Article

TREM2 Maintains Microglial Metabolic Fitness in

Alzheimer’s Disease
Graphical Abstract Authors
Tyler K. Ulland, Wilbur M. Song,
Stanley Ching-Cheng Huang, ...,
Maxim N. Artyomov, David M. Holtzman,
Marco Colonna

Correspondence
[email protected]

In Brief
The Alzheimer’s disease risk factor
TREM2 regulates microglial function
through modulation of cellular
biosynthetic metabolism.

Highlights
d TREM2-deficient microglia undergo increased autophagy in
an AD mouse model

d Microglia in humans with AD-risk-associated TREM2 alleles

display marked autophagy

d TREM2 deficiency impairs microglial mTOR activation and

metabolism

d Cyclocreatine improves microglia metabolism and

pathology in TREM2-deficient AD mice

Ulland et al., 2017, Cell 170, 649–663

August 10, 2017 ª 2017 Elsevier Inc.
http://dx.doi.org/10.1016/j.cell.2017.07.023
Article

TREM2 Maintains Microglial Metabolic

Fitness in Alzheimer’s Disease
Tyler K. Ulland,1,7 Wilbur M. Song,1,7 Stanley Ching-Cheng Huang,1 Jason D. Ulrich,2,3,4 Alexey Sergushichev,5
Wandy L. Beatty,6 Alexander A. Loboda,5 Yingyue Zhou,1 Nigel J. Cairns,2,3,4 Amal Kambal,1 Ekaterina Loginicheva,1
Susan Gilfillan,1 Marina Cella,1 Herbert W. Virgin,1 Emil R. Unanue,1 Yaming Wang,1 Maxim N. Artyomov,1
David M. Holtzman,2,3,4 and Marco Colonna1,8,*
1Department of Pathology and Immunology
2Department of Neurology
3Knight Alzheimer’s Disease Research Center
4Hope Center for Neurological Disorders

Washington University School of Medicine, St. Louis, MO 63110, USA

5Computer Technologies Department, ITMO University, Saint Petersburg 197101, Russia
6Department of Molecular Microbiology, Washington University School of Medicine, St. Louis, MO 63110, USA
7These authors contributed equally
8Lead Contact

*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.cell.2017.07.023

SUMMARY death, and ultimately cognitive decline (Holtzman et al., 2011;

Huang and Mucke, 2012). Rare familial AD is due to mutations
Elevated risk of developing Alzheimer’s disease (AD) in amyloid precursor protein (APP) and presenilins (PS) that pro-
is associated with hypomorphic variants of TREM2, a mote the generation of Ab peptides prone to aggregation (Tanzi,
surface receptor required for microglial responses to 2012). However, the risk for common late-onset AD is associated
neurodegeneration, including proliferation, survival, with rare variants of immune receptors expressed on microglia
clustering, and phagocytosis. How TREM2 promotes (Guerreiro and Hardy, 2014; Tanzi, 2012). One of these recep-
tors, TREM2, recognizes phospholipids, apoptotic cells, and
such diverse responses is unknown. Here, we find
lipoproteins (Atagi et al., 2015; Bailey et al., 2015; Wang et al.,
that microglia in AD patients carrying TREM2 risk
2015; Yeh et al., 2016). TREM2 transmits intracellular signals
variants and TREM2-deficient mice with AD-like pa- through two adapters, DAP12 and DAP10, which recruit protein
thology have abundant autophagic vesicles, as do tyrosine kinase Syk and phosphatidylinositol 3-kinase (PI3-K),
TREM2-deficient macrophages under growth-factor respectively (Peng et al., 2010). Arginine-to-histidine variants at
limitation or endoplasmic reticulum (ER) stress. Com- position 47 (R47H) or 62 (R62H) of TREM2 increase the risk for
bined metabolomics and RNA sequencing (RNA-seq) sporadic AD and impair binding to phospholipid ligands (Atagi
linked this anomalous autophagy to defective et al., 2015; Bailey et al., 2015; Guerreiro and Hardy, 2014;
mammalian target of rapamycin (mTOR) signaling, Wang et al., 2015; Yeh et al., 2016). These variants, as well as
which affects ATP levels and biosynthetic pathways. TREM2 deficiency and haploinsufficiency in mouse models of
Metabolic derailment and autophagy were offset AD, moderate microglial proliferation, survival, and accumulation
around Ab plaques, thereby facilitating Ab plaque buildup and
in vitro through Dectin-1, a receptor that elicits
injury of adjacent neurons (Jay et al., 2017; Ulrich et al., 2014;
TREM2-like intracellular signals, and cyclocreatine,
Wang et al., 2015, 2016; Yuan et al., 2016). TREM2 has also
a creatine analog that can supply ATP. Dietary cyclo- been implicated in microglial phagocytosis of dead neurons,
creatine tempered autophagy, restored microglial damaged myelin, and Ab plaques (Neumann and Takahashi,
clustering around plaques, and decreased plaque- 2007; Yeh et al., 2016). However, why defective TREM2 function
adjacent neuronal dystrophy in TREM2-deficient or expression affects microglia responses to AD lesions is
mice with amyloid-b pathology. Thus, TREM2 en- not known.
ables microglial responses during AD by sustaining To address this question, we first used electron and confocal
cellular energetic and biosynthetic metabolism. microscopy to analyze microglia from 5XFAD mice, which
develop Ab accumulation that mimics AD pathology due to the
expression of mutant APP and PS1 under neural-specific ele-
INTRODUCTION ments of the mouse Thy1 promoter. Microglia from 5XFAD
mice lacking TREM2 had many more autophagic vesicles than
Alzheimer’s disease (AD) is the most common cause of late- did microglia in 5XFAD mice. This observation was replicated
onset dementia. AD lesions in the CNS include plaques of in humans, as microglia in AD patients carrying TREM2 risk
amyloid b (Ab) peptides and neurofibrillary tangles of hyperphos- variants also had more autophagic vesicles than did microglia
phorylated tau protein, both linked to synapse loss, neuronal in AD patients with the common TREM2 variant. Autophagy is

Cell 170, 649–663, August 10, 2017 ª 2017 Elsevier Inc. 649
Figure 1. Defect in TREM2 Enhances Autophagy In Vivo in the 5XFAD Mouse Model and in AD Patients
(A) TEM images of microglia (CD45+, CD11b+, F4/80+ cells) sorted from 8-month-old WT, Trem2–/–, 5XFAD, and Trem2–/– 5XFAD mice.
(B) Average number of multivesicular and multilamellar structures/cell (30 cells analyzed/genotype).

(legend continued on next page)

650 Cell 170, 649–663, August 10, 2017

an intracellular degradation pathway essential for cellular and 5XFAD, Trem2–/–, or wild-type (WT) mice (Figures 1A and 1B).
energy homeostasis (Galluzzi et al., 2014). It provides a mecha- To determine whether these structures reflected ongoing
nism for the elimination of misfolded proteins and damaged autophagy in situ, we examined brain sections by confocal
organelles and compensates for nutrient deprivation during cell microscopy for the presence of LC3+ puncta, which denote
starvation through recycling of cytosolic components. Because autophagosomes decorated by lipidated LC3II (Klionsky et al.,
autophagy is partially regulated by mammalian target of rapamy- 2016). Many large LC3+ puncta were evident in microglia in
cin (mTOR)-dependent pathways (Saxton and Sabatini, 2017), Trem2–/– 5XFAD mice, whereas LC3+ puncta were sparse in
we assessed the impact of TREM2-deficiency on mTOR activa- microglia in WT, Trem2–/–, and 5XFAD mice (Figures 1C, 1D,
tion and found that, indeed, anomalous autophagy reflected and S1). These observations were translatable to human dis-
defective activation of mTOR signaling. ease. We observed dramatically more LC3+ microglia in post-
Similarly, enhanced autophagy was observed in TREM2-defi- mortem brain sections from both R47H and R62H heterozygous
cient macrophages in vitro, which was further amplified by AD patients than in those from case-matched AD patients
growth-factor limitation or endoplasmic reticulum (ER) stress; homozygous for the common TREM2 variant (Figures 1E and
this provided a model system for probing biochemical and meta- 1F; Table S1). Taken together, these data suggest that autopha-
bolic pathways in microglia during Ab accumulation. Combined gic-like vesicles accumulate in the microglia of TREM2-deficient
metabolomics, RNA sequencing (RNA-seq), and system ana- mice and humans with TREM2 risk alleles during the develop-
lyses of TREM2-deficient macrophages confirmed the impair- ment of AD.
ment of mTOR activation, energetic pathways, ATP levels, and
biosynthetic pathways. Thus, TREM2 sustains cell energetic TREM2 Deficiency Impairs mTOR Signaling and
and biosynthetic metabolism through mTOR signaling. Meta- Enhances AMPK Activation in Microglia
bolic derailment and autophagy were offset in vitro through To corroborate the association between TREM2 deficiency and
activation of Dectin-1, a surface receptor that triggers a signaling increased autophagy, we performed biochemical analyses on
pathway similar to that of TREM2-DAP12 (Dambuza and Brown, sorted microglia ex vivo. The ratio of lipidated LC3II to non-lipi-
2015). Metabolic abnormalities were also rescued by incubating dated LC3I was markedly higher in microglia from Trem2–/–
cells with the creatine analog 1-carboxymethyl-2-iminoimidazo- 5XFAD mice than in microglia from 5XFAD mice, consistent
lidine (cyclocreatine), which can passively cross membranes with the increased number of autophagic vesicles in TREM2-
and, upon phosphorylation by creatine kinase, generate a supply deficient microglia (Figures 2A and 2B). To determine whether
of ATP for energy demands independent of the TREM2-mTOR the increase in autophagosomes was due to activation of auto-
axis (Kurosawa et al., 2012). Remarkably, dietary administration phagy or blockade of lysosomal degradation, we measured pro-
of cyclocreatine in 5XFAD mice lacking TREM2 prevented micro- tein and mRNA levels of p62, an autophagy cargo protein that is
glial autophagy, enhanced microglia numbers and clustering digested by lysosomal enzymes. The amount of p62 protein was
around Ab plaques, and mitigated plaque-associated neurite lower in microglia from Trem2–/– 5XFAD mice than in microglia
dystrophy. This provides proof of principle that strategies aimed from 5XFAD mice. This difference was unrelated to transcription,
at sustaining basic microglial metabolism may be promising for as p62 (Sqstm1) mRNA levels were similar (Figures 2A and S2A).
treatment of AD and other neurodegenerative diseases associ- Thus, TREM2 deficiency results in bona fide autophagy.
ated with microglial dysfunction. Why is autophagic flux amplified in 5XFAD mice lacking
TREM2? Autophagy often reflects an adaptive response to
RESULTS stress that can occur when cells cannot satisfy increased de-
mands for energy and protein synthesis (Kroemer et al., 2010).
Defect in TREM2 Elicits Autophagy In Vivo in the 5XFAD Because the serine/threonine kinase target of rapamycin
Mouse Model and AD Patients (mTOR) has a crucial role in stimulating both energetic and
To determine the impact of TREM2 deficiency on microglia func- anabolic metabolism, cell growth, and proliferation (Laplante
tion, we examined the structure of microglia from the 5XFAD and Sabatini, 2012), we hypothesized that the autophagy
mouse model of AD by transmission electron microscopy observed in microglia in Trem2–/– 5XFAD mice might result
(TEM). Strikingly, microglia from Trem2–/– 5XFAD mice contained from a defect in mTOR signaling. mTOR signals through two
abundant multivesicular/multilamellar structures suggestive of distinct complexes, mTORC1 and mTORC2. Immunoblotting of
autophagosomes, which were largely absent in microglia from sorted microglia from Trem2–/– 5XFAD and 5XFAD mice revealed

(C) Confocal images of plaque-bearing regions of the cortex (1.1 mm Bregma to 0.8 mm Bregma) of 8-month-old WT, Trem2–/–, 5XFAD, and Trem2–/– 5XFAD mice
show Iba-1+ microglia (red), methoxy X04+ plaques (blue), and LC3 (green). Z stacks composed of
30 images taken at 1.2-mm intervals were analyzed. Results
are reported as an average of 2 regions of interest (ROI) analyzed.
(D) Quantification of the % of microglia that are positive for LC3 puncta.
150–400 microglia/HPF were analyzed depending on the genotype of the animal.
(E) Confocal images of sections from post-mortem brains of R47H+ AD patients and case-matched controls (CV, common variant of TREM2) show Iba-1+
microglia (red), methoxy X04+ plaques (blue), and LC3 (green). 3 ROIs/donor were analyzed and between 400–700 microglia/individual were analyzed.
(F) Percentages of LC3+ microglia in post mortem specimens of AD patients with different genotypes.
Error bar represents mean ± SEM. ***p < 0.005, ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test. 15 cells from 2 separate mice
were visualized for TEM (A and B). Confocal images are representative of 3 female mice per group (C) or 7 R47H, 4 R62H, and 8 case-matched AD patients for
post-mortem specimens (E). Arrowheads indicate multilamellar and multivesicular structures (A) or LC3+ vesicles (C and E).
See also Figure S1 and Table S1.

