As both the kidney proximal tubule and heart preferentially oxidize fatty acids for energy, it is worth considering the possibility that local HMGCS2 could be modifying fatty acid oxidation. Although ketones can enhance renal gluconeogenesis (27), we did not observe a difference in blood glucose levels in either L-KO or K-KO fasted mice (Fig. 3, F and K). Other ketone functions, such as chromatin remodeling via histone deacetylase inhibition and β-hydroxybutyrylation, could also be operative (28, 29). More research is needed to understand the function of kidney HMGCS2 and determine whether it is involved in homeostatic metabolic adaptation or dysfunction, either locally within the kidney or systemically.

GRANTS

This work was supported by The George M. O’Brien Kidney Center at Yale [National Institutes of Health (NIH) Grant P30DK079310, to S.C.H.], NIH Grant R35GM137984 (to S.C.H.), an American Society of Nephrology Carl W. Gottschalk Research Scholar Grant (to S.C.H.), NIH T32 Training Grant T32DK007257 (to A.H.V.), the Yale Diabetes Research Center-Molecular Genetics Core (NIH Grant P30DK045735), and the George M. O’Brien Kidney Research Core at the University of Texas Southwestern Medical Center (NIH Grant P30DK079328).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.H.V. and S.C.H. conceived and designed research; A.H.V., L.E.L., K.F., J.S., and T.B. performed experiments; A.H.V. and S.C.H. analyzed data; A.H.V. and S.C.H. interpreted results of experiments; A.H.V. and S.C.H. prepared figures; A.H.V. and S.C.H. drafted manuscript; A.H.V., L.E.L., and S.C.H. edited and revised manuscript; A.H.V., L.E.L., K.F., J.S., T.B., and S.C.H. approved final version of manuscript.