Background
Neuronal rhythmic electrical activity in the gamma-frequency band (⁓30–80 Hz; gamma oscillations) underlies cognitive processes such as memory storage and recall and plays a role in sensory perception [
1‐
5]. This cognition-relevant brain rhythm emerges from the finely balanced and cooperative activity of particular neuronal populations within neuronal networks. As such, gamma oscillations depend on recurrent activity between excitatory pyramidal cells (PCs) [
6] and the pacing of their activity by fast-spiking perisomatic GABAergic interneurons that express parvalbumin [
7‐
9]. Inhibitory (GABAergic) activity has been related to cognitive processes, including working memory, formation and consolidation of fear conditioning, and thought suppression, indicating a crucial role of GABAergic interneurons [
10]. The synchronized interplay between PCs and fast-spiking interneurons (FSNs) generates and depends on the finely-patterned fluctuations of excitatory and inhibitory currents that give rise to oscillations of the local field potential [
11,
12]. The gamma rhythm is prominent in the hippocampus, where its disruption has been reported to be linked to Alzheimer’s disease (AD) [
13‐
16]. Particularly, reduced levels of GABA in different brain areas of AD patients, including the hipocampus, the cingulate cortex, the amygdala, and the frontal and temporal cortices have been found (see ref. [
10] for review).
AD is a neurodegenerative disease characterized by diverse cellular and molecular pathological events that progressively lead to cognitive impairment. Such pathological events mainly, but not exclusively, comprise abnormal extracellular accumulation of beta-amyloid peptide (Aβ42) [
17‐
19], intraneuronal depositions of abnormally phosphorylated tau-tangles [
20,
21], defective synaptic transmission, and synaptic and neuronal loss, leading to marked instability of neuronal network and deficient information processing [
22]. In this regard, studies performed in AD models suggest a dynamic nature of synaptic and intrinsic dysfunctions destabilizing spontaneous spiking activity of diverse neuronal classes in the hippocampus and cortex, leading to aberrant activity of cortico-hippocampal circuits [
22].
There is growing evidence that in AD, synaptic failure and impaired network performance far precede Aβ plaque deposition and clinical expression of cognitive decline [
23‐
25]. Interestingly, diverse studies suggest that neuroinflammation, and specifically myeloid cells, play a major role in the initiation and progression of AD [
26‐
30] with a negative impact on brain circuit functioning [
31,
32]. Recently, we found galectin-3 (gal3) to be a central regulator of microglial immune response in Parkinson’s disease [
33,
34] and in prolonging sustained microglial activation in AD [
27]. As such, microglia, the resident phagocytic immune myeloid cells of the central nervous system, are a recognized part of the cellular phase of AD [
35].
Microglial activation has been reported to strongly correlate with amyloid deposition in patients with mild cognitive impairment [
36], and activated microglia have been found surrounding amyloid plaques [
27,
37,
38] and neurofibrillary tangles [
36,
39,
40], probably in an attempt to phagocytose them. However, their activation in AD becomes persistent and, eventually, ineffective [
41,
42]. Microglia-driven inflammatory responses include gal3 release, which has emerged as one of the foremost molecules in brain innate immunity associated with neurodegeneration [
27,
43]. Gal3 belongs to a protein family with at least 15 members that have substantial sequence similarity in their carbohydrate-recognition domain (CRD). It binds to β-galactosides with variable affinities and specificities [
44,
45] and has been found associated with microglial cells in close vicinity to Aβ42-containing plaques in AD patients as well as in animal and cellular models [
27].
Gal3 has been found to act as an endogenous ligand for paracrine/autocrine toll-like receptor 4 (TLR4) as well as the triggering receptor expressed on myeloid cells 2 (TREM2), binding through its CRD [
27,
33]. Moreover, gal3 has been found significantly upregulated in AD patients [
27], surrounding the atherosclerotic plaques in relation to atherogenesis [
46], and associated with other human diseases such as major kidney adverse events [
47], stroke [
48] and atopic dermatitis and psoriasis [
49]. Notably, it has been observed that gal3 inhibition decreases the inflammatory response of microglial cells to treatment with fibrillar Aβ. In the same study, a significant Aβ plaque reduction and improvement of cognitive performance were observed in 5×FAD mice lacking gal3 [
27]. Also, gal3 is capable of binding TREM2, a central microglial receptor important in AD and essential for a functional microglial response [
27]. However, possible functional involvement of gal3 in the disruption of cognition-relevant neuronal network oscillations remains unknown.