Cell 170, 649–663, August 10, 2017 651

Figure 2. Defect in TREM2 Impairs mTOR Activation and Elicits AMPK Activation, Autophagy, and Cell Death in Microglia from 5XFAD Mice
(A) Microglia were sorted as in Figure 1A. Immunoblots for LC3I/II, p62, phospho-Akt (serine 473), phospho-AMPK, phospho-NDRG1, phospho-4EBP1,
phospho-ULK1 (serine 757), and b-actin were performed on cell lysates.
(B) Quantification of the LC3II/I ratio in microglia from different genotypes.
(C) Single-cell suspensions of brain tissue were incubated with MitoTraker Green and stained for CD45+, CD11b+, and F4/80+. Representative histograms
comparing unstained 5XFAD microglia and stained 5XFAD and Trem2–/– 5XFAD microglia are shown.
(D) Quantification of the geometric mean fluorescence intensity (gMFI) of microglia from 3 mice of each genotype is shown.
(E) Confocal images of brain sections of 8-month-old WT, Trem2–/–, 5XFAD, and Trem2–/– 5XFAD mice were taken as in Figure 1C. Images depict Iba-1+ microglia
(red), methoxy X04+ plaques (blue), and cleaved caspase-3 (green).
(F) Quantification of the % of LC3+ microglia that are positive for cleaved caspase-3.
Error bar represents mean ± SEM. ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test (B and F). **p < 0.01 by Student’s t test (D).
Immunoblots are representative of 3 independent experiments from microglia from 3 separate mice per group (A). Confocal images are representative of 3 female
mice per group (E).
See also Figure S2.

decreased phosphorylation of 4EBP1, an mTORC1 effector, as with reduced mTOR activation and increased autophagy (Fig-
well as Akt (serine 473) and NDRG1, both mTORC2 effectors, ure 2A). Impaired mTOR signaling was associated with phos-
in the TREM2-deficient microglia (Figure 2A). Ulk1, a key inducer phorylation of AMP-activated protein kinase (AMPK), a sensor
of autophagy, which is inhibited by mTOR signaling through of low energy states (Figure 2A).
phosphorylation at serine 757 (Kim et al., 2011), was less phos- Providing further evidence for defective energetic and
phorylated in microglia from Trem2–/– 5XFAD mice, consistent anabolic metabolism associated with TREM2 deficiency,

652 Cell 170, 649–663, August 10, 2017

microglia from Trem2–/– 5XFAD mice had a lower mitochondrial targets in WT BMDMs; the residual amount of phosphorylation
mass than microglia from 5XFAD mice (Figures 2C and 2D). was similar to that seen in Trem2–/– BMDMs (Figures 3I and
Furthermore, gene expression microarray analyses of sorted S3A). In addition to limiting CSF1, other stressors may differen-
microglia from Trem2–/– 5XFAD and 5XFAD mice revealed that tially modulate mTOR signaling and autophagy in WT and
TREM2 deficiency was associated with decreased expression Trem2–/– BMDMs. Treatment with tunicamycin, which provokes
of genes encoding translation initiation factors, ribosomal pro- ER stress, unfolded protein response, and autophagy, induced
teins, glucose transporters, glycolytic enzymes, as well as the greater LC3II/LC3I ratios and less phosphorylated Akt (serine
transcription factor HIF1a that controls glycolysis (Figures 473) in Trem2–/– BMDMs than in WT BMDMs (Figures 3J and
S2B–S2G). Taken together, these data demonstrate that during 3K). Thus, TREM2 deficiency affects cell responses to multiple
the development of AD, TREM2 deficiency derails the mTOR stressors.
pathway, anabolic and energetic metabolism in microglia, which TREM2 signaling may not just have a more pronounced effect
induces compensatory autophagy in both mice and humans. on cells under stress conditions, but may actually increase under
To determine whether autophagy in TREM2-deficient micro- such conditions. When cultured in media containing 10% fetal
glia can effectively compensate for the metabolic defects in vivo bovine serum (FBS), TREM2 reporter cells, which express GFP
and prevent apoptosis, we examined brain sections by confocal upon TREM2 engagement (Wang et al., 2015), showed some
microscopy for the presence of cleaved caspase-3, an indicator activation (Figure S3B). However, upon serum starvation, a
of apoptosis. We found that cleaved caspase-3 was much more significantly higher proportion of reporter cells became acti-
abundant in microglia in Trem2–/– 5XFAD mice than in microglia vated, possibly due to exposure of the TREM2 ligand phospha-
in 5XFAD mice (Figures 2E and 2F). Additionally, the percentage tidylserine on the outer leaflet of stressed cells. This activation
of microglia with colocalization of cleaved caspase-3 and LC3+ could be blocked by inclusion of an anti-TREM2 antibody.
puncta was higher in Trem2–/– 5XFAD mice than in 5XFAD Thus, in multiple in vitro settings of stress, TREM2-expressing
mice (Figures S2H and S2I). Thus, autophagy may not be suffi- cells are more able to sustain mTOR activation and sup-
cient to sustain the microglial response to stress, at least at the press autophagy in a PI3-K-dependent manner than are cells
late time point of disease progression analyzed. lacking TREM2.

Macrophages Lacking TREM2 Inadequately Signal TREM2 Deficiency Curtails Anabolic and Energetic
through mTOR and Undergo More Autophagy Metabolism in BMDMs
We then asked whether TREM2-deficiency could derail mTOR To directly demonstrate the impact of TREM2 deficiency on
signaling in bone marrow-derived macrophages (BMDMs) energetic and anabolic pathways in BMDMs, we performed
in vitro. To mimic the metabolic stress that occurs during dis- mass spectrometry to quantify cellular metabolites and RNA-
ease, we used growth-factor deprivation. BMDMs were cultured seq to quantify mRNA levels of metabolic enzymes. Analysis of
overnight in concentrations of CSF1-containing L cell-condi- metabolite data alone, or in combination with RNA-seq data by
tioned medium (LCCM) ranging from optimal to limiting (10% a systems-based algorithm (Sergushichev et al., 2016), revealed
to 0.5%). Trem2–/– BMDMs contained more autophagic vesicles widespread differences between WT and Trem2–/– BMDMs in
(Figures 3A and 3B) and had a higher LC3II/LC3I ratio than WT various metabolic pathways. Compared to WT BMDMs in
cells when CSF1 was limiting (Figure 3C). Addition of the lyso- optimal CSF1, Trem2–/– BMDMs cultured under the same condi-
somal inhibitor bafilomycin greatly increased LC3II in Trem2–/– tions exhibited (1) a marked decrease of key intermediates in the
BMDMs, confirming that the increase in autophagosomes was synthesis of nucleotides (e.g., phosphoribosyl pyrophosphate),
due to increased autophagic flux rather than reduced autopha- N-glycosylated proteins (e.g., UDP-glucose), and phospholipids
gosome degradation (Figures 3D and 3E). (e.g., CDP-ethanolamine); (2) a decrease in glycolytic metabo-
As observed in sorted microglia, autophagy in BMDMs was lites (e.g., glucose 6-phosphate and fructose bisphosphate)
linked to impaired mTOR signaling. Trem2–/– BMDMs had less and tricarboxylic acid (TCA) cycle intermediates (citrate and suc-
phosphorylated 4EBP1, Akt (serine 473), and NDRG1 (Figure 3F) cinate); and (3) an increase in catabolic products of amino acids
and more phosphorylated AMPK (Figure 3G) in both optimal (e.g., indolacetate) and phospholipids (e.g., glycerol 3-phos-
CSF1 and limiting CSF1 than did WT BMDMs. In limiting CSF1, phate) (Figures 4A and S4A). Moreover, a selective increase in
Trem2–/– BMDMs had decreased inhibitory phosphorylation of malate and fumarate suggested an enhanced malate-aspartate
Ulk1 at serine 757, while activating phosphorylation of Ulk1 at shuttle to sustain defective NADH oxidation and NAD regenera-
serine 317 increased (Figure 3H). Thus, lack of TREM2 sup- tion (Figure 4A). Unbiased network analysis combining metabolic
pressed mTOR activation and elicited compensatory AMPK and RNA-seq data highlighted defects in metabolites and
and Ulk1 activation and autophagy in BMDMs in response to enzymes involved in glycolysis, TCA cycle, and pentose phos-
metabolic stress, very similar to our observations of microglia phate pathway in Trem2–/– BMDMs (Figure 4B).
in 5XFAD mice. CSF1 reduction further deteriorated energy and anabolic
As TREM2 signaling adapters DAP12 and DAP10 have been metabolism in Trem2–/– BMDMs. Under these conditions, again
shown to activate PI3-K, which in turn can activate mTOR, we in comparison to WT BMDMs, Trem2–/– BMDMs underwent a
asked whether enhanced mTOR signaling in WT BMDM marked increase in symmetrical dimethyl arginine, indicative
compared to Trem2–/– BMDM was dependent on PI3-K. Inhibi- of protein catabolism, as well as an increase in ADP-ribose,
tion of PI3-K with Wortmannin or LY294002 caused a major indicative of NAD degradation (Figures 4C, S4B, and S4C).
reduction in phosphorylation of mTOR and its downstream Furthermore, stores of high-energy phosphates, such as

Cell 170, 649–663, August 10, 2017 653

Figure 3. TREM2 Deficiency Affects mTOR Signaling and Induces Autophagy in BMDM
(A) TEM images of WT and Trem2–/– BMDM cultured overnight in either in 10% or 0.5% LCCM as source of CSF1.
(B) Number of multivesicular structures/cell observed in the TEM images. 30 cells/genotype and condition were analyzed.
(C) Quantification of the LC3II/LC3I ratio in BMDMs from WT and Trem2–/– mice cultured in 10% or 0.5% LCCM overnight or starved in HBSS for 4 hr prior to lysis.
(D) Immunoblots for LC3 and actin performed on lysates from WT and Trem2–/– BMDMs cultured in 10% or 0.5% LCCM overnight. Cell were treated with
bafilomycin for 5 hr prior to harvest at a final concentration of 0.5 mg/mL.
(E) Quantification of LC3II/LC3I ratio in BMDMs from WT and Trem2–/– mice treated as indicated.
(F–H) Immunoblots for phosphorylated Akt (serine 473), NDRG1, S6K, 4EBP1, (F), AMPK, (G), Ulk1 (serine 317 and serine 757) (H), and relative controls. Lysates
were from WT and Trem2–/– BMDM cultured overnight in 10% or 0.5% LCCM.
(I) Immunoblots for phosphorylated Akt (serine 473), NDRG1, S6K, 4EBP1, mTOR, total S6K, Akt, and actin performed on lysates from WT and Trem2–/– BMDMs
cultured overnight in 10% or 0.5% LCCM followed by the addition of Wortmannin for 3 hr prior to harvest.
(J) Immunoblots for LC3 and phosphoserine 473 Akt in WT and Trem2–/– BMDM cultured in 10% LCCM with the indicated concentration of tunicamycin.
(K) Bar graphs show quantification of the LC3II/LC3I ratios from (J).
Error bar represents mean ± SEM. *p < 0.05, **p < 0.01, or ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test (B, C, E, and K). Data are
representative of at least 3 independent experiments. Arrowheads indicate multilamellar and multivesicular structures (A).
See also Figure S3.

