Disorders involving neurometabolism can lead to both structural and functional disturbances of the nervous system through multiple mechanisms that include abnormal accumulation of toxic substrates, depletion of key energy or metabolic intermediates, or cell death.1 Although pediatric-onset brain diseases are often associated with genetic abnormalities, the link between metabolic impairments and brain disease has been associated with relatively common disorders of the nervous system.2 Especially in highly consanguineous populations, congenital malformations of the nervous system and neurometabolic disorders are often linked.3 Genes involved in inborn errors of metabolism often encode enzymes that catalyze specific biochemical reactions or required cofactors.4 Absent or abnormal functioning of such proteins leads to an accumulation of upstream substrates or deficiency of downstream products, with detrimental disease-causing effects.

Dioxygenases are oxidoreductase enzymes that utilize dioxygen (i.e., O2), most often complexed with iron, in oxidation reactions involved in metabolic pathways or as oxygen sensors.5 To date, 15 dioxygenases have been linked to diseases that are mostly autosomal recessive including encephalopathy, alkaptonuria (MIM 203500), and intellectual disability.6,7,8 Dioxygenases can oxidize either small molecules, as in tryptophan dioxygenase,9 or residues on macromolecules, as in EGLN1, a HIF proline dioxygenase acting as an oxygen sensor.10

4-hydroxyphenylpyruvate dioxygenase (HPD, MIM 609695) is a nonheme Fe(II)-dependent dioxygenase that catalyzes the second of four steps in L-tyrosine catabolism in nearly all aerobic cells: the conversion of 4-hydroxyphenylpyruvate into homogentisate.11 Biallelic loss associates with recessive tyrosinemia type III (MIM 276710), the least frequent form of three genetic tyrosinemias.12 Clinical features are milder than those of other genetic forms, and can include seizures, intellectual disability, and intermittent ataxia.13,14 Serum tyrosine level is variably elevated by 5- to 20-fold (148–1769 µmol/L, reference 29–86 μmol/L).14,15,16,17

In addition to biallelic variants, a particular gain-of-function HPD variant (p.Ala33Thr) associates with the dominant disease hawkinsinuria (MIM 140350),15 as mutant HPD can convert the normal substrate into a reactive epoxide intermediate but cannot rearrange the intermediate to homogentisate. As a result, the reactive epoxide is detoxified to either hawkinsin or hydroxycyclohexyl acetate by glutathione transferase and epoxide hydrolase, respectively. Hawkinsinuria manifests clinically with metabolic acidosis in early childhood, but most patients have normal long-term outcome.

HPDL encodes 4-hydroxyphenylpyruvate dioxygenase-like protein (HPDL), the only mammalian paralogue of HPD, and shares 44% sequence similarity. The origin and function of HPDL is not known, and the gene has not been linked to human disease previously. Here we describe that HPDL biallelic loss leads to a unique neurometabolic, infantile neurodegenerative condition.


Patient recruitment

The institutional review board (IRB) at the University of California–San Diego (UCSD) approved this study. All study participants signed informed consent documents, and the study was performed in accordance with the US Health Insurance Portability and Accountability Act of 1996 (HIPAA) Privacy Rules. The procedures followed for recruitment and data collection were in accordance with the ethical standards of the responsible committee on human experimentation at the respective participating institute, and proper informed consent was obtained.

DNA sequencing

Samples of the probands were subjected to exome sequencing. Sanger sequencing of polymerase chain reaction (PCR) products was performed for validation and segregation. Further details are in the Supplementary data.

Cell culture

HEK293T cells (ATCC) were transfected with expression vectors using Lipofectamine 2000 according to manufacturer’s protocol (100 ng of DNA and 0.5 µl of Lipofectamine 2000 per well) and immunostained 48 hours after transfection. Methods used for cloning mammalian expression vectors, transfection, and immunostaining are described in Supplementary Methods.

Methods used to generate HPDL knockout (KO) CACO-2 cell line are described in Supplementary Methods. Immunostaining and imaging of wild-type and KO CACO-2 cells details are described in Supplementary Methods.

