Targeted rescue of synaptic plasticity improves cognitive decline in sepsis-associated encephalopathy

Abstract

Sepsis-associated encephalopathy (SAE) is a frequent complication of severe systemic infection resulting in delirium, premature death, and long-term cognitive impairment. We closely mimicked SAE in a murine peritoneal contamination and infection (PCI) model. We found long-lasting synaptic pathology in the hippocampus including defective long-term synaptic plasticity, reduction of mature neuronal dendritic spines, and severely affected excitatory neurotransmission. Genes related to synaptic signaling, including the gene for activity-regulated cytoskeleton-associated protein (Arc/Arg3.1) and members of the transcription-regulatory EGR gene family, were downregulated. At the protein level, ARC expression and mitogen-activated protein kinase signaling in the brain were affected. For targeted rescue we used adeno-associated virus-mediated overexpression of ARC in the hippocampus in vivo. This recovered defective synaptic plasticity and improved memory dysfunction. Using the enriched environment paradigm as a non-invasive rescue intervention, we found improvement of defective long-term potentiation, memory, and anxiety. The beneficial effects of an enriched environment were accompanied by an increase in brain-derived neurotrophic factor (BDNF) and ARC expression in the hippocampus, suggesting that activation of the BDNF-TrkB pathway leads to restoration of the PCI-induced reduction of ARC. Collectively, our findings identify synaptic pathomechanisms underlying SAE and provide a conceptual approach to target SAE-induced synaptic dysfunction with potential therapeutic applications to patients with SAE.

Graphical abstract

Grünewald and colleagues report compromised ARC expression and BDNF signaling in a murine model of severe sepsis causing synaptic dysfunction and long-lasting cognitive impairment. Rescue of synaptic dysfunction with AAV-mediated ARC overexpression and general activation with enriched environment improved synaptic function and memory, thus representing a promising therapeutic avenue in sepsis-associated encephalopathy.

Introduction

Growing evidence suggests that severe systemic infections may lead to long-term brain dysfunction with the risk of developing cognitive dysfunction up to overt dementia and post-traumatic stress disorder (PTSD) with increased anxiety.^1,2 Sepsis, as the most severe manifestation of systemic infection, is defined as a life-threatening organ dysfunction caused by an overwhelming and dysregulated host response to inflammation.³ Sepsis is a worldwide health problem with increasing incidence and high mortality. Recent epidemiological studies revealed an incidence rate of 437 hospital-treated sepsis patients per 100,000 person-years and an estimated global incidence of 31.5 million sepsis cases with 5.3 million deaths annually.⁴ Sepsis-associated encephalopathy (SAE) is a frequent and severe complication and often presents as the first and leading disease sign. SAE occurs in an estimated 70% of patients with severe sepsis and has been identified as an independent risk factor for increased mortality.^5,6,7 Long-term sequelae of SAE include PTSD and marked cognitive deficits comparable to those found in patients with initial-onset Alzheimer’s dementia.⁸

The mechanisms of SAE are likely multifactorial. Postmortem findings in patients and in animal models provided evidence for severe systemic inflammation, blood-brain barrier disruption,^9,10 and activation of intracerebral immune cells, e.g., microglia and hematogenous macrophages.^11,12,13,14 This inflammatory milieu in the brain leads to the release of cytokines,^15,16 chemokines,¹⁷ oxygen radicals,¹⁸ and reactive nitrogen intermediates.¹⁹ In addition, ischemia,²⁰ hemorrhages,²¹ abnormal alterations in cerebral homeostasis, and dysregulation of cholinergic neurotransmission²² may all contribute to the development of SAE (for reviews see Gofton and Young,^7,23,24 Chung et al.,^7,23,24 and Mazeraud et al.^7,23,24), yet the neuronal pathophysiology underlying SAE and cognitive decline after systemic infection is only incompletely understood. Previous studies in a cecum ligation and puncture (CLP) mouse model of systemic infection and after lipopolysaccharide (LPS)-induced endotoxemia uncovered impairment of neuronal plasticity in an acute setting of 24 h and after 7 days.^25,26,27,28 These effects have been attributed to acute release of proinflammatory cytokines, e.g., interleukin-1β-mediated and brain-derived neurotrophic factor (BDNF)-mediated processes. However, it is unclear how severe systemic inflammation may cause persistent synaptic dysfunction that may finally lead to long-lasting defective plasticity and neurocognitive defects. To date, no specific treatment options have been developed.

Using a multidisciplinary approach, we here focused on the synaptic pathophysiology following severe systemic infection in a mouse model closely reflecting human peritoneal polymicrobial sepsis. In the acute stage and after full recovery from acute inflammation, we identified severely impaired synaptic pathways underlying disordered plasticity. These findings enabled us to establish therapeutic rescue strategies that successfully ameliorated neuronal functionality and overall disease signs. Targeting defective regulatory mechanisms of synaptic plasticity offers conceptual ways toward translational treatment strategies for SAE.

Results

Sepsis causes neurocognitive dysfunction and anxiety, defective synaptic plasticity, and synaptic scaling in excitatory synapses

We induced severe experimental sepsis in a total of 769 male C57BL/6J mice by polymicrobial peritonitis after intraperitoneal injection of a standardized and microbiologically validated human feces material (peritoneal contamination and infection model [PCI]; Figure 1A).²⁹ Mice showed a median disease severity of 12.3 on the cumulative 5-day PCI Clinical Severity Score (CSS), and the overall survival rate was 0.38 after 8 weeks (Figure 1B). For further experiments, we only considered surviving mice with severe sepsis (n = 309 PCI), which reached a cumulative 5-day CSS of ≥11.5. At a late stage, 8 weeks after PCI induction, the physically recovered mice still showed delayed learning and defective memory performance, as revealed by the Barnes maze test (BM), compared to saline-injected (SHAM) control mice. Deficits were present over the complete learning phase in the BM (Figure 1C, left), and memory dysfunction persisted also into the recall phase (BM probe trial) on day 8 (Figure 1C, right). Moreover, mice had increased anxiety-like behavior in the elevated plus maze test (EPM, Figure 1D) and in the open field (OF, Figure 1E). Since the overall phenotype was already normalized at week 8 after PCI as indicated by recovered body weight and regular locomotor activity, the behavioral abnormalities of defective memory and increased anxiety (Figures 1C–1E) can be attributed to neuronal dysfunction and not to post-sepsis-related general sickness.

Figure 1.

