Persistent DNA damage promotes microglial dysfunction in Ataxia-telangiectasia

The autosomal recessive genome instability disorder Ataxia-telangiectasia, caused by mutations in ATM kinase, is characterised by the progressive loss of cerebellar neurons. We find that DNA damage associated with ATM loss results in dysfunctional behaviour of human microglia, immune cells of the central nervous system. Microglial dysfunction is mediated by the pro-inflammatory RELB/p52 non-canonical NF-κB transcriptional pathway and leads to excessive phagocytic clearance of neurites. Pathological phagocytosis of neuronal processes by microglia has also been observed in multiple sclerosis, Alzheimer’s and progranulin deficiency, suggesting a common mechanism that promotes neuronal damage. Activation of the RELB/p52 pathway in ATM-deficient microglia is driven by persistent DNA damage and is dependent on the NIK kinase. These results provide insights into the underlying mechanisms of aberrant microglial behaviour in Ataxia-telangiectasia, potentially contributing to neurodegeneration.


INTRODUCTION
Ataxia-telangiectasia (A-T) is an autosomal recessive disease caused by loss-of-function mutations in the ATM gene (A-T mutated) (1). ATM is a serine/threonine protein kinase that plays a central role in coordinating the cellular response to genotoxic stress, in particular cytotoxic and mutagenic DNA double-strand breaks (2,3). Consequently, cellular processes regulated in an ATM-dependent manner include chromatin decondensation, apoptosis, senescence, cell cycle, redox balance, metabolism, and splicing (4).
A-T has a wide range of clinical manifestations, however, one of the most devastating clinical signs of the classical form of A-T is neurodegeneration. The disease manifests as motor dysfunction in young children and is predominantly characterised by the progressive loss of Purkinje and granule neurons in the cerebellum (5). Cerebellar degeneration is thought to be linked with defects in the neuronal DNA damage response, metabolic abnormalities, and epigenetic silencing of diverse neuronal genes (4,5). However, the molecular mechanisms that underpin cerebellar degeneration in A-T are poorly understood.
Microglia are resident immune cells of the central nervous system (CNS). Microglia play instrumental roles in the development and maintenance of the CNS, including regulation of neurogenesis, synaptic maintenance and plasticity, trophic support of other cell types, and clearance of apoptotic and dead cells. However, persistent microglial activation and consequent neuroinflammation are implicated in the pathology of diverse neurodegenerative disorders (6). For example, somatic mutations in the BRAF oncogene in erythro-myeloid precursors of microglia result in progressive neurological impairment with multiple features of cerebellar ataxia (7). In addition, microglial priming (an enhanced response to secondary stimuli) and activation, which are both linked to neuroinflammation, are seen in mouse models of Ercc1 Δ/--linked nucleotide excision repair and frataxin deficiencies (8,9).
In A-T defective mice and rats morphological changes associated with microglial activation have been observed (10)(11)(12). Pharmacological inhibition of Atm in cultured ex vivo murine microglia led to neuronal cell death, which was mediated by the secretion of the neurotoxic cytokine IL-1β (12). It was suggested that microglia mount a neurotoxic innate immune response to cytosolic DNA (10)(11)(12). In addition, links between ATM deficiency and activation of the innate immune signalling have been demonstrated in other cell types (13-15). Rodent models of ATM deficiency, however, do not fully recapitulate the human phenotype, possibly due to a higher sensitivity of the human CNS to oxidative stress and DNA damage (11, 16).
Moreover, microglia from rodents and humans show fundamental differences in their proliferation, response to extracellular signalling molecules and secretion, and exhibit highly divergent gene expression signatures (17)(18)(19).
In the present work, we utilised human cell models to investigate the underlying basis for the role of microglia in neurodegeneration in A-T. We find that loss of ATM, or its activity, promotes sustained microglial activation linked with increased expression of pro-inflammatory cytokines and phagocytic clearance. Microglial activation was shown to be mediated by the non-canonical RELB/p52 nuclear factor NF-κB transcriptional pathway. The RELB/p52 pathway is activated in response to persistent DNA damage associated with A-T and is driven by the NF-κB-inducing NIK kinase. Chronic activation of ATM-deficient microglia results in excessive phagocytosis of neurites, potentially contributing to neurodegeneration.
These data provide mechanistic insights into microglial dysfunction in human A-T.

