AIF3 splicing switch triggers neurodegeneration

Background Apoptosis-inducing factor (AIF), as a mitochondrial flavoprotein, plays a fundamental role in mitochondrial bioenergetics that is critical for cell survival and also mediates caspase-independent cell death once it is released from mitochondria and translocated to the nucleus under ischemic stroke or neurodegenerative diseases. Although alternative splicing regulation of AIF has been implicated, it remains unknown which AIF splicing isoform will be induced under pathological conditions and how it impacts mitochondrial functions and neurodegeneration in adult brain. Methods AIF splicing induction in brain was determined by multiple approaches including 5′ RACE, Sanger sequencing, splicing-specific PCR assay and bottom-up proteomic analysis. The role of AIF splicing in mitochondria and neurodegeneration was determined by its biochemical properties, cell death analysis, morphological and functional alterations and animal behavior. Three animal models, including loss-of-function harlequin model, gain-of-function AIF3 knockin model and conditional inducible AIF splicing model established using either Cre-loxp recombination or CRISPR/Cas9 techniques, were applied to explore underlying mechanisms of AIF splicing-induced neurodegeneration. Results We identified a nature splicing AIF isoform lacking exons 2 and 3 named as AIF3. AIF3 was undetectable under physiological conditions but its expression was increased in mouse and human postmortem brain after stroke. AIF3 splicing in mouse brain caused enlarged ventricles and severe neurodegeneration in the forebrain regions. These AIF3 splicing mice died 2–4 months after birth. AIF3 splicing-triggered neurodegeneration involves both mitochondrial dysfunction and AIF3 nuclear translocation. We showed that AIF3 inhibited NADH oxidase activity, ATP production, oxygen consumption, and mitochondrial biogenesis. In addition, expression of AIF3 significantly increased chromatin condensation and nuclear shrinkage leading to neuronal cell death. However, loss-of-AIF alone in harlequin or gain-of-AIF3 alone in AIF3 knockin mice did not cause robust neurodegeneration as that observed in AIF3 splicing mice. Conclusions We identified AIF3 as a disease-inducible isoform and established AIF3 splicing mouse model. The molecular mechanism underlying AIF3 splicing-induced neurodegeneration involves mitochondrial dysfunction and AIF3 nuclear translocation resulting from the synergistic effect of loss-of-AIF and gain-of-AIF3. Our study provides a valuable tool to understand the role of AIF3 splicing in brain and a potential therapeutic target to prevent/delay the progress of neurodegenerative diseases. Supplementary Information The online version contains supplementary material available at 10.1186/s13024-021-00442-7.

AIF is initially synthesized as a 67 kDa-precursor in the cytoplasm and imported into mitochondria where it is processed to a 62 kDa-mature AIF protein by removal of 53 amino acids at the N-terminus [22]. AIF mainly locates in the inter membrane space of mitochondria with its N-terminus inserted into the inner membrane [22]. A brain-specific AIF2 isoform, which contains an alternative exon 2B encoding a more hydrophobic mitochondrial inner membrane sorting signal than exon 2A in AIF mRNA, has been identified [23]. AIF2 is assumed to more strongly anchor to the inner membrane than AIF. Three other short AIF isoforms, including AIFsh, AIFsh2 and AIFsh3, have also been reported [2,3,6], although their biological functions and the molecular mechanisms accounting for the mRNA splicing remain unknown.
As constitutive deletion of Aifm1 in mice is embryonic lethal, several conditional AIF knockout mouse models have been developed using tissue-specific Cre recombinases to explore AIF function in vivo. Telencephalonspecific AIF knockout mice have defective cortical development and die by embryonic day 17 (E17), suggesting that AIF is required for neuronal cell survival during the cortical development [24]. Engrailed-Cre-mediated Aifm1 gene depletion in the midbrain and cerebellum at an early embryonic stage revealed that AIF is critical for both neuron survival and cerebellar development [25]. Harlequin (Hq) mice, which have an 80% reduction in AIF expression due to the proviral insertion into intron 1, develop oxidative stress-mediated neurodegeneration in cerebellum and retina during aging [26]. Apart from its cytoprotective role, AIF also plays an important role in parthanatos (poly(ADP-ribose) polymerase 1 (PARP-1)dependent cell death) after its release from mitochondria and translocation to the nucleus under ischemic brain injury and neurodegenerative diseases like Parkinson's disease and Alzheimer's disease [6,[27][28][29].
In the present study, we generated a floxed AIF mouse model by targeting exon 3 of Aifm1 gene with the original goal to understand AIF functions in adult mouse brain. Unexpectedly, the deletion of exon 3 in brain activates AIF splicing mechanism to produce a novel AIF isoform named as AIF3. Importantly, AIF3 is a natural isoform repressed under the physiological conditions but induced in mice and human brain following stroke. AIF3 splicing switch causes neurodegeneration in vitro and in vivo. The molecular mechanism underlying AIF3 splicing-triggered neurodegeneration involves AIF3 splicing-induced mitochondrial dysfunction and AIF3 nuclear translocation, which is attributed to the synergistic effect of both loss-of-AIF and gain-of-AIF3. These findings may provide new mechanistic insights into post-stroke neurodegeneration and other mitochondria-related neurological diseases.

Generation of tissue-specific Aif conditional knockout mice
The targeting vector for generating the AIF floxed (fl) allele consists of an 11 kb SphI fragment of the Aifm1 genomic DNA including 7.8 kb upstream and 3.2 kb downstream of the Aifm1 exon 3, and two LoxP sites at 114 bp upstream and 190 bp downstream of the Aifm1 exon 3, respectively. The G418 resistance gene (NEO) flanked by frt sites and the herpes simplex thymidine kinase (TK) gene inserted at 5′ of the targeting construct were used for positive and negative selection, respectively. The targeting construct was electroporated into 129/SV embryonic stem (ES) cells. Homologous recombination introduced a floxed exon 3 along with the Neo/ frt cassette in a correctly recombined ES clone. FlpE recombinase was overexpressed in the selected targeted 129/SV ES cells to remove the neomycin cassette, and Southern analysis was performed to screen for clones with the Neo cassette deletion. The selected clones were expanded and injected into C57BL/6 blastocysts to yield chimeric mice. Homozygous AIF fl/fl female mice and hemizygous AIF fl/Y male were backcrossed > 8 times onto the C57BL/6 background prior to initiation of the current studies. For neuron-specific deletion of Aifm1 exon 3, we crossed AIF fl/fl female with male mice expressing iCre recombinase under the control of the CamKIIα promoter (CamKIIα-iCre) [30,31] and hemizygous AIF fl/Y male mice were mainly used in this study. Alternatively, we crossed female (AIF fl/+ , CamKIIα-iCre+) mice with male (AIF fl/Y , CamKIIα-iCre-) mice to get homozygous female (AIF fl/fl , CamKIIα-iCre+) and its littermate (AIF fl/fl , CamKIIα-iCre-) for parallel studies. The body weight and growth curves were investigated in both homozygous AIF fl/fl female mice and hemizygous AIF fl/Y male mice. AIF protein expression was determined by AIF antibodies targeting different antigens (Table 1). Mouse genotyping primers and sizes of polymerase chain reaction (PCR) products were shown in Table 2.

