Erythropoietin promotes hippocampal mitochondrial function and enhances cognition in mice

Erythropoietin (EPO) improves neuronal mitochondrial function and cognition in adults after brain injury and in those afflicted by psychiatric disorders. However, the influence of EPO on mitochondria and cognition during development remains unexplored. We previously observed that EPO stimulates hippocampal-specific neuronal maturation and synaptogenesis early in postnatal development in mice. Here we show that EPO promotes mitochondrial respiration in developing postnatal hippocampus by increasing mitochondrial content and enhancing cellular respiratory potential. Ultrastructurally, mitochondria profiles and total vesicle content were greater in presynaptic axon terminals, suggesting that EPO enhances oxidative metabolism and synaptic transmission capabilities. Behavioural tests of hippocampus-dependent memory at early adulthood, showed that EPO improves spatial and short-term memory. Collectively, we identify a role for EPO in the murine postnatal hippocampus by promoting mitochondrial function throughout early postnatal development, which corresponds to enhanced cognition by early adulthood. Robert Jacobs, Mostafa Aboouf, et al. examined the effect of erythropoietin (EPO) in hippocampal mitochondrial function and memory in two mouse models: one overexpressing EPO in the brain, and juvenile mice treated during three days with a high dose of intraperitoneal EPO. Their results suggest that erythropoietin in the neonatal brain may impact spatial memory by increasing mitochondrial content.

E rythropoietin (EPO) continues to develop as a promising neuroprotective agent with widespread clinical relevance 1,2 since the discovery that EPO and its receptor (EPOR) are expressed in the central nervous system (CNS) 3 . EPO can mediate neuroprotection following traumatic injury, hypoxia-ischemic insults, excitotoxicity, inflammatory brain injury 4,5 , and retinal protection following hypoxia 6 or vascular injury 7,8 . EPO prevents hippocampal neurodegeneration and attendant cognitive impairment in rodents presenting with diabetes 9 or Alzheimerlike diseases 10,11 as well as reduces long-term spatial-memory deficits across various animal models of neonatal brain injury 12,13 . In humans, EPO has been shown to enhance memory retrieval by increasing hippocampal plasticity 4,14-17 and improving neurogenesis 10,14,[18][19][20] in both healthy individuals and those affected by various neurodegenerative diseases and dementia. The neuroprotective potential of EPO also extends to neonatology, since it has been reported to reduce perinatal brain injury 21,22 and improve neurological outcomes in very prematurely born neonates 21,23,24 .
The importance of mitochondria in neuronal development and function is highlighted by their role in promoting neuronal differentiation, neurotransmission, and synaptic pruning 29 . The close relationship mitochondria share with neurodevelopment and cognition is also evidenced by the influence of mitochondrial morphology and function on synaptic vesicle pool regulation, spatial memory, and working memory throughout neuronal maturation 30,31 . EPO has been reported to support neuronal mitochondria and prevent memory impairment in animal models of multiple sclerosis, sleep deprivation, neurodegeneration, and brain injury [32][33][34][35][36] . Furthermore, EPO prevents neurobehavioral deficits in the hippocampus of young rats exposed to intermittent hypoxia 37 as well as a murine model of sleep apnea 38 by managing cellular oxidative stress. However, questions relating to the influence of EPO regulating brain mitochondria during development remain to be answered.
We have recently reported that EPO signals on CA1 pyramidal cells in the hippocampus and improves hippocampal postnatal neuronal maturation by reducing cell death and promoting synaptogenesis 39 . Putative mitochondrial mechanism(s) in promoting these processes are unexplored. Additionally, whether EPO improves hippocampal-mediated cognitive function in healthy animals remains unknown.
Accordingly, the aim of this research was to analyze the influence of EPO on brain mitochondria throughout postnatal development and examine whether postnatal EPO availability influences cognition in mice. This work demonstrates that EPO in the CNS activates Erk1/2 and AKT pathways in the hippocampus. Constitutive cerebral EPO overexpression and high-dose intraperitoneal (i.p.) administration of EPO during three consecutive days, coincides with increased mitochondria number, size, and respiration. EPO overexpression increases mitochondria number and function at postnatal ages (P) 14 to 21. The postnatal rise in mitochondria was also identified with electron microscopy (EM) in presynaptic axon terminals, along with higher vesicle number, which suggests that EPO promotes neuronal mitochondrial function and enhances the putative reserve pool of the synaptic vesicles. Transcriptional signals controlling mitochondrial biogenesis and dynamics were altered with EPO-associated increases in peroxisome proliferator-activated receptor gamma coactivator 1-alpha (PGC1α), mitofusin 1 (MFN1), mitofusin 2 (MFN2), and DRP1-DNM1. Thus, EPO-mediated activation of Erk1/2 and AKT pathways throughout postnatal development corresponds with attendant improvements in mitochondrial function and hippocampal-specific cognition.
There was also an effect of age on mass-specific oxygen consumption rates (OCR), with an increase observed as age advanced in WT (1-way ANOVA, F(4,35) = 25.2, p < 0.0001; Fig. 2e) and Tg21 (1-way ANOVA, F(4,35) = 24.0, p < 0.0001; Fig. 2f) animals. In general, age-associated increases in hippocampal OCR occurred earlier and more abruptly in the postnatal development of Tg21 mice between P7 and P14 ( Fig. 2f) when compared to their WT counterparts that demonstrated a more gradual increase from P3 to P21 (Fig. 2e). Specific intra-genotypic differences in respiratory states across ages are shown in Fig. 2e, f for WT and Tg21 mice, respectively. Collectively, these data suggest that overexpression of cerebral EPO facilitates greater hippocampal respiratory capacity earlier in postnatal development.
EPO overexpression in the CNS facilitates changes in respiratory control by increasing mitochondrial content in the hippocampus during postnatal development. Cytochrome c oxidase (COX; complex IV) activity, validated as a surrogate of mitochondrial content in human skeletal muscle 40 , differed between genotypes throughout postnatal development with higher measures observed at P14, P21, and P60 in Tg21 animals compared to WT (2-way ANOVA, F(1,66) = 36.77, p < 0.0001; Fig. 3a). Like respiratory control (Fig. 2), an age-dependent increase in COX activity was observed in both genotypes with the greatest increases in COX activity occurring between P7 and P14 in Tg21 and between P14 and P21 in WT mice (2-way ANOVA, F(4,66) = 45.79, p < 0.0001; Fig. 3a). Mitochondria-specific respiratory analysis (mass-specific respiration normalized to a surrogate of mitochondrial content, e.g., COX activity) helps distinguish qualitative alterations in respiratory control from quantitative differences driven primarily by mitochondrial content. When normalizing massspecific OCR to COX, respiratory differences between genotypes were lost ( Fig. 3b-f). These data suggest that mass-specific OCR differences between genotypes are most likely attributable to quantitative disparities in mitochondrial content as opposed to qualitative variations in mitochondrial function.
