Ketogenesis controls mitochondrial gene expression and rescues mitochondrial bioenergetics after cervical spinal cord injury in rats

A better understanding of the secondary injury mechanisms that occur after traumatic spinal cord injury (SCI) is essential for the development of novel neuroprotective strategies linked to the restoration of metabolic deficits. We and others have shown that Ketogenic diet (KD), a high fat, moderate in proteins and low in carbohydrates is neuroprotective and improves behavioural outcomes in rats with acute SCI. Ketones are alternative fuels for mitochondrial ATP generation, and can modulate signaling pathways via targeting specific receptors. Here, we demonstrate that ad libitum administration of KD for 7 days after SCI rescued mitochondrial respiratory capacity, increased parameters of mitochondrial biogenesis, affected the regulation of mitochondrial-related genes, and activated the NRF2-dependent antioxidant pathway. This study demonstrates that KD improves post-SCI metabolism by rescuing mitochondrial function and supports the potential of KD for treatment of acute SCI in humans.

KD on energy metabolism following SCI we studied mitochondrial respiratory function and oxidative phosphorylation (OXPHOS) in the injured cervical spinal cord segments (Fig. 2a). For this, we used a substrate-uncoupler-inhibitor-titration protocol (see "Methods") based on a previous report by Sullivan et al. 10 . As expected, our respirometry results showed an overall decline in mitochondrial respiration (oxygen consumption) at 7 DPI. Following SCI, mitochondrial respiration measured in the presence of malate, pyruvate, glutamate, succinate and the uncoupler CCCP showed a significant decrease in Complex I (CI u ) and Complex II (CII u ) respiration (Fig. 2b). LEAK respiration measured after oligomycin addition to block ATP-synthase activity was significantly decreased after injury in both treated groups, CD and KD. The decrease in LEAK after injury was not prevented with the KD treatment (Fig. 2b). The ADP-stimulated OXPHOS for complex I showed greater variability and the apparent differences observed between the uninjured group and the injured groups (CD nor KD) did not reach significance. However, in the KD group, mitochondrial respiration with convergent electron supply through Complex I and Complex II were significantly higher compared to the CD group (48.12% increase, p = 0.028 for CI; 76.07% increase, p = 0.025 for CII). Mitochondrial respiration via Complex IV (CIV) also showed a significant increase in the KD treated group after the complex was fully inhibited with sodium azide in order to control for oxygen consumption through autoxidation (or spontaneous oxidation) of tetramethylphenylenediamine (TMPD) (35.82% increase, p = 0.0469) (Fig. 2b). Although respiratory rates and control ratios were somewhat low in the permeabilized spinal cord tissue compared to other tissues, previous literature reveals variability in the rates between different models, tissues and cell types. The finding that respiration with addition of the uncoupler CCCP showed only modest increases above coupled respiration indicates that tissues were sufficiently permeabilized.
To confirm the increase in cytochrome c oxidase (Complex IV) respiration, we decided to perform a colorimetric enzymatic assay in tissue homogenates from a second cohort of animals with the exact same injury and treatment conditions. Using this enzymatic approach we also saw significant changes in cytochrome c oxidase activity between treatment groups (Fig. 2c). The decrease in cytochrome c oxidase activity after injury (52.17% decrease, p = 0.087) was reversed to almost uninjured activity levels in the animals treated with KD (Fig. 2c). Similar results were observed when cytochrome c oxidase activity was measured in isolated mitochondria (See "Methods" section) from the epicenter of the injury (Supplementary Fig. S4).
Additionally, in order to identify changes in protein expression levels of the ETS complexes after injury and dietary treatment, a western blot analysis using a "total OXPHOS" antibody cocktail was performed in tissue homogenates obtained from the epicenter of the injury (Fig. 2d,e). These blots revealed an overall significant decrease in protein content at 7 DPI. Interestingly, the cords from the KD treated group showed a significant rescue in expression of complexes I, II and III compared to CD (68.8% increase, p < 0.001 for CI; 16.36% increase, p = 0.003 for CII; and 60.12% increase, p < 0.001) (Fig. 2d,e). No significant differences were observed in the protein content of complexes IV and V between the CD or KD groups, although there was a trend towards increased complex IV expression with KD (p = 0.0748) (Fig. 2d, www.nature.com/scientificreports/ Changes in mitochondrial markers after ketogenic diet treatment. We next asked whether KD rescues mitochondrial function via activation of the mitochondrial biogenesis PGC1α-SIRT3-UCP2 axis. An increase in this axis by KD was recently shown in hippocampal neurons in vitro 28 . We analyzed the expression of voltage-dependent anion selective channel protein 1 (VDAC1/Porin) and peroxisome proliferator-activated receptor γ-coactivator-1α (PGC-1α) by western blot and RT-qPCR respectively (Fig. 3). Our western blot analysis showed a significant decrease in VDAC1 content in the injured spinal cords in both CD and KD groups one week post-injury (Fig. 3a,b). Interestingly, treatment with KD significantly increased VDAC1 expression when compared to the CD treated group (20.71% increase, p = 0.0217). Furthermore PGC-1α mRNA levels showed a trend towards an increase in the KD animals compared to the CD treated animals at 7DPI (p = 0.101). This trend that was no longer seen at 2-weeks post-injury, suggesting that the possible effect of KD may be transient or time-dependent (Fig. 3c). Interestingly, our 2 weeks post-injury data also show a significant upregulation in both injured groups (Fig. 3c). In addition, using a histological approach, we wanted to evaluate if changes in expression of mitochondrial markers were taking place in neurons located in the vicinities of the injury site rostral and caudal to the injury (Fig. 3d,e). Interestingly, we observed a significant increase in the mitochondrial import receptor subunit TOM20 (TOM20) immunofluorescence mean intensity in NeuN + cells both rostral and caudal to the injury site (ipsilateral) when compared to the uninjured spinal cords (25% increase rostral, p = 0.0185 (CD); ~ 32% increase caudal, p = 0.048 (CD) p = 0.032 (KD)) . No difference between CD and KD treated groups was found. The contralateral side (non-injured) did not show differences between any of the three groups (Fig. 3d,e).
