Although mitochondrial and serotonergic dysfunctions have been implicated in the etiology of bipolar disorder (BD), the relationship between these unrelated pathways has not been elucidated. A family of BD and chronic progressive external ophthalmoplegia (CPEO) caused by a mutation of the mitochondrial adenine nucleotide translocator 1 (ANT1, SLC25A4) implicated that ANT1 mutations confer a risk of BD. Here, we sequenced ANT1 in 324 probands of NIMH bipolar disorder pedigrees and identified two BD patients carrying heterozygous loss-of-function mutations. Behavioral analysis of brain specific Ant1 heterozygous conditional knockout (cKO) mice using lntelliCage showed a selective diminution in delay discounting. Delay discounting is the choice of smaller but immediate reward than larger but delayed reward and an index of impulsivity. Diminution of delay discounting suggests an increase in serotonergic activity. This finding was replicated by a 5-choice serial reaction time test. An anatomical screen showed accumulation of COX (cytochrome c oxidase) negative cells in dorsal raphe. Dorsal raphe neurons in the heterozygous cKO showed hyperexcitability, along with enhanced serotonin turnover in the nucleus accumbens and upregulation of Maob in dorsal raphe. These findings altogether suggest that mitochondrial dysfunction as the genetic risk of BD may cause vulnerability to BD by altering serotonergic neurotransmission.
Bipolar disorder is a major mental disorder characterized by mania and depression. Dysregulation in both monoaminergic systems  and mitochondrial calcium signaling  have been proposed in the etiology of bipolar disorder. However, the relationship between these apparently unrelated metabolic signaling systems has not been elucidated . Clinical studies showed that around 20% of patients with mitochondrial disease have comorbid bipolar disorder [4,5,6], whereas 0.38% of patients with bipolar disorder had mutations of POLG (polymerase γ) causative for mitochondrial disease . In a family of bipolar disorder and autosomal dominantly inherited chronic external ophthalmoplegia (CPEO), the L98P mutation of ANT1 (adenine nucleotide translocator 1, SLC25A4) was identified [8, 9]. These findings suggest that central nervous system involvement in mitochondrial disease caused by ANT1 mutations confers a risk of bipolar disorder. ANT1 is named for its function as a translocator of adenosine triphosphate and adenosine diphosphate across the mitochondrial inner membrane. However, its modulatory role in the mitochondrial permeability transition pore (mPTP) has also drawn attention [10,11,12]. The mPTP plays a role in regulated cell death, and transient opening of the mPTP also regulates mitochondrial calcium signaling , which is consistent with the well-known calcium dysregulation hypothesis of bipolar disorder .
In this study, we searched for ANT1 mutations in patients with bipolar disorder, and identified two independent loss of function (LOF) mutations of ANT1. We investigated the relationship between heterozygous loss of function of ANT1 and bipolar disorder by generating a brain specific Ant1 conditional knockout (cKO) mouse. By behavioral screening, we identified that the heterozygous mice showed diminished delay discounting, that is the choice of smaller but immediate reward than larger but delayed reward and an index of impulsivity . Consistent with this finding, the mice had enhanced serotoninergic activity. These findings together shed new light on the mechanism of how ANT1 mutations may confer a risk for BD.
Materials and methods
The study was approved by the Wako first ethics committee of RIKEN. All animal care and experimental procedure were in accordance with the guidelines for proper conduct of animal experiments published by Science Council of Japan and approved by RIKEN Wako Animal Experiment Committee. Methods and materials are described in detail in Supplemental Experimental Procedures.
All four exons of the SLC25A4 gene were sequenced by PCR-direct sequencing in 324 probands of NIMH Genetics Initiative bipolar disorder pedigrees. Their diagnosis was bipolar I (n = 304), bipolar II disorder (n = 17), or schizoaffective disorder, bipolar type (n = 3).
