Loss of Frataxin activates the iron/sphingolipid/PDK1/Mef2 pathway in mammals

Friedreich’s ataxia (FRDA) is an autosomal recessive neurodegenerative disease caused by mutations in Frataxin (FXN). Loss of FXN causes impaired mitochondrial function and iron homeostasis. An elevated production of reactive oxygen species (ROS) was previously proposed to contribute to the pathogenesis of FRDA. We recently showed that loss of frataxin homolog (fh), a Drosophila homolog of FXN, causes a ROS independent neurodegeneration in flies (Chen et al., 2016). In fh mutants, iron accumulation in the nervous system enhances the synthesis of sphingolipids, which in turn activates 3-phosphoinositide dependent protein kinase-1 (Pdk1) and myocyte enhancer factor-2 (Mef2) to trigger neurodegeneration of adult photoreceptors. Here, we show that loss of Fxn in the nervous system in mice also activates an iron/sphingolipid/PDK1/Mef2 pathway, indicating that the mechanism is evolutionarily conserved. Furthermore, sphingolipid levels and PDK1 activity are also increased in hearts of FRDA patients, suggesting that a similar pathway is affected in FRDA. DOI: http://dx.doi.org/10.7554/eLife.20732.001


Introduction
FRDA is the most prevalent form of recessive cerebellar ataxia characterized by neurodegeneration and cardiomyopathy. Patients with FRDA exhibit a progressive degeneration of the dorsal root ganglia, sensory peripheral nerves, and dentate nuclei of the cerebellum. Most individuals develop several neurological symptoms at a juvenile age, but individuals with adult or late onset are occasionally reported (Koeppen, 2011). Mutations in FXN were discovered as the primary cause of FRDA (Campuzano et al., 1996). Studies of the yeast frataxin homolog identified a role for FXN in iron-sulfur cluster biosynthesis and iron homeostasis (Babcock et al., 1997;Mühlenhoff et al., 2002;Rö tig et al., 1997). Loss of FXN has been shown to cause mitochondrial dysfunction in several model organisms (Anderson et al., 2005;Puccio et al., 2001;Rö tig et al., 1997). Yet, whether iron accumulates in the nervous system is still under debate, as it has been reported that iron levels are not altered in the nervous system of a conditional knockout mouse model and in FRDA patients (Puccio et al., 2001;Simon et al., 2004;Solbach et al., 2014). However, in FRDA patients, iron was shown to accumulate in the dentate nuclei or in glia cells of dorsal root ganglia (Boddaert et al., 2007;Koeppen et al., 2009). In sum, the iron deposition phenotype is still controversial in mouse models and FRDA patients, and the role of iron in the pathophysiology of FRDA has not yet been determined.
It has been proposed that ROS, generated by impaired mitochondrial electron transport chain complexes or iron deposition, mediate the pathogenesis in FRDA (Bayot et al., 2011;Santos et al., 2010). However, emerging evidence suggests that ROS are not elevated in several model organisms, and several clinical trials based on antioxidant therapy have shown no or limited benefit in FRDA patients (Babcock et al., 1997;Llorens et al., 2007;Macevilly and Muller, 1997;Santhera Pharmaceuticals, 2010;Seznec et al., 2005;Shidara and Hollenbeck, 2010). Hence, the pathogenesis of this disease is still poorly understood.
In a recent study, we described the first fly mutant allele of frataxin homolog (fh), the fly homolog of FXN (Chen et al., 2016). We showed that loss of fh in Drosophila leads to iron accumulation in the nervous system, which in turn induces sphingolipid synthesis and ectopically activates Pdk1 and Mef2. Mef2 activation triggers the aberrant transcription of downstream target genes that causes the degeneration of fly photoreceptors (Chen et al., 2016). However, whether this pathway is also affected in vertebrates upon loss of FXN is unknown. Here, we removed the endogenous mouse Fxn gene by using an adeno-associated virus (AAV) and CRISPR/Cas9 strategy (Swiech et al., 2014). We show that the iron/sphingolipid/PDK1/Mef2 pathway is activated in mice upon loss of Fxn. Furthermore, sphingolipids and PDK1 activity are also up-regulated in hearts of FRDA patients, suggesting that this pathway is also affected in FRDA. Hence, down-regulation of this pathway provides possible new targets to interfere with disease progression.

