Studies of microglial gene manipulation in mouse models of Alzheimer’s disease (AD) amyloidopathy can cause unpredictable effects on various key endpoints, including amyloidosis, inflammation, neuritic dystrophy, neurodegeneration, and learning behavior. In this Correspondence, we discuss three examples, microRNA 155 (miR155), TREM2, and INPP5D, in which observed results have been difficult to reconcile with predicted results based on precedent, because these six key endpoints do not reliably track together (Table 1). miR155 is a microRNA in neurons and glia that modulates immune function and the response to viral infection [1,2,3,4,5,6]. TREM2 is a cell-surface protein in microglia that senses the presence of protein or lipid debris in the interstitial spaces of the brain [8,9,10,11,12,13,14,15,16,17, 19, 20]. INPP5D is a phospholipid and phosphoprotein phosphatase in microglia that regulates phagocytosis and clustering of microglia around amyloid plaques [21,22,23,24,25].

Table 1 Summary of confounding results from manipulations of miR155, TREM2, or INPP5D in various mouse models of Alzheimer’s pathology
Full size table

MicroRNAs are important regulators of many facets of physiological function while also being implicated in the pathogenesis of several systemic and neurological disorders, including immunity, inflammation, viral infection, cancer, cardiovascular disease, AD, and Down syndrome (for review, see ref. [1]). Elevation of miR155 levels in human AD brain was recently confirmed and localized to microglia and hippocampal neurons [2, 3], observations that were extended to an AD amyloidopathy mouse model [4]. These data made miR155 an attractive target for controlling neuroinflammation. However, three recent papers raise the possibility that therapeutic modulation of miR155 might not be straightforward [3, 5, 6]. In a 2020 study [3], APP/PS1 mice on a constitutive pancellular miR155 knockout background showed improvement in synaptic function and learning behavior despite an increase in amyloid burden. In one of the 2023 studies, selective and conditional deletion of miR155 from microglia in pre-symptomatic APP/PS1 mice caused new seizures to appear despite a decrease in amyloid burden [5]. In the other 2023 study [6], also using an inducible microglia KO strategy and APP/PS1 mice, the result was increased plaque compaction, reduced neuritic dystrophy, and improved behavior. Both 2023 papers used inducible, microglia-specific knockouts, and both gave tamoxifen at 6–8 weeks of age yet the phenotypes were different, perhaps because different PS1 mutations were used: in the former, the exon 9 deletion PS1 mutation was used [5], while the other used the L166P point mutation [6]. While we are unable to explain the conflicting data and complex effects, we note that such dramatic and irreconcilable differences are not unusual for studies in mouse models of AD amyloidopathy [7]. In these studies, key endpoints of amyloid burden, inflammation, neuritic dystrophy, neurodegeneration, and learning behavior are not consistently linked. We propose that the targeted cell type and timing of intervention can cause dramatic shifts in amyloid burden, microglial function, inflammation, and learning behavior and their relationships to one another.

Another dramatic example of discrepant results occurs with experimental manipulation of TREM2 (triggering receptor on myeloid cells-2). R47H is a well-known TREM2 mutation that increases the risk for AD [8, 9]. The molecular pathogenesis of this mutation is believed to relate to the inability of the mutant TREM2 to play its usual roles in modulating neuroinflammation and phagocytosis of Aβ. R47H shifts both endpoints in a damaging direction, leading to increased neuroinflammation and amyloid burden. Based on these data, it was initially hypothesized that TREM2 loss-of-function would increase pathology in mouse models, and that increased activity would mitigate pathology. The effects of the R47H mutation were confirmed following conditional overexpression of TREM2 R47H in mouse models, in which the mutant TREM2 transgene causes amyloid burden to increase [10]. These mutation-related effects contrast with those from TREM2H157Y, a genetic variant that also increases risk for AD; yet, H157Y reduces amyloidosis in a mouse model, suggesting that this mutation increases risk for AD by acting through an amyloidosis-independent pathway [11].

Since loss-of-function was predicted to exacerbate phenotype, increased TREM2 was predicted to ameliorate phenotype. In fact, constitutive or conditional overexpression of wild-type TREM2 is beneficial prior to amyloid deposition in models of early AD; yet, when TREM2 is overexpressed after amyloidosis is established, both amyloid burden and neuritic dystrophy are exacerbated [10]. Alternatively, when the 5xFAD mutations are on a constitutive Trem2 knockout background, accumulation of Aβ peptides is enhanced both within neurons and in the interstitial extracellular compartment [12]. After discovery of pathogenic mutations in the TREM2 gene, metabolism of the transmembrane holoprotein by α-secretase was discovered to release the TREM2 ectodomain as a protective factor known as soluble TREM2 (sTREM2), and its effects in mouse models have also been assayed. Experimental viral transduction with TREM2 and sTREM2 complicated matters further. In one study, adeno-associated virus (AAV) expressing sTREM2 was injected into symptomatic, 7-month-old 5xFAD mice, resulting in proliferation of microglia, reduction in amyloid burden, and preservation of normal learning behavior [13]. Viral overexpression of TREM2 in the hippocampi of wild-type mice resulted in significant synaptic impairment [14]. The same report showed that TREM2 overexpression enhanced microglial phagocytosis of synapses during the early-to-middle stage of amyloidosis. However, TREM2 overexpression and upregulated microglial phagocytosis came to play positive roles by the middle-to-late stage of pathology when TREM2 overexpression reduced amyloid deposition [14].

