In order to gain insight into the pathogenesis of neurodegeneration in MHBD deficiency and to obtain further information on the biological role of HSD10, we investigated additional patients with MHBDD. The condition was retrospectively diagnosed in one child (case 1, Supporting Information) with a very severe neonatal presentation, absent neurological development and death from progressive hypertrophic cardiomyopathy at the age of 7 months. MHBD activity in fibroblasts, as determined in two independent laboratories, was only partially reduced to approximately 30% of normal. This was considerably higher than for other patients with MHBDD. Sequence analysis of the complete coding region of the HSD17B10 gene in this patient revealed hemizygosity for a novel mutation p.D86G (c.257A>G) in exon 3; no other mutation was identified. MHBDD was also diagnosed in a boy (case 2, Supporting Information) who presented with pre- and postnatal failure to thrive but normal cognitive and motor development. Neurological examination in this boy and two affected relatives (brother and cousin) has been entirely normal up to the present age of 8 years. There was no measurable MHBD activity in fibroblasts; molecular studies identified hemizygosity for the novel mutation p.Q165H (c.495A>C) in exon 5 of the HSD17B10 gene.

Observations in these patients indicated that the development and severity of symptoms in MHBDD is unrelated to residual enzyme activity. In order to confirm these findings we carried out in vitro expression analyses of WT HSD10 and the relevant mutations R130C (the common mutation), D86G (associated with a very severe clinical phenotype) and Q165H (associated with normal development). WT and mutants were expressed as stable, soluble proteins (Supporting Information Fig 1). Further insights were gained through the 3D structure determination of human HSD10 (Fig 1A, B), which crystallized in spacegroup P3₂21 and was determined to high resolution (1.2 Å; data collection and refinement details in Supporting Information Table 1). Kinetic constants were determined with different substrates (Fig 1E, F). Kinetic parameters calculated for the WT enzyme were in line with earlier data reported for this enzyme (Shafqat et al, 2003). Residual dehydrogenase activities of 64 and 28% were measured for mutations R130C and D86G, respectively; however, mutation R130C was unstable at room temperature and steadily lost enzyme activity (Fig 1C). Protein stability was assessed with and without NAD⁺ or NADH cofactor as a function of temperature by differential scanning fluorimetry (Fig 1D). Cofactor binding of NAD⁺ or NADH, as reflected by increased thermal stability, was observed for both WT and D86G. In contrast, mutation Q165H located at the active centre of the enzyme showed neither residual activity nor cofactor binding in our experimental settings. Inability of the mutation Q165H to bind the cofactor indicates that regardless of the substrate, no enzymatic reaction requiring NADH or NAD⁺ (all known reactions) will be sufficiently catalysed by this mutant protein. WT and all mutations alike showed identical mitochondrial distribution (data not shown). The lack of correlation between disease severity and residual enzyme activity in MHBDD patients thus is not due to misdirection of intracellular transport or impaired mitochondrial localization of mutated HSD10. Mitotracker staining revealed a filamentous network-like structure of the mitochondria in WT cells and fibroblasts carrying the Q165H mutation. Cells with the R130C and D86G mutations showed punctate and fragmented mitochondrial organization (Fig 1G–J). Despite the differences in the Mitotracker staining the amount of mitochondrial material is not reduced in patient cells judged by the amount of mitochondrial marker Grp75 protein and TOMM20 mRNA (Supporting Information Fig 2). The amount of HSD10 transcripts was nearly identical in the cells analysed (WT, R130C and Q165H). In agreement with the finding that the R130C mutation is unstable (Fig 1C) less HSD10 protein could be detected in fibroblasts carrying this mutation (Supporting Information Fig 2).

Since HSD10 WT and mutations are localized to mitochondria and patients with MHBDD show signs of mitochondrial dysfunction, we analysed mitochondrial morphology by EM in fibroblasts derived from patients compared with control fibroblasts (Fig 1K and Supporting Information Fig 3). The majority of mitochondria in control cells showed WT morphology (dense, dark). Fibroblasts carrying the Q165H mutation maintained WT morphology of 45% of the mitochondria, 30% displayed an intermediate phenotype with loosely packed and/or swollen cristae and 27% showed depletion of cristae and appeared ‘empty’ (for details, see Material and Methods). In contrast, in cells carrying the D86G and R130C mutation 65–85% of the mitochondria displayed an aberrant phenotype (group 2 and 3). The EM analysis of mitochondria of HSD10 mutant human cells thus showed that HSD10 is required for mitochondrial integrity. This function is not correlated with residual enzyme activity.