Animals
Harlequin mouse (#000501; B6CBACa A w-J /A-Aifm1 Hq / J) and its control mouse (#000051, C57BL/6 J-A w-J /J) were purchased from Jackson Laboratory. CamkIIα-iCre mouse line (EM:01153) was purchased from EMMA. Ai-14 mouse (007908, Jackson Lab) was provided by Dr. Woo-Ping Ge at UT Southwestern. All mice were housed and cared in accordance with the NIH Health Guide for the Care and Use of Laboratory Animals in a pathogen-free facility with a 12 h light/dark cycle. UT Southwestern Medical School and The Johns Hopkins Medical Institutions are fully accredited by the American Association for the Accreditation of Laboratory Animal Care (AAALAC). All research procedures performed in this study were approved the Medical Institutions Institutional Animal Care and Use Committee (IACUC) in compliance with the Animal Welfare Act Regulations and Public Health Service (PHS) policy.

Middle cerebral artery occlusion (MCAO)
Cerebral ischemia was induced by 2 h of reversible MCAO as previously described [27]. Adult male C57BL/ 6 mice (2 to 4 month-old, 20-28 g) were first injected with Buprenorphine SR (0.5 mg/kg) right before the surgery and then anesthetized with isoflurane during the surgery and body temperature was maintained at 36.5 ± 0.5 o C by a feedback-controlled heating system. A midline ventral neck incision was made, and unilateral MCAO was performed by inserting a 7.0 nylon monofilament into the right internal carotid artery 6-8 mm from the internal carotid/pterygopalatine artery bifurcation via an external carotid artery stump. Laser-Doppler was used to measure the occlusion over the lateral parietal cortex in the core of the ischemic regions in the surgery mice. About 80-90% reduction of the blood circulation was controlled and monitored during the surgery. Sham-operated animals were subjected to the same surgical procedure, but the suture was not advanced into the internal carotid artery. Immediately following the surgery, mice were closely monitored in the  [32] presence of Buprenorphine SR, which provides up to 48 h of analgesia. Total 21 mice were used for this MCAO study and 18 mice were subjected to either sham or MCAO surgery. All mice waked up within 30 min after the surgery. After 6 h to 3 days of reperfusion, mice were perfused with PBS and stained with triphenyl tetrazolium chloride (TTC) or the brain tissues were harvested directly for RNA and protein analysis.

Human postmortem brain tissues
The frontal lobes from human postmortem brain without or with diffuse hypoxic ischemic injury or hemorrhage were collected during routine autopsy between 12 and 72 h after death and frozen at -80°C. The samples were provided by Children's Medical Center Pediatric Biospecimen Repository as well as Neuropathology tissue bank at UT Southwestern Medical Center. The consent for all autopsy samples includes use of tissue for research purposes. Only deidentified samples were provided with essential information including sex, age and the diagnosis as shown in Supplementary Figure 2e.

Open field tests
The open field tests were performed to evaluate the locomotor activity and anxiety in a rectangular plastic apparatus (25 cm × 25 cm × 25 cm) using the videotracking software system (SMART 3.0) from Pan Lab. The field was divided into peripheral zone (5 cm from the edge of walls) and central zone (36% of the total surface of the arena). The time spent in and entries into the center were measured as an anxiolytic indicator [34]. Mice were first transported to the behavior room and left undisturbed to get acclimated to the environment for at least 30 min. Then, each single mouse was placed  into the open field facing the center and allowed to explore freely for 8 min. After each experiment, the arena was thoroughly wiped with 70% ethanol solution to remove any feces, urine and odor. Finally, animal travel distance, walk speed and time in each zone for the first 5 min were analyzed using the SMART 3 software.

Mass spectrometry
Protein gel pieces was reduced and alkylated with DTT (20 mM) and iodoacetamide (27.5 mM). A 0.01 μg/μL solution of trypsin in 50 mM triethylammonium bicarbonate (TEAB) was added to completely cover the gel pieces on ice for 1 h, and then 50 μL of 50 mM TEAB was added to the gel pieces for digestion overnight at 37°C. Following solid-phase extraction cleanup with an Oasis HLB μelution plate (Waters), the resulting peptides were reconstituted in 10 μL of 2% (v/v) acetonitrile (ACN) and 0.1% trifluoroacetic acid in water. 2 μL of this resulting peptide solution were injected onto an Orbitrap Elite mass spectrometer (Thermo Electron) coupled to an Ultimate 3000 RSLC-Nano liquid chromatography systems (Dionex). Samples were injected onto a 75 μm i.d., 15cm long EasySpray column (Thermo), and eluted with a gradient from 1 to 28% buffer B over 60 min. Buffer A contained 2% (v/v) ACN and 0.1% formic acid in water, and buffer B contained 80% (v/v) ACN, 10% (v/v) trifluoroethanol, and 0.1% formic acid in water. The mass spectrometer operated in positive ion mode with a source voltage of 2.4 kV and an ion transfer tube temperature of 275°C. MS scans were acquired at 240, 000 resolution in the Orbitrap and up to 14 MS/MS spectra were obtained in the Orbitrap for each full spectrum acquired using collisionally-induced dissociation (CID) with a charge state 1 rejected. Dynamic exclusion was set for 15 s after an ion was selected for fragmentation. Raw MS data files were analyzed using Proteome Discoverer v2.2 (Thermo), with peptide identification performed using Sequest HT searching against the Mus musculus database from UniProt. Fragment and precursor tolerances of 10 ppm and 0.6 Da were specified, and three missed cleavages were allowed. Carbamidomethylation of Cys was set as a fixed modification and oxidation of Met as a variable modification. The falsediscovery rate (FDR) cutoff was 1% for all proteins and peptides.

Protein expression and purification
Full length mouse AIF cDNA (NM_012019) and AIF3 cDNA missing exons 2-3 (ΔE2-3) but still containing mitochondrial localization signal (MLS) in exon 1 were cloned into cFugw vector at AgeI and EcoRI sites with Flag tag at its C-terminus for mammalian expression system. cDNAs encoding mouse 62 kDa-mature AIF (Δ53 aa) and AIF3 (ΔE1-3, removing MLS) proteins were subcloned into a GST-tagged pGex-6P-1 vector (GE HealthCare) at EcoR I and Xho I restriction sites. All AIF constructs were verified by Sanger sequencing. The proteins were expressed and purified from E.coli using glutathione sepharose and the GST-tag was subsequently removed by the precision protease but with 8 aa linker remaining on the AIF. To exclude the interference from the tag, non-tagged recombinant AIFs were used for the in vitro biochemical determinations. The protein purity, stability and concentrations were determined in the Coomassie blue gel using bovine serum albumin (BSA, 98% purity) as the standard.

Lentivirus construction and virus production
Full-length cDNAs of AIF-Flag and AIF3-Flag (Supplementary Figure 1f) were subcloned into a lentiviral cFugw vector at Age I and EcoR I restriction sites, and their expression were driven by human ubiquitin C (hUBC) promoter. The lentivirus was produced by transient transfection of the recombinant cFugw vector into 293 T cells with three packaging vectors: pLP1, pLP2 and pVSV-G (1.3: 1.5:1:1.5). The viral supernatants were collected at 48 h and 72 h after transfection and concentrated by ultracentrifuge for 3 h at 50,000 g.

Mouse embryonic fibroblast (MEFs) culture
MEFs were prepared from the time pregnant AIF fl/fl female mice carrying embryos at E13 after crossing with AIF fl/Y male mice as described previously and maintained in Dulbecco's Modified Eagle Medium (DMEM) [35]. Then MEFs were infected with either polyomavirus simian virus 40 (SV40) or SV40 together with Cre lentivirus. Single MEF clones were picked and AIF3 expression was verified by western blot after Cre lentiviral infection.