High-dose intraperitoneal (i.p.) administration of EPO increases hippocampal Erk1/2 and AKT activity as well as mitochondrial respiration and content. To verify that hippocampal increases in pErk1/2, pAKT, respiratory rates, and mitochondrial content are not due to a non-specific effect of our transgenic (Tg21) model, WT mice received (50 μl) i.p. treatments of either saline or rhEPO (5 IU/g) for three consecutive days at the age of weaning (P23-25, Fig. 5a). This dose of EPO is high enough to facilitate transportation across the blood brain barrier 41 . At P26, twenty-four hours after the last i.p. injection, hippocampal Erk1/2 and AKT activation, respiratory control, and indices of mitochondrial content were assessed (Fig. 5c-l and Supplementary Fig. 5). Systemic administration of EPO resulted in the canonical increase in hematocrit from 38% to 50.4% (Student's t-test, p < 0.0001; Fig. 5b) that was not observed in saline-treated animals. While total AKT protein was higher in EPO-treated mice (Student's t-test, p = 0.049; Fig. 5d), there was no difference in Erk1/2 expression between EPO and saline-treated groups (Student's t-test, p = 0.26; Fig. 5e). pAKT and pErk1/2 were higher in EPO-treated mice (Student's t-test; p = 0.03, Fig. 5h and p = 0.01, Fig. 5i, respectively). EPO-treated mice expressed more mitochondrial protein including VDAC1 expression (Student's t-test, p = 0.004; Fig. 5f) and COX activity (Student's t-test, p = 0.0006; Fig. 5j) as well as mass-specific P CI , P CI+CII , and E CI+CII OCR (2-way ANOVA, F(1,56) = 40.27, p < 0.0001; Fig. 5k). When normalizing mass-specific respiration to COX, the differences between treatments were lost (2-way ANOVA, F(1,56) = 0.77, p = 0.38; Fig. 5l), further confirming that EPO increases mitochondrial content in the hippocampus. Collectively, these results are consistent with the effects of cerebral overexpression of EPO in the Tg21 line.
Mitochondrial profiles in the hippocampus, along with quantitative analysis of presynaptic mitochondria content and vesicle number, were examined by electron microscopy (EM; Fig. 6g-k). The upregulation of hippocampal mitochondria as suggested by COX activity (Fig. 3a), mtDNA/nDNA (Fig. 4a), mitochondrial protein (Figs. 4 and 5), and PGC1α mRNA ( Fig. 6a) was confirmed via EM as evidenced by a higher number of mitochondria structures (Fig. 6g, arrows and Fig. 6h; Student's t-test, p = 0.036) as well as greater size (Student's t-test, p = 0.002; Fig. 6i). Importantly, total mitochondrial mass and vesicle number were both higher in presynaptic axon terminals of Tg21 mice irrespective of the synapse type (symmetric or asymmetric) (Fig. 6g insets, Student's t-test, p = 0.0043, Fig. 6j; and Student's t-test, p < 0001, Fig. 6k). These latter results suggest Fig. 2 EPO overexpression in the CNS influences mass-specific respiratory control in the postnatal hippocampus. a Representative respirometric traces from a WT (top panel) and a Tg21 (bottom panel) hippocampal tissue, which illustrates the change in oxygen concentration (nmol·ml −1 , left y axis, blue line) and oxygen flux per mass (pmol O 2 /s * mg ww, right y axis, red line) in hippocampal tissue at P21. Respiratory states were achieved through the titration of various substrates, inhibitors, as well as a protonophore. The order of respiratory state analysis from beginning to end (left-to-right) with the respective substrates, inhibitors, or protonophores added, as fully explained in the methods, consisted of: leak without adenylates (L N ; addition of malate and octanoylcarnitine); coupled respiration with maximal electron input specific to mitochondrial complex I (P CI ; addition of ADP, pyruvate, and glutamate); maximal rates of coupled respiration with electron input specific to complex I and II (P CI+CII ; addition of succinate); maximal non-coupled respiration (E) with electron input from complex I and II (E CI+CII ; steps of carbonyl cyanide p-trifluoromethoxy phenylhydrazone, FCCP, addition until respiration ceases to increase); non-coupled respiration with maximal electron input specific to mitochondrial complex II (E CII ; addition of rotenone); and non-mitochondrial residual oxygen consumption (ROX; addition of antimycin). Following respiratory state analysis, ascorbate and N,N,N′,N′-tetramethyl-1,4-benzenediamine, dihydrochloride (TMPD) were simultaneously added to assess cytochrome c oxidase (COX; complex IV) activity. b Mass-specific P CI respiration in WT and Tg21 mice, showing higher oxygen consumption rates (OCR) at P14 and P21 in Tg21 mice; 2-way ANOVA, F(1.66) = 15.74, **p = 0.002. c Mass-specific P CI +CII respiration in WT and Tg21 mice, showing higher OCR at P21 in Tg21 mice; 2-way ANOVA, F(1,66) = 9.21, **p = 0.003. d Mass-specific E CI+CII in WT and Tg21 mice, showing higher OCR at P14 and P21 in Tg21 mice; 2-way ANOVA, F(1,66) = 22.22, ****p < 0.0001. e Mass-specific respiration for WT mice at postnatal ages of P: 3, 7, 14, 21, and 60. An age-dependent effect is observed in P CI , P CI+CII , and E CI+CII ; 1-way ANOVA, F(4,35) = 25.2, ****p <0.0001. f Mass-specific respiration for Tg21 mice at postnatal ages of P: 3, 7, 14, 21, and 60. An age-dependent effect is observed in P CI , P CI+CII , and E CI+CII ; 1-way ANOVA, F(4,35) = 24.0, ****p < 0.0001, with multiple comparisons: **p = 0.002, ***p < 0.001, ****p < 0.0001. Barplots with SD bars. a possible increase in synaptic transmission influencing hippocampal function in Tg21 mice.