Floxed exon 2–3 of Slc25a4 mouse line (Slc25a4tm1a(EUCOMM)Wtsi) was obtained from International Knockout Mice Consortium (IKMC). Flp-transgenic mouse line (B6 Tg(cat-Flpe)36) was previously generated . Nestin-Cre transgenic mouse line (B6.CgTg(Nes-cre)1Kln/J) was obtained from Jackson laboratory. Using these mice lines, heterozygous and homozygous cKO mice of Slc25a4 (Slc25a4fl/fl or Slc25a4fl/+) (Ant1 cKO mice) were generated.
Staining methods and antibodies are described in detail in Supplementary Methods.
Fluorescent in situ hybridization was conducted as previously described . Images were captured by a confocal microscope IX81 with FV1000 (Olympus Corporation, Tokyo, Japan), Observer Z1 with AxioVision 4.6 (Zeiss, Oberkochen, Germaney) or Nano Zoomer Digital Pathology system (Nanozoomer 2.0RS, Hamamatsu Photonics, Hamamatsu, Japan).
Mouse behavioral screening with IntelliCage
The IntelliCage apparatuses (NewBehavior AG, Zurich, Switzerland) were used for behavioral screening as described previously [18, 19]. Male mice including 8 heterozygous cKO mice (Slc25a4fl/+; Nes-Cre+), 10 homozygous cKO mice (Slc25a4fl/fl; Nes-Cre+) and 6 controls (Slc25a4fl/+ or Slc25a4fl/fl; Nes-Cre−), which were 20–27 week old, were used.
5-choice serial reaction time task (5-CSRTT)
The 5-CSRTT operant chamber (O’HARA & Co., Tokyo, Japan) was used as previously described with minor modification . Male mice including 8 controls (Slc25a4fl/+ or Slc25a4fl/fl; Nes-Cre−), 8 heterozygous cKO mice (Slc25a4fl/+; Nes-Cre+) and 8 homozygous cKO mice (Slc25a4fl/fl; Nes-Cre+), which were 8–13 week old at the beginning of training, were used for the analysis.
Quantification of mtDNA deletion and mtDNA copy number
Partially deleted mitochondrial DNA (ΔmtDNA) and copy number of mtDNA was measured by quantitative PCR methods using SYBR Premix Ex Taq Kit (Takara Bio, Otsu, Japan) and QuantStudio 12 K Flex (Thermo Fisher Scientific, Waltham, MA) as described . For quantification of mtDNA deletion 30–39 week old male mice were used. For mtDNA copy number analysis, 54–56 week old male mice were used. Control mice were Slc25a4fl/+ or Slc25a4fl/fl without Nestin-Cre.
Quantification of Ant1/Ant2 mRNAs
mRNA expression of Ant1 (Slc25a4) and Ant2 (Slc25a5) were measured by quantitative PCR methods using SYBR Premix Ex Taq Kit (Takara Bio, Kusatsu, Japan) and QuantStudio 12 K Flex (Thermo Fisher Scientific, Waltham, MA). For these analysis, female 78–114 week old mice were used (n = 3 for each group).
Measurement of calcium retention capacity
Brain mitochondria were isolated using a discontinuous Percoll gradient developed by Sims  with minor modifications . Extra-mitochondrial free Ca2+ concentration ([Ca2+]exm) was monitored with 200 nM Calcium Green-5N (Thermo Fisher Scientific, Waltham, MA) (Ex 480 nm, Em 540 nm) in a 96 well plate at 30 °C in a Drug Screening System (FDSS 3000, Hamamatsu Photonics, Hamamatsu, Japan). To evaluate the CRC, 10 µl of Ca2+ solution was repeatedly added at 1-min intervals. Heterozygous cKO mice (Slc25a4fl/+; Nes-Cre+) (n = 4), homozygous cKO mice (Slc25a4fl/fl; Nes-Cre+) (n = 4), and control mice (Slc25a4fl/fl; Nes-Cre−) (n = 3) aged 8–27 weeks were used for this analysis.
