Microglia proliferation and altered morphology in the cerebellum.
To investigate a synergistic effect of extrinsic stresses, we produced mouse models exposed to either MIA, RSDS, or both (Fig. 1a and Extended Data Fig. 1a-e). We administered polyinosinic:polycytidylic acid (poly(I:C)) at embryonic day (E)12.5 to the pregnant mothers to produce MIA mice. The male offspring received defeat stress for ten consecutive days from an aggressive stressor at early puberty (RSDS, 4-week-old) (Fig. 1a) or mature state (9-week-old) (Extended Data Fig. 9a,b). The proportions of susceptible mice were 65% in RSDS alone and 95% in 2HIT-conditioning when exposed to RSDS at 4 weeks old, whereas those were 50% in RSDS alone and 73.3% in 2HIT-conditioned male mice when exposed to RSDS at 9 weeks old, suggesting high vulnerability to RSDS in peripubertal mice (Fig. 1b and Extended Data Fig. 1f-o). In the following study, we adopted RSDS conditioning at 4 weeks old, unless otherwise stated. In contrast, female mice were only susceptible at the proportion of 44%, suggesting higher resilience to the combined stress (Fig. 1b). Since female mice cannot receive RSDS from male aggressor ICR mice, we administered a new protocol for generating 2HIT female mice (Methods), and the number of insistent chases and attacks were comparable to the male offspring (Extended Data Fig. 1p,q).
To assess the brain immune environment of the single-hit and 2HIT models through the developmental time course, we counted the microglia in the brain with Iba1 immunostaining (Fig. 1c,d and Extended Data Fig. 2). Figure 1c shows the microglia density of Control and 2HIT mice brains. We investigated across brain regions of both male and female mice at 2-week, 5-week, and 9-week-old in different stress conditions. For stable cell counting, we developed a cell counting program (Methods, Supplementary Figs. 1 & 2). The density of Iba1-positive microglia in the cerebellar cortex, cerebellar nuclei, and ventral tegmental areas (VTA) of male 2HIT mice was significantly increased at 5 weeks (Fig. 1d, Male), while 2HIT stress insults mainly exhibited microglia increase in the cerebellum and medial prefrontal cortex (mPFC) at nine weeks. In contrast, the microglia density was not significantly different in the hippocampus (: dorsal dentate gyrus), VTA, and periaqueductal gray (PAG) at 9 weeks old. In the female cerebellar cortex, 2HIT susceptible mice showed a significant increase in microglia similar to males, but resilient mice did not (Fig. 1d, Female). Considering the result of BALB/c mice, 2HIT stress-induced increase in cerebellar microglia was independent of mouse species (Fig. 1d, BALB). Co-labeling with Ki-67, a cell proliferation marker, showed a marked increase in microglia density in the cerebellum of 9-week-old 2HIT male mice (Fig. 1e,f). The increased proportion of Ki-67-positive microglia suggests that not only the total number of microglia increased but also their proliferation advanced in the 2HIT cerebellum (Fig. 1f). Therefore, 2HIT stress insult markedly increases cerebellar microglia.
To quantify morphological differences, we assessed the microglial processes. The Sholl analysis of the Iba1(+) microglia showed a significant reduction in their branching particularly in the cerebellar cortex and nuclei, depending on the stress conditioning (Fig. 2a-c and Supplementary Fig. 3). Accordingly, 2HIT stress responses of microglia appear in the cerebellum. Recent observation in MIA mice by poly(I:C) administration at E9.5 revealed alteration of microglia morphology in the ventral striatum32. Our finding suggests that later timing of MIA conditioning at E12.5 and additional RSDS at juvenile caused the regional difference.
The replacement of microglia prevents the increase and morphological changes of microglia.
