The microglial lineage: from molecular states to context-dependent functions

Abstract

Microglial states are now being cataloged far faster than their associated cellular functions can be tested, widening the gap between molecular description and causal understanding of tissue impact. The gap becomes especially apparent as microglial biology moves beyond the rodent central nervous system (CNS). Peripheral nervous system (PNS) microglia, identified in human and other large-vertebrate ganglia, share major molecular features with CNS microglia and contribute to neuronal growth, yet their adult maintenance and responses to injury and disease remain unresolved. TREM2 exposes a second limit of state-based interpretation. The receptor supports recurring cellular processes, including myeloid fitness, lesion-associated accumulation and cargo handling, but their tissue consequences vary with the substrate, model and disease stage. This review separates population identity, molecular state, cellular function and tissue outcome, and outlines an experimental path that combines state-informed genetic screens, direct functional assays, multicellular human models and temporally controlled in vivo validation to connect molecular discovery with context-dependent function.

1. Introduction

Pío del Río-Hortega’s separation of microglia from the other elements of neuroglia was one of the conceptual achievements of early neurobiology. In silver-carbonate preparations, he recognized small, ramified cells that became amoeboid phagocytes in diseased tissue and, remarkably, argued for their mesodermal origin. A century of work has since established microglia as yolk-sac-derived, locally maintained cells with roles in neural development, homeostasis, injury and disease. Yet most of the mechanistic evidence still comes from rodent CNS models. Several reviews have traced this history and discussed the immune and non-immune functions, molecular signatures and disease-associated states of microglia (Borst et al., 2021; Butovsky & Weiner, 2018; Li & Barres, 2018; Paolicelli et al., 2022; Prinz et al., 2019; Sierra et al., 2019). We will not repeat the existing content; instead, we focus on a gap opened by the rapid expansion of molecular profiling.

Molecular profiling substantially advances the field, but a molecular signature alone does not equate to biological identity or function. A shared transcriptional program does not necessarily establish a shared lineage, particularly when CNS-associated macrophages, infiltrating monocytes and peripheral neural macrophages show overlapping states (Prinz et al., 2021; Wang et al., 2023; Wu, Wang, et al., 2025). Nor does a disease-associated state tell us what a cell is doing or whether its response benefits the tissue; the same signature may accompany injury, adaptation or repair (Fumagalli et al., 2025; Paolicelli et al., 2022). State-associated pathways, by contrast, require perturbation, direct functional measurements and analysis at the relevant disease stage. The central challenge is to move from molecular description to causal, context-dependent function.

We use four terms throughout the review. Identity combines ontogeny, anatomical niche and coordinated molecular features to define a cell population. State describes a molecular configuration associated with development, region, age, sex, experience, or perturbation (Fumagalli et al., 2025; Paolicelli et al., 2022). Cellular function refers to an action demonstrated by perturbation and direct assay, such as survival, proliferation, engulfment, or lipid processing. Tissue or disease impact is the consequence of that action for neurons, myelin, synapses, lesions or disease progression. Following the proposed nomenclature (Chen et al., 2025), we use PNS microglia for the ganglionic population supported by a coordinated microglial molecular program, an inferred developmental relationship to CNS microglia, a defined neuron-associated niche and experimentally supported neural interactions (Wu, Wang, et al., 2025). The discussion follows this sequence. PNS microglia illustrate how molecular discovery can lead to anatomical and functional definition; microglial state studies define what is needed to establish cellular function; and TREM2 shows why the same recurring cellular function can have different tissue outcomes.

2. Expanding microglial biology beyond the mouse CNS

2.1. CNS microglia as the reference lineage

Fate-mapping studies traced adult parenchymal microglia to primitive yolk-sac macrophages that enter the embryonic brain before definitive hematopoiesis (Ginhoux et al., 2010) (Fig. 1). The identification of Myb-independent tissue-macrophage lineages and PU.1- and IRF8-dependent differentiation from erythromyeloid progenitors later reinforced this conclusion (Kierdorf et al., 2013; Schulz et al., 2012). Microglial ontogeny and the development of other CNS macrophages have been reviewed in detail elsewhere (Prinz et al., 2021). The key point here is lineage continuity: embryonic, adult and aged microglia represent successive stages of a resident population rather than gradual replacement by circulating monocytes during steady state.

Fig. 1.

From the established CNS microglial lineage to PNS microglial discovery.

CNS microglia arise from early yolk-sac macrophage progenitors, acquire their identity in the brain parenchyma and persist through local self-renewal with little monocyte contribution under steady-state conditions. The CNS panel depicts environmental sensing by motile processes, synaptic remodeling, phagocytosis of debris and dying cells, signaling with neurons, and neuroimmune responses. The peripheral panel uses the PNS microglia nomenclature proposed by Chen, Wu and Li (Chen et al., 2025) and shows these cells between neuronal somata and satellite glia in sensory and sympathetic ganglia of large vertebrates. Coordinated expression of P2RY12, TMEM119 and SALL1, together with epigenetic similarity, supports their relationship to CNS microglia. Low or absent MRC1, DAB2 and LYVE1 expression distinguishes the reported population from comparator peripheral macrophages. Depletion studies support contributions to neuronal growth and sensory function. Adult self-maintenance and disease-related functions remain unresolved. CNS, central nervous system; PNS, peripheral nervous system; CSF1R, colony-stimulating factor 1 receptor; PLX5622, a CSF1R inhibitor.

Embryonic origin, however, is not enough to define microglial lineage membership. Other resident macrophages, including CNS border-associated macrophages, also arise predominantly from early yolk-sac hematopoiesis. Even so, the trajectories of border-associated macrophages and microglia separate early in development, and their transcriptional programs are already distinct by E16.5 (Goldmann et al., 2016; Utz et al., 2020). Single-cell atlases separate them most clearly when anatomical position and local signals are considered together with transcription (Van Hove et al., 2019). Within the parenchyma, microglia contact neurons, synapses, myelin and the local vasculature; border macrophages instead encounter meningeal, cerebrospinal-fluid and blood-derived cues. We previously proposed a clear definition of the microglial lineage: microglial lineage cells refer to resident immune cells with a microglial molecular phenotype and ontogeny (Chen et al., 2025).

