Harnessing Immunoproteostasis to Treat Neurodegenerative Disorders

Abstract

Immunoproteostasis is a term used to reflect interactions between the immune system and the proteinopathies that are presumptive “triggers” of many neurodegenerative disorders. The study of immunoproteostasis is bolstered by several observations. Mutations or rare variants in genes expressed in microglial cells, known to regulate immune functions, or both, can cause, or alter risk for, various neurodegenerative disorders. Additionally, genetic association studies identify numerous loci harboring genes which encode proteins of known immune function that alter risk of developing Alzheimer’s disease (AD) and other neurodegenerative proteinopathies. Further, preclinical studies reveal beneficial effects and liabilities of manipulating immune pathways in various neurodegenerative disease models. Although there are concerns that manipulation of the immune system may cause more harm than good, there is considerable interest in developing immune modulatory therapies for neurodegenerative disorders. Herein, I highlight the promise and challenges of harnessing immunoproteostasis to treat neurodegenerative proteinopathies.

Despite major advances in understanding key initiating events underlying various neurodegenerative disorders, there are limited efficacious treatment options for patients (Cummings et al., 2014; Golde, 2016). Approved therapies are primarily symptomatic in nature and do not dramatically modify disease course. With rare exceptions such as deep brain stimulation or dopamine replacement strategies for Parkinson’s disease (PD), the clinical impact of such therapies is underwhelming.

Over the last two decades, there have been huge investments towards the development of therapies for neurodegenerative diseases that target the proteins which aggregate and accumulate (i.e., the proteinopathy). Clinical data generated to date, primarily with respect to altering amyloid β protein (Aβ) deposition in Alzheimer’s disease (AD), indicates such therapies that are intended to target the triggering proteinopathy may have limited efficacy in symptomatic patients (Doody et al., 2014; Egan et al., 2018; Sevigny et al., 2016). Indeed, there is growing recognition and evidence from clinical trials that targeting the proteinopathies will have the best chance for showing efficacy if treatment is initiated well before disease-defining symptoms appear (Bateman et al., 2017; Golde et al., 2013; Reiman et al., 2011; Sperling et al., 2014). There is cautious optimism that such therapies intended to target early steps in the disease process, when tested in primary or secondary prevention trials, will have major impact on disease course; however, the timelines for possible regulatory approval and widespread deployment of effective prophylactic therapies remain uncertain (Golde, 2016).

Indeed, there is a bleak outlook for individuals who are currently in the symptomatic phase of a neurodegenerative disorder, as well as those who might develop these diseases in the near future because of advanced age, genetic risk or environmental influences. Given this unmet medical need, it is imperative to identify novel strategies to treat neurodegenerative disorders that can significantly alter the disease course, even if initiated in symptomatic patients. Development of such therapies is almost certain to be challenging given our imperfect understanding of the pathological events downstream of the triggering proteinopathy. Herein, I examine the rationale for one of these potential strategies, namely, targeting immunoproteostasis (Chakrabarty et al., 2015). My colleagues and I have used this term previously to capture the complex, multidirectional, interplay between the immune system and the underlying triggering neurodegenerative proteinopathy. As much of the attention in the field is centered on immunoproteostasis in AD, this Perspective will focus on the efforts to understand and harness the immune system for therapeutic benefit in AD, although illustrative examples from other neurodegenerative diseases are discussed. The Perspective is also primarily focused on the role of innate immunity; nevertheless, the more limited evidence regarding adaptive immune responses in neurodegenerative disorders will be briefly discussed.

There has been long-standing interest in determining the role of the immune system in neurodegenerative proteinopathies. Initial studies in this area largely focused on a descriptive association of the reactive gliosis and a cataloguing of alterations in expression or accumulation of innate immune signaling molecules in the brains of individuals who died from neurodegenerative disease (Akiyama et al., 2000; Eikelenboom et al., 1998; Heneka et al., 2015; Rogers et al., 1996). Although the general gestalt has been that pro-inflammatory activation of the innate immune system within the brain is harmful, a much more nuanced role of the immune system in neurodegenerative disease is warranted. Perhaps the most cautionary note in this vein, is that clinical trials of anti-inflammatory agents in AD have either shown no benefit or, in some cases, actually showed anti-inflammatory agents may have worsened decline (Aisen et al., 2003; Breitner et al., 2011; de Jong et al., 2008; Group, 2015; Group et al., 2007; Reines et al., 2004; Soininen et al., 2007). The side effects of anti-inflammatory therapy in these studies were also quite marked in these elderly trial participants. Thus, these trials may actually have done more harm than good.

The origins of the hypothesis that pro-inflammatory immune stimuli are harmful in AD.

The hypothesis that proinflammatory drivers played a harmful role in AD led to clinical trials of anti-inflammatory agents. Thus, it is useful to reevaluate the data and interpretation of the data that led to development of that hypothesis. The presence of gliosis (“activated” microglial cells and a robust astrocytosis) in the AD brain was a foundational observation on which the “proinflammatory stimuli are bad in AD” hypothesis was built (Akiyama et al., 2000). As a reactive gliosis is present in most, if not all, neurodegenerative disorders, these observations still contribute to the concept that proinflammatory factors are harmful in other neurodegenerative disorders.