Loss of ATM function results in microglial activation
The effects of ATM deficiency on microglial function were studied in human HMC3 and C20 microglial cell lines (20)(21)(22). To provide an ATM-deficient model system, that is representative of A-T (23), an ATM knockout (KO) human microglial HMC3 cell line was generated using CRISPR/Cas9 (Supplementary Figure S1A and S1B). The loss of ATM function was verified by treating ATM KO microglia with camptothecin, a topoisomerase I inhibitor that induces ATM activation (24). As expected, ATM KO microglia displayed: i) no visible autophosphorylation of ATM at S1981 and grossly attenuated phosphorylation of ATM's downstream target, CHK2, in response to camptothecin ( Figure 1A) (25), and ii) reduced cell proliferation compared to wild-type (WT) cells (Supplementary Figure S1C). To determine whether the ATM KO microglia were activated, we first measured expression levels of two markers of microglial activation, CD40 and CD68 (26,27). Cell surface levels of CD40 were increased in ATM KO microglia to a similar extent as in WT cells stimulated with TNFα ( Figure 1B). Additionally, the levels of lysosomal CD68 were higher in ATM KO microglia as compared to WT (Supplementary Figure S1D). To further investigate the activation status of ATM KO microglia, we measured mRNA expression of the pro-inflammatory cytokines IL6, IL8, IL1B and TNFA. All One indicator of microglial activity is their ability to engulf cell debris, apoptotic and stressed cells, which expose the "eat-me" signal, phosphatidylserine, on their surface (6). Phagocytic properties of WT and ATM KO microglia were investigated using 5 µm carboxylated latex beads, which mimic the size of neuronal soma with externalised phosphatidylserine, as substrates (Supplementary Figure S2A). Both the percentage of phagocytic cells, further referred to as levels of phagocytosis, and phagocytic activity, which reflects the relative fluorescence intensity of engulfed beads per cell, were measured using flow cytometry (Supplementary Figure S2B). We observed an increase in: i) the percentage of phagocytic ATM KO cells compared to WT, over a 24 h time period, using immunofluorescence and flow cytometry ( Figure 1D and 1E), and ii) the fraction of highly phagocytic ATM KO cells, which engulfed more than 2 beads per cell, at different time points compared with WT ( Figure 1F).
The kinetics of phagocytic changes was linear in both cell lines over a 24 h period. However, to exclude any early apoptotic changes due to uptake of non-digestible substrates, the assays were carried out for 6 h. Also, given that the changes in phagocytic activities were modest, possibly due to the relatively large size of the substrates, levels of phagocytosis were used as the primary readout in subsequent assays. Upon re-expression of ATM, enhanced phagocytic properties of ATM KO cells were rescued ( Figure 1G; Supplementary Figure S2C). Similar to ATM KO microglia, increased levels of phagocytosis were observed in HMC3 and C20 human microglia, in which ATM was knocked down using siRNA (Supplementary Figure S2D-G). These data indicate that ATM-deficient microglia are more efficient phagocytes than their WT counterparts, and this effect is specific to ATM loss and cell line independent.
To investigate whether enhanced phagocytic properties of ATM-deficient microglia are mediated via the loss of ATM protein or its kinase activity, WT cells were treated with a reversible ATM inhibitor, AZD1390 (29), for up to 9 days (Supplementary Figure S3A and S3B).
To rescue the effects of ATM inhibition, AZD1390 was washed out and the cells were allowed to recover for 2 days (Supplementary Figures S3A and S3B, Release). The levels of ATM autophosphorylation and CHK2 phosphorylation following treatment with camptothecin indicated that kinase inhibition and inhibitor wash-out were successful (Supplementary Figure S3A). We then determined phagocytosis levels of AZD1390-treated WT cells and respective DMSO-treated controls. Although no significant changes in phagocytosis levels were observed at day 1 of inhibitor treatment, we found a 2-2.5-fold increase at days 4-6 in the ATM-inhibited cells compared to the control (Supplementary Figure S3B). It is likely that the long-term effects of ATM loss/inhibition, such as the accumulation of oxidative stress and DNA damage, drive the changes in phagocytic properties. Importantly, inhibitor removal resulted in a reduction in phagocytosis in ATM inhibitor-treated cells back to the levels of DMSO-treated microglia (Supplementary Figure S3B, Release).
Together, these data indicate that loss of ATM results in persistent microglial activation, which is characterised by increased expression of pro-inflammatory cytokines and enhanced ability to engulf synthetic substrates. These phenotypes are specific to ATM and driven by long-term changes associated with loss of ATM and/or its kinase activity.