Mouse cortical neuron culture, transfection, lentiviral transduction and cytotoxicity
Primary neuronal cultures from the cortex were prepared as described previously [36]. Briefly, the cortex was dissected and the cells dissociated by trituration in modified Eagle's medium (MEM), 20% horse serum, 30 mM glucose and 2 mM L-glutamine following a 10 min digestion in 0.027% trypsin/saline solution (Gibco BRL). Neurons were plated on 15 mm multiwell plates coated with polyornithine or on coverslips coated with polyornithine. Neurons were maintained in MEM with 10% horse serum, 30 mM glucose, and 2 mM L-glutamine in a 7% CO 2 humidified 37°C incubator. The growth medium was refreshed twice per week. In mature cultures, neurons represent 70-90% of the total number of cells. In DIV 7-9, neurons were infected by lentivirus carrying AIF-Flag, AIF3-Flag, Cre-GFP or GFP (1 × 10 9 T.U./ml) for 72 h. Cell viability was determined at DIV 10-13 by unbiased objective computer-assisted cell counting after staining of all nuclei with 7 μM Hoechst 33342 (Invitrogen) and dead cell nuclei with 2 μM propidium iodide (Invitrogen). The numbers of total and dead cells were counted with the Axiovision 4.3 software (Carl Zeiss). At least three independent experiments using at least six separate wells were performed with a minimum of 15,000-20,000 neurons or cells counted per data point.

Determination of NADH oxidase activity and FAD binding of AIF protein
The redox activity of AIF protein was determined as described [29]. In brief, the redox activity was measured at room temperature in the substrate solution containing 250 μM NADH in air-saturated 50 mM Tris-HCl, pH 8.0. The decrease in absorbance at OD340 nm was monitored right after the addition of recombinant AIF and AIF3 (3 μM in 50 μl reaction volume) into the substrate solution. FAD (25 nM) binding to AIF was monitored at 1 μM concentration of AIF and AIF3 by wave length scanning ranging from 250 to 800 nm using a Beckman Coulter DU800 UV/Vis spectrophotometer (Fullerton).

DNA gel retardation assay
AIF binding-mediated DNA mobility retardation was performed as described [37]. Recombinant AIF, AIF3 and bovine serum albumin (BSA,10 μg) were incubated with 200 ng 1 kb plus DNA ladder (Fermentas) for 30 min at room temperature and loaded on 2.5% agarose gel prestained with ethidium bromide (1 μg /mL). Mobility shift was visualized under the Alpha Innotech UV illuminator (San Leandro).

Electron microscopy
Mice were first perfused through the heart with PBS (pH 7.4) for 30 s, and then perfused with 3% paraformaldehyde, 1.5% glutaraldehyde, 100 mM cacodylate and 2.5% sucrose (pH 7.4), followed by postfixing for 1 h. Mouse cortex, hippocampus and cerebellum sections were processed as described [38]. They were postfixed in Palade's OsO 4 , en bloc stained in Kellenberger's uranyl acetate, dehydrated, and flat-embedded in epon. 80 nm en face sections were prepared on a Leica UCT ultramicrotome, collected onto 400 mesh high transmission nickel grids, and post stained with lead and uranyl acetate. Images were taken by a Philips EM 410 TEM equipped with a Soft Imaging System Megaview III digital camera.

Mitochondrial DNA copy measurement
Total DNA was isolated from different tissues. The amount of mitochondrial DNA relative to nuclear genomic DNA was determined by quantitative real-time PCR using primers ( Table 2) for mitochondrial DNAencoded cytochrome b and nuclear DNA-encoded ribosomal protein S2 (RPS2). Relative mitochondrial DNA levels were calculated based on the threshold cycle (Ct) Intracellular adenosine triphosphate (ATP) ATP levels were measured using an ATP assay kit (Promega) according to the manufacturer's instructions. Luminescence was measured using a GloMAX 20/20 single tube luminometer (Promega) and normalized to the protein concentration.

Measurement of oxygen consumption rate (OCR)
The OCR was measured using a cartridge containing an optical fluorescent O 2 sensor in a Seahorse Bioscience XF24 Extracellular Flux Analyzer (Seahorse Bioscience). Mouse cortical neurons prepared from Hq mice or AIFfl/Y /AIF fl/fl mice were seeded and cultured in XF24-well microplates at 2.5 × 10 5 cells per well for 3 days and infected with lentivirus as indicated for 3 days. OCR was measured after equilibration in assay medium containing 30 mM glucose for 1 h. Baseline rates were measured three times. Three testing chemicals, including oligomycin, FCCP, and rotenone, were preloaded in the reagent delivery chambers of the sensor cartridge and then pneumatically injected into the wells to reach the desired final working concentrations. After 2 min of mixing, postexposure OCR was measured for three times. The averages of three baseline rates and three test rates were used for data analysis. OCR was shown in the unit of picomoles per minute.

Nissl staining
Nissl staining was performed using standard methods as previously described [39]. For immunohistochemical analysis, mice were anesthetized with a lethal dose of pentobarbital (100 mg/kg) and perfused with 4% paraformaldehyde (Sigma). Brains were extracted, postfixed and cryoprotected in 30% sucrose. Free-floating sections (30 μm) were harvested with a microtome and processed for immunohistochemical analysis.

Experimental design and statistical analysis
AIF3 splicing mice and AIF3 conditional KI mice were housed with their own littermate controls in the same cage and identified by the unique number ID with genotyping. Animal experiments were performed in a blind fashion and decoded when data analysis. MCAO experiments with C57BL/6 mice were performed in a randomized fashion. All experiments were repeated at least three times. All statistical evaluation was carried out using GraphPad Prism 8 software. Student's t-test was performed between two groups in Figs. 2l, 3e-h, 5c, 8f and 10f-g. One-way analysis of variance (ANOVA) was performed in Figs. 2i, 4c, 6b, 8c, and 9c, e-f and two-way ANOVA were performed in Figs. 3a-d, 8h and 9a-b by GraphPrism software within multiple groups. Data are shown as mean ± S.E.M., p < 0.05 is considered significant.