Animal experiments

Animal use followed National Institutes of Health (NIH) guidelines and was approved by the Institutional Animal Care and Use Committee (IACUC) at UCSD. The frameshift variant in Hpdl (NM_146256.3:c.59_65del) was introduced to C57BL/6 mouse using CRISPR/Cas9 (single-guide RNA [sgRNA] sequence: CUU CCA GCC CCU GGC GGU GA) according to standard protocols at the UCSD Transgenic Mouse Core. Mice were genotyped by PCR followed by Sanger sequencing (Hpdl_KO_F/R primers; Table S1), and those with correct genotypes were backcrossed for at least five generations prior to experimental crosses of heterozygous animals. Timed pregnant animals were obtained by plug checks, where the day of the observed vaginal plug was determined to be E0.5. All mice were group-housed under a 12-hour light/dark cycle with access to chow and water. Breeding was performed until 30 homozygous pups were born to analyze survival, growth, histology, and immunofluorescence staining. The investigators were not blind to the analysis, since phenotypes of KO animals were very apparent. Methods used for histological analysis, immunofluorescence staining, and imaging are described in Supplementary Methods.

Direct infusion–based metabolomics

Collection of cell lysates and mouse brain for metabolomics experiments are described in Supplementary Methods.

A nonquantitative direct infusion high-resolution mass spectrometry metabolomics method was used as previously described with minor adaptations for cell and tissue extracts.18 In brief, 70 µl of cell lysate in methanol was diluted with 60 µl 0.3% formic acid (Emsure, Darmstadt, Germany) and filtered using a methanol preconditioned 96-well filter plate (Acro prep, 0.2 um GHP, NTRL, 1-ml well; Pall Corporation, Ann Arbor, MI, USA) and a vacuum manifold. The sample filtrate was collected in a 96-well plate (Advion, Ithaca, NY, USA).

Samples were analyzed using a TriVersa NanoMate system (Advion, Ithaca, NY, USA) controlled by Chipsoft software (version 8.3.3, Advion). Data acquisition was performed using Xcalibur software (version 3.0, Thermo Scientific, Waltham, MA, USA). A peak calling pipeline, developed in R programming language, annotated the raw mass spectrometry data according to the Human Metabolome Database (HMDB).

Phylogenetic tree reconstruction

Amino acid sequences were aligned using PROMALs3d.19 Three alignments were computed: (1) all sequences, (2) HPD sequences, and (3) HPDL sequences. We used a maximum likelihood method using RAxML20 on XSEDE via CIPRES to obtain the gene tree topology using the JTT model and 1000 bootstrap replicates. The gene tree was visualized using iTol,21 and the final rooted topology was obtained by rerooting the tree with a midpoint root.

Estimates of evolutionary rates

Branch lengths were obtained directly from the phylogenetic trees. To calculate dN/dS ratios over the tree and subtrees (HPD and HPDL), we used CODEML in the PAML package (version 4.8).22 To test multiple rates on the tree on CODEML, we used the model = 2 parameter to estimate an HPD, HPDL, HPD tetrapod, and an HPDL tetrapod rate. Pairwise dN estimates were used using the runmode = −2 parameter. To perform the CODEML analysis, the CDS alignment was performed using TranslatorX.23 The dist.alignment function was called in seqinr to compute the pairwise distance matrix using the amino acid identity and similarity.24


Identification of an infantile-onset neurodegenerative condition in eight families with biallelic variants in HPDL

We identified eight independent consanguineous families carrying HPDL biallelic variants leading to a range of neurological phenotypes, which included spastic tetraplegia, microcephaly, brain atrophy, epilepsy, and severe intellectual and motor disability (Fig. 1a and Table S2). After obtaining informed consent from all participating individuals in accordance with the ethical standards set by the UCSD IRB, we identified a total of eight distinct variants in HPDL in the eight families by exome sequencing. We recruited family 1 with three affected siblings showing limb spasticity and mild motor and intellectual disability. Although parental consanguinity suggested the presence of a homozygous variant as the cause, no homozygous variants in any gene passed filter criteria. Instead, we identified compound heterozygous p.Gly50Asp; p.Tyr118* variants in HPDL, which was the only candidate identified by exome sequencing and segregated in the family. Family 2 was recruited with documented parental consanguinity with one affected girl showing a nearly identical clinical pattern as family 1. We identified a homozygous p.Gly50Asp variant in the affected individual as the most likely cause of disease, one of the same variants identified in family 1 (Table S3).