Quantification of monoamine in tissue by HPLC
Dopamine, noradrenaline and serotonin and their metabolites were measured by HPLC with an EICOMPAK SC-5ODS with electrochemical detector ECD-300 (Eicom Corporation, Kyoto, Japan). In this experiment, control mice included Slc25a4+/+; Nestin-Cre+ and Slc25a4+/+, Slc25a4fl/+ or Slc25a4fl/fl without Nestin-Cre. For this experiment, 88–103 week old male mice were used.
Brain slices for experiments were prepared from 10–12-week-old, male mice as described previously . Whole cell patch-clamp recordings were acquired and controlled using the Axon 700B Multiclamp amplifer (Molecular Devices, CA, US) and pClamp11 acquisition software (Molecular Devices, CA, US). For this analysis, heterozygous Ant1 cKO mice (Slc25a4fl/+; Nes-Cre+) (n = 4) and control mice (Slc25a4fl/+ without Nes-Cre) (n = 3) aged 8–12 week old were used.
Data were analyzed by Prism 4 (Graphpad softoware Inc., San Diego, CA), IBM SPSS Statistics 20 (IBM Japan, Tokyo, Japan), “R” (https://www.r-project.org/) or Kyplot (Kyence, Tokyo, Japan). For genetic association analysis, Fisher’s exact probability test was used. In the comparison between the heterozygous or homozygous cKO mice and control mice, Student t-test was used. For place learning task and delay discounting task in IntelliCage, repeated measures ANOVA with Bonferroni’s post hoc test was used with main effects of genotype and delay or day.
Identification of loss of function mutations in bipolar patients
We sequenced the ANT1 gene in 324 NIMH probands with bipolar disorder and identified two patients carrying LOF mutations. One patient had a stop codon mutation p.Q85X, a substitution of C to T at Chr 4: 185,144,905 [hg38] while the other had p.Q175RfsX38, a single nucleotide deletion at Chr 4: 185,145,174, causing a frameshift and premature stop codon (Supplementary Figures 1A-B), which was reported in a pedigree of recessive cardiomyopathy with comorbid depression and anxiety . The frequency of LOF mutations (gain of stop codon or frame shift) for ANT1 in BD (2/324, 0.61%) was significantly higher than that in the exome and genome data in the gnomAD database (10 of 128,632 [2 stop codons and 8 frameshift mutations], 0.000069%, Fisher’s exact probability test, P = 0.00040, odds ratio = 79.7 [95%CI: 8.4–374.4]). Both mutations were on exon 3 and were thus predicted to undergo nonsense-mediated mRNA decay. Even when cDNAs of the predicted mutant mRNA encoding truncated proteins were constructed, protein expression of these mutants was markedly reduced in Neuro 2A cells (Supplementary Figure 1C). Although the two mutations did not show complete cosegregation with BD in the two pedigrees partly because of bilinear transmission (Supplementary Figure 1D), the significant association with high odds ratio suggested the role of these LOF mutations as a genetic risk factor for BD.
Generation of brain-specific Ant1 cKO mice
ANT1 mutations are known to cause a neuromuscular disorder, CPEO, and thus behavioral analyses must be performed using brain-specific mutant mice. We crossed floxed Ant1 mice (Slc25a4fl/fl or Slc25a4fl/+) with Nestin-Cre transgenic mice to generate a brain-specific cKO Ant1 mouse (Fig. 1a). We verified that the homozygous Ant1 cKO mice (Slc25a4fl/fl; Nestin-Cre+) have no Ant1 mRNA (Fig. 1b) and protein (Fig. 1c) expression in the brain, and heterozygous Ant1 cKO mice (Slc25a4fl/+; Nestin-Cre+) have reduced Ant1 mRNA (Fig. 1d) compared to control mice (Slc25a4+/+; Nestin-Cre+, Slc25a4fl/+ or Slc25a4fl/fl; without Nestin-Cre). No compensatory upregulation of Ant2 mRNA was observed (Fig. 1e).
