Background
Microglia are the primary innate immune cells of the central nervous system (CNS). These yolk sac-derived myeloid cells take up residence in the CNS during embryogenesis [
1]. Microglia perform both homeostatic functions such as synaptic pruning and environmental sensing and under disease states, can adopt transcriptionally distinct profiles that can mediate both neurotoxic as well as neuroprotective and phagocytic roles [
2‐
4]. In aging and in chronic neurodegenerative diseases such as Alzheimer’s disease (AD), microglia adopt unique disease-associated-microglial (DAM) transcriptional profiles and it is suspected that DAM can mediate phagocytic clearance of amyloid beta (Aβ), a process that becomes progressively dysregulated and ineffective with disease progression [
5,
6]. Although transcriptional profiling provides an in-depth understanding of gene-expression changes in disease, modest agreement between transcriptional level changes and protein level changes have been observed across various studies and cell types [
7,
8]. Since functional phenotypes are largely mediated by protein effectors, proteomic studies should provide an important perspective for understanding pathophysiological changes that result in neurodegeneration. Recent advances in mass spectrometry methods have made deep proteomic profiling of relatively smaller numbers of cells feasible [
9].
The proteome of microglia from young and aging wild-type (WT) mice has been described [
8,
10], and more recently, a limited proteomic analysis of acutely-isolated microglia from the 5xFAD mouse model of AD pathology identified early pro-inflammatory microglial activation prior to Aβ deposition [
11]. However, comprehensive and comparative proteomic studies of adult microglia under homeostatic conditions, acute neuroinflammatory and chronic neurodegenerative disease states have not been reported. We report the deepest quantitative and most comprehensive proteome of acutely isolated microglia derived from adult mice and from mouse models of acute neuroinflammation and AD pathology. We report unexpected similarities between murine models of acute neuroinflammation and chronic neurodegeneration, in addition to highlighting unique neurodegeneration-specific immune changes seen in 5xFAD mice. We also highlight the complementary nature of transcriptomic and proteomic data in understanding the dynamic nature of microglial gene and protein expression changes under different in-vivo conditions. Furthermore, we identify unappreciated microglia-specific markers relevant to human AD as and suggest novel AD-relevant biology for proteins that were not thought to be expressed by microglia. Therefore, this microglial proteome represents a valuable resource to the neuroimmunology and neurodegenerative disease research communities, serving as a reference for future comprehensive immune profiling and target validation studies.
Discussion
Advances in deep RNA sequencing and proteomic methods have provided unprecedented opportunities to characterize the molecular and cellular complexity of the CNS [
5,
8,
9,
25,
32,
38,
39]. Microglia are the principal immune effectors in the CNS [
1] that perform homeostatic functions under normal conditions but adopt diverse activated profiles as revealed by recent whole-cell and single-cell transcriptomic studies [
5,
6,
8,
25]. Several transcriptomic changes in microglia, however, may not reflect changes at the protein level due to post-transcriptional events that further regulate protein expression and degradation, leading to modest levels of agreement between transcript and protein expression data [
7,
8,
32]. Primary microglial proteomes from 8-week-old wild-type mice [
8] and aged mice [
10] as well as young adult 5xFAD mice (2–12 weeks of age) have been published [
11]. However, proteomic profiling of microglia from neuroinflammation models and later time points in neurodegenerative disease models has not been reported, representing a significant gap in the field. We present a comprehensive proteome of acutely isolated CD11b
+ microglia from older adult mice (6–7 mo og age) under normal, acute neuroinflammatory and chronic neurodegenerative disease conditions and highlight differentially expressed proteins specific to each state with relevance to human AD.