Here, we set out to perform ex vivo recordings of gamma oscillations in the cognition-relevant hippocampal CA3 area [
50] of wild-type (WT) and 5×FAD mouse brain slices. Gamma oscillations were induced by applying 100 nM kainic acid (KA) either in an interface- or a submerged-type recording chamber. Concomitantly to local field potential (LFP) recordings, we performed whole-cell patch-clamp recordings of FSNs and PCs with emphasis on FSN activity.
Discussion
Brain oscillations have been largely documented in humans aided by the use of EEG recordings [
79‐
84] and are on the focus as a diagnostic tool. Their manipulations have emerged as a putative therapeutic approach for AD. Particularly, gamma oscillations hold a prominent interest in AD, due to studies performed both in vivo and ex vivo [
73,
74,
82,
83,
85‐
89]. The mechanisms and gain-of-function found in these studies strengthen the notion that studying the neuronal network dynamics underlying gamma oscillations in different models provides useful tools as targets or effectors that may have therapeutic potential. Moreover, these studies reveal that neuronal network dynamics in animal studies and in humans may operate in a way similar to that observed experimentally ex vivo. For instance, the application of acetylcholine to brain slices increases neural excitability, mimicking cholinergic inputs to the hippocampus during exploratory behaviour in vivo [
90]. In addition, cholinergic gamma oscillations recorded ex vivo share numerous mechanistic commonalities with in vivo gamma oscillations, involving glutamatergic-mediated excitation and fast rhythmic GABAergic inhibition [
59,
90,
91] similar to kainate-induced gamma oscillations ex vivo [
25,
51‐
53,
55,
61,
92].
Thus, synchronized fast neuronal activity between PCs and FSNs within the hippocampal CA3 circuits gives rise to the emergence of periodic fluctuations of the LFP in the gamma frequency band. Such activities comprise phase-locked AP firing and rhythmic inhibitory and excitatory postsynaptic currents (IPSCs and EPSCs, respectively), as shown in the current study. Particularly, FSNs command the entrainment of PC activity during gamma oscillations and provide stability of the neuronal network performance. Therefore, FSNs are suitable targets to counteract deficient cognitive performance [
8,
9,
16,
25,
54,
55,
68,
75,
93]. The coordinated and cooperative activity within the network allows for the binding of neurons in neuronal ensembles that then process sensory information, thus supporting cognitive processes and memory formation [
94]. Therefore, any functional deviation such as impairment of FSN activity, disruption of inhibitory and excitatory synaptic transmission, and consequent aberrant PC activity, may negatively impact the overall network performance, with detrimental cognitive implications as observed in AD.
In this regard, we have previously observed that Aβ42 induces synaptic failure as well as desynchronization of AP firing of PCs [
51,
53], degradation of FSN spike phase-locking [
52,
62] relative to gamma oscillations, and overall neuronal network impairment [
51,
52,
62,
70,
71,
95]. Interestingly, the current study revealed for the first time that acute gal3 induces a functional network collapse that shares commonalities with the reported effects of Aβ42 on neuronal networks dynamics [
51‐
53,
62,
89]. Moreover, a possible association between gal3 and Aβ42 underlying detrimental microglial activation in AD has been described recently [
27]. Despite huge efforts towards counteracting the pathological events leading to progressive cognitive impairment in AD, disease-modifying approaches have failed or shown only modest advances [
76]. One possible shortcoming of past therapeutic attempts is the fact that their focus has been merely on Aβ42 plaque removal [
76], whereas growing evidence points to neuronal synchronization as well as neuroinflammation as critical targets to counteract the progression of AD [
26‐
29]. Circuit entrainment into physiological brain rhythms has shown prominence as a suitable and promising approach counteracting progressive neuronal network dysfunction and deficient cognitive performance in AD, with a focus on FSN activity [
52,
73,
75]. On the other hand, microglial gal3 has been proposed as a central regulator of microglia-driven neuroinflammatory responses in AD [
27]. We found here that inhibition or deletion of gal3 counteracts all of the functional deviations it induces (e.g., decrease of gamma power and rhythmicity, increase of gamma frequency variance, impairment of PC and FSN firing rate and phase-lock to gamma oscillation, hyperpolarization of FSN resting membrane potential, increase of the firing threshold in basal conditions, and disruption of rhythmic inhibitory and excitatory synaptic transmission), thus preserving normal neuronal network dynamics that are relevant for cognitive capabilities.
