ABSTRACT
Microglia are a fundamental component of pathogenesis in many neurological conditions and have specialized functions that vary by disease stage or specific pathology. Drugs targeting colony-stimulating factor-1 receptor (CSF1R) to block microglial proliferation in preclinical disease models have shown mixed outcomes, thus the therapeutic potential of this approach remains unclear. Here, we evaluated CSF1R inhibitors in tauopathy mice using multiple dosing schemes, drug analogs, and longitudinal measurements in the brain and plasma. In both spontaneous disease and in tau fibril inoculation models, we found a region-dependent reduction in insoluble phosphorylated tau and replication-competent tau in mice treated with CSF1R inhibitors. Surprisingly, despite greater drug exposure and microglial depletion in male mice, we observed a rescue of aberrant behavior, reduced plasma neurofilament light chain, and extended survival in female mice only. Gene expression patterns in CSF1R inhibitor-treated tauopathy mice reverted toward a normal wildtype signature, and in vivo imaging revealed suppressed astrogliosis. However, we observed drug dose-dependent upregulation of immediate early genes in male mice only, indicating excitotoxicity, which may have masked functional benefits. Drug-resilient microglia in tauopathy mice exhibited a ramified morphology similar to wildtype microglia but with greater territory occupied per cell, and their transcriptome was neither disease-associated nor homeostatic, suggesting a unique microglial subtype. Our data argue that complete or continuous microglial ablation is neither required nor desired for neuroprotection, and that selective depletion of detrimental, tauopathy-activated microglia may be achieved by precise timing and dosing of CSF1R inhibitors. Importantly, therapeutics targeting microglia must consider sex-dependent effects on functional outcomes when weighing their translational potential for neurological disease.
INTRODUCTION
Microglia, the resident innate immune cells of the central nervous system (CNS), are important for neurodevelopment and homeostasis, and are a fundamental component to pathogenesis in many neurological conditions. We now appreciate that microglia are heterogeneous cells, are influenced by the periphery, have sex-dependent biology, and can be helpful or harmful depending on the disease stage or specific pathology1–4. Gene mutations affecting the expression and sequence of microglial genes (e.g. TREM2, CD33, and MS4A) increase risk for Alzheimer’s disease (AD), and implicate microglia in several disease pathways including toxic protein aggregation (Aβ and tau) and neuroinflammation5,6. Thus, for the first time, there is unequivocal evidence in humans that certain microglial functions are robustly involved in the pathogenesis of neurodegenerative disease. However, the precise mechanisms governing microglia function in disease are still not well understood.
In tauopathy (a family of neurodegenerative disorders characterized by tau inclusions in neural cells), there is growing evidence that microglia play an early and constant role in tau aggregation and neuronal loss. Disease-activated microglia assemble an ‘inflammasome’ protein complex and secrete pro-inflammatory cytokines that regulate neuronal kinases and phosphotases causing tau hyperphosphorylation, aggregation and consequent neurodegeneration7–9. Genome-wide transcriptomic studies have identified innate immune pathways that implicate early and robust involvement of microglia in human tauopathy10,11 and related mouse models12–14. Deletion of microglia-specific genes or genetic ablation of microglial cells in rodents have been useful approaches to dissect microglia-mediated mechanisms in disease models, but pharmacologic tools to more dynamically manipulate microglial function have been limited. Recently developed small-molecule drugs targeting colony-stimulating factor-1 receptor (CSF1R), a receptor kinase critical for survival and proliferation of CNS microglia, peripheral tissue macrophages and blood myeloid cells15, are approved for clinical use in various oncology indications16, and have now been adopted by the neuroscience community to study microglia biology. In the past few years, there have been numerous studies using CSF1R inhibitors in models of neurological disease, but only a few studies in models of primary tauopathy17–20. While important first steps, these studies only explored single, static time points of treatment, or used only one sex. Given the dynamic nature and complexity of microglial activation, the timing of CSF1R inhibition in tauopathy and its translational relevance is still an open question.
Thus, the goal of our study was to define a therapeutic window that not only reduced pathological markers, but also led to functional improvement. Moreover, we questioned whether complete or continuous microglial ablation using CSF1R inhibitors was necessary given the important and diverse roles these cells play in brain health and disease. Here, we systematically test CSF1R inhibition using multiple drug analogs at several time points in transgenic mice developing spontaneous tauopathy, and in an inoculation model of induced tauopathy. We demonstrate a reduction of tau pathology in multiple dosing schemes without complete microglial ablation; drug exposure levels correlated with the extent of tau-prion21 and microglial reduction. Surprisingly, we observed suppressed plasma biomarkers of neurodegeneration, rescue of aberrant behavior, and extended survival in female mice only. These data reveal a previously unrecognized sex-dependent therapeutic benefit of pharmacological CSF1R inhibition. Transcriptome analyses showed that treated tauopathy mice exhibited a restored gene expression profile similar to wildtype mice; however, we observed a specific module of sex- and drug concentration-dependent gene expression that might explain the lack of functional rescue in male mice. Interestingly, residual microglia had a morphology similar to wildtype microglia and their gene expression pattern indicated a unique, homeostatic-like signature that was not responsive to tauopathy nor CSF1R inhibition. These data highlight yet another context for microglial heterogeneity with implications for novel microglial biology, and argue that tempering microglial activation with drugs, rather than cellular ablation, is a better therapeutic strategy with clinical relevance.
RESULTS
CSF1R inhibition reduces pathogenic tau in the brains of Tg2541 mice
Building on previous findings17–20, we first evaluated the effect of CSF1R inhibition on the levels of pathogenic tau in the brains of transgenic mice expressing human tau, using a cell-based tau-prion bioassay, enzyme-linked immunosorbent assay (ELISA), and immunohistochemical (IHC) analysis. Transgenic B6-Tg(Thy1-MAPT*P301S)2541 mice, referred to here as Tg2541 mice, express the 0N4R isoform of human tau with the familial frontotemporal lobar degeneration (FTLD)-linked P301S mutation22, which increases its aggregation propensity and prion-like characteristics23,24. We previously demonstrated that the levels of pathogenic tau in hindbrain regions of Tg2541 mice were greater than in forebrain regions25. Therefore, the forebrain and hindbrain regions were examined separately in this study (Fig. 1a). To deplete microglia, Tg2541 mice were dosed with one of two potent, orally bioavailable, and brain-penetrant CSF1R inhibitors: PLX3397 (pexidartinib), which binds receptor tyrosine kinases CSF1R, and to lesser extent, KIT and FLT326, and PLX5622, which selectively binds CSF1R27. Three different treatment paradigms were evaluated: acute (2–4 months old), chronic (2–7 months old), and terminal (2 months old until death) (Fig. 1b–d).
We confirmed that CSF1R inhibition effectively reduced microglial markers Iba1 and P2YR12 in both the forebrains and hindbrains of Tg2541 mice compared to vehicle treatment, and that they had similar effects in the brains of C57BL/6J wildtype mice (Supplementary Fig. 1). We found that sex did not have a statistically significant effect for most measures of microglia or pathogenic tau (Supplementary Table 1). Therefore, male and female mice were grouped together for analysis, unless otherwise noted.
