Research ArticleAlzheimer’s Disease

Chi3l1/YKL-40 is controlled by the astrocyte circadian clock and regulates neuroinflammation and Alzheimer’s disease pathogenesis

See allHide authors and affiliations

Science Translational Medicine  16 Dec 2020:
Vol. 12, Issue 574, eaax3519
DOI: 10.1126/scitranslmed.aax3519

Targeting astrocytes in AD

Astrocytes and microglia play a dual role in Alzheimer’s disease (AD), increasing neuroinflammation and limiting plaque growth through phagocytic activity. The astrocytic protein YKL-40 is increased in the cerebrospinal fluid (CSF) of patients with AD; however, its role in AD pathophysiology is unclear. Here, Lananna et al. used mouse model and in vitro systems to show that Chi3l1, the gene coding for YKL-40, plays a detrimental role in AD. Its deletion reduced amyloid plaque formation and promoted Aβ phagocytosis. A polymorphism in CHI3L1 resulting in reduced protein expression was associated with slower AD progression in patients. The results suggest that YKL-40 could be targeted for reducing AD progression in patients.

Abstract

Regulation of glial activation and neuroinflammation are critical factors in the pathogenesis of Alzheimer’s disease (AD). YKL-40, a primarily astrocytic protein encoded by the gene Chi3l1, is a widely studied cerebrospinal fluid biomarker that increases with aging and early in AD. However, the function of Chi3l1/YKL-40 in AD is unknown. In a cohort of patients with AD, we observed that a variant in the human CHI3L1 gene, which results in decreased CSF YKL-40 expression, was associated with slower AD progression. At baseline, Chi3l1 deletion in mice had no effect on astrocyte activation while modestly promoting microglial activation. In a mouse APP/PS1 model of AD, Chi3l1 deletion decreased amyloid plaque burden and increased periplaque expression of the microglial lysosomal marker CD68, suggesting that Chi3l1 may suppress glial phagocytic activation and promote amyloid accumulation. Accordingly, Chi3l1 knockdown increased phagocytosis of zymosan particles and of β-amyloid peptide in both astrocytes and microglia in vitro. We further observed that expression of Chi3l1 is regulated by the circadian clock, as deletion of the core clock proteins BMAL1 or CLOCK/NPAS2 strongly suppresses basal Chi3l1 expression, whereas deletion of the negative clock regulators PER1/PER2 increased Chi3l1 expression. Basal Chi3l1 mRNA was nonrhythmic because of a long mRNA half-life in astrocytes. However, inflammatory induction of Chi3l1 was gated by the clock. Our findings reveal Chi3l1/YKL-40 as a modulator of glial phagocytic activation and AD pathogenesis in both mice and humans and suggest that the astrocyte circadian clock regulates inflammatory Chi3l1 induction.

INTRODUCTION

Neuroinflammation plays a critical role in the pathogenesis of most neurodegenerative diseases, including Alzheimer’s disease (AD) (1, 2). In AD, triggering of the brain’s innate immune response, characterized, in part, by activation of astrocytes and microglia, can exert both degenerative and protective effects in a context-dependent manner (1, 3). Although astrocytes and microglia surround amyloid plaques and limit plaque growth through β-amyloid (Aβ) phagocytosis, degradation, and clearance, they can also exacerbate pathology via dysfunctional inflammatory responses (1, 2). Thus, a deeper understanding of factors affecting glial function and neuroinflammation in neurodegenerative conditions is needed.

YKL-40, a secreted glycoprotein encoded by the Chi3l1 gene, is a well-described human cerebrospinal fluid (CSF) biomarker of neuroinflammation, which is elevated in AD (46), as well as other neurologic diseases including multiple sclerosis, amyotrophic lateral sclerosis, and frontotemporal dementia (79). CSF YKL-40 expression steadily increases with age starting in middle age, even in amyloid-negative individuals (5). Chi3l1/YKL-40 is expressed primarily in astrocytes in the brain, in macrophages in the periphery, and is induced in the setting of inflammation (4, 10, 11). Animal studies suggest that it may mitigate inflammation and protect cells from oxidative stress in the periphery (1214). In AD, numerous studies show that CSF Chi3l1/YKL-40 increases in parallel with tau and other markers of inflammation and neurodegeneration and that elevated Chi3l1/YKL-40 predicts disease progression (4, 5). Although Chi3l1/YKL-40 has been studied extensively as a biomarker, little is known about its function in the brain and in AD.

Many patients with neurodegenerative diseases also exhibit circadian system dysfunction, including those with preclinical AD (15). This dysfunction has been implicated as a potential contributor to AD pathogenesis (16, 17). The circadian system orchestrates 24-hour rhythms in many physiological and behavioral parameters, including sleep, activity, and endocrine function (18). The core circadian clock consists of a transcription-translation feedback loop, driven by the master circadian transcription factor BMAL1 (Brain and Muscle ARNT-Like 1), which regulates circadian rhythms in transcription in a cell type–specific manner. The circadian clock regulates immune responses in a variety of peripheral innate and adaptive immune cell types (19). Astrocytes and microglia also have functioning circadian clocks, and circadian timing can affect their inflammatory responses (2022). Our group has shown that disruption of core circadian clock transcription via deletion of Bmal1 can promote neuroinflammation and that BMAL1 regulates astrocyte activation in a cell-autonomous manner (23, 24). However, the mechanisms linking glial clocks to neuroinflammation are poorly understood.

Here, we report data from humans and mouse models suggesting that Chi3l1/YKL-40 accelerates AD pathogenesis, potentially by altering glial function in the brain. We also show evidence from mice that the core circadian clock in astrocytes strongly regulates Chi3l1 transcription and gates its inflammatory induction. Together, our results reveal Chi3l1/YKL-40 as a modulator of AD pathogenesis and illuminate a link between glial circadian clocks and Chi3l1 expression.