To examine the functional consequences of the loss of ANT1, we isolated mitochondria from Ant1 cKO mice. The mitochondrial calcium retention capacity was significantly lower in homozygous Ant1 cKO mice (P = 0.02) (Figs. 1f, g) suggesting that the mitochondria of homozygous Ant1 cKO mice are vulnerable to the mPTP opening. The calcium retention capacity of heterozygous Ant1 cKO mice did not significantly differ from controls.
Because the identified patients with bipolar disorder were heterozygous for LOF mutations in ANT1, we performed the murine behavioral characterization in heterozygous Ant1 cKO mice to model the human disorder. Homozygous cKO mice were examined as a reference control, and Slc25a4fl/+ or Slc25a4fl/fl without Nestin-Cre, or Slc25a4+/+; Nestin-Cre+ were also examined as controls.
Behavioral phenotypes of brain-specific Ant1 cKO mice
We performed behavioral screening using the IntelliCage (Fig. 2a). Indices for spatial learning, reverse learning, and attention did not differ between genotypes (Supplementary Figures 2A-E), but the heterozygous cKO mice showed a significantly decreased delay discounting (Fig. 2b). The preference for saccharin did not differ between genotypes (F = 1.0, P = 0.36 by one way ANOVA) (Supplementary Figure 2F), suggesting that this finding is not due to altered reward value. Homozygous Ant1 cKO mice also showed a similar behavioral phenotype (Fig. 2b), although more similar to controls than heterozygous cKO mice.
To further characterize the behavioral phenotypes of heterozygous Ant1 cKO mice, we performed a 5-choice serial reaction time test (5-CSRTT), an established test to measure impulsivity  as a reflection of enhanced delay discounting (Figs. 2c–f, Supplementary Figures 2G-I). In this test, the number of premature nose pokes during long stimulus duration trials, which is an index of impulsivity, was significantly lower in heterozygous cKO mice than control mice (Fig. 2d). A decrease of impulsivity is equivalent to a decrease in delay discounting shown by the IntelliCage (Fig. 2b). The heterozygous cKO mice also showed better accuracy (Fig. 2e) and a lower number of perseverative responses than control mice in the long inter-trial interval trials (Fig. 2f). Homozygous cKO mice did not show similar phenotypes in the 5-CSRTT for unknown reasons.
We searched for brain regions with mitochondrial dysfunction due to the heterozygous knockout of Ant1 by performing COX (cytochrome c oxidase)/SDH (succinate dehydrogenase) co-staining. COX is a mtDNA-encoded protein whereas SDH is a nuclear-encoded mitochondrial protein, and COX-negative cells were detected in a model mouse of mitochondrial disease . In a sagittal section of aged homozygous Ant1 cKO mice, COX negative cells were detected preferentially in the dorsal raphe (DR) (Fig. 3a) and heterozygous Ant1 cKO mice also showed COX negative cells in the DR (Fig. 3b). Unexpectedly, however, a similar accumulation of COX negative cells was also detected in the DR of wild type mice (Fig. 3b). In this region, mtDNA deletions (ΔmtDNA) were not detectable both in cKO and control mice (Fig. 3c). Thus, the DR may have a selective vulnerability to mitochondrial dysfunction unrelated to the accumulation of ΔmtDNA, and the phenotype of the mutant mice might be caused by an interaction of the genotype and a general vulnerability of DR neurons to mitochondrial dysfunction. The copy number of mtDNA was significantly increased in DR of heterozygous cKO mice compared with control mice (Fig. 3d).