For AD, another factor was the comparison of AD to Serum Amyloid A (SAA) amyloidosis. In SAA amyloidoisis, inflammation is essential for SAA amyloid deposition in the periphery, as pro-inflammatory stimuli rapidly and massively upregulate SAA mRNA and protein expression in the liver (Malle and De Beer, 1996; Sack, 2018). Following the cloning of the amyloid protein precursor (APP) gene that encodes Aβ and resides on human chromosome 21, a series of genetic and modeling studies emerged indicating that simply increased levels of APP or Aβ could lead to Aβ deposition (reviewed in (Hardy and Selkoe, 2002)). The most straightforward example of this is in human trisomy 21 (Downs’s syndrome). Individuals with trisomy 21 develop Aβ amyloid deposition relatively early in life and other features of AD pathology later (Delabar et al., 1987; Mann, 1988; Masters et al., 1985). This accelerated Aβ deposition is linked directly to increased expression of APP and increased Aβ levels. Thus, when subsequent data emerged showing that various proinflammatory stimuli could increase APP and Aβ levels, at least in cell culture systems (Griffin et al., 1998), the proinflammatory stimuli drives AD pathology by increasing APP and Aβ hypothesis emerged.

In parallel studies, several groups were beginning to evaluate immune molecules present within the brain (Akiyama et al., 2000; Eikelenboom et al., 1998; McGeer et al., 1994). These studies, likely biased towards evaluating immune factors normally associated with activating the immune response (e.g., complement, inflammatory cytokines), revealed that numerous acute phase reactants and other immune molecules accumulate within the AD brain. Further, these factors were either associated directly with amyloid plaques, present in the reactive microglial surrounding these plaques, or in reactive astrocytes. Additional studies demonstrated higher levels of peripheral proinflammatory cytokines in AD, or at least all-cause dementia, results that are now further supported by several meta-analyses (Koyama et al., 2013; Swardfager et al., 2010) . Perhaps the final and pivotal pieces of support for the proinflammatory hypothesis emerged when i) non-steroidal anti-inflammatory drugs (NSAIDs) use was reproducibly associated with decreased risk for AD (Breitner et al., 1994; Breitner et al., 1995; in ‘t Veld et al., 1998; McGeer et al., 1996; Stewart et al., 1997; Zandi and Breitner, 2001) and II) preclinical studies in mouse models suggested that select NSAIDs might lower Aβ amyloid pathology in mice (Jantzen et al., 2002; Lim et al., 2000). Despite the fact that NSAIDs use contains some risk, especially for elderly individuals, the aforementioned clinical trials were initiated in symptomatic AD and, subsequently, a prevention study.

Therapeutic clinical trials of NSAIDs in individuals with symptomatic AD did not mimic the setting that was likely to account for the possible epidemiologic association of decreased risk for development of AD (>2 year history of NSAID use at an indeterminate time prior to diagnosis of dementia) (in ‘t Veld et al., 1998; Stewart et al., 1997). Further, the specific NSAIDs used in those trials were not necessarily the drugs most likely to account for the epidemiologic association. Thus, many questions, but few firm conclusions, emerge from a retrospective examination of these studies. Notably, the initial epidemiologic association that drove the studies on NSAID use and AD risk was preceded by studies showing reduced risk of individuals with rheumatoid arthritis (RA) and AD (McGeer et al., 1996). This raises the issue of reverse causation. Did the well-catalogued proinflammatory status of individuals with RA, or other peripheral conditions associated with chronic long-term activation of the immune system, and not the NSAID per se, drive the association with reduced risk for AD? Indeed, individuals with a chronic inflammatory condition likely represent a large portion of those with the >2 year history of NSAID use most reproducibly associated with reduced risk AD. Further, NSAIDs typically only minimally alter the immune phenotype and those NSAIDs linked to the epidemiologic data are poorly brain penetrant (Eriksen et al., 2003; Kukar et al., 2005). Finally, additional data emerged that at least some NSAIDs could have other effects (e.g., selectively lowering production of the longer, more pathogenic forms of Aβ) that might account for their apparent benefit in preclinical models (Kukar et al., 2005; Weggen et al., 2001).

Whether the epidemiologic data indicates a possible protective role of NSAID use or the underlying conditions such as RA that led to long-term NSAID use, represents an unsolved dilemma for the field. A protective role for NSAIDs in AD, possibly through an anti-inflammatory mechanism of action, remains a marginally plausible hypothesis, but only in the setting of primary prevention. A protective role of the underlying condition that necessitate long-term NSAID use would conversely suggest that the proinflammatory state is protective. In other words, depending on how one views the epidemiologic data, one ends up with the exact opposite conclusion for how to move forward with a therapy modulating the immune system in AD.

Can we resolve this dilemma?

This retrospective analysis serves as an exemplar of the challenges faced with respect to developing and testing immune modulatory therapies for AD and other neurodegenerative disease based on our current knowledge. Indeed, long-term therapeutic modulation of the immune system in the elderly will almost invariably lead to a higher incidence of side effects than what might be observed with that same therapy in younger individuals, no matter whether the therapy is intended to be immune activating or suppressing. Many of these potential side effects will be due to effects on the peripheral immune status. Potential for untoward side effects might be reduced by therapeutics with a selective effect on immune activation states in the central nervous system (CNS), but such targeted approaches are largely conceptual at this time. Perhaps, anti-Aβ antibodies with activating Fc domains that preferentially bind Aβ aggregates in the brain, focally activate microglial cells, and reduce Aβ loads, serve as examples of such therapies (Ostrowitzki et al., 2017; Sevigny et al., 2016).