Circulating or transplanted macrophages can enter a vacant CNS compartment, become ramified and acquire substantial parts of the microglial program (Mildner et al., 2007; Xu et al., 2020). In adult bone-marrow- or monocyte-derived cells, however, this conversion is extensive but incomplete. Even after prolonged residence in the CNS, engrafted cells retain ontogeny-associated transcriptional and chromatin features. They also differ from endogenous microglia in phagocytic capacity, cytokine production and responses to peripheral endotoxin (Bennett et al., 2018; Cronk et al., 2018; Lund et al., 2018; Shemer et al., 2018). The parenchymal niche can therefore impose much of microglial identity without erasing developmental history. Ontogeny and niche are closely intertwined, but they are not interchangeable. Whether these residual differences alter tissue outcomes during aging, infection or chronic protein aggregation requires further attention.

Lineage continuity does not mean that each microglial cell is static or indefinitely long-lived. Parabiosis and related experiments found little contribution from circulating cells in the unconditioned CNS, whereas irradiation, barrier disruption, inflammation and niche vacancy permit monocyte engraftment (Ajami et al., 2007; Mildner et al., 2007). Within the resident population, proliferation is balanced by apoptosis, and repopulation after injury may involve random or selected clonal expansion (Askew et al., 2017; Tay et al., 2017). The routine turnover of a self-maintained population is biologically different from repopulation of a depleted or damaged niche, where other cellular sources may gain access.

The molecular boundaries become especially blurred in diseased tissue. Homeostatic genes decline in plaque-associated microglia, while recruited and border macrophages acquire overlapping lipid-handling, phagocytic and inflammatory programs (Keren-Shaul et al., 2017; Krasemann et al., 2017; Van Hove et al., 2019). Loss of TMEM119 or P2RY12 is therefore poor evidence for a non-microglial origin, just as expression of either marker does not prove parenchymal ancestry. Anatomical position, fate information, and coordinated molecular programs provide independent, complementary clues. If origin has not been resolved, terms such as “myeloid cell” or “microglia/macrophage” are often more informative than a falsely precise label (Paolicelli et al., 2022). Functional differences also become visible when resident and recruited myeloid cells are experimentally separated. In experimental autoimmune encephalomyelitis, lineage-resolved imaging showed monocyte-derived macrophages initiating demyelination at nodes of Ranvier, while resident microglia mainly removed myelin debris (Yamasaki et al., 2014). Early after ischemic stroke, microglia produced more reactive oxygen species and TNF, whereas infiltrating monocytes were the main IL-1β-producing cells and had greater phagocytic capacity (Ritzel et al., 2015). The two populations can occupy the same lesion and share inflammatory markers without doing the same work.

2.2. PNS microglia: from population discovery to initial function

Cross-tissue atlases have prompted the field to reconsider the traditional CNS-restricted definition of microglia (Fig. 1). Human prenatal single-cell studies identified cells outside the CNS with coordinated microglia-related programs, leading to the proposal that brain microglia may be one member of a broader “microglial lineage” (Chen et al., 2025). Across postconception weeks 4–26 and 19 tissues, cells carrying combinations of P2RY12, TMEM119, SALL1 and other microglial core genes were found in fetal epidermis, heart-associated tissue, testes and peripheral neural structures (Wang et al., 2023; Wu, Wang, et al., 2025). Prenatal skin and gonadal atlases independently revealed specialized macrophage populations in these tissues (Garcia-Alonso et al., 2022; Gopee et al., 2024). The ganglionic myeloid population is proposed to be named PNS microglia, integrating its molecular and developmental relationship into CNS microglia, its neural niche, and function (Chen et al., 2025). We adopt this nomenclature because it captures a defined population supported by convergent evidence and facilitates comparison across the two nervous systems. It also avoids the ambiguity of the suffix "-like”, which has been applied to in vitro-derived cells, engrafted monocyte-derived cells and macrophages sharing only selected microglial features. It does not confer the same identity on every peripheral macrophage expressing microglial genes (Table 1).

Table 1.

Evidence for microglial lineage criteria in CNS and peripheral populations.

Population and source	Ontogeny and continuity	Core molecular program	Resident niche	Neural interaction and function	Adult maintenance
CNS microglia (Reference population) (Ginhoux et al., 2010; Ajami et al., 2007)	Established by genetic fate mapping and maintenance studies in mice.	Coordinated microglial transcriptional and regulatory program; individual markers vary with state.	Brain and spinal-cord parenchyma, including interfaces with neurons, synapses and myelin.	Directly tested roles in neural development, circuit regulation and cargo clearance; effects depend on context.	Local self-renewal with little blood-monocyte input in physiological condition.
PNS microglia (Ganglionic population) (Wu, Wang, et al., 2025; Chen et al., 2025)	Yolk-sac-associated route inferred from prenatal trajectories and transitional cells.	Convergent transcriptome, protein markers and chromatin features, including P2RY12, TMEM119 and SALL1.	Sensory and sympathetic ganglia; between neuronal soma and satellite-glial envelope in humans and other large vertebrates.	Direct anatomical contact. Regulate neuronal soma size.	Detected at adult ages; repopulation reported. Source of replacement and lifelong self-renewal unresolved.
Epidermal microglia-like cells (Wang et al., 2023)	Prenatal developmental relationship inferred; no direct genealogical tracing.	Microglia-related signature.	Early fetal epidermis; enriched dorsally and in the head.	Contact neural-crest cells.	Declined during gestation; not detected after 20 postconception weeks,
Gonadal microglia-like cells (Wang et al., 2023)	Embryonic association inferred.	Microglia-related transcriptional program.	Testicular/epididymal tissue, especially around developing efferent ducts; fewer cells in parenchyma.	No direct functional interaction with neural elements established for this population.	Cross-sectional detection cannot demonstrate continuity.
Cardiac-associated microglia-like cells (Wang et al., 2023)	Prenatal origin inferred; no direct fate mapping.	Microglia-related transcriptional program.	Developing aorta, including adventitia, media and intima.	Neural interaction and population-specific cardiac function untested.	Present through 26 postconception weeks; not detected in the adult aorta examined.

PNS microglia are found in the sensory and sympathetic ganglia of humans, macaques, pigs, and other relatively large animals, where they occupy a narrow space between neuronal somata and the satellite-glial-cell envelope (Wu, Wang, et al., 2025). Their transcriptomic and epigenetic programs, together with their protein-marker expression, converge substantially with those of CNS microglia. Depletion experiments in human fetal ganglion explants and neonatal pigs linked these cells to neuronal soma enlargement, axon growth and somatosensory function. Across 24 vertebrate species, the occurrence of these cells correlated more closely with neuronal soma and body size than with phylogenetic distance.