The non-canonical RELB/p52 NF-κB pathway is activated in ATM-deficient microglia
The NF-κB proteins are central mediators of inflammation in response to tissue damage and infection (30). If mis-regulated, normally protective NF-κB-mediated pro-inflammatory responses can amplify acute or chronic tissue damage, thus driving autoinflammatory and autoimmune disease (31). NF-κB family members include RELA (p65), RELB, c-REL, p50 and p52. The canonical NF-κB pathway provides a rapid and transient response to cytokines, mitogens, and growth factors. These stimuli activate the IκB kinase (IKK) complex, which phosphorylates IκBα, inducing its proteasomal degradation. Loss of inhibitory binding of IκBα to NF-κB proteins results in nuclear translocation of the canonical NF-κB dimers, the most abundant of which are p65-p50 and c-REL-p50. The precursor protein p105 also acts in the NF-κB-inhibitory manner and becomes active upon proteolytic processing into p50 upon stimulation (31).
Microglial activation and neuroinflammation are linked to the activities of the p65-p50 heterodimer (32). In addition, p65 is localised to the nucleus in rodent models of A-T (11,12).
We therefore investigated whether the canonical NF-κB pathway is activated in ATM-deficient human microglia. We first tested whether canonical NF-κB signalling can be induced in WT microglia upon treatment with the tumour necrosis factor TNFα, a well-established activator of the pathway (31). We observed TNFα-dependent: i) phosphorylation of p65 at S536 indicative of its activation (33) and relocalisation of p65 in the nucleus (Figure 2A and 2B; Supplementary Figure S4A and S4B), and ii) processing of p105 to p50 and nuclear translocation of p50 in WT cells ( Figure 2B). These data indicate that the canonical NF-κB pathway is inducible in WT microglia. In contrast, while we observed a modest increase in: i) basal p65 phosphorylation at S536 and ii) processing of p105 into p50 in ATM KO cells as compared to WT, we failed to detect relocalisation of p65 and p50 into the nucleus (Figure 2A and 2B; Supplementary Figure S4A and S4B). A similar lack of nuclear translocation of p65 was seen in ATM KO C20 microglia, confirming that the effect was not cell-line specific (Supplementary Figure S4C-E). To investigate the reasons of this effect, protein levels of the NF-κB-inhibitory protein IκBα were determined. We found that IκBα levels were only moderately reduced in ATM-deficient microglia as compared to WT cells ( Figure 2A), and propose that the remaining IκBα is sufficient to retain the p65-containing complexes in the cytoplasm.
Additionally, as c-REL-containing heterodimers play a role in the canonical NF-κB response, the cellular localisation of c-REL was determined in WT and ATM KO HMC3 microglia (31).
Although we failed to observe nuclear translocation of c-REL following TNFα treatment ( Figure 2A and 2B), HMC3 cells were previously shown to have proficient c-REL signalling (34).
Similarly to p65, c-REL was retained in the cytoplasm, confirming that the c-REL-dependent NF-κB pathway is not activated in ATM-deficient microglia ( Figure 2B). Together, these data indicate that proteasomal degradation of IκBα and the nuclear translocation of p65, p50 and c-REL serve as limiting factors to canonical NF-κB activation in the absence of ATM.
Non-canonical NF-κB signalling is activated downstream of cell surface receptors of the tumour necrosis factor receptor (TNFR) superfamily, such as BAFFR, CD40, LTβR, and RANK.
Pathway activation is dependent on stabilisation of the NIK (NF-κB-inducing) kinase, which phosphorylates IKKα, promoting IKKα-mediated phosphorylation of p100 and its cleavage-dependent processing to p52. RELB-p52 is the major non-canonical heterodimer that drives transcriptional programmes (35).
To investigate whether the non-canonical NF-κB pathway is activated in ATM-deficient microglia, we first determined the levels of NIK and the efficiency of p100 processing by immunoblotting. NIK protein levels were increased in the ATM KO cells, compared to WT, indicating its enhanced stability ( Figure 2A). In addition, a reduction in basal levels of p100 and a concurrent increase in p52 levels were observed in ATM KO microglia compared to the WT control, indicating enhanced p100 processing (Figure 2A and 2B). We also determined the basal levels of RELB and its cellular localisation in WT and ATM KO HMC3 cells ( Figure 2E). Finally, a similar basal increase in nuclear RELB was seen in ATM KO C20 microglia (Supplementary Figures S5B and S5C).
To confirm and extend these results, RELB levels and localisation were determined in WT cells treated with the ATM kinase inhibitor, AZD1390, for up to 9 days (Supplementary Figure S3C).
A gradual increase in RELB expression and nuclear translocation was observed at days 3-6 of treatment, reaching a plateau at days 6-9. No changes were seen in the DMSO-treated control (Supplementary Figure S5D). Together, these results unequivocally show that loss of ATM function results in specific activation of the RELB/p52 non-canonical NF-κB pathway in human microglia.