Since knock-out of Hsd17b10 in mice results in early embryonic lethality at gastrula stages (B. Arnold, unpublished data), we established two conditional Hsd17b10 knock-out mouse lines. In one line, Hsd17b10 was eliminated in endothelial cells and cells of the immune system by mating mice carrying a floxed allele of Hsd17b10 with a Tie2 promoter driven Cre recombinase strain. These mice are viable and fertile, but have defects in spleen and vasculature. The animals die rapidly around week 25. Supporting Information Fig 4 shows the recombination efficiency in spleen cells and dendritic cells of Tie2-Cre × Hsd17b10 mice as an example.

To generate the second conditional knock-out line, mice carrying a floxed allele of Hsd17b10 were mated with a DBH-Cre line which eliminates the Hsd17b10 gene in noradrenergic neurons. These mice are viable and fertile but die around week 26.

In the central nervous system (CNS) the locus coeruleus contains noradrenergic neurons and we set out to analyse the morphology of the mitochondria in this region by EM in conditional knock-out mice. In the loci coerulei of DBH-Cre HSD10 deficient mice, almost 30% of the mitochondria showed depletion of cristae and appeared ‘empty’, more than 50% of the mitochondria were loosely packed and had swollen cristae, while normal morphology (dense, dark) was only found in 20% (Fig 2, Supporting Information Fig 5). In the cerebellum of the same mice, which lacks noradrenergic neurons, 50% of the mitochondria had WT morphology, 40% displayed the intermediate phenotype and only 10% appeared ‘empty’ and without cristae. A similar picture was observed in the peripheral nervous system (PNS). Mitochondria were morphologically severely altered in superior cervical ganglia isolated from DBH-Cre conditional knock-out mice (Supporting Information Figs 6 and 7). The EM analysis of two independent primary cell types (neurons of the CNS and PNS) from Hsd17b10 DBH-Cre conditional knock-out mice thus confirmed in a genetic system that HSD10 is required for mitochondrial structural integrity in the CNS and the PNS and that the observed changes in mitochondria are cell type independent.

In order to study HSD10 loss of function in early vertebrate embryos we turned to Xenopus as an experimental model. We cloned the Xenopus homologue of HSD10 (xHSD10) and performed RT-PCR analysis, as well as whole mount in situ hybridization during the first 3 days of embryonic development. HSD10 mRNA is provided maternally, and zygotic HSD10 transcription was found from neurula stages onward (Supporting Information Fig 8). In Xenopus embryos, translation of mRNAs can be specifically blocked by Morpholino (Mo) antisense oligonucleotides. Two Mo antisense oligonucleotides were designed to inhibit HSD10 mRNA translation. Mo^5′UTR targeted the 5′UTR and Mo^ATG the coding region of xHSD10. Both oligos specifically suppressed HSD10 translation in Xenopus embryos (Fig 3A).

Next we tested whether HSD10 knock-down impaired mitochondrial function. Antisense Mo^ATG was microinjected into the prospective ectoderm and the injected region (animal cap) was explanted and cultured. Explanted animal caps differentiate into a homogeneous tissue, the so-called atypical epidermis. After 2 days in culture, when control embryos had reached tailbud stage, pyruvate turnover was measured as a parameter of mitochondrial function. Mo^ATG-injected samples displayed a 40% reduction in pyruvate turnover compared to the uninjected or control Mo (Mo^Co) injected tissue. Impaired mitochondrial function was specific because co-injection of a human HSD10 expression construct which could not be blocked by the HSD10 Mo^ATG rescued pyruvate turnover almost to the level of uninjected controls (Fig 3B).

EM analysis revealed that the morphology of mitochondria in animal cap explants was changed after HSD10 knock-down (Fig 3C, D, Supporting Information Fig 9). Mitochondria showed severe reduction of cristae and a generally irregular shape. This phenotype was rarely observed in uninjected samples or in tissue in which the Mo^ATG was injected together with a human HSD10 expression construct. This set of experiments demonstrated by LOF analysis that HSD10 is required for functional and structural integrity of mitochondria.

