Spatiotemporal heterogeneity and dynamic evolution of microglial glucose metabolic reprogramming in Alzheimer's disease

Abstract

Alzheimer's disease (AD) is accompanied by persistent disturbances in brain energy metabolism, and microglial glucose metabolic reprogramming constitutes an important link among pathological protein deposition, neuroinflammation, and impaired clearance. This review examines these metabolic changes across disease stages, brain regions, and plaque‐associated microenvironments and proposes a dynamic state‐composition model. In this model, microglia in different metabolic states coexist, and their relative proportions and spatial distribution change as AD progresses. In early disease or during active phagocytosis, some microglia show enhanced glycolytic and oxidative phosphorylation (OXPHOS) programs. Under persistent pathological stress, increased glucose uptake may not translate into effective energy production and can coexist with inflammation and impaired lipid metabolism. In late disease, microglia with reduced metabolic capacity and impaired phagocytosis become more prevalent. However, higher expression of OXPHOS‐related genes in some microglial states associated with tau pathology does not necessarily indicate improved mitochondrial respiration. These population‐level patterns do not imply that individual microglia follow a fixed sequence of metabolic changes. Longitudinal cell tracking and microglia‐specific metabolic flux measurements are needed to test this model and inform interventions tailored to disease stage and microglial state.

Highlights

• Alzheimer's disease progression reshapes coexisting microglial metabolic states across space and time.

• Plaque niche, disease stage, aging, sex, apolipoprotein E ε4, and triggering receptor expressed on myeloid cells 2 shape these states.

• The hypoxia‐inducible factor 1α– hexokinase 2/pyruvate kinase M2/lactate dehydrogenase A axis links glucose–lipid coupling and dysfunction.

• Therapies should calibrate microglial metabolism by disease stage and cellular state.

1. INTRODUCTION

Alzheimer's disease (AD) is an age‐related progressive neurodegenerative disorder whose pathological course is accompanied by abnormalities in brain energy metabolism and remodeling of microglial functional states. ¹ , ² , ³ Microglia exert state‐dependent functions in pathological recognition, plaque encapsulation, phagocytic clearance, and inflammatory amplification, and these energy‐intensive activities are jointly regulated by glucose uptake, glycolysis, oxidative phosphorylation (OXPHOS), and substrate selection. ⁴ , ⁵

The traditional view holds that, in response to amyloid beta (Aβ) deposition, inflammation, and related stimuli, microglia shift from an OXPHOS‐predominant metabolic state toward immunometabolic reprogramming characterized by enhanced glycolysis. ⁶ , ⁷ , ⁸ However, current evidence indicates that microglial metabolic states do not follow a single linear trajectory of increase or decline. A recent integrated single‐cell and spatial transcriptomic analysis spanning multiple human and mouse pathological conditions further suggests that microglial alterations in AD are better understood as a continuum of partially shared activation states composed of multiple functional programs, ⁹ providing an important basis for interpreting glucose metabolic reprogramming dynamically rather than statically.

Accordingly, this review examines microglial glucose metabolic reprogramming in AD across disease course, spatial location, aging, sex, genetic background, and key regulatory nodes, and proposes a dynamic state‐composition model. Against a background of age‐related metabolic vulnerability, the model summarizes disease‐associated changes as three overlapping state patterns: early compensatory co‐mobilization, metabolic instability during disease progression, and low metabolic fitness and exhaustion in late disease. Its central premise is not that the same microglial cell sequentially traverses three stages along a fixed timeline, but that AD progression entails continuous changes in the relative abundance, spatial distribution, and transition propensity of distinct metabolic states. The literature was primarily identified from PubMed, Web of Science Core Collection, and Scopus, supplemented by representative studies and reference tracking. Priority was given to recent advances in single‐cell omics, spatial omics, and immunometabolism, with selected classic studies included where necessary. Preprints closely related to the topic and informative for emerging directions were interpreted cautiously and are explicitly identified as such in the text.

2. MICROGLIAL GLUCOSE METABOLIC REPROGRAMMING AND AD

2.1. Multilevel evidence supports microglial involvement in AD‐associated glucose metabolic reprogramming

Evidence ranging from population‐level associations and human brain tissue to in vivo animal metabolism and cell‐based functional interventions supports microglial involvement in AD‐associated glucose metabolic reprogramming. Multiple large‐scale plasma and cerebrospinal fluid proteomic studies ¹⁰ , ¹¹ , ¹² , ¹³ have shown that glycolytic protein modules are significantly associated with Aβ and tau burden and cognitive impairment and are enriched in glia‐related components. ¹⁴ Human imaging studies have also identified a positive association between the soluble form of triggering receptor expressed on myeloid cells 2 (TREM2), termed sTREM2, and hippocampal fluorodeoxyglucose uptake, ¹⁵ suggesting a potential association between microglial activation state and glucose use at the level of whole‐brain metabolism. Multiplex immunohistochemistry and high‐resolution spatial analysis of the hippocampal‐entorhinal region in human AD further demonstrated a marked increase in pyruvate kinase M2 (PKM2)⁺/ionized calcium‐binding adaptor molecule 1 (Iba1)⁺ microglia, which predominantly colocalized with an ATP‐binding cassette transporter A7 (ABCA7)‐marked human AD microglia (HAM)‐like disease‐associated phenotype. ¹⁶

Animal studies further indicate that microglia exhibit higher glucose uptake and glycolytic activity than other neural cell types. ¹⁵ , ¹⁷ A study integrating fluorodeoxyglucose positron emission tomography (FDG PET) with single‐cell RNA sequencing (scRNA‐seq) showed that changes in glucose metabolic markers, including glucose transporters, glycolysis, and OXPHOS, occur predominantly in microglia. ¹⁸ In vivo microdialysis combined with ^{1 3}C‐glucose tracing in amyloid precursor protein/presenilin 1 (APP/PS1) mice showed that Aβ progression is accompanied by increased interstitial lactate, reduced glucose, and a relative shift of glucose carbon flux toward lactate production. ¹⁹ Integrated human and mouse single‐cell and spatial analyses further demonstrated enrichment of lactate dehydrogenase A (LDHA) in AD‐associated microglial states and plaque‐proximal regions, providing cellular localization evidence for microglial participation in local lactate metabolism. ¹⁹ Collectively, multilevel evidence identifies microglia as important participants and candidate cellular contributors to AD‐associated glucose metabolic reprogramming and the plaque‐proximal lactate microenvironment, although their independent contribution remains to be quantified by cell‐specific metabolic tracing.

RESEARCH IN CONTEXT

1. Systematic review: The authors searched PubMed, Web of Science Core Collection, and Scopus and screened reference lists for studies of microglial glucose metabolism in Alzheimer's disease. Evidence was appraised by study system and measurement level, prioritizing single‐cell and spatial omics, metabolic flux studies, and functional interventions.

2. Interpretation: The evidence is most consistent with a dynamic state‐composition model rather than a uniform metabolic trajectory. Selected early or phagocytically active states may co‐mobilize glycolytic and oxidative phosphorylation–related programs, whereas chronic pathology and aging favor metabolic instability and low‐fitness states. Most human evidence remains cross‐sectional and primarily transcriptomic or spatial; direct microglia‐specific metabolic‐flux measurements are lacking.

3. Future directions: Longitudinal studies should combine spatial mapping, lineage tracing, cell‐specific stable‐isotope tracing, bioenergetic measurements, and functional assays to distinguish cell‐intrinsic transitions from compositional shifts and define disease‐, niche‐, and state‐specific therapeutic windows.

2.2. Dynamic composition of AD‐associated microglial metabolic states against an aging background

The dynamic state‐composition model proposed here is anchored in age‐related metabolic vulnerability and summarizes disease‐associated changes as three overlapping state patterns: early compensatory co‐mobilization, increasing metabolic instability and state heterogeneity during disease progression, and low metabolic fitness and exhaustion in late disease. This framework describes shifts in the relative abundance, spatial distribution, and transition propensity of metabolic states at the population level rather than presupposing that the same cell sequentially traverses three stages along a fixed timeline.

First, microglia isolated at an early stage in an AD mouse model showed increased glycolysis and glycolytic capacity together with enhanced basal and maximal respiration, but also increased proton leak, consistent with a highly metabolic state in which glycolysis and mitochondrial respiration are co‐mobilized while oxidative coupling efficiency is reduced. ²⁰ In other AD animal studies, phagocytically active microglial populations likewise showed co‐enrichment of glycolytic and OXPHOS‐related programs. ²¹ , ²² Functionally, such parallel mobilization may reflect differential metabolic support for rapid effector responses and for sustained surveillance, chemotaxis, and phagocytic clearance. Studies of metabolic flexibility provide indirect support for this interpretation: under glucose deprivation, microglia in healthy mice and primary cultures can replenish the tricarboxylic acid (TCA) cycle through glutaminolysis, thereby maintaining mitochondrial respiration, process motility, and damage sensing. ¹⁷ These findings suggest that alternative substrate supply may preserve oxidative metabolism and sustained function when glycolytic conditions change, but they do not establish that both metabolic fluxes increase simultaneously within the same cell.