Experimental sepsis causes neurocognitive dysfunction and anxiety, defective synaptic plasticity, and synaptic scaling in excitatory synapses

(A) Experimental schematic. Acute phase of PCI is marked in red. Antibiotics (ATB) are delivered intraperitoneally until day 10, and behavioral testing (open field, OF; elevated plus maze, EPM; Barnes maze, BM) and ex vivo electrophysiological recordings (long-term potentiation, LTP; patch-clamp recordings, PC) are performed at the indicated time points. (B) (Left) Kaplan-Meier survival curve of all mice used in PCI experiments regardless of final experimental readouts. Mice intentionally taken out of the study, e.g., for experiments at early time points, were included up to the time when they were censored. (Right) Box plots and distribution of cumulative CSS on day 3 and day 5 after PCI (each circle represents an individual mouse after PCI). Note that subsets of PCI and SHAM mice were used in different experimental readouts in the study (detailed in the individual experiments). (C) In the BM, PCI mice (n = 30, red) show affected learning over the whole experimental period (two-tailed curve-permutation test) and a reduced memory recall in the probe trial (right; example tracking traces are shown, target quadrant marked in gray; SHAM mice n = 32, blue). (D) PCI mice spend less time in the EPM open arms (light gray). The locomotor activity as measured by total distance traveled within the EPM is unchanged (n = 12/7 SHAM/PCI). (E) PCI mice spent reduced time in the OF center (dark gray), whereas the distance traveled within the OF as a measure for locomotor activity was unchanged (n = 20/15 SHAM/PCI). (F) Long-term potentiation (LTP) in the hippocampal Schaffer collateral (SC)-CA1 pathway is severely impaired in PCI mice (n = 15/12 slices from 7/6 mice); two-tailed curve-permutation test, for time >0 min two-way repeated-measures ANOVA (F(1, 25) = 9,994), p = 0.0041). Two-tailed Student’s t test in the stable phase (min 25–30): p = 0.0123. Inset shows average traces of example recordings before and after theta burst stimulation. (G) Density of total synaptic spines and of mature mushroom spines in apical dendrites of CA1 neurons is reduced after PCI (n = 42/33 neurons). Scale bar, 3 μm. (H and I) Patch-clamp recordings of dentate gyrus granule cells indicates reduced frequency of miniature ESPCs (mESPCs) and spontaneous ESPCs (sEPSCs). (J) Peak amplitude of minimally evoked ESPCs (eEPSCs) is increased. PCI: n = 11/13/13, SHAM: n = 11/13/16 for mEPSCs, sEPSCs, and eEPSCs, respectively. Example traces in (H) and (I); average traces of minimal evoked eEPSCs in (H). Bar graphs show mean ± SEM; two-tailed Student’s t test, unless otherwise indicated.

To identify the underlying pathophysiology of cognitive dysfunction, we tested synaptic plasticity at the Schaffer collateral (SC)-CA1 synapse in the hippocampus by recording local field excitatory postsynaptic potentials (fEPSPs). We found a severely reduced long-term potentiation (LTP) at a late time point even at week 10 after PCI induction (Figure 1F), while the baseline single-stimuli fEPSC and short-term plasticity as measured by paired-pulse stimulation was unchanged after PCI (Figure S1). We additionally quantified the density and shape of synaptic spines in apical dendrites of hippocampal CA1 pyramidal neurons, providing morphological information on plasticity. Whereas the gross morphology of the hippocampus was not affected (Figure S2), the overall spine density and, more specifically, mature mushroom spines were reduced, thereby corroborating the findings of defective LTP (Figure 1G). Together, these data indicate affected long-term synaptic plasticity as the potential basis for persistent cognitive decline in a very late phase after surviving sepsis and physical recovery.

To elucidate the cellular and synaptic basis of persisting neuronal dysfunction, we next investigated synaptic transmission in the hippocampus from week 10 after PCI by whole-cell patch-clamp recordings in dentate gyrus granule cells. We found a reduced frequency of quantal excitatory postsynaptic currents (miniature ESPCs [mEPSCs], Figure 1H) and spontaneous EPSCs (sEPSCs, Figure 1I), which may point to an overall reduction of excitatory presynaptic terminals. Strikingly, peak amplitudes of mEPSCs, sEPSCs, and also of minimally evoked EPSCs (i.e., eEPSCs, arising from individual synapses after stimulation of the medial perforant pathway) were increased (Figures 1H–1J). EPSC rise-time and decay-time kinetics remained unchanged (Figure S3). These data thus provide important insights into synaptic regulation of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor expression and signaling following systemic inflammation. This might reflect increased recruitment of AMPA receptors to postsynaptic sites, thus suggesting a homeostatic scaling mechanism of inducing gain for correction of defective plasticity.

PCI leads to distinct and time-dependent changes in neuronal pathways relevant for synaptic signaling

To obtain mechanistic insights into the molecular mechanism of sepsis-induced neuronal dysfunction, we next investigated the transcriptional profile of brain tissue of acute septic PCI mice (day 3 after PCI) and of mice at week 10 after PCI compared to matched SHAM control mice. We found a differently regulated brain gene expression early after sepsis (Figures 2 and S4). Gene ontology (GO) term enrichment analysis of differentially expressed genes (DEGs) at day 3 revealed that transcripts related to neuronal function were markedly downregulated (Figure 3, light green). As expected, genes related to inflammation (Figure 3, red) and genes that promote enhanced transcription (Figure 3, blue) were highly upregulated. In contrast, in the late stage after sepsis, inflammatory pathways in the brain and the initially increased gene transcription pathways were no longer activated. Intriguingly, genes associated with neuronal function were then upregulated, indicating compensatory or maladaptive regulation once severe systemic inflammation had subsided. One of the most prominently downregulated gene was Arc, encoding activity-regulated cytoskeleton-associated protein ARC/ARG3.1 (ARC), a master regulator of synaptic plasticity³⁰ (Figures 4A and 4B). The protein interaction network of corresponding and differently expressed genes on day 3 after PCI revealed downregulation of genes encoding close ARC interaction partners, thus predicting an important role in dysfunctional synaptic plasticity and signaling (Figure 4A). Genes of the (1) early growth response protein (Egr) family, i.e., transcription-regulatory factors involved in neuronal plasticity,³¹ (2) Dusp6 encoding the ERK-directed dual-specificity phosphatase 6, (3) Nrgn encoding Neurogranin/RC3, a calmodulin-binding protein involved in CaMKII dependent synaptic plasticity,³² and (4) Mapk11, encoding p38 mitogen-activated protein kinase 11, were all downregulated at the early stage (Figures 4A and 4B). At the protein level and supporting our interpretation of sustained synaptic dysfunction after PCI, we indeed found a reduction of ARC and of phosphorylated ERK, an upstream mitogen-activated protein kinase (MAPK) relevant for transcription of immediate-early genes such as Arc,³³ at the early stage but also 10 weeks after PCI (Figures 4C, 4D, and S5). Together, these data indicate that severe neuroinflammation accompanies alterations of synaptic regulatory pathways in early stages, whereas sustained maladaptive regulation of neuronal pathways appears in late stages. Impaired neuronal functionality as shown by our electrophysiological recordings converges on these identified activity-regulated pathways that are relevant for synaptic plasticity.

Figure 2.

Hierarchically clustered heatmap of DEGs in a murine sepsis model

Expression of 144 DEGs (log₂FC > 1) in brain tissue on 3 days and 10 weeks after PCI with (color-coded) normalized Z-score visualization. Biological replicates from 3 days after PCI show a distinct and clustered expression profile (four lanes on the right).

Figure 3.

PCI leads to distinct and time-dependent changes in gene expression in the brain

Gene ontology (GO) terms enriched in 144 DEGs (log₂FC > 1) found by microarray analysis of brain tissue at 3 days and 10 weeks after PCI. DEGs were interrogated for enrichment in the gene sets of the GO definitions employing the piano package. Significantly enriched gene sets were clustered according to the score values of the five piano criteria distinct directional (up and down), mixed directional (up and down), and non-directional.

Figure 4.