We next employed a reproducible and rapid cell-based bioassay25,28,29 to measure the activity of replication-competent tau-prions in brain homogenates from Tg2541 mice. To ensure an appropriate dynamic range in this bioassay, we optimized for dilution factor and assay duration using aged Tg2541 mouse brain samples, which showed greater than 100-fold higher signal than wildtype mouse brain samples (Supplementary Fig. 2). Following acute, chronic, or terminal treatment with PLX3397 or PLX5622, tau-prion activity in the forebrains of Tg2541 mice was significantly decreased compared to vehicle-treated mice (Fig. 1e–g). Hyperphosphorylation and aggregation of tau occurs first in hindbrain regions of Tg2541 mice, especially in the brainstem and spinal cord, leading to motor deficits caused by severe paraparesis22. This is consistent with our previous report of early and aggressive tau-prion activity in hindbrain regions of Tg2541 mice25; as such, we found that acute CSF1R inhibition was insufficient to reduce tau-prion activity in the hindbrain (Fig. 1e). However, chronic or terminal treatment with PLX3397 did significantly reduce tau-prion activity in hindbrain regions (Fig. 1f,g) and the spinal cords of the Tg2541 mice (Supplementary Fig. 3). To examine other markers of pathogenic tau, we measured the levels of tau phosphorylated at Ser396 (pS396) by ELISA, and tau phosphorylated at Ser202/Thr205 (pS202/T205) by IHC. Acute, chronic, or terminal PLX3397 treatment robustly reduced pS396 tau in both forebrain and hindbrain regions of Tg2541 mice (Fig. 1h–j), and also reduced pS202/T205 tau in forebrain regions (Fig. 1k–m) and in the spinal cord (Supplementary Fig. 3d). Overall, we observed a largely consistent theme of pathogenic tau reduction by CSF1R inhibition, despite a few differences in the therapeutic efficiency of PLX3397 and PLX5622 administered in different treatment paradigms.
Having verified the benefits of microglial depletion at an early disease stage, we next wondered whether initiating CSF1R inhibition at a more advanced stage of disease would have similar effects, simulating an interventional drug treatment. Thus, we dosed Tg2541 mice with PLX3397 in a delayed treatment paradigm (4–7 months old). Similar to terminal treatment, interventional treatment also significantly reduced tau-prion activity in both the forebrain and hindbrain (Supplementary Fig. 4a,b). Although pS396 tau levels were unchanged after interventional treatment, levels of tau phosphorylated at Thr231 (pT231) were reduced in the forebrain (Supplementary Fig. 4c,d). Considering the potential off-target effects of continuous, long-term microglial depletion on brain function, we also wondered whether periodic CSF1R inhibition might provide a safer, yet similarly efficacious therapy. Thus, we also tested PLX3397 dosed intermittently by repeating dosing cycles of three weeks on followed by three weeks off. Intermittent treatment produced similar reductions in the levels of microglial markers in both brain regions as for continuous treatment, but tau-prion activity and pT231 levels were reduced only in the forebrain (Supplementary Fig. 4e–i). Taken together, these data suggest that intermittent/interventional dosing is sufficient to reduce pathogenic tau in the forebrain of Tg2541 mice, likely due to slower disease kinetics; however, continuous CSF1R inhibition is necessary for the extended reduction of pathogenic tau in the hindbrain.
Tau has been shown to propagate throughout the brain in a prion-like fashion along interconnected neural networks30–32. To test the hypothesis that microglial depletion may reduce the propagation of tau-prions21 in the brains of Tg2541 mice, we inoculated fibrils of the microtubule-binding repeat domain of tau, referred to as K18 fibrils33, into the hippocampus and overlying cortex (forebrain regions) of Tg2541 mice and then treated them with PLX3397.
Compared to un-inoculated mice, K18-inoculated mice had significantly increased tau-prion levels in the ipsilateral (inoculated) forebrain, as well as in the contralateral forebrain and in the hindbrain (Supplementary Fig. 5), which suggests that tau-prions had propagated from the inoculation site to those brain regions. However, acute PLX treatment was sufficient to significantly reduce tau-prion levels in the ipsilateral forebrain and hindbrain, as well as in the contralateral forebrain. Furthermore, tau-prion levels in the contralateral forebrain of PLX-treated mice were not significantly different from the forebrain of un-inoculated, vehicle-treated mice (Supplementary Fig. 5), which indicates that CSF1R inhibition prevented the spreading of tau-prions from the inoculation site to this brain region.
CSF1R inhibition can affect peripheral immune cells such as blood myeloid cells and tissue macrophages, in addition to microglia18,34. To determine if the effects of PLX3397 and PLX5622 on pathogenic tau in the brain are due, at least in part, to depletion of peripheral CSF1R-expressing cells we dosed Tg2541 mice with PLX73086, a non-brain penetrant CSF1R inhibitor analog of PLX3397 and PLX562235. Chronic treatment with PLX73086 had no significant effect on microglial markers Iba1 and P2YR12, or on levels of tau-prions or pTau[S396] or pTau[T231] in the forebrain or hindbrain of Tg2541 mice (Supplementary Fig. 6). Therefore, the effects of CSF1R inhibitors in peripheral compartments do not significantly contribute to their reduction of pathogenic tau in the CNS. Lastly, because there is limited data for CSF1R expression in neurons after injury36, we considered whether PLX3397 or PLX5622 might affect neurons or their expression of tau protein in Tg2541 mice. Acute, chronic, and terminal CSF1R inhibition did not significantly reduce levels of neuronal nuclei (NeuN), detected by IHC, or total tau, detected by ELISA (Supplementary Fig. 7). Therefore, CSF1R inhibitors do not directly affect measures of neuronal viability or tau expression, consistent with a prior report using PLX3397 in cultured primary neurons18. Together, these data argue that drug effects on biological and functional end points are due to inhibition of CSF1R in CNS microglia.
CSF1R inhibition extends survival and reduces behavioral deficits in female Tg2541 mice
We next focused on the terminal treatment paradigm with PLX3397 to evaluate the long-term effects of CSF1R inhibition on lifespan and behavior. Tg2541 mice develop paraparesis from 5–6 months of age22 which makes feeding difficult, resulting in a loss of body weight and thus, a greatly reduced lifespan compared to wildtype mice. We found that terminal PLX treatment significantly extended the median survival of female Tg2541 mice [16.5 days; P = 0.0004], but not male Tg2541 mice, compared to vehicle treatment (Fig. 2a,b). The extended survival in PLX-treated female mice was preceded by significantly reduced weight loss, which was not observed in male mice (Fig. 2c). Body weight at 180 days of age, irrespective of treatment, was predictive of lifespan in female mice but not in male mice, with less weight loss being correlated with longer survival (Fig. 2d). Lower forebrain tau-prion levels were also correlated with longer survival in female mice but not in male mice (Fig. 2e), suggesting that Tg2541 mice have a sex-specific physiological response to tauopathy. To confirm the effect of PLX treatment on survival in a different experimental paradigm, we used a midbrain inoculation model. Since Tg2541 mice spontaneously develop substantial tau pathology in the midbrain25, we predicted that K18 inoculation in the midbrain would accelerate and synchronize the disease course, which would be ideal for studying mouse survival. Indeed, female Tg2541 mice inoculated with K18 tau fibrils died significantly earlier than mice inoculated with diluent, though no difference was observed in male mice (Supplementary Fig. 8a,b). Consistent with our prior result, PLX treatment significantly extended the median survival of female mice inoculated with K18 tau fibrils [29.5 days; P = 0.0095], but not male mice, compared to vehicle treatment (Supplementary Fig. 8c). These data indicate that CSF1R inhibition robustly extends the lifespan of female Tg2541 mice, even during an accelerated disease course.