RESULTS

A human genetic variant causing decreased CHI3L1/YKL-40 is associated with slower disease progression in human patients with AD

We sought to determine whether CHI3L1/YKL-40 influences AD pathogenesis in humans. First, we examined a human single-nucleus RNA sequencing (RNAseq) dataset derived from cortical brain tissue from three patients with AD (25) and observed that CHI3L1 expression was strongly associated with the astrocyte cell cluster (Fig. 1A). Although a few CHI3L1-expressing microglia were present, differential gene expression analysis shows that CHI3L1 is enriched in astrocytes and minimally expressed in neurons (Fig. 1B). We have previously reported that CSF YKL-40 expression in humans is strongly controlled by genetic variation in the CHI3L1 locus and have described a common genetic variant in the CHI3L1 locus (rs10399931) associated with decreased CSF YKL-40 concentrations (26). Using the Knight Alzheimer’s Disease Research Center at Washington University clinical database, we next examined whether this rs10399931 single-nucleotide polymorphism (SNP) was associated with any changes in AD progression. We examined data from 778 participants enrolled in longitudinal observational studies and included only participants with AD (confirmed by clinical assessment and CSF biomarker profile), 26% of whom carried the CC_TT polymorphism at rs10399931. We examined clinical progression of AD as assessed by the rate of increase in the Clinical Dementia Rating Sum-of-Boxes score (CDR-SB). We found that the rs10399931 SNP was significantly associated (P = 0.031) with a slower rate of AD progression, suggesting that people with genetically lower CHI3L1/YKL-40 expression have 16% slower disease progression (Fig. 1C). For comparison, pathogenic variants in the Trem2 gene were previously associated with a 23% increased rate of progression using the same methods (27). This initial human data suggested that CHI3L1/YKL-40 might be a modulator of human disease pathogenesis. Thus, we sought to examine the mechanisms by which CHI3L1/YKL-40 might exacerbate AD pathogenesis, including glial activation, in mouse models.

Fig. 1 A polymorphism in human CHI3L1 affects rate of AD progression.

(A) Single-nucleus RNAseq t-distributed stochastic neighbor embedding (tSNE) plot showing cell clusters (top) or CHI3L1 expression (purple, bottom). Ex, excitatory neuron; In, inhibitory neuron; OPC, oligodendrocyte precursor cells; Endo, endothelial cells. Data are available at http://ngi.pub/snRNAseq/. (B) CHI3L1 expression by cell cluster from data in (A). ****P = 3 × 10−280. (C) AD progression (change in CDR-SB) from human patients with and without the CC_TT polymorphism of the rs10399931 SNP in CHI3L1.

Chi3l1 modulates the astrocytic and microglial inflammatory responses in vivo

Our data (Fig. 1A) and other studies show that Chi3l1 is highly enriched in human astrocytes (28) and only negligibly expressed in other cell types, whereas in the mouse central nervous system, at baseline, it is highly expressed in astrocytes and oligodendrocyte precursor cells but less so in microglia (29). Thus, we sought to investigate the role of astrocytic Chi3l1 in regulating neuroinflammation. We used small interfering RNA (siRNA) to knock down Chi3l1 (siChi3l1) in mouse primary astrocyte cultures, achieving a 91% decrease in Chi3l1 mRNA (Fig. 2A). Loss of Chi3l1 induced a small decrease in Gfap and a small increase in Il6. This suggests that Chi3l1 suppression may exert differential effects on astrocyte activation and cytokine expression. In vivo, constitutive Chi3l1 knockout (KO) (30) had no effects on expression of several inflammatory transcripts (Fig. 2B and fig. S1C). As a group, transcripts related to astrocyte activation were modestly altered at baseline by Chi3l1 KO, although no individual transcripts were changed (Fig. 2C and fig. S2A). Similarly, microglial activation genes (including Trem2, Spp1, Gpnmb1, and Apoe) were slightly increased at baseline as a group in Chi3l1 KO mice, although no individual genes were different (Fig. 2D and fig. S2B). Chi3l1 did not alter baseline glial fibrillary acidic protein (GFAP) immunoreactivity, a marker of astrocyte activation, in the hippocampus (Fig. 2, E and F). However, Chi3l1 deletion subtly increased staining for the microglial marker IBA1 (ionized calcium-binding adapter molecule 1) in the cortex (Fig. 2, G and H), suggesting that Chi3l1 differentially modulates astrocyte and microglial activation at baseline.

Fig. 2 Loss of Chi3l1 mildly shifts glial activation.

(A) qPCR gene expression from primary astrocytes transfected with control (siScr) or Chi3l1 (siChi3l1) siRNA. n = 6 to 10 replicates from three independent experiments. (B to D) Cytokine and chemokine (B), astrocyte activation marker (C), or microglia activation marker (D) expression from Fluidigm qPCR of 2- to 5-month-old Chi3l1−/− and WT control mouse cortex 6 hours after intraperitoneal PBS. Mean of six mice per group normalized to WT. Two-way ANOVA with Tukey correction for multiple comparisons. ns, not significant. (E to H) Representative images depicting GFAP (astrocyte) staining (E) and associated quantification (F) or IBA1 (microglia) staining (G) and associated quantification (H) in Chi3l1−/− and WT control mice. Scale bars, 400 μm. All data represent means ± SEM. *P < 0.05, **P < 0.01, and ****P < 0.0001 by two-tailed Student’s t test with Holm-Sidak correction for multiple comparisons when appropriate.

We next examined the effect of Chi3l1 on acute neuroinflammation induced by the inflamogen lipopolysaccharide (LPS). In primary astrocyte cultures, Chi3l1 siRNA exacerbated LPS-induced cytokine expression (fig. S1A). Chi3l1 KO mice exhibited a general exacerbation of the inflammatory response with increased hippocampal expression of several inflammatory transcripts, including several microglia-specific transcripts such as Cybb and Nlrp3 after intraperitoneal LPS injection (fig. S1, B to E). However, Chi3l1 KO did not alter LPS-induced expression of astrocyte activation markers or AD-associated microglial activation markers (fig. S2, A and B). Together, our data suggest that Chi3l1/YKL-40 deletion did not affect astrocyte activation but mildly enhances microglial activation at baseline and modestly potentiates LPS-induced inflammatory cytokine expression in astrocytes and microglia.