Serotonergic dysfunction in Ant1 KO mice
The activation of serotonergic neurons reportedly attenuates delay discounting , and therefore we examined serotonin turnover in the nucleus accumbens, which is innervated by DR serotonergic neurons and regulates impulsivity . We found that serotonin turnover was enhanced in the nucleus accumbens of heterozygous KO mice (Fig. 3e). There was no significant difference of the serotonin turnover between Nestin-Cre and wild type mice (0.55 ± 0.07 vs 0.75 ± 0.17, respectively, P = 0.15, n = 3 for both groups). We also analyzed the gene expression level of Maob (monoamine oxidase B) that encodes a monoamine metabolizing enzyme on the mitochondrial outer membrane enriched in the DR . To normalize expression levels of Maob by the number of serotonergic neurons within the sample, a serotonin neuron specific gene, Tph2 or Slc6a4, was used for a reference. We found that Maob mRNA was significantly increased in the DR of heterozygous Ant1 cKO mice (P < 0.05) (Figs. 3f, g) consistent with the elevated turnover of serotonin in the nucleus accumbens.
To test whether serotonergic neurons were activated in the Ant1 cKO mice, we performed electrophysiological recordings from midbrain slices containing the DR. As shown in Supplementary Table 1, both basic membrane properties and action potential properties of serotonergic neurons at the midline of the DR did not show significant differences between genotypes. However, the input-output relationship curves, which show the generation of action potentials by current injection was steeper in heterozygous Ant1 cKO mice than that of controls (Fig. 4) indicating that serotonergic neurons of heterozygous Ant1 cKO mice are more excitable.
In this study, we identified two patients with bipolar disorder carrying loss of function mutations in ANT1 (Supplementary Figure 1). Brain-specific heterozygous cKO mice of Ant1 showed diminished delay discounting or reduced impulsivity by two behavioral tests (Fig. 2). Enhanced serotonin turnover (Fig. 3e) and hyperactivity of serotonergic neurons (Fig. 4) were consistent with the behavioral phenotype . Increased mtDNA copy number in DR (Fig. 3d) might reflect an increased energy demand in the DR, and upregulation of Maob mRNA is considered a compensatory upregulation associated with enhanced serotonergic activity. These findings, together, suggest that the heterozygous loss of Ant1 in the brain causes a hyperserotonergic state and associated behavioral phenotypes.
The mechanism of how the heterozygous loss of function of Ant1 causes serotonergic hyperactivity is not known. Elevated intracellular calcium associated with depolarization is sequestered by mitochondria. The accumulation of intramitochondrial calcium results in a transient opening of the mPTP . Ant1 reportedly has a modulatory effect on the mitochondrial permeability transition pore (mPTP) . An altered mPTP function associated with heterozygous loss of Ant1 might affect the excitability of DR neurons.
A previous study reported exaggerated corticosterone responses to stress in homozygous conventional Ant1 KO mice . Enhanced serotonergic function might also underlie this finding, because serotoninergic stimulation is known to activate the hypothalamic-pituitary-adrenal axis .
Why are serotonergic neurons preferentially affected by the heterozygous loss of Ant1? Monoamines are metabolized by monoamine oxidases (MAOs) on the mitochondrial outer membrane. Metabolism of monoamines by MAOs is accompanied by the generation of hydrogen peroxide, a reactive oxygen species . Enhanced serotonin release by 3,4-methylenedioxymethamphetamine (MDMA) can decrease COX protein levels, which can be rescued by an MAO-B inhibitor . We speculate that serotonergic neurons may be intrinsically vulnerable to mitochondrial dysfunction, and the hyperserotonergic state may further impair mitochondrial function in a vicious cycle. The results of the COX/SDH staining in this study support this hypothesis. The degeneration of DR serotonergic neurons is frequently seen in Parkinson’s disease, in which mitochondrial dysfunction is implicated, and this might at least partly explain the non-motor symptoms of this disease . In Parkinson’s disease, 8-hydroxyguanosine is accumulated in substantia nigra dopaminergic neurons, and this is also true for the DR . Thus, the mitochondrial dysfunction phenotype due to the heterozygous cKO of Ant1 is predominantly seen in serotonergic neurons, which may have an intrinsic vulnerability to mitochondrial dysfunction. The mechanism of how COX immunoreactivity is reduced in the DR is not known because the present study did not show an increase of ΔmtDNA levels or a decrease of mtDNA copy number. Reduction of COX immunoreactivity in DR neurons might be regulated by other types of mtDNA abnormalities and/or at the protein level .