PCI induces downregulation of neuronal pathways relevant for synaptic plasticity

(A) Interaction network of proteins corresponding to DEGs on day 3 after PCI. The node color depicts the respective log₂FC from microarray data. Proteins related to ARC synaptic signaling are enlarged for better visualization. (B) Expression of Arc and the six highlighted DEGs (in A) that are related to ARC signaling and synaptic function. Three days SHAM and PCI and 10 weeks PCI n = 4, 10 weeks SHAM n = 3. Box plots show median and 75^th to 25^th interquartile range, whiskers show respective maximum or minimum values within 1.5 times of the interquartile range; two-tailed Student’s t test. (C and D) Quantitative analysis of hippocampal ARC protein and pERK/ERK protein ratio (n = 8, each; normalized to SHAM ARC or SHAM pERK/ERK ratio, respectively) shows downregulation of protein signaling pathways in the network shown in (A). Bar graphs show mean ± SEM; two-tailed Student’s t test.

Rescue strategies of impaired synaptic signaling and cognition

Based on these findings, we applied two interventional strategies with the intention to test pathophysiological causalities and to rescue defective synaptic signaling and cognitive dysfunction after PCI. First, we specifically targeted the PCI-induced downregulation of the synaptic master regulator ARC by overexpressing Arc using a neuron-specific adeno-associated virus (AAV). Second, in a completely different and non-invasive systemic behavior-modulating paradigm, we subjected mice after polymicrobial sepsis to enriched environment (EE), which is known to have beneficial effects for cognitive performance by enhancing synaptic plasticity. The underlying mechanisms for these effects are thought to be triggered by upregulation of activity-regulated genes and pathways.^34,35

Overexpression of Arc rescues sepsis-induced cognitive deficits and impaired long-term potentiation

We found above that AMPA-receptor-mediated EPSC peak amplitudes increase in mice after PCI (Figures 1H–1J), which is consistent with a deficiency of ARC leading to upscaling of AMPA-receptor expression.³⁶ To increase the PCI-induced synaptic downregulation of ARC levels already at basal and not activity-dependent states, we constitutively overexpressed Arc in the hippocampus of post-septic mice. Here, we used an AAV containing a neuron-specific bicistronic plasmid of the Arc and Venus transgene under the synapsin promoter (Arc-AAV; Figure 5A). Bilateral in vivo injection of Arc-AAV into three injection sites of the hippocampus (dentate gyrus, CA1 region, and CA3 region, see also materials and methods) led to a robust, constitutively driven and not activity-dependent upregulation of ARC in hippocampal neurons (Figures 5B and S6) without AAV-induced activation of microglia and astrocytes (Figure S7).

Figure 5.

Hippocampal overexpression of Arc rescues inflammation-induced cognitive dysfunction and defective long-term potentiation

(A) Experimental schedule indicating ARC overexpression in the hippocampus from day 10 after PCI. Stereotactic intrahippocampal microinjection of an adeno-associated virus (AAV) containing the bicistronic vector for ARC overexpression and Venus fluorescence under the synapsin promoter (bottom left). The injection coordinates are depicted on the bottom right. (B) Representative example of constitutive ARC overexpression in CA3 region of the hippocampus. Scale bars, 40 μm. (C) PCI mice receiving control AAV (PCI CTRL AAV; n = 17) show memory deficits in the Barnes maze compared to SHAM CTRL AAV (n = 18) as indicated by reduced time in the target zone in the probe trial. ARC overexpression in PCI mice (PCI ARC AAV; n = 18) results in rescue of learning and memory dysfunction, whereas memory after ARC overexpression in SHAM mice is unchanged (SHAM ARC AAV; n = 19). Example traces indicate tracks of individual mice during the probe trial; the target quadrant containing the open hole is marked in gray. Two-way ANOVA with Holm-Sidak post hoc test (F values: sepsis (1, 68) = 4.539, p = 0.368; ARC AAV (1, 68) = 0.887, p = 0.349; interaction (1, 68) = 8.791, p = 0.0042). (D) LTP as measured by slope of excitatory postsynaptic potentials (EPSPs) in single-cell whole-cell recordings in CA1 neurons after stimulation of Schaffer collaterals. Paired stimulation results in LTP in SHAM CTRL AAV (n = 11 slices from 4 mice) and PCI ARC AAV (n = 8 slices from 3 mice) mice compared to PCI mice with control virus injection (PCI CTRL AAV; n = 9 slices from 4 mice) and SHAM mice after ARC overexpression (SHAM ARC AAV; n = 11 from 5 mice), which show reduced LTP. Upper panel: average traces (blue: SHAM CTRL AAV; light blue: SHAM ARC AAV; red: PCI CTRL AAV; light red: PCI ARC AAV). Middle panel: LTP time course. Two-tailed curve-permutation tests with Benjamini-Hochberg correction applied on p values, for time >0 min. Lower panel: comparison of LTP at the indicated time points after theta burst stimulation. Two-way repeated-measures ANOVA with Tukey’s post hoc test (F values: 1^st 10 min: sepsis (1, 35) = 0.7281, p = 0.399; ARC AAV (1, 35) = 0.1733, p = 0.06797; interaction (1, 35) = 10.03, p = 0.0032; 3^rd 10 min: sepsis (1, 35) = 1.224, p = 0.2761; ARC AAV (1, 35) = 0.1566, p = 0.6947; interaction (1, 35) = 21.93, p < 0.0001). All graphs show mean ± SEM.

PCI surviving mice that received injection of the empty control vector expressing only fluorescent Venus protein again showed learning and memory dysfunction in the BM 8 weeks after PCI (Figure 5C). In contrast, hippocampal Arc overexpression in PCI mice induced improvement of spatial learning and memory recall (Figure 5C). As expected, Arc overexpression in SHAM mice did not improve cognitive function. Locomotor function was not affected in all experimental groups (Figure S8A). Fear-related behavior as measured in the EPM was present in PCI mice and did not improve upon Arc overexpression in the hippocampus, since it is primarily mediated by signaling pathways in the amygdala (Figure S8B).

Corroborating the behavioral findings, hippocampal LTP of PCI mice was rescued after Arc-AAV injection to almost normal values when compared to SHAM mice having received injections of control-AAV (Figure 5D). Notably, we applied single-cell LTP measurements using whole-cell patch-clamp recordings instead of LTP measurements by local field potential measurements to more specifically address these molecular regulations influencing LTP at the cellular level (Figure 5D and materials and methods). According to previous studies showing an inverse regulation of synaptic AMPA-receptor expression by synaptic scaling mechanisms after Arc overexpression,^37,38,39 we expected a diminished LTP in SHAM mice after Arc overexpression. Indeed, we found reduced LTP in the Schaffer collateral-CA1 pathway in single-cell recordings of SHAM mice after Arc-AAV injection (Figure 5D).

These data suggest that increasing ARC protein levels in the hippocampus may effectively improve sepsis-induced neuronal dysfunction and defective synaptic plasticity mechanisms, shown to be relevant for cognitive functions.

Enriched environment improves synaptic pathology and cognitive dysfunction induced by PCI

Having established that activity-induced synaptic signaling genes and proteins are markedly downregulated after PCI, we then used a second, completely different paradigm to address the hypothesis that cognitive dysfunction in post-sepsis mice may also benefit from stimulus-induced general activation. We therefore utilized an established EE paradigm³⁴ (Figure S9A) to investigate its rescuing potential for defective synaptic plasticity and memory dysfunction at late stage after PCI (Figure 6A). As a first step, we again reproduced PCI-induced long-term cognitive dysfunction, anxiety-related behavior, and defective LTP in a separate cohort of mice without special housing conditions (PCI versus SHAM non-enriched [NE]; Figures 6B–6D). Indeed, when PCI mice were subjected to 6 weeks of EE housing conditions (Figure S9A), they showed improved performance in the BM test 8 weeks after PCI and reduced anxiety-related behavior in the EPM (Figures 6B and 6C). Possible confounders such as PCI-induced sickness with locomotor hypoactivity were no longer present in any of the surviving mice in these experimental groups at 8 weeks after PCI (Figure S9B). Synaptic plasticity as measured by hippocampal LTP was equally preserved, and reduction of synaptic mushroom spines was no longer present in PCI mice following this EE paradigm (Figures 6D and 6E).