The current ambiguity with respect to directionality is highly problematic. If a therapy moves the activation state of the immune system in the wrong direction, then the therapy may actually accelerate the underlying neurodegenerative disease process. An emerging concept in the field is that a given immune activation state or immune manipulation may have different consequences on brain function, positive or negative dependent on the state of disease. This postulate is supported by limited, but nonetheless provocative, preclinical model data. For example, CX3CR1 or TREM2 knockout has been shown to have differential effects on amyloid and tau pathologies (Bhaskar et al., 2010; Lee et al., 2010; Leyns et al., 2017; Sayed et al., 2018). In another case, IFNγ overexpression has been shown to have a positive effect on amyloid pathology, but a negative impact on neuronal health (Chakrabarty et al., 2010; Chakrabarty et al., 2011). If bi-directional effects of an immune manipulation are conserved in humans, such data would suggest that it might be possible to intervene with an immune modulation at a defined disease stage (e.g., amyloid positive, tau negative) (Jack et al., 2016). Such stage-dependent treatment theoretically makes some sense, but is complicated by the fact that disease-state is heterogeneous in almost all neurodegenerative disorders across various CNS regions. Further, the knockout studies in mice that support this concept of “stage-specific” targeting of the immune system do not reflect a real-world intervention that would likely begin long after the pathology has emerged. Additionally, a thorough mechanistic understanding of the biology underlying these differential effects is lacking. For all of these reasons, clinical development of immune interventions that have beneficial effects on one pathology, but harmful effects on another in preclinical models should proceed very cautiously. It will be challenging to predict the likelihood that beneficial effects would outweigh negative effects.

With these caveats and concerns in mind, there are a number of paths forward that can help inform whether i) there are truly targets in the immune system that would provide disease modification in neurodegenerative proteinopathies and ii) determine directionality with respect to altering these targets. In the following sections. I will discuss these complimentary paths and review the progress that has been made in these areas.

Functional genetics studies can inform the directionality of immune therapies.

A key step in both identifying potential targets for immune therapies and whether agonism or antagonism of the target is desirable is to better understand how variants and mutations in immune genes implicated in altering risk for neurodegenerative disease alter immune function (see Table 1 and 2). AD and PD genome wide association studies (GWAS) and more recent meta-analyses of these GWAS implicate a number of loci that contain immune genes in mediating risk (Chang et al., 2017; Hamza et al., 2010; Hill-Burns et al., 2011; International Genomics of Alzheimer’s Disease, 2015; Lambert et al., 2010; Lambert et al., 2009; Nalls et al., 2014; Reitz et al., 2013; Wissemann et al., 2013; Witoelar et al., 2017). Unfortunately, there remains limited biological insight into how variants at these loci alter immune function, though some clues are emerging (Bradshaw et al., 2013; Brouwers et al., 2012; Carrasquillo et al., 2017; Huang et al., 2017; Katsumata et al., 2018; Siddiqui et al., 2017). Indeed, in many cases, though an immune gene is embedded within the region implicated by the GWAS association, the locus implicated is typically larger than a single gene and may contain both genes with known or purported immune function and genes with other functions. Of course, there are also examples outside of neurodegenerative disease where a disease-associated polymorphism, can have long-range effects outside of the region of association (Smemo et al., 2014; Tung et al., 2014). Further, even though evidence of association with the HLA locus is cited as genetic support for immune link to AD and PD, numerous non-immune genes are imbedded within the human HLA locus – potentially confounding this assertion. In addition, though the HLA associations are used as supporting rationale for studying adaptive immunity in these diseases, both MHC class I and class II are constitutively expressed on cells in the brain and these molecules can play roles in innate immune signaling (Frei et al., 2010; Shatz, 2009). Despite these caveats, given the large number of loci in both AD and PD, which contain genes that regulate immune function, it is reasonable to pursue studies designed to link genetic variation with functional variation in immune function.

Table 1. Candidate genes with known or possible immune function in loci implicated by GWAS in AD and PD.^A.