Transcriptional and translational regulation by ketogenic diet after SCI. We further analyzed whether the treatment with KD could regulate transcription factors and proteins associated with the general  www.nature.com/scientificreports/ control of the transcriptional and translational machineries in the injured spinal cord (Fig. 4). Using spinal cord homogenates, we first examined p44/42 MAPK (ERK1/2) and STAT3. Both kinases appear to play central roles in mitochondrial regulation and it has been suggested that they are translocated to the mitochondria and regulate the transcription of mtDNA 39,40 . Western blot analysis showed a significant decrease in phospho-p44/42 MAPK (ERK1/2) (Thr202/Tyr204) after SCI (56.6% decrease, p = 0.001) (Fig. 4a,b). The phosphorylation analysis of ERK1/2 revealed a significant increase in phosho-ERK1/2 in the KD group compared to the CD group (118.52% increase, p < 0.0001) (Fig. 4a,b). A significant increase in total STAT3 expression was observed in the  www.nature.com/scientificreports/ KD group compared to the CD group (93.22% increase, p = 0.026). However, no significant changes in STAT3 phosphorylation levels after SCI nor between treatment groups was detected (Fig. 4a,b). Next, we examined the phosphorylation levels of mTORC and two of its well-known downstream effectors: 4E-BP11 and p70-S6K. Amongst many other roles, these kinases also regulate the nutrient and growth factor protein biosynthesis in response to nutrient stimuli 41 . Furthermore, ketone bodies have been shown to induce mTORC inhibition in the hippocampus of rats fed with KD 42 . While a significant decrease in total mTORC content was observed after injury and this decrease was rescued by KD, we saw no differences in mTORC phosphorylation levels between KD and CD-treated groups (Fig. 4c,d). Despite no changes in mTORC activation after injury, we did see changes in its downstream effectors. After SCI, the phosphorylation of the translation initiation factor 4E-BP1 was remarkably increased in both CD and KD groups (253% increase, p < 0.0001). However, no differences were observed between the treatments (Fig. 4c,d). In contrast, the serine/threonine kinase p70-S6K showed a considerable decrease in phosphorylation on Thr389 epitope after injury (82.91% decrease, p = 0.0102) (Fig. 4c,d), and its phosphorylation was significantly increased after KD treatment when compared to the injured group treated with the CD (167.38% increase, p < 0.005) (Fig. 4c,d).
We further studied the expression of the mitochondrial transcription factor A (TFAM). TFAM is a nuclear DNA-encoded protein that binds to the mtDNA and promotes transcription 43 . Our data show that in tissue homogenates TFAM was significantly downregulated after injury (48.14% decrease, p = 0.037); the treatment with KD upregulated its expression to nearly uninjured levels (78.92% increase, p = 0.04) (Fig. 4a, b).
Modulation of intrinsic mitochondrial components by ketogenic diet after SCI. In order to study changes in the mitochondrial intrinsic signaling pathways after KD treatment, mitochondrial enriched Values represent relative expression levels of PGC1-α in uninjured rats (n = 5) compared to the injured spinal cord of KD rats (n = 6) compared with CD rats (n = 6). Two-way ANOVA, Tukey's Post-hoc test. All data are mean ± SEM. (d,e) Example high magnification micrograph of the ventral horn of an uninjured spinal cord stained with TOM20 (mitochondria), NF200 (Neurofilament) and NeuN (D). Intensity quantifications of the mitochondrial import receptor subunit (TOM20) from randomly outlined NeuN + cells (n = 9-12) (e). Oneway ANOVA, Fisher's LSD Post-hoc test. All data are mean ± SEM. Two different mitochondrial markers both targeting outer mitochondrial membrane proteins were used due to its differential effectiveness in detecting the target in western blots or in immunofluorescence. The blots from the same gels were cropped and rearranged in the following order: Uninjured, Control Diet, Ketogenic Diet; to make them more clear and enhance reader' understanding of the figure and the subsequent quantifications. Membranes were cut previous incubation with primary antibodies in order to optimize limited sample use. Full-length blots/gels are presented in Supplementary  Fig. 5 www.nature.com/scientificreports/ fractions from spinal cords were isolated by gradient centrifugation. Samples were analyzed by western blot and normalized to total VDAC1 content ( Supplementary Fig. S2a,b). Phosphorylation of mitochondrial ERK1/2 was significantly increased after SCI (86% increase, p < 0.01) (Supplementary Fig. S2c,d). Although we did not see significant differences between treatment groups, the KD treated group showed a slight trend towards a decrease in phosphorylation compared to the CD group ( Supplementary  Fig. S2c,d). STAT3 showed an overall increase in Ser-727 phosphorylation after injury. A significant increase in STAT3 activity was seen after KD treatment when compared to the injured group and the CD treated group (205.1% increase, p < 0.001 and 57.18% increase, p = 0.0294 respectively) ( Supplementary Fig. S2c,d). In isolated mitochondria, mTORC phosphorylation was not different between groups, consistent with the data obtained from the tissue homogenate samples ( Supplementary Fig. S2c,d). In contrast, when we examined total protein content, total mTORC protein was remarkably higher after injury in the CD group compared to the uninjured group (162.2% increase, p < 0.01), while its expression levels after KD treatment were reduced to levels that were comparable to the ones found in the uninjured group ( Supplementary Fig. S2c,d). The phosphorylation levels of p70-S6K followed a similar pattern of phosphorylation to that seen with ERK1/2: a significant increase in phospho-p70-S6K was observed after injury (88.5% increase, p = 0.01) and a trend towards a reduction in phosphorylation was seen in the KD treated group (p = 0.07) ( Supplementary Fig. S2c,d). Similar to the mTORC data from isolated mitochondria, these results differ from those observed in the tissue homogenate samples. Furthermore, our western blot analysis showed a significant increase in TFAM expression in the mitochondria that were isolated from the cords treated with KD.