Figure 6.

Enriched environment leads to improvement of synaptic pathology and cognitive dysfunction induced by PCI

(A) Mice were subjected to enriched environment (EE) or standard housing (NE) from day 10 after PCI. Behavioral tests and LTP recordings were performed as indicated. (B) EE improves cognitive dysfunction in the BM after PCI (PCI EE, red framed bars) to control levels (SHAM EE, blue framed bars; SHAM NE, filled blue bars) as indicated by increased time in the BM target zone during the probe trial compared to PCI mice in standard care (PCI NE, filled red bars). Example traces indicate tracks of individual mice during the probe trial; target quadrant is marked in gray. Two-way ANOVA with Holm-Sidak post hoc test (F values: sepsis (1, 57) = 9.510, p = 0.0031; cage type (1, 57) = 3.811, p = 0.0558; interaction (1, 57) = 3.968, p = 0.0512.) (C) EE leads to reduced anxiety-related behavior in the EPM after PCI as shown by increased time in the open arms and reduced time in the closed arms. Example traces indicate tracks of individual mice in the EPM (n = 14/18/13/16; SHAM NE/SHAM EE/PCI NE/PCI EE in B and C). Two-way repeated-measures ANOVA with Tukey’s post hoc test (F values: open arm: sepsis (1, 53) = 5.835, p = 0.0192; EE (1, 53) = 2.940, p = 0.0922; interaction (1, 53) = 2.311, p = 0.1344. Closed arm: sepsis (1, 53) = 5.143, p = 0.0274; EE (1, 53) = 9.816, p = 0.0028; interaction (1, 53) = 0.6826, p = 0.4124). (D) LTP as evaluated by field potential recording in CA1 is improved in PCI mice after EE. Upper panel: average traces before (baseline traces) and after theta burst stimulation (SHAM EE, light blue; PCI EE, light red; amplitudes of the baseline traces of each group were scaled to the amplitude of the SHAM NE trace; dark blue). Middle panel: comparison of LTP at the indicated time points after theta burst stimulation (n slices/N mice: SHAM NE: 8/3; SHAM EE: 7/3; PCI NE: 10/4; PCI EE: 6/3). Two-way repeated-measures ANOVA with Tukey’s post hoc test (F values: time point (2, 54) = 3.033, p = 0.0564; group (3, 27) = 5.473, p = 0.0045). Bottom panel: LTP time course. Two-tailed curve-shuffling tests with Benjamini-Hochberg correction applied on p values for time >0 min. (E) Quantification of synaptic mushroom spines in apical dendrites of CA1 neurons. Example images are provided (scale bar, 5 μm). PCI leads to a reduction of mature mushroom spine density (see PCI NE), which is rescued to basal levels (as in SHAM NE) after enriched environment (see PCI EE); n = 39/18/33/33 for SHAM NE/SHAM EE/PCI NE/PCI EE. Two-way ANOVA with Tukey’s post hoc test (F values: sepsis (1, 119) = 17.94, p < 0.0001; EE (1, 119) = 0.0016; interaction (1, 119) = 0.413, p = 0.5214). All graphs show mean ± SEM.

Since we could provide evidence that ARC plays a central role in the development of SAE (Figures 4 and 5), we next tested whether ARC expression is influenced by the EE paradigm. Indeed, ARC expression and the number of ARC-positive cells were increased in the hippocampus following EE in PCI mice (Figures 7A–7C). Furthermore, levels of BDNF, a growth factor inducing ARC expression via the trkB receptor, were elevated in mouse brain samples after EE (Figure 7D).⁴⁰ Together, these data imply that the beneficial effect of EE may be at least partially mediated by activation of the BDNF/TrkB signaling pathway, which in turn augments and rescues PCI-induced ARC expression.

Figure 7.

ARC and BDNF levels are increased after enriched environment

(A) Immunostaining for ARC in brain slices of post-septic mice with and without EE. Scale bar, 50 μm. (B) Increased density of ARC-positive granule cells in mice after EE. n = 7/8, two-tailed Student’s t test. (C) Arc expression increases in CA1 and CA3 pyramidal cell layer as measured from fluorescent intensity. n = 7/8, Mann-Whitney U test. (D) BDNF levels are increased in mouse brain tissue in SHAM and PCI mice after EE; n = 11/10/14/15 for SHAM NE/SHAM EE/PCI NE/PCI EE, pairwise permutation test with Benjamini-Hochberg correction. Bar graphs show mean ± SEM.

Discussion

From our findings, we can define three major aspects that may help us to understand synaptic pathology and neuropsychiatric sequelae after severe inflammation. Any of these may serve as a target for rescue strategies of impaired synaptic function and may eventually help to improve inflammation-induced neuropsychiatric symptoms. First, we identified defective synaptic plasticity as a crucial pathogenic mechanism of long-term cognitive dysfunction in SAE. Importantly, extending previous reports,^25,26,27,28 we showed that defective plasticity persists also at very late stages beyond acute brain inflammation. More specifically, we uncovered distinct changes in the regulation of important synaptic modulating proteins leading to abnormal basal synaptic transmission and synaptic scaling likely underlying defective LTP. These synaptic changes developed in early stages of SAE in parallel with severe sepsis-induced neuroinflammation and persisted even when the acute inflammation had subsided. Second, the constitutive overexpression of severely downregulated ARC, a prominent mediator of synaptic signaling, appears as a crucial and effective mechanism to largely rescue sepsis-induced synaptic and brain dysfunction. Third, SAE-induced synaptic changes and memory dysfunction could also be restored by non-invasive activation paradigms such as EE in a severe polymicrobial sepsis model. These effects may be at least partially mediated by a BDNF-mediated increase in ARC expression, thus corroborating the central role of the ARC signaling pathway in the development of SAE.

Direct interactions of proinflammatory mediators with synaptic function and excitatory signaling are well established. Endogenous tumor necrosis factor α (TNF-α) has been shown to enhance the membrane insertion of synaptic AMPA receptors, leading to increased EPSC amplitude⁴¹ and synaptic scaling. Moreover, proinflammatory cytokines are able to inhibit LTP via disturbed MAPK signaling,^42,43 and increased cytokine levels are associated with cognitive decline.⁴⁴ Here, we identified dysregulation of ARC together with its close interaction partners and the MAPK downstream mediator ERK to probably underlie the dysregulation of excitatory synaptic signaling. These data are consistent with previous reports showing the involvement of BDNF signaling in disturbed LTP early after CLP.²⁸ Accordingly, after EE we found increased BDNF levels that eventually contribute to the rescue and upregulation of ARC after SAE. In summary, we provide complementary evidence that disrupted synaptic ARC signaling and impaired long-term plasticity underlie hippocampal and neuronal network pathology during SAE. Our findings also clearly demonstrate that these signaling processes and synaptic dysfunction persist even after signs of inflammation and sickness have subsided, thus demonstrating enduring neuronal dysfunction after the initial trigger of systemic inflammation. Importantly, we also show that different intervention measures that converge on rescuing ARC-dependent synaptic signaling are sufficient to improve neurocognitive dysfunction during SAE.