As pathological burden persists, microglial populations may progressively develop more pronounced metabolic instability and state heterogeneity on the basis of early hypermetabolic mobilization. This heterogeneity is first evident in the non‐uniform metabolic programs of distinct disease‐associated states. Deep single‐nucleus transcriptomic profiling of the human AD dorsolateral prefrontal cortex identified multiple microglial states with different glycolytic programs accompanied by enrichment of stress‐, inflammation‐, or respiratory chain–related programs, ²³ indicating metabolic divergence, state heterogeneity, and compositional instability within the AD microglial population. A cross‐model animal transcriptomic atlas showed that disease‐associated microglia (DAM) as a whole can be enriched for OXPHOS‐related programs, whereas hypoxia‐inducible factor 1 (HIF‐1) signaling and broader metabolic‐remodeling pathways are preferentially enriched in DAM stage 2 (DAM2). ²⁴ This pattern suggests that OXPHOS‐related transcriptional mobilization may represent a relatively broad DAM background, while HIF‐1–associated metabolic remodeling is concentrated in specific stages or stress states.

Beyond the uneven distribution of metabolic programs, metabolic instability may also involve progressive uncoupling between substrate entry and downstream glucose use. As disease advances, hippocampal and sorted‐microglial FDG uptake increases in five familial AD mutations (5×FAD) mice, whereas single‐cell transcriptomics reveals an expanding microglial subset characterized by high glucose transporter (GLUT) expression but low glycolytic and OXPHOS programs. ¹⁸ Persistent substrate uptake therefore does not necessarily indicate effective energy production and may instead reflect metabolic mismatch or inefficient use under chronic stress. This mismatch may extend further to disrupted coupling between key metabolic nodes and cellular effector functions. In human AD brain tissue, PKM2⁺ microglia display a cluster of differentiation 68 (CD68)‐associated phagolysosomal phenotype together with perilipin 3 (PLIN3)‐associated lipid‐droplet accumulation and exhausted‐like phagocytic changes. ¹⁶ In light of related mechanistic studies, hexokinase 2 (HK2) dosage and mitochondrial localization, PKM2 conformation and nuclear translocation, and changes in the HIF‐1 α (HIF‐1α)–LDHA/lactate axis may propagate glucose metabolic abnormalities into inflammatory transcription, lactylation feedback, lipid use, and lipid‐droplet homeostasis, increasing the likelihood that metabolic mobilization coexists with inflammation, defective lipid handling, and phagocytic dysfunction. The state‐dependent actions of these metabolic nodes and their bidirectional coupling with neuroinflammation and lipid metabolism are discussed further in sections 3, 4, 5.

Against this background of metabolic instability and state heterogeneity, the combination of sustained Aβ burden and age‐related metabolic vulnerability may further restrict both glycolytic and oxidative support in selected microglial states. Studies using chronic Aβ exposure in vitro and microglia isolated from several aged AD models show that persistent Aβ burden can suppress glycolysis and OXPHOS and is accompanied by impaired phagocytosis and chemotaxis. ²⁰ , ²⁵ , ²⁶ In vivo stable‐isotope tracing of brain microglia from aged mice further demonstrated that a greater proportion of glucose carbon enters glycogen precursors, whereas less enters glycolysis and the TCA cycle, indicating that aging reduces effective glucose use and mitochondrial substrate support. ²⁷ Chronic pathological stimulation may therefore exacerbate mismatches among glycolysis, oxidative metabolism, and functional demand against a background of diminished metabolic reserve, increasing the likelihood that low‐fitness states emerge or accumulate. However, most available studies rely on single time points or comparisons across models and cannot determine whether glycolytic and OXPHOS decline occurs synchronously or establish a unified temporal sequence; distinct subsets and pathological niches may follow different trajectories of metabolic deterioration.

Low‐fitness states in late disease may also manifest through the accumulation of specific subpopulations. Millet et al. identified terminally inflammatory microglia (TIM), an exhausted‐like subset whose relative abundance increases with age and apolipoprotein E (APOE) ε4 burden. TIM show broad reductions in TCA cycle, glycolytic, and pentose phosphate pathway (PPP) programs together with defective Aβ clearance. ²⁸ These findings suggest that late‐stage changes may primarily reflect selective accumulation of low‐fitness states and population‐level compositional reorganization rather than synchronous entry of all microglia into a single terminal state.

Compared to glycolytic programs, changes in microglial OXPHOS‐related programs across AD progression appear more complex and strongly dependent on cellular state and pathological niche. Transcriptomic profiling of freshly isolated human prefrontal microglia showed that genes upregulated at high Braak stages (V–VI) relative to low stages (0–II) were enriched for OXPHOS‐related programs. ²⁹ A separate multiregion human single‐cell study further found that microglial OXPHOS modules were more strongly associated with neuritic plaques and neurofibrillary tangle (NFT) burden, whereas glycolytic modules were preferentially associated with diffuse plaques. ³⁰

Notably, both lines of evidence derive from human brain studies. Together, they indicate that disease progression does not entail a continuous linear decline in OXPHOS‐related programs. Over the longer course of human AD, under the combined influence of Aβ, tau, aging, genetic background, and regional niche, selected microglial states associated with high tau/NFT burden or specific plaque environments may retain or re‐mobilize oxidative metabolism‐related transcriptional programs. Such enrichment, however, may represent a compensatory transcriptional response to mitochondrial dysfunction or the relative expansion of subsets with higher OXPHOS transcriptional signatures and should not be equated with recovery of respiratory function.

Thus, OXPHOS‐related transcriptional enrichment in selected states is not necessarily inconsistent with functional metabolic decline in late disease. The former reflects transcriptional mobilization within specific microglial states or pathological niches, whereas the latter more often refers to population‐level reductions in mitochondrial respiration, reserve capacity, and energetic adaptability; the two observations therefore concern different cellular scopes and measurement layers. In late AD, low respiratory fitness states may become predominant while a smaller number of states retain compensatory OXPHOS transcription or niche‐specific enrichment. Metabolic changes across AD progression are consequently better understood as continuous reorganization of the relative abundance and spatial distribution of distinct metabolic states rather than synchronized changes in glycolysis and OXPHOS along a single temporal axis (Table 1).

TABLE 1.

Evidence mapping of a dynamic state‐composition model of microglial glucose metabolism across AD progression in the context of aging

