Microglia depletion after brain ischaemia

The authors showed that removing microglia lets astrocytes pour out inflammatory mediators after stroke, and that putting microglia back into the dish quiets the astrocytes, but they never identified the signal that does the quieting, and they say so in the Discussion.

Question

Does microglial signalling through astrocyte purinergic receptor P2Y1 account for the restraint of astrocyte inflammatory output after transient focal cerebral ischemia?

Hypothesis

Knocking down P2Y1 in astrocytes will reproduce microglial restraint and will abolish the extra infarct growth caused by microglia depletion after sixty minutes of middle cerebral artery occlusion.

Attempt type

wet-lab

Design

Two stages in male C57BL/6 mice on the published protocol: PLX3397 at 40 mg per kg per day in AIN-76A chow for 21 days before surgery and onward, or AIN-76A chow alone, then sham or 60 minutes of middle cerebral artery occlusion with the published reperfusion criteria. Stage one measures astrocyte P2Y1 (gene P2ry1) at 24 hours by RNA in situ hybridisation on glial fibrillary acidic protein positive cells and by the paper's own intracellular flow cytometry panel, asking whether depletion changes P2Y1 rather than assuming it. Stage two is a two by two design: adeno-associated virus carrying a glial fibrillary acidic protein promoter and either a P2ry1 short hairpin or a non-targeting sequence, injected into cortex four weeks before surgery, crossed with PLX3397 or control chow. Primary endpoint is infarct volume on 7 tesla T2 imaging at 24 hours, read blind. Secondary endpoints are the paper's astrocyte interleukin 1 alpha, interleukin 1 beta, inducible nitric oxide synthase, tumour necrosis factor alpha and interleukin 6 panel, and modified Neurological Severity Score. The test that matters is the interaction term, not either main effect.

Controls

• designated Flow cytometry count of CD11b positive CD45 intermediate microglia in every animal, with the published target of roughly 90 percent reduction.
  Rules out: That a null result came from a failed depletion rather than from an absent mechanism. Without this, a null is inconclusive-no-positive-control, not evidence of absence.
• internal-positive Sham versus occlusion arms on control chow, which must reproduce the ischemia-driven rise in astrocyte inflammatory mediators reported in Figure 5.
  Rules out: That the astrocyte flow cytometry panel was insensitive on the day, which would otherwise be misread as the knockdown having worked.
• vehicle Adeno-associated virus with the same promoter, titre and injection coordinates but a scrambled hairpin.
  Rules out: That cortical injection, viral load or promoter-driven overexpression in astrocytes, rather than loss of P2Y1, changed infarct size.
  Vocabulary note: A sequence-matched non-targeting adeno-associated virus is only loosely a vehicle in the sense the vocabulary intends; it is the same delivery vector with inert cargo.
• no-treatment Sham-operated mice on control chow with no virus.
  Rules out: That surgery and anaesthesia alone account for astrocyte activation or for the behavioural scores.

Sensitivity

With 15 animals per group, two-sided alpha 0.05 and power 0.8, a two-group comparison detects a standardised difference of about 1.0 standard deviations (2.80 times the square root of two fifteenths). The interaction contrast in a two by two design has twice the variance, so at 15 per cell the smallest detectable interaction is about 1.45 standard deviations, and 32 per cell are needed to bring that down to 1.0. The paper publishes no infarct volumes and no standard deviations, so this cannot be converted into cubic millimetres. A pilot of 8 control-chow occlusion mice measuring the standard deviation of 7 tesla infarct volume is needed before the stage two sample size is fixed.

Would count as

Supported → positive · opposed → negative-no-effect
Inconclusive if: The adeno-associated virus achieves less than a verified knockdown of P2ry1 in glial fibrillary acidic protein positive cells, or the depletion check falls short of the published roughly 90 percent, or the control-chow occlusion arm fails to show the Figure 5 astrocyte cytokine rise. In any of those cases the record should carry inconclusive-no-positive-control or inconclusive-reagent-failed rather than a negative.