EPO overexpression in CNS enhances hippocampal-dependent spatial navigation and short-term memory Morris water maze (MWM) test. Young adult mice from both genotypes successfully learned to find the hidden platform during the acquisition phase, as demonstrated by the escape latency and swimming path length (Fig. 7a, upper panels). A shorter escape latency was observed in Tg21 mice beginning on day 2 (d2) (2-   Novel object recognition (NOR). The cognitive impact of EPO on recognition memory was assessed via novel object recognition (NOR) tasks (Fig. 7b). Both Tg21 and WT mice were habituated to learn the location and become familiar with two equal objects over a single 5 min learning event. Subsequently, short-term and long-term memory was examined when testing the recognition of a novel object after 1 and 24 h inter-trial intervals (ITI), respectively (Fig. 7b, upper panel). During the habituation phase, Tg21 animals explored the empty arena more intensively than WT mice resulting in more total locomotor activity (4.96 ± 6.8 vs. 3.97 ± 6 m/min, respectively, Student's t-test, p < 0.001). During the training phase, total object exploration time was similar between genotypes. In the short-term memory test, both WT and Tg21 mice spent more time exploring the novel object. However, Tg21 mice exhibited a stronger preference for the novel object reflected by an object discrimination index (equation (1)) above 0.5. Longterm memory was then tested 24 h later when the mice were given 5 min in the arena with a novel displaced object. In this task, both genotypes equally recognized the novel object with a similar object discrimination ratio (equation (1)) of 0.28 ± 0.18 for WT and 0.25 + 0.2 for Tg21 (2-way ANOVA, F(1,68) = 6.18, p = 0.01; Fig. 7b, lower panel). Although this test is not exclusively hippocampal, as visual cues are involved in task analysis, it is a valid test to assess cognition in young adult mice with the speed in learning locations highlighting cognitive flexibility.
T-maze spatial working memory. Impact of EPO on spatial working memory showed spontaneous alternation to be more than 75% correct in both genotypes (Fig. 7c). Nevertheless, Tg21 mice were faster in learning the choice (10.13 ± 0.74 s for WT and 7.37 ± 0.70 s for Tg21; Student's t-test, p = 0.0004), and had less errors than WT in the second and third test of each. When the ITI was 24 h, however, no difference between genotypes was observed (t4-6; Student's t-test, p = 0.4; Fig. 7c, bottom panel) Thus, only an increase in short-term spatial working memory was observed in Tg21 mice.
Collective, cognitive testing demonstrates that EPO overexpression improves performance speed, hippocampal-mediated spatial learning and memory, and short-term memory while longterm memory remained unaltered.

Discussion
The aim of this research was to analyze the influence of EPO on brain mitochondria throughout early postnatal development and examine whether EPO availability improves cognition in mice. Our primary findings reveal that EPO promotes postnatal increases in hippocampal mitochondria, respiratory potential, and enhances cognition. More specifically, elevated cerebral EPO during early postnatal development increases hippocampal pErk1/2 and pAKT while hastening local mitochondrial maturation by approximately one week (~33% increase), which functionally results in greater oxidative potential. Expedited mitochondrial development in the hippocampus is also evident in presynaptic terminals along with greater vesicle numbers, both of which coincide with improved spatial and short-term memory at early adulthood. Collectively, these findings add to the clinical non-hematopoietic relevance of EPO on stimulating neural development and cognition (see Fig. 8 for results summary).
EPO/EPOR signaling is dependent upon the concentration of EPO and expression of EPOR. Expression of EPOR transcripts increases throughout postnatal development in the hippocampus, mainly in CA1 pyramidal neurons, with the greatest expression achieved at weaning (P21) 39 . This observation was confirmed in the current study, as EPOR mRNA increased throughout postnatal development up to P21. Alternatively, postnatal EPOR mRNA expression in cortical areas was mostly negligible. This implies a role of EPO in the postnatal maturation of CA1 principal neurons during the time of development when these cells undergo dendritic arborization, axon growth, spine formation, synaptogenesis 42 , and when GABAergic synapses form within the postsynaptic populations of CA1 neurons 43 . A transient hippocampal hyperexcitability in the second postnatal week has been reported as a result of axonal remodeling 44 . Hyperexcitability causes brief episodes of local hypoxia in CA1 pyramidal neurons 45 . Therefore, it is conceivable that these transient episodes of local hypoxia trigger EPO and EPOR production at this age, enhancing synaptogenesis, circuit formation, and, consequently, cognition. Evidence of a local hypoxic influence on the expression of both EPO and EPOR in pyramidal cells, as well as its impact in spine density, has been previously demonstrated 46 . EPOR expression in developing embryonic mouse brain is similar to adult hematopoietic tissue 47 , as similarly observed in humans 48 . However, high embryonic EPOR expression in the brain decreases~20-fold over 10 days leading up to birth 49 . Here we show in adult mice that transcriptional EPOR expression in the brain is lower than the spleen by approximately eight orders of magnitude. Since EPOR expression closely relates to cellular EPO sensitivity, the relatively high concentrations of EPO in the brain achieved in this study by means of transgene or high-dose EPO injections are suggested to be essential for promoting mitochondrial development in the hippocampus of mice.
