Erythropoietin restrains the inhibitory potential of interneurons in the mouse hippocampus

Severe psychiatric illnesses, for instance schizophrenia, and affective diseases or autism spectrum disorders, have been associated with cognitive impairment and perturbed excitatory-inhibitory balance in the brain. Effects in juvenile mice can elucidate how erythropoietin (EPO) might aid in rectifying hippocampal transcriptional networks and synaptic structures of pyramidal lineages, conceivably explaining mitigation of neuropsychiatric diseases. An imminent conundrum is how EPO restores synapses by involving interneurons. By analyzing ~12,000 single-nuclei transcriptomic data, we generated a comprehensive molecular atlas of hippocampal interneurons, resolved into 15 interneuron subtypes. Next, we studied molecular alterations upon recombinant human (rh)EPO and saw that gene expression changes relate to synaptic structure, trans-synaptic signaling and intracellular catabolic pathways. Putative ligand-receptor interactions between pyramidal and inhibitory neurons, regulating synaptogenesis, are altered upon rhEPO. An array of in/ex vivo experiments confirms that specific interneuronal populations exhibit reduced dendritic complexity, synaptic connectivity, and changes in plasticity-related molecules. Metabolism and inhibitory potential of interneuron subgroups are compromised, leading to greater excitability of pyramidal neurons. To conclude, improvement by rhEPO of neuropsychiatric phenotypes may partly owe to restrictive control over interneurons, facilitating re-connectivity and synapse development.


Summary
Erythropoietin (EPO) aids in rectifying hippocampal transcriptional networks and synaptic structures of pyramidal lineages, thereby mitigating mood and cognitionassociated disorders. An imminent conundrum is how EPO restores synapses by involving interneurons. By analyzing ~ 12,000 single-nuclei transcriptomic data, we generated a comprehensive molecular atlas of hippocampal interneurons, resolved into 15 interneuron subtypes. Next, we studied molecular alterations upon recombinant human (rh)EPO and saw that gene expression changes relate to synaptic structure, trans-synaptic signaling and intracellular catabolic pathways. Putative ligand-receptor interactions between pyramidal and inhibitory neurons, regulating synaptogenesis, are altered upon rhEPO. An array of in/ex vivo experiments confirms that specific interneuronal populations exhibit reduced dendritic complexity, synaptic connectivity, and changes in plasticity-related molecules. Metabolism and inhibitory potential of interneuron subgroups are compromised, leading to greater excitability of pyramidal neurons. To conclude, improvement by rhEPO of neuropsychiatric phenotypes may partly owe to restrictive control over interneurons, facilitating re-connectivity and synapse development.