Fig. 1: Variants in HPDL in eight independent consanguineous families lead to microcephaly and brain atrophy.
figure 1

(a) Pedigrees of families 1–8 showing consanguineous marriages (double line) with a total of 17 affected children. All unfilled members are without neurological disease, epilepsy, or neurodegeneration. Dashed line: deceased. (b) Panels show magnetic resonance imaging (MRI) scans for six affected individuals from five different families. Shown are sagittal and axial images demonstrating enlarged ventricles, thin corpus callosum, and severe cortical and cerebellar atrophy. (c) Schematic of the HPDL protein structure depicting a predicted mitochondrial targeting sequence (MTS), two vicinal oxygen chelate (VOC) domains from amino acids 7–135 and 160–328 (shown in dark gray) and three iron (Fe) binding sites (dashed lines). Solid black lines indicate the locations of the variants. (d) Amino acid alignments for all identified missense variants across different vertebrate species. Amino acids highlighted in gray indicate conserved residues. A.m. Alligator mississippiensis, B.t. Bos taurus, D.r. Danio rerio, G.g. Gallus gallus, H.s. Homo sapiens, M.m. Mus musculus, O.c. Oryctolagus cuniculus, X.l. Xenopus laevis.

Through collaboration with colleagues, we identified four additional families (3, 4, 5, and 8) with homozygous missense variants and two additional families (6 and 7) with homozygous truncating variants in HPDL independently identified as likely most pathogenic based upon inheritance pattern, computational prediction, population allele frequency, and segregation (Tables S1 and S2). These variants included homozygous p.Ala78Thr (family 3), p.Gly126Ser (family 4), p.Leu164Pro (family 5), p.Gly319Argfs*15 (family 6), p.Gln32* (family 7), and p.Gly301Val (family 8). Thus, we identified a total of eight families comprising 17 individuals with biallelic HPDL damaging variants with variable phenotypes. Because HPDL is encoded by a single exon, and thus not subject to alternative splicing, the presence of homozygous early truncating variants suggests HPDL loss of function as the mechanism of this infantile-onset neurodegenerative disease.

Affected children were born full-term without complications during pregnancy or delivery (Table S2). Where data were available, birth weight and head circumference were not remarkable. However, most affected individuals showed a trend toward smaller head circumference by 5 years of age; 7 of 14 affected individuals, where data were available, met criteria for microcephaly, defined as a head circumference <3 SD below mean (Table S2). Six of 14 affected individuals also displayed nonspecific facial dysmorphisms. Most individuals presented with minimal to absent psychomotor development, including gross motor skills, language, and social skills. Most displayed spontaneous epileptic seizures starting by 1 year. Seizures were typically myoclonic, focal, or tonic and occurred daily to weekly. Neurological findings included hypertonia, hyperreflexia, spastic gait, and positive Babinski sign—features that became more pronounced over time, suggesting a progressive course. Many individuals were unable to walk and displayed absent language skills. In six severely affected individuals, brain magnetic resonance images (MRIs) were available for review, demonstrating cortical atrophy, white matter hyperintensity, corpus callosum thinning, reduced size of the cerebellum, ventriculomegaly, and increased extra-axial fluid (Fig. 1b), suggesting loss of brain parenchyma. Brain atrophy and the progressive course were consistent with neurodegeneration.