Identification of a novel AIF isoform-AIF3
To explore AIF functions in adult brain, exon 3 of murine Aifm1 was flanked by loxP (AIF fl/fl ) ( Fig. 1a and Supplementary Figure 1a), and AIF fl/fl mice were crossed with CamKIIα-iCre mice [30]. The 62 kDa mature AIF protein in AIF fl/Y /CamKIIα-iCre+ mice was significantly reduced in the cortex, hippocampus, striatum, thalamus, hypothalamus, and midbrain, but not in the cerebellum and brain stem (Fig. 1b). This is consistent with the expression distribution of the CamKIIα gene predominantly in the adult forebrain [31]. Unexpectedly, an additional AIF protein of 57 kDa was detected in all brain regions of AIF fl/Y /CamKIIα-iCre+ mice but not in AIF fl/Y /CamKIIα-iCre-littermates, which was first observed at postnatal day 20 (P20) (Supplementary Figure  1b) and persistent for about 3 months till their death (Fig. 1b). The 57 kDa protein was recognized by at least three different AIF antibodies targeting different amino acid sequences of AIF (Fig. 1b, Supplementary Figure 1b, c, and Table 1). However, it was not observed in other organs of AIF fl/Y /CamKIIα-iCre+ mice, including heart, lung, liver and limb (Supplementary Figure 1d), or in any brain regions of CamKIIα-iCre+ control mice (Supplementary Figure 1e). These results suggest that the 57 kDa protein may be a novel AIF isoform.
To ascertain whether a novel AIF isoform was induced in AIF fl/Y /CamKIIα-iCre+ mice, we prepared cDNAs from AIF fl/Y /CamKIIα-iCre+ mice brain and their littermate controls (AIF fl/Y /CamKIIα-iCre-), and performed PCR using primers specific to exon 1 and 16 of the Aifm1 gene (PCR ID-mAIF, Table 2). A second smaller PCR product of 1697 bp was detected in the cortex of AIF fl/Y /CamKIIα-iCre+ mice in addition to the fulllength AIF cDNA (1940 bp) (Fig. 1c). Sanger sequencing analysis revealed that this smaller cDNA encoded AIF lacking exons 2 and 3 ( Fig. 1c and Supplementary Figure  1f, g), and is referred to as AIF3 to distinguish it from the other two previously characterized AIF isoforms.
To further determine whether this 57 kDa AIF3 protein was a novel AIF isoform but not a degradation product, we prepared MEFs from AIF fl/Y /AIF fl/fl mouse embryos at E13, and infected MEFs with Cre lentivirus to induce endogenous AIF3 splicing and established multiple immortalized single clones expressing AIF3 splicing instead of AIF (Fig. 1d). Then we  (Fig. 1e). The peptide link LPA-TEGGSVPQIR between exons 1 and 4 was uniquely detected in AIF fl/Y /Cre+;AIF fl/fl /Cre + clone 3 MEFs, but not in WT MEFs (Fig. 1f, g). Conversely, the peptide link AIASATEGGSVPQIR between exons 3 and 4 was only detected in WT MEFs (Fig. 1f, g). Peptides encoded by Aifm1 exons 4-16 were observed in both WT and AIFfl/Y /Cre+;AIF fl/fl /Cre + clone 3 MEFs. Collectively, mRNA analysis and bottom-up proteomic analysis revealed the presence of a novel 57 kDa AIF3 isoform in the brain of AIF fl/Y /CamKIIα-iCre+ and AIF fl/fl /CamKIIα-iCre+ mice.
AIF3 is a natural splicing isoform and induced following stroke in mouse and human To determine whether AIF3 is a natural splicing isoform, 5′ RACE assay, a method to detect the natural minor isoform, was performed using cDNAs prepared from C57BL/ 6 mouse brain to amplify the 5′ ends of AIF (Fig. 2a). All three AIF isoforms, including AIF, AIF2 and AIF3, were identified by 5′ RACE (Fig. 2b), indicating that AIF3 is a natural splicing isoform. Although AIF3 mRNA and protein levels were below detection limits in mouse brain under physiological conditions (Fig. 1b, c), we found that AIF3 protein was induced in the ipsilateral brain 24 h following 2 h middle cerebral artery occlusion (MCAO) (Fig. 2c & Supplementary Figure 2a). Using primers recognizing AIF exons 1 and 5 (mAIF-Fw34/Re648, Table 2), AIF mRNA was detected in both stroke and sham groups, but AIF3 mRNA was upregulated in brain with ischemic stroke (Fig. 2d). The resulting AIF3 PCR product from exons 1 to 5 was confirmed by Sanger sequencing. The link sequence between exons 1 and 4 was obviously detected in ischemic mouse brain (Fig. 2e). To avoid the interference from AIF, we designed an AIF3-specific PCR primer (mAIF-Fw2E1/4) recognizing the linkage of exons 1 and 4 ( Fig. 2f, g). Using AIF3 specific PCR primers (PCR ID-mAIF3, Table 2), we found that AIF3 mRNA was induced in mouse brain at 6 h reperfusion after 2 h MCAO and further elevated at 24 h reperfusion, whereas no AIF3 mRNA was obviously detected in sham mice (Fig. 2g). Consistent with mRNA induction, AIF3 protein levels were increased 24 h following stroke and its upregulation persisted for at least 72 h in mouse brain (Fig. 2h, i). In contrast, AIF3 was not induced in cultured cortical neurons following treatment with N-methyl-D-aspartate (NMDA, 500 μM for 5 min or 100 μM for 2 h), staurosporine (STS, 1 μM), oxygen-glucose deprivation (OGD, 90 min), or hypoxia (1% O 2 for 24-72 h) (Supplementary Figure 2b, c, d).
To determine the potential clinical relevance of AIF3, AIF3 expression was analyzed in brain frontal cortex from both male and female patients at ages varying from 20 days to 73 years who suffered either hypoxic-ischemic or hemorrhagic stroke (Supplementary Figure 2e). Using AIF3 specific primers (PCR ID-hAIF3, Table 2), the expression of AIF3 mRNA was observed in all 5 human patients tested, which was further confirmed by Sanger sequencing (Fig. 2j, Supplementary Figure 2e). AIF3 protein expression was also increased in these human patients with hypoxic-ischemic injury (Fig. 2k, l, Supplementary Figure  2e). In contrast, little AIF3 mRNA and protein expression was detected in control human brain tissues (Fig. 2j, k, l). These data indicate that AIF3 is a novel AIF splicing isoform and induced in brain following hypoxic-ischemic injury in both mice and humans.