Deletion of Chi3l1 reduces amyloid plaque deposition in a mouse model of AD

As YKL-40 is increased in the CSF of patients with AD and used as a biomarker of AD (4, 6), we next sought to test the hypothesis that Chi3l1/YKL-40 could also influence pathology in a mouse model of AD-related β-amyloidosis. We crossed amyloid precursor protein (APP)/PS1-21 mice, which express the human APP gene with the KM670/671NL (Swedish) mutation and human presenilin 1 (PS1) with the L166P mutation (31), with wild-type (WT) or Chi3l1 KO mice. All mouse brain tissues were harvested at 8 months of age when mice had developed substantial plaque pathology in both the cortex and hippocampus (31). Staining with X34, which selectively labels β-pleated sheet fibrillar Aβ plaques (32), revealed that loss of Chi3l1 reduced fibrillar plaque number by 21% (Fig. 3, A and C) and plaque area by 17% (Fig. 3, A and D) in the hippocampus but did not alter these measures in the cortex (fig. S3, A to C). Staining with an Aβ antibody (HJ3.4) revealed a much more pronounced 55% reduction in plaque burden in the hippocampus (Fig. 3, A and E) and a 42% decrease in the cortex (fig. S3, A and D). This discrepancy between staining methodologies led us to hypothesize that Chi3l1 deletion results in the selective reduction of nonfibrillar Aβ. Subtraction of X34 signal from total plaque (HJ3.4) staining indeed revealed a halo of aggregated, nonfibrillar Aβ surrounding the fibrillar plaque core in a majority of plaques in control APP/PS1 mice, as well as some X34-negative, HJ3.4+ plaques (Fig. 3, A and B, and fig. S3A). We observed an 85% loss of this nonfibrillar Aβ in the hippocampus of Chi3l1 KO mice (Fig. 3F), along with a 54% reduction in the cortex (fig. S3E). To more closely investigate this selective loss of nonfibrillar Aβ, regions with similar amounts of fibrillar (X34+) plaque load (fig. S3, F and G) were imaged using confocal microscopy and three-dimensional (3D) reconstruction in a subset of mice. Plaque volume measurements again revealed a loss in antibody (HJ3.4)–positive Aβ (fig. S3H) and confirmed the selective loss of nonfibrillar Aβ in the absence of Chi3l1 (Fig. 3, G and H). Moreover, loss of Chi3l1 altered the distribution of antibody (HJ3.4)–stained plaques in APP/PS1 mice (fig. S4, A and B) with a substantial decrease in plaque number and a small decrease in average plaque area (fig. S4, C and D). This pool of aggregated but nonfibrillar Aβ could not be separated biochemically via sequential fractionation with increasing concentrations of guanidine (fig. S4E). Loss of Chi3l1 did not result in changes in the amount of APP protein (fig. S4F) or in the expression of several key amyloid processing/metabolic genes, including App, Bace1, Ide, Mmp2 or Mmp9, Ldlr, Lrp1, Mme, Klk7, and Apoe (fig. S4G). These data, in combination with the fact that Chi3l1 is generally not expressed in neurons, indicate that YKL-40 likely does not alter the production of Aβ. Together, these data suggest that the loss of Chi3l1 mitigates the accumulation of Aβ plaque pathology, particularly nonfibrillar plaque material.

Fig. 3 Loss of Chi3l1 mitigates amyloid pathology.

(A) Representative hippocampal images from 8-month-old Chi3l1−/−:APP/PS1+ and APP/PS1+ control mice depicting staining by X34 (fibrillar plaques), HJ3.4 antibody (total Aβ), and subtraction of fibrillar (X34) from total (HJ3.4) Aβ. The hippocampus is outlined in yellow. The yellow rectangle denotes region of inset in (B). Scale bar, 300 μm. (B) Representative higher magnification images depicting stains from (A). “Halo” of nonfibrillar Aβ surrounding fibrillar plaque core (X34) substantially reduced in Chi3l1−/− mice. Scale bar, 50 μm. (C to F) Quantification of X34+ puncta (fibrillar plaque number) (C), X34+ area (fibrillar Aβ) (D), HJ3.4+ area (total Aβ) (E), or area covered by nonfibrillar Aβ (total Aβ in HJ3.4-fibrillar X34) (F) in hippocampal staining from (A). (G) X34 and HJ3.4 colocalization (top) with 3D surface rendering of X34 (red) and HJ3.4 (green) staining (middle) with 20-μm shells (pink) around each fibrillar plaque. Scale bar, 30 μm. (H) Quantification of nonfibrillar plaque volume as ratio of HJ3.4 to X34 (top) or total Aβ in HJ3.4-fibrillar X34 (bottom) in 20-μm shell surrounding each X34+ plaque in a subset of mice from (C) to (F). Data points represent average of two to four sections per mouse and 4 to 6 (Chi3l1+/+) and 4 to 12 (Chi3l1−/−:) mice per group. All data represent means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 by two-tailed Student’s t test.

Periplaque astrocyte activation is suppressed by Chi3l1 deletion

As astrocytes and microglia have been shown to play integral roles in both the clearance of and neuroinflammation induced by Aβ (1, 2), we next examined the consequences of Chi3l1 deletion on the glial response to Aβ plaques. Whereas previous reports disagree about whether impairing astrocyte activation increases or decreases Aβ plaque burden (33, 34), we found that Chi3l1 deletion resulted in decreased astrocyte activation as measured by GFAP staining in the hippocampus and throughout the cortex in APP/PS1+ mice (Fig. 4, A and B, and fig. S5A). Modest periplaque GFAP reductions were also observed (Fig. 4, A and C), as controlling for the decreased fibrillar plaque area (Fig. 4C) or number (fig. S5B) could not fully account for the lessened GFAP, particularly in the hippocampus. Thus, the amount of GFAP staining per plaque area is reduced with Chi3l1 deletion. In concordance with our staining data, transcriptional profiling of astrocyte activation markers revealed a slight attenuation of hippocampal astrocyte activation in Chi3l1 KO;APP/PS1 mice (Fig. 4D and fig. S5C). Chi3l1 mRNA was increased in APP/PS1 cortex at 8 months and was absent in Chi3l1 KO mice (fig. S5C).

Fig. 4 Loss of Chi3l1 mitigates astrogliosis but facilitates phagocytosis in the presence of Aβ.