HPDL encodes 371 amino acids containing a mitochondrial targeting sequence (MTS), two vicinal oxygen chelate (VOC) domains, and three iron-binding sites. The locations of all the variants occurred within the two VOC domains (Fig. 1c), thus predicted to impact enzymatic function. Of note, the phenotypes between homozygous truncating variants (p.Gln32* and p.Gly319Argfs*15), destroying at least one iron-binding site, and three of the missense variants (p.Ala78Thr, p.Gly126Ser, and p.Leu164Pro), were indistinguishable, suggesting that the missense variants cause loss of function of HPDL. The p.Gly50Asp variant was found only in families 1 and 2, and was associated with milder phenotypes compared with the more severely affected individuals in other families with brain atrophy and epilepsy, suggesting that this variant has a milder impact on HPDL function. The variants were unique in our data set of >5000 exomes from individuals with neurodevelopmental phenotypes, were predicted to be disease-causing, and were not represented (p.Ala78Thr, p.Leu164Pro, p.Gly301Val, and p.Gly319Argfs*) or very rare (p.Gln32* [allele frequency; AF = 8.46e-6], p.Gly50Asp [AF = 1.02e-5], p.Tyr118* [AF = 3.19e-5], and p.Gly126Ser [AF = 8.67e-6]), in the gnomAD database.25 No instances of biallelic damaging variants in HPDL among over 100,000 sequenced individuals occur in gnomAD (Fisher’s exact test P < 0.00001). Additionally, all missense variants are highly conserved among vertebrates (Fig. 1d). All variants were confirmed by Sanger sequencing and segregated according to a recessive mode of inheritance, with heterozygotes showing no apparent clinical features (Fig. S1).

HPD and HPDL have differing tissue-wide expression and subcellular localization

To functionally distinguish between HPDL and the paralogous HPD, we sought differences in tissue expression and subcellular localization patterns. The GTEx database indicates that HPD is almost exclusively expressed in the liver and kidney, whereas HPDL is widely expressed in most organs (Fig. S2).26 Importantly, the expression level of HPDL is much higher than HPD, specifically in the brain (Fig. S3).26 Publicly available human and mouse brain single-cell RNA-seq data suggests that HPDL transcripts are encountered ~5× more commonly than HPD in the brain, with brain astrocytes as the major source of HPDL transcripts.27,28,29

Protein databases suggest HPD localizes to the cytosol while HPDL localizes to mitochondria.30 We used Mitofates, an online MTS prediction tool, to assess for an N-terminal mitochondrial presequence or cleavable localization signal.31 While HPD had no predicted MTS, HPDL contained three different recognition motifs for TOMM20, a mitochondrial import receptor subunit responsible for the recognition and translocation of cytosolically synthesized mitochondrial preproteins. In addition, there was a region of maximum positively charged amphiphilicity, another essential feature for mitochondrial presequence function (Fig. 2a).32 To confirm this experimentally, we utilized CACO-2 cells, a human colon adenocarcinoma cell line expressing levels of HPDL higher than most other cell lines.33 To validate the specificity of the anti-HPDL antibody, we generated HPDL KO CACO-2 cells using CRISPR/Cas9, and confirmed correct biallelic targeting (Fig. S4A, B).

Fig. 2: HPD and HPDL display different subcellular localization patterns.
figure 2

(a) Schematic of HPD and HPDL N-terminal amino acid sequences highlighting the presence or absence of a predicted mitochondrial targeting sequence (MTS) as predicted by Mitofates. HPD has no predicted MTS, while HPDL has three different TOMM20 recognition motifs (yellow) and a region of max positively charged amphiphilicity (green). (b) Immunofluorescent staining of HPDL wild-type (WT) CACO-2 cells with DAPI (blue), HPDL (red), and TOMM20 (green). Scale bars: 10 μm (upper panels) and 2 µm (lower panels). (c) Immunofluorescent staining of HPDL knockout (KO) CACO-2 cells with DAPI (blue), HPDL (red), and TOMM20 (green). Scale bars: 10 μm (upper panels) and 2 µm (lower panels). (d) Immunofluorescent staining of HPDL WT CACO-2 cells with DAPI (blue), HPD (red), and TOMM20 (green). Scale bars: 10 μm (upper panels) and 2 µm (lower panels). (e) Immunostaining for TOMM20 (cyan) and DAPI (blue) of HEK293T cells transfected with both HPDL-Myc (red) and HPD-EGFP (green) or HPD-Myc (red) and HPDL-EGFP (green) to determine the subcellular localization of HPD and HPDL. White arrows: cells with mitochondrial localization of HPDL. Scale bar: 10 μm.