AIF3 splicing switch causes progressive neurodegeneration in vivo
To determine the role of AIF3 splicing in brain, we took advantage of AIF fl/Y /CamKIIα-iCre+ mice, in which AIF3 is spliced endogenously instead of AIF in brain. Comparing hemizygous male AIF3 splicing mice (AIFfl/Y /CamKIIα-iCre+) and homozygous female AIF3 splicing mice (AIF fl/fl /CamKIIα-iCre+) to their littermates (See figure on previous page.) Fig. 1 Identification of a novel AIF isoform-AIF3. a Strategy for conditional deletion of exon3 of mouse Aifm1 gene. A partial restriction map of Aifm1 gene locus and the targeting vector are shown on the first and second lines. Exons of the murine Aifm1 gene are indicated as a filled box and labeled. Exon 3 is flanked by loxP sequences which are indicated by triangles. The G418 resistance gene (Neo) flanked by frt sites is indicated with the open box. The PGK-herpes simplex thymidine kinase (TK) indicated by thick line is inserted at the 5′ end of the targeting vector. Restriction endonuclease sites used for cloning are shown (SphI, EcoRl; XbaI; BglII). Homologous recombination introduces a floxed exon 3 along with the Neo cassette, which is excised by transient overexpression of flpE recombinase in a correctly recombined ES clone. The floxed exon 3 in the Aifm1 allele is further excised by mating with mice expressing Cre recombinase in a brain specific manner. b Expression of AIF3 in different brain regions in AIF fl/Y mice after CamKIIα-iCre recombination at P70 determined by JHU-532 AIF antibody. The recombinant 62 kDa mature AIF (Δ53 aa) and 57 kDa truncated AIF (Δ103 aa), which have an extra 8 aa linker, were used as markers. c mRNA expression and sequence analysis of AIF3 using AIF gene specific primers on its exon 1 and exon 16 (PCR ID: mAIF, Table 2) and cDNA prepared from WT (AIF fl/Y /CamKIIα-iCre-) and AIF3 (AIF fl/Y /CamKIIα-iCre+) mice at P20 as the template. d Establishment of AIF3 MEFs single cell lines from AIF fl/Y /AIF fl/fl mice determined by E1-AIF antibody. e, AIF peptides bridging Exons 1/2, Exons 3/4, and Exons 1/4. f & g MS/MS spectra for AIF and AIF3 peptides bridging Exons 3/4 and Exons 1/4, respectively (AIF fl/Y /CamKIIα-iCre-, or AIF fl/fl /CamKIIα-iCre-), we found that both male and female AIF3 splicing mice were born at the expected Mendelian frequency and survived for the first 60 days (Fig. 3a, b, c, d). However, all AIF3 splicing mice died between P60 and P140 (Fig. 3a,  b). None of their littermate controls died before P140. Moreover, all AIF3 splicing mice were smaller than their littermate controls and showed severe weight loss at Fig. 2 AIF3 was induced following ischemic stroke in mouse model and human patients. a The 5′-RACE scheme of AIF cDNAs from C57BL/6 mouse brain. b All three AIF isoforms identified by 5′ RACE. c Expression of AIF3 protein in C57BL/6 mouse brain 24 h after 2 h MCAO. Contra, contralateral; Ipsi, ipsilateral. PC (positive control), cerebellum lysate of AIF3 (AIF fl/Y /CamKIIα-iCre+) mice. d Expression of AIF3 mRNA in ischemic stroke mouse cortex using the common AIF/AIF3 primers (mAIF-Fw34/Fe648, Table 2). Top image is the 1st PCR with 20 cycles, middle image is the 2nd PCR using AIF3 from 1st PCR as template, and bottom image is GAPDH. e Sanger sequencing of AIF3 mRNA in ischemic stroke mouse cortex prepared from d. f Scheme for AIF3 mRNA specific primer design. g Expression of AIF3 mRNA in ischemic stroke mouse cortex using AIF3 specific primers (PCR ID: mAIF3, Table 2). MEFs were used as controls. h Expression of AIF3 protein in ischemic stroke mouse cortex at 24 h, 48 h and 72 h following 2 h MCAO determined by AIF E1 antibody. i Quantification of AIF3 protein expression in ischemic stroke mouse cortex at 6 h-72 h following 2 h MCAO. AIF3 protein levels were normalized to actin and presented as the relative intensity. Data are shown as mean ± S.E.M and analyzed for statistical significance by one-way ANOVA. n = 3. ***p < 0.001. j Expression of AIF3 mRNA in human cortex with stroke using AIF3 specific primers (PCR ID: hAIF3, Table 2). k Expression of AIF3 protein in the cortex of human stroke patients. PC, cerebellum lysate of AIF3 (AIF fl/Y /CamKIIα-iCre+) mice. l Quantification of AIF3 expression protein in human stroke patients. Data are shown as mean ± S.E.M and analyzed for statistical significance by Student t test. n = 5. ***p < 0.001 P60-P90 (Fig. 3c, d). Open field tests were performed to evaluate the spontaneous locomotor activity and anxietylike behaviors. We found that the total travel distance and the speed of mouse walking were not obviously changed at P60-P75. However, the travel paths and percentage of time spent in the center zone were significantly decreased in AIF3 groups compared to their littermates (Fig. 3e, f, g,  h), indicating increase of anxiety in AIF3 splicing mice.
Next, we characterized AIF3 splicing mouse and found that ventricles were enlarged in AIF3 splicing mouse brain compared to its littermate control at P90 (Fig. 4a). Substantial neurodegeneration was observed in the forebrain of Data were shown as mean and analyzed for statistical significance by two-way ANOVA. *** p < 0.001. c-d Measurement of the body weight of male (c, n = 20) and female (d, n = 20) AIF3 splicing mice and their littermate AIF controls. Data are shown as mean ± S.E.M. and analyzed for statistical significance by two-way ANOVA. *** p < 0.001. e-h Open field tests of AIF3 (n = 13) and their littermate controls (n = 10) at P60-75. Central zone travel distance (e), percentage of time in the Central zone (f), total travel distance (g) and walk speed (h) were analyzed. Data are shown as mean ± S.E.M. and analyzed for statistical significance by Student t test. ** p < 0.01; *** p < 0.001, n = 10-13 Fig. 4 AIF3 splicing caused severe neurodegeneration in the forebrain of 3-month-old mice. a Representative overview images of neurodegeneration in mouse brain, including somatosensory cortex, piriform cortex, thalamus and hippocampus (i-ii, i'-ii', i"-ii") as well as cerebellum (iii, iii', iii"). i-iii, drawing images for coronal mouse brain sections. Somatosensory (orange), piriform cortex (pink) and thalamus (blue) are highlighted. i'-iii', Nissl staining images of the littermate control AIF mouse (AIF fl/Y /CamKIIα-iCre-). i"-iii", Nissl staining images of AIF3 splicing mouse (AIF fl/Y /CamKIIα-iCre+). b Nissl staining of somatosensory cortex (i, i'), piriform cortex (ii, ii'), thalamus (iii, iii'), hippocampus (iv-vii, iv'-vii') and cerebellum (viii, viii') from AIF3 splicing mice and their littermate AIF mice. The boxes in i-viii and i'-viii' are shown at higher magnification. Scale bar, 200 μm. c Quantification of the brain lesion volume (n = 5). Cortex, Somatosensory cortex; Hip, hippocampus; Piriform, piriform cortex. Data are shown as mean ± S.E.M. and analyzed for statistical significance by one-way ANOVA. **** p < 0.0001 AIF3 splicing mice, especially in the somatosensory cortex (Fig. 4b, i-i' & c), piriform cortex (Fig. 4b, ii-ii' & c), thalamus (Fig. 4b, iii-iii' & c), and hippocampus (Fig. 4b, iv-vii, iv'-vii' & c). In contrast, no granule cell loss was observed in the cerebellum (Fig. 4b, viii, viii' & c). CamKIIα is highly expressed in Purkinje cells in the cerebellum in addition to the forebrain [40]. Consistently, expression of AIF3 significantly reduced the number of calbindin-positive Purkinje cells in 3-month-old AIF3 splicing mice (Fig. 5). Taken together, these data indicate that AIF3 splicing switch leads to neurodegeneration in the mouse brain.
CamKIIα gene expression is negligible until birth and increased 10-fold during P1-P21 and an additional 2.5fold increase by P90 [31]. Thus, we compared the effect of AIF3 splicing switch on neuron loss at P20, P60 and P90 by Nissl staining. No obvious neuron loss was observed in the cortex or hippocampus of AIF3 splicing mice at P20, similar to WT littermates (Fig. 6a, b). At P60, partial neuron loss was observed in the cortex (25%), but not hippocampus, of AIF3 splicing mice (Fig.  6a, b). In contrast, there was substantial neuron loss in the somatosensory cortex (80%) and CA1 and CA3 regions of the hippocampus (48%) in AIF3 splicing mice at P90 (Fig. 6a, b). To further determine if there are specific types of neuron loss at P30 and P60, immunostaining using cortex layer markers TBR1 (mainly in layer VI), CUX1 (mainly in Layer II-III) and NeuN (general neuron marker), was performed. At P30, the expression of TBR1 and CUX1 was observed in both AIF3 splicing mice and its littermate controls (Supplementary Figure 3). The difference if any between AIF3 splicing mice and its littermate controls was not obvious. Remarkably, we found that the expression of TBR1 and CUX1 was almost completely eliminated from AIF3 cortex layers at P60 (Fig. 7a, b), although NeuN staining (Fig. 7c), DAPI staining (Fig. 7) and Nissl staining (Fig. 6) were only reduced moderately. These findings reveal that AIF3 splicing switch causes progressive neurodegeneration without obviously affecting neural differentiation process.

Expression of AIF3 in cortical neurons causes neurotoxicity in vitro
To distinguish the actions of AIF and AIF3 in neurodegeneration, we prepared cortical neurons from Hq mice and infected cells with lentivirus carrying Flag-tagged AIF or AIF3. Both AIF and AIF3 protein were similarly expressed in Hq cortical neurons 3 days after lentiviral infection (Fig. 8a). AIF3 expression caused neuronal cell death at day 4 following lentiviral infection, whereas expression of AIF or GFP did not induce neurotoxicity (Fig. 8b, c). To further determine the role of endogenous AIF3, cortical neurons were prepared from E15.5 AIFfl/Y /AIF fl/fl embryos and then infected with lentivirus carrying GFP or Cre-GFP (Fig. 8d). Three days after Crelentiviral transduction, AIF3 was successfully induced in neurons and AIF expression was decreased (Fig. 8d). Five days after Cre-GFP lentiviral transduction, over 50% of neurons died, while GFP-lentiviral transduced neurons exhibited minimal neuronal cell death (Fig. 8e, f). These data indicate that expression of AIF3 causes cytotoxicity in cultured neurons in vitro.