Intracellular EPO/EPOR signaling acts through multiple pathways including Erk1/2, AKT, and Janus kinase/signal transducers and activators of transcription (JAK/STAT). Analysis in this study focused on AKT and Erk1/2 pathways, as they are more  5 Intraperitoneal (i.p.) EPO injections in WT mice resulted in greater pAKT, pErk1/2, higher rates of respiration, and more mitochondrial content in the hippocampus. a Schematic diagram of EPO (red bars) or saline (white bars) treatment in mice. Male animals were injected for three consecutive days after weaning (P23 to P25) with either recombinant human (rh)EPO (5 IU/g) or saline solution. At P26 hippocampal tissue was taken for respirometry and protein analysis. b Changes in hematocrit (%) after treatment; unpaired t-test, ****p < 0.0001. c Representative images of total AKT, total Erk1/2, and voltage-dependent anion channel 1 (VDAC1) protein expression with vinculin as a loading control. d Quantification of AKT normalized to vinculin and saline; unpaired t-test, *p = 0.049. e Quantification of Erk1/2 protein expression normalized to vinculin and saline; unpaired t-test, p = 0.26. f Quantification of VDAC1 protein expression normalized to vinculin and saline; unpaired t-test, **p < 0.004. g Representative images of total phosphorylated pAKT and pErk1/2 protein expression with vinculin as a loading control. h Quantification of pAKT normalized to vinculin and saline; unpaired t-test, *p = 0.03. i Quantification of pErk1/2 normalized to vinculin and saline; unpaired t-test, *p = 0.012. j Cytochrome C oxidase activity (COX, pmol O 2 /s * mg ww) in the hippocampus of WT mice randomly treated with either saline or EPO (5 IU/g). EPO-treated mice show higher COX activity; unpaired t-test ***p = 0.0006. k Hippocampal mass-specific oxygen consumption rates (OCR) from WT control (saline) and EPO (5 IU/ml) injected mice. P CI respiration, P CI+CII respiration, and E CI+CII respiration is increased in EPO-treated mice; 2-way ANOVA, F(1,56) = 40.27, ****p < 0.0001, with multiple comparisons: *p < 0.05, ***p < 0.001, and ****p < 0.0001. l Mitochondria-specific OCR (mass-specific OCR normalized to COX activity, a surrogate for mitochondrial content). No differences in mitochondria-specific respiration are observed between saline and EPO-treated mice; 2-way ANOVA, F(1,56) = 0.77, p = 0.38. Barplots with SD bars.
Here we show that constitutive overexpression of cerebral EPO, as well as high-dose i.p EPO injections shown to traverse the blood-brain barrier 41 , increased pErk1/2 and pAKT levels in the hippocampus with negligible changes in total respective protein expression. During the early postnatal period, ERK participates in the maturation process of dendritic trees and synaptogenesis in Relative mRNA expression levels of the aforementioned genes are quantified by qPCR and normalized to ACTB. g Representative electron microscope (EM) images of hippocampal CA1 area from WT and Tg21 mice at P14. More mitochondrial profiles (yellow arrows) and a higher number of vesicles in presynaptic axon terminals are observed in Tg21 mice. Scale bar: 1 μm (right panels) and 100 nm (panel insets). h Quantification of mitochondria number in the total field of view; unpaired t-test, *p = 0.036. i Mitochondria size; unpaired t-test, p = 0.002. j Mitochondria total mass in presynaptic terminals; unpaired t-test, **p = 0.0043. k Vesicle numbers in presynaptic terminals; unpaired t-test, ****p < 0001. Barplots with SD bars.  50 . During neural differentiation, Erk1/2 signaling is associated with increased transcriptional expression of PGC-1α 51 . However, Erk1/2 generally promotes mitochondrial fission via DRP1 phosphorylation of S616 and suppressing MFN1 via phosphorylation of T562 27,28 under pro-apoptotic conditions 52 . Alternatively, AKT promotes a hyperfused mitochondrial network by suppressing DRP1 with reciprocal phosphorylation of DRP1 at S637 and reducing phosphorylation of S616 via proto-oncogene serine/threonine-protein kinase (Pim-1) 53 . EPO-activated AKT increases mitochondrial biogenesis through eNOS by which an increase in nitric oxide (NO) upregulates PGC-1α, as observed in cardiomyocytes 25,26 . The mechanistic/mammalian target of rapamycin complex 1 (mTORC1) has also been shown to upregulate mitochondria and increase respiration via inhibition of the eukaryotic translation initiation factor 4E (eIF4E)-binding proteins (4EBPs) 54 , and both Erk1/2 and AKT signaling promote the activation of mTORC1. Thus, EPO-EPOR signaling through Erk1/2 and AKT pathways , ****p < 0.0001 than WT controls during reversal. Multiple comparison: *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. b Representative diagram of the Novel Object Recognition (NOR) test, as fully explained in the methods. Object recognition relating to short-term memory function using 1 h inter-trial-intervals (ITI) and long-term memory function using 24 h ITI is shown with object discrimination index. Tg21 animals demonstrated longer total exploration time and an improved discrimination index when testing short-term memory. Animals from both genotypes appeared to recognize the replaced object equally when testing long-term memory; 2-way ANOVA, F(1,68) = 6.18, **p = 0.01. c T-maze spatial working-memory test shows spontaneous alternation above 75% in WT and Tg21 mice. Tg21 mice make less errors than WT at the second and third test from each day reflecting a better short-term but no improvement in long-term memory as observed in t4 after 24 h; 2-way ANOVA, F(1.132) = 14.17, ***p = 0.0003. Multiple comparisons, ***p < 0.001, and ****p < 0.0001. Barplots with SD bars. could increase hippocampal mitochondria and respiratory potential.
High-resolution respirometry analyses demonstrated higher OCR for 3 different respiratory states (P CI , P CI+CII , and E CI+CII ) during postnatal development (P14 and/or P21) in Tg21 mice as well as P26 WT mice treated with EPO when compared to their respective control groups. Additionally, COX activity, ratio of mtDNA to nDNA, an assortment of mitochondrial proteins, as well as mitochondrial number and size assessed via electron microscopy were all greater with EPO. The higher respiratory potential with EPO appears to result from quantitative mitochondrial increases/expansion and not qualitative alterations in respiratory control, as differences in mass-specific OCR were lost when normalizing respiration to a surrogate of mitochondrial content (e.g., COX activity).