A mitochondrial population of NRF2 has been described that forms a complex containing a KEAP1 dimer and the mitochondrial histidine phosphatase PGAM5 44,45 . Recent data also suggest that NRF2 might protect mitochondria from oxidative damage through direct interaction with the mitochondrial outer membrane 46 . www.nature.com/scientificreports/ Interestingly, our results showed a significant reduction in the levels of phosphorylated NRF2 in isolated mitochondria after injury (69.7%, p = 0.0131); after KD treatment we observed a trend towards a partial rescue (p = 0.13) ( Supplementary Fig. S2c,d). These results point out the importance of studying the different protein pools and populations from different organelles and compartments separately. We further analyzed the effect of KD in modulating the expression of the ETS complexes in isolated mitochondria. Similar to the tissue homogenates, we saw an overall decrease in expression of Complexes I to IV after SCI ( Supplementary Fig. S2e,f). However, contrary to the tissue homogenate results, no statistical differences were found in the expression levels between CD and KD treated samples. Even so, trends towards an increase in expression in the KD group can be seen for CI and CII ( Supplementary Fig. S2e,f).
Ketogenic diet reduces oxidative damage and induces activation of the NRF2-dependent antioxidant pathway after SCI. ROS formation after trauma contributes to the pathogenesis associated with SCI by affecting the mitochondrial respiratory system itself, and by inducing lipid peroxidation and protein nitration. Lipid peroxidation and protein nitration has long been known as a main contributor to secondary damage after SCI and it has been shown to increase after injury [47][48][49] . Our histological data showed an increase in lipid peroxidation and protein nitration using 4HNE and 3NT markers respectively at 7 days post injury ( Fig. 5a-f). Interestingly, when we compared the immunofluorescence mean intensity between treatment groups, we observed a significant reduction of both markers in the KD group compared to the CD group in the center of lesion (ROI) (17.2%, p = 0.046 (4HNE); 12.3%, p = 0.0047 (3NT)) area with most of the infiltrating immune cells (Fig. 5a-d,f). However, this reduction was no longer significant when the intensity of the entire contour of the ipsilateral side of the cord was measured, which included a large area with low signal (Fig. 5a,e).
ROS can also trigger activation of specific intracellular antioxidant pathways as a protective response to injury 50 . In order to assess the activation of intracellular antioxidant mechanisms after KD treatment, we used Western blots to quantify levels of phosphorylated (active) NRF2, an upstream transcription factor that regulates the expression of several antioxidant genes (Fig. 5g,h) in tissue homogenates. Our results showed that while the level of active NRF2 decreased after SCI (p = 0.051), those levels were fully rescued by KD (Fig. 5g,h).

β-hydroxybutyrate (βHB) increases NRF2 activation after H 2 O 2 -induced injury and downregulates p21 in non-H 2 O 2 -induced injury conditions in cortical neurons. β-hydroxybutyrate (BHB) is
the most abundant ketone body found in the blood as a result of exogenous KD ingestion 51 . We used in vitro cultures of primary cortical neurons to evaluate whether β-hydroxybutyrate (BHB) alone would be sufficient in promoting an increase in NRF2 activity in response to cellular ROS production. Hence, H 2 O 2 (10 μM) was added to the culture media to induce oxidative stress and initiate an antioxidative response. The cultures were incubated with BHB (4 mM) alone or in combination with H 2 O 2 for 24 h ( Supplementary Fig. S3a). Our immunofluorescence intensity analysis showed that addition of H 2 O 2 increases NRF2 activity-as shown by an increase in Serine 40 (S40) phosphorylation (162%, p < 0.0001) ( Supplementary Fig. S3b,d,g). Interestingly, when compared to untreated cultures exposed to H 2 O 2 alone, the H 2 O 2 exposed cultures treated with BHB showed an even higher increase in S40 phosphorylation (216.9%, p < 0.0001) ( Supplementary Fig. S3b,e,g). The treatment with BHB alone had no effect on pNRF2 levels ( Supplementary Fig. S3c,g). Moreover, to further assess the effect of KD on regulation of oxidative stress, we evaluated p21 intensity levels. p21 is a cyclin dependent kinase inhibitor (CDK) that regulates multiple cellular processes such gene transcription, cell differentiation and apoptosis. p21 has been also associated with a new protective role against oxidative stress in cells 52,53 . Our results showed that p21 expression in cortical neurons was significantly downregulated after BHB treatment (25%, p < 0.001) ( Supplementary  Fig. S3c,f). Cortical cultures treated with H 2 O 2 showed a dramatic decrease in p21 immunofluorescent intensity (83%, p < 0.0001); this decrease was not prevented with addition of BHB (4 mM) (Supplementary Fig. S3d-f).