INTRODUCTION
As the nomenclature reflects, the majority of studies on erythropoietin (EPO), a hypoxia-inducible growth factor, are dedicated to its hematopoietic function 1,2 .
Indeed, transgenic strategies demonstrated that constitutive overexpression of EPOR in principal neurons enhances cognitive functions in mice 18 . This improvement builds on the EPO-mediated maneuvering of neuronal differentiation trajectories, overpopulation of distinct pyramidal lineages, enhanced dendritic spine density and improved motor-cognitive execution, which is overlaid with induction of related gene regulatory networks 17,[19][20][21] .
While the plasticity of excitatory hippocampal circuits bears the most of rhEPOinduced changes in cognition, pyramidal neurons from the CA1 region are accurately timed and synchronized by a rich diversity of GABAergic interneurons [22][23][24][25][26] . It is consensus that the different classes of interneurons interact with specific cellular domains of pyramidal neurons to establish the excitatory:inhibitory (E:I) balance, properly modulating cell firing, plasticity, network activity, metabolism, and transsynapses. Like excitatory neurons, interneurons in the adult brain are capable of remodeling their structure and connectivity through changes in the expression of plasticity-related molecules, e.g. the polysialylated form of the neural cell adhesion molecule (polySia-NCAM), and specialized regions of the extracellular matrix denominated perineuronal nets (PNNs) [27][28][29][30][31][32] . How or to which extent interneurons remodel in response to rhEPO has remained enigmatic. Given the seminal papers in this field, it is imperative to determine the impact of rhEPO on interneurons to better resolve the mechanistic basis of cognitive advance. We wondered whether interneurons, subjected to rhEPO, undergo structural or physiological alterations, and molecular changes in gene regulatory networks, controlling crosstalk of pre-and postsynapses.
In the hippocampus, basket cells represent one of the main interneuronal subpopulations, given their abundance and strong inhibitory input on pyramidal neurons 22 . They extend their axonal arbor in the stratum pyramidale and vastly target the somata and proximal dendrites of pyramidal neurons 24,25,33 . This group is Interestingly, SST O-LM interneurons exhibit dendritic spines that remain highly plastic during adulthood 36,37 .
Although mouse studies have carefully detailed the characteristics and events that precede the formation and survival of interneurons, molecular criteria to define interneuron subgroups within major lineages seem sparse 25,[38][39][40] . This might either be due to their highly diverse and heterogeneous nature, or the conjecture that their specification is influenced by intrinsic or extrinsic factors [41][42][43] . High-resolution reference transcriptome of all hippocampal interneuron lineages following rhEPO treatment would greatly benefit the present study, but is currently lacking.
Even though the structural and molecular basis of interneurons in mammals is functionally similar, our understanding of their diversity and mechanisms of regulation in hippocampus remains limited 39,44 . Here, we first aimed to resolve the transcriptomic landscapes of interneuronal lineages in the murine hippocampus and then investigated alterations upon rhEPO to potentially reveal mechanistic details on its action.
Identifying the key interneuron lineages and their modified cell-to-cell communication with pyramidal neurons should help to discern how synapses are renovated by rhEPO.
With a strategic analysis of our single nuclei (sn)RNA-seq datasets 21 , we resolve 15 transcriptionally distinct interneuron lineages. Upon whittling the genes from the transcriptome of each lineage that are differentially expressed upon rhEPO, we show that EPO modulates the expression of genes involved in neurite morphology, synapses and metabolism in a lineage-specific manner. Also, we uncover the in-silico anatomical organization of pyramidal-interneuron connections mediated by ligand-receptor interactions that are re-bridged upon rhEPO. We overlaid our transcriptomics approach with robust in/ex vivo experiments to understand the impact of rhEPO over metabolism and plasticity of interneurons.

The snRNA-seq data analysis of hippocampal interneurons resolves 15 distinct lineages
A growing list of seminal studies has illustrated the enormous diversity of hippocampal interneurons. Accountability of rhEPO in altering the interneuronal transcriptome is still obscure 39,45 . Thus, we first classified all existing sub-lineages at single interneuron level. From the snRNA-seq of 12 hippocampi [6 each for rhEPO and placebo (PL) treatment], we obtained ~ 108,000 single-nuclei transcriptomes grouped into 11 major cell types 21 . Among them, ~16,000 cells (15%) nuclei were defined as interneurons, according to our cell-type classification and supported by their corresponding bonafide markers, allowing to identify interneuron subtypes.
Performing an unsupervised clustering using the most variable genes and graph-based learning on our snRNA-seq dataset 21 (see methods), we identified numerous distinct clusters of interneurons (Supplementary Figure 1A). Clusters were obtained after a rigorous batch-control algorithm, preventing sample biases (Supplementary Figure 1A), and classified based on differential expression of transcriptomes. To ensure clusters to be homogeneously interneurons, expression of genes marking both interneuron and non-interneuron lineages were screened. Though the majority of clusters were interneurons, i.e. co-expressing Gad1 and Gad2 genes but not markers of other lineages, a few clusters expressed markers of microglia, oligodendrocytes or Cajal-Retzius cells (Supplementary Figure 1B). Although it is tempting to resolve ancestral and derived cells of these heterogeneous clusters, we removed them in a stepwise process (see methods) and focused on mature homogeneous interneurons and not the transitory or yet-to-be-committed cells ( Figure   1A). These lineages expressed Gad1 and Gad2 at a higher level without displaying marker gene expression of other hippocampal lineages. For instance, Tgfbr1 for microglia, Gfap for astrocytes, Plp1 and Pdgfra for oligodendrocytes, Bsg for pericytes, Flt1 for endothelium and Slc17a6/Neurod6 for pyramidal neurons were tested in our classified interneuronal clusters. Again performing unsupervised clustering of the remaining ~9000 nuclei 21 , we identified 15 distinct clusters (Figure 1A-B). These nuclei did not cluster by batch or samples, indicating a robust control of such effects (Supplementary Figure 1A). Instead, we classified these 15 clusters based on the combination of known, bonafide and discovered gene expression markers 39 ( Figure  1C, Table S1). Our data resolved the distinct transcriptome signature separating three types of SST, two CCK, two PVALB, two IS, two Ivy cells, and two distinct lineages termed Trilaminar and Cholinergic interneurons. We also observed two previously uncharacterized interneuronal populations that we classified based on top marker genes as Nrg1/Ptprd and Zbtb20/Mgat4c interneurons (Figure 1A-C). Collectively, we provide a transcriptome reference frame as valuable resource to interpret and analyze the diversity and heterogeneity of hippocampal interneurons via comprehensive survey of snRNA-seq data postdating rhEPO and PL treatment.