(A) Representative high-magnification images from hippocampi of 8-month-old Chi3l1−/−:APP/PS1+ and APP/PS1+ control mice stained for X34 (fibrillar plaques) and GFAP (astrocytes). Scale bars, 20 μm. (B and C) Quantification of GFAP coverage (B) or GFAP coverage normalized to X34+ area in the same section (C) from mice in (A). Quantified from wide-field image in fig. S5. RS, retrosplenial. n = 6 (Chi3l1+/+) and 12 (Chi3l1−/−:) mice per group. (D) Astrocyte activation marker gene expression from Fluidigm qPCR of 8-month-old Chi3l1−/−:APP/PS1, WT:APP/PS1, Chi3l1−/−:APP/PS1+, and APP/PS1+ control mouse hippocampus. Mean of 4 mice (APP/PS1) or 6 to 10 (APP/PS1+) mice per group normalized to WT:APP/PS1. Two-way ANOVA with Tukey correction for multiple comparisons. (E and F) pHrodo-labeled zymosan bead (E) or TAMRA-Aβ (F) uptake by primary astrocyte cultures transfected with control (siScr) or Chi3l1 (siChi3l1) siRNA, ±cytochalasin D to inhibit phagocytosis (+cytoD). Each point represents one field of view with an average of 804 (E) or 517 (F) cells per field. Data are from two independent experiments. All data represent means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, analyzed by two-tailed Student’s t test (B and C) or one-way ANOVA (E and F). All data were subjected to Sidak correction for multiple comparisons unless otherwise noted.

As astrocyte activation state can affect phagocytosis (3) and astrocytes are known to phagocytose Aβ plaque material (35), we next sought to evaluate whether Chi3l1 may be regulating phagocytosis in astrocytes. siRNA-mediated Chi3l1 knockdown, which suppresses Chi3l1 mRNA expression in primary astrocyte cultures by ~91% (see Fig. 2A), increased the phagocytosis of zymosan-coated pHrodo-labeled beads by 13% (Fig. 4E) and fluorescent [TAMRA (carboxytetramethyl rhodamine)–labeled] Aβ42 peptide by 50% (Fig. 4F), each in two separate experiments. In combination, these data support the idea that loss of Chi3l1 tempers astrocyte activation while potentially increasing astrocytic phagocytosis in response to Aβ plaques.

Chi3l1 deletion promotes plaque-related microglial CD68 expression

To address the possibility that Chi3l1 could also be regulating the microglial response to Aβ, we used IBA1 to label microglia and the microglial lysosomal marker CD68 to assess microglial phagosome expression (Fig. 5A and fig. S6A). When normalized to plaque area, we did not observe any changes in absolute coverage of IBA1 in the hippocampus, motor cortex, or retrosplenial cortex (fig. S6, A and B). However, the amount of CD68 staining normalized to X34+ plaque area was elevated across all regions examined in Chi3l1 KO;APP/PS1 mice (Fig. 5B), indicating increased microglial phagocytic activation relative to plaque burden. This effect appeared to be driven by increased CD68 staining around amyloid plaques, as colocalized IBA1/CD68 area per X34+ plaque area was also increased in Chi3l1 KO mice (Fig. 5, A and C). These changes seem to be CD68 specific because there were no differences observed in transcript expression of other lysosomal markers (Fig. 5D and fig. S6C). Transcriptional analysis of hippocampal tissue for a selection of known microglia activation markers revealed an increase in the activation marker Spp1 in 8-month-old Chi3l1 KO mice without plaques. We also observed a slight overall dampening of microglial activation marker expression (Fig. 5D and fig. S6C) and inflammatory markers (fig. S7, A and B) with Chi3l1 deletion in the presence of Aβ pathology. These changes are very likely due to reduced plaque burden in Chi3l1 KO;APP/PS1 mice.

Fig. 5 Loss of Chi3l1 alters microglial activation and enhances Aβ phagocytosis.

(A) Representative high-magnification images from hippocampi of 8-month-old Chi3l1−/−:APP/PS1+ and APP/PS1+ control mice stained for X34 (fibrillar plaques), IBA1 (microglia), and CD68 (phagocytic microglia). Scale bars, 20 μm. (B and C) Quantification of CD68 (B) or colocalized IBA1/CD68 (C) from mice in (A) normalized to X34+ area in the same section. Quantified from wide-field image in fig. S6. n = 6 (Chi3l1+/+) and 12 (Chi3l1−/−) mice per group. RS, retrosplenial. (D) Microglia-associated gene expression from Fluidigm qPCR of 8-month-old Chi3l1−/−:APP/PS1, WT:APP/PS1, Chi3l1−/−:APP/PS1+, and APP/PS1+ control mouse hippocampus. Mean of 4 mice (APP/PS1) or 6 to 10 (APP/PS1+) mice per group normalized to WT:APP/PS1. Two-way ANOVA with Tukey correction for multiple comparisons. (E and F) pHrodo-labeled zymosan bead (E) or TAMRA-Aβ (F) uptake by primary microglia cultures transfected with control (siScr) or Chi3l1 (siChi3l1) siRNA, ±cytochalasin D to inhibit phagocytosis (+cytoD). Each point represents one field of view with an average of 209 (E) or 56 (F) cells per field. Data are from two (E) or one (F) independent experiments. All data represent means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 analyzed by two-tailed Student’s t test (B and C) or one-way ANOVA (E and F). All data were subjected to Sidak correction for multiple comparisons unless otherwise noted.

Because of the increase in microglial CD68 expression in Chi3l1 KO;APP/PS1 mice and the observed increase in astrocytic phagocytosis with loss of Chi3l1, we next examined phagocytosis in cultured primary mouse microglia. Chi3l1 siRNA (siChi3l1) suppressed Chi3l1 mRNA in microglial cultures by 97% (fig. S7C). Loss of Chi3l1 in microglia had an even greater effect than in astrocytes, with Chi3l1 knockdown increasing phagocytosis of pHrodo-labeled zymosan beads by 148% (Fig. 5E) and fluorescent (TAMRA-labeled) Aβ by 100% (Fig. 5F) in microglia in vitro. Together, our data demonstrate that loss of Chi3l1 leads to decreased amyloid plaque burden and increased microglial CD68 expression in vivo and enhances phagocytosis of Aβ by both astrocytes and microglia in vitro.