Ketogenic diet alters Citrate synthase activity and decreases Aconitase activity after SCI. The activities of the mitochondrial enzymes aconitase (AC) and citrate synthase (CS) were evaluated after SCI as previous work linked AC and CS activity to the regulation of the cellular redox state in response to ROS production [54][55][56][57] . Furthermore, it has already been established that ROS production increases after SCI 10,33 . Therefore, we analyzed the enzymatic activities of CS and AC in spinal cord homogenates after injury and treatment with KD (Fig. 5i,j). Our activity assay showed a significant decrease in CS activity after injury in the animals that were fed CD (45.37%, p < 0.01). There was a trend (p = 0.11) towards an increase in CS activity with KD treatment. (Fig. 5i). Using a different cohort of animals, we measured CS activity of isolated mitochondria from the injury site. Contrary to what we observed in tissue homogenates, CS activity in isolated mitochondria showed a significant decrease in activity in the KD treated group, but no differences were found between the CD group and the uninjured nor the KD group ( Supplementary Fig. S4). After SCI, AC activity increased significantly in the CD fed group (68.18%, p < 0.01) (Fig. 5j) and KD treatment reduced AC to nearly uninjured levels (Fig. 5j).
Effects of ketogenic diet on expression levels of mRNAs encoding proteins involved in mitochondrial metabolic and intracellular signaling pathways. RNA from the injured spinal cords was analyzed from rats fed either a KD or a CD for a week. The microarray analysis indicated significant changes (p < 0.05) in gene expression for a total of 1018 genes (530 increased; 488 decreased). The data for all genes detected as certain signals were deposited in the NCBI Gene Expression Omnibus (GEO, http:// www. ncbi. nlm. nih. gov/ geo/) under accession number GSE159555. Amongst those genes, in Table 1 we show a subset of significantly increased or decreased genes that includes genes associated with mitochondria, transcription and translation, metabolic pathways and oxidative stress responses. We found increased levels of mitochondrialspecific mRNAs that either encode for different subunits of complexes of the ETS (Ndufa12, Nda6, Atp5I) www.nature.com/scientificreports/ encode for proteins associated with the assembly or synthesis of its components (Coq3 (p = 0.055) and Cox11). Moreover, the microarray showed increased and decreased expression of mitochondrial ribosomal proteins such Mrps14 and Mrpl22 (up) and Ptcd3 (down); increase levels of Pdk4 that encodes for the pyruvate kinase responsible for inhibiting pyruvate dehydrogenase, regulation of glucose-fatty acid homeostasis, and an increase in the RNA encoding the outer mitochondrial membrane anti-apoptotic factor Bcl2.
Other interesting changes in gene expression were nuclear-coded genes linked to fatty acid and lipid metabolism such Mttp, Fads6 and Pank1, or genes associated with glucose homeostasis such Gcgr and Glp2r (both mRNA levels decreased). Furthermore, increasing mRNAs that encoded for transcription (Polr2f, Tceal1) or translation Values are shown as mean ± SEM. (g,h) Spinal cord homogenates were subjected to SDS-PAGE followed by western blot analysis using phospho-NRF2 antibody. Quantification is shown in (h). (i,j) Citrate synthase (CS) and Aconitase (AC) activities were measured in spinal cord tissue homogenates from uninjured (n = 3), control diet, (n = 4) and ketogenic diet (n = 4) treated animals. The epicenter of the injury was used for the analysis in the animals with SCI. The blots from the same gels were cropped and rearranged in the following order: Uninjured, Control Diet, Ketogenic Diet; to make them more clear and enhance reader' understanding of the figure and the subsequent quantifications. Membranes were cut previous incubation with primary antibodies in order to optimize limited sample use. Full-length blots/gels are presented in Supplementary Fig. 5 www.nature.com/scientificreports/ (Eif4e) factors were found. In addition, genes that contribute to the cell defense mechanisms against oxidative stress such Gsta5 (up) and Nrf1, (down) were also altered by KD.

Discussion
The beneficial effects of KD in ameliorating the metabolic deficits associated with some neurological disorders are widely recognized and accepted 58,59 . Ketone bodies can interact with specific receptors and trigger the activation of cell signaling pathways that have been shown to play important roles in models of stroke, Alzheimer's, Parkinson's disease and Multiple Sclerosis 60 .
Our respirometry data show that KD rescues complex I and II driven O 2 flux in the injured spinal cord, increases the total protein content of complexes I, II and III in tissue homogenates, and increases cytochrome c oxidase (or complex IV) activity at 7 days post injury. In fact, KD and low-carbohydrate diets have been shown to upregulate expression and activity of components of the mitochondrial ETS complexes 3 , which together with our microarray data and other results presented here suggest that modulation of transcription and the subsequent protein upregulation are linked to a rescue in respiratory function after KD treatment.