Arc is an immediate-early gene (IEG), and its transcription and translation are directly regulated by neuronal activity and neurotrophins, e.g., BDNF signaling.⁴⁵ ARC is abundantly expressed in cell bodies and dendrites of CamKII-positive excitatory neurons in the central nervous system (CNS).⁴⁶ Although recent studies suggest that ARC is not indispensable for LTP,⁴⁷ it is thought to be critical for memory formation and an important bidirectional regulator of long-term synaptic plasticity including LTP, long-term depression (LTD), heterosynaptic LTD, and homeostatic synaptic scaling.³⁰ ARC mediates activity-regulated endocytosis of AMPA receptors at postsynaptic sites by interaction with the endocytic machinery and PSD-95 complexes.^39,48,49 Indeed, memory formation and consolidation of LTP are impaired in Arc-knockout mice and after Arc silencing using antisense oligonucleotides.^50,51 Furthermore, ARC regulates synaptic AMPA-receptor expression inversely based on synaptic activity, thus representing the key mechanism of homeostatic plasticity.^52,53 In our PCI model, we identified substantially reduced ARC protein levels not only at early but also at late stages already under basal conditions without additional external stimulation. In late stages, Arc transcripts were normalized, thus pointing toward reduced local translation after PCI.⁵⁴ Accordingly, as one might expect from Arc-knockout studies, we found severe alterations with increased amplitudes of AMPA-receptor currents in basal conditions and defective LTP. Our experiments in mice with constitutive overexpression of Arc therefore aimed to rescue the sepsis-induced basal and persistent downregulation of ARC. This strategy was chosen following the observations by others that Arc overexpression in cell culture and in organotypic brain slices results in blockade of homeostatic upregulation of AMPA receptors and reduction of EPSC amplitude.^36,38 Moreover, recent findings suggest that ARC is capable of forming capsid-like structures and is able to transduce and affect the physiology of neighboring cells in vitro and in vivo.^55,56 Therefore, it is possible that the interneuronal trafficking of exogenous ARC leads to spreading effects starting from the initial AAV injection sites. Our present findings related to overexpression of Arc in vivo show that this approach was successful in rescuing PCI-induced hippocampal dysfunction, thus providing evidence for a successful strategic intervention in a crucial pathway of disturbed synaptic signaling.

Next, we questioned whether a completely different and non-invasive approach based on general activation, amenable to upregulate activity-dependent pathways, can induce beneficial effects on synaptic plasticity and memory in SAE. Utilizing an established EE paradigm in a comparative cohort, we were able to counteract inflammation-induced impairments of synaptic plasticity and memory dysfunction. EE has been proven to be effective in various models of neurodegenerative disorders, e.g., Alzheimer’s disease, Huntington’s disease, and others (for review see Nithianantharajah and Hannan³⁵) and also in an LPS endotoxemia model.⁵⁷ In these models, EE results in enhanced learning and memory and delayed disease progression.⁵⁸ Among a variety of mechanisms, EE can enhance adult neurogenesis and network incorporation of newborn neurons,⁵⁹ increase the expression of signaling molecules, induce synaptic plasticity pathways,⁶⁰ and increase the number of dendritic spines, finally leading to more efficient operation of neuronal networks.⁶¹ These changes are triggered by increased physical activity, social interaction, and sensory stimulation to various extents depending on the exact paradigm chosen.⁶² The underlying mechanisms are mainly mediated by increased neurotrophic signaling, in particular of BDNF via the trkB receptor and downstream activation of MAPK, e.g., the ERK pathway,⁴⁰ and the pathways that are affected after systemic inflammation as shown here and by others.^28,57 The increased BDNF levels observed here likely contribute to the positive effect of EE in this study. As the total increase of the BDNF levels was moderate compared to other studies using different paradigms of EE,⁴⁰ additional mechanisms, e.g., direct upregulation of Arc⁶³ and other IEGs, e.g., Egr 1, 2, and 4 and Dusp 6 in the hippocampus, may also occur.^58,64,65 Previous reports also discussed epigenetic mechanisms and enhanced expression of vasopressin possibly contributing to the positive effects of EE in systemic infection models.^66,67 Lastly, EE increases synaptogenesis and fosters synaptic plasticity like LTP,⁶⁸ which collectively may explain the beneficial effects of EE on cognitive function observed in our present study. Taken together, EE apparently acts via multiple synergistic mechanisms, and the effects on BDNF signaling, IEGs, and the MAPK pathway are plausible candidates to explain the beneficial impact on synaptic function, plasticity, and memory after severe systemic inflammation.

Several studies reported the effects of severe systemic inflammation on the immune mechanism in the brain in acute and subacute animal models.²⁸ Systemic application of LPS results in early inflammation-associated alterations and activation of cerebral endothelial cells.¹⁰ This initial event is followed by microglia activation, thus leading to an inflammatory state in the CNS characterized by increased production and release of proinflammatory cytokines and chemokines.^15,69 These include TNF-α and interleukin-1β among others. In parallel, anti-inflammatory mediators, e.g., lipocalin2, are induced to counteract LPS-induced neuroinflammation.^16,70,71 In these acute and subacute models of systemic LPS challenge and in other acute infectious rodent models, e.g., the CLP model, neuronal plasticity was shown to be impaired, and similar proinflammatory profiles have been identified.^25,26,27,28 However, systemic LPS challenge may not completely resemble polymicrobial sepsis.⁷² We suggest that the murine model, which we investigated here, may well reflect residual and chronic brain dysfunction in patients surviving severe inflammation.

Very recently, we and others could show that microglia mediate synaptic pruning after C1q tagging and HMGB1 activation, which finally results in synaptic loss.^73,74,75 These alterations were observed at early time points in polymicrobial sepsis models and after LPS challenge but induce behavioral abnormalities also in late stages after experimental sepsis. The structural changes of excitatory synapses may explain the reduction of sEPSC frequency observed here and likely contribute to the compensatory synaptic scaling mechanisms and functional synaptic alterations.

Previous studies reported a time-dependent improvement of cognitive dysfunction in rats after CLP, another rodent model reflecting polymicrobial sepsis.⁷⁶ In our study, we showed memory dysfunction in several independent experimental groups. We cannot rule out that cognitive dysfunction is even more pronounced in earlier stages after PCI, but this might always be superimposed by the general sickness induced by PCI. For this reason, we performed behavioral tests in surviving mice at a late time point (8 weeks) when systemic signs of sepsis had largely subsided and found that these cognitive and anxiety-related abnormalities are still present.

From a translational perspective, SAE-induced neuropsychiatric dysfunction becomes increasingly relevant. Due to improved intensive care medicine, patients more often survive severe systemic infections and sepsis. The majority of these patients are afflicted with SAE. Studies involving large patient cohorts provided robust evidence for the increased risk of developing dementia in survivors of severe systemic inflammation.² Postinfectious long-term cognitive dysfunction causes not only major disability of afflicted patients but also an immense burden on primary caregivers and the health care system.¹ The high clinical impact and socioeconomic importance warrants a better understanding of SAE pathophysiology with the ultimate development of target-directed therapeutic strategies.⁷⁷

Together, our findings may help to elucidate SAE-induced neuronal pathomechanisms at the molecular and cellular levels. Targeting synaptic pathways offers conceptual advances for translational treatment strategies in SAE. In practical terms, a general activation scheme equivalent to the EE paradigm, e.g., by personalized cognitive training approaches early in the course of SAE, may become the first to be studied in prospective randomized and controlled clinical trials. Our findings may also encourage future research addressing regulatory pathways in patients with SAE by means of targeted intervention strategies. From animal models like the ones reported here, various strategies may become feasible such as increasing ARC levels, e.g., by stimulation of upstream pathways such as neurotrophin BDNF receptor/TrkB activation, ERK signaling, or CREB phosphorylation.^37,78,79,80 Either of these may be promising starting points for target-directed therapies in future pilot trials.