Chi3l1 expression is nonrhythmic but controlled by the circadian clock

We next explored possible molecular mechanisms regulating Chi3l1 expression and potential link to glial activation. We previously reported that deletion of the master circadian clock gene Bmal1 caused astrocyte activation in a cell-autonomous manner (24). While examining an existing transcriptional dataset from control and brain-specific Bmal1 KO (Nestin-Cre;Bmal1f/f) cortex to identify circadian clock targets that may regulate astrocyte activation, we noticed Chi3l1 to be among the most down-regulated transcripts in Bmal1 KO mice. Although Chi3l1 is reported to be among the 50 most up-regulated transcripts in activated astrocytes after in vivo inflammation (LPS injection) (36), our transcriptomic data showed that Chi3l1 was markedly down-regulated in Bmal1 KO brain (−89%) and strongly up-regulated in Per1mut;Per2mut (+230%) mice across circadian time points (Fig. 6, A and B). This unexpected finding of reciprocal expression changes in mice with mutation of the positive limb (Bmal1) and negative limb (Per1/2) of the circadian clock suggested that Chi3l1 might be a clock-controlled gene. However, basal Chi3l1 mRNA did not show circadian oscillation in control mice in this array data (Fig. 6B). Follow-up quantitative polymerase chain reaction (qPCR) analysis of cortex tissue collected every 4 hours in constant darkness confirmed a lack of circadian oscillation in Chi3l1 mRNA in control mice but again showed an average of a 91% loss in Chi3l1 transcript in Bmal1 KO brain (fig. S8A). Moreover, we found that, similar to Bmal1 KO tissue, Chi3l1 mRNA was decreased in cortex from Clock/Npas2 double-KO mice, as compared to Clock KO alone (Fig. 6C). Clock/Npas2 double-KO mice lack a binding partner for Bmal1 and thus have a dysfunctional positive limb of the clock. This decrease in Chi3l1 expression mirrored that of the BMAL1-CLOCK/NPAS2 target Nr1d1 (fig. S8B) and occurred despite a compensatory increase in Bmal1 expression (fig. S8C). These data strongly support the requirement of a functioning positive limb of the clock (consisting of BMAL1/CLOCK or BMAL1/NPAS2 heterodimers) for Chi3l1 expression.

Fig. 6 Chi3l1 is regulated by the circadian clock.

(A) Microarray in Nestin-Cre;Bmal1f/f and Per1/2mut versus control, Bmal1f/f cortex [Lananna et al. (24)] cross-referenced with 50 genes most up-regulated in astrocytes with in vivo LPS [Zamanian et al. (36)]. CT, clock time. (B) Microarray data from (A) with two additional time points for Cre and Per1/2mut mice. (C) qPCR depicting gene expression in global Npas2 KO, Clock KO, Clock/Npas2 double KO, or Bmal1 KO mouse cortex. n = 2 mice per group, normalized to WT control. (D to F) qPCR showing Chi3l1 expression from Aldh1l1-Cre;Bmal1f/f (ALC) hippocampus (D) or Cx3cr1-Cre;Bmal1f/f (CX3) cortex (E) or gene expression from WT primary astrocytes 4 to 8 days after transfection with control (siSCR) or Bmal1 (siBmal1) siRNA (F). n = 3 mice per group (D), 6 to 8 mice per group (E), or 15 biological replicates from five independent experiments (F). (G) Enzyme-linked immunosorbent assay of Chi3l1 protein (YKL-40) in culture medium of WT primary astrocytes 6 days after transfection with control (siSCR) or Bmal1 (siBmal1) siRNA. n = 3 biological replicates per group. (H) ChIP-qPCR from mouse primary astrocytes with anti-BMAL1 antibody or IgG control. Three distinct E-box–containing regions in the Chi3l1 promoter were assayed. A known BMAL1 binding E-box–containing region in the Dbp promoter was assayed as positive control. (I) qPCR showing expression of Chi3l1 and Bmal1 mRNA in primary mouse astrocytes treated with actinomycin D at 0 hours and then harvested at intervals thereafter. n = 3 wells per time point, normalized to Actb mRNA. *P < 0.05 compared to time point 0. All data represent means ± SEM. *P < 0.05, **P < 0.01, and ****P < 0.0001 by two-tailed Student’s t test with Holm-Sidak correction for multiple comparisons when appropriate.

To test whether the circadian clock is regulating Chi3l1 in astrocytes, we measured Chi3l1 in inducible astrocyte-specific Bmal1 KO (Aldh1l1-CreERT2:Bmal1f/f) mice, which we have previously shown results in the loss of about 70% of astrocytic BMAL1 (24). Chi3l1 expression was reduced by 52% in the cortex of Cre+ animals (Fig. 6D), whereas we did not observe any change in Chi3l1 expression in microglia-specific Bmal1 KO (Cx3cr1-CreERT2:Bmalf/f) mice (Fig. 6E). Treating primary mouse astrocyte cultures with siRNA targeting Bmal1 resulted in a 64% loss in the direct BMAL1 transcriptional target Nr1d1 and a 71% loss in Chi3l1 (Fig. 6F), whereas the secretion of CHI3L1 protein (YKL-40) decreased by 72% (Fig. 6G). Knockdown of Bmal1 in primary microglia cultures resulted in a 79% loss in Nr1d1 but no change in Chi3l1 (fig. S8D). To determine whether BMAL1 directly regulates Chi3l1, we examined the putative Chi3l1 promoter (within 700 base pairs of the Chi3l1 transcriptional start site) and identified six E-boxes, three of which previously displayed weak binding of BMAL1 in an existing liver chromatin immunoprecipitation sequencing (ChIP-seq) dataset (fig. S8E) (37). ChIP-qPCR for BMAL1 binding to three of these E-boxes (fig. S8E) revealed enrichment when compared to immunoglobulin G (IgG) controls at all three sites, as well as at a known BMAL1 binding site in the Dbp promoter (Fig. 6H). Although Chi3l1 transcription is directly regulated by BMAL1, it is not rhythmic at baseline (Fig. 6B and fig. S8, A and F). One possibility for the lack of Chi3l1 oscillation in astrocytes is that its mRNA may have a long half-life. We measured mRNA degradation kinetics in primary astrocyte cultures after inhibition of transcription with actinomycin D and observed that the half-life of Chi3l1 transcript was much longer than that of Bmal1, a rhythmic gene (Fig. 6I). Thus, although the positive limb of the circadian clock directly regulates Chi3l1 expression, basal Chi3l1 mRNA is not rhythmic likely because of its long half-life.