GeneB	Disease	Evidence and Strength of Evidence for immune function in the brainC
CR1	AD	Strong, complement receptor, minimal expression in CNS, soluble form present in plasma and CSF, association region includes the CR1L gene, a homologue of CR1
INPP5D	AD	Strong, highly expressed in microglial cells, Is activated by immunoreceptor tyrosine-based inhibition motif receptors, acts as an intrinsic brake on activation signaling
HLA-DRB1	AD	Strong, one chain of the heterodimeric MHC class II molecule, highly expressed in microglial cells, region of association includes other MHC genes and non MHC genes, involved in antigen presentation, but may play a role in innate immune responses
TREM2	AD	Strong, an immunoreceptor tyrosine-based activation motif receptor, highly expressed on microglial cells (see main text for more details)
MS4A6A	AD	Strong, strongly expressed in microglial in the brain and in peripheral immune cells, but sparse functional data, other MS4A family members are within the region of association
SP1	AD	Strong, a transcription factor highly expressed in myeloid cells and microglial cells in the brain. C1qTNF4 a molecule with cytokine like functions is also within the region ofassociation.
BIN1	AD	Strong (peripherally) Weak (CNS), evidence for relatively high levels of RNA expression in microglial cells, indirectly influences function of immune cells in the periphery by regulating of expression of the immunomodulatory enzyme indoleamine 2,3-dioxygenase (IDO). IDO mRNA is barely detectable in the brain.
CLU	AD	Strong, a secreted chaperone, regulates Aβ deposition, regulates complement activity among many other activities, intracellular form can regulate NFKB
CD33	AD	Strong, an immunoreceptor tyrosine-based inhibition motif receptor that binds to sialic acid residues, highly expressed on microglial cells, activation of CD33 opposes activation of ITAM receptors such as TREM2.
ABI3	AD	Moderate, expressed selectively on microglial cells, may function as part of the WAVE complex, implicated in regulation of cell migration, rare variants associated with increased risk for AD (see Table 2)
ADAM10	AD	Moderate, cell surface proteins with a unique structure possessing both potential adhesion and protease domains, functionally studied for its effects as an α-secretase that cleaves APP and TNF-α, appears to be expressed highly in human microglial/macrophages, this and other α-secretases are known to regulate immune responses through their sheddase activity
APOE	AD	Moderate, highly expressed in microglial, APOE genotype, clear role in regulation of peripheral immune responses, interaction with Aβ aggregates can alter clearance of Aβ by microglial cells
CASS4	AD	Moderate, Docking protein that plays a role in tyrosine kinase-based signaling related to cell adhesion and cell spreading, selectively expressed in the brain in microglial cells.
ABCA7	AD	Moderate, widely expressed on many CNS cell types, can regulate microglial phagocytosis, deletions in ABCA7 have been associated with AD risk in African Americans
SORL1	AD	Weak, a multifunctional endocytic receptor, functionally studied in neurons for its effects on Aβ production, but appears to be expressed highly in human microglial/macrophages.
ALPK2	AD	Weak, a kinase, low expression in CNS cells but expressed in human microglial/macrophages
PICALM	AD	Weak, a clathrin assembly protein expressed in many CNS cell-types but expressed strongly in microglial cells.
CD2AP	AD	Weak, widely expressed in CNS cell types, has been shown to play a role in T-cell function
Il1R2	PD	Strong, binds interleukin alpha (IL1A), interleukin beta (IL1B), and interleukin 1 receptor, type I(IL1R1/IL1RA), and acts as a decoy receptor that inhibits the activity of its ligands,
		highly expressed on microglial cells
TLR9	PD	Strong, a member of the Toll-like receptor (TLR) family, which plays a fundamental role in pathogen recognition and activation of innate immunity, highly expressed on mouse microglial cells
STAB1	PD	Strong, a large, transmembrane receptor protein which may function in angiogenesis, lymphocyte homing, cell adhesion, or receptor scavenging, highly expressed on microglial cells
HLA-DRB6	PD	Strong, one chain of the heterodimeric MHC class II molecule, highly expressed in microglial cells, region of association includes other MHC genes and non-MHC genes, involved in antigen presentation, but may play a role in innate immune responses. This gene has been called a pseudogene, though there is evidence that it can generate a protein.
HLA-DQA1	PD	Strong, one chain of the heterodimeric MHC class II molecule, highly expressed in microglial cells, region of association includes other MHC genes and non MHC genes, involved in antigen presentation, but may play a role in innate immune responses
CTSB	PD	Strong, a lysosomal protease, highly expressed in microglial cells, plays a role in antigen processing and protein turnover.
GCH1	PD	Moderate, a member of the GTP cyclohydrolase family, highly expressed on microglial cells,
CD38	PD	Moderate, act as a receptor for cells in the immune system, primarily expressed by astrocyte in the brain.
CRHR1	PD	Moderate, Corticotropin Releasing Hormone Receptor 1 G-protein coupled receptor that binds neuropeptides of the corticotropin releasing hormone family that are major regulators of the hypothalamic-pituitary-adrenal pathway. This receptor regulates corticosteroid levels via the HPA axis, but is also present on immune cells and CRH has been shown to directly modulate immune cell function.
GBA	PD	Weak, a lysosomal membrane protein that cleaves the beta-glucosidic linkage of glycosylceramide expressed on many CNS cell types but highly expressed in mouse microglial cells
SIPA1L2	PD	Weak, a member of the signal-induced proliferation-associated 1 like family, highly expressed on microglial cells.
ARHGAP27	PD	Weak, member of a large family of proteins that activate Rho-type guanosine triphosphate (GTP) metabolizing enzymes, highly expressed on mouse microglia cells.

^AAdapted from (Chang et al., 2017; Hamza et al., 2010; Hill-Burns et al., 2011; International Genomics of Alzheimer’s Disease, 2015; Lambert et al., 2010; Lambert et al., 2009; Nalls et al., 2014; Reitz et al., 2013; Wissemann et al., 2013; Witoelar et al., 2017).

^BThese refer to the candidate immune genes within the region of association, in many cases there are other non-immune genes within the region of association.

^CEvidence for immune function based on data in genecards (https://www.genecards.org/) and pubmed https://www.ncbi.nlm.nih.gov/pubmed/ is highlighted, expression data is based on (Zhang et al., 2014; Zhang et al., 2016). The classification of evidence for function in the immune system in these disease (Strong, Moderate, Weak), is somewhat subjective but based on current evidence for functional role in the immune system and RNA expression data.

Table 2. Mutations and variants in immune genes that cause or alter risk^A for neurodegenertive disease.