Time-resolved human single-cell data place PNS microglia on an inferred developmental route from yolk-sac-related macrophage progenitors, running in parallel with the appearance of CNS microglia in humans. Pseudotime, differentiation trajectories and MRC1-positive/P2RY12-positive transitional cells all support this interpretation (Wang et al., 2023; Wu, Wang, et al., 2025). Direct genetic fate mapping was not possible because standard mouse and rat models lack this population. Together, these observations support a yolk-sac-associated developmental origin, while inferred relationship remains distinct from direct genealogical proof.

Developmental seeding and life-long maintenance are separate questions and require different experiments. Proliferation after injury and repopulation after withdrawal of a CSF1R inhibitor show that the PNS population can regenerate, but not which cells repopulate the niche (Wu, Wang, et al., 2025). Residual PNS microglia are one possibility. Another resident macrophage population or cells from definitive hematopoiesis could also contribute. Whether embryonically seeded cells persist throughout adult life by local self-renewal remains unknown.

For other peripheral populations with microglial molecular programs, evidence differs by tissue and criterion (Table 1). In human fetal epidermis, microglia-like cells contact neural-crest cells. These epidermal cells declined during gestation and were no longer detected after 20 postconception weeks. The cardiac-associated population was concentrated in the developing aorta, whereas the testicular population was enriched around efferent ducts in the epididymis; neither population has an established neural function (Wang et al., 2023). Coordinated expression programs are more informative than isolated markers, but direct perturbation and evidence for continuity into adult life remain sparse. Some of these cells may be transient prenatal macrophages that deploy a microglial molecular program; others may seed durable tissue populations. Current datasets cannot distinguish between these possibilities.

The distribution of PNS microglia across species adds a further dimension. Recent work has asked whether body size, lifespan, regenerative demand or neural architecture shaped peripheral microglial populations (Wu, Cui, et al., 2025). Their absence from standard laboratory rodents is both a practical limitation and a biological observation. Core programs may be conserved even when complement, phagocytic, metabolic and disease-susceptibility modules differ substantially among species (Geirsdottir et al., 2019). There is also a straightforward experimental problem: standard mouse models cannot reveal the normal function of a population they do not contain.

PNS microglia also offer an unusual cross-tissue comparison. They retain major elements of the CNS microglial program outside the brain parenchyma, yet sit between neuronal somata and satellite glia and are associated with neuronal size (Wang et al., 2023; Wu, Wang, et al., 2025). Comparison with CNS microglia may reveal which regulatory features belong broadly to neural macrophage niches and which depend on brain parenchyma, satellite glia, vascular access or neuronal geometry. This work shows how molecular atlases can uncover an unrecognized population and point toward its functions. It also brings the limits of molecular discovery: adult maintenance, responses to challenge, and the full range of effects on neurons and satellite glia still require longitudinal and perturbational experiments.

3. From molecular-state atlases to causal cellular function

3.1. State discovery and the validation gap

Comparative transcriptomics, sensome analysis and functional studies defined a TGF-β-dependent program associated with homeostatic microglia (Butovsky et al., 2014). P2RY12, TMEM119, HEXB and SALL1 are prominently expressed in microglia compared with other myeloid populations, but even these core signatures change under some conditions. Homeostatic genes decline during inflammation and neurodegeneration, while IBA1, CX3CR1, APOE, SPP1 and GPNMB are shared with other macrophage populations. Simply removing microglia from the brain rapidly remodels their transcriptome and enhancer landscape (Gosselin et al., 2017).

During development, microglia pass through successive embryonic and postnatal programs before acquiring adult homeostatic features (Matcovitch-Natan et al., 2016). Single-cell sequencing identified populations enriched for GPNMB, SPP1, CLEC7A, IGF1 and lysosomal and lipid-handling genes; several of these features reappear during aging and neurodegeneration (Li et al., 2019). Their recurrence in adult disease is consistent with renewed use of programs for proliferation, engulfment, growth support and lipid processing. It does not mean that the cells have reverted to an embryonic cell type.

Adult microglia are also heterogeneous. Regional analysis revealed differences in density, morphology, turnover and transcriptional programs, with white-matter, forebrain and border-associated macrophages acquiring distinct niche-related features (Van Hove et al., 2019). Across the lifespan, developmental, homeostatic, inflammatory, interferon-responsive and injury-associated configurations form shifting continua rather than a single activation axis (Hammond et al., 2019). Aging in white matter generates lipid-handling, inflammatory and antigen-presentation-associated states around myelin-rich substrates (Safaiyan et al., 2021). Human studies add species-specific and age-associated programs. Aged human microglia occupy several transcriptional states, accompanied by coordinated epigenetic remodeling (Li et al., 2023; Olah et al., 2018). Cross-species comparisons further show that a conserved core coexists with substantial primate divergence in complement, phagocytosis and regulatory networks (Geirsdottir et al., 2019). Sex-dependent transcriptional and metabolic differences add another layer of variation (Fumagalli et al., 2025).

3.2. Detecting and interpreting microglial states

Direct comparisons of tissue-processing methods show how easily the recovered microglial transcriptome can be altered by the experiment itself (Fig. 2). Marsh et al. found that conventional enzymatic dissociation of mouse brain at elevated temperature induced immediate-early and stress-response genes, with the strongest effect in microglia. Cold mechanical dissociation markedly reduced this program while preserving cellular diversity (Marsh et al., 2022). Single-nucleus RNA sequencing avoids whole-cell dissociation and can be used with frozen human tissue, but it samples nuclear rather than total cellular RNA. In matched cortical material from four neurosurgical donors, most transcripts were similarly represented in cells and nuclei. A small nuclear-depleted set, however, included several microglial state genes, among them APOE, SPP1, CST3 and CD74 (Thrupp et al., 2020). Whole-cell sequencing captures more cytoplasmic and activation-related RNA, but risks inducing part of that signal during isolation. Nuclear sequencing depends less on viable-cell recovery, but may miss elements of disease-associated programs. A rare or activated population is more convincing when it can be recovered with an independent preparation and then localized in intact tissue. In situ microglial Patch-seq takes a different route. Bakina et al. withdrew cytoplasm through a patch pipette from individual microglia in acute mouse brain slices, avoiding enzymatic dissociation and FACS while retaining anatomical origin, ramified morphology and membrane-current measurements (Bakina et al., 2024). The dataset contained 113 microglia from cortex, hippocampus and corpus callosum, with an average of 4 138 detected genes per cell. Two stress-associated clusters found in dissociated, FACS-isolated microglia were absent, and the FACS-associated marker Cdh23 occurred in fewer than 5% of microglia in intact tissue compared with 55% of isolated cells. These results indicate that isolation contributes to at least part of the stress program. Patch-seq, of course, is not artifact-free. Acute slicing perturbs tissue, extracellular RNA may contaminate the pipette, amplification from small RNA quantities is noisy, and the modest sample size limits the detection of regional differences. Even so, the method offers an orthogonal test of atlas-scale data by linking transcription, physiology and anatomical origin in the same cell.