rhEPO modulates gene expression of interneuronal subpopulations involved in E:I balance
The broader understanding of rhEPO mode of action warrants re-investigating the above identified interneuron lineages for differential expression of genes (DEGs) upon rhEPO. We first determined the aggregated expression of each gene from all nuclei within a lineage from rhEPO against PL samples, and then calculated the level of their differential expression using the recommended statistical set-up (see methods).
Given their relevance in the mouse hippocampus, we curbed our DEGs analysis to SST (O-LM and backprojection), PVALB (basket and axo-axonic cells), and CCK (basket and dendrite-targeting) lineages. Using the statistical threshold of adjusted pvalue <0.05, our analysis proclaims 1073 DEGs, accounting for all comparisons in these six lineages. While a few DEGs were shared between multiple lineages, most were unique to a particular lineage, suggesting that the impact of rhEPO on gene expression occurs in cell-type-specific manner (Figure 2A, Table S2).
To get a functional clue of the impact of rhEPO treatment, we conducted pathway and gene ontology analyses on the 1073 DEGs, applying KEGG databases and biological processes databases from Gene Ontology Consortium. These DEGs overrepresent genes related to glutamate receptors, synaptic structure, trans-synaptic signaling and intracellular catabolic pathways ( Figure 2B). We further asked if these DEGs also overrepresent gene sets associated with documented phenotypes at molecular and individual levels upon rhEPO [10][11][12][13][14][15][16][17] . Expectedly, we found phenotypic association with dendrite morphology, synaptic transmission and cognition ( Figure   2B), which are affirmed in previous studies to be positively regulated by rhEPO 12,17,19 .
Among the genes whose expression was toned down in rhEPO samples, we identified Gabra2 and Unc5d in the SST O-LM lineage, whose repression might lead to loss of the inhibitory effect (Figure 2A, and 2C). On one hand, Gabra2 keeps the inhibition of neurons intact at the synapse 46 ; on the other hand, Unc5d facilitates migration of interneurons to the synapse 47 . It is compelling to hypothesize that repression of these two genes could be one of the many causes of higher excitability of pyramidal neurons in rhEPO subjects, as shown previously 21 . Among the upregulated genes in rhEPO samples, we noticed Filip1 in both PVALB and SST lineages ( Figure 2C). Filip1 encodes a structural protein which shapes dendrite or neurite morphology and mediates neuronal migration 48,49 . Moreover, Bmpr1a, an essential receptor regulating downstream BMP signaling, is among the upregulated genes in both PVALB and SST lineages upon rhEPO ( Figure 2C). Of note, recent genetic experiments have illustrated that Bmpr1a guides specific synaptic connectivity in PVALB interneurons 50,51 . In addition, we found few other genes with opposite trend of differential expression between sister lineages, also associated with synaptic pruning and maturation of interneurons. For instance, Jak2, known to prune inactive synapses 52,53 appeared to be induced in SST-hippocamposeptal and CCK dendritetargeting cells; whereas its expression disappeared in the SST-backprojection lineage upon rhEPO ( Figure 2C). Similarly, Pde1c, a phosphodiesterase enzyme, mediating neuronal oxidative metabolism 54 , has gain and loss of expression in CCK basket and dendrite-targeting lineages, respectively, upon rhEPO ( Figure 2C). Notably, on the list of DEGs, we also identifed multiple phosphodiesterase genes possessing oxidative metabolic properties, further corroborating our hypothesis that rhEPO can palliate metabolic dysregulations (Figure 2A). The above results point towards an interplay of rhEPO with interneuronal physiology, much more diverse than previously appreciated.
It is enticing to hypothesize that the rhEPO-governed gene expression changes attenuate the interaction of interneurons with their target cells. This might underpin molecular criteria defining rhEPO-enforced interneuron plasticity, metabolism, and E:I balance.