Gene	Disease	Mutation/Variant
CSF1R	hereditary diffuse leukoencephalopathy with neuroaxonal spheroids (HDLS)	Numerous, more than 50 pathogenic variants, including missense, frameshift and non-sense mutations, but also deletions and splice-site mutations, all located in the intracellular tyrosine kinase domain, strongly support haploinsufficiency and partial loss of function in disease mechanism. Highly expressed on microglial cells and myeloid cells.
GRN	frontotemporal dementia (FTD)	Numerous, heterozygous result in haploinsufficiency and thus partial loss of function through nonsense mediated RNA decay. Highly expressed by microglial cells.
GRN	neuronal ceroid lipofuscinoses (NCL)	Numerous, homozygous or compound heterozygous mutations cause near complete loss of function
TREM2	Nasu-Hakola disease (NHD), also referred to as polycystic lipomembranous osteodysplasia with sclerosing leukoencephalopathy (PLOSL), also FTD	Numerous, homozygous or compound heterozygous mutations cause near complete loss of function. At least 11 mutations cause PLOSL/FTD. Not all mutation have been shown to have the osteodysplasia with the phenotype in those individuals largely FTD. Functionally mutations that cause PLOSL/FTD appear to be loss of function. Highly expressed on microglial cells.
TYROBP	NHD/PLOSL	Multiple, homozygous or compound heterozygous mutations cause near complete loss of function. Mutations include deletions, insertions and frameshift, resulting in either no or truncated non-functional TYROB. Strongly expressed on Microglial Cells
TREM2	Increased Risk for AD	R47H and R62H TREM2 variants are associated with partial loss of function. Other variants may also be associated with altering AD risk, but studies showing these associations have not been as widely replicated
ABI3	Increased Risk for AD	S209F variants is associted with increased risk for AD. Expressed selectively on microglial cells, may function as part of the WAVE complex, implicated in regulation of cell migration. Little data on functional effects of variant.
PLCG2	Decreased risk for AD	P522R variant is associted with increased risk for AD. Initial data suggest that this variant slightly increases PLCG2 activity. Selectively expressed ion microglial cells in the brain and in many cells in the periphery.

^ANumerous other variants in immune genes have been associated with AD, this table focuses only on the ones for which there is wide consensus in the field that these alter disease risk.

Potentially more immediately informative with respect to causality and directionality are studies that i) link genes with function in the immune system to rare familial forms of neurodegenerative disease and ii) rare functional variants in genes whose mRNAs are selectively expressed in microglia or other immune cells that are associated with risk for AD. Examples of mutations in immune genes that cause neurodegeneration are i) colony stimulating factor-1 receptor (CSF1R) mutations that cause dominantly inherited hereditary diffuse leukoencephalopathy with neuroaxonal spheroids (HDLS) (Mitsui et al., 2012; Rademakers et al., 2011; Stabile et al., 2016) ii) progranulin (GRN) mutations that cause of frontotemporal dementia (FTD) (Baker et al., 2006; Cruts et al., 2006) or neuronal ceroid lipofuscinoses (NCL) (Smith et al., 2012) and iii) triggering receptor expressed on myeloid cells 2 (TREM2) or TYRO Protein Tyrosine Kinase Binding Protein (TYROBP) mutations that cause Nasu-Hakola disease (NHD), also referred to as polycystic lipomembranous osteodysplasia with sclerosing leukoencephalopathy (PLOSL) (Paloneva et al., 1993; Paloneva et al., 2002; Soragna et al., 2003). Examples of coding variants in immune genes that alter risk for a neurodegenerative disease are the AD-associated variants in TREM2, phospholipase C gamma 2 (PLCG2), and ABI family member 3 (ABI3) genes (Guerreiro et al., 2013; Jonsson et al., 2013; Sims et al., 2017). Like TREM2, PLCG2 and ABI3 are thought to be selectively expressed on microglia in the CNS and monocytes/macrophages and other immune cells in the periphery. Notably, whereas TREM2 and variants are associated with increased AD-risk, the PLCG2 variant is protective (Sims et al., 2017).

Though our understanding of how these various mutations alter protein function is evolving, the weight of evidence suggests that those associated with risk for neurodegeneration result in loss of, or reduced protein function or signaling and, consequently, promote reduced microglial activity, dysfunction, and neurodegeneration. Functional effects of mutations in GRN and TREM2 have been most extensively studied. Heterozygous, mutations in GRN which encodes the secreted protein progranulin (PRGN), clearly result in haploinsufficiency, due to non-sense mediated RNA decay and this leads to FTD (Baker et al., 2006; Cruts et al., 2006). In contrast, homozygous and compound heterozygous mutations that result in even further reductions in PRGN levels, cause NCL (Smith et al., 2012). Thus, it is clear that GRN mutations are loss-of-function and that the relative loss of function, from partial to near complete, results in distinct forms of neurodegeneration. Unfortunately, it remains unclear whether loss of PRGN leads to degeneration due to effects on microglial function, paracrine effect on neurons or other CNS cells, or some combination of the two (Arrant et al., 2018; Filiano et al., 2013; Kao et al., 2017; Lui et al., 2016; Martens et al., 2012).

Because i) it is associated with a rare form of neurodegeneration (PLOSL) and a more common form (AD) and ii) the AD risk variants have a large impact on disease risk, functional impacts of the disease-associated variants in TREM2 have been intensively studied (Ulrich et al., 2017). TREM2 is member of a family of type 1 membrane protein, triggering receptors expressed on myeloid cells (TREMs). TREM2 is distinct in that it is highly expressed in brain microglia. It contains an intracellular immunoreceptor tyrosine-based activation motif that conducts signals following binding to a variety of ligands such as various lipids and lipoproteins through interaction with the intracellular adaptor TYROBP (Ulrich et al., 2017). These studies show that the mutations that cause PLOSL have more distinct effects on protein function than do the variants that cause AD (Filipello et al., 2018; Kleinberger et al., 2014; Korvatska et al., 2015; Lee et al., 2018; Lessard et al., 2018; Neumann and Takahashi, 2007; Schlepckow et al., 2017; Song et al., 2018; Xiang et al., 2018; Zhao et al., 2018). Indeed, there is emerging consensus that PLOSL-associated TREM2 mutations result in defective protein maturation and in almost complete loss of function, whereas distinct AD-associated TREM2 variants result in subtle, partial loss-of-function. Consistent with these findings are data showing that a variant in the TREM locus is associated with decreased risk for AD and increased expression of TREM2. Additionally, CSF1R mutations, that cause disease in a heterozygous state, appear to be loss-of function, though the mechanisms that result in decreased function are still debated (Mitsui et al., 2012; Rademakers et al., 2011; Stabile et al., 2016). The AD associated PLCG2 variant is somewhat unique as it is associated with protection from disease. Further, it is one of the few classically druggable proteins identified in AD to date. Emerging data suggests that the variant results in slightly increased phospholipase activity compared to the wild-type enzyme that would be predicted to enhance immune signaling, though this data will need to be confirmed and extended (Magno et al., 2018).