Next we asked whether Mo-mediated HSD10 knock-down would interfere with the establishment of the early body plan in Xenopus embryos. HSD10 Mo^ATG was injected into dorsal animal (d/a) blastomeres at the four-cell stage and the embryos were grown for 2 days until they had reached tailbud stages. The progeny of the d/a blastomeres contributes mainly to neural tissue and indeed we found reduced brain tissue and eyes in Mo^ATG-injected embryos (Fig 4A). This phenotype was specific because a human HSD10 expression construct which did not harbour the Mo binding site completely rescued the eye and brain defects. Despite the adverse effects of xHSD10 knock-down on neural tissue, the anterior/posterior (a/p) patterning of the brain was unaffected in these embryos. The spatial expression patterns of the forebrain marker BF1 and the fore-/midbrain marker otx2 were unchanged, but expression levels were reduced in Mo^ATG-injected embryos (Fig 4B, C). Transferase dUTP nick end labelling (TUNEL) staining of embryos which received the xHSD10 Mo^ATG only into the right side of the body showed enhanced apoptosis on the injected side (Fig 4D, Supporting Information Fig 10). Selective targeting of the Mo is possible in Xenopus embryos because the dorsal/ventral and the left/right axes can be seen already at the four-cell stage due to pigmentation differences of the animal blastomeres. Induction of apoptosis after siRNA-mediated knock-down of HSD10 was confirmed in human SHSY5Y cells (Supporting Information Fig 11).

HSD10 Mo^ATG-induced apoptosis could be rescued by human HSD10 WT which also rescued the defects in neural tissue (Fig 5). The mutated HSD10 proteins found in human patients, however, showed remarkable differences in their ability to rescue the apoptotic phenotype. The Q165H mutation, which has less than 3% residual enzymatic activity but was identified in three neurologically normal boys, partially rescued HSD10 Mo^ATG-induced apoptosis. In contrast, injection of the mutations R130C and in particular D86G failed to rescue, and instead further enhanced apoptosis. D86G has considerable residual enzyme activity despite being associated with a very severe clinical phenotype. The inability of the R130C and D86G mutants to rescue apoptosis was not due to protein mislocalization. In human HeLa and Xenopus A6 cells WT and all mutated proteins colocalized with mitochondria (data not shown and Supporting Information Fig 12). Next we performed rescue experiments using dendritic cells derived from Tie2-Cre HSD10 mice. Loss of HSD10 induced apoptosis was detected and quantified by TUNEL assay. In agreement with our Xenopus data, WT HSD10 and the Q165H mutant, but not the R130C and the D86G mutants, partially rescued apoptosis (Fig 5, Supporting Information Fig 4).

These experiments demonstrate that loss of HSD10 results in induction of apoptosis independent of the cell type and organism. The rescue experiments furthermore show that apoptosis after HSD10 loss of function is not dependent on the enzymatic activity of HSD10, again arguing for a non-enzymatic function of this protein which is required in mitochondria.

Mitochondrial dysfunction is a commonly observed feature of primary disorders of mitochondrial intermediary metabolism such as the classical organic acidurias. Neurological abnormalities in these conditions are usually attributed to accumulation of toxic substrates proximal to a specific enzymatic block, causing secondary interference with energy metabolism. Primary apoptotic nerve cell death is not a feature of organic acidurias. Here we report that neurodegeneration in MHBDD, a recently identified organic aciduria, is not due to loss of enzyme activity and, presumably, the accumulation of toxic metabolites, but rather involves a fundamentally different pathomechanism. Although the respective enzyme, HSD10 has an important role in mitochondrial metabolism, we show that the adverse effects of its deficiency are primarily caused by a non-enzymatic effect triggering mitochondrial disintegration, apoptosis and cell death.

Several lines of evidence presented here support this view. Measurements of enzymatic activity in patient fibroblasts revealed that the R130C and D86G mutations associated with classical and neonatal neurodegenerative forms of MHBDD, had up to 30% residual activity compared to the WT protein. In contrast, the Q165H mutation that was identified in three children with normal neurological development was associated with less than 3% residual enzymatic activity. Thus the severity of the clinical phenotype is not at all correlated with the enzymatic activity of the mutated HSD10 proteins. We subsequently set out to corroborate this observation with experimental evidence. Mitochondrial integrity and cell viability could be restored after HSD10 knock-down in Xenopus and in conditional knock-out mice by Q165H mutant HSD10 protein, but not by R130C or D86G. Complete loss of the HSD10 protein in mouse embryos causes lethality at very early stages of embryogenesis (day 5 p.f.). It is unlikely that the embryonic lethality is due to the lack of enzyme activity, because it was shown for defects in comparable metabolic enzymes that toxic metabolites are efficiently eliminated by the placenta.