rhEPO modifies the cell-to-cell interactions between interneuronal and pyramidal lineages
So far, we demonstrated that the impact of rhEPO might be associated with mitigation of intracellular signaling cascades in interneurons. We hypothesized that this impact might often be due to a differential receptor-ligand activity. To systematically study the interactions between interneuronal and pyramidal lineages, we employed LIANA framework, a specialized tool for ligand-receptor analysis, using snRNA-seq data to infer cell-cell communication 55 . LIANA leverages the expression levels of genes overlaying with the CellPhoneDB repository of ligand-receptor pairs 56  We next assessed rhEPO-specific differences in the ligand-receptor profiles.
While most of the interactions were similar in our catalogue of ligand-receptor repertoire, upon closer inspection, we observed a subtle loss and a gain of interaction in rhEPO samples. To elaborate, the Calm2-Pde1c pair, a known interaction partner 63 was lost between PVALB basket cells and mature CA1 pyramidal neurons upon rhEPO, suggesting the attenuation of calcium-dependent signaling between these two lineages ( Figure 3C). The loss of this interaction is consistent with the downregulation of a bunch of genes encoding phosphodiesterases in interneuron lineages.
Furthermore, Pdgfc-Pdgfrb paired between newly formed pyramidal neurons and CCKdendrite targeting interneurons was surfaced in rhEPO samples in contrary to PL ( Figure 3E). Interestingly, this ligand-receptor is known to regulate neurogenesis, cell survival and synaptogenesis in hippocampus [64][65][66] , which supports the notion that EPO mediates hippocampal neurogenesis and synaptogenesis 17,19,21 . Together, this data supports a potential control of rhEPO over trans-synaptic signaling and E:I balance through tinkering specific cell-to-cell communications between interneurons and pyramidal lineages.
While the segregation of rhEPO and PL samples through DEGs and ligandreceptor communications in interneurons is compelling, it is merely based on expression at RNA level and there are certain limitations to the snRNA-seq data, as noted in multiple reports before. Despite these limitations, the above observations suggest that the alterations in interneuron physiology are acquired due to rhEPO treatment. To concur this, we inquired whether interneuronal populations express EPOR, which should establish that they can directly respond to rhEPO. We employed a highly sensitive fluorescence in-situ hybridization (FISH) method, able to detect Epor mRNA ( Figure 3G-H). We analyzed the expression of Epor in GAD67+ (encoded by So far, the differential expression profiles and Epor expression seem to agree with the hypothesis that rhEPO modulates the interneuronal population; however, these findings still lack validation at cellular and physiological levels. We therefore performed a series of in/ex vivo experiments and specifically asked whether E:I balance, structural and synaptic properties, and metabolism of specific interneuronal subpopulations are changed upon rhEPO.