Representative study	Study system/disease window	Key methods and evidence type	Key findings	Positioning and boundaries within the dynamic state‐composition model
Aging background: aging‐associated metabolic vulnerability
Minhas (2021) 27	Brain microglia from aged C57BL/6 mice; macrophages derived from older adults	Stable‐isotope tracing, glucose carbon‐flow analysis, OCR/ECAR measurements, and EP2 signaling perturbation	Aging diverted less glucose‐derived carbon into glycolysis and the TCA cycle and more toward glycogen precursors, suggesting that aging may precondition microglia for impaired glucose use and reduced mitochondrial substrate support.	Provides foundational background evidence for the model; primarily indicates an age‐related loss of metabolic reserve rather than an AD‐specific disease‐stage change.
Compensatory co‐mobilization
Sangineto (2023) 20	Microglia isolated from 6‐month‐old 3×Tg‐AD mice	Seahorse OCR/ECAR analysis, mitochondrial functional assays, inflammatory mediator measurements, and SDH perturbation	Glycolysis, glycolytic capacity, basal respiration, and maximal respiration were all increased, accompanied by greater proton leak.	Supports coordinated early mobilization of glycolysis and mitochondrial respiration.
March‐Diaz (2021) 21	Microglia from APP751SL/+, MAPT P301S, 5×FAD, and other AD mouse models	Cross‐model transcriptomic analysis, immunohistochemistry, mitochondrial morphology analysis, and HIF‐1α/VHL perturbation	Disease‐associated microglia across models were enriched for OXPHOS‐related transcriptional programs and exhibited altered mitochondrial morphology.	Suggests that OXPHOS‐related transcriptional mobilization may represent a shared background across subsets of DAM; the evidence is primarily transcriptional and morphological.
Grubman (2021) 22	Methoxy‐X04+ Aβ‐phagocytosing microglia from 5×FAD mice	Methoxy‐X04 labeling, FACS, bulk RNA‐seq, scRNA‐seq, and functional assessment	Microglia with active Aβ uptake were concurrently enriched for HIF‐1, OXPHOS, phagocytic, and lipid‐processing programs.	Supports parallel mobilization of glycolytic and oxidative‐metabolic programs in phagocytically active states; simultaneous dual‐flux measurements in the same cell remain lacking.
Metabolic destabilization and increasing state heterogeneity
Liu (2026) 24	Single‐cell/single‐nucleus transcriptomic data from 10 AD mouse models	Integrated multi‐model scRNA‐seq/snRNA‐seq analysis, DAM subcluster analysis, and pseudotime analysis	DAM were broadly enriched for OXPHOS programs, whereas HIF‐1 and broad metabolic‐remodeling pathways were concentrated mainly in DAM2‐associated substates.	Supports distinct metabolic configurations across disease‐associated states; HIF‐1‐related remodeling may be concentrated in specific stress‐associated substates.
Prater (2023) 23	Microglia‐enriched nuclei from the dorsolateral prefrontal cortex of 12 patients with AD and 10 controls	Deep microglia‐enriched snRNA‐seq, subcluster analysis, gene regulatory network analysis, and trajectory analysis	Multiple microglial states with distinct glycolytic programs were identified, accompanied by enrichment of stress‐, inflammation‐, or respiratory chain–related programs.	Supports the coexistence of parallel metabolic states and destabilization of population composition in human AD microglia; cross‐sectional transcriptional evidence cannot substitute for metabolic‐flux measurements or longitudinal validation of state transitions.
Choi (2021) 18	Hippocampi across age points and sorted microglia from 5×FAD mice; ADNI participants with FDG‐PET and cerebrospinal fluid soluble triggering receptor expressed on myeloid cells 2 (sTREM2)	FDG PET, FDG‐uptake assays in sorted cells, hippocampal scRNA‐seq, and ADNI imaging–biomarker association analyses	FDG uptake was increased in 5×FAD hippocampi and sorted microglia; with disease progression, a microglial subpopulation characterized by high GLUT but low glycolytic and OXPHOS programs expanded.	Provides core support for a mismatch between glucose uptake and downstream metabolic programs. FDG uptake reflects glucose uptake and early phosphorylation, whereas GLUT, glycolysis, and OXPHOS changes are derived from transcriptional modules; neither substitutes for direct metabolic‐flux measurements.
Lu (2025) 16	Human AD hippocampal–entorhinal cortex tissue	Multiplex immunohistochemistry, spatial analysis, and localization relative to Aβ/p‐tau pathology	Increased PKM2+ microglia co‐occurred with a CD68‐associated phagolysosomal phenotype, PLIN3+ lipid‐droplet accumulation, and phagocytic exhaustion‐like changes.	Suggests potential aberrant coupling between key metabolic nodes and phagocytic and lipid‐handling functions; based on cross‐sectional colocalization evidence in human brain tissue.
Pan (2022) 64	5×FAD mice, human AD brain tissue, and microglia‐related experimental systems	H4K12la measurement, CUT&Tag/transcriptional analysis, functional glycolysis assessment, pharmacological PKM2 inhibition, and microglia‐specific Pkm2 deletion	H4K12la was increased in plaque‐proximal microglia and enriched at glycolytic‐gene promoters, forming a glycolysis–H4K12la–PKM2 positive feedback loop that promoted aberrant microglial activation and AD pathology.	Supports lactylation‐dependent feed‐forward amplification of inflammation and pathological progression through the glycolytic axis; provides evidence of aberrant coupling between metabolic nodes and effector function.
Codocedo (2024) 57	Microglial Hk2 gene‐dosage models in 5×FAD mice; human AD brain tissue; pharmacological microglial intervention models	Microglia‐specific monoallelic/biallelic Hk2 deletion, pharmacological inhibition with lonidamine, pathological and cognitive assessments, and analyses of inflammation and mitochondrial function	HK2 was increased in plaque‐associated microglia. Partial antagonism or monoallelic deletion reduced Aβ pathology and inflammation, whereas complete deletion conferred no protection and was accompanied by mitochondrial injury and exacerbated inflammation.	Supports dose‐ and state‐dependent effects of HK2; indicates that this metabolic node may drive pathology, whereas excessive inhibition may precipitate energetic and inflammatory dysregulation.
Low metabolic fitness and exhaustion‐like states
Baik (2019) 26	Primary microglia chronically exposed to Aβ; 5×FAD mice	OCR/ECAR analysis, RNA‐seq, multiphoton imaging, and phagocytic functional assays	Chronic Aβ exposure reduced both glycolysis and OXPHOS and was accompanied by impaired phagocytosis and chemotaxis.	Supports the induction of low metabolic fitness by sustained pathological burden; derived mainly from chronic in vitro exposure and animal models and should therefore be extrapolated to the human brain with caution.
Lu (2021) 25	Primary microglia rendered tolerant to chronic Aβ oligomers; microglia‐specific Trpv1‐deficient APP/PS1 mice	OCR/ECAR analysis, AKT/mTOR/HIF‐1α pathway measurements, phagocytosis/autophagy assays, conditional deletion, and pharmacological TRPV1 activation	Chronic Aβ tolerance suppressed glycolysis and OXPHOS and impaired phagocytosis and autophagy; TRPV1 activation restored metabolic and immune functions.	Provides functional evidence for chronic Aβ‐induced low metabolic fitness and its reversibility; derived mainly from primary cells and APP/PS1 mice and cannot establish a uniform temporal sequence for late‐stage human AD.
Sangineto (2023) 20	Microglia isolated from 18‐month‐old 3×Tg‐AD mice	Seahorse OCR/ECAR analysis, mitochondrial functional assays, and SDH perturbation	Compared to age‐matched controls, basal and maximal respiration were significantly reduced; glycolysis and glycolytic capacity also showed downward trends; and dimethyl malonate, an SDH inhibitor, no longer produced a clear effect.	Together with the 6‐month findings, this cross‐sectional age comparison is consistent with a population‐level shift from early metabolic mobilization toward reduced respiratory fitness, but does not establish a cell‐intrinsic longitudinal transition.
Millet (2024) 28	Humanized‐APOE 5×FAD mice and human brain data	Longitudinal scRNA‐seq, scATAC‐seq, multi‐omics integration, Compass transcriptome‐constrained metabolic‐flux inference, spatial transcriptomics, and Aβ‐uptake assays	The proportion of TIM exhaustion‐like microglia increased with age and APOE ε4 burden. Compass inferred broad reductions in energy‐generating pathways, including the TCA cycle, glycolysis, and the PPP, together with impaired Aβ uptake and clearance.	Supports selective accumulation of late low‐fitness states and compositional remodeling. The metabolic abnormalities were recapitulated by Compass flux modeling across single‐cell data but remain transcriptome‐constrained computational inferences. Aβ‐uptake assays provide direct evidence of clearance defects.
Advanced pathology: Maintenance or relative enrichment of OXPHOS transcriptional programs
Kosoy (2025) 29	Freshly isolated post mortem human frontal‐cortex microglia	FACS, bulk RNA‐seq, co‐expression‐module analysis, and associations with Braak stage	Genes upregulated at high Braak stages (V–VI) relative to low stages (0–II) were enriched for OXPHOS‐related programs.	Suggests that some microglial states in advanced human disease may maintain or show relative enrichment of OXPHOS transcriptional programs; actual recovery of respiratory function cannot be inferred.
Mathys (2024) 30	ROSMAP snRNA‐seq data from six human brain regions	Multi‐region snRNA‐seq, metabolic‐module analysis, and associations with pathological burden	The microglial OXPHOS module showed stronger associations with neuritic plaque and NFT burden, whereas the glycolysis module was more closely associated with diffuse plaques.	Supports pathology‐type and niche dependence of late‐stage metabolic states; based on cross‐sectional associations between transcriptional modules and pathology.

Note: APP751SL/+ denotes a heterozygous human amyloid precursor protein 751 transgene carrying the Swedish and London mutations; MAPT P301S denotes human microtubule‐associated protein tau carrying the P301S mutation. Italicized Hk2, Pkm2, and Trpv1 denote mouse gene symbols.

Abbreviations: 3×Tg‐AD, triple‐transgenic Alzheimer's disease mouse model; 5×FAD, mouse model carrying five familial Alzheimer's disease mutations; AD, Alzheimer's disease; ADNI, Alzheimer's Disease Neuroimaging Initiative; AKT, protein kinase B; APOE, apolipoprotein E; APP/PS1, amyloid precursor protein/presenilin 1 transgenic mouse model; CD68, cluster of differentiation 68; CUT&Tag, cleavage under targets and tagmentation; DAM, disease‐associated microglia; ECAR, extracellular acidification rate; EP2, prostaglandin E₂ receptor subtype 2; FACS, fluorescence‐activated cell sorting; FDG, fluorodeoxyglucose; FDG PET, [^{1 8}F]fluorodeoxyglucose positron emission tomography; GLUT, glucose transporter; H4K12la, histone H4 lysine 12 lactylation; HIF‐1, hypoxia‐inducible factor 1; HK2, hexokinase 2; mTOR, mechanistic target of rapamycin; NFT, neurofibrillary tangle; OCR, oxygen consumption rate; OXPHOS, oxidative phosphorylation; PKM2, pyruvate kinase M2; PLIN3, perilipin 3; PPP, pentose phosphate pathway; p‐tau, phosphorylated tau; RNA‐seq, ribonucleic acid sequencing; ROSMAP, Religious Orders Study and Rush Memory and Aging Project; scATAC‐seq, single‐cell assay for transposase‐accessible chromatin using sequencing; scRNA‐seq, single‐cell ribonucleic acid sequencing; SDH, succinate dehydrogenase; snRNA‐seq, single‐nucleus ribonucleic acid sequencing; sTREM2, soluble triggering receptor expressed on myeloid cells 2; TCA, tricarboxylic acid cycle; TIM, terminally inflammatory microglia; TRPV1, transient receptor potential vanilloid 1; VHL, von Hippel–Lindau tumor suppressor protein.