654 Cell 170, 649–663, August 10, 2017

phosphocreatine and ATP, were depleted in Trem2–/– BMDMs reduction in the LC3II/LC3I ratio (Figures 5D and 5E) along with
cultured in limiting CSF1 conditions (Figure S4D). A luciferase- increased amounts of p62 (Figure 5F). Treatment with zymosan
based ATP assay confirmed an ATP deficiency in Trem2–/– also enhanced cellular ATP levels in Trem2–/– BMDMs, restoring
BMDMs, which was exacerbated at low CSF1 concentrations them to WT BMDM levels (Figure 5G). Thus, alternative energetic
(Figure 4D). We further assessed the energy metabolism of WT and signaling pathways can compensate for lack of TREM2
and Trem2–/– BMDMs using the Seahorse analyzer. A lower signaling.
extracellular acidification rate (ECAR), indicative of less glyco-
lytic flux, was noted in Trem2–/– BMDMs both at baseline and Cyclocreatine Rescues Microgliosis and Clustering and
after induction of maximal glycolytic capacity by oligomycin Moderates Neurite Dystrophy In Vivo
and fluoro-carbonyl cyanide phenylhydrazone (FCCP). This Because cyclocreatine rescued metabolism and viability and
deficit widened relative to WT cells as the CSF1 concentration suppressed autophagy in Trem2–/– BMDMs in vitro and given
was reduced (Figure 4E). Trem2–/– BMDMs had only a slightly previous studies showing that cyclocreatine is passively trans-
reduced oxygen consumption rate (OCR) compared to WT ported across membranes and can accumulate and function as
BMDMs when cultured in standard CSF1 concentrations, indi- a phosphagen in the mouse brain in vivo (Kurosawa et al.,
cating relatively intact oxidative phosphorylation; however, a 2012), we asked whether dietary supplementation with cyclo-
deficit in OCR emerged as the CSF1 concentration was reduced creatine could rescue microglial function and suppress auto-
(Figure 4E). Trem2–/– BMDMs also had fewer mitochondria than phagy in vivo in Trem2–/– 5XFAD mice. The drinking water of
WT BMDMs on a per cell basis as measured by MitoTracker WT, Trem2–/–, 5XFAD, and Trem2–/– 5XFAD mice was supple-
Green fluorescence and by the mitochondrial-to-nuclear DNA mented with cyclocreatine from 10 weeks of age until 8 months
ratio (Figures 4F and 4G). These findings were not restricted to of age. Remarkably, significantly fewer multivesicular/multila-
BMDMs, as resting and thioglycolate-elicited peritoneal mellar structures were seen by TEM in microglia of Trem2–/–
TREM2-deficient macrophages also had a lower mitochondrial 5XFAD mice treated with cyclocreatine than in microglia of
mass than did WT macrophages (Figures S4E–S4H). Cultured untreated mice (Figures 6A and 6B). Confocal microscopy
adult primary Trem2–/– microglia recapitulated the deficiencies corroborated that cyclocreatine treatment ameliorated many
in energetic metabolism and mTOR signaling as well as auto- autophagy and viability parameters in Trem2–/– 5XFAD mice,
phagy observed in Trem2–/– BMDMs (Figures S4I–S4M). Thus, such as the number of LC3 puncta/high powered field (HPF)
lack of TREM2-mTOR signaling impairs the energy status and (Figures 6C and 6D), the percentage of LC3+ microglia (Fig-
anabolism of BMDMs and other primary macrophages both in ure S5A), and the percentage of cleaved caspase-3+ microglia
steady state and under energetic stress. (Figure 6E), although other parameters did not reach statistical
significance (Figures S5B–S5D). Furthermore, clustering of
Enhanced Energy Storage or Dectin-1 Signaling Can microglia around plaques (Figures 6C and 6F) and the number
Compensate for TREM2 Deficiency In Vitro of microglia/HPF in plaque-bearing regions of the cortex
Given the dramatic effect of TREM2 deficiency on mTOR activa- (Figure S5E) were both significantly increased in Trem2–/–
tion and energy utilization in BMDMs, we tested whether bypass- 5XFAD mice treated with cyclocreatine compared to untreated
ing TREM2 and directly compensating for these deficits by Trem2–/– 5XFAD mice. These findings indicate that dietary sup-
alternative means could restore the cellular energy status of plementation with cyclocreatine is sufficient to partially rescue
Trem2–/– BMDMs. Muscle physiology studies have extensively the defect in microgliosis and microglial clustering around pla-
demonstrated that creatine phosphate contributes to the regen- ques in Trem2–/– 5XFAD mice, while concomitantly mitigating
eration of ATP and to the maintenance of uniformly high ATP/ autophagy and death of the microglia.
ADP ratios in muscle fibers (Walker, 1979). Moreover, the To assess the impact of cyclocreatine on microglial activation,
creatine analog 1-carboxymethyl-2-iminoimidazolidine (cyclo- which is also impaired in Trem2–/– 5XFAD mice, we quantified the
creatine) can, upon phosphorylation, generate a long-acting percentage of microglia that expressed the activation marker os-
phosphagen that can effectively sustain cellular ATP levels teopontin (Spp1), a protein that, in the brain, is upregulated in mi-
during increased energy demand (Kurosawa et al., 2012; Woz- croglia in the context of Ab deposition (Orre et al., 2014; Wang
nicki and Walker, 1979). Thus, we tested whether addition of et al., 2015). Untreated Trem2–/– 5XFAD mice had very few
cyclocreatine to the culture medium could rescue energetic Spp1+ microglia, while cyclocreatine-treated Trem2–/– 5XFAD
metabolism in Trem2-deficient BMDMs. Indeed, incubation mice had significantly more Spp1+ microglia, as did TREM2-
with cyclocreatine improved ECAR, which was accompanied sufficient 5XFAD mice (Figures 7A and 7B). Moreover, biochem-
by less autophagy, increased mTOR signaling and cell viability ical analysis of microglia isolated ex vivo demonstrated that
(Figures 5A–5C and data not shown). cyclocreatine treatment of Trem2–/– 5XFAD mice also restored
To test whether engagement of receptors that elicit signaling microglial mTOR signaling and limited autophagy compared to
pathways similar to those of TREM2 could also mitigate auto- untreated Trem2–/– 5XFAD mice (Figures 7C and 7D).
phagy and support cell survival, Trem2–/– and WT BMDMs As a major function of TREM2 in vivo is enabling microglia to
were cultured with depleted zymosan, a selective ligand of form a barrier around plaques that prevents spreading of Ab
Dectin-1, which activates Syk and PI3-K signaling independent fibrils and alleviates dystrophy of plaque-adjacent neurites
of DAP12 (Dambuza and Brown, 2015). Dectin-1 activation (Wang et al., 2016; Yuan et al., 2016), we asked whether cyclo-
curbed autophagy in CSF1-starved TREM2-deficient BMDMs creatine treatment of Trem2–/– 5XFAD mice impacted plaque
to levels seen in WT BMDMs in low CSF1, as indicated by a morphology and/or neuronal dystrophy. While plaques in

Cell 170, 649–663, August 10, 2017 655

Figure 4. TREM2 Deficiency Reduces Anabolic and Energetic Metabolism in BMDM
(A) Top most changed metabolites between WT and Trem2–/– BMDM cultured overnight in 10% LCCM. Defined as p % 0.01 and identified in the mouse metabolic
network analysis in (B).
(B) Shiny-genes and metabolites (GAM) output for network analysis combining mass spectrometry and RNA-seq data highlights differences between WT and
Trem2–/– BMDM cultured in 10% LCCM. Enzyme-encoding mRNAs and metabolites downregulated or upregulated in Trem2–/– cells versus WT cells are indi-
cated with green or red nodes and connectors, respectively.
(C) Top most changed metabolites between WT and Trem2–/– BMDM cultured in 0.5% LCCM. Defined as p % 0.01 and identified in the mouse metabolic network
analysis in Figure S4C.
(D) ATP content of WT and Trem2–/– BMDM cultured in the indicated concentration of LCCM overnight.
(E) Extracellular acidification rate (ECAR) and baseline oxygen consumption rate (OCR) by WT and Trem2–/– BMDM cultured overnight in the indicated con-
centration of LCCM.

(legend continued on next page)

656 Cell 170, 649–663, August 10, 2017

Figure 5. Enhanced Energy Storage or Dectin-1 Signaling Can Compensate for TREM2 Deficiency
(A) ECAR of WT and Trem2–/– BMDM incubated overnight in 0.5% LCCM ± 10 mM cyclocreatine.
(B) Viability of WT and Trem2–/– BMDM incubated for 40 hr in 0.5% LCCM ± cyclocreatine.
(C) Immunoblots of LC3, phosphorylated mTOR, phosphorylated Akt (serine 473), and actin in WT and Trem2–/– BMDM incubated overnight in 0.5%
LCCM ± 5 mM cyclocreatine.
(D and F) LC3, phosphoserine 473 Akt (D), p62 (F), and actin (D and F) immunoblots from WT and Trem2–/– BMDM incubated overnight in the indicated
concentration of LCCM ± depleted zymosan.
(E) Quantification of the LC3II/LC3I ratio derived from immunoblots of LC3 as shown in (D).
(G) ATP content of WT and Trem2–/– BMDM cultured in the indicated concentration of LCCM ± zymosan overnight.
Error bar represents mean ± SEM. *p < 0.05 or ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test (B, E, and G). Data are repre-
sentative of results from at least 3 independent experiments.

untreated Trem2–/– 5XFAD mice had a lower density than those did not moderate plaque accumulation or the engulfment of pla-
in 5XFAD mice as measured by methoxy-X04 staining intensity, que particulates by microglia, at least at this time point (Figures
the density of plaques in cyclocreatine-treated Trem2–/– 5XFAD S5G–S5I). As APP is known to accumulate as distinct rounded
mice resembled that of plaques in 5XFAD mice (Figure 7E), particles in dystrophic neurites, we used APP deposition around
although plaque shape complexity was not significantly altered plaques to assess neurite dystrophy (Masliah et al., 1996; Wang
(Figure S5F). Despite reducing plaque density, cyclocreatine et al., 2016; Yuan et al., 2016). Cyclocreatine treatment of

(F and G) Mitochondrial mass of WT and Trem2–/– BMDM assessed by Mito Tracker Green incorporation (F) and by the ratio of mitochondrial-to nuclear DNA (G).
Error bar represents mean ± SEM. *p < 0.05, **p < 0.01, or ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test (C) or Student’s t test
(F and G). Data are representative of at least 3 independent experiments.
See also Figure S4.