Fig. 2.

From molecular-state discovery to cellular function.

Tissue sources represent development, adult homeostasis, aging, amyloid and tau pathology, demyelination, and injury. Bulk, single-cell, single-nucleus, spatial and Patch-seq approaches provide complementary views of microglial molecular state, each with distinct sampling and technical limits. The heatmap, colors and dendrogram in the “Inferred functions” panel are illustrative only. Molecular patterns nominate candidate functions. Direct cellular assays measure the proposed activities, and genetic perturbation, pathway modulation and rescue test their mechanisms. Matched measurements of neurons, synapses, myelin or lesions establish tissue effects in the relevant context. The workflow depicts an experimental strategy; biologically, cellular activity can also feed back onto molecular state. RNA-seq, RNA sequencing; Patch-seq, combined patch-clamp recording and transcriptomic profiling; TBI, traumatic brain injury.

Single-cell atlases are snapshots, although many of the questions they raise concern the history of individual cells. Serial sampling can order population changes across development, aging or injury. Pseudotime and RNA velocity go further by inferring likely transitions from similarities among destructively sampled transcriptomes, including the ratio of unspliced to spliced RNA (Hammond et al., 2019; La Manno et al., 2018; Matcovitch-Natan et al., 2016). The catch is that these trajectories remain predictions: cells placed at the beginning and end of a path are not the same cells, and the result depends on sampling, gene selection and assumptions about transcriptional kinetics (Bergen et al., 2020; Gorin et al., 2022). Fate mapping records a different kind of information. After stroke, Spp1-based tracing showed that neonatal Spp1-history-positive microglia could downregulate the injury program, regain homeostatic markers and reintegrate into the microglial network. In juvenile mice, the comparable DAM-like population did not reverse and was eventually eliminated (Lan et al., 2024). The neonatal cells nevertheless remained sensitized to a later immune challenge, showing that molecular homeostasis need not erase cellular history. Multicolor fate mapping revealed another route to recovery: selected clones expand after injury, and excess reactive cells are later removed by apoptosis and egress as the network is restored (Tay et al., 2017). Pseudotime can nominate these paths, but state-specific fate mapping is needed to distinguish reprogramming of surviving cells from selective survival, clonal replacement or loss.

State nomenclature captures both the value and the limits of molecular classification. In amyloid models, disease-associated microglia (DAM) describe a two-stage transition: an initial TREM2-independent response includes loss of homeostatic features and APOE induction, whereas a later TREM2-dependent program induces lipid-handling and phagolysosomal genes, including LPL, CST7 and CLEC7A (Keren-Shaul et al., 2017). Related responses have been called the microglial neurodegenerative phenotype (MGnD), activated-response microglia (ARM) and interferon-response microglia (IRM). Studies of aging and anatomical specialization have added white-matter-associated microglia (WAM) and lipid-droplet-accumulating microglia (LDAM) (Krasemann et al., 2017; Marschallinger et al., 2020; Safaiyan et al., 2021; Sala Frigerio et al., 2019). At first glance, this reads like a catalogue of separate populations. It is not. These labels describe overlapping responses defined across different species, models, regions, ages, isolation methods and clustering strategies, and they share combinations of lipid-handling, phagolysosomal, inflammatory, interferon and antigen-presentation modules. Human single-cell and single-nucleus studies have resolved additional antigen-presentation-enriched and TREM2-independent responses, while spatial transcriptomics has placed such programs relative to plaques, neuritic injury and neighboring cells (Chen et al., 2020; Mathys et al., 2019; Zhou et al., 2020). The Human Microglia Atlas likewise identifies recurrent as well as disease-biased states across neurodegenerative disorders (Martins-Ferreira et al., 2025). A state label summarizes a molecular or phenotypic pattern recovered in a defined context. It helps locate cells and nominate candidate activities, while direct functional assays determine which activities those cells perform. Table 2 compares the defining contexts, methods, markers and limits of functional evidence for six widely used labels.

Table 2.

Definitions and functional evidence for commonly used terms of microglial states.