Mitochondrial dysfunction has been implicated in bipolar disorder based on several lines of evidence  including altered energy metabolism detected by magnetic resonance spectroscopy , comorbidity with mitochondrial diseases [4, 7], and findings in postmortem brains including an accumulation of ΔmtDNAs [39, 40], altered gene expression of mitochondria-related genes , altered morphology of mitochondria , and decreased activity of mitochondrial complex I . On the other hand, serotonergic dysfunction in bipolar disorder has been implicated by evidence including the mania-inducing effect of antidepressants that inhibits the serotonin transporter and thereby activates serotonin , the efficacy of atypical antipsychotics that inhibits serotonergic neurotransmission , altered mRNA expression levels of serotonergic receptors in postmortem brain , altered DNA methylation of serotonin transporter gene , altered serotonin transporter binding in the brain by positron emission tomography , and levels of cerebrospinal fluid metabolites, among others . The present findings provide a potential missing link between these two lines of evidence. Because the two LOF mutations identified in this study were not cosegregated with bipolar disorder in the two pedigrees, they are not “causative” mutations. However, their significant association suggests that the heterozygous LOF mutations of ANT1 confer the risk of bipolar disorder.
There are several limitations in this study. Notably, it is unknown why homozygous Ant1 cKO mice do not show behavioral alterations in some experiments. However, such non-linear dynamics are inherent to complex biological systems such as the brain. Secondly, Nestin-Cre mice reportedly have some behavioral alterations , which could in principle confound the results. However, we verified that this transgene did not affect serotonin turnover and excitability of DR serotonergic neurons (Supplementary Figure 3). Thirdly, behavioral analysis is confounded by genetic background . Although all the mice used were generated and/or kept under the background of C57BL/6, there are subtle behavioral differences even among C57BL/6 substrains . We therefore used the mice after backcrossing with C57BL/6J, though the number of generations might not be enough to rule out a possible effect of substrains. Finally, the results in calcium retention capacity are not consistent with previous studies that showed a loss of Ant1 causing an increase in calcium retention capacity . This discrepancy might be due to differences in experimental conditions or mouse strains.
In summary, our current findings suggest that mitochondrial dysfunction caused by heterozygous loss of Ant1 can cause alterations of serotonergic activity, a first step toward understanding the complex neurobiological processes underlying bipolar disorder subtypes.
We are grateful to Ms. Mizue Kametani and Dr. Hirochika Kawakami for technical assistance. We thank Dr. Katsuhide Igarashi and Ms. Maky Otsuka (Hoshi University School of Pharmacy and Pharmaceutical Sciences) for advising us the method for the dissection of dorsal raphe. We also thank Research Resource Center and BSI-Olympus Collaboration Center, RIKEN Brain Science Institute, for technical assistance. The work was supported by JSPS KAKENHI Grant Number JP27860956 to T.M.K. and JP17H01573 to T.K. and the Advanced Genome Research and Bioinformatics Study to Facilitate Medical Innovation (GRIFIN) from the Japan Agency for Medical Research and Development (AMED) (16815678) to TK.
TMK and TK conceived the study design. TMK and NFT performed the molecular, anatomical, and behavioral experiments. MKS performed the experiments on isolated mitochondria and a part of the immunohistochemical analyses. FS and HS performed the electrophysiological experiments. SF sequenced the Ant1 gene in human samples. AM and SI conceived the method for the IntelliCage analysis. TMK and TK wrote the paper.
Electronic supplementary material
About this article
These authors contributed equally: Tomoaki M. Kato, Mie Kubota-Sakashita, and Noriko Fujimori-Tonou