Results
To test if the iron/sphingolipid/PDK1/Mef2 pathway is activated in vertebrates in vivo upon loss of Fxn, we reduced Fxn levels in the mouse nervous system by using the AAV and CRISPR/Cas9 system (Swiech et al., 2014). We designed two different Fxn single guide RNAs (sgRNAs) and tested their efficiency. The Fxn-sgRNAs induce lesions in the Fxn locus and reduce the levels of FXN protein when compared to the LacZ-sgRNA control in Neuro-2a cells (Figure 1-figure supplement 1A and B). To deliver LacZ-or Fxn-sgRNA into the nervous system, we injected AAV that carried sgRNA into the ventricle of Rosa26-Cas9 knock-in newborn mouse brains (Platt et al., 2014), and the efficiency of AAV infection into the cortical neurons was examined. As shown in Figure 1-figure supplement 1C, the majority of cortical neurons are infected with AAV, and the mRNA levels of Fxn in Fxn-sgRNA mice are less than 40% of the LacZ-sgRNA control mice (Figure 1-figure supplement 1D).
At approximately four months of age the Fxn-sgRNA mice (mice with Fxn-sgRNA AAV injection) develop an abnormal appearance when compared to LacZ-sgRNA control mice. These include a smaller body size and a hunchback phenotype at P130 ( Figure 1A). Interestingly, these features are similar to the neuronal conditional knockout mice that were previously reported (Puccio et al., 2001). In addition, the Fxn-sgRNA mice become much less mobile than their LacZ-sgRNA littermates (Videos 1 and 2). The Fxn-sgRNA mice also display an aberrant reflex when lifted by the tail as they fail to spread their hind limbs ( Figure 1B). To evaluate the sensorimotor coordination, we performed a rotarod test at two different ages. At P50, we observe no difference between LacZ-sgRNA and Fxn-sgRNA mice. However, Fxn-sgRNA mice show a significant reduction of latency to fall from the rod at age of P130 ( Figure 1C). Besides the deficit on the rotarod test, Fxn-sgRNA mice also show a severe impairment in the wire hang test at P130 ( Figure 1D). In summary, a removal of Fxn in a substantial neuronal population in the newborn mouse brain causes a severe impairment of neuronal function.
We further characterized the neuronal phenotype by examining brain sections at P130. We observed that the activating transcription factor 3 (ATF3), a marker that is typically associated with cellular stress response and neuronal damage (Tsujino et al., 2000), is up-regulated in cortical neurons of Fxn-sgRNA mice ( Figure 2A). However, no ATF3 positive cells were observed in LacZ-sgRNA mice (Figure 2A). In addition, the axon initial segments, specialized membrane regions where Na + channels cluster to initiate action potentials (Szu-Yu Ho and Rasband, 2011), are lost in cortical neurons of Fxn-sgRNA mice. As shown in Figure 2B, the scaffolding protein required for the formation and maintenance of the axon initial segment, ankyrin-G (AnkG), is severely reduced or lost in cortical neurons in Fxn-sgRNA mice, especially in ATF3 positive neurons ( Figure 2B). These data suggest that loss of Fxn in cortical neurons leads to severe neuronal injury and damage.
To assess if the iron/sphingolipid/PDK1/Mef2 pathway is activated in the brains of Fxn-sgRNA mice, we assessed if iron levels are increased upon loss of Fxn. To examine Fe 3+ levels, we performed Perls' staining and used 3,3'-diaminobenzidine (DAB) to enhance the signal. At P130 the DAB signal is strongly increased in brain sections of Fxn-sgRNA mice when compared to control mice ( Figure 3A). To further characterize the iron distribution, we co-stain iron with neuronal and glia markers in the same brain sections. Since strong DAB signal interferes with the fluorescence, we  chose the cortex region with relatively weak DAB deposits of Fxn-sgRNA mice. Interestingly, the majority of DAB signals are co-localized with a neuronal marker but not glial markers, suggesting that Fe 3+ accumulates in neurons, although there are some weak diffuse signals distributed outside of the cell body ( Figure 3 and Figure 3-figure supplement 1). These data show that Fe 3+ accumulates in the nervous system upon loss of Fxn. The fact that we observed an obvious Fe 3+ accumulation, whereas previous reports did not, is likely due to the use of DAB to enhance the Perls' Prussian blue staining, as we did not observe any signal with Perls' Prussian blue staining alone ( We next assessed Fe 2+ levels by using RPA (Petrat et al., 2002). RPA is a cell-permeable fluorescent dye that selectively accumulates in mitochondria, and its fluorescence is stoichiometrically quenched by Fe 2+ (Petrat et al., 2002). Interestingly, we observed Fe 2+ accumulation in the brain of Fxn-sgRNA mice at P60, a time point at which we did not observe behavioral defects in the Fxn-sgRNA mice. As shown in Figure 3B, the RPA fluorescence is quenched in cells of cortical layer IV and V of Fxn-sgRNA mice when compared to LacZ-sgRNA controls, indicating an accumulation of Fe 2+ in the nervous system upon loss of Fxn. As a negative control, we used RPAC, a dye that is structurally very similar to RPA but is insensitive to Fe 2+ mediated fluorescence quenching. RPAC is present at similar levels in both Fxn-sgRNA and LacZ-sgRNA brains ( Figure 3B), suggesting that the decreased fluorescence of RPA in Fxn-sgRNA mice is not due to impaired dye uptake but due to Fe 2+ quenching.
It was previously proposed that iron deposition triggers free radical production via Fenton's reaction. We therefore measured oxidative stress by examining the levels of 4-hydroxynonenal (4-HNE), a major product of lipid peroxidation (Ayala et al., 2014). However, we did not observe an increase of 4-HNE in Fxn-sgRNA brain tissues (Figure 3-figure supplement 2B), consistent with previous observations in flies (Chen et al., 2016). We next determined if iron triggers the activation of the PDK1/Mef2 pathway in Fxn-sgRNA mice. Phosphorylation of S241 in the PDK1 activation loop is required for its activity (Casamayor et al., 1999). We observed an increase in phosphorylation levels of S241 in Fxn-sgRNA mice compared to the LacZ-sgRNA mice at P60, prior to the onset of symptoms ( Figure 3C and Figure 3-figure supplement 2C). To assess if Mef2 is activated in Fxn-sgRNA mice, we determined the mRNA levels of several validated Mef2 downstream targets (Anderson et al., 2004;Ewen et al., 2011;Potthoff et al., 2007;Woronicz et al., 1995). As shown in Figure 3D, many, but not all, transcripts are up-regulated 1.5-3 fold in Fxn-sgRNA mice at P60. This is consistent with previous findings that only about half of Mef2 targets are up-regulated (to 2-3 fold) when Mef2A is overexpressed (Ewen et al., 2011). In summary, these results provide evidence that the PDK1/Mef2 pathway is activated in the vertebrate nervous system upon loss of Fxn prior to the onset of phenotypes that are observed about two months later.
Last, we tested if the same pathway is also activated in FRDA patients. As we were unable to obtain cerebella and dorsal root ganglia of patients, we used heart samples that were previously shown to accumulate Fe 3+ (Bradley et al., 2000;Koeppen, 2011). We first investigated the sphingolipid levels in frozen heart tissues of five controls and six FRDA patients. As shown in Figure 4A, a substantial fraction of the sphingolipids are increased in FRDA patients when compared to controls, including short chain ceramides (C14-C18), different dihydroceramides, dihydrosphingosines, dihydrosphingosine-1-phosphate, and sphingosine. Finally, we tested if PDK1 activity is also increased in FRDA patients. The phosphorylation levels of S241 in PDK1 are elevated in FRDA patients when compared to controls ( Figure 4B). Note that patient six showed no obvious increase in PDK1 phosphorylation when compared to the other patients. This patient had relatively mild   symptoms and a later onset of disease (age 18) than other patients (ages 4 to 9). The age of death was also much later in patient 6 (age 67) compared to the other patients (age 25-37). Similarly, the increased levels of sphingolipids in patient six are also less pronounced than other patients. In sum, these data suggest that the iron/sphingolipid/PDK1/Mef2 pathway is activated in FRDA patients and that it may contribute to the pathogenesis of FRDA.