By contrasting microglia-specific changes in 5xFAD mice to existing human brain proteomes, we have found that App, Apoe, Clu, Htra1 and Cotl1 were among the most highly upregulated proteins in microglia isolated from a mouse model of AD pathology. Of these, App, Apoe, Clu and Htra1 have known AD-relevant biology although their expression and role within microglia in AD is not clear [
36,
40,
41]. Increased expression of Apoe, Htra1 and Clu proteins, but not of Cotl1, has been reported in 5xFAD microglia as early as 10 weeks of age when Aβ deposition is first seen [
11]. Additionally, we identified Cotl1 as a microglia-specific marker with increased expression in human AD. Cotl1 was also found in a recent network analysis study of post-mortem human AD brains to be a key member of a glial module that is positively correlated with AD pathology [
32]. In this module, Cotl1 was co-expressed with leukotriene A4 hydrolase (Lt4ah). Since Cotl1 stabilizes and promotes the activity of lipoxygenase 5 (5-LOX) to convert arachidonic acid to LT-A4 [
30,
31], a co-upregulation of Lt4ah in AD microglia is predicted to facilitate LT-A4 conversion to LT-B4, a pro-inflammatory leukotriene implicated in neurotoxicity and Aβ formation [
42‐
44]. Another protein co-expressed with Cotl1 was PG-D2 synthase (Ptgds), an enzyme that catalyzes the conversion of PG-H2 to PG-D2 in the cyclooxygenase pathway of arachidonic acid metabolism [
45]. Non-specific upstream inhibition of cyclooxygenase-2 (COX2) as well as specific inhibition of PG-D2 signaling results in neuroprotection in animal models [
46,
47]. These converging lines of evidence suggest that at least some pro-inflammatory effects of microglia in AD may be mediated by LTB4 and PG-D2 [
45]. Since Cotl1 is involved in actin dynamics, immune synapse function and regulation of eicosanoid biosynthesis, we suspect that Cotl1 may regulate motility and neuroinflammatory functions mediated by microglia in AD [
30,
31,
45].
Apoe expression, like Cotl1, was also highly increased in 5xFAD microglia. This is consistent with observations from transcriptomic studies of microglia where Apoe was upregulated in neurodegenerative disease models and was identified as a marker of disease-associated-microglia [
5,
6,
20]. Our immuno-histochemical studies show that Apoe protein, unlike Cotl1, is not expressed by ramified/homeostatic microglia, but is seen within microglial processes that surround and infiltrate Aβ plaques in 5xFAD mice, and specifically co-localizes with intra-microglial Aβ. Since activated microglia engulf Aβ and Apoe strongly co-localizes with Aβ within microglia, our findings suggest that microglia may either phagocytose Aβ and Apoe that co-aggregate in the AD brain or that Apoe’s co-localization with Aβ indicates functional interaction between Aβ and Apoe during Aβ phagocytosis and clearance [
48,
49]. In support of the latter possibility are recent findings demonstrating the role of Apoe, along with Trem2, in the transition of microglia from homeostatic to DAM states in AD [
27,
29], as well as the consistent identification of Apoe within microglia in our as well as other microglia proteomes [
8,
10,
11].
In addition to validating specific proteins with increased expression in 5xFAD microglia, our bioinformatics analyses provide clues regarding the general metabolic and inflammatory states of microglia in AD. We observed that dysfunctional lipid metabolism and trafficking seem to be features of microglia in AD brain. Since dysfunction in lipid metabolism and clearance are implicated in AD pathogenesis [
50,
51], comprehensive lipidomic characterization of microglia from AD models may provide novel insights into mechanisms of impaired lipid clearance in AD. Our observation of general suppression of IL4-related pathways in 5xFAD microglia along with similarities between protein expression profiles of microglia in 5xFAD mice and LPS-induced neuroinflammatory states suggest that microglia in AD adopt a profile more consistent with that of pro-inflammatory microglia. However, the identification of p53 and AP2 as upstream transcriptional regulators of AD-associated microglial activation are distinct from regulators of canonical LPS-induced pro-inflammatory activation such [
52‐
54] suggesting that unique inflammatory activation pathways are also upregulated by microglia in AD models. Beyond changes in AD models, our results from LPS-treated WT mice also demonstrate robust effects of systemic inflammation on energy metabolism and oxidative stress in microglia. GO analyses showed that LPS-induced systemic inflammation suppressed mitochondrial oxidative phosphorylation in microglia. In LPS-treated mouse microglia, 21 mitochondrial proteins were reduced while cellular stress response proteins were upregulated, suggesting that stress-induced mitochondrial damage may occur in microglia in response to systemic inflammation. Furthermore, LPS upregulated proteins involved in focal adhesion formation, proliferation, exocytosis and apoptosis confirming a state of microglial activation [
13]. Since systemic LPS does not cross the blood brain barrier [
55], our observations are more likely to represent indirect effects of LPS (involving Tnf or other soluble factors) rather than direct toll-like-receptor-mediated effects of LPS on microglia [
56].