This functional impact of gal3 on neighboring neurons and synapses farther from Aβ42 plaques should also be considered. Here, we observed that preincubation of the hippocampal neuronal network with gal3 induced a drastic degradation of gamma oscillations mediated by gal3 CRD. Interestingly, gal3 has been reported to bind to and activate different microglial receptors, which include TREM2 [
27] and TLR4 [
33]. Particularly, gal3 binding to TREM2 and TLR4 has been described to be mediated by its CRD [
27,
33]. Accordingly, the consistent prevention of gal3-mediated disruption of neuronal and network function observed in the presence of the gal3 inhibitor TD139 reinforces our findings since TD139 is a 3,3’-Bis-(4-aryltriazol-1-yl) thio-digalactoside gal3 inhibitor with high affinity for gal3 CRD.
In this study, we tested whether the disruption of gamma oscillations observed following exposure of the hippocampal network to gal3 is accompanied by changes in the expression of some genes related to microglial activity. Overall, we did not observe relevant changes in the expression of the analyzed genes, which could be ascribed to the exposure time (45 min total), which appears long enough to prevent an efficient induction of gamma oscillations but too short to trigger a significant change of gene expression. However, the overall quantification of the expression of the listed transcripts in the present study should just be taken as a descriptive clue and future experiments should be performed to further assess a wider range of microglia activation-related genes with longer exposure to gal3.
The electrophysiological changes observed here in neurons and the network appear to rely on a mechanism linked to the cell membrane (i.e., a particular microglial receptor), which probably induces an early microglial activation with further consequences if the application lasts longer, as expected from the time-dependence shown in Fig.
1 versus Fig.
2, and as it may happen in vivo during sustained gal3 release. Again, the overall preventive effect of TD139 reinforces this notion. TD139 mainly acts extracellularly within the time of application employed here. The inhibitor does not reach intracellular compartments unless incubation/wash-in lasts 24 h or longer [
96].
Furthermore, we found that gal3 is less efficient in disrupting gamma oscillations if the network has already properly entrained into the gamma rhythm (see Fig.
2). However, applied to the network prior to FSN engagement into strong spike-phase coupling, gal3 prevented the proper establishment of a coordinated activity that led to gamma oscillations of physiological relevance (see Fig.
4). In an AD scenario, it is tempting to hypothesize that at a certain point during very early pathology progression, cognition-relevant neuronal networks preserve their functionality due to homeostatic mechanisms (a high concentration of gal3 is needed to disrupt cellular and neuronal normal activities). As the pathology progresses, such homeostatic mechanisms become overwhelmed, perhaps even deleterious (i.e., microglial shift to damage-associated microglia phenotype) [
22] and gal3 concentration largely increases. Then, a progressively weakened neuronal network appears more susceptible to gal3, which could also reach non-damaged neurons/synapses by diffusing throughout the brain parenchyma, once secreted by the already activated microglia. At this point, a lesser gal3 concentration could drive major dysfunction, which is in line with the damage amplification role of gal3. Such hypothesis deserves further experimental confirmations.