Already done

• Microglia transform astrocytes into a neuroprotective, scar-forming phenotype by downregulating the astrocyte P2Y1 purinergic receptor, and astrocyte-specific P2Y1 overexpression increases neuronal damage while knockdown reduces it. (Shinozaki et al. 2017, Cell Reports, doi:10.1016/j.celrep.2017.04.047)
  So: This supplies the candidate mechanism and the astrocyte-specific genetic tools, but it was done in traumatic brain injury, so the ischemia version is still untested and is the gap this design fills.
• Activated microglia induce neurotoxic reactive astrocytes by secreting interleukin 1 alpha, tumour necrosis factor and complement C1q, which are necessary and sufficient together. (Liddelow et al. 2017, Nature, doi:10.1038/nature21029)
  So: This is the opposite direction of travel from the Jin result and was published the same year, so the design has to measure astrocyte state rather than assume a single sign, and any mechanism proposed must explain why microglia restrain astrocytes here and provoke them there.
• Binary A1 versus A2 or neurotoxic versus neuroprotective labels for reactive astrocytes are not supported; multiple molecular and functional parameters plus multivariate statistics are recommended instead. (Escartin et al. 2021, Nature Neuroscience, doi:10.1038/s41593-020-00783-4)
  So: The six-cytokine flow panel the paper used is too thin on its own, so this design keeps it only as a secondary endpoint and anchors the primary endpoint on infarct volume.
• Astrocyte-derived interleukin 33 drives microglial synapse engulfment, establishing bidirectional astrocyte to microglia signalling. (Vainchtein et al. 2018, Science, doi:10.1126/science.aal3589)
  So: Any single-direction mechanism claim has to contend with a feedback loop, which is why the interaction term and not a main effect is the primary test here.
• Minocycline-based microglial inhibition in a rat chronic post-surgical pain model shifted astrocyte phenotype through a CXCR7, phosphoinositide 3-kinase and Akt pathway. (Li et al. 2020, Journal of Neuroinflammation, doi:10.1186/s12974-020-01891-5)
  So: A second named candidate pathway exists in a different model, so a null result on P2Y1 would not license the conclusion that microglia act on astrocytes by no identified route.
• Genetic microglia depletion with CX3CR1-CreER and an inducible diphtheria toxin receptor in a Rose Bengal photothrombotic stroke model reduced infarct volume and raised anti-inflammatory transforming growth factor beta 1, Arg1, interleukin 10, interleukin 4 and Ym1 messenger RNA. (Li et al. 2021, Journal of Neuroinflammation, doi:10.1186/s12974-021-02127-w)
  So: The direction of the depletion effect flips with model and timing, so this design must stay inside the published 60 minute occlusion and reperfusion model and must not be read as settling the mechanism for stroke in general.

Searched: Searched Europe PMC and OpenAlex for the 350 works citing this paper filtered on microglia and astrocyte crosstalk, then direct searches for microglia restraining or suppressing astrocyte reactivity in middle cerebral artery occlusion, for microglia-derived transforming growth factor beta acting on astrocytes in ischemia, and for the P2Y1 and interleukin 33 routes. Found the candidate mechanisms listed above but found no paper that identifies the microglia-derived factor restraining astrocyte inflammatory output specifically after focal cerebral ischemia, so this issue appears still open as of October 2026.

Feasibility

A rodent stroke laboratory with a 7 tesla small-animal magnet, intracranial adeno-associated virus injection and intracellular cytokine flow cytometry, which is the capability set the original authors already have at Barrow Neurological Institute and Tianjin Neurological Institute. · 9 to 15 months, driven by the four week pre-surgical virus expression window on top of the 21 day chow pre-treatment · high, because the two by two design with an interaction endpoint needs roughly twice the animals the paper used and adds custom virus production and 7 tesla imaging time
Blockers: No published infarct volume standard deviation in this model, so the stage two sample size cannot be fixed without a pilot; Astrocyte-specific promoters driven by adeno-associated virus leak into other cell types and the leak has to be quantified per batch; Cortical virus injection is itself an injury near the territory being measured

Strongest objection

A specialist would say the design assumes the astrocyte response is a cause of the larger infarct when the paper only shows it alongside, and that astrocyte inflammatory output at 24 hours is plausibly downstream of the bigger infarct rather than upstream of it. Knocking down P2Y1 in astrocytes could change infarct size through calcium signalling, glutamate handling or blood flow with no involvement of microglia at all, and the interaction term would not distinguish that from the proposed mechanism.

Not derivable from what is published

• Standard deviation of 7 tesla infarct volume in this model, needed for any power calculation in absolute units
• Whether astrocyte P2Y1 is even changed by microglia depletion after ischemia, which is why stage one exists
• Adeno-associated virus serotype, titre, volume and coordinates that give reliable cortical astrocyte transduction in the peri-infarct territory
• Whether the paper's astrocyte flow cytometry gate on glial fibrillary acidic protein captures peri-infarct astrocytes or whole hemisphere astrocytes

Microglia depletion raised brain-infiltrating neutrophils, macrophages, CD4 positive T cells and natural killer cells, yet removing lymphocytes with Rag2 and gamma c double mutant mice and blocking monocyte homing with bindarit both failed to rescue the injury, and the authors concede the infiltration was never pinned to a mechanism.