Immunofluorescence in WT cells confirmed the colocalization of TOMM20 and HPDL within mitochondria (Fig. 2b), while staining in the KO cells confirmed absent HPDL signal (Fig. 2c). HPD showed no colocalization with TOMM20, but instead was localized diffusely to the cytoplasm (Fig. 2d). To assess localization of tagged versions of the proteins, we cotransfected HEK293T cells with C-terminally Myc-tagged HPDL and C-terminally EGFP-tagged HPD and costained with anti-TOMM20. We used C-terminally tagged vectors to ensure that any N-terminal MTS would not be disturbed. Again, TOMM20 overlapped in localization with HPDL, but not HPD. We also flipped the vectors and repeated the experiment (C-terminally Myc-tagged HPD and C-terminally EGFP-tagged HPDL) to ensure that the tag did not affect localization, which provided further confirmation (Fig. 2e). We utilized a Seahorse assay to compare the oxygen consumption rate in these cells, which illustrated slightly lower oxygen consumption rate in the HPDL KO cells compared with WT (Fig. S5). Together, these results demonstrate that HPDL, but not HPD, localizes to mitochondria, suggesting that these enzymes act in different locations on either the same or on different substrates. The slightly reduced oxygen consumption rate in the KO cells further suggests that the absence of HPDL may have an effect on mitochondrial metabolic function.

Hpdl KO mice display epilepsy, early lethality, smaller brain sizes, and cellular apoptosis in the brain

To investigate the physiological role of HPDL, we created a KO mouse model with a homozygous frameshift deletion of Hpdl generated using CRISPR/Cas9 (Hpdl KO, Fig. S6). KOs were born healthy and at Mendelian ratios (WT:heterozygote:homozygote = 30:68:26, test for Hardy–Weinberg equilibrium P = 0.368). Although KOs were indistinguishable at birth, by postnatal day 5 (P5), KOs exhibit lethargy and spontaneous movements suggesting of epileptic seizures (Supplemental video S1). Almost all KOs died by P10, and none survived to P15 (Fig. 3a). Although body weights and brain size were similar at birth, by P9–10 KO mice weighed approximately half that of littermates (Fig. 3b, c). Immunostaining of P9 brain tissue sections for DAPI and Cleaved Caspase 3 (CC3) demonstrated significant cellular apoptosis (CC3+ cells) in the KO brains compared with WT pups (23 vs. <1 CC3+ cell per 500-µm column), which correlated with smaller brain size (Fig. 3c–e). To determine the cellular identity of CC3+ cortical cells, we performed coimmunostaining with CTIP2 (upper layer cortical neurons), OLIG2 (oligodendrocytes), and GFAP (astrocytes). CTIP2/CC3 staining revealed the most colocalization compared with OLIG2/CC3 and GFAP/CC3 (5.9% vs. 1.9% and 3.8%, respectively, Fig. 3f, g). Taken together, these data suggest cellular apoptosis in the brain occurs in the absence of HPDL function.

Fig. 3: Hpdl knockout (KO) mice display early lethality, smaller brain sizes, and cellular apoptosis.
figure 3

(a) Kaplan–Meier curve displaying the survival probability of Hpdl heterozygous mice (n = 10, blue) versus Hpdl KO mice (n = 6, red). (b) Graph depicting body weight measurements of Hpdl heterozygous mice versus Hpdl KO mice at postnatal days 0, 2, 5, and 9. P values were obtained by an independent t-test. (c) Whole-brain images of wild-type (WT) versus Hpdl KO mice at P0 and P10. Scale bar: 1 mm (d) Coronal sections of P9 WT and KO mouse brain were stained for DAPI (blue) and Cleaved Caspase 3 (CC3, green). The P9 KO shows strong CC3+ immunoreactivity. Note the absence of CC3 immunoreactivity in the P9 WT. Scale bar: 50 µm (e) Quantification of CC3+ cells in Hpdl WT vs KO mouse brain per 500-μm column. Statistical analysis (two-way analysis of variance [ANOVA]) was performed between the categories. N = 3 independent experiments with two sections per mouse quantified (****P < 0.0001). (f) Coronal sections of P9 KO mouse brain were stained for DAPI (blue) and CC3 (green) in combination with either CTIP2 (red), OLIG2 (red), or GFAP (red). White arrows depict colocalization of the cell identity markers with CC3. Scale bar: 50 µm. (g) Quantification of the number of CTIP2+, OLIG2+, or GFAP+ cells which were also CC3+. Analysis shows CTIP2+ cells to be most affected, followed by GFAP+ cells, followed by OLIG2+ cells. N = 2 independent experiments. Two sections per mouse were used for quantification.