State and source	Species and model	Region and substrate	Isolation and profiling	Representative genes or features	TREM2 dependence	Function tested
Disease-associated microglia (DAM) (Keren-Shaul et al., 2017)	Primarily 5xFAD mice; related signatures in other neurodegeneration models; human histological comparison.	Amyloid-bearing brain; plaque-proximal cells and Aβ cargo.	Dissociated, FACS-sorted brain immune cells; MARS-seq single-cell profiling; in situ validation.	Apoe, Lpl, Cst7, Clec7a and Itgax up; P2ry12 and Tmem119 down during transition.	Initial response is partly independent; progression to the later DAM program requires TREM2 in this model.	Intracellular Aβ and plaque localization documented; Trem2 loss tests disease progression.
Microglial neurodegenerative phenotype (MGnD) (Krasemann et al., 2017)	Mouse APP-PS1, SOD1 and EAE models; apoptotic-neuron challenge; human disease comparisons.	Diseased CNS and sites of neuronal death; apoptotic neuronal material.	Isolated/sorted microglia; expression profiling including bulk RNA-seq; histological validation.	Apoe, Axl, Clec7a, Itgax and Lgals3 up; P2ry12, Tmem119 and Sall1 down.	TREM2-APOE signaling required for the full induced program in tested paradigms.	Apoptotic-neuron uptake and Trem2/Apoe perturbation link cargo exposure to program induction; immune-regulatory assays performed.
Activated-response microglia (ARM) (Sala Frigerio et al., 2019)	AppNL-G-F mice across age and sex; APP/PS1 comparison and Apoe deletion.	Cortex and hippocampus; aging and amyloid plaques.	Cold-prepared, FACS-sorted live CD11b-positive cells; Smart-seq2; in situ validation.	Apoe, Cst7, Clec7a, Itgax; subsets enriched for Cd74/H2 genes or Spp1/Gpnmb.	Not directly tested in the defining study; Apoe deletion markedly reduced ARM acquisition.	Age, sex and Apoe effects on state occupancy and plaque association measured.
Interferon-response microglia (IRM) (Sala Frigerio et al., 2019; Roy et al., 2020)	AppNL-G-F and aging mice; related interferon responses in other amyloid models and human tissue.	Cortex and hippocampus; interferon-associated stimuli; nucleic-acid-containing plaques in related work.	Cold preparation and FACS/Smart-seq2 in the defining atlas; separate in situ and pathway-intervention studies.	Ifit2, Ifit3, Ifitm3, Irf7 and Oasl2; representative interferon-response genes.	Not directly established in the defining atlas; IRM persisted after Apoe deletion.	IFN-β and IFNAR interventions test pathway activity and complement-associated responses in related amyloid studies.
White-matter-associated microglia (WAM) (Safaiyan et al., 2021)	Aged mice; Trem2 and Apoe loss; amyloid-model comparisons.	White matter, including corpus callosum; degenerating myelin.	Separate white/gray-matter dissection; FACS and Smart-seq2, with bulk profiling and imaging.	Apoe, Lpl, Cst7, Clec7a, Lgals3 and Spp1; phagolysosomal and lipid-handling modules.	TREM2-dependent but APOE-independent in aging; APOE dependence differs in amyloid models.	Myelin uptake and phagocytic nodules visualized; perturbations test WAM emergence.
Lipid-droplet-accumulating microglia (LDAM) (Marschallinger et al., 2020)	Aged mice; LPS and Grn-deficient models; aged human hippocampal histology.	Prominent in aged hippocampus; intracellular neutral-lipid droplets.	BODIPY-high/low microglia sorted by FACS; bulk RNA-seq, lipidomics and imaging.	Droplets are defining; Plin3 and Acly are enriched in mouse lipid-high cells; inflammatory program. No universal gene-only classifier.	Not established in the defining study; neither droplet abundance nor overlap with DAM proves dependence.	Mouse phagocytosis, reactive oxygen species and cytokine assays; CRISPR screen for droplet regulators.

3.3. From state-associated candidates to cellular function

Expression programs point to processes that can be tested. Enrichment of receptors, lysosomal enzymes, metabolic proteins, cytokines or trophic factors may suggest altered sensing, cargo handling or intercellular communication. Transcript abundance alone, however, does not establish pathway activity or tissue outcome (Fumagalli et al., 2025; Paolicelli et al., 2022). A causal study follows a simple sequence: reproduce and localize the state, select a candidate regulator or effector, perturb it in the relevant population and time window, measure the predicted cellular process, and then measure the tissue consequence. Rescue offers stronger evidence that these steps belong to the same causal chain (Fig. 2).

The DAM transition illustrates how a molecular state can nominate a cellular mechanism. In the original amyloid model, an initial TREM2-independent response preceded a later TREM2-dependent program (Keren-Shaul et al., 2017). Genetic perturbation studies linked TREM2 to microglial survival, lipid sensing, energetic fitness (Ulland et al., 2017; Wang et al., 2015). These experiments establish receptor-dependent cellular processes. 3, 4 consider how these cellular effects influence disease outcomes.

In developing white matter, a transient CD11c-positive population expressed Igf1, Spp1, Lgals1 and other genes associated with neuronal support and myelin formation (Wlodarczyk et al., 2017). Depleting Igf1 from CD11c-positive cells reduced myelin-gene expression and increased the frequency of less-myelinated axons in the corpus callosum. This experiment linked a state-associated trophic factor to a developmental tissue outcome without assigning the same function to every co-expressed gene.

Interferon-stimulated genes define another recurrent state in amyloid models. Microglia expressing these genes accumulate around nucleic-acid-containing plaques in mice and human AD tissue (Roy et al., 2020). Administration of IFN-β to the brain induced microglial activation and complement C3-dependent synapse elimination, whereas IFNAR blockade reduced microgliosis and synapse loss. The intervention links interferon signaling to complement engagement and synaptic pathology. Because it was not restricted to microglia, however, the phenotype cannot be assigned exclusively to IFN-stimulated gene-expressing microglia.

Lipid-droplet-accumulating microglia (LDAM) in aged mouse brains show an innate-inflammatory transcriptome together with impaired phagocytosis, increased reactive oxygen species and enhanced cytokine secretion; lipid-droplet accumulation was also observed histologically in aged human microglia (Marschallinger et al., 2020). Persistent droplet accumulation is associated with dysfunction, but low lipid accumulation may not necessarily mean functionally fit microglia. The amount of lipid inside a cell does not tell us how efficiently that lipid is being cleared. In an amyloid model, inhibiting DGAT2-dependent lipid synthesis restored Aβ uptake ex vivo, providing a perturbational test of a cellular defect (Prakash et al., 2025). Nevertheless, droplet abundance is a pool size, not a measure of flux: it reflects uptake, esterification, storage, lipolysis, oxidation and export. Lipid-species measurements and uptake-chase or efflux assays are therefore needed to distinguish adaptive storage from failed processing. The contrasting TREM2 phenotypes and their consequences for repair are considered together in 3, 4.

Functional assays need to follow the process suggested by the molecular data. Complement and phagocytosis programs can be examined through synapse number, cargo handling and circuit activity (Badimon et al., 2020; Parkhurst et al., 2013; Schafer et al., 2012), whereas trophic programs call for measurements of neuronal survival, myelination or neurogenesis (Sierra et al., 2010; Ueno et al., 2013; Wlodarczyk et al., 2017). Lipid-handling states are better tested by metabolic flux, cholesterol export, debris clearance and remyelination than by droplet abundance alone (Hughes & Appel, 2020; Kent & Miron, 2024; McNamara et al., 2023). Molecular states identify functions worth testing; perturbation shows whether those predictions are correct.