Genetic deletion of Trem2 exerted different outcomes depending on the transgenic line, timing of the intervention and the age of the mouse [15]. Constitutive deletion in APP/PS1 mice led to decreased amyloid accumulation [15]. Knockdown of Trem2 at the early-to-middle stage of amyloid accumulation (2–6-month-old APP/PS1 mice) prevented synaptic phagocytosis by microglia, whereas knockdown of Trem2 at the middle-to-late stage (6–10-month-old APP/PS1 mice) caused impairment of microglial phagocytosis, exacerbated amyloid burden, and destroyed synapses via amyloid toxicity [15]. On the other hand, deletion in PS2APP mice led to increased amyloid accumulation in 7-month-old female mice, but reduced accumulation in both sexes by 21–22 months, whereas neuritic dystrophy was increased [16].

More examples of divergent results have arisen from modeling of therapeutic intervention using anti-TREM2 antibodies. Antibody 4D9 blocks shedding of sTREM2, activates protective TREM2 signaling, and improves pathology in a mouse model of AD amyloidopathy [17]. Agonistic antibodies such as 4D9 and others are currently being tested in clinical trials [17, 18]. It is worth noting that while TREM2-activating antibodies show promise in reducing amyloid burden at early stages, if treatment is continued into later stages of pathology, TREM2 agonism may exacerbate seeding and spreading of tauopathy [19]. This means that for the TREM2-activating antibody to be useful as a long-term therapy, it is likely that complete purging of amyloid (e.g., with one of the anti-amyloid antibodies) will be required prior to initiation of TREM2-activating antibody treatment. TREM2 and tauopathy represent another interaction with conflicting results. Both TREM2 activation and TREM2 inhibition have been reported to attenuate tauopathy [19, 20]. Finally, the level of TREM2 downregulation also led to unexpected results, in that genetic haploinsufficiency vs total deletion led to opposite effects on tauopathy rather than causing similar effects in a dose-dependent fashion [20].

Studies on the role of INPP5D in amyloidosis studies have likewise yielded a mixture of results. INPP5D is another microglial AD risk gene identified by GWAS, and four recent Inpp5d knockdown studies in cerebral amyloidosis models were each slightly different in design, and each showed different effects [21,22,23,24]. Despite disparate impact on amyloid burden, three related results were similar across several Inpp5d knockdown studies: (1) increased plaque-associated microglia; (2) increased microglial barrier; and (3) attenuated neuritic dystrophy. Castranio et al. selectively and inducibly knocked down Inpp5d in the microglia of early symptomatic APP/PS1 mice crossed with floxed Inpp5d mice, using the CX3CR1-Cre mouse; this genotype and treatment exacerbated the amyloid burden and increased expression of disease-associated microglial genes[21]. A similar study by Samuels et al. used the same approach to target microglial Inpp5d in 1 month-old 5xFAD mice, but, in those mice, amyloid burden was unchanged by the Inpp5d deficiency[22]. Iguchi et al. (2023) studied the effects of Tyrobp deficiency and Inpp5d deficiency (singly and in combination) on the AppNL−G−F/NL−G−F knock-in mouse[23]. As with the Samuels et al. [22] study, Inpp5d deficiency alone had no effect on amyloid burden in the Iguchi et al. study [23]. In the fourth study, Lin et al. [24] used a constitutive pancellular Inpp5d knockdown approach and reported increased plaque clearance and reduced amyloid burden in the presence of Inpp5d haploinsufficiency. Taken together, these papers suggest that modulation of amyloid burden may not be the most important effect of alterations in INPP5D, which apparently differs by stage of disease. This implies that clinical trials of INPP5D inhibitors might show cognitive benefits even if amyloid burden is unchanged or increased. Further complicating the miR155 and INPP5D data interpretation is that INPP5D is a direct target of miR155, and is increased twofold in the brain in the absence of miR155 [25].

Our point is not that some results are right, and some are wrong, but rather that the pathogenesis and staging of AD are so complex that it can be challenging to determine whether increasing or decreasing the level or activity of a target is more likely to cause benefit and that the beneficial direction may even change with the targeted cell type or disease stage. The two miR155 knockout studies and the multiple TREM2 and INPP5D studies provide examples for how disease stage and/or cell type can differentially alter the outcome as well as examples where cognitive benefit does not track with amyloid burden. Defining this complexity in mouse models enables us to take these issues into account when we design human clinical trials.