The non-canonical NF-κB pathway controls microglial activation in the absence of ATM
Having observed the stabilisation of NIK kinase and nuclear translocation of RELB, but not p65 or C-REL, in the absence of ATM, we aimed to determine whether RELB-containing complexes might be responsible for microglial activation. Therefore, the phagocytic properties and expression levels of several pro-and anti-inflammatory cytokines were investigated in WT and ATM KO microglia following siRNA-mediated knockdown of RELB. In addition, siRNA that targets all five NF-κB subunits (NF-κB pan ) was used to probe for synergistic effects of the canonical and non-canonical pathways ( Figure 3A-C). The knockdown of RELB significantly reduced phagocytosis levels in ATM KO cells to levels typical of WT cells. No effect of RELB loss on phagocytic properties of WT cells was observed, indicating that the phagocytic properties of ATM-deficient microglia are modulated via RELB ( Figure 3B).
Next, we studied the effect of RELB depletion on cytokine expression. As expected, siRNA transfection moderately enhanced the fold change difference in mRNA levels of pro-inflammatory cytokines IL6, IL8, TNFA, CCL5 and IL1B but not anti-inflammatory IL4 in ATM KO vs WT control, as compared to that seen in unchallenged cells ( Figure 3C, compare Ctrl siRNA-treated ATM KO vs WT cells with the results in Figure 1C). Importantly, upon RELB knockdown, the high expression levels of NF-κB-dependent cytokines IL6, IL8 and TNFA in Ctrl-siRNA treated ATM KO microglia was reduced to similar levels as observed in Ctrl-siRNA treated WT cells ( Figure 3C). Expression of the chemokine CCL5 was reduced in the absence of ATM and RELB but to a lesser extent, in agreement with it being cooperatively regulated by NF-κB and interferon-regulatory factors IRF1, IRF3 and IRF7 (36). In contrast, mRNA expression levels of IL1B that acts upstream of NF-κB (37), and anti-inflammatory IL-4 (38), remained unchanged in ATM KO cells treated with RELB siRNA ( Figure 3C). One important observation is that both phagocytic properties and expression of pro-inflammatory cytokines were reduced at similar levels following treatment of ATM-deficient microglia with siRNA against RELB and all NF-κB subunits, confirming the dependency of these phenotypes on RELB-containing complexes ( Figure 3B and 3C).
These data demonstrate that the RELB/p52 non-canonical NF-κB pathway promotes microglial activation in the form of enhanced phagocytic clearance and expression of pro-inflammatory cytokines in the absence of ATM. We therefore further investigated the mechanisms of RELB/p52 activation in ATM-deficient microglia.