3.4. A staged path to human functional validation

Human postmortem and surgical tissues establish clinical existence, anatomy and molecular context, but they are largely cross-sectional and are influenced by treatment, agonal state, postmortem delay and sampling (Bao et al., 2024). These materials can define a state and its niche, but cannot provide the sole evidence for mechanism. Even the choice between cells and nuclei changes how activation-associated transcripts are recovered (Marsh et al., 2022; Thrupp et al., 2020).

Cultured human microglia do not reproduce an entire in vivo disease state. Their role is narrower: to test proximal cellular processes under conditions informed by human tissue data. Isogenic iPSC-derived microglia allow scalable genetic perturbation of survival, proliferation, cargo-specific uptake, lysosomal processing and inflammatory responses (Abud et al., 2017). CRISPRi and CRISPRa screens have already linked candidate genes to microglial survival, activation and synaptosome phagocytosis, with CRISPR droplet sequencing (CROP-seq) relating perturbations to state occupancy (Dräger et al., 2022). These readouts are cellular phenotypes, not disease outcomes.

Selected hits can then move into a model that restores the relevant niche and substrate. Neuron-glia co-cultures and multicellular neurospheres permit controlled measurements of human cell-cell interactions (Wendt et al., 2025). Neuroimmune organoids add tissue organization, while xenotransplantation promotes further maturation in a living brain (Han et al., 2023; Hasselmann et al., 2019; Schafer et al., 2023). Isogenic TREM2-knockout microglia illustrate this staged approach. In vitro assays identified defects in survival, phagocytosis and chemotaxis; xenotransplantation into an amyloid model then tested plaque association in vivo (McQuade et al., 2020).

A practical workflow begins by establishing the molecular state and anatomical context in human tissue, measures candidate cellular activities in suitable human-cell models, and uses perturbation and rescue to test the underlying mechanisms. Multicellular or xenograft systems can then establish how these activities affect neighboring cells and tissue. Disagreement among systems may itself be informative, pointing to inadequate maturation, missing ligands, different lesion kinetics or unequal substrate burdens. Until perturbation connects a state to a cellular process and tissue effect, neutral molecular labels remain preferable to functional names (Fumagalli et al., 2025; Paolicelli et al., 2022). And even when a cellular function has been established, its contribution to disease still needs to be tested separately.

4. From cellular function to context-dependent disease impact

4.1. TREM2: recurring cellular actions, divergent disease outcomes

Establishing a cellular function by perturbation still leaves a second gap: that function does not, by itself, predict tissue or disease impact. TREM2 makes the problem particularly clear (Fig. 3). Across amyloid deposition, demyelination, ischemia and glioma, the receptor repeatedly influences myeloid-cell survival, accumulation, metabolic adaptation and responses to lipid-rich material. The cellular effects recur; their consequences do not. Plaque compaction and removal of damaged myelin can be beneficial, while persistent post-ischemic inflammation or support of tumor-associated vasculature is not. TREM2 can therefore be understood only in relation to the substrate, anatomical context and disease stage.

Fig. 3.

TREM2: recurring cellular actions and context-dependent disease outcomes.

TREM2 controls a core set of microglial programs — phagocytosis and clearance, migration and clustering, metabolic reprogramming, survival and proliferation, and homeostatic-to-disease state transition. These proximal cellular actions interact with the microenvironments and disease phase to produce different tissue effects. TREM2-dependent plaque compaction in Alzheimer’s disease and myelin-debris processing during demyelination are generally protective. Stroke outcomes depend on the experimental context: TREM2 deficiency worsens recovery after middle-cerebral-artery occlusion (MCAO), whereas deficiency reduces infarct size and impairment after photothrombosis. TREM2 deficiency reduced myeloid accumulation, vascular density and tumor growth in male mice; the same tumor-growth effect was not detected in females. The panels summarize the direction of the reported TREM2-associated tissue effect relative to deficiency and do not imply a universal effect across disease stages or sexes. TREM2, triggering receptor expressed on myeloid cells 2; GL261, a murine glioma cell line.

TREM2 is selected here as a particularly well-characterized example, not a unique route from state to function. APOE perturbation changes the neurodegeneration-associated microglial program, the AD-protective PLCG2 P522R variant alters Aβ and synaptosome uptake in an allele-dose-dependent manner in human iPSC-derived microglia, and reduced INPP5D activity activates the NLRP3 inflammasome and IL-1β/IL-18 release in human microglial models (Chou et al., 2023; Krasemann et al., 2017; Solomon et al., 2022). These examples reinforce the need to test a defined allele or perturbation, substrate and tissue endpoint rather than infer benefit from a state label.

Microglial proliferation is often used as one of the readouts for the TREM2 phenotype, although cell number may reflect division, survival or migration. In amyloid models, TREM2-deficient microglia accumulate poorly around plaques and show increased apoptosis; impaired mTOR signaling and energetic fitness further limit their response to tissue damage (Ulland et al., 2017; Wang et al., 2015). TREM2 deficiency also prevents the usual age-associated increase in microglial numbers in several brain regions (Poliani et al., 2015). The cuprizone model has a more complicated time course. Early microgliosis is attenuated in TREM2-deficient mice, but microglia persist during recovery while myelin debris remains unresolved (Poliani et al., 2015). IBA1-positive cell numbers are also reduced in the TREM2-deficient photothrombotic model (Chen et al., 2026). A cell count at any one time point combines proliferation, survival, recruitment, retention and persistence of the lesion. Taken together, these experiments support a broader interpretation of TREM2 as a regulator of myeloid fitness under challenge.

In amyloid pathology, this support contributes mainly to a protective tissue response. Rare coding variants first linked TREM2 to AD risk (Jonsson et al., 2013), and experimental studies connected the receptor to lipid sensing, survival, plaque-associated expansion, chemotaxis, lysosomal activity and metabolic adaptation (Ulland et al., 2017; Wang et al., 2015). TREM2-dependent microglia surround early plaques, restrict amyloid diffusion, compact the deposits and reduce the exposure of nearby neurites to fibrillar Aβ (Wang et al., 2016; Yuan et al., 2016). Restoring metabolic fitness in TREM2-deficient microglia improves plaque engagement and reduces neuronal dystrophy, implicating cellular fitness in this barrier response (Ulland et al., 2017). The pathway is not unitary, however. TREM2 can signal through SYK-dependent and SYK-independent routes, and Fc-receptor stimulation increases antibody-assisted Aβ uptake without fully restoring the overall clearance capacity of TREM2-deficient cells (Wang et al., 2022; Xiang et al., 2016).