Previous studies have demonstrated a common hyperactive phenotype in the early stages of tauopathy in transgenic rodent models37,38. While the precise mechanism that leads to this deficit is unclear, this phenotype is causally linked with tau aggregate burden39. Based on the reduction of tau deposition we observed with PLX treatment, we sought to also examine its effect on this hyperactive phenotype. Using an automated home-cage monitoring system, we longitudinally tracked the activity levels in Tg2541 mice at different ages, measuring their amounts of rearing, locomotion, and wheel running. We confirmed previous reports that at early ages the Tg2541 mice displayed a hyperactive phenotype (90–150 days old in females, 90–120 days old in males), while at later ages their activity was significantly reduced (Fig. 2f), likely due to the accumulation of pathogenic tau in brain regions associated with motor function. PLX treatment led to a consistent reduction in Tg2541 mouse hyperactivity, but did not change their hypoactivity at later ages (Fig. 2f), indicating the activity reduction is not due to a general weakening effect. Detailed examination of the individual activity measurements revealed that PLX treatment normalized the amounts of wheel running (Fig. 2g) and active time (Fig. 2h). These data indicate that PLX treatment corrects the aberrant behavior of Tg2541 mice towards that of wildtype mice.
PLX3397 exposure is predictive of disease course in male and female Tg2541 mice
To examine the relationship between drug exposure and markers of disease progression more closely, we collected blood plasma at monthly intervals from mice receiving terminal treatment with PLX3397 or vehicle (Fig. 3a). Consistent with previous reports18, male Tg2541 mice had higher (25.3%; P < 0.0001) plasma concentrations of PLX than female mice (Fig. 3b and Supplementary Fig. 8d); we also observed this difference in wildtype mice (Supplementary Fig. 9a,b). Brain concentrations of PLX were also higher (44.9%; P = 0.0250) in male mice (Supplementary Fig. 8e). Male and female mice had ad libitum access to food, and had similar rates of food consumption relative to body weight, independent of whether it contained PLX or vehicle (Supplementary Fig. 10a,b). However, female mice were consistently more active than male mice (Supplementary Fig. 10c,d). Thus, the reduced PLX exposure in female mice is likely due to a higher metabolic and drug clearance rate compared with male mice. As only female mice benefitted from PLX, having extended survival and reduced weight loss (Fig. 2b,c), we hypothesize that drug exposure was excessive in male mice, resulting in adverse effects. In line with this hypothesis, we observed a trend towards reduced body weight in male wildtype mice receiving terminal PLX treatment, but not in female wildtype mice (Supplementary Fig. 9c).
Independent of sex, higher plasma concentrations of PLX were correlated with greater microglial depletion in both forebrain and hindbrain regions (Fig. 3c). Furthermore, higher PLX exposure was correlated with reduced tau-prion levels in the forebrain, but not in the hindbrain of Tg2541 mice (Fig. 3d). We also evaluated the plasma levels of neurofilament light chain (NfL), a validated blood-based biomarker of neuronal injury40, which correlates with disease progression and tau burden in human tauopathy41,42. Female PLX-treated mice had reduced levels of NfL compared to vehicle-treated mice (Fig. 3e). Conversely, NfL levels were increased in male mice following PLX treatment (Fig. 3f). We found no correlation between NfL level and survival in female mice (Fig. 3g), but in male mice, higher NfL levels were strongly correlated with reduced survival in both PLX and vehicle treatment groups (Fig. 3h). PLX treatment resulted in significantly increased NfL levels in male mice that received midbrain inoculation of K18 tau fibrils, compared to vehicle treatment (Supplementary Fig. 8f,g). Likewise, we found that PLX treatment increased the plasma NfL levels in male wildtype mice (Supplementary Fig. 9d,e), providing additional evidence of CSF1R inhibitor toxicity. Interestingly, intermittent treatment, resulting in a 50% lower total dosage of PLX, produced a significant decrease in NfL levels in male mice (Supplementary Fig. 4j–l). Taken together, these data further confirm that the drug exposure during chronic and terminal treatment with CSF1R inhibitors was appropriate for female Tg2541 mice to reduce tauopathy and significantly extend survival; however, the drug exposure was too high for male Tg2541 mice and likely resulted in adverse effects such as anemia, leukopenia, and hepatotoxicity, which have been observed in clinical trials43, that outweighed the therapeutic benefit.
CSF1R inhibition shifts gene expression patterns in Tg2541 mice towards wildtype
To better characterize global molecular changes in the CNS due to CSF1R inhibition, we used the Nanostring platform to analyze a curated panel of gene transcripts related to neuroinflammation, myeloid cell function and neuropathology in bulk brain tissue following chronic treatment of Tg2541 mice with PLX5622. We measured mRNA transcripts of 1,841 genes, many of them shown to be regulated by tau or Aβ pathology in previous genome-wide gene expression studies12,14,44. To validate the Nanostring approach, we identified 53 genes with a broad range of expression level changes and measured mRNA transcripts with quantitative reverse-transcription PCR (RT-qPCR) in the same sample used for sequencing. The RT-qPCR results matched the trends shown in the Nanostring data (Supplementary Fig. 11), indicating that our transcriptomic data was robust.
Since microglial cells are directly impacted by PLX treatment, we first excluded the microglial-specific genes (see Methods) and examined the general trend of expression patterns among different treatment groups. Pearson’s correlation matrix showed high similarity among wildtype brains with or without PLX treatment (Fig. 4a), indicating that the gene expression pattern we measured is not affected by the treatment itself. In contrast, PLX treatment in Tg2541 mice caused a distinct shift in the gene expression pattern away from the vehicle-treated group. Interestingly, the correlation coefficients for PLX-treated Tg2541 mice are higher with wildtype mice than with vehicle-treated Tg2541 mice (Fig. 4a dashed boxes and Fig. 4b). To further quantify this shift, we performed partial-least squares (PLS) regression analysis using the gene expression data from vehicle-treated Tg2541 and wildtype mice (Fig. 4c filled circles), and projected the data from PLX-treated mice onto the PLS dimensions (Fig. 4c hollow circles). This allows us to represent the transgene-specific gene expression pattern in a relatively low-dimensional space, and to quantify the changes associated with treatment by calculating the population vector distances and angles in this space. We found that PLX treatment significantly normalized gene expression patterns in Tg2541 mice towards those of wildtype mice (Fig. 4d,e, only two out of five dimensions are shown, covering >95% of the total variance). The normalization in gene expression was further confirmed by similar trends in neuronal-specific genes (Supplementary Fig. 12). Importantly, PLX treatment in wildtype mice showed negligible changes in the gene expression patterns. These results indicate that PLX treatment specifically suppresses the abnormal transcriptome associated with transgene overexpression, consistent with its effects ameliorating pathogenic tau deposition.
Evidence for excitotoxicity with increased drug exposure
As described above, although we observed consistent reduction in the levels of pathogenic tau in the brains of both male and female Tg2541 mice with PLX treatment (Fig. 1), only female mice benefited from extended survival, functional rescue, and reduced NfL levels (Fig. 2 and Fig. 3). We hypothesize that excessive PLX dosing may underlie this sex-specific effect, as male mice consistently had higher plasma and brain drug concentrations (Fig. 3 and Supplementary Fig. 8), and also benefitted from a lower, intermittent PLX dosing paradigm (Supplementary Fig. 4). Indeed, in our transcriptomics analysis we identified individual genes whose expression was associated with brain PLX concentration or with sex (Fig. 4f). Using a PLS regression of all non-microglia genes to brain PLX concentration and sex for each sample, we calculated the variable importance score along each of these dimensions. Interestingly, many immediate early genes (IEGs) showed high importance scores (Fig. 4g), suggesting that IEGs might be a module that is altered by drug exposure. To further examine this possibility, we examined all immediate early genes45,46 (56 genes overlapped in our dataset) and found that their expression patterns fit closely with the brain PLX concentration (Fig. 4h). Importantly, when we excluded the IEGs and examined the PLX-induced transcriptome shift along the WT to TG dimension, the gene expression changes and brain PLX concentration were no longer correlated (Fig. 4i), indicating that the IEGs contribute substantially to the PLX treatment effects. Notably, relative to WT vehicle-treated mice, only male PLX-treated Tg2541 mice had significantly upregulated expression of IEGs (Fig. 4j). As increased IEG expression can be indicative of neuronal hyperactivity, these data provide a plausible mechanism by which excessive PLX dosing may have led to excitotoxicity, thereby masking its therapeutic effect in male mice.