2.3. Spatial heterogeneity and functional associations

2.3.1. Regional differences

Consistent with the pronounced spatiotemporal heterogeneity of microglia. ³¹ , ³² AD‐associated glucose metabolic programs exhibit clear regional differences and lesion‐proximity dependence. Human single‐nucleus studies identified the glycolysis‐associated MG7 state, ³³ whose proportional distribution varies across brain regions. A multiregion atlas further demonstrated marked regional heterogeneity in microglial glycolysis‐related transcriptional programs, with greater glycolytic enrichment in the prefrontal cortex (PFC) and entorhinal cortex (EC) and asynchronous timing of peak enrichment across regions. ³⁰ These findings indicate that regional differences exist in the composition of microglial metabolic states and their association with disease stage, supporting regional temporal organization of AD‐associated glucose metabolic reprogramming. Here, “regional temporal organization” refers to stage‐dependent differences in the enrichment and cellular composition of metabolic states rather than an identical rhythm of change across all brain regions.

2.3.2. Multiscale stratification of the peri‐plaque metabolic niche

Beyond brain region, pathological proximity also shapes microglial metabolic states. Current findings do not support classifying all peri‐plaque microglia as a fixed “high‐glycolytic” or “high‐OXPHOS” population. Animal studies show that selected plaque‐associated microglia can mobilize TREM2‐dependent OXPHOS‐related programs and exhibit mitochondrial elongation, ²¹ while methoxy‐X04‐labeled, Aβ‐internalizing microglia show simultaneous enrichment of OXPHOS‐ and HIF‐1–related programs. ²² Integrating cell‐sorting and functional findings, it can be proposed that some phagocytically active peri‐plaque subsets use glycolysis to meet rapid energy and biosynthetic demands while retaining mitochondrial oxidative support for sustained phagocytosis and lysosomal degradation. However, because the two metabolic fluxes have not been measured simultaneously in the same cell, this functional complementarity remains an interpretive hypothesis.

Using lineage tracing, scRNA‐seq, and assay for transposase‐accessible chromatin with sequencing (ATAC‐seq), Ardura‐Fabregat et al. ³⁴ showed that non–plaque‐associated microglia (non‐PAM) can transition through a colony‐stimulating factor 1 (CSF‐1)–related program into plaque‐associated microglia (PAM) with greater plaque‐restricting capacity, accompanied by upregulation of OXPHOS‐related programs. These findings provide a potential cellular‐state basis for the retention or enhancement of oxidative metabolic support in selected peri‐plaque microglia. Nevertheless, PAM conversion cannot account for all peri‐plaque OXPHOS enrichment; local recruitment, proliferation, selective survival, and heterogeneity of the plaque microenvironment may also shape peri‐plaque cellular composition.

At finer spatial resolution, the plaque‐proximal zone may contain additional metabolic stratification. A high‐resolution spatial transcriptomic preprint stratified microglia by distance from thioflavin S‐positive (ThioS⁺) plaques and reported that cells within 50 µm were enriched for glycolytic and lipid/cholesterol metabolic programs relative to more distant cells, whereas OXPHOS‐, TCA‐, and PPP‐related programs were comparatively reduced. ³⁵ This pattern suggests that entry into the plaque‐proximal niche favors glycolytic and lipid‐stress states. Importantly, the analysis captures a broad contrast between the peri‐plaque zone and distant tissue and does not establish a monotonic increase in glycolytic programs with decreasing distance to the plaque core within the 50 µm zone. A human hippocampal–entorhinal study likewise found an overall increase in PKM2⁺Iba1⁺ microglia near plaques, but within the 50 µm peri‐plaque zone their density increased with distance from the plaque center, suggesting preferential localization immediately outside the pathological core. ¹⁶ These observations are not contradictory: at a larger spatial scale, glycolysis‐related programs are higher in the peri‐plaque niche than in distant tissue, whereas at a finer scale, glycolysis‐associated cells do not necessarily reach their highest abundance closest to the plaque center.

Further work showed that LDHA is enriched not only in the plaque‐proximal microenvironment and reactive microglia but is also more strongly associated spatially with diffuse Aβ deposits than with dense plaque cores. ¹⁹ Peri‐plaque metabolic heterogeneity may therefore depend not only on geometric distance from the plaque center but also on the aggregation morphology of Aβ. Diffuse Aβ surrounding a dense core may form a pathological interface with relatively active glycolytic and lactate metabolism, consistent with the preferential peripheral distribution of PKM2‐associated cells. The peri‐plaque zone is thus more likely to comprise non‐linear stratification among the pathological core, diffuse Aβ interface, and distant tissue than a simple radial metabolic gradient.

2.3.3. Spatial plasticity after Aβ clearance

Aβ immunotherapy studies further indicate that peri‐plaque metabolic states exhibit spatial plasticity dependent on pathological burden. In human microglia–chimeric AD mice, lecanemab induces focal metabolic reprogramming in plaque‐proximal microglia. ³⁶ In APP‐SAA knock‐in mice, long‐term anti‐Aβ treatment reduces widespread glycolysis‐related signals and glucose uptake as overall Aβ burden declines, while persistent metabolic responses remain around residual plaques. ³⁷ Spatial transcriptomics of post mortem brain tissue from patients with AD after Aβ immunotherapy further showed relative preservation of glycolytic and inflammatory responses in regions with limited clearance, whereas regions with more complete clearance exhibited attenuated glycolysis and enrichment of OXPHOS‐related programs. ³⁸ These findings suggest that anti‐Aβ therapy reorganizes the spatial distribution of glycolytic and oxidative programs according to local residual pathology: residual lesions maintain a stronger glycolytic response, whereas regions with more complete Aβ clearance display lower stress and a state composition relatively enriched for OXPHOS‐related programs.

Collectively, peri‐plaque glucose metabolism does not form a fixed gradient along the plaque radius but undergoes non‐linear spatial reorganization according to pathological distance, Aβ aggregation morphology, and clearance efficiency. Because current evidence derives primarily from cross‐sectional spatial omics and a limited set of metabolic markers, this spatial stratification requires validation through cell‐specific flux measurements and longitudinal studies.

2.4. Sex and species differences in AD‐associated microglial glucose metabolic reprogramming: boundaries of model extrapolation

Genetic, hormonal, and psychosocial factors are thought to contribute to sexual dimorphism in AD. ³⁹ Microglia may also be important contributors to this difference, ⁴⁰ , ⁴¹ although direct clinical evidence regarding glucose metabolism in human brain microglia remains limited. Brain sections from female patients show a marked increase in dystrophic microglia together with sex‐related differences in iron staining, implicating oxidative stress, disrupted iron homeostasis, and mitochondrial stress. ⁴² , ⁴³ , ⁴⁴ Single‐nucleus RNA sequencing (snRNA‐seq) has identified sex differences in interleukin‐10 (IL‐10) pathway activation in AD microglia, ⁴⁵ while classic macrophage studies indicate that IL‐10 secretion depends on glycolytic metabolic reprogramming. ⁴⁶ , ⁴⁷ , ⁴⁸

Animal studies provide more direct evidence of sex‐related metabolic stratification. In 18‐month‐old APP/PS1 mice, microglia isolated from females showed higher expression of glycolytic enzymes, extracellular acidification rate (ECAR), and lactate levels, together with reduced Aβ uptake and greater plaque burden. ⁴⁹ During normal aging, microglia from aged females likewise exhibited higher basal and compensatory glycolysis than those from males, accompanied by reduced basal respiration and adenosine triphosphate (ATP) production. Single‐cell reanalysis suggested that this population‐level difference may partially reflect a relative increase in DAM with higher glycolytic programs, but could not distinguish cell‐intrinsic metabolic changes from compositional effects, and metabolic flux was not measured directly in sorted DAM. ⁵⁰

Human data ⁵¹ indicate sex‐related remodeling of glucose metabolism‐immune response coexpression modules during aging, while animal studies ⁵² , ⁵³ more directly show a stronger tendency for female microglia to shift toward high‐glycolytic and DAM‐like states and that Aβ accelerates activation trajectories already present in normal aging. Although these findings are directionally consistent, human coexpression modules, animal cellular states, and direct metabolic flux are not interchangeable. Aging and sex may jointly shape microglial immunometabolic states by altering metabolic reserve, inflammatory response thresholds, and the propensity toward exhaustion, but the magnitude and disease window of these effects in human AD require confirmation through cell‐specific functional studies.

Species differences likewise substantially influence the specific manifestation of microglial glucose metabolic reprogramming and must be considered when extrapolating animal findings to human AD. Integrated transcriptomic, proteomic, and metabolic‐flux analyses comparing human induced pluripotent stem cell (iPSC)–derived microglia to mouse microglia showed that the key molecular nodes required for inflammation‐induced glycolytic activation are not fully conserved between species. ⁵⁴ Proteomic profiling further revealed marked differences in the basal metabolic protein composition of isolated mouse and human microglia, with a higher proportion of glycolytic proteins in mouse microglia. ⁵⁵ Together, sex and species differences demonstrate that microglial glucose metabolic reprogramming is not a single invariant process but a dynamic phenotype jointly regulated by intrinsic biological background and the external pathological environment. Metabolic states in human AD therefore cannot be inferred directly from rodent models and should instead be interpreted in conjunction with human omics evidence and the relevant pathological context.