rhEPO treatment decreases the structural complexity of specific hippocampal interneurons
Following the above leads, we analyzed -after 3-week PL/rhEPO treatmentstructural changes in two major subpopulations of interneurons ( Figure 4A

rhEPO treatment marginally affects excitatory/inhibitory (E:I) balance in the hippocampal parenchyma
We subsequently investigated the impact of rhEPO on E:I balance in the hippocampus. As the phenomenon of brain homeostasis might compensate for such imbalance, we did not expect a dramatic E:I disturbance in whole parenchyma upon rhEPO, but a noticeable difference. We asked whether rhEPO may affect the expression of excitatory and inhibitory presynaptic markers ( Figure 5A) and thus analyzed the density of puncta expressing VGLUT1 and VGAT as a proxy for synaptic density, reflected as E:I ratio (Figure 5B-E). While our results show, in general, a trend of higher E/I ratio after rhEPO, expression of these excitatory and inhibitory markers was not significantly different. Nonetheless, we observed a slight increase in VGAT+ puncta density in stratum radiatum ( Figure 5D; t=2.178; p=0.0574), leading to reduction of E/I balance (density of VGLUT1+ puncta/density of VGAT puncta) (t=2.205; p=0.0549). Collectively, we add another layer to our above findings: EPO seems to reconfigure synapses, probably by fine-tuning the balance between excitatory and inhibitory neurons.

Electrophysiological recordings elucidate a shift in the hippocampal excitatory/inhibitory balance (E:I) upon rhEPO
We next explored functional facets of rhEPO treatment on hippocampal activity ( Figure 6A). In a virtual reality environment, we obtained in awake mice local field potential (LFP) recordings of the pyramidal layer of the dorsal CA1 subfield. With this design, we induced exploratory behaviors in the animals, a condition evoking an active hippocampal state in form of theta rhythmicity.
Selecting theta epochs on hippocampal LFP, we filtered the raw signal to detect presence of coupled gamma waves ( Figure 6B), a typical oscillatory profile of hippocampal theta activity, reflecting periodic bouts of excitation and inhibition 68 .
According to Gao and colleagues 69 , we explored a slope-fitting model of the power spectra on the 30-50Hz band to capture differences in E:I ratio ( Figure 6C). We found a decrease of the 30-50Hz slopes in rhEPO treated mice (placebo: sPL=-2.91; rhEPO: sEPO=-2.52; t=14.96, p<0.001; Figure 6D), calculated by regression models on the mean power spectral density (PSD) for both groups. We confirmed this result by comparing the means of 30-50Hz slopes for both groups, calculated from the PSD of each theta cycle (placebo: sPL=-3.06; rhEPO: sEPO=-2.75; t=4.26, p<0.001; Figure   6E).

rhEPO decreases the metabolic activity of hippocampal PV-expressing cells
We used nanoscale secondary ion mass spectrometry (NanoSIMS) 19,20,70 for analyzing the metabolic turnover in the CA1 PVALB subpopulation (PV+) through 15 Nleucine incorporation ( Figure 6F). This method provides indirect evidence of protein synthesis, commonly associated with e.g. growth processes or cellular activity. When we compared the 15 N/ 14 N ratio in PV+ cells, we observed a significant decrease after rhEPO (p=0.0016), indicative of an abridged metabolic turnover. This contrasts to the increased 15 N-leucine incorporation previously observed upon rhEPO in NeuN+ cells of the stratum pyramidale 19 . In parallel to the rhEPO stimulated neurodifferentiation of pyramidal cells in CA1, resulting in substantial numbers of new functional neurons with increased metabolism and density of dendritic spines 17,19 , we observed a decrease in structural complexity, connectivity, and metabolism in specific interneuronal subpopulations.

rhEPO alters the expression of plasticity-related molecules associated with hippocampal interneurons
Certain types of plasticity-related molecules play key roles in the structural remodeling and connectivity of neurons, specifically interneurons 31,32,71,72 . We initially performed a densitometric analysis of polySia-NCAM expression in the different layers of the CA1 region ( Figure 7A). Increases after rhEPO treatment were observed in the strata oriens (t=2.602; p=0.0315), radiatum (t=2.782; p=0.0239) and lacunosummoleculare (t=3.076; p=0.0152; Figure 7B). A tendency towards an increase was also