Discussion
Here we show that removal of Fxn in the central nervous system using AAV and CRISPR/Cas9 causes behavioral and neurological phenotypes in mice. This new FXN mouse model exhibits several similar phenotypes to the previous neuronal conditional knockout mouse model, including smaller body size, hunchback phenotype, impaired locomotion, and shortened life span (Puccio et al., 2001). By using RPA and Perls'/DAB stainings, more sensitive iron staining assays than Perls' Prussian blue staining alone, we found that Fe 2+ and Fe 3+ levels are increased in the cerebral cortex of Fxn-sgRNA mice, and the activities of PDK1 and Mef2 are up-regulated, suggesting that the iron/sphingolipid/ PDK1/Mef2 pathway, previously identified in fly fh mutants, is also activated in Fxn-sgRNA mice. Importantly, sphingolipids and PDK1 activity are also up-regulated in the hearts of FRDA patients, indicating that a similar pathway is also activated and may contribute to the pathogenic mechanism of FRDA.
Iron toxicity has been long considered as a possible pathogenic contributor of FRDA. However, the iron accumulation phenotypes have never been reported in the nervous system in an inducible conditional knockout mouse model (Martelli and Puccio, 2014;Puccio et al., 2001;Simon et al., 2004). We provide the first direct evidence that both Fe 2+ and Fe 3+ accumulate in the brain upon loss of Fxn. In addition, we noticed that irons are not only deposited inside the neuronal cell body but also present in the extracellular space. For example, DAB staining showed a diffuse pattern in and surrounding the neuronal cell body ( Figure 3A and Figure 3-figure supplement 1), and the RPA fluorescence is similarly reduced in the extracellular space ( Figure 3B), suggesting that iron accumulation occurs both intracellularly and extracellularly.
It is worth noting that the iron/sphingolipid/PDK1/Mef2 pathway is conserved from yeast to fly (Chen et al., 2016;Lee et al., 2012), and that this pathway is also activated in Fxn-sgRNA mice and the hearts of FRDA patients. Interestingly, genetic variations at the 5'UTR of Mef2C that promote Mef2C transcription are associated with hypertrophic cardiomyopathy (Alonso-Montes et al., 2012). In addition, overexpression of Mef2A or Mef2C causes cell sarcomeric disorganization and elongation in cultured cardiomyocytes (Xu et al., 2006). These data suggest that an activation of iron/ sphingolipid/PDK1/Mef2 pathway may play a role in the induction of cardiomyopathy in FRDA patients. Inhibition of this pathway may not only suppress neurodegeneration, but also prevent or delay cardiomyopathy, the major cause of lethality in FRDA. In sum, this pathway may provide us with therapeutic targets based on antisense oligonucleotide strategies (Ackermann et al., 2012), an option that should be explored.