Ketone bodies preserve the function of mitochondrial complexes after Traumatic Brain Injury (TBI) 33 , in Purkinje cells in aged rats 61 , in a cellular model of mitochondrial Encephalomyopathy 62 , and after a moderate T9-T10 spinal cord contusion in mice 38 . Moreover, TBI decreases mitochondrial content due to an imbalance between mitochondrial biogenesis and mitophagy that in turn leads to deficiencies in energy metabolism 63 . Our results show a trend towards an increase in the mRNA levels for the mitochondrial biogenesis regulator PGC-1α, and an increase in protein content for the mitochondrial biogenesis marker VDAC1 after SCI in the animals treated with KD. Surprisingly, at the 2 weeks after injury time-point, the data shows a major increase in PCG-1α in both treated groups, suggesting that the injury itself might lead to a compensatory mechanism in which a delayed increase in mitochondria biogenesis might compensate for an initial energetic deficiency. The implications of this compensation, and if this increase in mRNA is truly translated into protein stills to be determined. In regards to a proposed increase in mitochondrial biogenesis by KD, KD increases mitochondrial mass in other neurological disorders such as epilepsy 64 and ischemic stroke 65 ; and recent publications highlight how mitochondrial biogenesis could be a potential pharmaceutical target to treat SCI 66,67 . VDAC1 has also been linked to an increase in the transcription of multiple ETS complexes by interacting with TOM20, a component of the receptor complex responsible for the recognition and translocation of cytosolically synthesized mitochondrial pre-proteins or enzymes required to promote the activation of the transcriptional machinery of the mitochondria 68 . This suggests that an increase in VDAC1 might concurrently promote the expression of ETS complexes. However, it should be noted that the relationship between respiratory function and markers of mitochondrial biogenesis is not always linked 69,70 . Interestingly, our histological analysis using TOM20, a different mitochondrial marker, showed that the injury itself increases mitochondria density in neurons in the intermediate grey/ventral horn area -similarly to what others have shown 71 . However, there were no differences in the expression between CD and KD in neurons indicating that they are unlikely to contribute to the overall changes in mitochondrial content we observed in tissue homogenates after KD treatment. Although with a generally low mitochondrial content compared to other mammalian cells 72 , inflammatory cells might be linked to changes in mitochondrial content after neurotrauma. However, our observations suggest that they would not be major contributors to the increase in mitochondrial density in our model. Indeed, growing evidences suggest that KD or ketones might reduce inflammation 73 , implying that a reduction in mitochondrial content after our treatment should have been expected. It remains to be determined whether (1) other cells (i.e. inflammatory cells, oligodendrocytes or astrocytes) in the lesion site (and vicinities) or/and increase in tissue sparing would show differences and lead to changes in mitochondrial density; (2) if the improvements in mitochondrial respiration that we observed in the injured cord may be partially associated with a global increase in total amount of mitochondria, or (3) what seems more likely, that activation of several signaling pathways involved in the regulation of nuclear and mitochondrial transcription and translational in multiple cell types, as discussed further below.
Mitochondrial DNA (mtDNA) encodes for 13 subunits of the complexes composing the oxidative phosphorylation system (OXPHOS) or ETS 74 . Yet, mitochondria require nuclear DNA coding the remainder of constituent proteins 75 . Extracellular signal-regulated kinase 1/2 (ERK1/2) is a serine/threonine kinase implicated in signal transduction that is activated in response to growth factors and other intercellular messengers 76 . Interestingly, ERK1/2 can regulate mitochondrial function [77][78][79] . In HeLa cell cultures, activation of ERK1/2 leads to its interaction with VDAC1 in the mitochondrial outer membrane promoting its translocation into the mitochondrial matrix. Once inside the mitochondrial matrix activated ERK1/2 interacts with the mitochondrial specific transcription factor TFAM in order to promote the transcription of the ETS complexes 40 . Our results show that KD treatment induces an increase in ERK1/2 phosphorylation in spinal cord tissue homogenates, and TFAM expression in tissue homogenates and isolated mitochondria. It is conceivable, that after activation, ERK1/2 is translocated into the mitochondria via VDAC1 to bind to TFAM. However, this mechanism was not fully supported in this study as we were unable to detect ERK1/2 in the mitochondrial fraction. Since TFAM is a nuclear encoded gene 80 , one possibility is that an increased availability of TFAM would also increase its translocation into the mitochondria. Hence, an increase in ERK1/2 activation (and translocation) together with an increase in mitochondrial TFAM could enhance the transcription of some of the ETS complexes in order to compensate for the deficits observed after injury.