A related cellular repertoire operates during the repair of demyelinated lesions. TREM2 detects lipids exposed by damaged myelin and sustains programs for uptake, intracellular processing, trophic support and lesion-associated expansion (Cantoni et al., 2015; Poliani et al., 2015). Receptor agonism enhances myelin uptake and degradation, improves recovery of the oligodendrocyte lineage, and promotes remyelination and axonal integrity (Cignarella et al., 2020). Competent phagocytes are helpful here because inhibitory myelin debris must be removed before repair can proceed. Expansion alone is not enough: the ingested material must be degraded, and the microglial response must be coupled to oligodendrocyte differentiation.

Stroke illustrates just how much the interpretation depends on the lesion model and observation window. After middle-cerebral-artery occlusion (MCAO), TREM2 deficiency impairs chemotaxis or phagocytic activity and worsens neurological outcome (Kawabori et al., 2015; Mazaheri et al., 2017). In a photothrombotic model with a stronger TREM2 response, however, TREM2 depletion reduces infarct volume, inflammatory cytokines and neurological impairment. Soluble TREM2 aggravates injury, with GPNMB acting as a downstream effector (Chen et al., 2026). The models differ in vascular insult, lesion kinetics and the magnitude of TREM2 induction. These studies demonstrate different outcomes under different experimental conditions. Fig. 3 therefore identifies the model-specific findings explicitly.

Glioma presents a very different picture. In male mice, TREM2 deficiency reduces the accumulation of glioma-associated microglia and macrophages, lowers blood-vessel density and restrains tumor growth. Notably, the same study did not find this effect in female mice (Chen et al., 2023). TREM2 deficiency also failed to alter tumor size in an organotypic slice system lacking functional vascularization and peripheral immune input. This result points to macrophage-vascular interactions rather than a direct effect on glioma-cell growth (Chen et al., 2023).

Cargo handling provides a mechanistic explanation for some context-dependent tissue effects. Clearance proceeds through recognition, internalization, lysosomal degradation, metabolic processing or export, and finally resolution of the lesion. Finding cargo inside a cell proves that uptake occurred; it does not answer whether the remaining steps were completed. Myelin debris delays repair when it is not removed, whereas defective cholesterol export generates intracellular crystals and impairs remyelination even when uptake continues (Cantuti-Castelvetri et al., 2018; Lampron et al., 2015). During acute lysolecithin demyelination, TREM2-deficient phagocytes internalized myelin but failed to mount an adequate lipid-droplet response, accumulated free cholesterol and endoplasmic-reticulum stress, and supported poorer remyelination; ACAT deficiency likewise impaired droplet formation and repair (Gouna et al., 2021). During chronic cuprizone challenge, TREM2-deficient microglia instead accumulated cholesteryl esters because cholesterol processing was defective (Nugent et al., 2020). The fate of stored lipid is equally important. Plin2 deficiency accelerated lipid-droplet turnover and cholesterol efflux in myelin-loaded phagocytes and improved remyelination (Loix et al., 2022). In 5xFAD mice, deficiency of DGAT2, a key enzyme that facilitates triacylglycerol biosynthesis, reduced plaque-associated microglial lipid droplet formation as well as amyloid burden and neuritic dystrophy (Prakash et al., 2025). Therefore, the useful question is not simply whether microglia contain more or less lipid. It is which step of lipid flux has changed, and what cargo remains inside and outside the cell.

The therapeutic implication is direct: a conserved microglial pathway may drive a similar cellular response yet produce a protective or harmful outcome depending on the disease context, substrate and timing.

4.2. From context-dependent mechanisms to therapy

TREM2-directed antibodies have been reviewed in detail elsewhere (Colonna & Holtzman, 2025; Schlepckow et al., 2023). The question here is when signaling should be enhanced and when it should be restrained. Preclinical support for agonism is strongest where TREM2-dependent microglial fitness and cargo processing contribute to tissue protection. A stalk-directed antibody stabilizes surface TREM2, reduces shedding and activates SYK-dependent responses (Schlepckow et al., 2020), while receptor agonism enhances myelin-debris processing and remyelination (Cignarella et al., 2020). These findings support testing TREM2 activation when plaque engagement, metabolic adaptation or removal of damaged myelin is limiting. They do not establish clinical efficacy or show that sustained activation benefits at every stage of disease; the unsuccessful INVOKE-2 trial raised further cautions against extrapolating from target engagement to patient benefit (Colonna & Holtzman, 2025).

TREM2 antagonism addresses the opposite setting. Genetic loss of TREM2 reduced secondary injury in a photothrombotic stroke model and restrained vascularized tumor growth in male mice in one glioma model. These findings raise the possibility that transient inhibition could be useful when TREM2 sustains damaging inflammation or macrophage support of pathological tissue (Chen et al., 2023, 2026). The evidence is far less mature than that for agonism. Constitutive deficiency is not equivalent to timed pharmacological blockade, and the reported glioma phenotype was sex-dependent. Moreover, acute antibody-mediated TREM2 inhibition during cuprizone demyelination depleted a MAFB-high microglial population and impaired remyelination (Hou et al., 2025). Antagonism would therefore need to be reversible and restricted to the appropriate disease stage, with the intended cellular effect and tissue outcome measured together. It should not be treated as the general inverse of agonist therapy.

5. Future directions

Microglial biology has advanced rapidly, but much of its mechanistic foundation still rests on rodent studies of the CNS. Progress now depends on human-relevant systems, direct analysis of populations absent from standard rodents, and experiments that connect molecular states to cellular actions and tissue outcomes. No single model spans the entire sequence. Human tissue anchors a state in disease; scalable human-cell systems test causal cellular functions; multicellular models reveal effects on neighboring cells; and studies in an intact neural niche establish the tissue consequence.

5.1. Human and non-human primate studies

Human postmortem, biopsy and surgical materials reveal which microglial states occur in patients and where they are positioned, although their cross-sectional nature limits causal interpretation. Their value increases when transcriptomic, spatial, epigenetic and proteomic analyses are matched across brain regions, ages, genotypes and disease stages. Replication across cohorts and preparation methods can separate recurrent programs from signals introduced by sampling or clinical history. Cross-species comparisons can then identify which parts of a state are conserved and which are primate-specific. A conserved homeostatic core, for example, coexists with marked divergence in complement, phagocytic and disease-susceptibility programs (Geirsdottir et al., 2019; Masuda et al., 2019). Non-human primates and other naturally occurring species are especially important when rodents lack the relevant cell population or neural architecture.