Given stress synergy leads to the distribution and transformation of reactive microglia in the brain, the removal of activated microglia and replenishment of a new population may revert the immune environment. To investigate if primed microglia by MIA change the number and their branching in response to defeat stress at 4 weeks, we depleted microglia by CSF1R inhibition during the period of RSDS conditioning (Fig. 1a). Repopulation of microglia has been shown primarily through a proliferation of the remaining subpopulation45. We therefore depleted microglia across the brain once (Extended Data Fig. 1r,s). We found that the number of microglia, Ki-67(+) microglia density, and their branching reverted to the control level after four weeks of the repopulation period (Fig. 1d,f and Fig. 2a-c, Control + microglia-replacement (rMG) & 2HIT + rMG). The results suggest that a combination of inflammatory stress along a developmental timeline spanning early puberty leads to the onset of the inflammatory milieu in the 2HIT model and that the replacement of microglia resets the tuned state.
Purkinje cell loss in the 2HIT cerebellum.
Next, we asked if the increase in reactive microglia links to disruptions of the neuronal circuit in stress accumulation, leading to cerebellar dysfunction. Since the loss of 20–33% of Purkinje cells in the MIA-exposed animals was previously reported46, we investigated the possibility of neurodegeneration in the cerebellar lobules by the accumulative stress (Fig. 2d-h and Supplementary Fig. 4). First, we found a prominent reduction in Calbindin(+) Purkinje cells in lobule VIa-VIb of the 2HIT cerebellum (Fig. 2d,e). The number of Purkinje cells decreased by 38.4% in lobule VIa-VIb, spanning to the entire cortex of the male 2HIT (lobules II-IV and IX-X, Fig. 2e), suggesting a promotion of Purkinje-cell death in the stress-exposed cerebellum compared to MIA. Female MIA also showed a 29.2% reduction in Purkinje cells. Although the neurodegeneration by MIA was progressive in the course of development46, the second hit RSDS accelerated the neuron loss (Fig. 2e). In contrast, the density of Calbindin(+) interneurons in mPFC was comparable across stress conditions and sexes (Supplementary Fig. 5), implying cerebellum-specific disruption. Notably, the microglia replacement ameliorated the cerebellar neurodegeneration (Fig. 2d,e, 2HIT + rMG), suggesting the involvement of microglia reactivity.
In identical cerebellar cortical slices, the density of Calbindin(+) Purkinje-cell axons decreased, and the density of Iba1(+) microglia increased, respectively (Fig. 2f,g). Principle component analysis (PCA) showed a distinct separation of 2HIT (Fig. 2h). MANOVA (*p < 1.6 × 10− 11) of the three measurements across Control, 2HIT, and 2HIT + rMG indicates a strong correlation between Purkinje cell loss, the axon degeneration, and microglia increase in cerebellar cortical regions as 2HIT phenotypes. Therefore, reactive microglia are crucial for the Purkinje cell neurodegeneration41,43.
The emergence of stress-associated microglia/macrophages in the 2HIT cerebellum.
Recent microglia studies highlighted the spectrum of transcript expression and the functional heterogeneity22–25, though it had not been addressed if different forms of microglia show specific characteristics underpinned by multiple protein expression. Next, we sought the molecular signature and cell association in the single- and double-hit brains, using imaging mass cytometry (IMC) (Fig. 3, and Extended Data Fig. 3) and single-cell proteomics analyses (Fig. 4 and Extended Data Figs. 4–6). IMC with designed 24-antibody markers and the standard Iridium (Ir)-intercalators to FFPE sections (Methods) from the stress-model brains successively demonstrated highly multiplexed images of different microglia/macrophage subpopulations, neurons, astrocytes, blood-brain barrier (BBB), pericytes, and vasculatures, with 1-µm precision (Fig. 3d and Extended Data Fig. 3). Remarkably, we found unique immune cells expressing distinct molecules (: MHC-class II, soluble (s)IL-6R (i.e., gp130), TREM2, TGFBR2, APOE, and MMP-9) dominantly in the cerebellar cortex, dentate (DN), fastigial (FN), and medial vestibular nuclei (MVN) (Fig. 3d and Extended Data Fig. 3a-d). We called this type of immune cell stress-associated microglia/macrophage (SAM) by multiple-protein expression with IMC. MHC-class II(+) immune cells were scarcely found in other brain regions (Fig. 3e,f). The caveat is that formerly found disease-associated microglia (DAM) by transcriptome studies in Alzheimer’s disease26,47–49 and other neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS), tauopathy, and multiple sclerosis50–53 appear in our human postmortem brain-sample with neurodegenerative disease (: diagnosed to amyotrophic lateral sclerosis) in the IMC (Extended Data Fig. 3e). The 2HIT SAM puncta have distinct cell size and spatial localization. The classification by manually-measured puncta diameters suggests the existence of small- (: Φ < 10 µm, 87.2% of total 78 cells) and large-size (: Φ > 10 µm, 12.8% of total) populations among MHC-class II(+) immune cells in both the cortex and nucleus (Fig. 3e-h). We found a close deposition of SAMs to claudin-5-positive vasculatures and ventricular surface (Fig. 3i), as described below. Our result indicates the spatially distinct subsets of microglia/macrophages in 2HIT brains at a multiple-protein expression level.