DISCUSSION
Improvement of mood and higher cognition following rhEPO treatment can in principle be an indirect effect due to increase in hematocrit and thus oxygen delivery.
Although we do not discard this possibility entirely, our previous findings support a hematopoiesis-independent, specific mode of rhEPO action in brain. For instance, elimination of EPOR from pyramidal neurons prevented rhEPO-induced hippocampal and behavioral changes 17 . Transgenic overexpression of spontaneously active EPOR in hippocampal pyramidal neurons enhanced higher cognition 18 . Moreover, we reported very recently that rhEPO markedly influences the transcriptome of excitatory hippocampal neurons, thereby inducing e.g. associated signaling cascades to tackle metabolic challenges 21 .
Despite the complex nature of synapses, it takes both pyramidal cells and interneurons to form viable circuits. In the present work, we identified interneuron lineages that exhibit core functional differences upon rhEPO, thus taming synapses.
Employing the relevant computational approaches, we here provide a detailed transcriptomic analysis of hippocampal interneurons in mice, treated with either rhEPO or PL. Although most lineages classified in this study are consistent with previously reported interneuronal subpopulations, we found two distinct lineages that are not documented to date. These lineages exhibit distinct transcriptomes, co-express Gad1 and Gad2 genes at high level, but not markers of other lineages, supporting their interneuronal origin. We named them according to the top marker genes they express, Zbtb20/Mgat4c and Nrg1/Ptprd, respectively.
Upon rhEPO, we observe a substantial difference in the gene expression profiles of distinct subgroups of interneurons. To determine whether the changes in transcriptional readouts are secondary events or whether rhEPO is directly involved in modulating them, we asked if these subgroups of interneurons do express Epor.
Indeed, we found that a generous portion of interneurons is expressing Epor, enough to be detected in our FISH experiments. These results affirm that the interneuron population is an addition to the known arena of EPO action, which are excitatory neurons and glial cells 17,[19][20][21] . Therefore, this premise is pleading for further During all the procedures, the experimenter was blind to group assignment.

Perfusion and Microtomy
Mice were deeply anesthetized with sodium pentobarbital (150 mg/kg) and
After washing, sections were incubated for 2 h at room temperature with different secondary antibody cocktails (Table S4)  To analyze the expression of polySia-NCAM in conventional light microscopy, sections were processed using the ABC method. After several washes, sections were first incubated for 1 min in an antigen unmasking solution (0.01 M citrate buffer, pH 6) at 100 °C. After cooling down the sections to RT, they were incubated with 10% methanol and 3% H2O2 in PBS for 10 min to block endogenous peroxidase activity.
After this, sections were blocked and incubated with the primary antibody as mentioned before. After washing, sections were incubated for 2 h (at RT) with a secondary biotinylated antibody followed by an avidin-biotin-peroxidase complex (ABC; Vector Laboratories, Peterborough, UK) for 1 hour in PBS. See Table S4 for antibody information. Color development was achieved by incubating with 0,05% 3,3′-diaminobenzidine tetrahydrochloride (DAB; Sigma-Aldrich) and 0.033% H2O2 for 4 min.
Finally, sections were mounted on slides, dried for 1 day at RT, dehydrated with ascending alcohols and rinsed in xylene. After this, sections were coverslipped using Eukitt mounting medium (PANREAC).

Surgery
At 24 hours after the last PL/rhEPO injection, mice were anesthetized with isoflurane (4% for induction, 2% for maintenance, both in 0.5 ml O2/min flow rate).
Additionally, atropine (0.1 mg/kg) and buprenorphine (0.1 mg/kg) were s.c. injected to minimize respiratory pitfalls and suffering during surgery. Ophthalmic ointment was applied to protect the eyes during surgery. Animals were placed in a stereotaxic frame (Narishige), then lidocaine (2%, Normon) was s.c. injected above the incision area and the skull surface was exposed and cleaned with 70% ethanol. Afterwards, the skull was trephined in two locations in the following coordinates relative to bregma: AP -2.00, ML -1.50 for the subsequent in vivo recording of the right dorsal hippocampus, and AP +0.30, ML +0.50 to place a reference electrode (formvar insulated stainless steel monopolar macroelectrode (120 μm in diameter, WPI) in the subarachnoid space.
Then, a polymeric resin (OptiBond; KERR) was applied on the skull surface and polymerised using a UV light lamp (Woodpecker Led D). Afterwards, a custom-made 3D printed resin headplate was placed on the skull and dental cement (DuraLay, Reliance Dental) was applied to cover all the tissue exposed except the cranial window, sealed with SILASTIC (Kwik-cast, WPI). After surgery, mice were s.c. injected with buprenorphine (0.1 mg/kg) and meloxicam (3 mg/Kg) and placed under a heat lamp and monitored until complete recovery.