CSF1R inhibition ameliorates pathological activation of astrocytes
As suggested by the reduction of tau deposition, we hypothesized that astrocyte-driven neuroinflammation would also be reduced by PLX treatment. Therefore, we examined transcriptome shifts in astrocyte-specific genes upon PLX treatment. Similar to the neuronal-specific genes, we observed a normalization of astrocyte-specific gene expression patterns towards WT in both forebrain and hindbrain regions (Fig. 5a, b). In addition, we measured astrogliosis over time using longitudinal bioluminescence imaging (BLI) methods based on a previously established transgenic reporter system of glial fibrillary acidic protein (GFAP)-driven luciferase47,48. To perform reliable BLI in Tg2541 mice, we intercrossed each transgenic line to an albino background and refined the method, using a synthetic luciferin substrate to increase signal from deep hindbrain regions (see Methods and Supplementary Fig. 13). This technique allowed us to non-invasively measure astrogliosis in live mice longitudinally over the course of PLX treatment. In vehicle-treated Tg2541 mice, the BLI signal gradually increased with age, in accordance with the accumulation of tau pathology and gliosis reported in Tg2541 mice22,49. Consistent with our hypothesis, CSF1R inhibition suppressed the BLI signal in both forebrain and hindbrain (Fig. 5c–e), suggesting that astrocytic inflammation driven by microgliosis was attenuated, thus leading to a general neuroprotective effect.
CSF1R inhibition preferentially eliminates a highly activated microglia subpopulation
We next examined the morphological and transcriptional changes in microglia after CSF1R inhibition. In tauopathy, microglia acquire an activated morphology in brain regions where neurons contain tau aggregates (Fig. 6a), a phenomenon seen in many focal neuropathologies50,51. Interestingly, the elimination of microglia in the Tg2541 mouse brain following PLX3397 treatment was not uniform nor complete, but was the most effective in the vicinity of tau aggregates. The microglial density near tau-laden neurons was reduced by more than 60%, but in distal regions (>200 microns) the microglial density was not significantly changed (Fig. 6b), indicating that microglia in the vicinity of the tau aggregates may have increased sensitivity towards CSF1R inhibition. We then compared the morphologies of PLX-resistant (residual microglia in PLX-treated mice) and PLX-sensitive microglia (in vehicle-treated mice). Notably, we found that PLX resistance is associated with more abundant and intricate microglial cell processes, close to the levels seen in wildtype mice (Fig. 6c–f). These data suggest that microglia associated with tau pathology may be in an “activated” state, with a reduced number of processes. This view is consistent with previous reports that activated microglia adopt a “disease-associated microglia” (DAM) phenotype that is pro-inflammatory and detrimental to adjacent neural cells13,52. Interestingly, our data suggests that DAMs are more vulnerable to PLX treatment, and that surviving microglia are less neurotoxic and potentially neuroprotective.
In support of this view, transcriptome analysis showed that many microglial-specific genes were upregulated in Tg2541 mice (Fig. 7a), among which the most notable were signature DAM genes such as Tyrobp, Clec7a, Trem2 and CD68. By correlation analysis among different samples using a generalized Louvain algorithm53, we found that the microglia-specific genes in our dataset were clustered into three groups (Fig. 7b and Supplementary Fig. 14). The red-cluster genes showed the highest degree of modulation by transgene overexpression, while the blue-cluster genes showed a moderate degree of modulation, and the green-cluster genes showed almost no difference between Tg2541 and wildtype mice (Fig. 7c). Transgene-modulated genes clustered into red and blue groups, consistent with a recent finding that tau pathology activates both immune-activation and immune-suppression gene expression modules10. Interestingly, when we compared the gene expression patterns in vehicle- and PLX5622-treated brains, the expression of red- and blue-cluster genes were substantially diminished by PLX, while the green-cluster genes were largely unchanged by the treatment (Fig. 7c). The stability of gene expression in the green cluster is remarkable considering the >50% reduction in microglial cell number. Given that PLX predominantly eliminated microglia in the vicinity of the tau deposits (Fig. 6b), these data suggest that red-cluster genes are preferentially expressed by tau-associated microglia. On the other hand, the lack of changes in green-cluster genes suggests that they are preferentially expressed by the quiescent population of microglia located further away from tau deposition, which are more resistant to PLX-mediated elimination. Moreover, we examined all previously reported DAM signature genes52 and we found a partial match with the activation markers in each of our identified gene clusters (Fig. 7d). Regardless of the designation of homeostatic or activation genes reported in previous studies, the red-cluster genes showed a stereotypical pattern of transgene activation and sensitivity to PLX treatment, while green- and blue-cluster genes did not appear to be modulated by these factors. Taken together, our data show morphological and transcriptional changes in microglia associated with tau deposition, consistent with a pattern of pathological activation. CSF1R inhibition appears to preferentially eliminate these microglia, leaving the brain with a more quiescent and less inflammatory microglial population.
DISCUSSION
Our study revels several major findings from a comprehensive evaluation of CSF1R inhibitors in preclinical models of tauopathy. Importantly, we present the first line of evidence that CSF1R inhibition reduces pathology that leads to functional improvements associated with longer lifespan in tauopathy mice (Fig. 2 and Supplementary Fig. 8). Overall, our data showing a reduction of pathogenic tau is consistent with prior studies using a different drug scaffold targeting CSF1R (JNJ-527; edicotinib) in Tg2541 mice19, or using PLX3397 in a different mouse model of tauopathy (TgPS19)18. However, in our study, neuroprotection occurred despite incomplete microglia depletion (∼60%); upon deeper analysis, we identified distinct microglial-specific gene clusters suggesting subsets of microglia responsive to tauopathy or resilient to CSF1R inhibition (Fig. 7). This finding is in line with the wealth of data demonstrating that microglia exist as unique subsets in different brain regions, sexes, ages or disease states2–4. From this perspective, our data suggest that it may be possible to target specific subsets of activated microglia in tauopathy, while leaving other beneficial microglia alone. Taken together, we argue that CSF1R inhibition causing complete microglia ablation is unnecessary for therapeutic benefits, and may possibly be detrimental in humans given that microglia are important for brain homeostasis and defending against other insults. This notion is at odds with Shi et al., who argue that after PLX3397 treatment, the residual, activated microglia are detrimental18. This discrepancy could be explained by differences in mouse models such as disease kinetics, the genetic promoter driving the transgene, or the background genetics of the mouse54. Another possibility is that Shi et al. use a higher drug dose (400 mg/kg) to achieve ∼100% microglial ablation, and this may negatively shift the phenotype of residual microglia, or cause concentration-dependent off-target effects on other kinases such as KIT in neurons or PDGFRα in oligodendrocyte precursor cells (OPCs)55, leading to reactive microglia. In contrast, a study by Bennett et al. showed that partial depletion (∼30%) of microglia in Tg4510 mice did not rescue any pathological phenotype20, but PLX3397 was administered to aged mice with advanced pathology, which could explain the negative result, as has been seen in Aβ models56,57; additionally, Bennett et al. used drug formulated in a different chow which may have reduced its exposure and efficacy, as shown previously18. Furthermore, Tg4510 mice are confounded by genetic factors other than tau causing pathology58, thus making it altogether difficult to interpret this data. Nevertheless, these disparate findings in prior literature are now more interpretable alongside our study, which sheds light on the intricate relationships between CSF1R inhibitor dosing, microglial depletion and therapeutic outcomes.