Cellular and molecular signature via IL-6 and TGF-β pathways unveiled by the single-cell proteomics.
Proliferated parenchymal or border-associated microglia would be responsible for the emergence of SAMs24,26,28,47. Yet, the spatial proteomic signature and the transition states in 2HIT models are elusive. To this end, we performed cell segmentation and spatial single-cell proteomics analysis of the multiplexed imaging data (Extended Data Fig. 4 and Fig. 4). We used Ir-images as a seed for the segmentation of the cellular nuclei and measured the IMC signal intensity of the segmented cells (Methods). For clustering, we used signals of αSMA, Olig2, IBA1, Sall1, MMP-9, MAP2, Lyve1, TGFBR2, Claudin-5, GFAP, sIL-6R, IL-6, AQP4, TMEM119, Caspase-1, PDGFR beta, Ki67, TGF-β1, IL-17R, TREM2, CX3CR1, MHC-class II, APOE, and IL-17. In Supplementary Figs. 6–8, we provide a representative marker-positive cells and signal intensity profile of the marker-molecules. We identified 4284 and 5895 cells from the Control and 2HIT cerebellar cortex, respectively (Methods). The cell segmentation of Control and 2HIT indicates cell-type specific clusters (12 clusters in Control and 14 clusters in 2HIT) in the cerebellar cortex (Extended Data Fig. 4a and Fig. 4a). Single-cell clustering with tSNE across experiments showed the clusters of characterized cell populations by cell-type specific marker subsets (Extended Data Fig. 4b-d in Control and Fig. 4b-d in 2HIT). In the control cerebellar cortex, by using 24-marker signal intensity and the localization pattern, we annotated segmented cell populations to 12 clusters: granule-cell layer neurons, molecular layer interneurons, immune cells, astrocytes, vasculatures, pericytes, and oligodendrocytes (Extended Data Fig. 4c). Immune cells were identified from Iba1, Sall1, and CX3CR1 expression, and neurons were identified by MAP2 expression and less immune-cell marker expression. Please note that the thickness of IMC is 4-µm, and we hardly detected the complete shape of large cells such as Purkinje cells. In the immune cells of Control, TMEM119, CX3CR1, and IL-17R characteristically express as markers, whereas there is less expression of sIL-6R, TGFBR2, TREM2, MHC-class II, and MMP9 (Extended Data Fig. 4b-d). In contrast, in the 2HIT cerebellar cortex, we identified certain immune cell clusters characterized by expression of sIL-6R, TGFBR2, TREM2, MHC-class II, MMP9, APOE, Ki67, TGF-β1 Caspase-1, and Lyve1 (clusters 12 and 14) (Fig. 4b-d). The proportions of cells composing clusters 12 and 14 were 2.2% and 0.9%, respectively. Ki-67 expression is high in clusters 6 and 12, whose summed proportion was 27.9% of immune cells (see also Fig. 1f). The single-cell spatial proteome of the RSDS and MIA cerebellum found no SAM in the cerebellum (data not shown). Thus, in the cerebellum, there is a 2HIT stress-associated cluster. The localization of spatially segmented clusters corresponds to the location of SAMs in the superimposed images of multiplexed staining (Fig. 4a, and Extended Data Fig. 5a). Interestingly, the immune cells of clusters 12 and 14 (Extended Data Fig. 5a) are located close to each other, and Extended Data Fig. 5a Cluster 12 (squares) includes SAM-puncta. Accordingly, we annotated Cluster 12 as SAM. From Lyve1 expression, Cluster 14 was as border-associated macrophages (BAMIMC 28,47).