Collectively, these genetic studies begin to paint a picture whereby decreased function of activating immune receptors or signaling molecules results in disease. In contrast, and with far fewer examples, promoting immune activation appears to be protective. Whether these functional associations will hold up as more data emerges is uncertain; however, there clearly is enough data to suggest that the long-standing association of neuroinflammation and immune activation as a driving, solely pathogenic force in neurodegeneration is misplaced. There are simply too many examples where loss of immune protein function and signaling that is functionally linked to less immune mobilization appears to trigger neurodegeneration.

Account for possible differences in immune function between human and mouse models.

Mouse models of neurodegenerative proteinopathies have limitations, but they have provided pivotal support for the hypothesis that many neurodegenerative disorders are fundamentally proteinopathies (Dawson et al., 2018). Although many of the alterations in immune cells and proteins documented in human neurodegenerative disease are noted in the counterpart rodent models, there are important differences in immune repertoire, responses and priming between rodent and human (Heneka et al., 2015; Labzin et al., 2018). Such differences need to be accounted for in order for observations made in the rodent model to be used to inform how a given manipulation of the immune system might alter the human disease or even how a given genetic alteration causes disease.

There are several examples of neurodegenerative diseases where genetic knockout of the mouse counterpart of the human gene does not produce a close phenocopy of the human disease. One example is the heterozygous Grn (Grn+/−) mice, which develop very mild phenotypes that do not mimic human FTD (Lui et al., 2016; Minami et al., 2014). In contrast, the full knockout (Grn−/−), produces a model of NCL that closely phenocopies the human disease (Martens et al., 2012). Another example is mice lacking Trem2, which exhibit select immune phenotypes and osteoclastogenesis, but do not closely phenocopy the neurodegenerative aspects of human PLOSL (Otero et al., 2012). Similarly, mice lacking Tyrobp that do show brain and bone abnormalities do not model the overt phenotype of human PLOSL (Haure-Mirande et al., 2017; Nataf et al., 2005). Differences in immune repertoire between mice and humans can impact the understanding how genetic variants alter risk for select human neurodegenerative diseases. For example, there are considerable differences between humans and mice in the major histocompatibility (MHC/HLA) locus, which has been genetically associated with AD and PD. Further, both the TREM and the MS4A locus, which are both implicated by GWAS in AD, are quite divergent between mice and humans (Colonna, 2003; Liang et al., 2001).

Efforts to standardize housing conditions and limit pathogen exposure means that our mouse models are typically reared for generations in relatively sterile environments that restrict immunologic priming, including the priming of microglial cells in the brain (Erny et al., 2015; Perry and Holmes, 2014). Variation in normal microbiome and pathogen exposure could, in theory, alter immunoproteostasis in unpredictable ways, not only through direct effects on the brain, but also through effects on peripheral immune cells (Prinz and Priller, 2017). Though recent studies do show that altering the microbiome can influence immunoproteostasis in model systems, for the most part, the effects observed to date are rather modest (Harach et al., 2017; Minter et al., 2017; Sampson et al., 2016).

Concerted efforts and careful study, including development of mice with humanized immune genes or gene loci and exposure of mouse models to more diverse microbiomes, will likely provide much needed insight into how some immune genes may modulate risk for neurodegenerative disease and how environmental factors may contribute to the neurodegenerative phenotype; however, these types of studies are challenging and may still not overcome all the inherent limitations of modeling complex human neurodegenerative disorders that typically occur in late life in mouse models with much shorter lifespans and only partially conserved physiology.

A need for more human data.

Given the aforementioned limitations of mouse model studies, when possible, we should try to study how variants in immune genes alter the immune system in living humans. Most of the immune genes implicated in neurodegenerative disorders are expressed in peripheral immune cells. Further, in some instances, there are peripheral manifestations associated with the neurodegenerative disease, for example, bone cysts in PLOSL (Paloneva et al., 1993). In parallel to modeling studies, there should be a concerted effort to catalogue peripheral immune function in humans that carry these variants and also to look for other signs of altered immune function (e.g, look for bone cysts in those who carry TREM2 AD risk variants). This phenotyping might rapidly provide data on how variants in immune genes implicated in neurodegeneration alter immune function in living humans and serve as translational biomarkers. There is, of course, continued value in looking at human postmortem brain using multi-omic approaches to provide a systems level “compendium” of changes in immune gene expression and protein levels; however, such studies have limited ability to stage changes in immune system within the disease continuum and are confounded by many factors including the inability to distinguish effects that may be the result of the agonal state or long-standing disease state.