Cell 170, 649–663, August 10, 2017 657

Figure 6. Enhanced Energy Storage Can Correct Microglial Defects in TREM2-Deficient 5XFAD Mice
(A) TEM images of microglia sorted from 8-month-old 5XFAD and Trem2–/– 5XFAD mice ± cyclocreatine.
(B) Quantification of the number of multivesicular and multilamellar structures/cell from (A).
(C) Confocal images of brain sections of 8-month-old 5XFAD and Trem2–/– 5XFAD mice ± cyclocreatine show Iba-1+ microglia (red), methoxy X04+ plaques (blue),
and LC3 (green).
(D) Quantification of the number of LC3 puncta per HPF in the cortexes of the indicated mice.
(E) Quantification of the percentage of microglia that were cleaved caspase-3-positive from the indicated mice.
(F) Clustering analysis quantifying the number of microglia per mm3 within 15 mm of the surface of plaques.
Error bar represents mean ± SEM. *p < 0.05, ***p < 0.005, and ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test (B and D–F) results
pooled from 2 independent experiments representing a total of 5–8 male and female mice per treatment group. Arrowheads indicate multilamellar and
multivesicular structures (A) or LC3+ vesicles (C).
See also Figure S5.

Trem2–/– 5XFAD mice significantly reduced plaque-associated DISCUSSION

neurite dystrophy compared to untreated Trem2–/– 5XFAD
mice to levels observed in 5XFAD mice (Figures 7F and 7G). Increasing evidence supports the hypothesis that the microglial
Taken together, these data indicate that cyclocreatine adminis- response to AD lesions controls disease progression (Gold and
tration improves microglial metabolism and the protective El Khoury, 2015; Hong et al., 2016; Meyer-Luehmann and Prinz,
response to Ab plaques in TREM2-deficient 5XFAD mice. 2015; Perry and Holmes, 2014; Tejera and Heneka, 2016; Wang

658 Cell 170, 649–663, August 10, 2017

Figure 7. Energy Supplementation Can Offset Neuronal Damage in TREM2-Deficient 5XFAD Mice
(A) Representative images depicting plaques (X04 in blue), nuclei (To-Pro3 in white), microglia (Iba-1 in red), and Spp1 (in green) staining in cortexes of mice from
the indicated genotypes.
(B) Quantification of the percentage of microglia that were Spp1+ in the indicated genotypes of mice. Confocal images were taken as in Figure 1C.

(legend continued on next page)

Cell 170, 649–663, August 10, 2017 659

et al., 2016; Yuan et al., 2016). Toll-like receptors and NOD-like TREM2-deficient microglia have been thought to improperly
receptors have been previously implicated in the microglia remain in a homeostatic state during neurodegenerative dis-
response to Ab accumulation and shown to mediate an inflam- ease rather than responding appropriately to pathology, a para-
matory response that contributes to pathology (Freeman and digm that has been supported by transcriptomic analysis of
Ting, 2016; Heneka et al., 2013, 2015). To sustain cytokine these cells. However, for the first time, we demonstrate that
secretion, these receptors induce a striking metabolic reprog- on a biochemical and ultrastructural level, TREM2-deficient mi-
ramming, which consists of a switch from fatty acid metabolism croglia adopt a severely divergent cellular state that does not
and oxidative phosphorylation to glycolysis (O’Neill and Pearce, reflect homeostasis during neurodegeneration, with a dramatic
2016). In our study, TREM2 emerges as an innate immune recep- loss of mTOR signaling and robust induction of autophagy.
tor that impacts microglia metabolism in AD through a distinct These results suggest that TREM2-deficient microglia are not
mechanism, which consists of basic activation of mTOR simply ignoring plaque pathology, but rather that they are being
signaling that supports long-term cell trophism, survival, growth, actively driven into a stressed state that is normally compen-
and proliferation, rather than drastic metabolic reprogramming. sated by TREM2-dependent survival signals. An important
This function of TREM2 is reminiscent of the tonic function of implication of this finding is that microglia in a neurodegenera-
the B cell antigen receptor in mature B cells, which delivers tive environment probably receive not only positive activating
survival signals through PI3-K (Werner et al., 2010). Likewise, signals but also negative cytotoxic signals. Thus, it may not
cell membrane phospholipids and lipoprotein particles may be microglial activation per se that is required to protect against
continuously engage TREM2, inducing tonic mTOR signaling neurodegeneration, but rather avoidance of a dysfunctional,
through upstream activators, such as PI3-K, PDK1, and Akt, low-energy state induced by the disease. Based on our find-
which are recruited by the TREM2-associated signaling subunits ings, previous reports of impaired microglial activation in a
DAP12 and DAP10 (Ford and McVicar, 2009; Peng et al., 2010). variety of settings may be due to either impaired recognition
This concept provides a unifying mechanism to explain the of activating signals or to impaired resistance to cytotoxic
reported broad and long-term impact of TREM2 on diverse signals—two possibilities that can be distinguished by the
microglial functions, such as survival, proliferation, clustering strength of mTOR signaling. Counteracting such dysfunction
around plaques, as well as phagocytosis of apoptotic cells and by metabolic compensation may also represent a fundamen-
myelin debris (Jay et al., 2017; Neumann and Takahashi, 2007; tally distinct therapeutic approach.
Wang et al., 2015; Yeh et al., 2016; Yuan et al., 2016). Along these lines, our study shows that the defect in mTOR-
We found that the defective mTOR signaling in TREM2-defi- mediated metabolic activation in TREM2-deficient cells can be
cient microglia is associated with a compensatory increase of corrected in vitro through the creatine kinase pathway or by trig-
autophagy in vitro and in vivo in AD. Reduced glycolysis and gering the Dectin-1 pathway, which transmits intracellular sig-
autophagy are known to attenuate inflammation (Netea-Maier nals similar to those of TREM2. Based on these results, we adop-
et al., 2016) and, indeed, microglia from 5XFAD mice lacking ted a therapeutic strategy based on the use of cyclocreatine, an
TREM2 weakly express inflammatory mediators in comparison analog of creatine that crosses membranes, enters the brain
to microglia from 5XFAD mice (Wang et al., 2015). Moreover, (Woznicki and Walker, 1979), can be phosphorylated and de-
autophagy may also enhance microglial clearance of Ab (Cho phosphorylated by creatine kinases (McLaughlin et al., 1972),
et al., 2014; Lucin et al., 2013; Shibuya et al., 2014), as it and can generate a supply of ATP (Kurosawa et al., 2012).
does in neurons (Hara et al., 2006; Komatsu et al., 2006; Remarkably, we found that administration of dietary cyclocrea-
Yang et al., 2011). However, a long-term defect in mTOR acti- tine throughout the progression of Ab accumulation improves
vation results in global microglial dysfunction, reduced cell microglia viability, numbers, and clustering around Ab plaques.
viability, and proliferation, as demonstrated by increased cas- As a result, plaques are denser and, most importantly, plaque-
pase-3 activation in microglia and by the previously reported associated neurite dystrophy is greatly reduced. Although cyclo-
increase in dying microglia around plaques in Trem2–/– 5XFAD creatine treatment was not sufficient to reduce the overall Ab
mice (Wang et al., 2015). Thus, while increased autophagy plaque accumulation, this may depend on time point chosen
may be beneficial in reducing inflammation and Ab load in the for analysis and/or cyclocreatine dosage and duration of treat-
short-term, a defect in mTOR signaling is detrimental and ment. While the creatine kinase pathway has been previously
severely impairs microglia fitness and capacity to respond to recognized to play an important role in the CNS in neuro-
Ab accumulation in the long-term. transmitter release, membrane potential maintenance, Ca2+

(C) Immunoblots performed on lysates of microglia sorted from the indicated genotype and treatment group of mice. Immunoblots for phosphorylated Akt (serine
473), NDRG1, total LC3, Akt, and actin.
(D) Quantification of the LC3II/LC3I ratio observed in immunoblots from 3 mice of each of the indicated genotypes and treatment groups.
(E) Average intensity of the plaques observed in the cortexes of mice from the indicated genotypes and treatment groups.
(F) Representative images depicting plaques (X04 in blue), nuclei (To-Pro3 in white), and N terminus APP (green) from the indicated mice and treatment groups.
Confocal images were taken as in Figure 1C.
(G) Quantification of the number of dystrophic neurites/plaque in the indicated mice and treatment group.
Error bar represents mean ± SEM. N.S. indicates not significant. *p < 0.05, and ****p < 0.001 by one-way ANOVA with Holm-Sidak’s multiple comparisons test
(A, C, D, and F) results pooled from 2 independent experiments representing a total of 5–8 male and female mice per treatment group.
See also Figure S5.

660 Cell 170, 649–663, August 10, 2017

homeostasis, and ion gradient restoration (Snow and Murphy, AUTHOR CONTRIBUTIONS
2001; Wyss and Kaddurah-Daouk, 2000), our results indicate
T.K.U., W.M.S., S.C.-C.H., J.D.U., A.K., E.L., Y.W., Y.Z., and W.L.B. performed
that this system may also be exploited for sustaining microglial
in vitro and in vivo studies. N.J.C. provided CERAD Reagan plaque scores.
metabolism. It should be noted that in certain settings cyclocrea- T.K.U., W.M.S., S.C.-C.H., A.S., and W.L.B. analyzed data. S.G. provided
tine can inhibit creatine kinase and can also have systemic and maintained all mouse lines. J.D.U. and D.M.H. provided human samples.
effects such as alteration in pancreatic hormones and glucose S.G., E.R.U., H.W.V., M.Ce., D.M.H., and M.N.A. provided critical insights
metabolism (Ara et al., 1998; Kuiper et al., 2008). Thus, it must and advice. M.C. supervised the research. T.K.U., W.M.S., and M.Co. pre-
be emphasized that our findings provide proof of principal pared the manuscript with input from all other authors.
studies and the use of cyclocreatine as a preventative treatment
for AD is not advisable. Future studies will be required to pre- ACKNOWLEDGMENTS
cisely define the mechanisms through which cyclocreatine
We would like to thank Dr. Edward Driggers at General Metabolomics for
impacts microglial responses to Ab. Additionally, it will be impor- providing metabolite analysis services. We also would like to thank Drs.
tant to determine whether cyclocreatine has any impact on Edward J. Pearce and David E. Sanin at the Max Planck Institute of Immunobi-
proteolytic shedding of TREM2 from microglia, which results ology and Epigenetics for discussions and experimental assistance. Finally,
in the release of soluble TREM2 with potential pro-survival we would like to thank Dr. Robert E. Schmidt for assistance in interpreting
functions. Altogether, our study indicates that strategies aimed TEM images. Confocal imaging was performed at the Washington
University Center for Cellular Imaging. This work was funded by the NIH
at sustaining microglial metabolism may be promising for
(RF1AG05148501 to M.Co., 5T32CA009547-30 to T.K.U., DK058177 to
therapeutic intervention in AD and other neurodegenerative E.R.U., and P01 AG03991 and P01 AG026276 to D.M.H.), Knight ADRC (P50
diseases linked to TREM2 deficiency and microglial dysfunction AG0568 1 to M.Co.), the Cure Alzheimer’s Fund (to M.Co.), the National
in general. Multiple Sclerosis Society (RG4687A1/1 to M.Ce.), and the Government of
Russian Federation (074-U01 to A.S. and A.A.L.). Y.W. is currently employed
by Eli Lilly and Company.
STAR+METHODS
Received: April 4, 2017
Detailed methods are provided in the online version of this paper Revised: May 26, 2017
and include the following: Accepted: July 14, 2017
Published: August 10, 2017
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING REFERENCES
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
Ara, G., Gravelin, L.M., Kaddurah-Daouk, R., and Teicher, B.A. (1998). Anti-
B Mice
tumor activity of creatine analogs produced by alterations in pancreatic hor-
B Human post-mortem samples
mones and glucose metabolism. In Vivo 12, 223–231.
B Cell lines and primary cells
Arase, H., Mocarski, E.S., Campbell, A.E., Hill, A.B., and Lanier, L.L. (2002).
d METHOD DETAILS Direct recognition of cytomegalovirus by activating and inhibitory NK cell re-
B Mice ceptors. Science 296, 1323–1326.
B Human post-mortem brain tissue Atagi, Y., Liu, C.C., Painter, M.M., Chen, X.F., Verbeeck, C., Zheng, H., Li, X.,
B Immunohistochemistry of human post-mortem brain Rademakers, R., Kang, S.S., Xu, H., et al. (2015). Apolipoprotein E is a ligand
tissue for triggering receptor expressed on myeloid cells 2 (TREM2). J. Biol. Chem.
B Cell culture and biochemical assays 290, 26043–26050.
B Microglia sorting Bailey, C.C., DeVaux, L.B., and Farzan, M. (2015). The triggering receptor ex-
B Immunoblotting pressed on myeloid cells 2 binds apolipoprotein E. J. Biol. Chem. 290,
26033–26042.
B Metabolite profiling by EIS-MS/MS
B RNA-seq analysis Cho, M.H., Cho, K., Kang, H.J., Jeon, E.Y., Kim, H.S., Kwon, H.J., Kim, H.M.,
Kim, D.H., and Yoon, S.Y. (2014). Autophagy in microglia degrades extracel-
B Network analysis
lular b-amyloid fibrils and regulates the NLRP3 inflammasome. Autophagy
B qRT-PCR 10, 1761–1775.
B Metabolism assays
Dambuza, I.M., and Brown, G.D. (2015). C-type lectins in immunity: recent de-
B Transmission electron microscopy velopments. Curr. Opin. Immunol. 32, 21–27.
B Preparation of brain samples and confocal microscopy
Ford, J.W., and McVicar, D.W. (2009). TREM and TREM-like receptors in
B Microglia clustering analysis inflammation and disease. Curr. Opin. Immunol. 21, 38–46.
B Reporter cell assay Freeman, L.C., and Ting, J.P. (2016). The pathogenic role of the inflammasome
B Quantification of methoxy-X04 coverage in neurodegenerative diseases. J. Neurochem. 136 (Suppl 1 ), 29–38.
B Plaque morphology analysis Galluzzi, L., Pietrocola, F., Levine, B., and Kroemer, G. (2014). Metabolic con-
d QUANTIFICATION AND STATISTICAL ANALYSIS trol of autophagy. Cell 159, 1263–1276.
d DATA AND SOFTWARE AVAILABILITY Gold, M., and El Khoury, J. (2015). b-amyloid, microglia, and the inflamma-
some in Alzheimer’s disease. Semin. Immunopathol. 37, 607–611.