RELB/p52-dependent activation of ATM-deficient microglia is mediated by NIK kinase
To determine whether enhanced RELB/p52 signalling in the absence of ATM is dependent on NIK kinase, we used siRNA to knock down NIK and investigated the microglial activation phenotypes ( Figure 4A-D). As expected from the dependence of processing of p100 to p52 on the NIK-IKKα axis (35), NIK downregulation resulted in the modest accumulation of p100 and consequent reduction in p52 protein levels in both WT and ATM KO cells ( Figure 4A and 4B).
Importantly, NIK loss partially rescued: i) the nuclear translocation of RELB ( Figure 4A), ii) the expression of NF-κB-dependent pro-inflammatory cytokines IL8, TNFA and CCL5, whereas there was no effect on the levels of anti-inflammatory IL4 ( Figure 4C), and iii) the levels of phagocytosis ( Figure 4D) in ATM KO, and not WT microglia. These data show that RELB/p52 non-canonical NF-κB signalling and microglial activation in the absence of ATM are dependent on NIK kinase and are likely to be regulated, at least in part, via extracellular signalling cues.

Persistent DNA damage promotes RELB/p52 NF-κB signalling and microglial activation
The continuous occurrence of DNA damage requires proficient DNA damage signalling and repair, including ATM-dependent responses, to maintain genome stability. A-T cells are therefore characterised by enhanced oxidative stress and delayed repair of a subset of specialised DNA lesions (2,3,39). Importantly, DNA damage is thought to activate the non-canonical NF-κB pathway, although the mechanism of this activation is unclear (30). We therefore set out to investigate whether DNA damage might drive RELB/p52-dependent activation of microglia.
To determine whether ATM KO microglia possess increased levels of spontaneous DNA damage, we analysed the cells for several DNA damage markers. Protein poly[ADP-ribosyl]ation (PAR) and phosphorylation of histone variant H2AX at S139 (γH2AX), prototypical signalling events that occur during the DNA damage response (4,40), were increased in ATM-deficient cells ( Figure 5A). Additionally, the basal levels of DNA damage were directly measured using alkaline single-cell gel electrophoresis, which detects alkali labile sites, DNA single-and double-strand breaks (41). Basal levels of DNA damage were elevated in ATM KO microglia compared to the WT control ( Figure 5B). We also analysed the  Figure 5F). The kinetics of DNA damage-induced changes in nuclear translocation of RELB and phagocytosis was similar to that observed in response to long-term treatment with AZD1390 (Supplementary Figure S3B and S5D). These results indicate that persistent low-level DNA damage associated with ATM loss underpins microglial activation mediated by the non-canonical RELB/p52 pathway.

ATM-deficient microglia excessively engulf neurites
In addition to secretion of pro-inflammatory cytokines and oxidative species, chronically activated microglia can contribute to neurodegenerative pathologies by: i) defective phagocytosis of dead and dying cells that leads to uncontrolled release of neurotoxic compounds and ii) excessive phagocytosis of neuronal synapses, axons and dendrites (43)(44)(45)(46).  To investigate the types of phagocytic events that occurred in co-cultures of WT or ATM KO microglia with WT or ATM-deficient neurons we used confocal microscopy. ATM-deficient microglia were found to uptake both apoptotic (positive for cleaved caspase-3) and viable (negative for cleaved caspase-3 and non-pyknotic) neuronal soma, although both types of events were relatively rare ( Figure 6C). To determine whether some of the material readily phagocytosed by the ATM KO microglia included neurites, neuronal networks were visualised using an antibody against β3-tubulin following co-culture with CFSE-labelled microglia ( Figure 6D). We observed that the ATM KO microglia demonstrated a significantly increased uptake of neurites (DAPI negative, β3-tubulin positive) compared to neuronal soma (DAPI positive, β3-tubulin positive) ( Figure 6E). In addition, microglial clusters that formed specifically in co-cultures with neurons were detected (Supplementary Figure 7G). The number of microglial clusters that contained phagolysosomes showed a tendency to increase in co-cultures of both WT and ATM-deficient neurons with ATM KO, but not WT, microglia (Supplementary Figure 7H). Importantly, clusters of ATM KO microglia, compared with WT, were associated with disruptions in the neuronal network and damaged neurites when cultured with ATM-deficient neurons. Similar effects, albeit to a lesser extent, were observed in the case of healthy neurons ( Figure 6D and 6F). Taken together, these data demonstrate the intrinsic capacity of activated ATM KO microglia to aberrantly engulf neuronal soma and neurites in vitro.