A plausible explanation for the broad range of gal3-induced effects observed could be found in the complex purinergic signaling in the neuron-microglia crosstalk. It has been proposed that the initial increase of extracellular adenosine to levels far greater than reached in physiological conditions initially leads to a burst of adenosine receptor 1 (A
1R)-mediated inhibition, and the continuous massive overflow of extracellular adenosine then overcomes the restricted activation of A
2ARs, which results in a predominant role of A
2ARs in the development of neurodegeneration [
97]. In line with this proposal, a paramount impact of ATP/adenosine signaling on hippocampal circuitry function has been observed. At mossy fiber-CA3 synapses, microglia-derived ATP differentially modulates synaptic transmission and short-term plasticity through activation of presynaptic P2X4 receptors and A
1R, respectively, with the latter converting to adenosine extracellularly [
98]. Additionally, blockade of A
2ARs prevents lipopolysaccharide-induced impairment of long-term potentiation in rats in vivo, by counteracting the shift of microglia towards a pro-inflammatory phenotype [
99]. Notably, A
2AR is upregulated in the APP/PS1 mouse model [
100] as well as in cortical areas [
101] and the hippocampal formation of AD patients [
102]. It has been found that adenosine inhibits KA-induced and spontaneous gamma oscillations, particularly via the activation of A
1R [
103]. Consistent with our hypothesis and results, it has been observed that endogenous ATP release drastically reduces PC AP firing rate as well as spike synchronization during KA-induced gamma oscillations in the CA3 area. Moreover, both KA- and acetylcholine-induced gamma oscillations are inhibited by ATP/adenosine receptor activation [
104]. Here we also observed that gal3 impairs cholinergic-induced gamma oscillations (Additional file
1: Fig. S1), which indicates that the effect of gal3 is independent of the method of gamma oscillation induction. Of note, Aβ42 also induces degradation of cholinergic gamma oscillations and gamma-theta rhythm interaction in the hippocampus [
62]. In addition, the degeneration of basal forebrain cholinergic neurons (one of the relevant cholinergic inputs to and for the hippocampal rhythmogenesis) has been ascribed as a common feature of AD [
105].
In our study, we found impairment of both excitatory and inhibitory synaptic drive onto PCs as well as a decrease of excitatory input onto FSNs. Interestingly, gal3 differentially affected the amplitude or frequency of excitatory synaptic events in FSNs compared to PCs, which suggested that gal3 affects synaptic transmission in a synapse-specific manner with differential effects and sites of action (presynaptic and postsynaptic sites), depending on the synapse type and the circuit activity. In this regard, hippocampal A
1Rs are known to hyperpolarize FSNs, reduce the excitability of PCs and interneurons, and also reduce neurotransmitter release [
106,
107]. Changes in neurotransmitter release are mostly associated with changes of frequency of synaptic events, observed in this study as a decrease of EPSCs onto PCs (Fig.
3b) and FSNs in either basal or activated state once the circuit is challenged with gal3 prior to rhythm entrainment (Additional file
1: Fig. S6e and 7b). Notably, gal3 reduced the occurrence of larger IPSC components in PCs probably as a reflection of the observed FSN impairment since the major perisomatic inhibition of PCs is driven by FSNs [
8,
9]. The overall collapse of the network was also observed in the gal3-induced increased variability of the phase relation of both EPSCs and IPSCs with the corresponding LFP-gamma as well as the loss of rhythmicity of the postsynaptic currents. This loss of rhythmicity of synaptic transmission likely accounts for the degradation of the network rhythm since the generation of gamma oscillations depends on balanced excitatory and inhibitory interplay [
12]. However, due to the diverse evidence of purinergic signaling in the modulation of the operational capacity of the hippocampal circuitry, and commonalities found in our study regarding gal3-induced neuronal network collapse, a possible underlying mechanism involving ATP/adenosine receptor activation deserves further research, without exclusion of an indirect involvement of astrocytes, nitric oxide, metabolic arrestment [
31,
32,
102,
108] and a direct effect of gal3 on PCs and FSN. Also, effects of microglial activation due to the slicing procedure appear negligible in our study since microglia seem to be less relevant for moderate tissue repair at the slice cut surfaces as well as for synaptic remodelling and neuronal network formation, at least during the second and third postnatal weeks of hippocampal maturation in situ [
109,
110]. Moreover, application of the gal3 specific inhibitor TD139 alone (Fig.
1c, d), or deletion of gal3 (Fig.
5e, f), did not affect neuronal network dynamics prior to gamma oscillation induction.
Finally, there is mounting evidence suggesting that in AD, synaptic and network failure starts long before the establishment of Aβ42 deposits into solid plaques and manifestation of cognitive decline [
22,
74,
111‐
113]. Interestingly, in the APP/PS1 mouse model, associative long-term synaptic plasticity is impaired in CA3 PCs at the early onset of AD-like features due to the postsynaptic activation of upregulated A
2AR [
100]. Moreover, focal glial activation has been reported to precede amyloid plaque deposition in APP transgenic mice associated with a vicious cycle of APP proteolytic cleavage that gives rise to soluble and amyloidogenic immunostimulatory mediators [
28]. Using a proteomic approach, it has been found that immune alterations in microglia in 5×FAD mice occur prior to plaque deposition [
114].