SUPPLEMENTAL INFORMATION Guerreiro, R., and Hardy, J. (2014). Genetics of Alzheimer’s disease. Neuro-
therapeutics 11, 732–737.
Supplemental Information includes five figures and two tables and can be Hara, T., Nakamura, K., Matsui, M., Yamamoto, A., Nakahara, Y., Suzuki-Mi-
found with this article online at http://dx.doi.org/10.1016/j.cell.2017.07.023. gishima, R., Yokoyama, M., Mishima, K., Saito, I., Okano, H., and Mizushima,

Cell 170, 649–663, August 10, 2017 661

N. (2006). Suppression of basal autophagy in neural cells causes neurodegen- Neumann, H., and Takahashi, K. (2007). Essential role of the microglial trig-
erative disease in mice. Nature 441, 885–889. gering receptor expressed on myeloid cells-2 (TREM2) for central nervous
Hedrick, S.M., Matis, L.A., Hecht, T.T., Samelson, L.E., Longo, D.L., Heber- tissue immune homeostasis. J. Neuroimmunol. 184, 92–99.
Katz, E., and Schwartz, R.H. (1982). The fine specificity of antigen and Ia deter- O’Neill, L.A., and Pearce, E.J. (2016). Immunometabolism governs dendritic
minant recognition by T cell hybridoma clones specific for pigeon cytochrome cell and macrophage function. J. Exp. Med. 213, 15–23.
c. Cell 30, 141–152. Oakley, H., Cole, S.L., Logan, S., Maus, E., Shao, P., Craft, J., Guillozet-Bon-
Heneka, M.T., Kummer, M.P., Stutz, A., Delekate, A., Schwartz, S., Vieira- gaarts, A., Ohno, M., Disterhoft, J., Van Eldik, L., et al. (2006). Intraneuronal
Saecker, A., Griep, A., Axt, D., Remus, A., Tzeng, T.C., et al. (2013). NLRP3 beta-amyloid aggregates, neurodegeneration, and neuron loss in transgenic
is activated in Alzheimer’s disease and contributes to pathology in APP/PS1 mice with five familial Alzheimer’s disease mutations: potential factors in am-
mice. Nature 493, 674–678. yloid plaque formation. J. Neurosci. 26, 10129–10140.
Heneka, M.T., Golenbock, D.T., and Latz, E. (2015). Innate immunity in Alz- Orre, M., Kamphuis, W., Osborn, L.M., Jansen, A.H., Kooijman, L., Bossers, K.,
heimer’s disease. Nat. Immunol. 16, 229–236. and Hol, E.M. (2014). Isolation of glia from Alzheimer’s mice reveals inflamma-
Holtzman, D.M., Morris, J.C., and Goate, A.M. (2011). Alzheimer’s disease: the tion and dysfunction. Neurobiol. Aging 35, 2746–2760.
challenge of the second century. Sci. Transl. Med. 3, 77sr1. Peng, Q., Malhotra, S., Torchia, J.A., Kerr, W.G., Coggeshall, K.M., and Hum-
Hong, S., Beja-Glasser, V.F., Nfonoyim, B.M., Frouin, A., Li, S., Ramakrishnan, phrey, M.B. (2010). TREM2- and DAP12-dependent activation of PI3K requires
S., Merry, K.M., Shi, Q., Rosenthal, A., Barres, B.A., et al. (2016). Complement DAP10 and is inhibited by SHIP1. Sci. Signal. 3, ra38.
and microglia mediate early synapse loss in Alzheimer mouse models. Science Perry, V.H., and Holmes, C. (2014). Microglial priming in neurodegenerative
352, 712–716. disease. Nat. Rev. Neurol. 10, 217–224.
Huang, Y., and Mucke, L. (2012). Alzheimer mechanisms and therapeutic stra- Saxton, R.A., and Sabatini, D.M. (2017). mTOR signaling in growth, meta-
tegies. Cell 148, 1204–1222. bolism, and disease. Cell 168, 960–976.
Jay, T.R., Hirsch, A.M., Broihier, M.L., Miller, C.M., Neilson, L.E., Ransohoff, Sergushichev, A.A., Loboda, A.A., Jha, A.K., Vincent, E.E., Driggers, E.M.,
R.M., Lamb, B.T., and Landreth, G.E. (2017). Disease progression-dependent Jones, R.G., Pearce, E.J., and Artyomov, M.N. (2016). GAM: a web-service
effects of TREM2 deficiency in a mouse model of Alzheimer’s disease. for integrated transcriptional and metabolic network analysis. Nucleic Acids
J. Neurosci. 37, 637–647. Res. 44 (W1), W194–W200.
Kim, J., Kundu, M., Viollet, B., and Guan, K.L. (2011). AMPK and mTOR regu- Shibuya, Y., Chang, C.C., Huang, L.H., Bryleva, E.Y., and Chang, T.Y. (2014).
late autophagy through direct phosphorylation of Ulk1. Nat. Cell Biol. 13, Inhibiting ACAT1/SOAT1 in microglia stimulates autophagy-mediated lyso-
132–141. somal proteolysis and increases Ab1-42 clearance. J. Neurosci. 34,
Klionsky, D.J., Abdelmohsen, K., Abe, A., Abedin, M.J., Abeliovich, H., Ace- 14484–14501.
vedo Arozena, A., Adachi, H., Adams, C.M., Adams, P.D., Adeli, K., et al. Snow, R.J., and Murphy, R.M. (2001). Creatine and the creatine transporter: a
(2016). Guidelines for the use and interpretation of assays for monitoring auto- review. Mol. Cell. Biochem. 224, 169–181.
phagy (3rd edition). Autophagy 12, 1–222. Tanzi, R.E. (2012). The genetics of Alzheimer disease. Cold Spring Harb. Per-
Komatsu, M., Waguri, S., Chiba, T., Murata, S., Iwata, J., Tanida, I., Ueno, T., spect. Med. Published online July 14, 2017. http://dx.doi.org/10.1101/cshper-
Koike, M., Uchiyama, Y., Kominami, E., and Tanaka, K. (2006). Loss of auto- spect.a006296.
phagy in the central nervous system causes neurodegeneration in mice. Tejera, D., and Heneka, M.T. (2016). Microglia in Alzheimer’s disease: the
Nature 441, 880–884. good, the bad and the ugly. Curr. Alzheimer Res. 13, 370–380.
Kroemer, G., Mariño, G., and Levine, B. (2010). Autophagy and the integrated Turnbull, I.R., Gilfillan, S., Cella, M., Aoshi, T., Miller, M., Piccio, L., Hernandez,
stress response. Mol. Cell 40, 280–293. M., and Colonna, M. (2006). Cutting edge: TREM-2 attenuates macrophage
Kuiper, J.W., Pluk, H., Oerlemans, F., van Leeuwen, F.N., de Lange, F., Fran- activation. J. Immunol. 177, 3520–3524.
sen, J., and Wieringa, B. (2008). Creatine kinase-mediated ATP supply fuels Ulrich, J.D., Finn, M.B., Wang, Y., Shen, A., Mahan, T.E., Jiang, H., Stewart,
actin-based events in phagocytosis. PLoS Biol. 6, e51. F.R., Piccio, L., Colonna, M., and Holtzman, D.M. (2014). Altered microglial
Kurosawa, Y., Degrauw, T.J., Lindquist, D.M., Blanco, V.M., Pyne-Geithman, response to Ab plaques in APPPS1-21 mice heterozygous for TREM2. Mol.
G.J., Daikoku, T., Chambers, J.B., Benoit, S.C., and Clark, J.F. (2012). Cyclo- Neurodegener. 9, 20.
creatine treatment improves cognition in mice with creatine transporter defi- Vincent, E.E., Sergushichev, A., Griss, T., Gingras, M.C., Samborska, B.,
ciency. J. Clin. Invest. 122, 2837–2846. Ntimbane, T., Coelho, P.P., Blagih, J., Raissi, T.C., Choinière, L., et al.
Laplante, M., and Sabatini, D.M. (2012). mTOR signaling in growth control and (2015). Mitochondrial phosphoenolpyruvate carboxykinase regulates meta-
disease. Cell 149, 274–293. bolic adaptation and enables glucose-independent tumor growth. Mol. Cell
Lucin, K.M., O’Brien, C.E., Bieri, G., Czirr, E., Mosher, K.I., Abbey, R.J., Mas- 60, 195–207.
troeni, D.F., Rogers, J., Spencer, B., Masliah, E., and Wyss-Coray, T. (2013). Walker, J.B. (1979). Creatine: biosynthesis, regulation, and function. Adv.
Microglial beclin 1 regulates retromer trafficking and phagocytosis and is Enzymol. Relat. Areas Mol. Biol. 50, 177–242.
impaired in Alzheimer’s disease. Neuron 79, 873–886. Wang, Y., Cella, M., Mallinson, K., Ulrich, J.D., Young, K.L., Robinette, M.L.,
Masliah, E., Sisk, A., Mallory, M., Mucke, L., Schenk, D., and Games, D. (1996). Gilfillan, S., Krishnan, G.M., Sudhakar, S., Zinselmeyer, B.H., et al. (2015).
Comparison of neurodegenerative pathology in transgenic mice overexpress- TREM2 lipid sensing sustains the microglial response in an Alzheimer’s dis-
ing V717F beta-amyloid precursor protein and Alzheimer’s disease. ease model. Cell 160, 1061–1071.
J. Neurosci. 16, 5795–5811. Wang, Y., Ulland, T.K., Ulrich, J.D., Song, W., Tzaferis, J.A., Hole, J.T., Yuan,
McLaughlin, A.C., Cohn, M., and Kenyon, G.L. (1972). Specificity of creatine P., Mahan, T.E., Shi, Y., Gilfillan, S., et al. (2016). TREM2-mediated early micro-
kinase for guanidino substrates. Kinetic and proton nuclear magnetic relaxa- glial response limits diffusion and toxicity of amyloid plaques. J. Exp. Med.
tion rate studies. J. Biol. Chem. 247, 4382–4388. 213, 667–675.
Meyer-Luehmann, M., and Prinz, M. (2015). Myeloid cells in Alzheimer’s dis- Werner, M., Hobeika, E., and Jumaa, H. (2010). Role of PI3K in the generation
ease: culprits, victims or innocent bystanders? Trends Neurosci. 38, and survival of B cells. Immunol. Rev. 237, 55–71.
659–668. Woznicki, D.T., and Walker, J.B. (1979). Formation of a supplemental long
Netea-Maier, R.T., Plantinga, T.S., van de Veerdonk, F.L., Smit, J.W., and Ne- time-constant reservoir of high energy phosphate by brain in vivo and
tea, M.G. (2016). Modulation of inflammation by autophagy: Consequences for in vitro and its reversible depletion by potassium depolarization.
human disease. Autophagy 12, 245–260. J. Neurochem. 33, 75–80.