Metabolomic analysis reveals disrupted metabolic signature but normal tyrosine catabolism

Deficiencies of specific enzymes of the tyrosine catabolism pathway are known to give rise to a number of severe metabolic disorders in humans. In the oxidative degradation of tyrosine by vertebrates, the first conversion step is to 4-hydroxyphenylpyruvate in a reaction catalyzed by tyrosine transaminase (Fig. S7A).34 HPD then catalyzes the oxidation of 4-hydroxyphenylpyruvate to homogentisate.12 Defects in these two steps in tyrosine catabolism lead to tyrosinemia type II and type III, respectively. Normally, homogentisate is further catabolized by the next enzyme in the pathway, homogentisate 1,2-dioxygenase.8 In alkaptonuria, however, the enzyme is deficient, leading to this rare hereditary metabolic disorder. In tyrosinemia type I, fumarylacetoacetate hydrolase is mutated, causing the subsequent buildup of byproducts.35 Because both HPD and HPDL have highly similar amino acid homology, we speculated that HPD and HPDL may have similar substrates. To test this, we performed mass spectrometry analysis on HPDL WT and KO CACO-2 cell extracts and conditioned cell culture media. The analysis revealed several significantly downregulated and upregulated metabolites in both the cell extracts and culture media of KO cells compared with WT; however, we did not detect an abnormal accumulation of any metabolites involved in tyrosine catabolism (Fig. S7B, C, Table S4). We then performed a similar analysis in both Hpdl WT and KO mouse brain extracts and serum at P5, which again showed no abnormal accumulation of tyrosine catabolites (Fig. S7D, E, Table S4). Taken together, these data suggest that HPDL likely acts on a different substrate than HPD, and perhaps on a different cellular pathway than tyrosine catabolism.

Independent HPD duplication events gave rise to HPDL in vertebrates

HPD, conserved in all eukaryotes, is encoded on human chromosome 12, with its major isoform consisting of 14 coding exons (Fig. 4a). One alternative splice isoform is described; however: the major isoform encodes a protein 393 amino acids long. HPDL, on the other hand, is a single-exon gene located on chromosome 1, with no alternative splice isoforms reported, encoding 371 amino acids, apparent in Xenopus tropicalis (frog) but not in lower species like Danio rerio (zebrafish) (Fig. 4b). There are no other HPD paralogues in mammals, suggesting that HPDL arose as a result of a retrotransposition event at the base of the tetrapod lineage (Fig. 4c). However, there is an HPDL paralogue in other lower deuterostomes, including Ciona intestinalis (sea vase) and zebrafish, but this HPDL contains introns. Thus, other models are possible by which HPDL arose from HPD during evolution.

Fig. 4: HPD and HPDL evolutionary analysis reveals two separate, yet significant, evolutionary events leading to the emergence of vertebrate HPDL with novel function.
figure 4

(a) Schematic of the major human HPD isoform depicting the coding sequence spanning 14 exons using the University of California–Santa Cruz (UCSC) Genome Browser as a reference. (b) Schematic of human HPDL depicting the coding sequence spanning a single exon and the 5’ and 3’ UTRs using the UCSC Genome Browser as a reference. Scale bars are shown. (c) Species tree of genomes used in this analysis. Arrows indicate HPD gene duplication and a possible retrotransposition event leading to the loss of exons in HPDL. (d) Gene tree for the HPD and HPDL genes constructed with RAxML with the JTT model. Branch lengths are on top of each branch, bootstrap support of 65 and above are shown below branches. Gray box indicates a loss of HPDL exons in tetrapods. dN/dS ratios are estimated using codeml and the model showing four different dN/dS ratios is shown (Fig. S8).