Recently, direct involvement of microglia-released gal3 and Aβ cross-seeding agents in plaque formation has been validated [
27,
115,
116]. Particularly, the 5×FAD mouse model lacking gal3 (5×FAD-Gal3KO) fails to develop prominent Aβ plaques and cognitive impairment typical of the 5×FAD model at 6 months of age [
27]. Here we found that a possible underlying functional reason explaining these previous reports could be the degradation of gamma oscillations (observed in the 5×FAD model at 6 months of age), which is absent in the age-matched 5×FAD-Gal3KO mice that also show significant reduction of Aβ load in the hippocampal CA3 area (Fig.
5). 5×FAD mice also showed a significant slowing of gamma oscillation frequency compared to WT control, but the age-matched 5×FAD-Gal3KO retained values similar to WT. Interestingly, slower gamma oscillations have been observed also in the CA3 area of organotypic hippocampal cultures under microglial priming with interferon gamma [
117]. Inhibition of gal3 by co-application of TD139 prevented the decreases of both gamma oscillation power and frequency in the acute Aβ application model (see Fig.
5). 5×FAD mice also displayed gamma oscillations of higher frequency variance (Additional file
1: Fig. S9), which is in line with the less rhythmic behavior observed during gal3 acute application (e.g., see Fig.
1d, f). This decrease of gamma rhythm quality was absent in 5×FAD mice lacking gal3. Interestingly, acute Aβ decreases gamma rhythm quality evidenced by a decrease of the Cr (Fig.
5c) [
51,
55] or an increase of the frequency variance observed under acute Aβ application [
52] or in the App
NL−G−F mouse model [
25]. This suggests a highly congruent parallelism between the underlying mechanisms of gal3- and Aβ-induced degradation of gamma oscillations. Moreover, we have observed that the cognitive impairment correlates with the impairment of gamma oscillations ex vivo [
25]. Recently, we found that the recovery of cognitive functions is accompanied by recovery of the hippocampal oscillatory activity and that counteracting proinflammatory triggers counteracts degradation of gamma oscillations in our model [
61,
92].
Conclusion
In summary, here we report for the first time that gal3, a proposed central microglial/neuroinflammatory regulator in AD, causes degradation of a neuronal network rhythm and reveal its underlying neuronal synchronization mechanisms by performing
ex-vivo recordings of gamma oscillations, which may well serve as an appropriate prototype for cognition-relevant neuronal network dynamics. The impairments observed are mediated by the gal3-CRD and prevented by the gal3 inhibitor TD139 in a dose-dependent manner. Additionally, gal3 prevents neuronal network entrainment into proper gamma rhythm. Such disruption is accompanied by the impairment of FSN and PC gamma spike-phase locking and disturbances of excitatory and inhibitory synaptic transmission. Interestingly, we found a possible functional link for gal3 to AD since (1) TD139 prevents Aβ42-induced degradation of gamma oscillations ex vivo and (2) gamma oscillations are impaired in the hippocampal CA3 area of the 5×FAD mouse model at 6 months of age while gamma oscillations recorded from 5×FAD mice lacking gal3 (5×FAD-Gal3KO) remain similar to age-matched WT counterparts. In parallel, 5×FAD-Gal3KO display lower Aβ plaques in the recorded hippocampal CA3 area. Moreover, our results bridge the gap between cellular and molecular notions on the central role of gal3 in AD progression and behavioral studies by providing functional evidence that is relevant for those behaviors. This reinforces the therapeutic potential of inhibiting/removing gal3 to counteract AD progression and putatively as disease-modifying interventions for other neurodegenerative disorders involving microglial activation and neuroinflammation. In vivo recordings performed at different time points during disease progression (e.g., in the App
NL−G−F mouse model [
25] lacking gal3 signaling) will ultimately cross-validate the therapeutic potential of our current findings. Moreover, such studies on the follow-up of the molecular and cognitive changes linked to the functional parameters studied here, will establish a timely rescue approach including effective inhibitor dosage, safety, and treatment schedule, and will further our notions on the therapeutic potential of gal3 inhibition in AD.