662 Cell 170, 649–663, August 10, 2017

Wyss, M., and Kaddurah-Daouk, R. (2000). Creatine and creatinine meta- Yeh, F.L., Wang, Y., Tom, I., Gonzalez, L.C., and Sheng, M. (2016). TREM2
bolism. Physiol. Rev. 80, 1107–1213. binds to apolipoproteins, including APOE and CLU/APOJ, and thereby facili-
Yang, D.S., Stavrides, P., Mohan, P.S., Kaushik, S., Kumar, A., Ohno, M., tates uptake of amyloid-beta by microglia. Neuron 91, 328–340.
Schmidt, S.D., Wesson, D., Bandyopadhyay, U., Jiang, Y., et al. (2011). Yuan, P., Condello, C., Keene, C.D., Wang, Y., Bird, T.D., Paul, S.M., Luo, W.,
Reversal of autophagy dysfunction in the TgCRND8 mouse model of Alz- Colonna, M., Baddeley, D., and Grutzendler, J. (2016). TREM2 haplodeficiency
heimer’s disease ameliorates amyloid pathologies and memory deficits. Brain in mice and humans impairs the microglia barrier function leading to decreased
134, 258–277. amyloid compaction and severe axonal dystrophy. Neuron 90, 724–739.

Cell 170, 649–663, August 10, 2017 663

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Antibodies
anti-Pan Actin Cell Signaling Technologies Cat# 4968; RRID: AB_2313904
anti-p62 Cell Signaling Technologies Cat# 8025; RRID: AB_10859911
anti-phospho Akt 473 Cell Signaling Technologies Cat# 9271; RRID: AB_329825
anti-phospho Ulk1 317 Cell Signaling Technologies Cat# 12753
anti-phospho Ulk1 757 Cell Signaling Technologies Cat# 6888; RRID: AB_10829226
anti-phospho NDRG1 Cell Signaling Technologies Cat# 3217
anti-phospho S6K Cell Signaling Technologies Cat# 2710
anti-phospho 4EBP1 Cell Signaling Technologies Cat# 2855; RRID: AB_560835
anti-Akt Cell Signaling Technologies Cat# 9272; RRID: AB_329827
anti-S6K Cell Signaling Technologies Cat# 9202; RRID: AB_331676
anti-phospho AMPKa Cell Signaling Technologies Cat# 2535; RRID: AB_331250
anti-AMPKa Cell Signaling Technologies Cat# 5832
anti-LC3 Cell Signaling Technologies Cat# 4108; RRID: AB_2137703
anti-phospho mTOR 2448 Cell Signaling Technologies Cat# 2971; RRID: AB_330970
anti-Spp1 R and D Systems Cat# AF808; RRID: AB_2194992
anti-Xbp-1 s BD Biosciences Cat# 562642
anti-Iba1 Wako Cat# 019-19741; RRID: AB_839504
anti-cleaved caspase-3 Alexa 647 Cell Signaling Technologies Cat# 9602
anti-LC3 488 Cell Signaling Technologies Cat# 13082
anti-APP Millipore Cat# MAB348; RRID: AB_94882
anti-CD45 BV421 Biolegend Cat# 103134; RRID: AB_2562559
anti-CD11b FITC eBioscience Cat# 11-0112
anti-F4/80 APC eBioscience Cat# 17-4801
anti-Trem2 Colonna Lab M1178 doi: 10.1016/j.cell.2015.01.049.
anti-rabbit DyLight 549 Vector Laboratories Cat# DI-1549; RRID: AB_2336407
Bacterial and Virus Strains
Murine Trem2 expressed in pMXs- Cell Biolabs https://www.cellbiolabs.com/pmxs-ires-
IRES-Puro Retroviral Expression Vector puro-retroviral-expression-vector
Biological Samples
Human brain tissue for Alzheimer’s Knight Alzheimer’s Disease Research Protocol numbers: Healthy Aging and Senile
disease patients (TREM2CV/CV (8), Center at Washington University Dementia (HASD) Morphology Core: 89-0555
TREM2CV/R47H (7), TREM2CV/R62H (4)) and Program Project: Alzheimer’s Disease
Research Center (ADRC): 89-0556.
Chemicals, Peptides, and Recombinant Proteins
Cyclocreatine Santa Cruz Biotechnology Cat# SC-217964 S
Methoxy X04 Tocris Biosciences Cat# 4920
Mitotraker green Invitrogen Cat# M7514
Wortmannin Millipore Cat# 12-338
Ly294002 Calbiochem Cat# 440202
Tunicamycin Sigma-Aldrich Cat# 654380
Bafilomycin A1 from Streptomyces Sigma-Aldrich Cat# B1793-10UG
griseus
ToPro3 Iodide Life Technologies Cat# T3605
Zymosan, depleted Invivogen Cat# tlrl-zyd
(Continued on next page)

e1 Cell 170, 649–663.e1–e6, August 10, 2017

Continued
REAGENT or RESOURCE SOURCE IDENTIFIER
Oligomycin Cayman Chemical Cat# 11341
FCCP Cayman Chemical Cat# 370865
Critical Commercial Assays
ATP Assay Invitrogen Cat# A22066
Neural Tissue Dissociation Kit (T) Miltenyi Biotech Cat# 130-093-231
Deposited Data
d GEO/GSE65067 Microarray data N/A
GEO/GSE98563 RNA-seq data N/A
Experimental Models: Cell Lines
2B4 cells retrovirally transduced with Generated in the laboratory of doi: 10.1016/j.cell.2015.01.049.
an NFAT-GFP reporter and retrovirally Dr. Marco Colonna.
transduced with TREM2
2B4 cells expressing NFAT-GFP Generated in the laboratory of doi: 10.1126/science.1070884
Dr. Lewis Lanier
Experimental Models: Organisms/Strains
Mouse: C57BL/6J WT The Jackson Laboratory Cat# 000664
Mouse: 5XFAD/Tg6799 The Jackson Laboratory Cat# 34840-JAX
Mouse:Trem2/ Generated in the Laboratory of N/A
Dr. Marco Colonna
Oligonucleotides
For nuclear DNA FW: CTTCCCCACTGGCCTCAAG N/A
RV: CCAAAACCCAGTGATCCAGC
For mitochondrial DNA FW: TGCTAGCCGCAGGCATTAC N/A
RV: GGGTGCCCAAAGAATCAGAAC
For beta-Actin (RNaseq) FW: GGA GGG GGT TGA GGT GTT N/A
RV: TGT GCA CTT TTA TTG GTC
TCA AG
Linker primers for RNaseq CCTACACGACGCTCTTCCGATCT- N/A
XXXXXXXX-T15
AGATCGGAAGAGCACACGTCTG
Software and Algorithms
MATLAB MathWorks N/A
Morpheus Broad Institute N/A
Prism 7 Graphpad N/A
Fiji 2.0 ImageJ N/A
Imaris 7.7 Bitplane N/A
IPA QIAGEN N/A
Shiny GAM https://artyomovlab.wustl.edu/ N/A
shiny/gam/
Other
XF96 Extracellular Flux Analyzer Agilent N/A
Nikon A1Rsi Confocal Microscope Nikon N/A

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Marco
Colonna ([email protected]).

Cell 170, 649–663.e1–e6, August 10, 2017 e2

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice
Mice were of mixed sexes. Mice within experiments were age and sex matched. For studies using 5XFAD and Trem2–/– 5XFAD
animals all animals were 8 months of age at the time of use. For bone marrow and primary microglia mice were used from 6 weeks
of age until 12 weeks of age. Mice used in this study include WT C57BL/6J, 5XFAD, Trem2–/–, and Trem2–/– 5XFAD animals. All
animals were backcrossed until at least > 98% C57BL/6J confirmed by genotype wide microsatellite typing. Mice were housed under
specific pathogen free conditions. Mice from different genotypes were cohoused. Mice did not undergo any procedures prior to their
stated use. For cyclocreatine treatment mixed litters of sex matched mice were randomly assigned to experimental groups. All
studies performed on mice were done in accordance with the Institutional Animal Care and Use Committee at Washington University
in St. Louis approved all protocols used in this study.

Human post-mortem samples

Characteristics of donors of human post-mortem brain tissue at the time of collection is indicated in Table S1. Samples from 7 R47H,
4 R62H, and 8 case matched AD patients were examined. Samples were obtained from the Knight Alzheimer’s Disease Research
Center at Washington University. Protocol numbers: Healthy Aging and Senile Dementia (HASD) Morphology Core: 89-0555 and Pro-
gram Project: Alzheimer’s Disease Research Center (ADRC): 89-0556.

Cell lines and primary cells

Bone marrow derived macrophages and microglia were prepared from sex and age matched mice. To prepare bone marrow-derived
macrophages, femurs and tibias were removed and flushed with PBS. Cells were counted and plated at 2.5 3 106 cells/100 mm petri
dish in RPMI supplemented with Glutamax, penicillin/streptomycin, non-essential amino acids, pyruvate, and 10% heat inactivated
fetal bovine serum (complete RPMI) and 10% L-cell conditioned medium (LCCM). Cells were cultured for 4-5 days before use. Micro-
glia were prepared as previously described (Wang et al., 2015). Briefly, brains were dissociated by using a Neural Tissue Dissociation
Kit (T) (Miltenyi Biotech Cat. Number 130-093-231). Cells suspensions were labeled with anti-mouse CD45 magnetic beads and
isolated on LS columns (Miltenyi Biotic). Cells were plated onto poly-L-lysine coated polystyrene plates in complete RPMI supple-
mented with 20% LCCM and 10 ng/ml human TGF-b. Media was changed on day 3 post plating and cells were used 5-7 days post
plating.
TREM2 reporter cells were maintained in 10% FBS in RPMI 1640 supplemented with sodium pyruvate, GlutaMAX, and penicillin/
streptomycin. TREM2 reporter cells were based on the 2B4 NFAT-GFP cells developed by Arase et al. (Arase et al., 2002). The sex of
the mouse from which 2B4 T cell hybridoma cells were derived has not been reported (Arase et al., 2002; Hedrick et al., 1982).