Materials and methods

Mouse model of experimental sepsis

All animal experiments were approved by the state authorities of Thuringia (UKJ-02-085-14) and were performed in accordance with animal welfare and the ARRIVE guidelines.⁸¹ Experimental sepsis was induced by the established PCI model as described previously.²⁹ PCI mice received an intraperitoneal injection of human fecal slurry (3 μL/g body weight). Control mice (SHAM) were injected with 0.9% saline solution. We exclusively used male C57Bl6J mice, as PCI disease severity varies in female mice due to hormonal fluctuations. Mice were closely monitored, and the CSS (grades 1–4) was assessed²⁹: grade 1: no signs of illness, active and curious, quick movements upon exogenous stimuli, normal posture; grade 2: low-grade illness, less active with occasional interruptions in activity, reduced alertness, but adequate response to exogenous stimuli, posture slightly hunched; grade 3: moderately severe illness, slow and sleepy, movement difficulty, limited and delayed reaction to exogenous stimuli, hunched posture; grade 4: severe illness, mouse lethargic, motionless, no spontaneous movements, no reaction to exogenous stimuli, severely hunched posture. Floating numbers between two grades are possible. Mice were euthanized when a CSS of grade 4 was reached at two consecutive time points within 3 h. To avoid premature death due to severe peritonitis, mice were treated with the β-lactam antibiotic meropenem (650 μL, 1 mg/mL) when the CSS reached 3. Subsequently, mice received meropenem twice daily for 7 days and once per day on the following 3 days. Thereafter, 2 mg/mL enrofloxacin (Baytril 2.5%; Bayer, Leverkusen, Germany) in sweetened drinking water was provided. Control animals were treated in the same way. For the sepsis groups, only animals with severe sepsis (cumulative 5-day CSS of ≥11.5) were included.

Stereotactic intrahippocampal microinjection

Stereotactic surgery for injection of AAVs into the hippocampus was performed as previously described.⁸² In brief, the skull of anesthetized mice (1.5%–2% isoflurane), head-fixed into a stereotactic apparatus (Lab Standard; Stoelting, Wood Dale, IL, USA), was exposed, and holes were drilled above the stereotactic coordinates. Injections of AAV were performed at three sites within the hippocampus to cover CA1, CA3, and dentate gyrus regions (Figure S6) using thin injection pipettes (resistance >20 MΩ) with coordinates with respect to bregma (AP/L/DV): 1^st (2.8/2.8/2.8 mm); 2^nd (2.1/1.4/1.4 mm); 3^rd (2.1/1.4/2.1 mm). One microliter of the AAV preparation (concentration 1 × 10¹⁰ genome copies/μL) in 0.9 M saline solution was injected (0.4 nL/s; Nanoliter 2000 + SYS-Micro4 Controller; WPI, Sarasota, FL, USA) into each injection site. After injections, the skin of the animals was sutured and mice were taken back to their home cage.

Behavioral analyses

Behavioral analyses were performed at 8 weeks after PCI or SHAM treatment when surviving mice had recovered from systemic disease signs. All tests were performed by a blinded investigator who was not involved in PCI induction and scoring in the acute phase of PCI.

Open field

OF testing was performed similarly to a previously described protocol.⁸³ A custom-made square-shaped box (50 × 50 cm) was used (illumination decreased from center to periphery, 200 lux to 120 lux). The arena was divided into a center zone (25 × 25 cm) and the surrounding periphery zone. Mice were allowed to explore the OF for 5 min while the animals were videotaped and tracked (EthoVision [Noldus, Wageningen, The Netherlands] or ANYmaze [Stoelting Europe, Dublin, Ireland]). After completion of a trial, the OF was cleaned with 70% ethanol.

Elevated plus maze

A custom-made EPM elevated 100 cm above floor level with two opposing open and two closed arms (31 × 5.8 cm) as well as an intersection (5,8 × 5.8 cm) was used. The closed and open arms and the intersection were illuminated with 40, 160, and 120 lux, respectively. Mice were placed in the intersection facing an open arm, and locomotion was tracked for 5 min. The EPM was cleaned using 70% ethanol after each trial to ensure equal experimental conditions.

Barnes maze

A circular platform (diameter 120 cm), mounted on a rotatable stand, elevated 100 cm above floor level, and consistently illuminated by two light sources (900–1,000 lux) was used.

The maze comprised 40 pseudo-randomly distributed and equally sized holes. During the habituation (day 0) and training phase (days 1–6), an escape box was mounted below one of the holes, while the other holes were closed. The location of the escape box was kept constant for each mouse but varied between mice. After each trial the maze was cleaned with 70% ethanol. In the habituation phase (day 0), mice were placed in a transparent plastic cylinder located in the center of the maze for 60 s. Thereafter, mice were allowed to explore the maze for 60 s and then gently led to the escape box. Mice were allowed to spend 2 min in the escape box. In the training phase (days 1–6), an opaque instead of transparent plastic cylinder was used. The cylinder was lifted after 20 s, and mice were videotaped and tracked (EthoVision) for 4 min or until they entered the escape box. In the testing phase (day 7), all holes were closed and the mice were videotaped for 60 s to assess the time spent in the target quadrant (previous location of the escape box).

Enriched environment

From day 10 after PCI or SHAM injection, mice were housed in an EE cage (6 animals/cage, 85 cm × 75 cm × 40 cm; groups PCI EE and SHAM EE) with different objects offering diverse stimuli (tunnels, different housings, nesting material, running wheels). The configuration of the objects was changed on a weekly basis⁸⁴ (see also Figure S9A). The respective non-stimulated control groups (PCI NE and SHAM NE) remained in the normal standard housing conditions over the entire experiment. (2 animals/cage; 54 cm × 38 cm × 19 cm).

Patch-clamp and local field potential recordings

Preparation of acute hippocampal slices

For fEPSP recordings, acute coronal hippocampal slices were prepared from adult mice sacrificed under isoflurane anesthesia by decapitation. Brains were dissected and sliced (400 μm in ice-cold high-sucrose solution: 20 in mM NaCl, 25 in mM NaHCO₃, 10 in mM glucose, 150 in mM sucrose, 4 in mM KCl, 1.25 in mM NaH₂PO₄, 0.5 in mM CaCl₂, 7 in mM MgCl₂). Slices were stored in artificial cerebrospinal fluid #1 solution (aCSF#1): 124 mM NaCl, 26 mM NaHCO₃, 10 mM glucose, 3.4 mM KCl, 1.2 mM NaH₂PO₄, 2 mM CaCl₂, 2 mM MgSO₄) at 32°C for 30 min and subsequently at room temperature.