The precise mechanism of CSF1R inhibitors causing reduced tau pathology is still unclear, but our data indicates that activated microglia are the primary target resulting in reduced numbers of cells producing pro-inflammatory cytokines7–9 and other disease-associated microglial factors (Fig. 7) that stoke tau pathogenesis in neurons, such as apolipoprotein E18,59,60 and complement proteins61–63. In addition, it seems plausible that CSF1R inhibitors may also block microglia-mediated activation of astrocytes64, which in turn secrete factors that also drive tau pathogenesis; blocking this cellular feed-forward pathway using a different drug targeting microglia led to neuroprotection in a synucleinopathy model of Parkinson’s disease65. Consistent with this view, PLX-treated mice exhibited a restored astrocyte phenotype (Fig. 5), suggesting that therapeutic benefits in our study may also be due, in part, to quelling disease-associated astrocytes. While there is cross-talk between peripheral immune cells and microglia1, we show that a non-brain penetrant analog, PLX73086, did not affect CNS microglia or tau pathology (Supplementary Fig. 6), and it is thus unlikely that CSF1R inhibition in the periphery contributes to the phenotypic rescue observed in our study. Lastly, it is possible that chronic PLX treatment caused depletion of some OPCs in our study, but we expect that PLX did not affect mature oligodendrocytes or myelination55. The relationship between OPC biology and tau pathology in neurons is largely unknown, and thus it remains unclear how CSF1R inhibition in OPCs contributes, if at all, to the mechanism of action. Nevertheless, this topic warrants further investigation.
Tau pathology in Tg2541 mice is associated with moderate microgliosis and an up-regulation of transcriptomic signatures of microglial activation10,13. Our transcriptome analysis showed activation of two major clusters of microglial-specific genes in Tg2541 mice. These two clusters had a high degree of overlap with the immune activation and suppression modules recently described in tauopathy mice and FTLD patients10, indicating a specific reactive transcriptional program of microglia towards tau pathology. Consistent with this view, genes in the activated clusters also matched transcriptome modules described in activated microglia in neurodegeneration models (such as Itgax and Clec7a), but not in tumor or acute inflammation models13. We found an additional cluster of microglial genes that had similar expression in Tg2541 and wildtype mice. Intriguingly, this cluster was not affected by PLX treatment, while the other two clusters showed significant down-regulation (Fig. 7). Considering that PLX eliminates more than half of the total microglia population, a parsimonious explanation for this sustained gene cluster is that they are preferentially expressed by an inert microglial subpopulation that does not respond to tauopathy or CSF1R inhibition. Consistent with this notion, we found that PLX treatment preferentially eliminates activated microglia in the vicinity of tau deposits, and thus most surviving microglia are not in direct contact with tau-laden neurons (Fig. 6). This is in contrast to Aβ mice, in which the surviving microglia are usually associated with Aβ plaques following PLX treatment27,57. We found that surviving microglia were non-inflammatory, and had longer and more elaborate processes compared to vehicle-treated microglia, showing functional and morphological features more similar to those of wildtype microglia (Fig. 6). In sum, our data describe a microglial genetic signature that remains stable in Tg2541 mice with or without PLX treatment, likely representing a “dormant” microglial subpopulation that are less dependent on CSF1R for survival, or are less sensitive to CSF1R inhibition at the doses administered in our study.
Sex-specific differences exist in mouse microglial function, gene expression, and response to tauopathy, and the differences increase with age60,66,67. Our data identify a sex-dependent effect on therapeutic exposure and efficacy of CSF1R inhibition in Tg2541 mice. A difference in the plasma levels of PLX3397 during ad libitum oral dosing in male and female mice has been noted previously18; however, only male mice were evaluated further. We examined both male and female Tg2541 mice and found that, despite similar food intake, plasma (and brain) levels of PLX3397 were higher in male mice compared to female mice (Fig. 3 and Supplementary Fig. 8). However, at this level of drug exposure, only female mice received a functional benefit from CSF1R inhibition, an unexpected and clinically relevant outcome that would have been overlooked had our analysis been focused on a single sex. In male Tg2541 mice, despite a robust reduction of microglia and pathogenic tau, PLX treatment did not slow weight loss or extend survival, and plasma NfL levels were significantly increased, indicative of neuronal damage40. It has been suggested before that microglia from male animals may exhibit an increased responsiveness to CSF1R depletion compared to microglia from female animals68. Our results indicate that despite robust on-target effects for microglial depletion, male mice are more susceptible to toxic off-target effects (on non-microglial cells) of CSF1R inhibitors than previously known. Furthermore, we observed a concentration-dependent activation of IEGs in PLX treated Tg2541 mice, suggesting that excessive PLX dosing in male mice may lead to excitotoxicity (Fig. 4), thus masking the beneficial effect of tau reduction. Curiously, IEGs were not significantly upregulated in male WT mice, indicating that their activation may not be due to high concentration of PLX alone, but may also be dependent on tau deposits. Previous studies have linked tau accumulation and neural activity in vivo69. On the other hand, microglia are known to mediate neuroprotection against excitotoxicity70,71 and elimination of microglia can exacerbate seizures and related neuronal degeneration72. Therefore, the concurrent tau removal and microglial elimination may increase the risk for hyperactivity, resulting in excitotoxicity in male mice with high PLX concentrations. Other sex-specific differences may also contribute to microglial sensitivity to CSF1R inhibition by a currently unknown mechanism. Future translational studies of pharmacological CSF1R inhibitors will need to carefully evaluate the role of sex in both safety and therapeutic outcomes.
CSF1R inhibitors were shown to be protective in mouse models of other neurodegenerative diseases, such as AD and Down syndrome27,73–75. However, under different treatment conditions, CSF1R inhibition did not affect Aβ plaque burden, but did rescue some functional deficits56,57. Therefore, microglia play a dynamic role in the brain’s response to Aβ pathogenesis, and their attenuation may impart distinct benefits at different stages of disease. Our results suggest that, in primary human tauopathies, a subset of microglia play a net negative role before, during, and after disease onset and that their removal may be a viable therapeutic strategy. It remains to be determined if similar benefits should be expected for tauopathy in AD given the preceding comorbid Aβ pathology, but this may be elucidated in rodent models co-expressing human tau and Aβ. Nevertheless, because CSF1R inhibition has not been reported to be detrimental in Aβ mice, CSF1R inhibitors could ameliorate AD-related tauopathy even if caused by different disease mechanisms. Ongoing human clinical trials of CSF1R inhibitors in AD (e.g. NCT04121208) may provide additional mechanistic insights.