Electrophysiological Recording
Training: At 48 hours after surgery, mice were trained for two consecutive days to get familiarized with our custom-made virtual reality platform (CIBERTEC, Spain). After a water restriction period, ranging randomly from 1 to 3 hours, mice were placed in a head-fixed frame on a low-friction styrofoam cylinder and positioned in front of a TFT were configured and designed with colours within the range of wavelengths perceived by the mouse visual system 95,96 . The virtual corridors were split into four virtual sectors (S1-S4) based on distinct cue patterns on the walls. A distinctive cue (white cross) identified a final reward area, where a 1% sucrose dilution on water drop was provided.
Animals were left undisturbed to explore twice the same corridor without time limitation.
Recording: Twenty-four hours after the end of training, electrophysiological recording of the mice was carried out in one session with full corridor exploration. After placing them in the head fixation frame, a cranial window was exposed, and the local field potential (LFP) of the dorsal hippocampus (CA1) was recorded with a 120 μm diameter Teflon-coated steel monopolar macroelectrode (WPI), whose tip was located 1.30 mm ventral to the brain surface. The signal was referenced against the electrode previously placed in epidural space and acquired using the Open-ephys system 97  It is well-known that during active exploration, theta (5-12 Hz) oscillations dominate the hippocampal CA1 area of the rodent brain 98 . We extracted the theta periods by decomposing the normalized signal into the different predominant frequency bands by the empirical mode decomposition (EMD, Hilbert-Huang transform), thereby preserving the time domain during periods of active exploratory behavior (animal velocity > 2 cm/s). We isolated the low-frequency (< 5Hz), theta (5-12 Hz) and supratheta (> 12 Hz) signals by combining the components with mean instantaneous frequencies into the ranges < 5 Hz, 5-12 Hz and >12 Hz, respectively.
We then identified individual theta cycles by detecting each cycle's local maxima and minima with an absolute value above the envelope of the low-frequency time series. A theta cycle was thus defined by a candidate central peak surrounded by a pair of consecutive valleys, separated at least by 71 ms (~14 Hz) and no more than 200 ms (~5 Hz). test with a significance level of p < 0.05 was used to determine significant differences.
All results were expressed as mean ± SEM. Graphs were obtained using the R package ggplot2. Data access (DOI: 10.5281/zenodo.7885936).

Structural Analysis of GAD-EGFP and PV Expressing Interneurons
All the structural parameters of GAD-EGFP+ (~SST O-LM) and PVALB interneurons were studied using a laser scanning confocal microscope (Leica TCS SPE) as described before 67

Analysis of the Excitatory/Inhibitory Balance
We analyzed the density of neuropil puncta expressing the vesicular glutamate (VGLUT1) and GABA (VGAT) transporters, as a redoubt of the excitatory and inhibitory level, respectively, in the different layers of the hippocampal CA1 region. Confocal zstacks covering the whole depth of the sections were taken with 0,38 μm step size (63x oil objective and 2x digital zoom magnification) and a laser scanning confocal microscope (Leica TCS SPE, Germany). Only confocal planes with the optimal penetration level for each antibody were selected. On these planes, 16 small squares of the neuropil (336 μm 2 ) per layer and animal were selected for analysis to avoid blood vessels and cell somata. Images were processed using FIJI/ImageJ software as described before 102,103 : the background was subtracted with rolling value of 50, converted to 8-bit deep images and binarized using a determined threshold value. This value depended on the marker and the area analyzed and was kept the same for all images with the same marker and area. The images were then processed with a blur filter to reduce noise and to separate closely apposed puncta. Finally, the number of the resulting dots per layer was automatically counted and expressed as a density.
The E/I ratio was calculated as the density of VGLUT1 expressing puncta divided by the density of inhibitory VGAT expressing puncta.