2.5. Integrated interpretation and applicability boundaries of the dynamic state‐composition model

The disease‐course, spatial, and host‐background evidence summarized above supports interpreting microglial glucose metabolic changes in AD as dynamic compositional remodeling across coexisting states. Differences observed in cross‐sectional atlases may arise simultaneously from cell‐state transitions, changes in the relative proportions of subpopulations, selective proliferation or survival of specific states, and brain region‐ or Aβ/tau pathology–specific enrichment. The state labels used here therefore describe relatively dominant transcriptional and functional features rather than fixed, mutually exclusive cell types; substantial overlap and continuous transitions may occur among them. Consequently, cells resembling homeostatic microglia, cells with enhanced glycolysis, cells with relatively active oxidative metabolism, and cells with globally restricted metabolic function may coexist within the same disease stage or brain region. These metabolic states may not merely represent sequential stages traversed by the same cell but may also exist as relatively distinct yet interconvertible parallel states whose proportions are reorganized as disease and local environments evolve.

Support for the coexistence and proportional reorganization of parallel states derives mainly from cellular atlases spanning ages and spatial scales. TIM are detectable at low abundance early and increase with age and APOE ε4 burden rather than emerging through synchronous conversion of all microglia into a single terminal state, ²⁸ indicating that disease progression partially reflects accumulation of low‐fitness states and population‐level compositional change. Within local peri‐plaque niches, non‐PAM, PAM, and transitional states may coexist, and conversion toward PAM is accompanied by upregulation of OXPHOS‐related programs. ³⁴ At broader tissue scales, different brain regions, pathological morphologies, and degrees of Aβ clearance may preferentially enrich glycolytic or OXPHOS‐associated states. ³⁰ , ³⁸ These observations show that disease progression does not require all microglia to change in the same direction.

Accordingly, the three disease‐associated state patterns in the model describe the relative composition of population‐dominant states under different conditions rather than prescribing a fixed fate sequence for individual cells (Figure 1). Age‐related metabolic vulnerability serves as a host background that persists throughout disease and influences the probability and compensatory capacity of each state. Because the model remains based largely on cross‐sectional evidence, it cannot yet resolve the relative contributions of state transitions, subpopulation expansion, selective survival, or local recruitment, nor can it establish a uniform temporal order for glycolytic and oxidative metabolic instability.

FIGURE 1.

Dynamic state‐composition model of microglial glucose metabolism in AD against an aging background. AD progression may involve homeostatic surveillance, early compensatory co‐mobilization, increasing metabolic instability and state heterogeneity during disease progression, and a relative increase in low‐fitness and exhausted‐like states in late disease. Teal, orange‐red, and gray‐brown cells denote OXPHOS‐biased, glycolysis‐biased, and low‐fitness/exhausted‐like states, respectively; their distribution illustrates relative changes in state composition across disease stages. The blue and red curves summarize population‐level activity of OXPHOS‐ and glycolysis‐related programs, respectively. Maintenance or compensatory mobilization of OXPHOS transcription in selected states during disease progression does not indicate recovery of respiratory function. Spatial heterogeneity, APOE ε4, and female‐associated background can modulate the relative abundance of each state. The model does not represent a fixed trajectory followed by the same cell. Made using BioRender. AD, Alzheimer's disease; APOE, apolipoprotein E; ATP, adenosine triphosphate; OXPHOS, oxidative phosphorylation.

3. KEY METABOLIC NODES AND STATE‐DEPENDENT REGULATORY NETWORKS

3.1. HK2: dose‐dependent effects and mechanistic controversies

3.1.1. Conflicting outcomes and dose dependence after HK2 reduction

Post mortem human brain and 5×FAD mouse studies consistently show increased HK2 in Aβ plaque–associated microglia, implicating HK2 in plaque‐related metabolism and immune responses. ⁵⁶ , ⁵⁷ However, studies have not reached a consistent conclusion regarding the functional consequences of reducing HK2. Leng et al. found that microglial Hk2 deletion or pharmacological inhibition was accompanied by increased lipoprotein lipase (LPL) expression, lipid use, ATP levels, and Aβ phagocytosis, together with reduced plaque burden and improved cognition. ⁵⁶ By contrast, Hu et al. reported under basal physiological conditions and acute cerebral ischemia that efficient Hk2 deletion reduced glycolysis, glucose‐supported mitochondrial respiration, and ATP production without sufficient compensation from fatty acids or glutamine, and impaired microglial surveillance, migration, and repopulation. ⁵⁸ In 5×FAD mice, Codocedo et al. further identified a gene‐dosage effect: heterozygous Hk2 deletion attenuated inflammation, Aβ pathology, and cognitive impairment, whereas complete biallelic deletion did not confer the same protection and was accompanied by reduced mitochondrial membrane potential and enhanced inflammation. ⁵⁷ These findings indicate that the effects of HK2 intervention may depend on the magnitude of reduction, disease context, and the functional state of microglia.

3.1.2. Evidence hierarchy and methodological heterogeneity

The divergent findings may arise from differences in evidence hierarchy and experimental design. Through metabolite supplementation, LPL inhibition, and blockade of fatty acid oxidation, Leng et al. established a relatively complete mechanistic evidence chain linking HK2 inhibition, LPL upregulation, increased lipid use, and enhanced phagocytosis. ⁵⁶ However, ATP measurements primarily reflected steady‐state abundance, direct stable‐isotope evidence that fatty‐acid carbon entered the TCA cycle and was converted into ATP was lacking, and the pharmacological tools used were not entirely specific. Hu et al. combined ECAR and oxygen consumption rate (OCR) measurements, substrate‐dependency assays, ATP quantification, and in vivo imaging to provide strong evidence that complete HK2 loss compromises basal metabolism and microglial motility. ⁵⁸ Nevertheless, the principal in vivo disease validation was performed in acute cerebral ischemia and cannot be directly equated with chronic Aβ pathology. Codocedo et al. provided more direct evidence of gene dosage in 5×FAD mice, ⁵⁷ but some mitochondrial and metabolic conclusions were based primarily on membrane‐potential measurements or fluorescent probes, and pharmacological benefit was observed mainly in relatively early‐stage male mice.

Deletion efficiency, timing of intervention, animal sex, disease model, culture substrate, and outcome selection may therefore all influence the phenotype produced by HK2 manipulation. In addition, these studies largely used Cre systems driven by the C‐X3‐C motif chemokine receptor 1 promoter to target microglia. Although this approach is efficient in the adult brain, recombination may also occur in some brain border‐associated macrophages, and a contribution from non‐parenchymal myeloid cells to selected in vivo phenotypes cannot be fully excluded. ⁵⁷ , ⁵⁸

3.1.3. Dependence on subcellular localization and substrate environment

Beyond expression abundance and deletion efficiency, the subcellular localization of HK2 may be central to its bidirectional effects. HK2 is enriched in mitochondria‐containing microglial processes and can bind voltage‐dependent anion channels (VDACs) at the outer mitochondrial membrane, coupling glucose phosphorylation to mitochondrial energy production. By contrast, motile distal processes lacking mitochondria may depend more heavily on local glycolysis to support surveillance, chemotaxis, and phagocytosis. ⁵⁹ , ⁶⁰ High fructose exposure can also promote glucose transporter 5 (GLUT5)‐dependent recruitment of HK2 to mitochondria and partially suppress phagocytosis, indicating that HK2 localization is modulated by the substrate environment. ⁶¹ Increased total HK2 expression therefore does not necessarily indicate more effective energy production; its function may depend on the relative distribution of cytosolic and mitochondria‐bound HK2. Findings involving optic atrophy 1 (OPA1) regulation and astrocytes derived from patients with AD further support a potential role for hexokinase–mitochondrial coupling in Aβ clearance and metabolic homeostasis. ⁶² , ⁶³

3.1.4. Consensus, controversy, and boundaries of interpretation

A consistent conclusion is that HK2 is increased in Aβ plaque–associated microglia and participates in glycolytic energy supply, mitochondrial homeostasis, and inflammatory signaling; its function cannot be inferred from expression alone but must be interpreted in relation to intervention intensity, subcellular localization, and pathological context. ⁵⁶ , ⁵⁷ , ⁵⁸ The central controversy is not whether HK2 is important, but the source and reproducibility of the protective effects observed after HK2 reduction. Leng et al. attributed increased ATP and Aβ phagocytosis to LPL upregulation and enhanced lipid use. ⁵⁶ Codocedo et al., however, concluded that some benefits of HK2 antagonism do not depend on increased ATP or LPL upregulation and may instead arise from modulation of the inhibitor of nuclear factor κB α (IκBα)–nuclear factor κB (NF‐κB) axis. ⁵⁷ Hu et al. showed that complete HK2 loss causes energetic insufficiency and impaired motility. ⁵⁸ LPL‐mediated lipid compensation, suppression of the IκBα–NF‐κB axis, and energetic failure after complete deletion should therefore be regarded as competing or context‐dependent mechanisms proposed in different experimental systems rather than combined into a single continuous pathway.