Our study also identified several highly abundant and microglia-specific proteins including Bin1, Msn, Cotl1 and Hexb [
3]. The identification of Bin1 as a highly abundant protein in our microglial proteome agrees with a reference mouse microglial proteome [
8] and is of particular relevance in AD [
57]. At the transcript level, Bin1 is highly and specifically expressed by homeostatic microglia and Bin1 down-regulation occurs as microglia adopt DAM profiles in neurodegenerative disease [
5,
25]. Although Bin1 polymorphisms have been identified as independent risk factors for late-onset AD [
57], the pathophysiological roles of Bin1 in AD are unclear. Based on our findings that are consistent with existing expression data for Bin1 in the brain [
8,
25], it may be particularly important to further investigate the role of Bin1 in microglia-mediated neuroinflammatory mechanisms in neurodegenerative disorders. We also identified Hexosaminidase B (HexB) as abundant microglial protein that was also previously shown to be microglia-specific at transcriptomic and proteomic levels [
3,
5,
8,
25]. We observed that HexB was upregulated by 5xFAD microglia. Hexb is an endosomal-lsysomal enzyme that regulates ganglioside metabolism and is involved in autophagy [
58]. Hexb mutation results in a rare lysosomal storage disorder called Type II GM2 gangliosidosis or Sandhoff disease that is characterized by progressive hearing loss, seizures, intellectual disabilities and motor deficits [
59]. In the Hexb knock-out mouse, accumulation intra-neuronal Aβ-like and synuclein-like protein aggregates [
58]. Our findings of Hexb upregulation in microglia from AD mouse models further support the possibility that dysregulation of lipid metabolism, lysosomal functions and autophagy in microglia may contribute to AD neuropathology [
60‐
62].
Despite the identification of microglia-specific proteins discussed above, some of the more well-known microglial proteins such as Tmem119 and P2ry12 were not identified in our study [
63,
64]. It is also interesting to note that the reference microglial proteome [
8] did not identify Tmem119 in microglia even though Tmem119 protein expression by microglial has been confirmed by immuno-histochemical methods [
25,
64]. This may be explained by our use of whole cell lysates, rather than membrane-enriched fractions for our study, leading to a lower likelihood of identifying cell membrane anchored proteins. This bias in our dataset is further supported by the lack of enrichment of GO terms such as “cell membrane” in our analyses. Another possibility for the relative absence of membrane proteins in our study is potential loss of cell membrane fragments during the cell isolation process, a likely result of heavy myelination and architectural complexity of the aging mouse brain as contrasted with post-natal or younger adult mice. Since membrane receptors, ion channels and transporters may represent drug targets for neurodegeneration, future studies using more advanced proteomics technologies with deeper coverage coupled with enrichment of membrane-associated microglial proteins will likely help overcome this limitation.
Additional limitations of this study also deserve discussion. In addition to ubiquitous and microglia-specific proteins, several proteins highly expressed by astrocytes, oligodendrocytes and neurons were also identified in our microglial proteome despite > 95% enrichment of CD11b
+ cells in our samples. While this could indicate cell or protein contamination by adherent non-microglial cell fragments during CD11b
+ enrichment, this could also be indicative of constant pruning of neuronal synapses or dendrites as well as phagocytic uptake of cellular debris, myelin and astrocytic processes by microglia. Indeed, an age dependent increase in myelin basic protein (Mbp) peptides in mouse microglia has been previously reported [
65], which is more likely to explain why Mbp was highly abundant in our dataset (top 10 percentile). As compared to the reference proteome in which over 7000 proteins were identified across 4 cell types by label-free quantitative methods [
8], we identified 4133 proteins by TMT. This lower number of proteins identified could be attributed to lower microglial yield from the older mouse brain as compared to younger mice. Thus, pooling additional brain samples as well as enzymatic digestion prior to microglial isolation may be considered in future studies to improve protein yield. A more likely explanation for lower number of proteins identified is that our single batch of fractionated TMT, 10-plex-quantified by SPS-MS3 mass spectrometry did not contain non-microglial enriched samples from which to borrow or trigger identification and quantification of shared proteins which may be higher expressed in other cell types (i.e., neurons, oligodendrocytes and astrocytes). Lastly, our analyses using 3 pools per group were statistically underpowered to detect statistically significant changes using FDR corrections for pair-wise comparisons because of which we considered unadjusted
p-values < 0.05 as statistically significant. To overcome this inherent limitation, we applied an additional fold-change threshold of 25% which exceeded the overall coefficient of variation for all protein measurements (6–7%) in each of the three groups.