AAV production and purification
AAV vectors were produced at the Gene Vector Core, Baylor College of Medicine. Dulbecco's modified Eagle's medium (DMEM) with high glucose (4.5 g/l) and antibiotic-antimycotic were purchased from GenDEPOT (Katy, TX), and fetal bovine serum (FBS) was purchased from Sigma (St. Louis, MO). HEK293T cells were grown in DMEM supplemented with 10% FBS and 1x antibioticantimycotic. DJ/8 Rep-Cap plasmid (Grimm et al., 2008) was purchased from Cell Biolabs (San Diego, CA). AAVs were packaged by three plasmids transfection (transfer vector, DJ/8, and helper plasmid) using iMFectin Poly DNA Transfection Reagent (GenDEPOT). In brief, 80-90% confluent HEK293T cells in a 15 cm dish were split into 6Â15 cm dish one day before transfection. Next day, the media was removed and replaced with 9 ml of fresh DMEM/5% FBS. 24 ml of iMFectin reagent were diluted to 0.5 ml by DMEM and then added to 0.5 ml of DMEM containing 4 mg of helper plasmid, 2 mg of DJ/8 and 2 mg of transfer vector. The transfection mixture was vortexed for 10 s and incubated for 15 min at room temperature before overlay on HEK293T cells without removing media. After 4 hr, 10 ml of DMEM/5% FBS were added and cells were incubated in humidified CO 2 incubator. AAV purification was performed by the method developed by Ayuso et al (Ayuso et al., 2010) with modifications. Three days after transfection, cells were collected by centrifugation at 2500 x g for 10 min. Cell culture supernatant was retained for subsequent precipitation. The cell pellet was re-suspended in 1 ml of 50 mM Tris pH8.0 containing 5 mM MgCl 2 and 0.15 M NaCl per plate, lysed by adding 0.1 vol of 5% sodium deoxycholate at room temperature for 30 min and then incubated with 10 mg/ml of DNase I and RNase A for 1 hr at 37˚C. Cell lysates were clarified by centrifugation at 5000 x g for 10 min at 4˚C. Cell culture supernatant was incubated with 10 mg/ml of DNase I and RNase A for 1 hr at 37˚C and then incubated at 4˚C overnight after addition of a stock solution of 40% polyethylene glycol (PEG) 8000 containing 2.5 M NaCl to final concentration of 8% PEG. AAV in cell culture supernatant was collected by centrifugation at 2500 x g for 30 min. The pellets containing AAV were re-suspended in a minimal volume of HBS (50 mM HEPES, 0.15 M NaCl, 1% Sarcosyl, 20 mM EDTA, pH8.0). Cell associated and secreted AAVs were combined for the subsequent iodixanol density centrifugation (Zolotukhin et al., 1999). AAVs were dialyzed against phosphate buffered saline without Mg 2+ and Ca 2+ and the titer was determined by real time PCR using vector specific primers (Ljungberg et al., 2012).