Transcriptional promotion of mitochondrial biogenesis was also higher in Tg21 animals as evidenced by more hippocampal PGC-1α mRNA at P21. Astrocyte differentiation and synaptogenesis have been shown to be influenced by PGC-1α in the postnatal prefrontal cortex 55 . We have also previously shown EPO-facilitated improvements in synaptogenesis during postnatal hippocampal development 39 . Thus, EPO-directed PGC-1α upregulation may influence synaptogenesis in the hippocampus. Transcriptional control of mitochondrial dynamics in the hippocampus was also altered with EPO, as pro-fusion (MNF1 and MNF2) and pro-fission (DRP1-DNM1) signals were higher at P21. Recent evidence identified distinct DRP1mediated signaling profiles that result in mitochondrial degradation or biogenesis in which the origin of fission differs 56 . Midzone mitochondrial fission promotes mitochondrial proliferation whereas peripheral mitochondrial fission sequesters dysfunctional components of the organelle for mitophagic control. Although we are unable to verify the specific fission profiles or origins of fission in the current study, peripheral fission is preceded by evidence of mitochondrial dysfunction (i.e., oxidative stress, drop in membrane potential, and/or loss of proton motive force) as well as lysosomal-mitochondrial contact 56 . Respirometric analyses in the current study suggest no difference in ROS production (L N state respiration) as well as greater membrane potential and proton motive force in Tg21 and EPO-treated WT mice (P CI , P CI+CII , and E CI+CII respiration). Also, we found no indication of the difference in lysosomal activation as evidenced by the transcriptional consistency of ATG5 between genotypes throughout postnatal development. Given these findings, along with the observations that mitochondrial number and size were greater in Tg21 vs WT controls, EPO appears to increase mitochondrial biogenesis in the hippocampus early in life.
Mitochondrial plasticity is a key factor in metabolic adaptation, and alterations in shape and content can reflect changes in respiratory potential. Greater mitochondrial content was observed specifically in presynaptic terminals in combination with increased vesicle numbers at P14 in Tg21 mice. Although total vesicle number in synaptic terminals were counted and reported, not just reflecting those in contact with the active zone, synaptic vesicles are recruited to the active zone to discharge neurotransmitters during intense stimulation. This process is dependent on mitochondrially derived ATP 57 . Therefore, higher OCR and mitochondrial content in presynaptic neurons suggest a higher energetic potential to support greater vesicle release. A greater number of glutamatergic and GABAergic synapses and increased inhibitory postsynaptic current amplitude reported at P14 in Tg21 mice 39 also support the concept that increased respiratory potential is coupled with more synaptogenesis during postnatal development, and these EPO-facilitated differences may enhance cognition.
Mitochondrial dysfunction leads to severe defects in hippocampal-dependent cognitive function, such as memory and learning 58 , associated with several neurodevelopmental disorders and degenerative diseases 59,60 . EPO has been shown to be an effective drug to restore mitochondrial dysfunction and cognition after brain traumatic injury 61 and in neurodegeneration 11,[62][63][64][65] . EPO treatment improves neuronal mitochondrial activity and upregulates brain hemoglobin expression in mouse models of multiple sclerosis (MS) 35 . It is therefore likely that EPO overexpression causes an increase in neural hemoglobin during development stimulating neuronal mitochondrial function. Our paired observations of concentrated EPOR expression on CA1 cells 39 and higher rates of cellular respiration in the hippocampus of Tg21 mice led us to speculate that EPO overexpression may also improve hippocampal-mediated cognition.
It is well documented that EPO administration or overexpression enhances spatial memory across several models of disease 30,35,36,66 . Hippocampal CA1 principal cells have a welldefined role as being cognition-relevant but hypoxia-vulnerable 67 . Mitochondrial function, especially as it pertains to the hippocampus, has also been shown to influence spatial memory throughout life 68,69 . Spatial memory and short-term memory were tested using the MWM, NOR, and T-maze spatial workingmemory tests. The MWM test showed significantly shorter escape latency during acquisition and reversal phases with reduced swim path length. These measures were all influence by the genotypic difference in task performance speed. We previously reported improved exercise performance in Tg21 mice with a longer time to exhaustion that was independent of systemic hematological differences (e.g., hematocrit) 70 . Perhaps the faster swim speed observed in the current study could be partially explained by the diminished influence of central fatigue on physical function in Tg21 animals, as metabolic control in the brain plays a critical role in limiting maximal exercise capacity 71 . While higher mitochondrial content was observed in Tg21 mice earlier in life (P14-21), maximal coupled respiration, P CI+CII , the respiratory state that best reflects in vivo oxidative phosphorylation potential, did not differ between WT controls and Tg21 mice at P60. Moreover, the shorter swim path length of Tg21 mice to find the platform, suggests improved spatial orientation most likely attributed to enhanced plasticity as shown previously 13,16,72 . The NOR test with spatial cues examined the influence of EPO on short-term and long-term memory. Throughout the training phase, Tg21 mice showed increased activity, but total exploration time was like WT controls. Ultimately, our findings suggest that EPO overexpression may improve short-term memory, but long-term memory remains uninfluenced. The role of the hippocampus, cerebral cortex, and their integrated communication in developing short-and long-term memory is not fully understood. However, the neural mechanisms used to establish short-and long-term memory are known to be somewhat independent 73 . While we cannot currently provide a clear description explaining the disparate influence of EPO on short-and long-term memory, our observations suggest a role for EPO to enhance short-term memory. Finally, T-maze spatial working memory showed no differences in total spontaneous alternation between genotypes, but Tg21 mice were faster in the choice and made fewer errors in the second and third test from each day, further reflecting improved learning and short-term memory.
In this study, we provide insights on the mechanisms by which EPO may enhance cognition in early life. We show that EPO signaling through Erk1/2 and AKT corresponds with improvements in respiratory potential and mitochondrial content, such as the greater number and size of mitochondria identified in presynaptic terminals of the hippocampus, which are observable by the second postnatal week of development. This enhanced energetic potential could support greater synaptogenesis and provide a cognitive advantage later in life. Indeed, constitutive expression of cerebral EPO also improved spatial and short-term memory in early adulthood. These findings highlight a therapeutical potential of EPO to stimulate hippocampal metabolism during maturation, which may improve neurodevelopment and cognition in, for example, premature children afflicted with hippocampal dysfunction following neonatal hypoxic-ischemic brain injury.

Methods
Animals. Mitochondrial analyses were conducted with both sexes using WT controls and transgenic Tg21 74,75 , mice at postnatal (P) ages: 3, 7, 14, 21, and 60 (n = 8). Behavioral tests were performed in early adulthood (P45-60). Tg21 hemizygous mice were backcrossed with C57Bl/6 mice for more than ten generations to obtain the corresponding control mice and Tg21 mice were bred to homozygosity.
All mice were bred and housed under standard conditions at the University of Zurich. Animals were kept on a 12/12-h light/dark cycle with food and water provided ad libitum. Animal experiments were performed following the ARRIVE guidelines and were approved by the Cantonal Veterinary Office of Zurich, Switzerland (ZH177_16).

EPO intraperitoneal injections.