Question

Does depleting neutrophils abolish the extra infarct growth caused by microglia depletion after sixty minutes of middle cerebral artery occlusion?

Hypothesis

Antibody depletion of neutrophils will remove most of the microglia-depletion-induced increase in infarct volume at 24 hours, the arm of leukocyte infiltration the original study never tested.

Attempt type

wet-lab

Design

A two by two design in male C57BL/6 mice. Factor one is PLX3397 at 40 mg per kg per day in AIN-76A chow for 21 days before surgery and continuing, versus AIN-76A chow alone. Factor two is anti-Ly6G clone 1A8 versus the matched rat IgG2a isotype clone 2A3, both given intraperitoneally starting 24 hours before surgery and repeated daily. All animals undergo 60 minutes of middle cerebral artery occlusion with the published laser Doppler criteria of residual flow below 20 percent and recovery above 50 percent within 10 minutes. Primary endpoint is infarct volume on 7 tesla T2 imaging at 24 hours, read blind. Secondary endpoints are the modified Neurological Severity Score, blood and brain neutrophil counts by the paper's flow panel, and the luminol reactive oxygen species signal at 200 mg per kg. Analysis is a two-way analysis of variance with the depletion by neutrophil interaction as the pre-registered primary test. Randomisation and blinding follow the paper, as do the exclusion criteria.

Controls

• isotype Rat IgG2a clone 2A3 at the same dose and schedule as anti-Ly6G clone 1A8.
  Rules out: That repeated antibody load, Fc receptor engagement or the injection schedule rather than neutrophil loss changed infarct size. This is the single most important control, because anti-Ly6G also marks some monocytes.
• vehicle AIN-76A standard chow alone, the paper's own comparator for PLX3397.
  Rules out: That the chow formulation or the 21 day feeding regime, rather than CSF1R inhibition, drives the effect.
• designated Two separate efficacy checks in every animal: the roughly 90 percent reduction in CD11b positive CD45 intermediate microglia, and a blood neutrophil count confirming antibody depletion.
  Rules out: That a null came from a failed depletion on either factor. The registry rule applies strictly here: with two interventions, a null licenses evidence of absence only if both efficacy checks pass.
• no-treatment Sham-operated mice in each of the four cells.
  Rules out: That neutrophil depletion or microglia depletion alone produces a deficit or an imaging signal without ischemia, which the paper established for PLX3397 but not for anti-Ly6G.

Sensitivity

The same arithmetic as the first design. With 15 per cell, alpha 0.05 and power 0.8, the interaction contrast detects about 1.45 standard deviations, and 32 per cell are needed for 1.0. Because the question is whether neutrophil depletion removes an effect rather than whether it creates one, a null here needs an equivalence margin, and the paper gives no infarct volume variance to set one. Two anchors are available instead: the roughly 90 percent depletion efficiency the paper reports, and the 60 percent increase in infarct size that Szalay and colleagues report for microglia elimination in a comparable ischemia model. If the real microglia-depletion effect is of that order, a rescue that leaves less than a third of it standing would be detectable at 15 per cell; a partial rescue would not be.

Would count as

Supported → positive · opposed → negative-no-effect
Inconclusive if: Blood neutrophil counts show incomplete depletion, or the control-chow occlusion arm does not reproduce the published aggravation, or mortality in the double-intervention cell exceeds the paper's own attrition pattern of 14 surgical deaths and 18 inadequate reperfusions in 294 animals, which would leave the interaction cell underpowered. A null with intact neutrophils is inconclusive-no-positive-control; a null with verified depletion and verified aggravation is negative-no-effect and does license the narrow claim that neutrophils are not the main carrier in this model at this timepoint. It does not license the claim that infiltrating leukocytes as a class are irrelevant, since the paper has already closed the lymphocyte and monocyte arms only partially.