Induction of Chi3l1 in astrocytes is gated by the circadian clock

As Chi3l1/YKL-40 has been shown to increase during inflammatory conditions (10, 38) and is regulated by nuclear factor κB (11), we next sought to evaluate the regulation of Chi3l1 induction by BMAL1 in the setting of inflammation. In cultured astrocytes, Chi3l1 was induced by LPS stimulation, but both basal and LPS-stimulated Chi3l1 expression was suppressed in Bmal1 KO cells (although some LPS-induced increase in Chi3l1 was still observed) (Fig. 7A). To more closely examine the possibility that the inflammatory induction of Chi3l1 is gated by the astrocyte circadian clock, we measured basal and LPS-induced Chi3l1 expression in synchronized primary astrocytes across circadian time points. Consistent with our in vivo data, Chi3l1 transcript did not oscillate in synchronized primary astrocytes (fig. S8F). However, LPS-mediated induction of Chi3l1 was highly dependent on circadian phase, as LPS-induced Chi3l1 expression varied antiphase to Bmal1 mRNA expression and closely mirrored the expression pattern of Nr1d1, which is dependent on BMAL1 transcriptional activity (Fig. 7, B and C). This time-of-day variation in Chi3l1 induction was not observed in cells treated with Bmal1 siRNA, indicating a requirement for a functioning circadian clock (Fig. 7D). These data suggest that circadian oscillations in BMAL1 transcriptional activity gate the inflammatory induction of Chi3l1 in astrocytes (Fig. 7E). The induction of Chi3l1 in astrocytes was not limited to LPS as exposure of primary astrocyte cultures to Aβ42 fibrils resulted in an about 4.5-fold increase in Chi3l1 (Fig. 7F).

Fig. 7 Chi3l1 is induced during inflammation in a Bmal1-dependent manner.

(A) qPCR showing gene expression from WT and Bmal1 KO primary astrocytes ± LPS (500 ng/ml) for 6 hours. n = 5 independent experiments. (B and C) qPCR showing LPS-induced Chi3l1 gene expression (B) or expression of Nr1d1 (C) in WT primary astrocytes synchronized with high serum shock. Cells were treated with LPS (500 ng/ml) (B) or PBS (500 ng/ml) (C) at designated time points and collected 3 hours after treatment. n = 6 to 9 replicates from two to three independent experiments per time point. Data were normalized to basal Chi3l1 expression in PBS control cells (depicted by dashed line) (B) or expression at 18 hours (C). Main effect P = 0.0150 (B) and P = 0.0013 (C). Multiple comparison tests are depicted on graphs. (D) qPCR showing LPS-induced Chi3l1 expression in WT primary astrocytes after transfection with control (siSCR) or Bmal1 (siBmal1) siRNA, treated with LPS (500 ng/ml) at 24 or 36 hours after synchronization [as in (B)]. n = 3 replicates per genotype per time point. (E) Diagram depicting hypothesis that induction of Chi3l1 expression (blue) is dependent on the circadian phase of BMAL1 transcriptional activity (green curve). (F) qPCR showing gene expression from WT primary astrocytes ± 10 μM Aβ fibrils for 48 to 72 hours. n = 6 replicates from two independent experiments. All data represent means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001. Analyzed by two-way ANOVA (A and D) or one-way ANOVA (B and C) with Tukey correction for multiple comparisons or two-tailed Student’s t test (F).

DISCUSSION

Because glia can exert either protective or degenerative influences on the brain, striking a delicate balance of glial activation and inflammatory signaling is critical for maintaining brain health. Thus, factors that alter glial activation state may disrupt this balance and promote neurodegeneration. Although Chi3l1/YKL-40 is an AD biomarker, its role in the progression of AD remains unknown. Here, we show that in mice and cells, Chi3l1 deletion alters glial inflammatory responses, promotes astrocyte and microglial Aβ phagocytosis, and mitigates amyloid plaque formation. Furthermore, we show that a genetic polymorphism associated with lower CSF CHI3L1/YKL-40 concentrations in humans is associated with slower AD progression. These data suggest that increases in Chi3l1/YKL-40 that occur during aging and AD (46) may have a detrimental impact on AD pathogenesis by altering glial function and plaque deposition.

Glial activation is thought to be a double-edged sword in AD, as activated glia can phagocytose Aβ and tau and prevent proteopathy, whereas excessive inflammatory activation can accelerate plaque accumulation and synapse loss (2). Several studies find Chi3l1 to be anti-inflammatory and/or neuroprotective in the setting of bacterial infection (39), traumatic brain injury (14), and experimental autoimmune encephalomyelitis (13). However, pharmacologic inhibition of Chi3l1 has been reported to suppress Aβ deposition in a rat Aβ infusion model, albeit through a purported anti-inflammatory mechanism (40). We observed that Chi3l1 KO enhances inflammation in response to LPS but not amyloid plaques. Moreover, the transcriptional signature of microglial activation in response to LPS versus amyloid plaques is very different, with opposite regulation of key mediators such as Trem2 (1). Thus, the effect of Chi3l1/YKL-40 on the balance of glial activation and neuroinflammation appears to be context dependent such that loss of Chi3l1 could be neuroprotective in AD but destructive in settings of acute inflammation.

In APP/PS1 mice, we observed that Chi3l1 deletion causes decreased periplaque astrocyte clustering. Attenuation of astrocyte activation in APP/PS1 mice has previously yielded mixed results, as Gfap:Vimentin double KO increases (33), whereas Stat3 KO decreases (34) plaque burden. Our finding that Chi3l1 knockdown enhances phagocytosis of zymosan beads and Aβ by cultured astrocytes suggests that Chi3l1/YKL-40 is a general regulator of astrocyte phagocytosis and reveals a potential mechanism for the decrease in amyloid plaques in vivo. Attenuated astrocyte activation in APP/PS1 models has previously been associated with increased microglial plaque clustering (33, 34) and, in one case, with increased microglial phagocytosis of Aβ (34), which is consistent with our data. Decreased astrocyte activation in Chi3l1 KO mice (and in these other models) could relent a physical barrier that previously limited microglial access to the plaque. Alternatively, astrocyte-secreted YKL-40 may signal to restrain microglial phagocytic activation.