Primary human tauopathies constitute a class of neurodegenerative diseases caused by tau misfolding and aggregation and include progressive supranuclear palsy (PSP), corticobasal degeneration (CBD) and Pick’s disease, among others76. When combined with AD, in which tau aggregation follows Aβ deposition, tauopathies afflict a significant proportion of the human population, and thus novel approaches to directly or indirectly block tau pathogenesis or its downstream effects are urgently needed. Our study highlights several aspects of pharmacological CSF1R inhibition that bolster its therapeutic potential for human tauopathies. First, we demonstrated that interventional dosing of Tg2541 mice, initiated at a stage when robust tau deposition had already occurred22,25, led to a significant reduction in pathogenic tau (Supplementary Fig. 4). Therefore, our data support the clinical benefit of CSF1R inhibitors for both treatment, and prevention, of tauopathy. This is important because prophylactic treatment of non-autosomal dominant neurodegenerative diseases is difficult due to a lack of definitive prognostic biomarkers paired with the fact that aggregation of the causative proteins can occur years or decades prior to symptom onset29,77. Second, we found that intermittent dosing of Tg2541 mice at three-week intervals produced a significant reduction in pathogenic tau (Supplementary Fig. 4). Despite relatively minimal off-target effects from continuous, long-term dosing of CSF1R inhibitors in mice27, non-human primates78, and humans26,79, intermittent dosing would be clinically preferable if a similar therapeutic outcome was achieved, given the important functions for microglia and related peripheral cells in innate immunity. Because neurodegenerative tauopathies are slow, protracted diseases and microglia are long-lived80, it is conceivable that breaks in dosing may occur on the order of months or years and be informed by medical imaging probes for microglia activation81. Third, we found CSF1R inhibition to extend the survival of female Tg2541 mice (Fig. 2 and Supplementary Fig. 8), indicating that the reduction in pathogenic tau in this model system translates to an improved clinically relevant outcome. We postulate that if CSF1R inhibitor dosing was optimized for male Tg2541 mice, any adverse effects in the CNS or periphery would likely be diminished and their survival extended. Fourth, we showed that complete microglial depletion is not necessary, or even desirable, for a therapeutic benefit. As discussed above, the microglia that survive CSF1R inhibition represent a unique microglial sub-population that likely serves important functions in brain homeostasis. Future preclinical studies may pinpoint the precise level, timing, and frequency of CSF1R inhibition such that the detrimental effects of microglial activation are minimized while an appropriate number of homeostatic microglia remain for brain surveillance. Lastly, CSF1R inhibitors applied in conjunction with tau immunotherapy82,83 may prove to be a successful combination therapy; because microglia are not needed for antibody effector function84, removing tauopathy-activated microglia would slow tau pathogenesis and may also increase the efficacy of tau immunotherapy. Taken together, our data strongly support the therapeutic modulation of microglial activation by CSF1R inhibitors as a promising approach to treating human tauopathies.
METHODS
Animals
The Tg2541 transgenic mouse line expresses the human 0N4R tau isoform under the Thy1.2 genetic promoter. Tg2541 mice were originally generated on a mixed C57BL/6J × CBA/Ca background22 and were then bred onto a congenic C57BL/6J background using marker-assisted backcrossing for eight generations before intercrossing to generate homozygous mice. Albino Tg2541 mice were generated by intercrossing Tg2541 with C57BL/6J mice expressing a spontaneous mutation in the tyrosinase gene (homozygous for Tyrc-2J) causing albinism (Jackson Laboratory; 000058). To generate mice for in vivo bioluminescence imaging, we employed Tg(Gfap-luc) mice, which express firefly luciferase under the control of the murine Gfap promoter (gift from Caliper Life Sciences). These reporter mice were originally on the FVB background, but we backcrossed them to a congenic B6 background, and then crossed them to the B6-albino background. To create bigenic mice, albino Tg2541 mice were crossed with albino Tg(Gfap-luc) animals to produce double hemizygotes; next, double hemizygotes were crossed and the progeny were screened for the presence of both transgenes expressed at homozygosity. Animals were maintained in a facility accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International in accordance with the Guide for the Care and Use of Laboratory Animals. All procedures for animal use were approved by the University of California, San Francisco’s Institutional Animal Care and Use Committee.
PLX compound formulation in mouse chow
PLX3397 was provided by Plexxikon Inc. and was formulated in AIN-76A standard chow by Research Diets Inc. at 275 mg/kg as previously described85. PLX5622 was provided by Plexxikon Inc. and was formulated in AIN-76A standard chow by Research Diets Inc. at 1200 mg/kg as previously described57. PLX73086 was provided by Plexxikon Inc. and was formulated in AIN-76A standard chow by Research Diets Inc. at 200 mg/kg as recommended by Plexxikon Inc.
Immunohistochemistry and slide scanning
Formalin-fixed samples were embedded into paraffin using standard procedures and microtome-cut into 8 µm sagittal brain sections or coronal spinal cord sections and mounted onto slides. To reduce tissue autofluorescence, paraffin slides were photobleached for 48 hours86. Slides were deparaffinized in a 61°C oven for 15 minutes and rehydrated through alcohols. Antigen retrieval was performed by autoclaving for 10 minutes at 121°C in 0.01 M citrate buffer. Sections were blocked in 10% normal goat serum (NGS) (Vector Labs) for 1 hour at room temperature. All primary antibodies were used at 1:250 dilution and included rabbit monoclonal anti-Iba1 (Abcam, ab178847), rabbit polyclonal anti-P2YR12 (Atlas, HPA014518), mouse monoclonal anti-NeuN (Millipore, MAB377), and mouse monoclonal anti-pS202/T205 tau (AT8; Thermo Fisher, MN1020). Primary antibodies were diluted in 10% NGS in PBS and allowed to incubate on the slides overnight at room temperature. Primary antibody detection was performed using goat secondary antibodies with conjugated AlexaFluor 488, AlexaFluor 555, or AlexaFluor 647 (Life Technologies) at 1:500 dilution. Slides were cover-slipped using PermaFluor mounting medium (Thermo). Whole-section tiled images were acquired with an Axioscan.Z1 slide scanner (Zeiss) at 20× magnification, and quantification was performed with Zen 2.3 software (Zeiss).
Cellular bioassay to measure tau-prion activity
A HEK293T cell line expressing the repeat domain of 4R human tau (aa 243–375) containing the P301L and V337M mutations and C-terminally fused to YFP was previously generated as described28,87. A stable monoclonal line was maintained in DMEM, supplemented with 10% FBS and 1% penicillin/streptomycin. To perform the bioassay, 3,000 cells (containing 0.1 μg/ml Hoechst 33342) were plated in 70 μl per well into 384-well plates (Greiner) and incubated for 2 hours before treatment with samples. Clarified brain lysate at a final concentration of 1.25 µg/mL total protein was first incubated with Lipofectamine 2000 (0.2% final concentration) and OptiMEM (9.8% final concentration) for 90 minutes, and then added to the plated cells in quadruplicate. Plates were incubated at 37°C for 1–3 days, and then the live cells were imaged using an INCell Analyzer 6000 Cell Imaging System (GE Healthcare) and custom algorithms were used to detect fluorescent YFP-positive puncta (aggregates).
Mechanical tissue homogenization
Postmortem brains and spinal cords were thawed and weighed to determine the mass in grams. Brains were bisected into forebrain and hindbrain pieces using a single cut with a scalpel blade between the striatum and hypothalamus. Tissue was homogenized in nine volumes of cold DPBS containing Halt Protease Inhibitor Cocktail (1x, Thermo Fisher Scientific) using a Precellys 24-bead beater (Bertin Instruments) with metal bead lysing matrix (MP Biomedical). Where necessary, brain lysates were clarified by centrifugation at 10,000 × g for 10 min at 4°C. All tissue and samples were stored at −80°C until further use.