DISCUSSION
A-T is a prototypical genomic instability disorder characterised by progressive cerebellar neurodegeneration. Neuroinflammation associated with genomic instability is increasingly recognised as a possible driver of progressive neurodegeneration (48). Previously, activation of microglia was observed in rodent models of A-T (10-12). In addition, morphological observations indicative of microglial activation were made in post-mortem brain samples from individuals with A-T (49). Here, using human cell models, we show that ATM-deficient microglia exhibit multiple features of activation, including increased expression of pro-inflammatory cytokines, cell surface CD40 and lysosomal CD68 markers and increased phagocytic clearance.
In the context of increased oxidative stress and DNA damage associated with ATM deficiency, the NF-κB proteins represent the fundamental mediators of inflammation. Surprisingly, although the canonical NF-κB pathway was inducible in both WT and ATM KO cells, we did not observe substantial degradation IκBα or the nuclear translocation of p65 or p50 in the absence of ATM. These data are consistent with previous observations that ATM plays multifaceted roles in activation of the IKK complex in response to genotoxic stress (30). For example, cytoplasmic ATM is essential for proteasomal degradation of IκBα via binding and recruitment of the SKP1/CULL 1/F-box protein-βTRCP (SCF βTRCP ) E3 ubiquitin ligase complex to phosphorylated IκBα (50). In contrast, nuclear localisation of RELB, stabilisation of the NIK kinase, and proteolytic processing of p100 to p52 were apparent in ATM-deficient microglia, confirming basal activation of the non-canonical NF-κB pathway. Notably, SCF β-TRCP is also involved in ubiquitin-mediated proteolysis of p100 to p52. However, the p100-SCF β-TRCP complex formation was not defective in the absence of ATM (50).
At the present time, the role of non-canonical NF-κB signalling in microglia is poorly understood. We find that microglial activation in the absence of ATM, including pro-inflammatory gene expression signatures and enhanced phagocytic clearance, was dependent on RELB and mediated by NIK kinase. In ATM-proficient microglia, the RELB/p52 signalling is activated following induction of low-levels unrepaired DNA damage. The results lead us to suggest that specific activation of the RELB/p52-mediated non-canonical NF-κB pathway in microglia is linked with persistent DNA damage associated with loss of ATM function ( Figure 6G). The detailed mechanistic understanding of the primary events that trigger this activation is presently unclear. One possibility is that aberrantly localised cytosolic DNA, which arises from micronuclei and damaged mitochondria in A-T (10,12,13,15), can stimulate the non-canonical NF-κB pathway. However, mechanistic links between cytosolic DNA signalling and the non-canonical NF-κB pathway are yet to be defined (48). Alternatively, poly[ADP-ribose] polymerase-1, a DNA strand break sensor, which cooperates with ATM to regulate the canonical NF-κB pathway, might play a role in RELB/p52 NF-κB signalling in A-T (51,52).
The non-canonical NF-κB pathway promotes cell survival albeit at the expense of disease-promoting chronic inflammation and autoimmunity (35). Activation of canonical p65 signalling represses expression of anti-apoptotic genes following DNA damage (53). In addition, NIK-and RELB/p52-mediated pathways supress p65-dependent production of type I interferons and pro-inflammatory cytokines (54,55). It is therefore possible that RELB/p52-mediated signalling in the absence of ATM provides a tolerance mechanism for limiting p65-driven pro-apoptotic responses. This mechanism is nevertheless accompanied by aberrant clearance of neurites, and to a much lesser extent neuronal soma, by ATM KO, but not WT, human microglia in the context of chronic inflammation. Our results indicate that highly phagocytic clusters of ATM KO microglia are associated with, and may drive, the damage of neuronal network via aberrant phagocytosis. Interestingly, clusters of activated microglia were previously detected in the cerebellum of A-T patients, although their function was not determined (49). Importantly, aberrant microglial pruning of neurites and synapses drives neurodegeneration in murine models of progranulin deficiency, Alzheimer's disease and multiple sclerosis, and blocking microglial phagocytosis mitigates the pathological consequences (56-58).