AIF3 splicing switch mediates mitochondrial dysfunction in vitro and in vivo
AIF has flavin adenine dinucleotide (FAD)-binding and DNA-binding capabilities and also acts as an NADH oxidase in mitochondria [6]. GST-tagged AIF (62 kDa mature form) and AIF3 proteins were expressed and purified from bacteria, and then the GST tag was removed (Fig. 8g). NADH oxidase activity of purified AIF3 and AIF were compared by monitoring changes in absorbance in vitro. NADH oxidase activity of AIF3 (3 μM) was significantly lower than that of AIF at the same concentration (Fig. 8h). Both AIF and AIF3 had typical FAD-binding peaks at 378 nm, 450 nm and shoulder at 467 nm (Fig. 8i). However, the FAD-binding affinity of AIF3 (1 μM) was substantially reduced (Fig. 8i). Unlike their NADH oxidase activity and FAD-binding affinity, both AIF3 and AIF had a similar binding ability to DNA and caused DNA retardation (Fig. 8j). These data indicate that AIF3 has reduced NADH oxidase activity and FAD binding ability in vitro, but its DNA binding activity is retained.
Oxidation of NADH in mitochondria is critical for ATP synthesis. To study the role of AIF3 splicing switch in mitochondrial respiration, cortical neurons were prepared from AIF fl/Y /AIF fl/fl embryos and infected with Cre-GFP lentivirus. AIF3 splicing switch significantly decreased the basal oxygen consumption rate as compared to controls transduced with GFP lentivirus (Fig. 9a). Hq cortical neurons also had lower oxygen consumption rate due to loss-of-AIF (Fig. 9b). Expression of AIF in Hq cortical neurons significantly increased the basal oxygen consumption rate, whereas expression of AIF3 failed to rescue the basal oxygen consumption rate (Fig.  9b). Consistently, overexpression of AIF in Hq cortical neurons increased ATP production, whereas AIF3 Fig. 7 Characterization of neuron loss in the cortex of AIF3 splicing mice at P60. a Representative images of TBR1 + neurons in AIF3 splicing mice (AIF fl/Y /CamKIIα-iCre+) and littermate control AIF mice (AIF fl/Y /CamKIIα-iCre-) at P60. Scale bar, 200 μm. b Representative images of CUX1 + neurons in AIF3 splicing mice (AIF fl/Y /CamKIIα-iCre+) and littermate control AIF mice (AIF fl/Y /CamKIIα-iCre-) at P60. Scale bar, 200 μm. c Representative images of NeuN + neurons in AIF3 splicing mice (AIF fl/Y /CamKIIα-iCre+) and littermate control AIF mice (AIF fl/Y /CamKIIα-iCre-) at P60. Scale bar, 200 μm. Representative images are from at least three independent experiments expression in Hq cortical neurons did not rescue the ATP production (Fig. 9c). In contrast, it even significantly reduced ATP levels (Fig. 9c). These data suggest that AIF3 at least loses part of AIF functions and contributes to the dysregulation of mitochondrial oxygen consumption rate and ATP production. by Student t test. g AIF (62 kDa) and AIF3 protein purification. BSA was used as purity and quantity control. h The NADH oxidase activity of purified AIF (3 μM) and AIF3 (3 μM) was determined by monitoring changes in absorbance at OD340 nm. Data were shown as mean ± S.E.M. n = 3. ***p < 0.001 by two-way ANOVA. i FAD binding of purified AIF (1 μM) and AIF3 (1 μM) was determined by spectrophotometric wavelength scanning. j DNA retardation assay of purified AIF (3 μM) and AIF3 (3 μM) by incubating with 1 kb plus DNA ladder (200 ng) for 30 min. BSA was used as a negative control To further study the role of AIF3 splicing switch in mitochondrial functions, we examined mitochondrial ultrastructure in AIF3 splicing mouse brain at P70 by electron microscopy. The majority of neurons in the cortex, hippocampus, and cerebellum of AIF3 splicing mice had severe defects in mitochondrial morphology and cristae organization (Fig. 9d, e). Mitochondria in AIF3 splicing mice were larger in diameter and exhibited vesiculation of their inner membranes, whereas littermate control mitochondria were intact and well organized (Fig. 9d, e). Mitochondrial DNA mass was significantly reduced in the forebrain, cerebellum, and brain stem of AIF3 splicing mice compared to littermate controls (Fig. 9f). No difference in mitochondrial mass was observed in heart and liver of AIF3 and AIF mice where AIF3 was not expressed (Fig. 9f; Supplementary Figure 1c). Taken together, our findings demonstrate that AIF3 splicing switch inhibits ATP production, oxygen consumption, and mitochondrial biogenesis, leading to mitochondrial dysfunction in mouse brain.