Similarly to ERK1/2, several studies have shown that STAT3 can also be found inside the mitochondria (mitoSTAT3) where it regulates the activity of the ETS [81][82][83][84][85][86] . For example, ciliary neurotrophic factor (CNTF) enhances respiratory capacity, membrane potential and complex IV activity in sensory neurons via STAT3 signaling 87,88 , and cultured astrocytes lacking STAT3 show decreased mitochondrial function 81 . However, the precise mechanism by which STAT3 locally modulates mitochondrial function is unknown. Interestingly, some findings suggest that STAT3 can directly interact with complexes I and II to regulate their activity 84 . Our results indicate that KD treatment increases total STAT3 content-but not the relative phosphorylation levels at Ser727in tissue homogenates. Interestingly, KD increases the amount of phosphorylated STAT3 (Ser727) in isolated www.nature.com/scientificreports/ mitochondria, which could be due to an increase in total STAT3 in the tissue homogenate driving the observed increase in mitoSTAT3 that would in the end contribute to an improvement in mitochondrial bioenergetics. Clearly, the efficient function of the ETC requires complex layers of regulation (mitochondria-to-nucleus communication and vice versa) to coordinate the expression of the different subunits, including a combination of transcriptional coordination, sub-cytoplasmic localization of translation and the intricate assembly of the ETC enzyme complexes 89,90 . KD was found to inhibit the mTOR pathway in liver and brain 42 . mTORC1 regulates intracellular signaling associated with protein synthesis and acts as a nutrient/energy sensitive regulator through, for example, activation of p70-S6K, a serine/threonine kinase that via S6 ribosomal phosphorylation leads to protein synthesis; and inhibition of 4E-BP11, a member of the family of translation repressors 91,92 . Interestingly, our data indicate that a 7 day KD treatment after injury had no effects on PI3 kinase/Akt-dependent mTOR activation (Ser 2448), but increased mTOR total protein content. Although our findings do not support an increase in activation through KD, we cannot preclude that other epitopes such as the Ser2481 (autophosphorylation) could still be phosphorylated and active 93 . Interestingly, we observed different results between the two downstream effectors: p70-S6K and 4E-BP11. KD increased the phosphorylation of p70-S6K after injury but did not affect 4E-BP11 phosphorylation levels. Several mechanisms such as association/dissociation of mTOR associated proteins or post-translational modifications on mTORC1 have been considered as possible mechanisms to explain such differential effects in regulating its downstream effectors 94,95 . mTOR and p70S6K are also found in mitochondria 96,97 . Similar to our observations in tissue homogenates, we found no differences in mTOR phosphorylation at Ser2448 after injury or between treatment groups, but on the contrary there was an increase in total mTOR and p70-S6K phosphorylation in the mTOR mitochondrial pool after injury, both of which are reduced after KD treatment. This reduction in total mTOR and phospho-p70-S6K in the mitochondrial pool after KD might contribute to a reduction of ROS production as previously described 98 , while the increase in the mTOR cytosolic pool might contribute to an increase in nutrient-related protein translation 99 .
Two possible mechanisms could explain the differences in activities and expression of proteins and kinases observed when comparing our tissue homogenate samples to our samples from isolated mitochondria; the first one is the differential rate of kinase degradation by mitochondrial proteases, thus some kinases might have been degraded faster than others 100 ; and the second is that some of those kinases might have been removed from the mitochondria after the initial stimulation and downstream signaling activation as previously shown 101 . It is increasingly recognized that proteins and kinases can be translocated to or associating with mitochondria, modulating their differential expression in the mitochondrial pools versus the cytosolic pools (i.e. faster degradation, different activation mechanisms, and different activity levels) 102 . The detailed mechanisms on how these kinases function or are interconnected after SCI or KD treatment would require further study.
Secondary damage after SCI is associated with an increase in oxidative stress 10,103,104 . An overproduction of ROS leads to deficits in enzymatic activity, increase in lipid peroxidation, activation of apoptosis and activation of the antioxidant response, which, together contribute to deficits in mitochondrial bioenergetics after trauma 105 . Similarly to a recently published work from Lu et al. 106 , our data indicate that KD treatment decreases lipid peroxidation (4-HNE) and protein nitration (3NT) at the center of the lesion -where most of the infiltrated inflammatory cells are localized, and activates the NRF2-dependent antioxidant pathway after SCI in vivo and in vitro. In response to stimuli, NRF2 detaches from KEAP1 in the cytosol and translocates into the nucleus to bind to the antioxidant response element (ARE) leading to the expression of antioxidant-associated genes 107 . Furthermore, NRF2 increases the mitochondrial membrane potential and ATP levels, the rate of respiration and the efficiency of oxidative phosphorylation 108,109 . Indeed, NRF2 plays a key role in regulating the balance between mitochondrial function and the redox state of the cells by modulating the utilization of substrates for mitochondrial respiration (NADH and FADH2), which also appears to be important for the function of antioxidant enzymes 110 . Our data from isolated mitochondria show that KD treatment also induces NRF2 activation. NRF2 can associate with the mitochondria through a complex with KEAP1 and the mitochondrial outer membrane serine/threonine kinase PGAM5. Interestingly, PGAM5 interacts with the antiapoptotic protein BCL-XL 111 suggesting that KD, through an increase in NRF2 might also be important in reducing apoptosis after SCI. In addition, ROS production changes the activity of TCA cycle enzymes involved in the regulation of cellular metabolism. In skeletal muscle aconitase (AC) undergoes oxidative modification and inactivation after oxidative stress 54,112 . Interestingly the opposite has been reported in the CNS where AC activity is increased at 7 and 28 days after optic nerve transection in rats 113 . Similarly, our results show increased AC activity after SCI at 7 DPI which was reduced by KD treatment. Hence, after SCI KD treatment could decrease AC activity, increase citrate production and thus protect the mitochondria from more oxidative damage leading to recovery of mitochondrial metabolism. These experimental differences also suggest that TCA enzyme function may differ between tissues or after different types of insults.
Finally, our microarray data are in line with previous work from the brain 114 , and as briefly commented at the beginning of the discussion, supports the hypothesis that KD treatment after injury might overall regulate the transcriptional and translational machineries (i.e. Eif4e, Mrps14, Nfyb) that control the expression of proteins encoding subunits of some of the complexes (i,e Ndufa12), or proteins required for their assembly (i.e. Cox11); which would contribute to the bioenergetic rescue after SCI. Additionally, our microarray data also indicate that KD might promote activation of antioxidative stress pathways complementary to the NRF2 pathway, seen by the up-and down-regulation of Gsta5 and Nrf1 respectively; and it implies that KD treatment triggers changes in expression of genes encoding proteins with roles in the glucose-fatty acid metabolism as previously reported 114,115 .