The possibility that clusters of highly phagocytic ATM KO microglia contribute to neuronal damage via local secretion of cytokines cannot be excluded. Indeed, in murine model of A-T, microglia-induced neurotoxicity was dependent on secretion of IL-1β (12). We propose that initial activation of the non-canonical NF-κB pathway in the presence of unrepaired DNA damage could promote the release of cytokines in the extracellular compartment, thus establishing the local pro-inflammatory environment. Such inflammation could lead to cellular stress and cooperate with the aberrant phagocytic activities of microglia ( Figure 6G).
Our findings that activation of ATM KO microglia is partially rescued NIK kinase depletion confirm the contribution of cell surface receptor-cytokine interactions to inflammation. These interactions may involve the TNFR receptor CD40, expression of which is increased in ATM KO microglia. The result of this positive feedback loop in A-T microglia would be the amplification of non-canonical NF-κB signaling and chronic neuroinflammation contributing to disease progression. The phagocytic clearance of neuronal material would further perpetuate the non-canonical NF-κB signalling pathways and inflammation ( Figure 6G).
The observation that loss of ATM drives normally neuroprotective microglia to become dysfunctional, are in general agreement with the neurotoxicity of Atm-inhibited murine microglia (12). There are, however, mechanistic differences. In mouse models of A-T, neurotoxicity is dependent on secretion of IL-1β by microglia in the absence of contact with neurons and is controlled by the STING-p65 canonical NF-κB pathway. In contrast, in human ATM KO microglia, we failed to observe global damage of ATM-deficient neurons via secretion. One possibility is that the long-term use of an ATM inhibitor, which acts as a dominant negative mutant of ATM (59), exacerbated the toxicity of murine microglia and contributed to neuronal susceptibility to inflammation. In the present work, the levels of phagocytosis of both damaged and live neuronal material by ATM KO microglia were on average 40-50% higher compared to WT cells. Although the contribution of ATM KO microglia to neuroinflammation and phagocytosis may appear to be modest, we suggest that even small increases in dysfunctional microglial clearance could lead to damage of PC and granule cells in a human brain. These findings are consistent with the slow progressive nature of neurodegeneration observed in individuals with A-T. Human microglia are highly immune reactive compared to their rodent counterparts (60). The differences in the threshold for execution of pro-survival strategies may explain why our findings of non-canonical NF-κB activation in ATM-deficient human microglia differ from the previously observed activation of p65 in rodent models of A-T (10,12).
Microglia are known to dynamically interact with both the soma of PCs and their dendritic arborisations in vivo, and there is increasing evidence that these interactions might be dysfunctional in disease (61). For example, in a mouse model of Niemann Pick Type-C disease, which is associated with genetic defects in lysosomal storage, activated cerebellar microglia showed increased engulfment of PC dendrites, contributing to neuronal degeneration (62).
Interestingly, microglia show high regional diversity, with cerebellar microglia being the most distinct based on their gene-expression signature. They exist in a highly immune-vigilant state, and this profile is further exacerbated with ageing (60). In addition, microglia in the cerebellum are enriched in genes associated with active cell clearance (63). Indeed, it is tempting to speculate that persistent DNA damage associated with the loss of ATM might exacerbate the phagocytic abilities of highly reactive cerebellar microglia in vivo, resulting in excessive clearance of neurites that contributes to neurodegeneration in A-T.