AIF3 splicing switch promotes chromatin condensation
AIF is translocated to the nucleus to induce chromatin condensation and nuclear shrinkage, leading to cell death [28,29]. The subcellular localization of AIF3 was studied by immunochemical and MitoTracker staining. We showed that exogenous Flag-AIF and Flag-AIF3 were colocalized with endogenous AIF in cortical neurons (Fig. 10a). Likewise, AIF3 was localized at mitochondria as indicated by MitoTracker and colocalized with endogenous AIF in MEFs, where expression of AIF3 did not obviously alter mitochondrial size or morphology (Fig. 10b). In vivo immunohistochemistry of Fig. 9 Effects of AIF and AIF3 expression on mitochondrial functions in vitro and in vivo. a OCR in AIF fl/Y /AIF fl/fl cortical neurons 3 days after GFP or Cre-GFP lentivirus infection was measured after exposure to 1 μM oligomycin, 3 μM FCCP, and 1 μM rotenone. Data are shown as mean ± S.E.M. n = 3. ***p < 0.001 by two-way ANOVA. b OCR in Hq cortical neurons 3 days after AIF or AIF3 lentiviral infection was measured by Seahorse Bioscience XF24 Extracellular Flux Analyzer after exposure to 1 μM oligomycin, 3 μM FCCP, and 1 μM rotenone. Data are shown as mean ± S.E.M. n = 3. ***p < 0.001 by two-way ANOVA. c Intracellular ATP levels were measured in Hq cortical neurons 3 days after AIF or AIF3 lentiviral infection using a luminescence ATP assay kit. Data are shown as mean ± S.E.M. n = 3. *p < 0.05, ***p < 0.001. d Mitochondrial structure of neurons in the cortex, hippocampus and cerebellum of AIF3 splicing mice (AIF fl/Y /CamKIIα-iCre+) and their littermate control AIF mice (AIF fl/Y /CamKIIα-iCre-) by electron microscopy. e Quantification of abnormal mitochondria (mito) in the cortex, hippocampus (Hip) and cerebellum of AIF3 splicing mice and their littermate control AIF mice. Data are shown as mean ± S.E.M. n = 5. ****p < 0.0001 by one-way ANOVA. f AIF3 splicing reduced mitochondrial biogenesis in brain, but not heart or liver of AIF3 splicing mice. The ratio of mitochondrial DNA relative to nuclear genomic DNA was determined by quantitative real-time PCR using primers for cytochrome b (mitochondrial) and ribosomal protein S2 (RPS2, nuclear). The data observed in AIF3 splicing mice were normalized to the littermate control AIF mice. Data are shown as mean ± S.E.M. n = 3. ***p < 0.001, **p < 0.01, *p < 0.05 by one-way ANOVA the cortex further showed that AIF3 was localized at mitochondria, and only a small portion of AIF3 was observed in the nucleus at P20. In contrast, the majority of AIF3 was translocated into the nucleus at P60 (Fig. 10c,  d). At P90, little AIF3 staining was detected in the cortex of AIF3 splicing mice due to severe neuron loss (Fig. 10c,  d). Distinct from AIF3, AIF was primarily found at mitochondria from P20 to P90 (Fig. 10c, d). Electron microscopy revealed that approximately 81% of nuclei in the cortex of AIF3 splicing mice displayed chromatin condensation and nuclear shrinkage, which was significantly higher than that in the littermates (5.3%) (Fig. 10e,  f). Similar phenotypes were also observed in the cerebellum of AIF3 splicing mice (Fig. 10e, g). Taken together, Fig. 10 Expression of AIF3 increased chromatin condensation and nuclear shrinkage. a Immunostaining of Flag-tagged AIF and AIF3 and their colocalization with endogenous AIF in cortical neurons. OE, overexpression. Scale bar, 20 μm. b Immunostaining of Flag-tagged AIF and AIF3, endogenous AIF, and mitotracker (mito) in MEFs. OE, overexpression. Scale bar, 20 μm. Representative images are from at least three independent experiments. c Representative AIF3/AIF localization changes in the cortex of AIF3 splicing mice (AIF fl/Y /CamKIIα-iCre+) and littermate control (AIF fl/Y /CamKIIα-iCre-) AIF mice at P20, P60 and P90. Scale bar, 50 μm. d High magnification images for highlighted areas in b. i, ii, iii, AIF mice at P20, P60 and P90 respectively, and i', ii", iii", AIF3 splicing mice at P20, P60 and P90 respectively. Scale bar, 50 μm. e, Representative images of nuclei of neurons in the cortex and cerebellum of AIF3 splicing mice (AIF fl/Y /CamKIIα-iCre+) and littermate control AIF mice (AIF fl/Y /CamKIIα-iCre-) by electron microscopy. f & g Quantification of nuclear shrinkage of neurons in the forebrain (f) and cerebellum (g) of AIF3 splicing mice and their littermate control AIF mice. Data are shown as mean. n = 4. Representative images are from at least three independent experiments these data indicate that AIF3 is prone to translocate to the nucleus and promote chromatin condensation and nuclear shrinkage, thereby increasing neuronal susceptibility to death in vivo.
The synergistic effect of loss-of-AIF and gain-of-AIF3 contributes to AIF3 splicing switch-induced neurodegeneration in vivo AIF3 splicing switch from AIF to AIF3 involves both gain-of-AIF3 and loss-of-AIF events. We next dissected whether loss-of-AIF, gain-of-AIF3, or both contributes to neurodegeneration in vivo. To study whether loss-of-AIF mainly contributes to neurodegeneration observed in AIF3 splicing mice, we used Hq mice as a model as conditional deletion of AIF in mouse brain is embryonic lethal and Hq mice have 80% of AIF reduction (Fig. 1,  Supplementary Figure 1b, c, e). We found that no obvious neuron loss was detected in the cortex or hippocampus of male Hq mice at P100 (Fig.11a, b), despite Purkinje neuron loss in the cerebellum of 6-month-old Hq mice reported previously [10,26,41]. However, 19.4% of Hq mice showed enlarged ventricle (Fig. 11c), which was a feature observed in AIF3 splicing mice. Hq hemizygous males and homozygous females showed a smaller body weight at 3-6 months old [10,41]. These data suggest that loss-of-AIF is necessary, but not sufficient, to cause robust neurodegeneration observed in AIF3 splicing mice.
To explore the role of gain-of-AIF3 in neurodegeneration in the presence of AIF, we established a conditional AIF3KI mouse model by inserting the loxP-stop-loxP-AIF3 (LSL-AIF3) sequence into Rosa26 locus using CRIS PR/Cas9 technique and crossed these mice with CamkIIα-iCre mice (Fig. 12a, b, c). Expression of heterozygous AIF3 in CamkIIα-iCre+ mice was induced in the forebrain (Fig. 12c). Both male and female AIF3KI mice appeared normal for at least 7 months. No obvious neuron loss was observed in the cortex and hippocampus of AIF3KI mice at P105 (Fig. 12d). However, the lateral ventricles in AIF3KI mice were enlarged (Fig. 12d). Moreover, the neuronal cell density was reduced in both cortex and hippocampus (Fig. 12e, f). The consistent results were also found in another AIF3 KI mouse model expressing neuronal cell-specific nestin-Cre (Fig. 12g, h). AIF3KI mice with nestin-Cre+ had a smaller body weight but survived for at least 7 months (Fig. 12i). Although enlarged lateral ventricles and reduced neuron density were detected in the cortex layer and hippocampus of AIF3 KI mice (AIF3 KI/nestin-Cre+), no substantial neuron loss was observed from the overview of AIF3KI mice at P83, distinct to what we had observed from AIF3 splicing mice (Fig. 12j, k, l), indicating that gain-of-AIF3 alone is not sufficient to cause severe neurodegeneration in vivo. Together, these data suggest that the synergistic effects of lossof-AIF and gain-of-AIF3 trigger robust neurodegeneration in AIF3 splicing mice.