METHOD DETAILS

Mice
The generation of Trem2–/– and Trem2–/– 5XFAD mice has been described previously (Oakley et al., 2006; Turnbull et al., 2006; Wang
et al., 2015). All mice were on a C57BL/6 background. Age and sex matched mice were used for all experiments; experimental
cohorts of mice were cohoused from birth to control for the microbiota. For in vivo cyclocreatine treatment 10-week old mice
were put on cyclocreatine-containing water, treatment was continued until mice reached 8 months of age (Santa Cruz
SC-217964 S). Desired intake of cyclocreatine was approximately 0.28 mg/g of body weight/day, which is approximately the
same as the standard creatine dose used in humans of 285 mg/kg of body weight/day (Kurosawa et al., 2012). Cyclocreatine was
administered in drinking water at a final concentration of 2.33 mg/ml. The Institutional Animal Care and Use Committee at Washington
University in St. Louis approved all protocols used in this study.

Human post-mortem brain tissue

Paraffin-embedded sections (8 um) from the frontal cortex of individuals carrying the common variant (CV) of TREM2 (8) or hetero-
zygous for the CV and either R47H (7), R62H (4) were obtained from the Knight Alzheimer’s Disease Research Center at Washington
University. Protocol numbers: Healthy Aging and Senile Dementia (HASD) Morphology Core: 89-0555 and Program Project:
Alzheimer’s Disease Research Center (ADRC): 89-0556. R47H and R62H carriers were case matched for age, gender, and
CERAD-Reagan plaque score to CV TREM2 control individuals. Detailed demographic characteristics are provided in Table S1.

Immunohistochemistry of human post-mortem brain tissue

Brain sections were deparaffinized with xylene and rehydrated with decreasing concentrations of ethanol. Antigen retrieval was
performed by incubating sections for 20 min in a 95 C citrate buffer bath (10 mM sodium citrate, 0.05% Tween-20, pH 6.0) prior
to staining. Sections were blocked in 3% goat serum in PBS for 30 min at room temperature (RT) followed by incubation with rabbit
anti-Iba1 (1:250, Wako; catalog no. 019-19741) overnight at 4 C. Sections were washed in PBS and incubated for 1 hr at room
temperature (RT) with methoxy-XO4 (20 mg/ml) (Tocris Bioscience #4920) and anti-rabbit DyLight 549 (Vector Laboratories
DI-1549). Sections were washed and incubated overnight in anti-LC3A/B Alexa Fluor 488 (Cell Signaling Technologies #13082).

e3 Cell 170, 649–663.e1–e6, August 10, 2017

Sections were washed and mounted using Fluoromount G (SouthernBiotech #0100-01) and images were collected using a Nikon
A1Rsi+ confocal microscope. Images were then processed with Imaris 7.7 (Bitplane).

Cell culture and biochemical assays

To prepare bone marrow-derived macrophages, femurs and tibias were removed and flushed with PBS. Cells were counted and
plated at 2.5 3 106 cells/100 mm petri dish in RPMI supplemented with Glutamax, penicillin/streptomycin, non-essential amino acids,
pyruvate, and 10% heat inactivated fetal bovine serum (complete RPMI) and 10% L-cell conditioned medium (LCCM). Cells were
cultured for 4-5 days before use. Microglia were prepared as previously described (Wang et al., 2015). Briefly, brains were dissoci-
ated by using a Neural Tissue Dissociation Kit (T) (Miltenyi Biotech Cat. Number 130-093-231). Cell suspensions were labeled with
anti-mouse CD45 magnetic beads and isolated on LS columns (Miltenyi Biotic). Cells were plated onto poly-L-lysine coated polysty-
rene plates in complete RPMI supplemented with 20% LCCM and 10 ng/ml human TGF-b. Media was changed on day 3 post plating
and cells were used 5-7 days post plating. ATP concentrations were determined with an ATP Determination Kit (Invitrogen).

Microglia sorting
Microglia were isolated from the indicated animals as previously described (Wang et al., 2015). CD45+, CD11b+, F4/80+ (Biolegend
Cat. Number 103134, eBioscience Cat. Numbers 11-0112 and 17-4801) cells in the brain were fluorescence-activated cell-sorted
(FACS) directly into RLT-plus lysis buffer for microarray or 2% FBS in PBS for TEM or immunoblot lysates. For microarray RNA extrac-
tion was performed using a RNeasy micro kit (QIAGEN). Microarray hybridization (Affymetrix MoGene 1.0 ST array) and data process-
ing were performed at the Washington University Genome Center. For normalization, raw data was processed by Robust Multi-Array
(RMA) method and genes were pre-filtered for expression value R 120 expression units, a cut-off above which genes have a 95%
chance of expression demonstrated in Immgen dataset, which uses the same array platform (Wang et al., 2015). QIAGEN IPA anal-
ysis was preformed by comparing fold change and p values for all genes 5XFAD and Trem2-deficient 5XFAD microglia. Heatmaps
and hierarchical clustering were generated from preselected gene-lists using Morpheus. Microarray data has been deposited at
GEO: GSE65067.

Immunoblotting
BMDM or microglia were lysed in RIPA buffer (50 mM Tris, 150mM NaCl, 1% SDS, and 1% Triton X-100) containing PMSF, leupeptin,
activated sodium orthovanidate, apoprotinin, and phosphatase inhibitor cocktail 3 (Sigma Aldrich Cat. Number P0044). Lysates were
flash frozen on dry ice and stored at 80 C until use. Lysates were thawed and 4x LDS running buffer and 10% b-mercaptoethanol
were added. Lysates were heated to 95 C for 10 min and run on either a 15% polyacrylamide gel with a 4% stacking gel, a 12% bis-
tris gel (Nupage), or a 4%–12% bis-tris gel (Nupage). Proteins were transferred to nitrocellulose and blocked for 1 hr at RT in 5% milk
in Tris buffered saline with 0.05% Tween 20 (TBST). Membranes were incubated in primary antibody overnight at 4 C (Table S2).
Membranes were subsequently washed and incubated in Lienco anti-rabbit HRP for 1 hr at RT, washed, and developed using either
SuperSignal West Pico Chemiluminescent Substrate or a combination of SuperSignal West Pico Chemiluminescent Substrate and
SuperSignal West Femto Chemiluminescent Substrate.

Metabolite profiling by EIS-MS/MS

BMDMs were cultured in either 0.5% or 10% LCCM overnight in complete RPMI. Polar metabolites were extracted according to
General Metabolics protocol for extraction of polar metabolites from adherent mammalian cell culture. Briefly, cells were washed
in pre-warmed 75mM ammonium carbonate in water. Metabolites were extracted by addition of 70 C 70% ethanol for 3 min. Ethanol
was removed and plates were washed with additional 70 C 70% ethanol. Debris was pelleted by spinning at 14,000 rpm in a tabletop
microcentrifuge for 10 min at 4 C. Extracts were moved to a fresh tube and shipped to General Metabolics for assessment by
EIS-MS/MS. Differential expression analysis was done using limma.

RNA-seq analysis
Cells were cultured as described in the metabolite profiling by EIS-MS/MS section above. mRNA was extracted from cell lysates
using oligo-dT beads (Invitrogen). For cDNA synthesis, we used custom oligo-dT primer with a barcode and adaptor-linker sequence
(CCTACACGACGCTCTTCCGATCT-XXXXXXXX-T15). After first-strand synthesis, samples were pooled together based on Actb
qPCR values and RNA-DNA hybrids were degraded with consecutive acid-alkali treatment. Subsequently, a second sequencing
linker (AGATCGGAAGAGCACACGTCTG) was ligated with T4 ligase (NEB) followed by SPRI-beads (Agencourt AMPure XP,
BeckmanCoulter) clean-up. The mixture was enriched by PCR for 12 cycles and SPRI-beads (Agencourt AMPure XP,
BeckmanCoulter) purified to yield final strand-specific RNA-seq libraries. Libraries were sequenced using a HiSeq 2500 (Illumina)
using 50 bp x 25 bp pair-end sequencing. Second read (read-mate) was used for sample demultiplexing. Reads were aligned to
the GRCm38.p2 assembly of mouse genome using STAR aligner. Aligned reads were quantified using quant3p script (https://
github.com/ctlab/quant3p) to account for specifics of 30 sequencing. RefSeq genome annotation was used and DESeq2 was
used for differential gene expression analysis. RNaseq data has been deposited at GEO:GSE98563.

Cell 170, 649–663.e1–e6, August 10, 2017 e4

Network analysis
Network analysis was performed as previously described utilizing Shiny GAM (https://artyomovlab.wustl.edu/shiny/gam/) (Sergush-
ichev et al., 2016; Vincent et al., 2015). We considered a network of chemical mappings between carbon atoms in substrates and
products for all annotated reactions in KEGG database using RPAIRs entries. The scores for nodes and edges were assigned
according to log(p value), such that highly significant gene or metabolite signals had positive scores and not significant had negative
scores. Using an exact solver (https://github.com/ctlab/sgmwcs-solver) we found a module with a maximal weight, with counting
positive scores maximum once for a measured entity (a mass-spectrometry peak or a gene). For clarity, addition edges between
nodes in the module were added if the corresponding gene was highly expressed (was in a top 3000 expressed genes).

qRT-PCR
Total RNA was isolated with TRIzol Reagent (Invitrogen) and single-strand cDNA was synthesized with a High Capacity cDNA
Reverse Transcription Kit (Applied Biosystems). Genomic DNA was extracted using the QIAamp DNA micro kit (QIAGEN) to deter-
mine mtDNA/nDNA ratios. Real-time PCR was performed using SYBR Green real-time PCR master mix (Thermo-Fisher) and
LightCycler 96 detection system (Roche). mtDNA primers were to cytochrome c oxidase subunit 1 and nDNA primers were to
NADH:ubiquinone oxidoreductase core subunit V1.

Metabolism assays
For real-time analysis of extracellular acidification rates (ECAR) macrophages were analyzed using an XF96 Extracellular Flux
Analyzer (Agilent Technologies). Cells were incubated overnight in complete RPMI in the indicated concentration of LCCM with or
without cyclocreatine (10 mM). Measurements were taken under basal conditions and following the sequential addition of 1 mM
oligomycin and 1.5 mM fluoro-carbonyl cyanide phenylhydrazone (FCCP) (purchased from Sigma-Aldrich).

Transmission electron microscopy

For ultrastructural analyses, cells were fixed in 2% paraformaldehyde/2.5% glutaraldehyde in 100 mM sodium cacodylate buffer, pH
7.2 for 1 hr at RT (Polysciences Inc., Warrington, PA). Samples were washed in sodium cacodylate buffer and postfixed in 1% osmium
tetroxide for 1 hr (Polysciences Inc.). Samples were then rinsed extensively in deionized water prior to en bloc staining with 1%
aqueous uranyl acetate for 1 hr (Ted Pella Inc., Redding, CA). Following several rinses in dH20, samples were dehydrated in a graded
series of ethanol and embedded in Eponate 12 resin (Ted Pella Inc.). Sections of 95 nm were cut with a Leica Ultracut UCT ultrami-
crotome (Leica Microsystems Inc., Bannockburn, IL), stained with uranyl acetate and lead citrate, and viewed on a JEOL 1200 EX
transmission electron microscope (JEOL USA Inc., Peabody, MA) equipped with an AMT 8 megapixel digital camera and AMT Image
Capture Engine V602 software (Advanced Microscopy Techniques, Woburn, MA).
For quantitation of multivesicular/multilamellar structures, 30 cells that were cross-sectioned through the nucleus (indicating
cross-section through the middle of cell) were randomly chosen, and images of each cell were taken at 6,000X and 20,000X magni-
fication. The cross-sectional area of each of the multivesicular structures were determined using ImageJ 1.38 g (National Institutes of
Health, USA, customized for AMT images). Data is expressed as the 1) total number of a multivesicular/multilammelar structures per
cross-sectional area of cytosol and 2) the total cross-sectional area of multivesicular/multilamellar structures per area of cytosol.