Male mice received i.p. injections with 5 IU/g rhEPO (NeoRecormon, Roche) or placebo (saline solution) for three consecutive days. EPO/saline treatment was initiated in P23 animals. Animals were sacrificed at P26, 24 h after the last treatment followed by respirometry or tissue collection for later protein analyses.
Blood samples for measures of hematocrit, hemoglobin, and plasma EPO. Mice were anesthetized with a subcutaneous injection containing 100 mg/kg ketamine (Ketasol-100; Dr. Graub, Switzerland), 20 mg/kg xylazine (Rompun; Bayer, Germany), and 3 mg/kg acepromazine (Sedalin, Switzerland). Blood samples were taken by cardiac puncture with a 27 G needle attached to a 1 ml heparinized syringe. The blood was immediately transferred into an Eppendorf tube (1 ml) to determine hematocrit, hemoglobin, and plasma EPO concentrations.
Duplicate measures of hematocrit were collected using microcapillary tubes (Micro haematocrit tube 100, Assistant) and immediately centrifuged (Hettich, Tuttlingen, Germany) for 5 min at 10 4 rpm (Autokit II, Pharmap, Geneva, Switzerland). Hemoglobin concentration was determined using Abbott Cell Dyn 3500 (Abbott Diagnostic Division, Santa Clara, CA, USA). For assessing EPO in plasma, blood was collected in pre-cooled EDTA tubes, immediately spun for 10 min at 1500×g in a cooled centrifuge, plasma was collected and transferred to dry ice. All plasma samples were kept at −80°C until assayed. Plasma EPO concentrations were assessed using a commercial radioimmunoassay (RIA) kit (Epo-Trac TM125 I RIA kit; DiaSorin, USA). A known concentration of EPO antibody is incubated with 125-I radiolabelled antigens for binding. Blood plasma is added and unlabeled EPO antigens from plasma compete with the radiolabelled antigens for antibody binding sites. The higher the concentration of the EPO plasma antigens, more binds to the antibodies displacing more free radiolabelled antigens to the media. Radioactivity of the media is measured using a gamma counter (Quanta Smart for Tri-Carb 4910TR, PerkinElmer).
Measurement of mouse EPO (mEPO) and recombinant human EPO (rhEPO) in brain, kidney, liver, and spleen tissues. Concentrations of rhEPO were determined from frozen samples of brain, kidney, liver, and spleen from WT (P5) and Tg21 (both P5 and P21) mice using a commercial kit (Quantikine IVD ELISA, human erythropoietin, R&D Systems). Fresh wet weights of brain and spleen samples were registered for comparison. mEPO and rhEPO levels in the brain were additionally assessed at P60 in WT and Tg21 mice using commercial kits (Quantikine ELISA, mouse erythropoietin, and human erythropoietin, R&D systems).
Measurement of hippocampal EPO protein expression. Hippocampal EPO from WT and Tg21 mice was assessed in snap-frozen samples throughout development using a radioimmunoassay (RIA) kit (Epo-Trac 1251 RIA kit; DiaSorin, Saluggia, Italy) as previously described 76 .
DNA extraction and mitochondrial copy number. To extract genomic and mitochondrial DNA, 5-10 mg hippocampal tissue was digested by 100 µg/ml proteinase K in a lysis buffer (50 mM KCl, 10 mM Tris-HCl pH 8.3, 10 µg/ml gelatin, 0.045% Nonidet P-40, 0.045% Tween 20) at 55°C overnight. After heat inactivation of proteinase K and centrifugation to remove cellular debris, DNA was precipitated by sodium acetate and ethanol. After centrifugation, DNA pellets were air-dried then resuspended in nuclease-free water. DNA concentration was analyzed using Nanodrop 2000 (ThermoFisher Scientific). The amount of mitochondrial DNA (mtDNA) in tissue was estimated by the ratio of the mitochondrial MT-ND1 gene copy number and the nuclear N-B2M gene copy number 77,78 . Primers against both genes (Table 1) and SYBR Green (ThermoFisher Scientific, #A25741) were used for a semi-quantitative analysis by quantitative realtime PCR (7500 Fast Real-Time PCR System, ThermoFisher Scientific). The ratio of genomic (N-B2M gene) and mitochondrial (MT-ND1) DNA was determined by the ΔΔCt method 79 .
RNA extraction and mRNA expression analyses. Five-to-ten mg of hippocampal and cortical tissues from both genotypes (WT and Tg21) and sexes (male and female) brain tissue at P3, P7, P14, and P21 were used to extract RNA using ReliaPrep RNA Tissue Miniprep System (Promega, #Z6110). Purity was determined spectrophotometrically (Nanodrop 2000, ThermoScientific). First-strand cDNA was obtained by reverse transcription using RevertAid First Strand cDNA Synthesis Kit (ThermoFisher Scientific, #K1622). Samples (5 ng/µl cDNA) were analyzed by a SYBR Green (ThermoFisher Scientific, #A25741) semi-quantitative real-time PCR (qRT-PCR) (7500 Fast Real-Time PCR System, ThermoFisher Scientific). Primers for mRNA expression analyses were designed with Primer 3.0. Software to amplify either human or murine genes without cross-specificity (Table 1). Oligo properties were calculated using Oligo Analyzer 3.1. Before using primers for mRNA expression analyses, they were validated by qRT-PCR by (i) melting curve analyses (mode integrated in the 7500 Fast Real-Time PCR System) as well as on (ii) acrylamide gels to confirm the size and purity of PCR products. mRNA expression levels were calculated using the ΔΔCt method 79 and normalized to either ACTB or GAPDH. Each group of samples was normalized to WT P3 hippocampus. Annealing temperatures are shown in Table 1.
Detection and analysis of fluorescent bands were completed with the Li-COR Odyssey Platform (Biosciences).
High-resolution respirometry. Mass-specific (pmol O 2 /s * mg tissue wet weight) OCR were collected using a high-resolution Oxygraph-2k respirometer (Oroboros, Innsbruck, Austria). Standardized instrumental calibrations were performed to correct for the back-diffusion of oxygen into the chamber from various internal components, leak from the exterior, oxygen consumption by the chemical medium, and sensor oxygen consumption. All experiments were carried out in a hyper oxygenated environment (>200 nmol/ml) to prevent any potential oxygen diffusion limitations and oxygen flux was resolved by software allowing nonlinear changes in the negative time derivative of the oxygen concentration signal (DatLab, Oroboros, Innsbruck, Austria). All measures were collected at 37°C in a respiration buffer MiR06 (MiR05 + 280 iU/ml catalase) with saponin (50 μg/ml) to facilitate cell membrane permeabilization 81 . All substrate, uncoupler, and inhibitor titrations described below were added in series (Fig. 2a).