The exacerbation of LPS-induced microglial transcriptional changes with Chi3l1 deletion matches previous reports showing that Chi3l1 can cell-autonomously suppress the macrophage inflammatory response in the periphery (39). Together, these data suggest that Chi3l1 could also be important in microglia. Recently, a single-nucleus RNAseq study identified Chi3l1 as a gene that is strongly up-regulated in microglia in the brains of human patients with AD (41). Whereas our single-nucleus RNAseq data show clear expression of Chi3l1 in astrocytes, a small number of microglia do appear to express Chi3l1, and this could increase in the setting of disease. Thus, a cell-autonomous effect of Chi3l1 in microglia is highly possible. Our observation that Chi3l1 knockdown in cultured microglia increases phagocytosis of beads and Aβ suggests that the increase in periplaque CD68 expression in Chi3l1 KO;APP/PS1 mice may be due to a cell-autonomous effect in microglia. In this case, increased Chi3l1 expression in microglia in AD [as reported by Zhou et al. (41)] would be expected to suppress microglial, and potentially astrocytic, Aβ phagocytosis and accelerate plaque growth, in keeping with our data.

In patients with AD, we observed that a common variant in the CHI3L1 gene that causes decreased CSF YKL-40 concentrations is associated with slower AD progression. This observation was made in a cohort of extremely well-characterized individuals confirmed to have AD by both clinical evaluation and biomarkers (based on CSF Aβ/tau or amyloid positron emission tomography imaging). Because CSF YKL-40 increases in response to aging, inflammation, and neurodegeneration, it is difficult to determine how YKL-40 itself might be affecting these processes simply by measuring it in CSF. By examining disease progression in carriers of this genetic variant, which lowers CHI3L1/YKL-40 expression throughout life, we can assume that changes in progression are likely caused by reduced CHI3L1/YKL-40 signaling, providing a unique opportunity to assess a possible causal relationship between Chi3l1/YKL-40 expression and AD. Our results suggest that inhibition of Chi3l1/YKL-40 may be a potential future therapeutic target for limiting plaque accumulation, optimizing the glial phagocytic response to plaques, and slowing progression of AD.

Research into the relationship between circadian disruption and neurodegenerative disease has begun to uncover a role for the clock in regulating astrogliosis (24), microgliosis (22), and plaque deposition (17). Our findings provide an example of a clock-controlled gene (Chi3l1) that does not oscillate at the mRNA level, likely because of a long mRNA half-life. However, because Chi3l1 induction by LPS is suppressed with Bmal1 deletion and is greatest at times when BMAL1 transcriptional activity is highest, it appears that BMAL1 transcriptional activity gates the induction of Chi3l1 in astrocytes, revealing a role for circadian timing in Chi3l1 regulation.

As Chi3l1 is strongly suppressed in Bmal1 KO astrocytes, it is important to reconcile the reduction in Aβ plaque burden that we have observed in Chi3l1 KO mice with our previous data showing that global Bmal1 deletion increased fibrillar plaque burden (17). Global Bmal1 deletion affects every cell type in the brain while also disrupting peripheral clocks, sleep-wake cycles, and whole-animal rhythmicity. The resultant phenotype in global Bmal1 KO mice is thus a summation of many smaller and possibly divergent effects. It is likely that in global Bmal1 KO mice, the potentially beneficial effect of decreased Chi3l1 on plaques is overwhelmed by effects on sleep and other processes, resulting in a net increase in plaques. Astrocyte-specific effects of Bmal1 deletion on amyloid plaque deposition remain to be explored. Elucidating the intricacies of competing cell-specific pathways regulated by the clock in the brain is vital in understanding how circadian dysfunction may affect the course of neurodegeneration (42).

There are several limitations to this study. Our experiments do not differentiate the relative effects of Chi3l1 in astrocytes versus microglia in vivo. Such disentanglement would require cell type–specific Chi3l1 manipulation. The APP/PS1 mouse used in this study models amyloid plaque formation due to rare familial mutations and does not recapitulate all aspects of human AD, especially tau aggregation. Future studies in alternate models, such as tau transgenic mice, will be needed. The effects of Chi3l1 on phagocytosis of various Aβ and tau species, such as oligomers, were not investigated. Last, our clinical data do not establish how CHI3L1 polymorphisms affect different aspects of AD pathology or glial activation in humans. Human pathological and biomarker data will need to be integrated with CHI3L1 genotyping in future studies.

In summary, we have provided evidence that Chi3l1 regulates glial activation, Aβ phagocytosis, and amyloid plaque deposition in mice and influences AD progression in humans. These findings identify Chi3l1/YKL-40 as a potential therapeutic target for slowing disease progression in AD and provide insights into regulation of neuroinflammation by the astrocyte circadian clock.

MATERIALS AND METHODS

Study design

The goal of this study was to elucidate the role of the well-known biomarker of AD, Chi3l1/YKL-40, in neuroinflammation and AD pathogenesis. We used data from a large observation study of AD to determine whether a known genetic variant in the CHI3L1 gene in humans, which was associated with lower CSF YKL-40 levels, might influence the rate of AD progression. This analysis method was based on previous work by members of our group in identifying the variant in CHI3L1 that affects YKL-40 levels and in developing a method to accurately detect single-gene influences on clinical AD progression (26). Mouse studies were then carried out using constitutive Chi3l1−/− mice, which were crossed to an APP/PS1 model of β-amyloidosis for some experiments. Last, the regulation of Chi3l1 expression by the circadian clock was investigated using a variety of tissue-specific Bmal1 KO mice, as well as other circadian clock gene mutant mice. Cell culture experiments using primary mouse glial cultures and siRNA were also used for mechanistic studies. Sample size for APP/PS1-21 experiments was determined at the outset and based on power calculations derived from previous analysis of amyloid plaque pathology in this line from our laboratory. Mice were not randomized but were grouped on the basis of genotype with groups containing roughly equal numbers of male and female animals. Mouse tissue samples were all processed together and analyzed to prevent batch effects. Investigators were blinded to genotype throughout the data analysis process. All mice were housed in the same facility, and cohorts of mice were bred at the same time such that they aged together. The number of biological replicates is indicated in the figure legends.