Additional clinical observations support our conclusions. Attempts to improve the condition of MHBDD patients by a restrictive diet that avoids accumulation of potentially toxic precursor metabolites failed (unpublished data). Slow progressive neurodegeneration without metabolic crises is atypical for organic acidurias caused by defects in mitochondrial intermediary metabolism, and deficiency of the enzyme 2-methylacetoacetyl-CoA thiolase, positioned immediately after MHBD in the isoleucine pathway, causes accumulation of the same metabolites in sometimes even higher concentrations but is associated with a completely different disease picture (episodic ketoacidosis, no neurodegeneration). In contrast to most other metabolic disorders, no null mutations that completely eliminate the protein have been identified in MHBDD. Indeed, one particular mutation, R130C, is found in more than half of cases and has usually occurred de novo in the respective patients. There is no evidence that the corresponding nucleotide change c.388C>T represents a hypermutable position, raising the possibility that most other mutations in the HSD17B10 gene are not observed because they are incompatible with life. This again is highly unusual for organic acidurias that usually manifest after birth only, since prenatally toxic metabolites are removed via the placenta. Finally, a splicing variant of the HSD17B10 gene that causes moderately reduced production of normal HSD10 without significant loss of enzyme activity and without production of an abnormal protein was found to cause an unusual non-progressive syndrome of mild mental retardation, choreoathetosis and abnormal behaviour (Lenski et al, 2007).

Reduction or loss of HSD10 in Xenopus, mouse and humans has adverse effects on mitochondrial integrity and leads to enhanced apoptotic cell death and reduced viability. Thus, HSD10 may be a component serving to protect mitochondria at times of increased metabolic stress. This interpretation is compatible with the observation that infections or other stress situations in patients with MHBDD may trigger clinical deterioration. In response to metabolic stress, HSD10 protein translocates from the mitochondrial matrix to the inner mitochondrial membrane and this might be important for maintaining mitochondrial integrity as this is an essential site for assembly of the mitochondrial membrane permeability transition pore (Tieu et al, 2004, and unpublished own data). A protective effect of overexpressed HSD10 under conditions of oxidative stress was found in the setting of acute brain damage (Yan et al, 2000) and in a mouse model of Parkinson disease (Tieu et al, 2004). It appears plausible that interaction of HSD10 with an integral protein of the mitochondrial membrane could mediate its protective effect and is disturbed by altered HSD10 protein structure due to specific mutations.

Cellular energy failure caused by mitochondrial dysfunction is thought to play an important role in the development of AD, and several studies indicated that this effect is caused by a direct interaction of Aβ with HSD10 (Lustbader et al, 2004; Yan et al, 1997). HSD10 protein or transcript was found to be increased in affected neurons in human AD as well as in AD mouse models and it has been suggested that overexpression of HSD10 amplifies Aβ-mediated mitochondrial toxicity (Chen & Yan, 2007). However, a recent study showed that a mitochondrial bioenergetic deficit precedes HSD10 overexpression and Alzheimer's pathology in an AD mouse model (Yao et al, 2009). The results of our present study, the first to examine the effect of HSD10 loss-of-function, indicate that adverse effects observed as a consequence of the binding of Aβ to HSD10 are more likely due to Aβ-mediated loss of a non-enzymatic HSD10 function, and that increased amounts of (non-functional) HSD10 protein may be produced as a secondary or compensatory mechanism. It thus remains possible that the mitochondrial interaction of Aβ with HSD10, triggering early dysfunction of energy homeostasis, is one of the first steps in the pathogenesis of AD.

Our conclusion that toxic metabolites are not responsible for the clinical symptoms observed in MHBDD patients implies that treatment targeting reduction of precursor metabolites through dietary measures is not indicated. This is in line with the clinical observation that dietary reduction of the isoleucine load has no beneficial effect on the course of this disease. In contrast, treatment of patients with MHBDD should aim at reducing mitochondrial stress and maintaining mitochondrial homeostasis through proactive management of infections and fever and possibly the administration of vitamins and cofactors. Our results may also contribute to a better understanding of neurodegeneration in the context of AD. Strategies to reduce Aβ toxicity through inhibition of its interaction with HSD10 could make HSD10 an interesting protein for Alzheimer's therapy.