Respiratory titration protocol. The respiration protocol used examines individual aspects of respiratory control by initiating a specific sequence of respiratory states. Here we analyzed various respiratory states representative of mitochondrial proton leak (L), maximal rates of coupled oxidative phosphorylation (P), and maximal rates of uncoupled respiration (E). Each respiratory state is specific to the substrate(s), uncoupler(s), and inhibitor(s) included in the respiration medium at any given time in addition to the proton motive force and the relation of respiration to ATP production. L-state respiration, with kinetically saturating substrate and oxygen concentrations, represents uncoupled respiration due to proton leak and slippage across the inner mitochondrial membrane with a maximal proton motive force in the absence of ATP production and is comparable to the classical definition of either state 2 or 4 respiration 82 . Mitochondrially derived oxidant production is highest during L-state respiration 83,84 . P-state respiration, with kinetically saturating substrates, oxygen, and thermodynamically favorable adenylate concentrations to facilitate oxidative phosphorylation, represents respiratory rates that are well-coupled to ATP synthesis with a high proton motive force and are comparable to the classical definition of state 3 respiration 82 . E-state respiration, with kinetically saturating substrate and oxygen concentrations along with optimal exogenous protonophore provision, represents non-coupled respiratory rates independent of ATP synthesis with a collapsed proton motive force. Respiratory analysis began with the collection of L-state respiration without exogenous adenylates (L N ) following the addition of malate (2 mM) and octanoylcarnitine (0.2 mM). P-state respiration driven by electron input from the electron transfer flavoprotein complex and mitochondrial complex I (P ETF+CI ) was initiated with the addition of ADP (5 mM). As the brain poorly oxidizes lipid substrates, P ETF+CI can serve as an internal control of sample purity. Indeed, we found negligible changes in respiration between basal, L N , and P ETF+CI respiration (see result section). Accordingly, any reference of electron input from the electron transfer flavoprotein complex will be excluded from respiratory state identification throughout the results and discussion. Pstate respiration with maximal NADH-linked electron input from mitochondrial complex I, P CI , was induced following the additions of pyruvate (5 mM) and glutamate (10 mM). Maximal rates of P-state respiration, and the best representation of maximal oxidative phosphorylation potential in vivo, were then initiated with the addition of succinate (10 mM), which adds additional electron input from mitochondrial complex II, P CI+CII . Maximal E-state respiration, E CI +CII , was achieved with titrations of the protonophore, carbonyl cyanide p-(trifluoromethoxy) phenylhydrazone (FCCP), in 0.5 μM steps up to an optimum concentration (ranging from 1.5 to 3 µM). E-state respiration primarily reflecting electron input via mitochondrial complex II was then determined following the addition of rotenone (0.5 µM) and subsequent inhibition of mitochondrial complex I. Finally, non-mitochondrial residual oxygen consumption (ROX) was induced with the addition of antimycin A (2.5 μM) and attendant inhibition of mitochondrial complex III. All respiratory states were corrected for measures of ROX. Upon completing respiratory state analyses, ascorbate (2 mM) and N,N,N′, N′-tetramethyl-1,4-benzenediamine, dihydrochloride (TMPD, 500 μM) were simultaneously titrated into the chambers to assess cytochrome c oxidase (COX; complex IV) activity via mass-specific OCR (pmol O 2 /s * mg ww). Chemical calibrations were performed to determine and control for the auto-oxidation of TMPD occurring during measures of COX activity. For this, OCR is assessed over time using MiRO6 + saponin free of any biological sample with cytochrome c (10 μM), ascorbate (2 mM) and N,N,N′,N′-tetramethyl-1,4-benzenediamine, dihydrochloride (TMPD, 500 μM) added to the chamber.

Electron microscopy
Perfusion and fixation. Mice P14 were i.p. anesthetized with 40 μl pentobarbital (50 mg/ml) and, immediately following loss of pain perception, transcardially perfused with ice-cold PBS followed by 2.5%PFA/2.5%Gluteraldehyde in 150 ml sodiumcacodylate solution (0.13 M, pH = 7.4). Brains were removed from the skull and placed in a 50 ml Falcon tube containing the same fixative for post-fixation for at least 2 days, at 4°C.
Tissue block preparation. 80-100 μm thick hippocampal CA1 sections were cut in a vibratome and kept overnight in Na-cacodylate buffer (0.1 M) at 4°C.
Osmification. Na-cacodylate buffer was removed from the small PE tube and the tissues were incubated with 1% Osmium tetroxide (OsO 4 ) solution for 60 min. Tissues were then washed rapidly twice with Na-cacodylate buffer (0.1 M) and a third time for at least 30 min.
Tissues were transferred on a glass slide to create the inclusions, covered with Aclar embedding film (ACLAR®33 C Film, 7.8 Mil, Electron Microscopy Sciences, Hatfield, PA), and incubated for 48 h, at 60°C, for resin polymerization.
Cutting of semithin sections. When the resin was completely polymerized, very small regions of interest were dissected with a blade under the stereomicroscope (ZEISS, SteREO Lumar.V12) and mounted on small pyramids of resin with glue. They were left to air-dry overnight.
The mounted tissue was placed on the ultramicrotome (Leica Ultracut EM UC6, Vienna, Austria) and with a razor-sharp blade, they were given the shape of a trapezoid. The tissue was then aligned with a glass knife and semithin sections of 1µm-thickness were cut. Sections were collected with non-magnetic tweezers and placed in small drops of ddH 2 O on gelatinized glass slides. Glass slides were shortly placed on the heater and when the water drop was completely dried, then the tissue was contra stained with toluidine blue and observed with the brightfield microscope. Cutting of thin sections: Thin sections (70 nm thickness) were cut with a diamond blade and collected in ddH 2 O and gathered on a nickel grid (diameter: 3.05 mm, G300-Ni, 300 lines/inch square mesh, No. 100, Science Services GmbH, Munich, Germany). The grids were put in a grid box and air-dried overnight.