To answer whether the different protein-expression phenographs result from individual differentiation or consecutive steps in stress synergy, we analyzed a cell transition47,49 from the IMC expression levels (Extended Data Fig. 5b-e). Using PHATE mapping (Methods), we found that immune cells composing different clusters of 2HIT, but not Control, aligned in a continuous trajectory, indicating in-series differentiation of microglia/macrophage in 2HIT cerebellum with a direction (Extended Data Fig. 5c). The PHATE transition entropy predicted a direction of the cell transition or differentiation (an arrow in Extended Data Fig. 5c). Cluster 14 did not align in the transition course. Therefore, we considered cluster 12 in 2HIT as the mature SAM. Cluster 6 (immune cells) has a molecular signature in the transition state from resting state to SAM cluster, and it is assumed the progenitors of SAMs (proSAMs). RSDS datasets suggested a consecutive transition from naïve microglia to proSAMs, while the transition was minor in MIA from entropy (Extended Data Fig. 5d,e).
We further asked if there were any spatial links between clusters. We performed a pairwise distance analysis of cells (Extended Data Fig. 6), and we found that, in Control, Cluster 10 (Vasculature) and Cluster 11 (Pericytes) were located closely. In 2HIT, pairwise distance was significantly close between Cluster 10 (Vasculature), Cluster 12 (SAM), and Cluster 14 (BAMIMC). Results suggest that SAMs and BAMs, not the other immune cells, are localized close to the vasculature in the 2HIT cerebellum, supporting the result of Fig. 3i.
Together, we postulated that MIA generates proSAMs, and the subsequent RSDS impacts on proSAMs to transit to mature SAMs. Mixture analysis of immune cell datasets of Control and 2HIT also provides a distinct cluster in Fig. 4e and f. The protein expression pattern of Cluster 13 consists of cells only from the 2HIT cerebellum, reflecting the SAM gene expression pattern24,26,48,51. The volcano plot comparing the expression patterns of immune cells and SAM indicates significantly high expression of TGF-β1, APOE, IL-6, and TREM2 (Fig. 4f), suggesting promotion of the molecular signaling. However, due to the large number of granule cells present, there is an experimental limitation in which contamination potentially causes overestimation.
Finally, we performed a conventional triple immunostaining of SAMs (Fig. 4g-i) using markers of Iba-1, TREM2, and MHC-class II. In the male 2HIT cerebellar cortex, 10.4 ± 2.7% of Iba-1(+) microglia are the double-marker positive microglia (Fig. 4h), comparable to IMC. Female 2HIT susceptible mice (female 2HIT sus) also have the marker positive microglia (5.5 ± 1.5%) in the cerebellum, dominantly, but not in other regions and female resilient. Immunohistochemistry images collaborate the blunted morphology of 2HIT and the remedy by microglia replacement (Fig. 4g-i).
Inflammatory stress synergy reduces intrinsic excitability.