For whole-cell patch-clamp recordings, previously published protocols were adapted.^85,86 Isoflurane-anesthetized mice were transcardially perfused with 95 mM N-methyl-D-glucamine, 30 mM NaHCO₃, 20 mM HEPES, 25 mM glucose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 2 mM thiourea, 5 mM Na-ascorbate, 10 mM MgSO₄, 0.5 mM CaCl₂, and 12 mM N-acetyl-L-cysteine. Brains were sliced and subsequently stored for 12 min in the same solution at 34°C. Thereafter, slices were transferred to aCSF#2 solution (125 mM NaCl, 25 mM NaHCO₃, 25 mM glucose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 2 mM CaCl₂, 1 mM MgCl₂ supplemented with 2 mM thiourea, 5 mM Na-ascorbate, 3 mM Na-pyruvate, 12 mM N-acetyl-L-cysteine). All solutions were continuously purged with carbogen (95% O₂, 5% CO₂), and the pH was maintained at 7.3.

Whole-cell patch-clamp recordings in the dentate gyrus

Whole-cell patch-clamp recordings were performed as described previously.⁸³ Slices were superfused with aCSF#2 at room temperature. Dentate gyrus granule cell were visually identified. Patch pipettes (3–4 MΩ) were filled with 115 mM potassium gluconate, 40 mM KCl, 10 mM HEPES, 0.1 mM EGTA, 2 mM Na₂ATP, and 2 mM MgCl₂. EPSCs were recorded in the presence of bicucillin and CGP 52432. For the measurement of miniature EPSCs, 1 μM tetrodotoxin was added. Minimal stimulation was applied at 0.3 Hz via an aCSF#2-filled glass pipette within the medial perforant path.⁸⁷

Analysis of long-term potentiation by recording fEPSPs

Measurements of fEPSPs at the SC-CA1 synapses were performed as previously described.⁸⁸ Slices were perfused with aCSF#1 at 2.5 mL/min at 32°C–33°C. Stimulation within SC was performed using custom-made bipolar platinum electrodes. aCSF#1-filled glass electrodes in the stratum radiatum were used to record the fEPSPs. Stimulation strength was varied between 25 and 400 μA to record the input-output curve. For subsequent recordings, the stimulation strength was adjusted by reducing the stimulation by 10% from the lowest strength evoking an fEPSP with a population spike. Stimuli were delivered at 0.03 Hz. Baseline was recorded for 15 min. LTP was induced by 20 theta bursts of 4 pulses of 100 Hz with an interval of 200 ms. After induction of LTP, fEPSPs were recorded for 30 min. The fEPSP slopes were measured directly after the fiber volley to the peak or until the beginning of a clearly distinguishable population spike.

Analysis of LTP by whole-cell patch-clamp recording in CA1 pyramidal neurons

Whole-cell patch-clamp recordings of CA1 pyramidal cells were performed in aCSF#2 with 100 μM picrotoxin. Glass pipettes were filled with 115 mM potassium gluconate, 20 mM KCl, 10 mM HEPES, 4 mM Mg-ATP, 0.3 mM Na-GTP, 10 mM Na-phosphocreatine, and 0.001 mM CaCl₂ (pH 7.4, 280–290 mosmol/L). SCs were stimulated using an aCSF#2-filled pipette and a stimulus duration of 70 μs. Input-output curves were measured, and stimulation strength was adjusted to evoke an EPSP of 30%–50% of the maximum EPSP. Baseline was recorded for 10 min (stimulation frequency 0.05 Hz). Spike-timing-dependent plasticity was induced by pairing an evoked EPSP with four action potentials induced by postsynaptic current injection (3 ms duration, 1 nA, 200 Hz) with 35 repetitions at 0.5 Hz, as previously reported.⁸⁹ Thereafter, EPSPs were recorded for further 30 min. EPSP slopes were calculated from the initial 2 ms after EPSP onset.

Data acquisition and data analysis of electrophysiological recordings

All data were obtained using an EPC10 amplifier controlled by Patchmaster software (HEKA, Lambrecht, Germany). mEPSCs were analyzed using Clampfit 10 (Molecular Devices, San Jose, CA, USA). All other data were analyzed using the NeuroMatic plugin (http://www.neuromatic.thinkrandom.com) of Igor Pro Software (Wavemetrics, Portland, OR, USA).

Generation of Arc-expressing constructs and of negative control adeno-associated viral constructs

To generate a negative control adeno-associated viral (pAAV) coding for murine Arc mRNA (indicated as ARC-AVV), the Arc cDNA was cut from a pCMV6-Entry_Arc vector (Origene, MR206218) and inserted into a p-AAV-Syn-iCre-2A-Venus vector.⁹⁰ As negative control adaptor, oligonucleotides were generated and subcloned into the p-AAV-Syn-iCre-2A-Venus vector (indicated as CTRL AVV). Sequences of constructs were verified by Sanger sequencing (Eurofins Genomics, Ebersberg, Germany).

Virus production, isolation, and quantification

AAV-293 cells (Cell Biolabs, San Diego, CA, USA) at a confluence of 70%–80% were transfected using a reaction mix containing pRV1 (AAV2), pH21 (AAV1), pFdelta6 (Ad Helper), and one of the pAAV constructs described above (pEGFP-N3) with a ratio of 1.5:1.5:6:3:1.5, respectively, together with 150 mM CaCl₂ and HEPES-buffered saline. The medium was replaced after 6 h. Seventy-two hours after transfection, cells were harvested in cold PBS and centrifuged at 800 × g for 10 min. Cells were lysed in a buffer containing 20 mM Tris, 159 mM NaCl (pH 8.0), 0.5% sodium deoxycholate, and 50 U/mL benzonase for 1 h at 37°C. After centrifugation, the AAV was isolated from the supernatant using heparin columns (GE Healthcare). Eluted virus was then concentrated using Amicon Ultra-4 centrifugal filter units (Merck-Millipore) and filtered.

For virus quantification, the virus load was determined using quantitative real-time PCR using an AAV plasmid construct with a known concentration as a standard, primers WPRE-F 5′-TGC TTC CCG TAT GGC TTT CAT-3′ and WPRE-R 5′-CAG CAA ACA CAG TGC ACA CC-3′, and SYBR Select Master Mix (Thermo Fisher Scientific, Rockford, IL, USA). The measurements were performed with a CFX384 instrument (BioRad, München, Germany), and analyses were done using CFX Manager software (BioRad).

Immunohistochemistry

Isoflurane-anesthetized mice were transcardially perfused with PBS followed by 4% paraformaldehyde (PFA). The brains were post-fixed in 4% PFA for 24 h at 4°C, cryoprotected in 10% and 30% sucrose in PBS, and stored at −80°C. Coronal 40-μm slices were prepared on a sledge microtome (Thermo Fisher Scientific) and blocked with Tris-buffered saline (TBS; 0.1% Triton X-100, 3% serum, 2% skimmed milk powder) for 30 min. The first antibody was incubated in blocking buffer overnight at 4°C (anti-ARC/Arg3.1 [#156 005], anti-NeuN [#266 004, 1:1,000], anti-GFP [#132 006, 1:500], all Synaptic Systems, Göttingen, Germany; anti-Iba1 [019-19741, 1:500], Wako Pure Chemical, Richmond, VA, USA; anti-glial fibrillary acidic protein [GFAP; #BNUM0789-50, 1:500], Biotium, Fremont, CA, USA). Slices were washed in TBS and incubated with second antibody for 2 h at room temperature (rhodamine@ guinea pig or rabbit [#706-296-148; #711-297-003, 1:500], Jackson ImmunoResearch; AF488@ guinea pig, chicken, or mouse [#A11073, #A11039, #A21202, 1:500], Invitrogen). Slices were mounted on gelatin-coated glass slides, stained with DAPI, and embedded using Mowiol. Analysis of GFAP and IBA were performed using Fiji. Maximum-intensity projections from four confocal images (BC43; Oxford Instruments Andor, Belfast, UK) within the hippocampus were taken for analysis. Cells were counted manually by a blinded investigator. For measurement of IBA staining density, a constant threshold was set and the staining intensity was measured.