Contrasting. A small drop of double distilled water was placed on sections, followed by drops of uranyl acetate for 7 min, washed with double distilled water; subsequent drops of citrate for 7 min; and wash with double distilled water. Samples were then left to dry.
Image acquisition and data analysis. Semithin sections were observed with a brightfield Axioscope 2 microscope (Carl Zeiss AG, Oberkochen, Germany) equipped with a color digital camera (AxioCam MRc5) and its corresponding software, AxioVision 4.5 (Carl Zeiss AG, Oberkochen, Germany), and then EM micrographs were taken with the 100 kV transmission electron microscope (TEM -Philips CM100 and Telos) connected to a digital CCD camera.
Mitochondria number, mitochondria size, and vesicle numbers were quantified in each field of view with automatized particle counter from Fiji ImageJ (National Institutes of Health, USA).
Cognitive functions tests. Morris water maze (MWM), to test hippocampaldependent spatial learning and memory 85 , novel object recognition (NOR), which combines spatial and visual memory 86 , and a T-maze test for assessment of spatial working memory 87 were all performed with male and female animals of both genotypes at postnatal ages P45-60. After weaning, animals were housed in the testing room with room a 12 h light/dark inverted cycle and access to food and water ad libitum in standard laboratory conditions. Mice were controlled three times per week and handled once weekly to minimize handling-related stress that could impact performance in the task. Tests were performed during the dark phase under dimly-lit conditions (12 lux). Animals were video tracked at 4.2 Hz and 256 × 256-pixel spatial resolution using a Noldus EthoVision 1.96 system (Noldus Information Technology, Wageningen NL, www.noldus.com) throughout testing.
Morris water maze (MWM) test (P45-50). The MWM, test to test hippocampalmediated spatial memory, was employed as previously described 85 . The circular black arena was 150 cm in diameter with a wall height of 50 cm. High contrast spatial cues were placed on walls of the encompassing room and inside the arena above the water surface. The arena was filled with water to a height of 15 cm and maintained at 25°C. Day 1 consisted of teaching the mice to identify the escapable endpoint of the task by training them to find a visible Plexiglas platform (16 × 16 cm) placed randomly 0.5 cm underneath the water surface. Days 2-5 consisted of training the animals to find the hidden platform, which remained in a fixed position (acquisition phase). Throughout the acquisition phase, the water was made opaque by the addition of milk and the target platform was hidden 0.5 cm underneath the water surface 35 cm away from the surrounding walls in one quadrant (NW, NE, SE, or SW). Animals performed 6 trials of 120 s per day, with an inter-trial interval (ITI) of 30 min. On day 6, the platform was moved to the opposite quadrant (reversal phase) and four trials of 120 min were registered to measure spatial retention. The swim pathway of each mouse was automatically tracked and the time to reach the platform (escape time), total swim distance (m), and speed (m/s) were calculated.
Novel object recognition (NOR) test (P52-55). The NOR test relies on the animal's intrinsic preference for novelty. The choice to explore a novel object reflects learning and recognition memory, which also involves the hippocampus 88,89 together with cortical areas such as the visual cortex.
Mice were first habituated to an open field-testing arena (60 × 60 × 50 cm) with two spatial environmental cues on the walls for 10 min on two consecutive days. Mice were free to explore during habituation and the total activity time was registered. On the third day, animals were allowed 5 min to explore two equal objects placed in specific quadrants of the arena. The mice were subsequently removed from the arena for a 1-h ITI. One of the objects was replaced by another object similar in height and volume but different in shape and texture. Following this brief 1-h ITI, mice were reintroduced to the testing arena and allowed to freely explore the new object for 5 min. The next day (ITI of 24 h), mice were again allowed to explore the open testing field in the presence of a familiar and another new object. Time spent exploring each object and the number of explorations was recorded. An object discrimination ratio was calculated by (1): TN/(TN + TF). TN = time exploring the new object, TF = time exploring the familiar object. Animals that explored each object less than 10 s were excluded from the test.
T-maze spatial working-memory test (57)(58)(59)(60). Spatial memory was assessed by a rewarded alternation task in a T-maze made of gray poly-vinyl-chloride/plexiglas (each arm measures 30 × 10 cm) with a removable central partition and one guillotine door for each arm of the maze. Each mouse was habituated to the T-maze for two consecutive days before testing by allowing two mice to explore the T-maze with a food reward (pellets). The test was performed with 6 trials over two consecutive days. One trial consisted of two successive runs through the maze. In the first run (forced run) a food reward (R) was placed in an open arm. The other arm was closed with a barrier. Mice were then immediately returned to the start arm for 5 s and allowed a second run in which both arms were available, so the animals could enter either arm (choice run), however, food was available only in the arm closed on the first run. Consequently, spatial working memory is necessary to remember which arm was open in the first trial and to alternate. A run was terminated if a choice was not made within 2 min. The sample arm available to the mice on the first forced run varied randomly from trial to trial. Following each trial, the mice were removed from the T-maze and returned to their cage for a 1-h ITI (short-term memory). This spatial working-memory task is quantified by spontaneous alternation. The percentage of alternation and time to reach the reward was calculated per animal.
Statistics and reproducibility. For all statistical evaluations included in this study, unless otherwise specified, an α of p < 0.05 was considered significant and data are reported as mean ± SD.
Comparisons of rhEPO expression in brain, kidney, liver, and spleen in WT and Tg21 mice ( Supplementary Fig. 1c), and the general differences across respiratory states in WT (Figs. 2e, 3e and Supplementary Fig. 2c, d) and Tg21 mice (Figs. 2f, 3f and Supplementary Fig. 2c, d), were analyzed with a one-way repeated measures ANOVA and individual differences identified using Tukey's HSD post hoc test.
The main effects of genotype for outcome variables derived from the behavioral tests were analyzed using a 2-way ANOVA and post hoc analyses identifying individual group differences were conducted with Sidak's correction for multiple comparisons (GraphPad Prism 8.0.1.).
All tests were performed using Prism 8.0.1. GraphPad Software, San Diego, CA, USA.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available within the paper and its supplementary information files. Source data underlying figures are presented in Supplementary Data 1. Individual data sets generated during the cognitive tests are available from the corresponding author on reasonable request.