Primary fixation was done with 2.5% glutaraldehyde in PBS. For easier handling tissue culture cells were scraped off and embedded in 2% agarose. All samples were postfixed with 1% osmium tetroxide (Serva Electrophoresis) in 100 mM phosphate buffer pH 7.2 for 1 h on ice, washed extensively with water, block-stained with 1% aqueous uranyl acetate (Serva Electrophoresis) for 1 h at 4 °C, dehydrated in a graded series of ethanol at ambient temperature, infiltrated with mixtures of ethanol/Epon and finally embedded in Epon. Ultrathin sections were stained with uranyl acetate and lead citrate (Serva Electrophoresis) (Venable & Coggeshall, 1965) and viewed in a Philips CM10 electron microscope at 60 kV using a 30 µm objective aperture. Mitochondria were quantitatively classified into groups according to their morphology by examiner who was blinded to the nature (WT/mutant, etc.) of the samples. Group 1 is distinguished by a regular arrangement and shape of the cristae stacks. Group 3 characteristics are a reduced cristae compartment resulting in less electron dense material in the mitochondria and an irregular arrangement and shape of the cristae stacks. Group 2 is represented by an intermediate morphological phenotype (see Figs 1K, 2B and 3C).

All samples were sectioned at random angles resulting in cross and longitudinal sections of mitochondria. For the morphological analyses the plain of sections was not relevant.

As mice deficient for HSD10 display embryonic lethality, two conditional knock-out mouse lines were generated. The construct used to generate floxed HSD10 alleles contained the coding sequence for HSD10 with exon 1 being flanked by loxP sites. This construct was introduced into embryonic stem cells (E14.1 of the 129P2/Ola Hsd stem) via electroporation. 20 µg DNA were mixed with 10⁷ cells in PBS and electroporated at 240 V and 960 µF. Stable clones were selected with G418 and cultured in DMEM containing LIF (myeloid leukemia inhibitory factor) on a layer of Feeder cells. Blastocysts were gained from pregnant C57Bl/6N mice at day 3.5 p.c. Stable transfected ES cells (20–25) were injected into the blastocoel and the blastocysts were transferred into pseudo-pregnant NMRI females at day 2.5 p.c. Resulting chimeric male animals were backcrossed with C57Bl/6 N females to homozygosity.

To generate an Hsd17b10 DBH-Cre conditional knock-out mouse, a transgenic mouse line carrying the Cre recombinase under the control of the DBH promoter was mated with a mouse line carrying a floxed allele of Hsd17b10. In the DBH-Cre line Hsd17b10 is eliminated specifically in noradrenergic neurons of the peripheral and CNS. The generation of DBH-Cre mice has been described previously (Stanke et al, 2006).

The second conditional knock-out (Hsd17b10 Tie2-Cre) in endothelial cells and hematopoietic stem cells was achieved using a transgenic mouse line carrying the Cre recombinase under the control of the Tie2 promoter. Generation of these mice has been described elsewhere (Constien et al, 2001).

Eggs were obtained by injecting human chorionic gonadotropin (Sigma) into female Xenopus laevis and fertilized artificially. Embryos were dejellied using 2% l-cysteine (pH 7.8–8.0, Biomol) and cultured in 0.1× MBSH. Embryo stages were determined according to Nieuwkoop and Faber (1967). Animal caps were dissected at stage 8 or 9 and cultured in 0.5× MBSH until stage NF 33/34.

Fibroblasts from MHBDD patients and control fibroblasts were cultivated in DMEM ReadyMix (Paa Laboratories) at 37°C and 5% CO₂. Mitochondrial staining in patient fibroblasts was achieved within 20 min with 300 nM Mitotracker Green FM (Invitrogen) on cells fixed with 3.7% formaldehyde on coverslips. Mitochondria were visualized after mounting in Mowiol (Merck) on a Perkin Elmer spinning disc confocal ERS-FRET on Nikon TE2000 inverted microscope.

The study was approved by the Ethical Committee of the Medical Faculty of Heidelberg University.