Reagents and materials
Reagents and materials used in this work are described in Supplementary Table S1.

Cell culture
Immortalised human fetal microglial HMC3 cells were established by Marc Tardieu and kindly provided by Brian Bigger (University of Manchester) (20). HMC3 cells were grown in DMEM (Gibco; 4.5 g/L glucose, no pyruvate) supplemented with 10% heat-inactivated fetal bovine serum (HI-FBS; Merck). Immortalized human C20 microglia were kindly provided by David Alvarez-Carbonell (21). C20 cells were grown in DMEM:F12 (Lonza) supplemented with 1% N-2 (ThermoFisher) and 1% HI-FBS. Microglia were routinely seeded at the density of 3,000-6,000 cells/cm 2 reaching about 80% confluency at the end of an experiment. All cells were cultured at 5% CO2, 37°C and 95% humidity. Cell lines were routinely tested for viability and mycoplasma, and kept in culture for no longer than 40-50 population doublings. The cells were selected using 250 ng/mL hygromycin B for 7 days and maintained as a polyclonal population in standard growth medium supplemented with 50 ng/mL of hygromycin B. Expression of ATM was induced using 25-100 ng/mL doxycycline.

RNA interference
Cells were transfected with 50 nM final siRNA for 72 h using Lipofectamine RNAiMax according to the manufacturer's guidelines (Thermo Fisher Scientific). siRNA Oligonucleotides were synthesised by Merck (sequences are given in the Supplementary Table S1).

Cell treatments
To induce NF-κB signalling, human recombinant TNFα (PeproTech) was used at 50 ng/mL for

Protein extraction, biochemical fractionations and immunoblotting
Whole-cell and cytoplasmic and nuclear extracts were prepared as previously described  Table S1. RT-qPCR was performed using qPCRBIO SyGreen Blue Mix Lo-ROX (PCR biosystems) and the QuantStudio 5 Real Time PCR system (ThermoFisher).
The data were analysed using QuantStudio Design and Analysis software (ThermoFisher). The following cycling program was used: initial denaturation and polymerase activation at 95°C for 2 min; 40 cycles with 5 s at 95°C followed by 30 s at 65°C; followed by a melting curve step (15 s at 95°C; 1 min at 60°C; 15 s at 95°C). Ct values for all mRNAs of interest were normalised to RPS13 or IPO8 reference mRNA and the data were expressed as fold-change relative to the control using the 2 -DDCt method.

Comet (alkaline single cell electrophoresis) assays
Suspension cells were left untreated or treated with 2 μM hydrogen peroxide for 5 min on ice (positive control) and embedded in low-melting point agarose for 2 min on a microscope slide.

Phagocytosis assays
At 24 h prior to phagocytosis assays, the growth medium on HMC3 microglia was replaced with 0.5% HI-FBS DMEM. Fluorescent carboxylated 5 µm beads (Spherotech) were added to cells at a ratio of 5:1 and incubated for 6 h at 37°C, 5% CO2 and 95% humidity. As a negative control, the assays were carried out in the presence of 1 µM Cytochalasin D. Cells were In all immunofluorescent analyses, images of at least 5 fields of view (field of view: 318 x 318 µm) per coverslip with at least a total of 100 cells were taken in a blinded manner using an epi-fluorescent Zeiss Axio Observer Z1 (images in a single Z-plane) or a confocal Nikon Eclipse Ti (compressed Z-stack images) microscope. Confocal imaging was sequential for different fluorophore channels to obtain a series of axial images. A secondary antibody only control was used to subtract the non-specific background during quantifications. Analyses were performed using ImageJ.
To determine the levels of CD68, integrated intensity was measured using compressed Z-stack images.
The percentage of cells containing nuclear p65 or RELB staining was determined using nuclear (defined using DAPI) vs cytoplasmic (defined using vinculin) MFI.
Phagocytic events in microglial-neuronal co-cultures were identified using compressed Z-stack confocal images and XZ/YZ orthogonal projections. Phagosome-like structures, which were identified by the disruption of microglial CFSE staining, containing neuronal material were quantified in a randomly acquired tile of approximately 7 x 5 fields of view (2194 x 1567 µm). Events positive for both cleaved caspase-3 and CTV were considered as the uptake of apoptotic neurons, whereas events negative for cleaved caspase-3 and positive for CTV included phagocytosis of healthy cells and/or their parts. Additionally, events positive for β3-tubulin and DAPI represented the uptake of neuronal soma, whereas events positive for β3-tubulin and negative for DAPI were considered as the uptake of neurites. The relative frequency of these events was quantified. Microglial clusters were defined as a minimum of five proximal cells as above. Colocalisation of microglial clusters and the damaged neuronal network was quantified.

Quantification and statistical analysis
Quantitative data are expressed as the mean of at least 3 independent biological experiments ± standard deviation (SD) unless stated otherwise. The data were tested for normality using the Shapiro Wilk test and the analysis informed the choice of parametric or non-parametric statistical tests. When normalisation of data to the respective control was carried out, a one-sample t-test with a theoretical mean of 1 was used. For comparison of two samples, paired or unpaired two-tailed t-test with Welch's correction for unequal variances was used.
To compare three or more unmatched samples, a one-or two-way ANOVA was used. The comparison of data obtained from different fields in immunofluorescence was carried out using a non-parametric Wilcoxon test (paired samples -variables counted from the same field). The exact p-values are indicated in figure legends, ns -not significant (p ³ 0.05). The analyses were performed using GraphPad Prism.