3.2. PKM2: conformational state, subcellular localization, and context‐dependent functions

PKM2 is a key regulatory enzyme at the terminal step of glycolysis. It is increased in plaque‐proximal microglia in 5×FAD mice and patients with AD and participates in a glycolysis–lactate/lactylation positive‐feedback loop. ⁶⁴ Its mechanistic interpretation cannot be reduced to total expression. The highly active tetramer primarily performs pyruvate kinase activity, whereas low‐activity monomers or dimers are more likely to translocate to the nucleus and regulate transcription. Evidence that PKM2 promotes anterior pharynx‐defective 1a (APH‐1a) transcription and γ‐secretase complex formation derives mainly from neuronal systems ⁶⁵ and therefore indicates only a potential cross‐cellular mechanism rather than directly demonstrating that microglial PKM2 increases Aβ production. Microglia‐specific studies show that genetic deletion or conformational modulation of PKM2 can alter DAM transition, inflammation, and phagocytic phenotypes, ⁶⁶ but these interventions act at different mechanistic levels.

Current PKM2 intervention studies cannot be reduced to the simple categories of “inhibition is beneficial” or “activation is beneficial.” Genetic deletion simultaneously alters enzymatic activity, conformational equilibrium, and nuclear non‐metabolic functions; promotion of tetramerization primarily enhances pyruvate kinase activity while restricting nuclear translocation; and direct enzymatic inhibition may reduce ATP production and the supply of metabolic intermediates. PKM2 deficiency promotes transition from homeostatic microglia to DAM stage 1 (DAM1) and upregulates protective factors, ⁶⁶ suggesting that PKM2 may restrain adaptive state conversion. In other inflammatory models, however, inhibition of nuclear translocation or promotion of tetramerization attenuates inflammation. The most defensible conclusion is that PKM2 conformation, subcellular localization, and mode of intervention are more informative than total expression for determining its metabolic or inflammatory effects, and that the outcome further depends on disease stage, substrate conditions, and the initial microglial state.

3.3. LDHA and lactate metabolism

Lactate is an important terminal product of glycolysis, and LDHA determines carbon flux from pyruvate toward lactate. Aβ progression is accompanied by increased brain interstitial lactate, while LDHA is enriched in the plaque‐proximal microenvironment and peri‐plaque microglia. ¹⁹ , ³⁵ In vivo ^{1 3}C tracing supports altered brain glucose‐to‐lactate flux, whereas spatial and single‐cell data identify microglia as a possible contributing source without quantifying their independent contribution. Functional intervention studies showing that microglial glycolysis‐derived lactate forms a PKM2‐related positive‐feedback loop through histone H4 lysine 12 lactylation ⁶⁴ provide stronger causal support than colocalization alone. Nevertheless, lactate can also originate from astrocytes, neurons, and vascular‐associated cells and may function as a fuel, signaling molecule, or acidifying factor. Current evidence therefore supports a role for LDHA/lactate in plaque‐proximal immunometabolic regulation but does not justify treating local lactate elevation as a microglia‐specific marker of pathological glycolysis.

3.4. State‐dependent regulatory networks

3.4.1. Upstream state regulators: TREM2, APOE ε4, and insulin signaling

Genome‐wide association studies (GWASs), human cell models, and animal functional studies collectively implicate TREM2, APOE ε4, and insulin signaling in microglial state transition and metabolic regulation, although the evidence supporting each differs in strength and level. TREM2 deficiency can impair mitochondrial quality and disrupt HIF‐1α–associated metabolism, ⁶⁷ , ⁶⁸ whereas TREM2 expression or agonism is associated with enhanced glucose use, OXPHOS‐related programs, and phagocytic function. ⁶⁹ The TCA cycle intermediate succinate partially rescues OXPHOS defects in TREM2 R47H human iPSC‐derived microglia, ⁷⁰ providing human cellular functional evidence linking TREM2 to mitochondrial metabolism. More recently, the small‐molecule TREM2 ligand sulfavant A enhanced microglial motility and Aβ phagocytosis, reduced plaque burden, and did not induce a canonical proinflammatory response. ⁷¹ These findings support the concept that pharmacological TREM2 engagement can reprogram function without broadly activating inflammation, although the evidence remains preclinical. TREM2 is therefore better conceptualized as a state‐dependent tuner of metabolic fitness than as a simple glycolytic switch.

Transcriptomic profiling of microglia freshly isolated from human tissue showed a downward trend in glucose metabolism‐related coexpression modules among APOE ε4 carriers. ⁵¹ As the strongest genetic risk factor for AD, APOE ε4 promotes the accumulation of DAM‐like and exhausted‐like microglial states ²⁸ , ⁷² , ⁷³ and is closely associated with immunometabolic dysfunction. Integrated evidence from primary studies and reviews ⁷² , ⁷³ , ⁷⁴ suggests that APOE ε4 functions both as an accelerator of disease progression and as a regulator of the immunometabolic response threshold, making microglia more likely to enter metabolic reprogramming under cellular stress and to progress more rapidly toward reduced metabolic fitness and functional mismatch. Microglia‐specific attenuation of insulin signaling likewise triggers metabolic reprogramming, ⁷⁵ characterized by enhanced glycolysis, increased mitochondrial fission, impaired Aβ clearance, and increased release of proinflammatory mediators. TREM2, APOE ε4, and insulin signaling may therefore influence the transition from metabolic adaptation to functional mismatch through receptor‐mediated metabolic fitness, genetic risk background, and hormone‐signal sensitivity, respectively.

3.4.2. Key regulatory factor: HIF‐1α

HIF‐1α is a central transcriptional regulator of cellular responses to hypoxia and chronic cerebral hypoperfusion‐related stress and is also a major hub for activation of glycolytic programs. ⁷⁶ , ⁷⁷ HIF‐1 directly regulates the expression of multiple glycolytic and metabolic‐branching genes, including phosphofructokinase 1 (PFK1), hexokinase 1/2 (HK1/2), and PKM2, ⁷⁸ and contributes to glucose transporter 1 (GLUT1)‐mediated enhancement of glucose uptake, thereby shifting cellular metabolism from mitochondrial oxidation toward glycolysis.

Aβ deposition and local stress can activate HIF‐1α–related programs. ²¹ , ²² , ³³ Human single‐nucleus data localize HIF‐1 activity to glycolytic, stress‐associated, and inflammatory states. Genetic stabilization of HIF‐1 signaling in vitro and in vivo reduces the accumulation and proliferation of Aβ‐associated microglia (AβAM) and aggravates Aβ pathology, ²¹ while intervention targeting complement C3a receptor (C3aR)–HIF‐1α alters metabolic fitness, lipid droplets, and phagocytosis. ⁷⁹ A biphasic model in which moderate or transient HIF‐1 activation supports stress adaptation, whereas persistent excessive stabilization promotes metabolic quiescence, mitochondrial damage, or impaired phagocytosis, offers a plausible explanation for these divergent findings; however, direct validation through systematic control of activation magnitude and duration within the same AD model is lacking. A HIF‐1α high microglial subset has also been associated with neuroinflammation. ⁸⁰ HIF‐1α should therefore be viewed as a context‐dependent regulatory hub connecting hypoxia, metabolic remodeling, and inflammatory state, whose functional direction must be determined according to disease window, cellular state, and direct metabolic flux.

3.5. Sugar transport and substrate entry: potential roles of glucose transport and fructose metabolism

Microglia are relatively enriched for expression of the fructose transporter GLUT5 in the brain and express components of the polyol pathway and fructolytic enzymes, providing a molecular basis for fructose uptake, use, and potentially endogenous production. ⁸¹ , ⁸² Available reviews indicate that the direction of GLUT5 change varies across microglial subsets, Aβ‐ or tau‐based models, ages, sexes, and APOE genotypes, suggesting that GLUT5 is a state‐dependent substrate‐transport node rather than a uniformly increased metabolic marker of AD. ⁸³ , ⁸⁴ , ⁸⁵ Under basal conditions, only a limited fraction of peripheral fructose enters the brain, ⁸⁶ and GLUT5 is a facilitative transporter driven by concentration gradients. ⁸⁷ Its expression level therefore cannot determine the direction of net fructose transport or substitute for direct metabolic flux measurements.

Fructose can also be generated from glucose through the cerebral polyol pathway. Human brain metabolic studies have demonstrated this process, and glucose, sorbitol, and fructose are elevated in multiple regions of post mortem AD brain tissue, ⁸⁵ , ⁸⁸ , ⁸⁹ although their cellular sources remain unresolved. Functional evidence from non‐AD models shows that ketohexokinase (KHK)‐dependent fructose metabolism in microglia promotes nicotinamide adenine dinucleotide phosphate oxidase 4 (NOX4) translocation to mitochondria and oxidative stress in diabetic mice, thereby aggravating cognitive impairment. ⁹⁰ Exogenous fructose tracing further demonstrates that fructose enters microglia through GLUT5 and contributes carbon to lactate and TCA‐cycle intermediates; excessive exposure is accompanied by mitochondrial recruitment of HK2, reduced ATP, and impaired phagocytosis. ⁶¹ These findings indicate that fructose carbon can enter glycolysis and the TCA cycle and alter mitochondrial function. However, expression of polyol‐pathway and fructolytic enzymes differs between human and mouse microglia, ⁸¹ , ⁸² , ⁸⁵ and the KHK isoform expressed by microglia remains uncertain. ⁸⁵ , ⁹¹ Current evidence therefore supports the possibility that endogenous fructose may be increased in the AD brain and that microglia can use and respond to this substrate, but does not establish autonomous fructose production or increased KHK‐dependent fructose flux in human AD microglia.