One example of how data from such studies could be transformational would be study of how the AD-associated PLCG2 variants alter peripheral immune function (Sims et al., 2017). PLCG2 is a potentially tractable drug-target, but more information on how the AD-associated variant alters both enzymatic and immune function is needed. Notably, other coding region variants in PLCG2 have been linked to peripheral immune disorders. A PLCG2 mutation (p.Ser707Tyr) causes an autosomal dominant inflammatory disorder (Zhou et al., 2012). This mutation is hypermorphic with respect to enzyme activity, strongly increasing phospholipase activity relative to wild type enzyme. In contrast, exon skipping due to genetic deletions in PLCG2 result in a distinct peripheral immune disorder characterized by cold-induced urticaria, antibody deficiency, and susceptibility to infection and autoimmunity (Ombrello et al., 2012). These PLCG2 deletions, located within an autoinhibitory domain, result in constitutive phospholipase activity, but diminished cellular signaling at 37°C and enhanced signaling at lower temperatures. Given the initial data that the AD-associated variant is also a slightly hypermorphic with respect to enzyme activity (Magno et al., 2018), one might postulate that carriers of this variant also have a mild autoimmune disorder and altered responsiveness of peripheral immune cells.

There are challenges to such human studies. In the case of PLCG2, the AD-associated variant is rare (allele frequency <1%). To identify individuals with and without AD, who harbor this variant would require a significant investment. There would also need to be careful consideration for what assays, clinical assessments and other studies might be most informative. In contrast to rare variants, more common variants implicated by GWAS might be studied as well; however, as most of those variants have rather weak effects on disease risk, functional effects of such genetic variation may be muted and difficult to detect.

Much of our data on immune responses in human neurodegenerative disease comes from cross-sectional autopsy studies, which have inherent limitations. In contrast to mouse models, we have limited insight into the evolution of the immune response during disease progression. Studies using a positron emission tomography (PET) ligands that binds translocator protein 18 kDa (TSPO) which is strongly upregulated in “activated” microglial have some utility (Cerami et al., 2017; Edison et al., 2018); however, due to increases in TSPO binding as a simple consequence of age, ability to discriminate between diseased brain and normal brain in the aged is limited. Cerebrospinal fluid (CSF)-based biomarker studies have also suggested that there may be various distinct phases of the immune responses in AD, with different markers peaking at different stages of disease (Liu et al., 2018; Piccio et al., 2016; Rauchmann et al., 2018; Suarez-Calvet et al., 2016a; Suarez-Calvet et al., 2018; Suarez-Calvet et al., 2016b); however, there remains a critical need to develop more sensitive and specific PET- and CSF-based markers that can provide a more global assessment of immune activation state in the living brain and immune function as disease progresses.

Revisit epidemiological studies to look at how incidence of neurodegenerative disease is influenced by inflammatory conditions and immune modulatory medicines.

Given the rather uncertain impact and concerns over interpretation of epidemiologic studies relating to NSAID use and risk of AD, there may be some hesitancy to simply revisit this type of study; however, there are good reasons to reexamine a broader association between inflammatory disorders, immune modulatory therapy and neurodegenerative diseases. There are now many widely used immunomodulatory drugs that have much more profound functional effects on the immune system than NSAIDs. Immunosuppressive drugs, such as etaneracept, adilimumab, natalizumab and others, have potentially been used for long-enough periods that some insight into whether they influence risk for various neurodegenerative disorders may be gleaned from large-scale epidemiologic studies (Miller et al., 2003; Moreland et al., 1999; Weinblatt et al., 2003). As these are all immunosuppressive agents and are used to treat various pro-inflammatory immune disorders, one could argue that the same reverse causation confound regarding whether it is the disease itself or the treatment that might be associated with any alteration in risk; however, these biologic agents are not universally applied therapeutics. Thus, theoretically, any confound could be lessened by stratifying risk of neurodegeneration based on both underlying disease and type of medication used. A second reason is the emergence of many large databases, where millions or even tens of millions of individual’s medical data is housed and could be mined to evaluate this issue, potentially increasing statistical power to the point where there was much more confidence in the data that is generated. A final reason is simply that some of the therapies such as natalizumab (Miller et al., 2003), which binds α4 integrin and blocks T-cell infiltration from the periphery through the blood brain barrier, might have an unexpected impact, positive or negative, on neurodegenerative disease. Although there is some preclinical data implicating T-cells in modulation of some neurodegenerative disorders (Marsh et al., 2016), there is only sparse genetic and pathological data linking T-cells to most forms of neurodegeneration. Highly powered, appropriately controlled epidemiologic data could result in highly translatable insights that may never be resolved by studies in model systems.

Use genetics to broadly inform immune therapies.

There has been increasing emphasis in the pharmaceutical industry to only pursue genetically validated targets for disease modifying therapies. This is especially true for CNS disorders where therapeutic successes have been extremely limited. With respect to AD, the convergence of genetic data implicating the immune system has dramatically increased interest in the development of immune modulating therapies, specifically those that might target TREM2. Though TREM2 is a potentially tractable therapeutic target, development of therapeutics that increase CNS TREM2 function and signaling, which has emerged as the desired therapeutic mechanism of action, will not be trivial. Indeed, antagonism of a type 1 membrane receptor with a plethora of endogenous ligands that is also rapidly shed from the membrane poses both pharmacologic and pharmacokinetic challenges (Ulland and Colonna, 2018).

An alternative to a strict focus on proteins or other factors directly implicated by genetic studies is to broadly understand the signaling pathways implicated through genetics. Guided by this experimentally deduced functional understanding, it may be possible to identify more tractable targets that are likely to provide a similar therapeutic effect. A simplistic example relating to TREM2, is that TREM2 and other immunoreceptor tyrosine-based activation motif (ITAM) based signaling receptor function is often opposed by immunoreceptor tyrosine-based inhibition motif (ITIM) based signaling receptors. Inhibition of an ITIM based receptor, which opposes TREM2 signaling theoretically could result in a similar functional effect as activation of TREM2 (Barrow and Trowsdale, 2006; Linnartz and Neumann, 2013). As receptor antagonism or inhibition of downstream signaling is often easier to achieve pharmacologically than agonism; rigorous mechanistic studies of immune pathways in the brain that build off of genetic associations are almost certain to reveal additional, potentially tractable, therapeutic targets.