Discussion
Here we identified a novel AIF3 isoform, which lacks the hydrophobic domain encoded by exons 2 and 3 of the Aifm1 gene. AIF3 is a nature splicing isoform in C57BL/ 6 mouse brain containing 14 exons and is distinct from five other known AIF isoforms identified by RACE using human kidney cDNA library [2,3]. Although AIF3 is not detectable under physiological conditions, our studies reveal that AIF3 is a disease-inducible splicing isoform produced from the endogenous Aifm1 gene. AIF3 expression is induced in the cortex of human postmortem brain with hypoxic-ischemic or hemorrhage stroke and mouse brain following ischemic stroke. However, AIF3 expression cannot be induced by hypoxia or OGD in cortical neuron cultures. Glial cells in vivo not only provide structural support but also regulate the release of gliotransmitters, neuroinflammation and metabolism [42]. Therefore, it might be due to the fact that in vitro neuron cultures lack neuron-glia interactions and cannot mimic the hypoxic-ischemic environment in brain. Interestingly, we showed that AIF3 splicing is robustly induced in mouse brain when exon 3 of the Aifm1 gene is deleted, suggesting that exon 3 of the Aifm1 gene is critical for AIF3 splicing regulation. Mouse with Aifm1 exon 3 deletion appears to be a valuable model to understand biological functions of AIF3 splicing in brain.
Another key finding of this study is that AIF3 splicing switch in mouse brain causes neurodegeneration. AIF3 splicing mice display an early onset of aging phenotypes supported by our observations of neuron loss, enlarged ventricles, and alerted animal behavior at young ages. All AIF3 splicing mice eventually die between 2 to 4 months after birth. Given that AIF3 is induced in patients following hypoxic-ischemic brain injury, AIF3 splicing in these patients may trigger neurodegeneration. The Stroke-Dementia Association showed that about 10% of stroke patients develop post-stroke-neurodegeneration and signs of dementia within 1 year and stroke has been recognized as the second most common cause of dementia [43,44]. In future studies, a tamoxifen inducible CamkIIα-CreERT mouse line might be used to induce comparable levels of AIF3 splicing observed in hypoxic-ischemic brain and directly study its effect on hypoxic-ischemic brain injury, post-stroke-associated neurodegeneration as well as other mitochondrial dysfunction-associated neurodegenerative diseases [45].
Hq and AIF3KI mice are valuable models allowing us to directly dissect the importance of loss-of-AIF and gain-of-AIF3 in neurodegeneration. We found that neither loss-of-AIF nor gain-of-AIF3 alone is sufficient to cause robust neurodegeneration as observed in AIF3 splicing mice at similar ages, although loss-of-AIF in aged Hq mice causing neurodegeneration in the cerebellum has been well appreciated [10,26,41]. The less obvious neurodegeneration effects in forebrain of Hq mouse might be partially due to the fact that Hq mice still have a small portion of AIF (20%) supporting its basic functions in each single cell, whereas AIF3 splicing mice completely switch AIF expression to AIF3 in CamkIIα-iCRE expressing cells. Since AIF knockout and telencephalon-specific AIF knockout both are embryonic lethal, this possibility cannot be excluded. On the other hand, it is also no doubt that gain-of-AIF3 contributes to mild neurodegeneration, which is supported by the observation of enlarged ventricle and mild neuron reduction in the forebrain. Moreover, gain-of-AIF3 in Fig. 11 Loss of AIF in Hq mice did not obviously cause neurodegeneration in the forebrain. a Representative overview images of Hq mouse brain (P100), including somatosensory cortex, piriform cortex, thalamus and hippocampus. Nissl staining images from C57BL/6 J-A w-J /J mice with the same genetic background were used as the control. Scale bar, 200 μm. b Nissl staining of somatosensory cortex and hippocampus from Hq and their control mice at P100. Scale bar, 200 μm. c Quantification of mouse number with either obvious neuron loss or enlarge ventricle cultured Hq cortical neurons leads to neurotoxicity. Taken together, our three animal models provide the direct evidence that the synergistic effects of loss-of-AIF and gain-of-AIF3 contribute to neurodegeneration in AIF3 splicing mice.
Interestingly, AIF3 splicing switch in MEFs prepared from AIF fl/fl /AIF fl/Y mice does not display obvious cytotoxicity, which is different from that in AIF3-overexpressed Hq cortical neurons. These AIF3 splicing MEFs even grow faster than WT MEFs with obvious increase in extracellular acidification, indicating altered cellular metabolism may overcome cytotoxicity in AIF3 splicing MEFs.
Mitochondrial dysfunction has been often linked to neurodegeneration, although the precise mechanism is less clear. AIF plays a pivotal role in mitochondrial bioenergetics that is critical for cell survival [26,46,47]. Decreased Fig. 12 Characterization of AIF3KI mouse model and its role in neurodegeneration in the forebrain. a Strategy for insertion of the LoxP-eGFP/STOP-LoxP-AIF3 (LSL-AIF3) cassette into the mouse Rosa26 locus. sgRosa26-1 and Cas9 introduce a double-strand break between 1 kb and 4 kb fragments used as homology arms in the targeting vector. Homology-directed repair (HDR) leads to the insertion of the cassette. Cre recombination induces C-terminal Flag-tagged AIF3 expression. LA, left arm. RA, right arm. b PCR analysis of CamKIIα-iCre recombination at the Rosa26 locus in 7-day-WT and AIF3KI male and female mice (PCR ID: iCre, Rosa, AIF3, Table 2). c Expression of AIF3 in the cortex of male WT (AIF3KI fl/+ , CamkII-iCre-) and AIF3KI (AIF3KI fl/+ , CamkII-iCre+) at P105 determined by E1-AIF antibody. d Nissl staining overview of WT (AIF3KI fl/+ , CamkII-iCre-) and AIF3KI (AIF3KI fl/+ , CamkII-iCre+) male mice at P105. e-f Nissl staining of cortex and hippocampus from WT (AIF3KI fl/+ , CamkII-iCre-) and AIF3KI (AIF3KI fl/+ , CamkII-iCre+) male mice at P105. Box indicates the zoom in images shown in f. Scale bar, 200 μm. n = 3. g PCR analysis of nestin-Cre recombination at the Rosa26 locus in 7day-WT and AIF3KI male and female mice (PCR ID: nestin-Cre, Rosa, AIF3, Table 2). h Expression of AIF3 in the cortex of male WT (AIF3KI fl/+ , nestin-Cre-), and AIF3KI (AIF3KI fl/fl , nestin-Cre + and AIF3KI fl/+ , nestin-Cre+) at P83 determined by E1-AIF antibody. i Measurement of the body weight of AIF3KI male mice (AIF3KI fl/fl , nestin-Cre+, n = 17) and their littermate control mice (AIF3KI fl/fl , nestin-Cre-, n = 12). Data are shown as mean ± S.E.M. and analyzed for statistical significance by two-way ANOVA. *** p < 0.001. j Nissl staining overview of WT (AIF3KI fl/fl , nestin-Cre-) and AIF3KI (AIF3KI fl/fl , nestin-Cre+) male mice at P83. k-l Nissl staining of cortex and hippocampus from WT (AIF3KI fl/fl , nestin-Cre-) and AIF3KI (AIF3KI fl/fl , nestin-Cre+) male mice at P83. Box indicates the zoom in images shown in l. Scale bar, 200 μm mitochondrial oxidative phosphorylation is a common feature observed in different AIF-deficient mouse models [26,46,47], supporting that mitochondrial defects in AIF3 splicing mice is at least partially caused by the loss-of-AIF, as AIF3 lacks NADH oxidase activity and FAD-binding affinity. However, loss-of-AIF alone is not sufficient to fully phenocopy mitochondrial dysfunction observed in AIF3 splicing mice as the effects of loss-of-AIF on mitochondrial structural alteration in various animal models are quite different from that in AIF3 splicing mice. For example, loss-of-AIF in telencephalon alters its mitochondrial cristae but does not disturb the overall mitochondrial morphology [24] and deletion of AIF in liver and muscle fails to obviously alter mitochondria size, distribution or ultrastructural structure [47]. AIF3 splicing mice showed the disorganization of overall mitochondrial structures with the larger diameter and vesiculation of their inner membranes. In line with this, overexpression of AIF3 in Hq cortical neurons reduces ATP production, supporting the notion that AIF3 gains additional detrimental functions to promote mitochondrial dysfunction.
AIF has been recognized as a key cell death mediator of parthanatos (caspase-independent cell death) when AIF is released from mitochondria and translocated to the nucleus under pathological conditions, especially ischemic brain injury [6,[27][28][29]. It eventually causes chromatin condensation and cell death [6,[27][28][29]. AIF3 lacks the hydrophobic domain and is expected to be easily released from mitochondria and translocated to nucleus. Indeed, our studies showed increased AIF3 nuclear translocation, chromatin condensation, and neuronal cell death in a time dependent manner in AIF3 splicing mice. Together, these findings uncover that both mitochondrial dysfunction and AIF3 nuclear translocation resulting from the synergistic effects of loss-of-AIF and gain-of-AIF3 contribute to AIF3 splicing-triggered neurodegeneration.

Conclusions
We identified a novel disease-inducible AIF splicing isoform AIF3 in brain and established a conditional inducible AIF3 splicing mouse model. AIF3 splicing switch in brain plays a detrimental role in neurodegeneration and the underlying mechanism involves mitochondrial dysfunction and AIF3 nuclear translocationtriggered neuronal cell death, which is attributed to the synergistic effects of loss-of-AIF and gain-of-AIF3. The discovery of AIF3 isoform and establishment of AIF3 splicing mouse model may provide a valuable tool to understand the role of AIF3 splicing in aged brain and a potential therapeutic target to prevent or delay the progress of neurodegenerative diseases.