Mice

All mouse experiments were conducted in accordance with protocols approved by the Washington University Institutional Animal Care and Use Committee. Bmal1−/− and Aldh1L1-CreERT2+, CX3CR1-CreERT2+, and Bmal1f/f mice were obtained from the Jackson laboratory and bred at Washington University. Tissue from NPAS2 KO, CLOCK KO, and CLOCK/NPAS2 double KO was provided by D. Weaver (University of Massachusetts, Worcester, MA). Chi3l1−/− mice were obtained from J. Elias (Brown University, Providence, RI). APP/PS1-21 mice were obtained from M. Jucker (University of Tübingen, Tübingen, Germany). Timed-pregnant WT CD1 mice for culture experiments were obtained from Charles River Laboratories (Wilmington, MA). All mice except those used for perinatal cultures were maintained on a C57BL/6 background and housed under 12-hour light/12-hour dark conditions, unless otherwise specified. All mice expressing any Cre or APP/PS1 transgene were heterozygous for these transgenes. Aldh1L1-CreERT2+;Bmal1f/f and CX3CR1-CreERT2+; Bmal1f/f mice were given tamoxifen (dissolved in corn oil, 2 mg per mouse per day for 5 days; Sigma-Aldrich) by oral gavage at 1 or 2 months, respectively, to induce Bmal1 deletion in the applicable tissue. Cre, Bmal1f/f control littermates were given identical tamoxifen treatment.

Human studies

For human studies, data were collected as part of several ongoing longitudinal observational studies of aging and dementia carried out at the Knight Alzheimer’s Disease Research Center. Participants are evaluated annually by clinical staff who are blinded to the participant’s previous diagnosis and all previously collected data, allowing an unbiased assessment of AD diagnosis CDR each year (43). The inclusion/exclusion criteria for our analyses were predetermined, and only data from participants meeting these criteria were queried. Inclusion required a clinical AD diagnosis at the last visit, available CSF biomarkers with a profile compatible with AD (based on CSF Aβ/tau profiles with established cutoffs), a CDR > 0 at last assessment, and at least 1.5 years of follow-up. Exclusions included a clinical diagnosis of a non-AD form of dementia or diagnoses of another coexistent neurological diseases. In total, 778 participants were included.

Statistical analysis

For mouse and cell experiments, statistical analyses were performed using GraphPad Prism version 8.02. When multiple t tests were performed, Holm-Sidak correction test for multiple comparisons was applied unless otherwise noted. Fluidigm qPCR data were analyzed by two-way [individual gene or gene groups provided no baseline change in phosphate-buffered saline (PBS) or APP/PS1 mice] or three-way [if a significant main effect of genotype was found by two-way analysis of variance (ANOVA) at baseline in PBS or APP/PS1 mice] ANOVA with Tukey correction for multiple comparisons. Other tests are noted in figure legends. In mouse and cell experiments, data points were determined to be outliers (and thus excluded) based on the ROUT method in Prism 8, Q = 1%, performed post hoc where appropriate.

Statistical analysis of human AD progression was carried out using R statistical software, and the package nlme was used for a linear mixed model. A linear mixed-model repeated measure framework was used to account for correlation between repeated measures in the same individual. Disease progression was modeled as follows(Y)=β_1[SNP*time]+β_2 [CDR_baseline8*time]+β_3age_baseline β_4gender+β_5 education+β_6 SNP+β_7CDR_(baseline+) β_8time+β_9 PC1+β_10 PC2where Y was CDR-SB, the change in CDR-SB per year baseline CDR, baseline age, gender, follow-up time, level of education, and, to avoid the possibility of spurious association due to population substructure, the two first principal components scores were included as covariates.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/12/574/eaax3519/DC1

Materials and Methods

Fig. S1. Loss of Chi3l1 exacerbates the LPS-induced inflammatory response.

Fig. S2. Loss of Chi3l1 does not change astrocyte reactivity gene signature but does modulate the microglial response to LPS.

Fig. S3. Loss of Chi3l1 mitigates amyloid pathology.

Fig. S4. Plaque number and size reduced with Chi3l1 deletion without affecting APP processing.

Fig. S5. Loss of Chi3l1 mitigates plaque-related astrogliosis.

Fig. S6. Loss of Chi3l1 alters plaque-related microglial activation.

Fig. S7. Loss of Chi3l1 reduces inflammation in APP/PS1+ mouse hippocampus.

Fig. S8. Chi3l1 is regulated by the circadian clock and expressed in astrocytes.

REFERENCES AND NOTES

Acknowledgments: We thank D. Weaver (UMass Medical School) for Clock and Npas2 KO mouse tissue and the Washington University Center for Cellular Imaging (WUCCI) for imaging assistance. Funding: This work was supported by NIH grants R01AG054517 (E.S.M) and R01AG044546, P01AG003991, RF1AG053303, R01AG058501, U01AG058922, U01AG052411, and R01AG05777 (C.C.). WUCCI is supported by the Washington University School of Medicine, The Children’s Discovery Institute of WU and St. Louis Children’s Hospital (CDI-CORE-2015-505), and the Foundation for Barnes-Jewish Hospital (3770). The Knight ADRC at Washington University is funded by NIA grants P50-AG05681, P01-AG03991, and P01-AG026276. Author contributions: B.V.L. and E.S.M. designed and conducted mouse and cell experiments, analyzed data, and wrote the manuscript. M.W.K. conducted and analyzed phagocytosis experiments. C.A.M., J.M.D, C.J.N., D.D.X., and A.J.C conducted mouse and cell experiments. C.G. and J.Z. performed and analyzed ChIP-qPCR. J.L.D.-A., F.H.G.F., and C.C. analyzed human genetic data and contributed to the manuscript. J.A.E. provided Chi3l1 KO mice. Competing interests: E.S.M. has consulted for Eisai Pharmaceuticals. No patents related to this work are pending. The other authors declare that they have no competing interest. Data and materials availability: All the data associated with this study are present in the paper or the Supplementary Materials. Microarray data are freely available on the EMBL-EBI ArrayExpress database, accession no. E-MTAB-7151.

Stay Connected to Science Translational Medicine

Navigate This Article