Several lines of evidence suggest that mammalian HPDL arose as a separate gene duplication event from the event that yielded HPDL in lower deuterostomes. First, only mammalian HPDL contains a predicted MTS. Second, mammalian HPDL shows evidence of stronger positive selective pressure than fish hpdl, as evidenced by faster accumulation of nonsynonymous to synonymous base changes (Figs. 4d and S8A). Third, mammalian HPDL is in a region of conserved synteny in frog but not in zebrafish (Fig. S8B). Finally, and most convincingly, it is more likely that a retrotransposition event yielded an intronless gene than removal of introns from an intron-containing gene. Human HPDL is more similar to zebrafish hpdl than to human HPD (40% vs. 25% amino acid sequence identity). The data suggest that lower deuterostome HPDL was lost contemporaneously with its retrotransposition to intronless HPDL, which subsequently acquired an MTS and amino acid changes that provided a new essential function.


Here, we implicate HPDL in a recessive infantile-onset neurodegenerative disease, characterized by intellectual and motor disability, epilepsy, spastic tetraplegia, and brain atrophy. Radiographic phenotypes include severe cortical and cerebellar atrophy, thinning of corpus callosum, and ventriculomegaly. There are likely a range of developmental brain phenotypes resulting from loss of HPDL, although our subjects are likely to be at the most severe end of the spectrum given the nature of the alleles. HPDL is not the first mitochondrial dioxygenase to be implicated in a neurometabolic disorder. ETHE1, another mitochondrial dioxygenase, is linked to ethylmalonic encephalopathy, an inherited disorder defined by elevated excretion of ethylmalonic acid, leading to developmental delay, seizures, and hypertonia.36 Future studies will be required to elucidate the full range of phenotypes, genotype–phenotype correlations, and mechanisms of pathogenicity.

We demonstrate that HPD and HPDL not only have different tissue-wide expression patterns, but also show different subcellular localizations. Localization of HPDL to mitochondria, together with the disrupted metabolomes in HPDL KO cells and mice, implicates this disease not only as a mitochondrial neurodegenerative condition, but also as a likely neurometabolic condition. In a direct comparison between Hpdl WT and KO mice, we highlight the pathophysiological importance of the HPDL enzyme for neurodevelopment and survival. Interestingly, the fact that the mice are born healthy and begin regressing soon after birth suggests that the mice lacking Hpdl may depend on an environmental or some other factor provided by their mother to stay well. Further, although astrocytes have been shown to express HPDL at the highest levels in the brain,27,28,29 in HPDL KO brain we found more neurons than glia with evidence of apoptosis, suggesting that neurons are ultimately more sensitive to HPDL loss.

There are several possible reasons why we were unable to identify a specific substrate for HPDL despite comprehensive metabolic profiling. First, the metabolite may be unidentifiable by the current approach, where we focused on small molecules that have been preannotated in public databases. Second, the amount of the potential substrate may have been too small to detect due to large sample volume. Therefore, future work should aim toward more targeted approaches (i.e., analyzing specific brain regions rather than the whole brain, or analyzing mitochondrial-specific extracts rather than whole-cell extracts). Finally, it is still possible that HPDL is not an enzyme, since we did not confirm its enzymatic activity. However, given its close sequence similarity to the enzyme HPD, presence of iron-binding motifs, disease caused by biallelic null variants, and metabolic dysregulation seen in the cell and mouse models, this hypothesis is less likely.

This finding was further supported by our evolutionary analysis, suggesting that HPDL and HPD are functionally different. We have shown that the HPDL found in humans likely emerged through a series of HPD gene duplication events, the last of which probably involved a retrotransposition event at the transition from deuterostomes to tetrapods. This suggests that at some point in time, HPDL existed as a multiexonic gene before losing its introns and gaining an MTS.

Taken together, we describe a novel, early-onset neurodegenerative disorder caused by variants in HPDL, and also illustrate the importance of this protein for neurodevelopment and survival. Through evolutionary modeling and analysis, we illustrate the route by which human HPDL emerged as a paralogue of HPD, with proposed novel biochemical functions. Although further work is needed to identify the specific function of mitochondrial HPDL, we believe our work is a significant cornerstone to elucidate the downstream mechanisms of HPDL, and to potentially lead to a treatment for this disease.