Preparation of brain samples and confocal microscopy

Confocal microscopy analysis was preformed as previously described (Wang et al., 2015). Briefly, mice were anesthetized with
ketamine/xylazine and perfused with ice-cold PBS containing 1 U/ml of heparin. Brains were fixed in 4% PFA overnight at 4 C rinsed
in PBS and incubated overnight at 4 C in 30% sucrose before freezing in a 2:1 mixture of 30% sucrose and optimal cutting temper-
ature compound. Serial 40 mm coronal sections were cut on a cryo-sliding microtome. Floating sections from 1.1 mm Bregma to
0.8 mm Bregma for cortical imaging or slides with fixed human sections were stained with Iba-1 (Waco Chemicals Cat. Number
019-19741) overnight at 4 C followed by staining with anti-rabbit IgG DyLight 549 (Vector Laboratories Cat. Number DI-1549) and
methoxy-X04 (Tocris Cat. Number 4920) for 1 hr at RT. Finally, sections were stained for with anti-LC3 Alexa 488 ± anti-cleaved
caspase 3 (Cell Signaling Technologies Cat. Number 13082 and 9602). Images were collected using a Nikon A1Rsi+ confocal micro-
scope. Images were then processed with Imaris 7.7 (Bitplane).

Microglia clustering analysis

Positions of microglia and positions and volumes of plaques within z stacks were derived from analysis in Imaris, and
microglia-plaque association was determined using automated scripts in MATLAB. Briefly, each plaque in the z stack was modeled
as an idealized sphere with the same volume and center of mass. Microglia density within 15 mm of the plaque surface was deter-
mined by isolating the voxels of the image that fall within 15 mm of the edge of the idealized plaque. The number of microglia contained
in these voxels was divided by the total volume of those voxels to obtain density for a single plaque. Densities of all plaques in a
z stack were averaged together, and the resulting values were averaged together for all z stacks corresponding to a single animal.

e5 Cell 170, 649–663.e1–e6, August 10, 2017

Reporter cell assay
The 2B4 T cell hybridoma cell line was retrovirally transduced with an NFAT-GFP reporter construct, and TREM2 reporter cells were
generated by a second retroviral transduction with a TREM2 overexpression construct and selected by puromycin resistance, as
previously described (Wang et al., 2015). Cells were cultured routinely in complete media (10% FBS in RPMI-1640 supplemented
with sodium pyruvate, GlutaMAX, and penicillin/streptomycin). For serum starvation, cells were plated at a density of 25,000
cells/well in a 96-well plate in either complete media or RPMI-1640 in the presence of 20% anti-TREM2 hybridoma supernatant (clone
M178, generated in house) or 20% isotype control hybridoma supernatant (Wang et al., 2015). After 16 hr, the percent of GFP+ cells
among live cells was measured by flow cytometry.

Quantification of methoxy-X04 coverage

To measure total plaque area, brain sections were stained with methoxy X04. Images were collected using a Nikon Eclipse 80i
microscope. For quantitative analysis, images were converted to 8-bit greyscale and stitched using the ‘‘Stitching’’ plugin in ImageJ.
Cortex (
1.1 mm Bregma to 0.8 mm Bregma) and hippocampus (
-1.7 Bregma to –2.4 Bregma) were determined by manual selec-
tion. The threshold of selected images were set at 1.5x mean intensity of the selected area to highlight plaques and analyzed using the
‘‘Measure’’ function in ImageJ to calculate the percent area covered. Identified objects after thresholding were individually inspected
to confirm the object as a plaque or not. Two brain sections per mouse were used for quantification. The average of two sections was
used to represent a plaque load for each mouse.

Plaque morphology analysis

Methoxy-X04-stained sections were imaged by confocal microscopy using a 60X objective and 1.5X digital zoom in the cortex at

1.1 mm Bregma to 0.8 mm Bregma. Z images were taken at 1.2 mm intervals. 20-30 mm z stacks were z-projected by maximum
intensity projection and individual plaques were selected in ImageJ. Each individual plaque was traced using a combination of thresh-
olds and edge detection and smoothened using image erosion. The average intensity was determined by averaging values of pixels
within the plaque trace. The shape index was calculated as 4p*(perimeter pixels)2/(all pixels).

QUANTIFICATION AND STATISTICAL ANALYSIS

Data in figures are presented as mean ± SEM. Unless otherwise stated statistical analysis was performed using Prism (GraphPad).
Quantification of confocal microscopy, immunoblots, and electron microscopy images were performed using Imaris, ImageJ,
MATLAB, and FIJI. Differential metabolite expression was analyzed using limma. Pathway analysis of microarray data was performed
using IPA software. RNaseq analysis was performed by using Second read (read-mate) for sample demultiplexing. Reads were
aligned using STAR aligner and quantified using quant3p scirpt. RefSeq genome annotation was used and DESeq2 was used for
differential gene expression analysis. Combined RNaseq and metabolite network analysis was performed utilizing Shiny GAM. Sta-
tistical analysis to compare the mean values for multiple groups was performed using Prism by one-way ANOVA with Holm-Sidak’s
multiple comparisons test. Comparison of two groups was performed in Prism using a two-tailed unpaired t test (Mann Whitney).
Values were accepted as significant if p % 0.05. Intragroup variation compared between groups was similar in all experiments.

DATA AND SOFTWARE AVAILABILITY

Microarray data has been deposited at GEO: GSE65067.

RNA-seq data has been deposited at GEO: GSE98563.

Cell 170, 649–663.e1–e6, August 10, 2017 e6

Supplemental Figures

Figure S1. Wider Field of View of LC3 in Microglia, Related to Figure 1

Lower magnification confocal images of cortexes of WT, Trem2–/–, 5XFAD and Trem2–/–5XFAD mice. Confocal images are representative of 3 female mice
per group.
(legend on next page)
Figure S2. TREM2 Deficiency in an AD Model Affects Microglia Expression of Genes Involved in Metabolic Pathways, and Microglia from
Trem2–/– 5XFAD Mice Undergo More Cell Death, Related to Figure 2
(A) Sqstm1 expression taken from microarrays of sorted microglia from 8 month old WT, Trem2–/–, 5XFAD and Trem2–/– 5XFAD mice.
(B) Top 10 pathways in IPA analysis of differentially expressed genes from 5XFAD and Trem2–/– 5XFAD microglia. Negative log10 p values are shown.
(C) Gene enrichment plots for genes included in the eIF2, glycolysis, and mTOR signaling modules of IPA. Plots were generated utilizing gene-set enrichment
analysis (GSEA) software.
(D–F) Heatmaps comparing WT, 5XFAD, Trem2–/–, and Trem2–/– 5XFAD microglia for expression of genes included in the eIF2 (D), glycolysis (E) and mTOR
signaling (F) pathways.
(G) Illustration of the glycolytic pathway: proteins indicated in red correspond to genes upregulated in 5XFAD but not Trem2–/– 5XFAD microglia compared to WT
microglia.
(H) Mosaic of images depicting representative images of microglia (Iba1 red), plaques (methoxy-X04 blue), cleaved caspase-3 (aqua), LC3 (green), and total
merged images from a 5XFAD and a Trem2–/– 5XFAD animal.
(I) Quantification of the percentage of microglia that are both LC3 and cleaved caspase-3 positive.
Microarray data represents analyses of microglia sorted from 3 WT, 4 Trem2–/–, 5 5XFAD, and 5 Trem2–/– 5XFAD mice (A-G). Confocal images are representative
of 3 female mice per group (H). Error bar represents mean ± SEM. ***p < 0.005 by One-way ANOVA with Holm-Sidak’s multiple comparisons test (I).
Figure S3. Trem2 Is Activated In Vitro and Contributes to PI3K-Dependent mTOR Activation, Related to Figure 3
(A) Immunoblots for phosphorylated Akt (serine 473), NDRG1, S6K, 4EBP1, mTOR, total S6K, Akt, and actin performed on lysates from WT and Trem2–/– BMDMs
cultured overnight in 10% or 0.5% LCCM followed by the addition of Ly294002 for 3 hr prior to harvest.
(B) Reporter cell assay assessing TREM2 activation in reporter cell line incubated at optimal and low serum conditions with or without soluble anti-TREM2.
Error bar represents mean ± SEM. N.S. indicates not significant and ****p < 0.001 by One-way ANOVA with Holm-Sidak’s multiple comparisons test (B). Results
are representative of at least 3 independent experiments.
Figure S4. Macrophages and Microglia from Trem2–/– Mice Are Less Energetically Active, Related to Figure 4
(A and B) Heatmaps representing the total metabolic profiles of WT compared to Trem2–/– BMDMs culture overnight in 10% LCCM (A) or 0.5% LCCM (B).
(C) Shiny-GAM output for network analysis combining RNA-seq and mass spectrometry data highlights differences between WT and Trem2–/– BMDM cultured
overnight in 0.5% LCCM. Enzymes-encoding mRNAs and metabolites that are downregulated or upregulated in Trem2–/– cells versus WT cells are represented by
green or red nodes and connectors, respectively.

(legend continued on next page)

(D) Quantification of phosphocreatine from WT and Trem2–/– BMDMs cultured in the indicated concentration of LCCM.
(E and F) Assessment and quantification of the mitochondrial mass of resident peritoneal macrophages from WT and Trem2–/– mice by MitoTraker Green
incorporation.
(G and H) Assessment and quantification of the mitochondrial mass of thioglycolate elicited peritoneal macrophages from WT and Trem2–/– mice by MitoTracker
Green incorporation.
(I) ATP content of WT and Trem2–/– microglia cultured in 10% LCCM overnight.
(J) Mitochondrial content of WT and Trem2–/– microglia assessed by the ratio of mitochondrial-to-nuclear DNA.
(K) Extracellular acidification rate (ECAR) of WT and Trem2–/– microglia cultured overnight in 10% LCCM.
(L) Primary WT and Trem2–/– microglia were incubated overnight in the indicated concentration of LCCM. Immunoblots for LC3, phosphoserine 757 Ulk1,
phosphoserine 317 Ulk1, phosphoserine 473 Akt, phospho-NDRG1, and b actin were performed.
(M) Quantification of the LC3II/LC3I ratio from immunoblot preformed on primary microglia shown in L.
Error bar represents mean ± SEM. *p < 0.05 or ****p < 0.001 by One-way ANOVA with Holm-Sidak’s multiple comparisons test (D, F, I, H, K). Results are
representative of at least 3 independent experiments (E-K) or 2 independent experiments (L, M).
Figure S5. Enhanced Energy Storage Improves Microglial Response in Trem2–/– 5XFAD Mice In Vivo, Related to Figures 6 and 7
(A) Quantification of the percentage of microglia that contain LC3 puncta in the cortexes of brains from mice of the indicated genotype and treatment group.
(B) Quantification of the number of LC3 puncta per LC3+ microglia.
(C) Quantification of the relative enrichment of LC3+ microglia within 15 mm of a plaque surface in the cortexes of brains from mice of the indicated genotype and
treatment group.
(D) Quantification of the relative enrichment of cleaved caspase-3+ microglia within 15 mm of a plaque surface in the cortexes of brains from mice of the indicated
genotype and treatment group.
(E) Quantification of the number of microglia per high-powered field in the cortexes of brains from mice of the indicated genotype and treatment group.
(F) Assessment of the complexity of plaques in the cortexes from mice of the indicated genotype and treatment group.
(G and H) The percent of the overall area of the cortex (F) and hippocampus (G) 1.5X brighter than the mean in methoxy-X04 stained sections from mice of the
indicated genotype and treatment group.
(I) Quantification of the percentage of microglia containing methoxy-X04 in plaque bearing regions of the cortexes from mice of the indicated genotype and
treatment group.
Error bar represents mean ± SEM. N.S. indicates not significant, **p < 0.01 or ****p < 0.001 by One-way ANOVA with Holm-Sidak’s multiple comparisons test
(A, B, D, E). Results pooled from 2 independent experiments representing a total of 5-8 male and female mice per treatment group.