In conclusion, our findings-see Fig. 6 for graphical summary-indicate that KD mitigates mitochondrial oxidative damage and dysfunction in acute SCI. Furthermore, we provide new insights into the possible underlying regulatory mechanisms through which KD ameliorates mitochondrial bioenergetics after spinal cord injury (Fig. 6). Additional studies are necessary in order to evaluate whether these early metabolic changes are long www.nature.com/scientificreports/ lasting or if a long term exposure to ketone bodies, either via KD or ketone esters, might be necessary to sustain these early beneficial effects in cell metabolism seen after SCI. Data analysis. The data analysis was performed following the methodology previously described by Oliveira et al. 117 . The Feature Extraction Software v9.1.3.1 (Agilent Technologies) was used to extract and analyze the assay signals and subsequently determine the signal-to-noise ratios from the microarray images. Microarray raw data (.txt files) were imported into R v. 3.0.1 (Team RDC, 2012) and analyzed with the Bioconductor (Gentleman et al., 2004) packages Agi4 × 44PreProcess and limma (Smyth, 2005). Briefly, after quality check, the microarray probes were filtered and their median foreground intensity was normalized within and between arrays according to Agi4 × 44Preprocess and limma user guides, respectively. Finally, the probes were tested for differential expression using a linear model followed by Bayes moderated t-Student test (Smyth, 2005) for the comparisons of interest. Genes with nominal p < 0.05 were accepted to be differentially expressed. The data was deposited in the NCBI Gene Expression Omnibus (GEO, http:// www. ncbi. nlm. nih. gov/ geo/) under accession number GSE159555. Genes whose absolute fold change was > 1.1 were selected in Table 1.

RNA isolation and quantitative real-time polymerase chain reaction (Q-PCR). A 5 mm segment
of the ipsilateral side of the spinal cord centered on the impact site was removed for extraction of RNA. Total RNA was extracted using a Trizol Reagent (Invitrogen, CA, USA) according to the manufacturers' manual. Quantitative PCR reactions were performed using the TaqMan Fast Universal PCR Master Mix Kit (Applied Biosystems, CA, US) 35 . The mRNA for peroxisome proliferator-activated receptor-gamma coactivator 1α (PGC-1α) was measured by real-time quantitative RTPCR using PE Applied Biosystems prism model 7700. Triplicate reactions were run for each sample for both the gene of interest and the endogenous control (glyceraldehyde-3-phosphate dehydrogenase; GAPDH). Results are presented as 2-dCt, where dCt was difference between cycle threshold (Ct) values of GAPDH and gene of interest, a calculation that compensates for loading errors. Subtrac- Figure 6. Graphical summary of the proposed mechanism of mitochondrial bioenergetics rescue and activation of antioxidant pathways by ketogenic diet after SCI. After SCI, an increase in ROS production would affect several complexes of the Electron Transport System (ETS) leading to a global dysfunction in cellular bioenergetics. Treatment with KD would lead to an increase in ketone bodies that would bind to specific receptors (i.e. HCA2, FFR2 or unknown). The activation of these receptors would trigger phosphorylation of ERK1/2 and the increase in expression of Stat3 and mTOR . mTOR would regulate ribosomal translation and the subsequent synthesis of proteins and growth factors that are nutrient-dependent. Simultaneously, an increase in VDAC1 expression, might contribute to the translocation of phospho-ERK1/2 (and other essential proteins) into the mitochondrial matrix where by binding the mitochondrial transcription factor (TFAM) would lead to an increase mtDNA transcription and a subsequent expression of several components of the ETS contributing to the mitochondrial bioenergetics rescue. Furthermore, the presence of the mitochondrial pools of Stat3 and mTOR/p70-S6K would contribute to the activation of some of the complexes of the ETS and to reduce ROS production inside the mitochondria. KD treatment would also increase activation of the NRF2dependent antioxidant pathway that not only would activate expression of antioxidant and DNA repair genes, but whose mitochondrial-associated pool might also help reducing cellular apoptosis. Moreover, KBs might act as an alternative substrate and promote neuroprotection. Overall, our data indicates that KD treatment regulates mtDNA transcription, DNA transcription and ribosomal translation, which altogether contributes to the improvement of mitochondrial bioenergetics after SCI. Created with BioRender.com. www.nature.com/scientificreports/ tion of dCt of CD rats from dCt of KD rats gives the ddCt value that was used to calculate relative expression levels in KD animals (2-ddCt) as previously described 35 .