4. BIDIRECTIONAL COUPLING BETWEEN MICROGLIAL GLUCOSE METABOLIC REPROGRAMMING AND NEUROINFLAMMATION

4.1. Glucose metabolic reprogramming regulates neuroinflammation and phagocytic function

Metabolic and mitochondrial alterations in microglia are important regulators of immune state and are closely linked to neuroinflammation in the central nervous system. ⁹² , ⁹³ Microglia in different activation and functional states exhibit distinct bioenergetic profiles, and metabolic programs are tightly coupled to immune function. ⁹⁴

Mechanistic evidence derives mainly from in vitro inflammatory models and animal interventions. Glycolytic inhibitors reduce lipopolysaccharide (LPS)‐induced NF‐κB signaling and inflammatory mediator expression, ⁹⁵ supporting a role for glycolysis in inflammatory programs, although pharmacological inhibitors may also affect cell viability and non‐glycolytic targets and therefore cannot establish target‐specific causality on their own. Histone H3 lysine 18 lactylation is increased in senescent microglia and AD models and promotes NF‐κB–related transcription through promoter enrichment, ⁹⁶ providing a causal chain stronger than lactate association alone. Reduced glucose use, changes in the reduced/oxidized nicotinamide adenine dinucleotide ratio (NADH/NAD⁺), and mitochondrial hexokinase localization can also regulate inflammatory transcription and phagocytosis. ⁹⁷ , ⁹⁸ HK2 haploinsufficiency or partial pharmacological antagonism attenuates inflammation and Aβ pathology, whereas complete loss may induce mitochondrial dysfunction and inflammasome activation. ⁵⁷ , ⁵⁸ These differences involve dosage, subcellular localization, and substrate‐compensation capacity, as discussed in section 3.1.

LPS‐induced PKM2 exists predominantly as low‐activity monomers or dimers and therefore more readily translocates to the nucleus to perform non‐metabolic functions. ⁹⁹ HIF‐1α directly upregulates PKM2, whereas dimeric PKM2 phosphorylates and coactivates HIF‐1α; the resulting complex promotes transcription of proinflammatory genes such as interleukin‐1β (IL‐1β). ¹⁰⁰ Pharmacological activation of PKM2 tetramerization reduces basal and maximal glycolytic capacity and attenuates mitochondrial damage and neuroinflammation in multiple models. ⁶⁶ , ⁹⁹ , ¹⁰¹ , ¹⁰² , ¹⁰³ , ¹⁰⁴

Recent studies ¹⁰⁵ , ¹⁰⁶ have also identified an immune‐responsive gene 1 (IRG1)/itaconate/nuclear factor erythroid 2‐related factor 2 (NRF2) glycolytic metabolic negative‐feedback axis in inflammatory microglia that can partially counteract oxidative stress and neurotoxicity caused by excessive glycolytic activation. Excessive glycolysis and energy insufficiency are therefore not mutually exclusive but may characterize different stages of inflammatory imbalance. HK2, the PKM2–HIF‐1α axis, and energy‐sensing mechanisms such as NADH/NAD⁺ collectively determine the direction and magnitude of this process.

4.2. Neuroinflammation and trained immunity reshape metabolic thresholds

Microglia can undergo trained immunity–like immunometabolic remodeling after short‐term stimulation. Aerobic glycolysis induced by the mechanistic target of rapamycin (mTOR)–HIF‐1α pathway forms the metabolic basis of trained immunity. ¹⁰⁷ , ¹⁰⁸ In AD model mice, immune training aggravates Aβ pathology, whereas immune tolerance temporarily alleviates related pathology. ¹⁰⁹ In a Parkinson's disease model, microglia‐specific deletion of HIF‐1α blocks immune memory and reduces pathology. ¹¹⁰ These findings suggest that prior inflammatory exposures, including maternal inflammation, may alter the metabolic‐response threshold of microglia when they subsequently encounter protein aggregation or injury. ¹¹¹ , ¹¹² This concept has recently been introduced into microglial research and proposed to contribute to microglial dysfunction in AD and other neurodegenerative diseases. ¹¹¹ , ¹¹³ However, direct evidence for a trained‐immunity framework currently derives mainly from non‐AD or animal models. Whether it constitutes an important mechanism of dysregulated microglial responses in AD requires direct validation in additional human studies.

5. BIDIRECTIONAL COUPLING BETWEEN MICROGLIAL GLYCOLYSIS AND LIPID METABOLISM

Current evidence suggests that microglial glycolysis and lipid metabolism form a bidirectional regulatory relationship in AD and related neuroinflammatory models. In the direction from glucose metabolism to lipid remodeling, Aβ promotes PKM2 nuclear translocation and activates sterol regulatory element‐binding protein 1 (SREBP1)‐mediated fatty‐acid synthesis and lipid‐droplet accumulation, indicating that PKM2 can link glucose metabolism to lipogenesis through non‐canonical transcriptional regulation. This effect may depend primarily on PKM2 conformation and subcellular localization rather than on increased glycolytic flux per se. ¹¹⁴ Glycolysis‐derived lactate can also remodel metabolic gene expression through histone lactylation and promote neuronal lipid‐droplet accumulation through a microglia–neuron axis. ¹¹⁵ This cross‐cellular effect supports a role for lactate in propagating glucose‐lipid metabolic signals but does not establish that lactate directly drives lipid‐droplet formation within microglia themselves. Conversely, in selected experimental contexts, HK2 loss compensates for restricted glucose metabolism by upregulating LPL and increasing lipid use, indicating that suppression of glucose metabolism can trigger compensatory redistribution toward lipid substrates.

In the reverse direction, from lipid dysregulation to glycolytic reprogramming, the atypical sphingolipid metabolite 1‐deoxysphinganine induces HK2‐, PKM2‐, and LDHA‐associated glycolytic programs, ¹¹⁶ demonstrating that specific lipid metabolites can reshape microglial glucose metabolism. A recent preprint further reported that APOE ε4‐mediated impairment of lipoprotein secretion is accompanied by restricted fatty‐acid oxidation and increased glycolytic dependence, suggesting that the glycolytic shift may arise not only from inflammatory signaling but also secondarily from defects in lipid transport, lysosomal processing, and mitochondrial substrate use. ¹¹⁷ Collectively, microglial glycolysis and lipid metabolism constitute a bidirectional, state‐dependent coupling network: glucose‐metabolic nodes can promote lipid synthesis, alter lipid‐substrate use, and propagate metabolic effects across cells through lactate, while impaired lipid transport, sphingolipid metabolism, and fatty‐acid oxidation can reciprocally enhance glycolytic programs or dependence. This relationship is not a fixed linear pathway but is jointly regulated by cellular state, substrate environment, and pathological context (Figure 2).

FIGURE 2.

Regulatory network of microglial glucose metabolic reprogramming in Alzheimer's disease. Aβ deposition and inflammatory stimuli, APOE ε4‐associated effects, impaired insulin signaling, and context‐dependent TREM2 signaling can influence the HIF‐1α–HK2/PKM2 axis and microglial glycolysis. Glucose enters through GLUT1 and supplies glycolysis, while its conversion through sorbitol to fructose via the polyol pathway is depicted as a potential additional substrate route. Exogenous fructose enters through GLUT5 and can contribute to glycolysis through KHK‐dependent metabolism. HK2 is shown in both cytosolic and mitochondrial outer membrane–associated locations, and dashed bidirectional arrows depict context‐dependent interactions between glycolysis and mitochondrial dysfunction. Nuclear PKM2 participates in the regulation of pro‐IL‐1β expression and SREBP1‐mediated lipogenesis, linking metabolic regulation to inflammatory responses and lipid‐droplet accumulation. Glycolysis‐derived lactate contributes to histone lactylation, which is depicted as a link to NF‐κB–associated inflammatory signaling and feedback reinforcement of glycolysis. Black solid arrows indicate the depicted pathway connections, dashed arrows indicate indirect or context‐dependent relationships, and the red arrow indicates pathological positive feedback. Made using BioRender. Aβ, amyloid beta; APOE, apolipoprotein E; GLUT1, glucose transporter 1; GLUT5, glucose transporter 5; HIF‐1 α, hypoxia‐inducible factor 1 alpha; HK2, hexokinase 2; KHK, ketohexokinase; la, lactylation; NF‐κB, nuclear factor κB; PKM2, pyruvate kinase M2; pro‐IL‐1β, interleukin‐1β precursor; SREBP1, sterol regulatory element‐binding protein 1; TREM2, triggering receptor expressed on myeloid cells 2

6. TRANSLATIONAL IMPLICATIONS AND RESEARCH PRIORITIES: FROM UNIDIRECTIONAL INHIBITION TO STATE‐DEPENDENT METABOLIC CALIBRATION

Current preclinical evidence supports a metabolic‐calibration framework in which disease course, spatial location, and cellular state are treated as key variables. ⁶⁷ , ⁷⁴ , ¹¹⁸ The objective is not to broadly “activate” or “inhibit” microglia but to distinguish among peri‐plaque clearance, metabolic instability, and lipid droplet–associated dysfunction. A current priority is to develop biomarkers that resolve microglial metabolic state, direct flux, and response windows; the strategies below should therefore be regarded as testable stratification hypotheses rather than established therapeutic recommendations.