How do reactive microglia with SAMs alter neurophysiological properties in stress accumulation models? To determine the independence or the additivity of multiple developmental stresses with MIA and RSDS in the neurophysiological properties of Purkinje neurons, we first examined the firing of action potentials evoked by depolarization step-pulses under suppression of synaptic transmission. In Extended Data Fig. 7a,b, the firing frequency of Purkinje cells was reduced in male MIA, 2HIT, and female 2HIT sus, but not in male RSDS and female 2HIT resilient. This result supports the idea that the intrinsic excitability of Purkinje cells is lowered in the autism spectrum disorder (ASD) model mice36. Next, waveform analyses of an action potential showed an increase in input resistance, half width (i.e., FWHM), 10–90% rise time of action potential in MIA and 2HIT Purkinje cells, suggesting the modulation of voltage-gated K+ channels and voltage-gated Na+ channels, which corroborated the reduced intrinsic excitability (i.e., hypoexcitability) (Extended Data Fig. 7c-o). Female 2HIT sus Purkinje cells also showed a decrease in action potential amplitude. Therefore, these results indicate that the intrinsic excitability of Purkinje cells is lower in the stress models. Incidentally, firing frequency and action-potential waveform were comparable between Control and 2HIT mPFC layer 5 pyramidal neurons (Supplementary Fig. 10).
To investigate if the reactive microglia by 2HIT conditioning modulated the intrinsic excitability of Purkinje cells, we depleted microglia during 2HIT conditioning and performed a patch-clamp after recovery at nine weeks of age. Then, the firing frequency of action potential reverted to the control level (Extended Data Fig. 7, 2HIT + rMG). The results indicate that the 2HIT-stress-exposed inflammatory milieu disrupts the membrane properties of the cerebellar neurons and reduces their excitability.
2HIT stress insults during the developmental period induce functional dysconnectivity and behavioral anomalies.
Ejection impairment of action potentials of Purkinje neurons (Extended Data Fig. 7) and axon degeneration (Fig. 2d-f) may lead to disturbance of long-ranged neural transduction in living animals. Next, we performed resting-state fMRI to parcellate pathways with malfunction. Extended Data Fig. 8a shows correlation matrices of the blood-oxygen-level dependent (BOLD) signals of 90 seeds across the brain (Supplementary Table 1). We obtained the Z-scores of the Pearson correlation coefficient as the index of functional connectivity. The correlations between the cerebellum (e.g., cortex and nucleus), midbrain (e.g., VTA and PAG), and prefrontal cortex (e.g., mPFC, cingulate cortex, and primary motor cortex) significantly decreased in the stress model mice (Extended Data Fig. 8a). From the connectivity matrices, we generated network models for visualizing interaction networks at different significant levels (Extended Data Fig. 8b), and we found that the node pairs in 2HIT reduced at specific brain regions: cerebellar cortex, cerebellar nuclei, pons, VTA, PAG, the insular region, and medial prefrontal cortex (Extended Data Fig. 8c). Thus, the cerebellum-involved pathways reduce functional connectivity in in vivo resting state.
Since the Purkinje-cell axons degenerate in 2HIT mice (Fig. 2f), we investigated axon structure in the different brain regions and associated changes in myelin sheath using electron microscopy. In 2HIT, both the diameter of the axon and g-ratio significantly reduced, compared to Control in the cerebellar nucleus (: DN) (Extended Data Fig. 8d). The results suggest a population of the myelinated axon with thick caliber decreased more in 2HIT than Control cerebellar nucleus (Extended Data Fig. 8d-f). Similar results were obtained in VTA, a projected region from the dentate nuclei, although the axon diameter and g-ratio have no change in PAG and mPFC. Next, we investigated if the microstructural changes in the 2HIT brain were rescued after depletion of microglia (Supplementary Fig. 11). Microglia replacement in the 2HIT brain remained an increase in smaller-caliber axons in the cerebellar nuclei, while there were no significant changes in g-ratio (Supplementary Fig. 11). The remaining reduction of the nuclei axon diameter implies the irrelevance of microglia to recovery from the axon shrinkage in the cerebellar nuclei. Considering that oligodendrocyte differentiation and axon diameter in the developing male cerebellum underlie the social behavior54, our results suggest the disruption of the axon transduction in the 2HIT brain across the cerebellum-including network.