5.2. Niche-informed human models

iPSC-derived microglia offer controlled genetics and experimental scale, but they are not substitutes for human microglia in vivo. Culture conditions reshape their transcriptional and epigenetic programs, and monocultures omit signals from neurons, astrocytes, oligodendrocytes, vascular cells and extracellular matrix (Abud et al., 2017; Gosselin et al., 2017). Defined brain substrates can nevertheless generate useful experimental diversity. Exposure to synaptosomes, myelin debris, apoptotic neurons or fibrillar amyloid-β produced transcriptional states that mapped partly onto microglial states in the human brain and allowed candidate state regulators to be manipulated functionally (Dolan et al., 2023). In practice, these models are most informative when baseline identity, substrate response and the direction of the induced program are benchmarked against patient material before the functional phenotype is interpreted.

Different human models address successive levels of causality. Isogenic iPSC pairs isolate the effect of a disease variant; neuron-microglia and oligodendrocyte-microglia cultures reveal direct cellular interactions; and neurospheres, organoids and microfluidic systems add multicellular organization, allowing live measurements of neuronal activity, myelination or barrier function (Schafer et al., 2023; Wendt et al., 2025). Xenotransplantation exposes human microglia to a living brain niche and allows plaque association and other spatial responses to be studied, although the surrounding tissue remains non-human (Hasselmann et al., 2019; McQuade et al., 2020). Organotypic slice cultures prepared from neurosurgical human brain tissue retain local cellular architecture and permit short-term imaging, physiological measurements and experimental perturbation. Adult human cortical slices can remain viable in culture, and single-nucleus analysis has shown substantial preservation of microglial and other cell-type identities for up to two weeks ex vivo (McGinnis et al., 2026; Park et al., 2022). The model has clear constraints: resection and slicing cause injury, circulation is lost, tissue availability is limited and patients differ. Even with these caveats, it provides a valuable bridge between cultures and the intact patient brain. For the PNS, recreating the corresponding niche will require sensory or autonomic neurons together with satellite glial cells.

5.3. A faster route from molecular state to tissue impact

Combining different experimental models can accelerate the functional analysis, rather than asking one system to reproduce an entire disease. Human atlases can first identify recurrent states, candidate regulators and the anatomical substrate. CRISPRi, CRISPRa and knockout screens in isogenic human microglia can then test proximal actions such as survival, proliferation, migration, cargo uptake, lysosomal processing, lipid export and secretion (Dräger et al., 2022). Pooled CRISPR screens mix many perturbations in one culture, but the guide RNA carried by each cell can be recovered by sequencing and linked to phenotypes measured in that same cell. Effects mediated by secreted factors or interactions with neighboring neurons and glia are easier to resolve in arrayed screens, where each well contains a single defined perturbation and can be assessed by automated imaging (Dräger et al., 2022). In either case, state occupancy and cellular activity are complementary readouts. A shift in the abundance of a transcriptional state becomes functionally interpretable when linked to a measured change in activity.

The strongest candidates can move into multicellular human models and then into organotypic preparations or xenografts. Cellular and tissue endpoints need to remain paired throughout. A lipid-handling experiment, for example, can quantify uptake, degradation and export alongside extracellular debris and myelin recovery. A trophic or synaptic mechanism can likewise be tested through both the microglial response and the resulting neuronal structure or activity. Rescue of the cellular defect and the tissue phenotype in the same experiment provides the crucial causal link. Focused in vivo Perturb-seq libraries offer another route for a limited number of atlas-prioritized genes. Current AAV approaches, however, recover relatively few microglia per perturbation and require careful control of transduction efficiency and vector-induced responses (Okada et al., 2022; Santinha et al., 2023).

The time window for intervention is another consideration. Perturbation before pathology, during the emergence of a state and after disease is established separates initiation from maintenance and repair. Fate mapping and state-history labeling distinguish reprogramming of surviving cells from selective survival, clonal expansion or loss (Lan et al., 2024; Tay et al., 2017). For a context-dependent pathway such as TREM2, the therapeutic window becomes visible only when the proximal cellular action and tissue consequence are measured at the same disease stage.

5.4. Functional specialization of PNS microglia

PNS microglia are a major test of this human-centered strategy. They form a defined interface with neuronal somata and satellite glia and regulate neuronal growth during development, yet the causal contributions of specific cellular actions to adult sensory physiology, peripheral neuropathy, chronic pain, nerve injury and regeneration remain to be established (Wu, Wang, et al., 2025). Human ganglion tissue can establish molecular state and anatomy. Fetal ganglion explants, sensory-neuron-satellite-glia cultures and organotypic ganglion preparations then permit perturbation and live functional analysis. Pigs and non-human primates provide the naturally occurring population and intact peripheral niche needed to connect these cellular effects to sensory and regenerative outcomes.

Adult maintenance remains a separate problem from developmental origin. Long-term labeling, clonal analysis and depletion-repopulation experiments in species that contain PNS microglia can distinguish local self-renewal from input by another resident or hematopoietic source. The answer will show whether the developmental relationship inferred from prenatal atlases extends through adult life and which experimental systems are suitable for studying PNS microglia in disease.

6. Concluding remarks

The discovery of PNS microglia exemplifies how molecular atlases can identify novel populations and direct initial functional inquiry, moving the field beyond its traditional CNS-centric focus. Within an identified population, molecular state shapes functional capacity, while the activities performed by the cell can reshape that state. The relationship is reciprocal and allows several states to support a shared activity, or one state to support several activities. Molecular measurements therefore generate specific functional hypotheses; direct assays establish activity, and perturbation tests its mechanisms and consequences. As the case of TREM2 clearly demonstrates, the same cellular action can yield divergent tissue outcomes depending on the substrate, context, and disease stage. Consequently, translating molecular discoveries into meaningful biological insight requires a staged experimental strategy that rigorously connects a molecular state to a defined cellular function and ultimately to its context-dependent impact on tissue health.

CRediT authorship contribution statement

Cuihua Li: Writing – original draft, Writing – review & editing, Investigation, Visualization. Zhisheng Wu: Writing – original draft, Investigation. Xianyuan Xiang: Writing – review & editing, Writing – original draft, Investigation, Funding acquisition, Conceptualization. Hanjie Li: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.