Formic acid extraction of insoluble proteins in brain tissue for ELISA
Fifty microliters of formic acid were added to 25 µL of 10% brain homogenate and placed in an ultracentrifuge tube. The samples were vortexed, sonicated for 20 minutes at 37°C in a water-bath sonicator, and then centrifuged at 100,000 x g for 1 hour. Fifty microliters of supernatant were recovered to a low-binding tube and neutralized with 950 µL of neutralization buffer (1 M Tris base and 500 mM dibasic sodium phosphate). Samples were aliquoted into low-binding tubes and flash frozen in liquid nitrogen. The following ELISA kits from Thermo Fisher Scientific were used according to the manufacturer’s protocols: total tau (KHB0041), p-tau S396 (KHB7031), and p-tau T231 (KHB8051). Each sample was analyzed in duplicate. Raw ELISA values were adjusted to total brain protein (grams) in the clarified 10% brain homogenate as determined by bicinchoninic acid (BCA) assay (Pierce/Thermo Fisher Scientific).
Quantification of total protein in brain homogenate
Total protein content in the PBS-soluble (clarified 10% brain homogenate) and detergent-soluble fractions was quantified using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific) following the manufacturer’s protocol.
Generation of tau K18*P301L fibrils
Production, purification and fibrillization of recombinant tau K18*P301L fibrils were performed as previously described88.
Stereotaxic injections in Tg2541 mice
Forebrain inoculation: Ten-week-old Tg2541 mice received unilateral inoculations of 10 µl of 1.5 mg/ml tau K18 P301L fibrils using stereotaxic methods. Injections followed a two-step process: the needle was first advanced to the hippocampus (Bregma -2.5mm, Lateral 2.0mm; Depth -2.3 mm from the skull surface) to deliver 5 µl over three minutes, then the Hamilton syringe pump was paused for five minutes to allow for diffusion prior to retracting the needle to the overlying cortex (Depth -1.3 mm) where the remaining 5 μl was injected. After fibril injection, the needle remained in place for five minutes to allow for diffusion of fibrils before retraction, patching the skull and suturing the scalp. Midbrain inoculation: Ten-week old Tg2541 mice received bi-lateral inoculations of 10 µl of 1.5 mg/ml tau K18 P301L fibrils using stereotaxic methods. Five microliters was injected at each site in the midbrain (Bregma, -4.3 mm; Lateral, 1.0 mm, Depth, - 2.5 mm) and (Bregma, -4.3 mm; Lateral, -1.0 mm, Depth, -2.5 mm).
Automated home cage monitoring of behavior
Total activity measurements of freely moving mice were made every 30 days after PLX dosing in Promethion cages (Sable Systems International). At each time point, mice were first randomized and placed individually in Promethion cages for 4 to 6 days. Real-time cage activity recording was continuous during the entire session using a combination of a running wheel with sensors to measure speed and distance traveled, three balances to measure body weight, food and water consumption, and a matrix of infrared light beams to measure XYZ movements with 0.25 cm resolution. Analysis of these metrics was used to detect behaviors such as sleep, rearing and general locomotion. For each mouse, data used for analyses were average readings per light or dark cycle. Data from the first circadian cycle were excluded due to variable behavior during habituation. To calculate the activity scores, wheel use, locomotion and rearing were first normalized to a 0–1 scale by the maximum value in the whole dataset, and then the geometric mean of the normalized values for each session was calculated.
Quantification of PLX compound levels in brain tissue and plasma
Brain homogenates (20% w/v) were prepared in PBS by one 30-second cycle of bead beating at 5500 rpm with a Precellys 24-bead beater (Bertin Instruments) or plasma samples were prepared by dilution to 25% with PBS. Compounds were recovered by mixing equal parts of brain homogenate with a 50/50 (v/v) solution of acetonitrile (ACN) and methanol containing 1 mM niflumic acid. Precipitated proteins were removed by vacuum filtration (Captiva ND, Agilent). Analysis was performed using a liquid chromatography-tandem mass spectrometry system consisting of an API4500 triple quadra-pole instrument (AB Sciex, Foster City, CA) interfaced with a CBM-20A controller, LC20AD 230 pumps, and a SIL-5000 auto-sampler (Shimadzu Scientific, Columbia, MD). Samples were injected onto a BDS Hypersil C8 column maintained at room temperature. The amount of ACN in the gradient was increased from 75– 95% ACN over two minutes, held for one minute, and then re-equilibrated to 75% ACN over 1.4 minutes. Data acquisition used multiple reaction monitoring in the positive ion mode. Specific methods were developed for each compound (PLX3397 and PLX5622), enabling the determination of absolute concentrations.
Blood plasma neurofilament light (NfL) protein measurement using SIMOA
At monthly time points, 150 µl blood was collected in EDTA-coated tubes. The plasma was centrifuged at 1,000 x g for ten minutes to clarify the samples, and was then diluted with sample diluent buffer included in the kit by 25-fold and 100-fold, respectively, prior to the measurement. Plasma NfL concentration was measured and analyzed using the NfL kit (Quanterix) with the SIMOA HD-1 analyzer (Quanterix). Briefly, samples, magnetic beads coated with capture antibody, and biotinylated detector antibodies were combined. Thereafter, the capture beads were resuspended with streptavidin-β-galactosidase (SBG) and resorufin β-D-galactopyranoside (RGP) and transferred to the SIMOA disk. Each bead fit into a microwell in the disk and if NfL was captured then the SBG hydrolysed the RGP substrate which generated a fluorescent signal, and then the concentration was measured against a standard curve derived from known concentrations of recombinant NfL included in the kit. The lower limit of quantification of the assay for plasma was 17.15 pg/mL.
RNA extraction and Nanostring RNA expression measurements
RNAlater-preserved samples were homogenized in PBS and total RNA was extracted from samples using the Quick-RNA Miniprep Kit (Zymo Research). RNA extracts were evaluated for concentration and purity using a Nanodrop 8000 instrument (Thermo Fisher Scientific) and diluted to a concentration of 20 ng/µl. Hybridizations were performed for the mouse Neuroinflammation, Myeloid cell, and Neuropathology panels according to the nCounter XT Assay user manual (Nanostring). The hybridizations were incubated at 65°C for 16 hours, and then were added to the nCounter SPRINT Cartridge for data collection using the nCounter SPRINT Profiler. Counts were analyzed using the nSolver Analysis Software.
RNA expression analysis
In total, there were 10 mice in the Tg2541 vehicle group, 10 mice in the Tg2541 PLX5622 group, 6 mice in the wildtype vehicle group, and 6 mice in the wildtype PLX5622 group. Each mouse had separate forebrain and hindbrain samples and three panels of Nanostring sequencing were performed on each sample. Data from the three panels were pooled together to form the final dataset. When pooling data, if a gene appeared in more than one panel then the average read value was used in subsequent analysis, unless one panel failed to detect the gene.
To assign cell-type specificity of each gene, we used the transcriptome dataset reported in a previous study89, inspired by previously reported approaches in bulk tissue samples13. We set a specificity threshold in which a gene qualifies to be cell-type specific if its expression in a cell type is greater than five times the sum in all other cell types. Using this standard, our dataset had 242 microglia-specific genes, 47 astrocyte-specific genes and 70 neuron-specific genes. All cell-type specific gene analyses were repeated with a three-time threshold and all results were consistent (data not shown).