Golgi staining and dendritic spine analyses of CA1 pyramidal neurons

Morphological spine analysis was performed in pyramidal neurons of the CA1 region as described previously.⁹¹ Golgi silver impregnation was conducted with the FD Rapid GolgiStain Kit (FD NeuroTechnologies, Columbia, MD, USA) according to manufacturer’s instructions. Hemispheres were stored at 80°C before slicing and mounting of 150-μm-thick coronal slices. The further staining procedure was performed according to manufacturer’s instructions. Slices were coverslipped with Entellan (Merck, Darmstadt, Germany). Dendritic morphology was reconstructed and analyzed using an integrated microscope system (Neurolucida; MBF Bioscience, Williston, VT, USA). Cells were selected for reconstruction if they fulfilled the following criteria: (1) located within the pyramidal cell layer of the CA1 region; (2) distinguishable from neighboring cells; (3) the dendrites were not truncated or broken; and (4) cells were well filled throughout whole dendrites. Third-order branches of 20–50 μm stretch, 50–150 μm away from the soma of putative CA1 pyramidal cells, were selected for analysis. Spines were classified as follows: 1, thin: shorter than 1.5 μm without swellings; 2, mushroom: spine head larger swelling than 0.6 μm; 3, stubby: length/width smaller than 1.0 μm without swelling. Number and type of spines (mushroom, stubby, thin) were counted manually by the same blinded investigator.

Western blot

Lysates were obtained using RIPA buffer supplemented with complete protease inhibitor kit and PhosSTOP inhibitor cocktail (both from Roche Diagnostics, Mannheim, Germany). Protein samples (30 μg per lane) were subjected to SDS-PAGE and transferred to an Amersham Hybond membrane (GE Healthcare Life Science, Marlborough, MA, USA). First antibodies were incubated overnight at 4°C (ARC/ARG3.1, 1:7,500, #156005; Synaptic Systems; GAPDH, 1:10,000, #2118L, Cell Signaling Technology, Danvers, MA, USA). Horseradish peroxidase-labeled secondary antibodies (1:1,000, Dianova, Hamburg, Germany) were incubated for 2 h at room temperature. Chemiluminescent signals (Pierce ECL Western Blotting Substrate, Thermo Fisher Scientific) were visualized using a luminescent image analyzer (LAS 3000; Fujifilm Life Science, Cambridge, MA). Densitometric analysis was performed using ImageJ software. Regarding ARC expression 10 weeks after the PCI, the summed signal of the ARC double band was used for analysis.

Capillary western immunoassay

Mouse hippocampi were homogenized in lysis buffer (0.32 M sucrose, 4 mM Tris-HCl [pH = 7.4], 1 mM EDTA, 0.25 mM dithiothreitol) with additional ultrasound sonication (5× pulses). For each target, the antibody dilution and dynamic range was determined independently to assure quantitative measurements by capillary western immunoassay (ProteinSimple, San Jose, CA, USA). Proteins (0.2 μg/μL) were separated in the 12–230 kDa detection module (ProteinSimple, #SM-W004) and detected in multiplex assays with α-tubulin (#2144S, 1:10), Erk (#4696, 1:10), and pErk (#4370, 1:15) antibodies, respectively (all from Cell Signaling Technology). Primary and secondary antibodies were incubated for 30 min each. The data from PCI cohorts were normalized to averaged SHAM levels. The normalized data were used for statistical analysis.

ELISA analysis of brain BDNF levels

Brain tissue samples were prepared as described above. BDNF levels were determined using the Total BDNF Quantikine ELISA Kit (#DBNT00; R&D Systems, Minneapolis, MN, USA) according to the manufacturer’s instructions. Technical duplicates were measured on a plate reader (Enspire 2300; PerkinElmer, Waltham, MA, USA), and BDNF concentration was calculated using a 4PL regression model.

Analysis of gene-expression and regulatory networks

RNA isolation

Total brain RNA extraction was carried out using QIAzol lysis reagent and RNeasy Mini Kit (Qiagen, Hilden, Germany) according to manufacturer’s instructions; quality control of isolated RNA was performed on a QIAxcel capillary electrophoresis system (Qiagen).Quantification of total RNA was performed on a Nano-Drop spectrophotometer ND-2000 (Thermo Fisher).

Microarray analysis

RNA transcription and amplification were performed using the TargetAmp-Nano Labeling Kit for Illumina Expression BeadChips (Biozym Scientific, Hessisch Oldendorf, Germany) on a thermal Cycler (BioRad) according to the manufacturer’s instructions. cRNA samples were purified with the NucleoSpin RNA Clean-up system (Macherey-Nagel, Düren, Germany) and quantified before hybridization. Samples were hybridized 20 h at 58°C on two Illumina MouseRef-8 v2.0 Expression BeadChips (Illumina, San Diego, CA, USA). Data readout, data preprocessing, and data analyses were performed with Illumina GenomeStudio software v.2011.1).

Microarray data were preprocessed by log₂ transformation and robust spline normalization (rsn) using the R lumi package.⁹² One sample (“9298740075_G”) was identified as an outlier and was removed (Figures S4A and S4B). DEGs were determined using false-discovery-rate/Benjamini-Hochberg correction and log₂(fold change) (log₂FC) ≥ 1 cutoff. Gene set enrichment analysis for enriched GO terms was performed using the piano package.⁹³

Microarray data were processed using the R environment (R Studio version 1.1.383, R version 3.6.1; RStudio, Boston, MA and R Core Team, Foundation for Statistical Computing, Vienna, Austria) with the packages reshape2⁹⁴ and ggplot2. Networks were built and analyzed with the aid of Cytoscape (version 3.8.0), Legend Creator (version 1.1.5), and stringApp (version 1.6.0).⁹⁵ Protein interactions were imported from the string database for the species Mus musculus with a confidence cutoff of ≥0.4.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 9 and MATLAB 2020a. Data represent the median or the mean ± SEM as indicated in the figure legends. Normality and variance homogeneity of data were checked using the Shapiro-Wilk test and F test with a significance level of 0.05, respectively. An unpaired, two-tailed Student’s t test or a two-way ANOVA (>2 groups) or equivalent non-parametric tests (for non-normal data) were used to evaluate statistical significance using p value less than 0.05 as the cutoff. Curve-permutation tests were performed by comparing the difference in the area-under-curve of the group-averaged curves, with those obtained after shuffling the individual curves across groups, with repetition 1 million times (see Ceanga et al.⁹⁶ for more details). For cases with >2 groups, the Benjamini-Hochberg procedure was used to correct the p values (two-tailed) obtained by curve-permutation tests for multiple-comparison problems. Details of the applied statistical tests (including the post hoc tests for ANOVA), F and p values) and the sample sizes are provided in figures or figure legends.

Data and code availability

The raw and preprocessed microarray data have been deposited in NCBI GEO DataSets: GSE167610. All further data are available in the main text or supplemental information.