Hypoactivity of the cerebellum and disrupted cerebellum-associated pathways may lead to behavioral anomalies. We performed a battery test of mouse behaviors: social avoidance, open field, elevated plus maze, 3-chamber, social dominance, marble burying, novel object recognition, and prepulse inhibition (PPI) in the experiments of Control, RSDS, MIA, 2HIT, Control + rMG, and 2HIT + rMG (Fig. 5a-j). We further tested for motor coordination on the rotarod, balance beam, and footprint of strides (Fig. 5k-n). Animals showed multiple behavioral deficits among groups of stress models: RSDS, MIA, and 2HIT (Fig. 5). For instance, the social-avoidance test indicates that RSDS and 2HIT showed heavy aversive responses but not Control and MIA (Fig. 5a). In Fig. 7j, a reduction of PPI in MIA and 2HIT indicates less habituation to the startle response, representing the retardation of sensorimotor gating as a schizophrenic-like stress response. The 2HIT animals do not have a weak auditory response but rather have higher startle responses against loud acoustic stimulation (Supplementary Fig. 12). The timing of RSDS insult affected behavioral change since animals exposed to RSDS at nine weeks old showed relatively mild anomalies (Extended Data Figs. 1k-o & 9). Further, the motor performance tests showed clear motor discoordination in 2HIT and RSDS without any ataxia (Fig. 5k-n). Female 2HIT sus also showed multiple deficits in behaviors with a minor difference against male phenotype (e.g., OF, 3C, and MB), and female 2HIT res was comparable to female control. Altogether, our behavior tests suggest the synergistic anomalies in sociability, spontaneous explorative behavior, anxiety, stereotypy and compulsion, episodic-like memory, and startle reflex between single-hit (: RSDS and MIA) and 2HIT animals (Fig. 5o). Our results also suggest motor dyscoordination in stress-exposed mice possibly disrupts the higher-order cognitive behaviors.
Lastly, we aimed to restore the behavioral anomalies in the psychiatric disorder models by microglia depletion. As expected, the transient depletion of microglia and the replenishment enabled a remedy of behaviors (Fig. 5, 2HIT + rMG). The results are consistent with the recovery from disruption of the microglia reactivity and neuronal physiology, implying cumulative stress-induced microglial anomaly was ameliorated by their replacement.
Cerebellum-specific microglia replacement suffices for 2HIT phenotype remedy.
To specify the microglia reactivity in the cerebellum, we performed the cerebellum-specific depletion of microglia and investigated the consequences. To remove microglia in the cerebellum, we injected clodronate disodium salt (CDS) into the peripheral vermis (total 1.2µL vol. of 50 mg/ml CDS) (Fig. 6a). Then, we counted the number of microglia one, five, and thirty days after injection, comparing CDS and phosphate-buffered saline (PBS) injection. The microglia significantly reduced at 5 days post-injection (dpi) and reverted to the control level at 30-dpi (Fig. 6b,c). Figure 6c displays the density of microglia across six brain regions at 5-dpi, indicating cerebellum-specific microglia depletion with the protocol successful. Then, we injected CDS one day before RSDS (Fig. 6d). At nine weeks of age, the number of microglia in the cerebellum reverted to the control level (Fig. 6e), while 2HIT mPFC microglia remained to show a slight increase (119.2%). The density of MHC class II(+) TREM2(+) was scarce in the cerebellum (Fig. 6f, 2HIT + CDS), implying less microglia reactivity. To confirm the recovery, we examined the electrophysiological property and found that the firing frequency of Purkinje cells (Fig. 6g,h) and action-potential waveform parameters (Fig. 6i-o) were comparable to PBS-injected Control, indicating the amelioration by CDS injection. Moreover, the behavior test battery demonstrated the remedy of the anomalies of cognitive-affective behaviors and motor coordination by CDS injection (Fig. 6p-z), except for an improvement of rotarod test score (Fig. 6w). Results indicate cerebellar microglia replacement is successful for the remedy of 2HIT phenotypes.