In vitro cultures, treatment, immunocytochemistry and immunohistochemistry. Cortical neurons from C57BL/6 P0-P2 mouse pups were dissociated by combined trypsinization. Cells were placed in 24-well tissue culture dishes (Nunc, Roskilde, Denmark) on coated coverslips and grown for 7 DIV in Neurobasal A medium supplemented with N2 and B27. Cultures were treated then for 24 h with H 2 0 2 (10 μM, Sigma-Aldrich), β-Hydroxybutyrate (BHB) (4 mM, Sigma-Aldrich) alone or a combination of both. Cultures were stained with anti-p21, anti-phospho-NRF2 and anti-β-III-Tubulin, fixed with 4% PFA mounted in Fluoromount (Vector Labs, Burlingame, CA, USA). The immunofluorescence intensities of the randomized confocal (TCS SPII, Leica, Bannockburn, IL, USA) images (n = 5) taken from each well were analyzed. A total of 14 rats were perfused transcardially with PBS, followed by PFA 4%. The harvested spinal cords were post-fixed for 12-24 h and cryoprotected in a PBS-30% sucrose solution. The cords were subsequently snap frozen in tissue freezing media for cryostat cryosectioning (20 μm section thickness). A full set of sections of each harvested cord was stained for each animal (200 μm between sections). After thawing the sections, the following primary antibodies were used against: TOM20, 3NT, 4HNE, NeuN, NF200, CD45, Iba-1, and GFAP. These were applied overnight at room temperature. Secondary antibodies (1:500, Jackson) conjugated to DyLight 594, 488 and 647, respectively, were applied for 2 h at room temperature. Digital images were captured with an Imager M2 microscope (Zeiss, Jena, Germany) (Immunocytochemistry and Immunohistochemistry protocols as per used in our previous publication 118 ). For the lipid peroxidation and protein nitration analysis, 2 different sections of the epicenter of the lesion per animal were evaluated. The spline contour of the whole ipsilateral (injured) side of Table 2. List of antibodies, vendors and used dilutions. www.nature.com/scientificreports/ each section and a ROI in the center of the injury were selected for the intensity measurements. NeuN + staining was used to identify neurons in the intermediate gray/ventral horn area, randomized 9-12 NeuN + cells were outlined and assessed for intensity measurements (TOM20) from four different sections per animal rostral and caudal to the injury (intensity values were averaged). The sections selected were the 4 closest sections to the injury where we did not find neuronal loss. Next, 2 different sections where the injury was still present (rostral and caudal, 1000-1200um from epicenter) were used to outline the injury site (GFAP was used as a reference) and the intensities of the different markers (IBA1, CD45, TOM20) were assessed. The images were analyzed using Zen Software (Zeiss). For more detailed information about the antibodies used, see Table 2 at the end of the "Methods" section.
Enzymatic activities. Tissue homogenates or isolated mitochondria lysates from the epicenter of the injury were used to quantify the activities of complex IV (Cytochrome C oxidase) as well as citrate synthase by spectrophotometry using the Cytochrome C oxidase Kit (Abcam, ab239711) and Citrate Synthase Assay Kit (Sigma, CS0720) respectively. Citrate Synthase activity defined by the rate of change in the linear range per mg of protein in tissue homogenates corresponding to the epicenter of the injury. Unit definition: 1 unit would oxidize 1 μmol of reduced cytochrome c per min at 25ºC and pH 7.2.
Mitochondrial respiration. Mitochondrial respiration in permeabilized spinal cord segments was measured using high-resolution respirometry (Oroboros Instruments, Innsbruck, Austria). Experiments were performed at 37 °C in respiration medium containing EGTA (0.5 mM), MgCl 2 (3 mM), K-lactobionate (60 mM), taurine (20 mM), KH 2 PO 4 (10 mM), HEPES (20 mM), sucrose (110 mM), and BSA (1 g/l), with the addition of mitochondrial substrates and inhibitors (Substrates-Uncouplers-and Inhibitors (SUIT) Protocol) to measure coupled and uncoupled respiration, flux through complex I and II, maximal respiration of the ETS, and cytochrome oxidase (complex IV respiration). Mitochondrial respiration was expressed as weight-specific oxygen flux (pmolO 2 s −1 mg −1 ) (See Supplementary Fig. S1 for an example of a tracing). Chamber oxygen levels were maintained between 240 and 400 nmol ml −1 to avoid O 2 limitation. b routine respiration was first measured, then malate (2 mM), pyruvate (5 mM), and glutamate (10 mM) were added to measure LEAK respiration (respiration due to proton leakage and the circuit of electrons and cations that is not dependent on ATP synthase activity). ADP (2.5 mM) was then added to measure NADH-dependent coupled respiration through complex I, followed by Oligomycin (2.5 μM) to measure LEAK respiration at high membrane potential during inhibition of ATP synthase. Complete, non-physiological uncoupled respiration of the ETS was measured with the titration of carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (CCCP) (0.5 + 0.5 μM). After complex I inhibition with rotenone (0.5 μM), succinate (10 mM) was added to measure reduced flavin adenine dinucleotide (FADH 2 )-dependent complex II respiration. Residual oxygen consumption (ROX) attributed to non-mitochondrial respiration was measured with addition of Antimycin A (2.5 μM) to inhibit complex III of the ETS. Finally, ascorbate (2 mM) and N,N,N′,N′-tetramethyl-p-phenylenediamine dihydrochloride (TMPD) (0.5 mM) were added to measure the isolated respiratory rate of cytochrome oxidase (complex IV respiration) (See Supplementary Fig. S1 for the quantification of the oxygen flux after sequential titration of the components in our SUIT Protocol).
Statistical analysis. Statistics were analyzed by using the GraphPad Prism version 6.0.0 for Windows (GraphPad Software, San Diego, California USA, www. graph pad. com). The Shapiro-Wilk test was performed to determine the normality of the data. Wherever appropriate, an unpaired Student's t-test , one-way or two-way ANOVA, followed by Turkey's or Fisher's LSD post hoc comparisons test were used to determine differences between group mean values. Each test used is specified in the figure legends. Data are expressed as means ± SEM. The minimum level of statistical significance was set at P < 0.05. www.nature.com/scientificreports/