We used partial least-square (PLS) regression (MATLAB) to extract the gene expression pattern aligned with Tg2541-wildtype axis, using individual gene reads from each mouse as predictors and genotype as responses. Only vehicle groups were used in constructing the PLS regression. Forebrain and hindbrain were calculated separately. Five output dimensions were chosen for all PLS analyses, as they covered 99.99% of the total variance in all cases. The scores in the first two dimensions were plotted. To project PLX3397-treated groups to the PLS dimensions, we used the following formula:
To calculate population vector distance, we use the “mhal” command in MATLAB. All five dimensions were used for each mouse. The wildtype vehicle group was used as a target.
To calculate the vector angle, each mouse’s gene expression pattern was regarded as a five-dimension vector in the PLS space, and the angle between each mouse and the average vector of the wildtype vehicle group was calculated with the following formula:
To calculate the PLS regression along the PLX concentration and sex-correlated dimensions, we constructed regressions using all non-microglial genes or only immediate early genes45 to measured brain PLX concentrations and sex of each sample. We then calculated variable importance in projection to isolate the genes important for the regression. To calculate the projected PLX concentrations, we used the products of gene expression levels and coefficients estimated from PLS regression.
To calculate clusters in the microglial-specific genes, we calculated pairwise Pearson’s correlation coefficients across 32 samples among each gene. The resulting similarity matrix was then processed with a generalized Louvain community detection algorithm53.
Gene expression analysis by RT-qPCR
Mouse brains were collected at endpoints and flash frozen in DNA/RNA shield reagent. Tissue was homogenized as described above and total RNA was purified using a commercial isolation kit (Zymo Research). RNA concentration and the RNA integrity number (RIN) were determined using a Bioanalyzer 2100 instrument and an Agilent RNA 6000 Pico Kit (Agilent 5067-1513). Only samples with a RIN score ≥7.0 were used for gene expression analysis. To confirm transcriptome profiling results, 2.5 ng of sample mRNA was applied to triplicate RT-qPCR reactions consisting of 1x TaqPath 1-Step Multiplex Master Mix (ThermoFisher Scientific A28526), Taqman primer/probe sets and a normalizing human MAPT Taqman assay. Reactions were run on a QuantStudio 6 and 7 Pro instrument and amplification yielding cycle threshold (CT) values were corrected with Mustang Purple passive reference dye for each target gene. Gene expression of PLX-treated mice relative to vehicle-treated mice was determined by the comparative CT method and values were expressed as fold-change90.
Comparative CT equation:
2−ΔΔCT=[(CT gene of interest – CT hMAPT internal control)]PLX-treated mice − [(CT gene of interest – CT hMAPT internal control)]vehicle-treated mice
In vivo bioluminescence imaging
Bioluminescence imaging was performed on the brains and spinal cords of albino bigenic Tg(2541:Gfap-luc) homozygous mice after receiving an intraperitoneal injection of 25 mg/kg cyclic luciferin-1 (CycLuc1) sodium salt solution (Aobious; AOB6377) prepared in PBS, pH 7.4. After CycLuc1 injection, mice were placed in an anesthetization chamber and exposed to an isoflurane/oxygen gas mix for ten minutes. During this time, the heads of the mice were shaved to enhance the bioluminescence signal. After anesthetization, mice were placed in an IVIS Lumina III small animal imaging system (PerkinElmer) and were kept under constant anesthesia. Mice were imaged for 60 s duration at three time points (14, 16 and 18 minutes) following CycLuc1 injection as determined in one-hour time-lapse calibration studies. After image acquisition, the mice were allowed to recover in their home cages. Brain and spinal cord bioluminescence values were calculated from images displaying surface radiance using standardized regions of interest and were then converted to total photon flux (photons per second) using Living Image software version 4.4 (PerkinElmer).
Confocal imaging of thick tissue sections
Vibratome-sectioned brain slices (40 µm thick) were immunolabeled with Iba1 and AT8 antibodies using standard protocols for free-floating sections in multi-well plates. Sections were mounted using PermaFluor and #1.5 coverglass. Using a Leica SP8 confocal microscope equipped with HyD detectors and an AOBS, samples were first visualized using Navigator function to acquire an overview image of each slice using a 20× water-immersion lens (0.95 NA). From the mosaic image, smaller tiled-ROIs were marked in the forebrain and hindbrain to acquire high-resolution, sequential-scanned image stacks using a 63× water-immersion lens (1.2 NA). Eight-bit image z-stacks (1 μm steps) were collected at 512×512-pixel resolution. Images were processed using custom MATLAB code.
Microglial morphology analysis
Microglia morphology was analyzed using a custom script in MATLAB. Briefly, raw confocal image stacks were smoothed and then maximally projected. Isolated microglia cells were manually selected for analysis. The selected microglia region was binarized with an intensity threshold, and then the cell body was detected by fitting a largest circle in the binary mask. After excluding the cell body region, the remaining microglia processes were skeletonized and branch number, branch length and bounding box were measured using “regionprops” and “bwmorph” commands.
Statistical analysis
All statistical analyses were performed using GraphPad Prism 8. Comparisons between two groups were performed by two-tailed unpaired t test or by Mann-Whitney nonparametric test. For comparisons of more than two groups, one-, two-, or three-way ANOVA was performed with Holm-Šidák post hoc analysis. Following ANOVA, residuals were evaluated for normal distribution using the Anderson-Darling test and the data were evaluated for equal variance using the Brown-Forsythe test. If both assumptions were violated (P < 0.05), the data was reanalyzed using Welch’s ANOVA with Dunnett T3 post hoc analysis. For repeated-measures ANOVA, sphericity was not assumed and the Geisser-Greenhouse correction was applied. If any data points were missing, a mixed-effects model (Restricted maximum likelihood; REML) was used instead. Pearson’s correlation tests were performed as one-tailed tests as, in each case, we had a directional hypothesis of either positive or negative correlation. Sample sizes are shown in graphs with each data point representing an individual mouse, or are reported in the figure legends. Experimental replication and exact statistical tests used are detailed in the figure legends.
AUTHOR CONTRIBUTIONS
C.C. conceived the study and designed experiments. N.J., T.P.L., W.Y., A.B., B.M.R., H.M., K.G., A.A., and C.C. performed experiments and prepared data. N.J., P.Y., and C.C. analyzed and interpreted data. N.J., P.Y., and C.C. wrote the paper. C.C. supervised the study.
SUPPLEMENTARY INFORMATION
ACKNOWLEDGEMENTS
We thank Plexxikon, Inc for providing the PLX3977, PLX5622 and PLX73086 compounds, and Andrey Reymar and Brian West for consulting on drug dosing and chow formulation. We thank Julian Castaneda, Karina Walker and Lyn Batia (and their staff) at the UCSF Hunter’s Point animal facility for coordinating transgenic mouse development, breeding, and drug efficacy studies. We thank Masahiro Inoue (Daiichi Sankyo, Inc.) for insightful discussions on pharmacokinetics and pharmacodynamics in our study. We thank Stanley Prusiner and David Ramsay at the UCSF Institute for Neurodegenerative Diseases (IND) for access to equipment and technical resources critical for the completion of this study. We thank the following UCSF IND staff for technical assistance: Abby Oehler, Rigoberto Roman-Albarran, Julia Becker, Marta Gavidia and Manuel Elepano. The study was funded by grants from the National Institutes of Health (NIH): (# RF1 AG061874 and P01 AG002132), as well as by the Rainwater Charitable Foundation, the Sherman Fairchild Foundation, and Daiichi Sankyo, Inc. Competing interests: The Institute for Neurodegenerative Diseases (UCSF) had a research collaboration with Daiichi Sankyo, Inc. (Tokyo, Japan).
REFERENCES
- 1.↵
- 2.↵
- 3.
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