AAV injection into the neonatal mouse brain
Rosa26-Cas9 knockin mice (Gt(ROSA)26Sor tm1.1(CAG-cas9,-EGFP)Fez h /J, Stock No. 024858, the Jackson Laboratory, gift from Huda Zoghbi) were used for AAV injection. P1 neonates were cryo-anesthesized at 0˚C for 3 min before injection. AAVDJ-8 carrying Fxn-sgRNA or control LacZ-sgRNA (2 Â 10 14 gc/ml) was injected into the ventricles using a pulled glass needle. After injection, pups were laid on a warm pad until recovery. The brains were dissected at P60 and P130 after virus injection. All mouse work and procedures were approved by the Animal Care and Use Committee at Baylor College of Medicine and were performed in accordance with the guidelines of the U.S. National Institutes of Health (animal protocol: an-4634).

Behavior test
LacZ-and Fxn-sgRNA mice were subjected to accelerating rotarod and wire hang tests as previously described (Zhang et al., 2013).

Iron staining
To detect Fe 2+ levels of AAV injected mouse brain, the brains were dissected and embedded into 2.5% low-melting agarose (Sigma-Aldrich). The 100 mm brain sections were acquired by a vibratome (1000 Classic; Warner Instruments) and transferred into 24 well plate with ice-cold artificial cerebrospinal fluid (ACSF) (125 mM NaCl, 2.5 mM KCl, 1 mM MgCl 2 6H 2 O, 1.25 mM NaH 2 PO 4 , 2 mM CaCl 2 .2H 2 O, 25 mM NaHCO 3 , 25 mM Glucose, pH 7.4). The brain sections were then incubated with 10 mM RPA or RPAC (Squarix Biotechnology) in ACSF for 20 min at room temperature in the dark. The samples were washed subsequently three times by ACSF. To detect Fe 3+ levels, brains were dissected and fixed in 4% paraformaldehyde for 1 hr on ice followed by immersion in 20% sucrose overnight at 4˚C. Brains were then sectioned using a freezing stage Microtome (Microm KS 34, Thermo Scientific) and spread out on glass coverslips. Brain slices were then incubated with Perls' solution (1% K 4 Fe(CN) 6 and 1% HCl in 0.4% Triton-PBS) for 10 min. After three quick washes in 0.4% Triton-PBS, samples were incubated for 5 min with DAB solution (10 mg DAB with 0.07% H 2 O 2 in 0.4% Triton-PBS) to enhance the signal. The brains were then quickly washed three times with 0.4% Triton-PBS and mounted. The images were obtained with an Axio-imager Z1 microscope (Carl Zeiss) fitted with an AxioCam digital camera (Carl Zeiss). The images were analyzed by ZEN 2012 (Carl Zeiss).

Mass spectrometry
Sphingolipid profiling of human heart tissues was performed by the Lipidomics Core at the Medical University of South Carolina as previously described (Chen et al., 2016).

Control and FRDA patient heart specimens
Frozen heart specimens of FRDA patients were obtained from Dr. Arnulf Koeppen, Veterans Affairs (VA) Medical Center, Albany, NY, USA. Our receipt and analyses of these postmortem tissues was conducted with the approval of the Institutional Review Board at Baylor College of Medicine (protocol H-32191). Table 1 shows clinical and genetic data of six patients with FRDA. Five frozen normal control heart specimens were obtained from National Disease Research Interchange, Philadelphia, PA, USA.