6.1. Early protective metabolic support: maintaining clearance and peri‐plaque defense

During early disease or local increases in Aβ burden, selected microglia may show co‐enrichment of glycolytic and OXPHOS‐related programs. If this state is confirmed in the same cells through flux and functional measurements, intervention should not indiscriminately suppress metabolic activity but should preserve the energetic support required for peri‐plaque clustering, encapsulation, and phagocytosis while limiting excessive inflammation. At present, this concept remains a state model–derived therapeutic hypothesis and lacks population‐stratification biomarkers and clinical outcome evidence.

Among candidate targets, the TREM2 pathway has comparatively extensive functional intervention evidence. Antibodies or small‐molecule ligands enhance glucose use and Aβ phagocytosis in animal models, but the effects depend on TREM2 expression, receptor occupancy, and cellular state. ⁶⁹ , ⁷¹ , ¹¹⁹ A phase 1 human study showed preliminary safety and target‐engagement profiles for AL002. ¹²⁰ However, the subsequently published phase 2 INVOKE‐2 randomized trial demonstrated sustained TREM2 target engagement and cerebrospinal fluid pharmacodynamic responses without meeting the primary Clinical Dementia Rating Sum of Boxes (CDR‐SB) endpoint or showing clear clinical benefit. ¹²¹ Future work should prioritize state biomarkers and pharmacodynamic readouts before identifying patient subgroups most likely to benefit.

6.2. Metabolic correction of mid‐ to late‐stage unstable states: limiting inflammatory amplification and lipid‐droplet accumulation

Cross‐sectional and preclinical evidence suggests that metabolic instability and functional mismatch become relatively enriched when chronic Aβ stimulation converges with aging, APOE ε4, and mitochondrial vulnerability. This pattern may reflect either subsequent conversion of early hyperresponsive states or expansion and selective survival of exhausted‐like subpopulations. Mid‐ to late‐stage candidate interventions should therefore aim to constrain glycolysis–lactate‐inflammation feedback, persistent lipid‐droplet accumulation, and declining phagocytic function rather than broadly “inhibiting” or “activating” microglia.

In this context, conformational modulation of PKM2 ⁶⁶ has mechanistic value and could theoretically restrict nuclear proinflammatory functions while redirecting glycolysis–lactate and lipid‐metabolic branching. The PKM2‐targeted positron emission tomography (PET) tracer [^{1 1}C]7d detects increased signals in AD‐related brain regions in 5×FAD mice, ¹²² providing proof of concept for development of a metabolic imaging biomarker. However, the signal has not been shown to be microglia specific and cannot distinguish PKM2 abundance, conformation, or enzymatic activity. It should therefore be viewed as a candidate state readout rather than a tool for defining an actionable mid‐ to late‐stage intervention window.

The HIF‐1α–HK2/PKM2 axis is linked at multiple levels to enhanced glycolysis, inflammatory amplification, lipid‐droplet formation, and abnormal phagocytosis and may therefore represent a candidate regulatory network in unstable states. Current evidence, however, does not support universal unidirectional inhibition across disease stages. Excessive suppression may compromise residual clearance and necessary stress adaptation, while complete HK2 deletion and direct PKM2 inhibition may produce outcomes distinct from partial reduction or conformational modulation. In the absence of cell‐state and pharmacodynamic biomarkers, a more rational research strategy is to compare intervention intensity, conformational effects, and disease windows rather than prespecifying a uniform therapeutic direction.

Lesion enrichment and microglia‐selective delivery provide technical opportunities for state‐dependent metabolic intervention. ¹²³ , ¹²⁴ , ¹²⁵ However, available studies vary substantially in cargo, targeting strategy, model stage, and endpoints and lack long‐term toxicity assessment, in vivo biodistribution, and independent replication. Current evidence supports delivery feasibility but does not establish superiority over systemic or single‐target interventions.

6.3. Opportunities and boundaries of systemic interventions

Systemic interventions such as near‐infrared photobiomodulation and exercise are associated with improvements in microglial glucose metabolism, mitochondrial indices, and inflammatory markers in AD or aging models. ¹²⁶ , ¹²⁷ , ¹²⁸ , ¹²⁹ , ¹³⁰ , ¹³¹ Because these interventions simultaneously affect neurons, astrocytes, the vasculature, and the peripheral metabolic environment, current findings do not establish that benefit is primarily mediated by a directed microglial shift from glycolysis toward OXPHOS. These approaches are better viewed as adjunctive means of calibrating the overall metabolic environment rather than substitutes for state stratification and targeted therapy.

7. DISCUSSION

Current evidence supports three relatively robust conclusions regarding microglial glucose metabolic reprogramming in AD. First, metabolic states depend strongly on spatial location, pathological subtype, disease stage, and genetic background. Second, glycolysis and OXPHOS are not mutually exclusive switches: their associated programs may be co‐enriched in selected early, peri‐plaque, or phagocytically active states, but such co‐enrichment does not establish simultaneous increases in both metabolic fluxes within the same cell. Third, the effects of metabolic nodes including HK2, PKM2, HIF‐1α, and TREM2 depend on intervention intensity, subcellular localization, substrate environment, and cellular state. The principal controversies are whether glycolysis and OXPHOS can increase simultaneously within individual cells, whether late OXPHOS transcriptional enrichment corresponds to genuine recovery of respiratory function, and whether the direction of HK2, PKM2, or HIF‐1α intervention can be extrapolated across disease stages and cellular states.

Based on these areas of consensus and controversy, this review proposes a dynamic state‐composition model that interprets AD progression as continuous reorganization of the relative abundance and spatial niches of distinct metabolic states rather than uniform temporal changes in glycolysis and OXPHOS or conversion of individual cells along a fixed fate sequence. The model integrates multilevel evidence but does not establish longitudinal metabolic trajectories within individual microglia. Human brain studies rely mainly on cross‐sectional single‐cell or single‐nucleus transcriptomics, spatial omics, and pathological colocalization. These approaches identify associations between metabolic states and disease stage, brain region, or pathological niche but do not directly measure microglia‐specific metabolic flux in the human brain or establish causal direction. Animal and ex vivo studies provide functional support for selected metabolic changes, while genetic or pharmacological interventions strengthen causal inference. However, these findings are constrained by the pathological spectrum of each model and by species differences and cannot substitute directly for human evidence. In vitro experiments facilitate mechanistic dissection but often use acute LPS, high concentrations of Aβ, or culture systems lacking the brain microenvironment. Furthermore, transcriptomic enrichment, protein expression, enzymatic activity, OCR/ECAR, and stable‐isotope tracing measure different layers and are not interchangeable. Upregulation of an OXPHOS gene set may represent enhanced respiration, stress‐induced mitochondrial biogenesis, or inefficient compensation, just as upregulation of glycolytic genes does not necessarily indicate a parallel increase in lactate carbon flux. Finally, a substantial proportion of the evidence linking metabolic states to plaque distance, lipid‐droplet accumulation, impaired phagocytosis, and cognitive decline remains correlational. Causal chains can be progressively closed only through cell‐specific interventions that measure metabolic and functional outcomes concurrently.

Future research should prioritize falsifiable longitudinal designs rather than continue to accumulate static pathway‐enrichment observations. Ideally, spatial localization, cellular lineage tracing, stable‐isotope tracing, OCR/ECAR, metabolite quantification, subpopulation proliferation or survival, and phagocytic function should be integrated across the same model and disease sequence, with cross‐validation among human iPSC‐derived microglia, acute brain tissue, and post mortem samples. The model generates at least three testable predictions. First, the plaque core and periphery should exhibit distinct direct glucose and fatty‐acid carbon fluxes at the same disease stage. Second, selected late states with high tau/NFT burden may show increased OXPHOS transcriptional programs without parallel recovery of respiratory reserve. Third, if dynamic change is driven mainly by cell‐intrinsic state transitions, lineage tracing should reveal movement of the same cells across states; if it is driven mainly by population composition, differential proliferation, death, or recruitment of subpopulations should predominate. For HK2 and PKM2, partial reduction, complete loss, enzymatic inhibition, and conformational modulation should also be compared directly. Only after disease windows, cellular states, and direct metabolic flux have been defined can metabolism‐stratified interventions advance from an explanatory framework to a testable therapeutic strategy.

CONFLICT OF INTEREST STATEMENT

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Author disclosures are available in the Supporting Information.

Supporting information

Supporting Information