Analysis of Inhibitory Perisomatic Puncta on Excitatory Neurons
To analyze the perisomatic innervation that PVALB and CCK basket cells exert, the density of puncta immunoreactive for PV or CB1r surrounding pyramidal neuron somata (identified by CAMKII expression) was analyzed as described previously 104 .
Briefly, 20 CAMKII expressing neurons per animal located in the stratum pyramidale of the CA1 region, were randomly imaged. Confocal z-stacks covering the whole depth of the neuron somata were taken with 0.38 μm step size using a 63× oil objective with 2× digital zoom magnification (Leica TCS SPE, Germany). Stacks were then processed using FIJI/ImageJ software. Single confocal planes from each neuron, in which the penetration of each antibody was optimal, were selected. Briefly, the profile of the soma of these neurons was drawn manually and the selection was enlarged 1.25 μm to cover the area surrounding the somata. The region of interest (ROI) was then binarized and puncta was defined as a structure displaying an area not smaller than 0.15 μm 2 and not larger than 2.5 μm 2105 . The density of puncta (number of puncta per micron of soma perimeter) was analyzed as described above (see "Analysis of the excitatory/inhibitory balance"). The percentage of area covered by PV+ puncta was additionally analyzed in P90 mice.

Analysis of PolySia-NCAM Immunoreactivity
Sections starting at Bregma -1.94 mm were selected to study the immunoreactivity of polySia-NCAM in the different layers of the hippocampal CA1 region as described previously 106 . Samples were examined with an Olympus CX41 microscope under bright-field illumination, homogeneously lighted and digitized using a CCD camera. Photographs of the different layers were taken randomly at 20x magnification using the same exposure time and ISO. Grey levels were converted to optical densities using FIJI/ImageJ software (NIH) and normalized with respective internal white matter regions.

Quantification of PVALB Cells, PNNs and Analysis of Hippocampal Volume.
To study the number of PVALB neurons (PV+), PNNs and their co-localization in CA1 hippocampal region, we used a modified version of the fractionator method 107,108 . Briefly, within each 50-µm-thick section of one from the six systematicrandom series of sections, all labeled cells covering the 100% of the sample area were counted. Confocal z-stacks covering the whole depth were taken with a confocal microscope (Olympus FV-10) using a 10× objective to obtain the 2D projections. The images were processed afterward using FIJI/ ImageJ software 99 .
A volumetric analysis of the hippocampal CA1 region was also performed on these sections. Microphotographs at 10x were obtained with a confocal microscope, and then processed using the stitching plugin in FIJI/ImageJ 99 in order to have the whole nucleus on a single microphotograph. All the sections in a 1/6 subseries containing the region of interest were analyzed and the areas estimated using Cavalieri's principle 109 . To obtain the volume of the CA1 region, we multiplied the area by the thickness of the slice (50-µm) and the number of series. We finally summed up the volumes across all slices for each animal.

Expression of EPOR in GABAergic Neurons
RNAscope Multiplex Fluorescent assay v2 (CatNo. 323100) from Advanced Cell Diagnostics (ACD; Hayward, CA, USA) was used for the detection of Epor mRNA in GAD67 and PVALB positive cells, encoded by Gad1 and Pvalb genes, respectively.

Statistical Analysis
Data is shown as mean ± standard error of the mean (SEM), with N numbers (i.e., number of mice/group) and statistical test specified in the text or corresponding     All graphs show mean±SEM; N numbers depicted as dots in the bars; unpaired twotailed Student's t-test. All analyses were conducted following treatment scheme in Figure 4B, except Figure 5F that followed the 3-week PL/rhEPO treatment at older age (P90). Scale bar: 5μm for A; 25μm for F-G panoramic views and 5μm for insets.  All graphs show mean±SEM; N numbers depicted as dots in the bars; unpaired twotailed Student's t-test (except 7E, one-tailed). All analyses were conducted following treatment scheme in Figure 4B, at same age, and same area evaluated; scale bar: 60μm for A; 12.5μm for overview in C and 9μm for inset magnifications.