We also need to think about the therapeutic implications of preclinical studies in the context of human genetics. For example, as noted above, it appears that partial loss of function of CSF1R can cause a neurodegenerative disorder. In contrast, several preclinical studies indicate that small molecule CSF1R inhibition may have beneficial effects in APP and ALS mouse models (Elmore et al., 2018; Martinez-Muriana et al., 2016; Olmos-Alonso et al., 2016; Sosna et al., 2018; Spangenberg et al., 2016). The juxtaposing of these findings would certainly suggest that therapeutic studies of CSF1R inhibition in AD, or any neurodegenerative condition, should only proceed with appropriate caution and rigorous evaluation of potential liabilities, but also reinforce the need to truly understand how the variants within CSF1R cause disease.

Develop a more rigorous, translational preclinical road map for development of immune therapies targeting neurodegenerative disease.

Given the imperfect nature of many of our mouse models of neurodegenerative disease, their predictive value with respect to development of novel disease modifying therapies has been drawn into question. Despite these translational shortcomings, which in many cases are not necessarily an inherent defect with the model itself, these models have enabled us to understand critical aspects of disease pathogenesis and progression (Dawson et al., 2018). At least for the foreseeable future, these models are the best we can do with respect to providing a realistic system to explore immune modulating therapies designed to alter immunoproteostasis. In the future, it may be possible to develop ex vivo models, perhaps using differentiated human induced pluripotent stem cells (iPSCs) (Abud et al., 2017; Douvaras et al., 2017; Garcia-Reitboeck et al., 2018; Haenseler et al., 2017; Park et al., 2018; Takata et al., 2017). However, it is difficult to envision how one can recreate all of the complex interactions between various cells in the brain, the blood brain barrier, the circulatory system, the peripheral immune system, and environmental factors from iPSCs in a dish. Nevertheless, there is exciting work in this area and valuable insights may emerge from such studies in the future (Berry et al., 2018).

Because a given immune manipulation may have effects that are contextually dependent, the net effect may be dependent on both the timing with respect to disease stage and the intensity and length of the manipulation. This means that intensive preclinical studies are needed. Indeed, an interventional preclinical study with a single therapeutic dose that only looks at effects on a limited set of phenotypes and in one treatment paradigm is probably not sufficient to reliably predict efficacy in a different disease stage, or reveal possible side effects. Further, emerging data that an immune modulation may have different effects on different human neurodegenerative disease models indicates that preclinical studies of immune modulation should include multiple diseases. During these studies, there should be concerted efforts to develop translational theragnostic biomarkers that inform on both target engagement and potential efficacy. Some monitoring of the peripheral immune status is also warranted, as well as a survey of possible immune related effects. With increased rigor and more systematic preclinical studies, it may be possible to identify immune targeting therapies that can safely provide disease modification in select neurodegenerative disorders.

Notably, mouse-modeling studies conducted to date often reveal conflicting results of effects of modulating innate immune activation states in neurodegenerative disease models (reviewed in (Czirr and Wyss-Coray, 2012). Numerous studies show beneficial effects of an immune suppressing manipulation and harmful effects of an immune activating manipulation, but multiple studies show the opposite effects, or show that that the outcome is dependent on the disease model; for select recent examples of such conflicting data see (Ayers et al., 2014; Chakrabarty et al., 2015; Guillot-Sestier et al., 2015; Heneka et al., 2013; Venegas et al., 2017; Wendeln et al., 2018)). Future studies, which more systematically evaluate how a given immune manipulation alters the phenotype, as described above, will hopefully help build consensus regarding how these preclinical studies inform immune modulatory therapeutic approaches to neurodegenerative disease.

Change our lexicon regarding the role of the immune system in neurodegenerative disease.

Because of the i) potentially misplaced assertion that neuroinflammation is a pro-inflammatory state that is invariably harmful, and ii) the growing recognition that immune activation states are incredibly complex and only rarely, dichotomously, pro- or anti-inflammatory in nature, perhaps, in the future, there should be an effort to limit the use of term neuroinflammation. Similarly, as suggested by others previously, blanket descriptors of immune activation states in the brain as either pro- or anti-inflammatory, or M1 or M2, are probably not particularly useful (Graeber et al., 2011; Masgrau et al., 2017; Ransohoff, 2016). Perhaps as an alternative, we should simply be more precise in describing the immune response and resultant phenotype with respect to a particular disease stage or immune targeting intervention. With ever-increasing technological advances that make systems biology levels studies more feasible, this should become a best practice in the field.

Summary

In planetary science, the concept of a “Goldilocks zone” is used to describe planetary orbits that are potentially compatible with life - neither too close to the sun, “too hot”, or too far away, “too cold”. A similar concept likely applies to immunoproteostasis (see Figure 1). Although functional genetics clues are emerging that suggest that we may want to activate the immune system for therapeutic benefit in some neurodegenerative disorders, it is not at all clear if such an inference is generalizable. As manipulation of the immune system has potential for harm as well as benefit, caution and due diligence is warranted before testing novel immune therapies designed to treat a neurodegenerative disorder in humans. No matter whether “activation” or “inactivation” appears to be desirable effect, it is almost certain that pushing the system too far in one direction has potential to do no more harm than good.