Already done

• Reactive microglia engulf neutrophils at the edge of the ischemic lesion, local microglial loss in the core is associated with neutrophils accumulating first in perivascular spaces and then in parenchyma, and long-term CSF1R inhibitor depletion increased neutrophil numbers and enlarged the lesion in permanent middle cerebral artery occlusion. (Otxoa-de-Amezaga et al. 2019, Acta Neuropathologica, doi:10.1007/s00401-018-1954-4)
  So: This largely answers the how for neutrophils, published two years after Jin, so the remaining gap is the causal rescue test, which is what this design is: nobody has asked whether removing neutrophils removes the injury penalty.
• A microglia depletion and repopulation paradigm with PLX5622 for seven days after ischemia followed by seven days of withdrawal suppressed T cell exhaustion, limited neutrophil terminal differentiation and promoted a phagocytic macrophage phenotype, improving recovery. (Shi et al. 2025, CNS Neuroscience and Therapeutics, doi:10.1111/cns.70565)
  So: Microglia-leukocyte interaction is now known to run in both directions and to depend on timing, so this design must keep the paper's continuous pre-treatment schedule and must not claim to speak about the repopulation phase.
• Early-phase microglia attenuation with PLX5622 in transient middle cerebral artery occlusion amplified a reparative monocyte-derived macrophage subset high in GPNMB and CD63 and promoted long-term functional recovery, with Mafb as the controlling transcription factor. (Zhang et al. 2025, Nature Communications, doi:10.1038/s41467-025-58254-y)
  So: Infiltrating myeloid cells that arrive after microglia are gone can be reparative rather than harmful, which means an increase in infiltration counts is not evidence of harm and the design has to measure injury, not cell numbers.
• Regulatory T cell derived osteopontin acts on microglial integrin receptors to drive white matter repair, and microglia depletion but not T cell lymphopenia abolished the benefit of transferred regulatory T cells. (Shi et al. 2021, Immunity, doi:10.1016/j.immuni.2021.04.022)
  So: A lymphocyte to microglia signal exists that the Rag2 and gamma c double mutant experiment in the original paper could not have detected, so the lymphocyte arm is less closed than the paper's null suggests.
• Microglia are the main immune cells of the brain and their selective elimination increased infarct size by 60 percent, reversed by microglial repopulation. (Szalay et al. 2016, Nature Communications, doi:10.1038/ncomms11499)
  So: This is the only published effect size available for the magnitude a rescue would have to remove, and it is the number the sensitivity statement above leans on.

Searched: Searched Europe PMC and OpenAlex across the works citing this paper for microglia and infiltrating leukocyte interaction after stroke, then direct searches for microglia regulating neutrophil accumulation, for microglia and T cell interaction in ischemic stroke, and for anti-Ly6G rescue experiments combined with CSF1R inhibition. Found the neutrophil phagocytosis mechanism and two 2025 papers on the repopulation phase, but found no study that combines microglia depletion with neutrophil depletion in the same cohort, which is the specific untested cell.

Feasibility

Any rodent stroke laboratory already running filament middle cerebral artery occlusion with flow cytometry and small-animal magnetic resonance imaging; no new capability is required beyond the antibodies. · 6 to 9 months · moderate, because the only additions to the published protocol are two antibodies and roughly double the animal numbers for an interaction endpoint
Blockers: Neutrophil depletion plus microglia depletion may raise infection risk and surgical mortality enough to unbalance the cells; Anti-Ly6G clone 1A8 is not perfectly neutrophil specific, so the isotype arm carries the whole weight of interpretation; No published variance for infarct volume in this model to set an equivalence margin for a partial rescue

Strongest objection

A specialist would object that neutrophil depletion by anti-Ly6G is incomplete in tissue even when blood counts fall to near zero, so a failed rescue may reflect residual tissue neutrophils rather than neutrophil irrelevance. They would also point out that the paper's own two null branches show the effect survives removal of whole leukocyte classes, which makes it likely that no single infiltrating population carries it and that a one-cell-at-a-time strategy will keep returning uninformative nulls.

Not derivable from what is published

• Dose and schedule of anti-Ly6G that gives verified tissue as well as blood neutrophil depletion in this strain and model
• Infarct volume variance needed for an equivalence margin on a partial rescue
• Whether the increased brain neutrophils in Figure 3 were perivascular or parenchymal, which the flow cytometry cannot distinguish and which the Otxoa-de-Amezaga result makes decisive
• Expected mortality in the double-depletion cell

This is the authors' own stated next step. Their microglia-conditioned medium reduced lactate dehydrogenase release from oxygen-glucose-deprived neurons, and they propose that microglia-derived factors calm ischemic neurons, but they never named a factor or measured neuronal activity at all.

Question

Is the protection that microglia-conditioned medium gives oxygen-glucose-deprived neurons carried by adenosine acting on the neuronal adenosine A1 receptor?

Hypothesis

Removing adenosine from microglia-conditioned medium, or blocking the neuronal adenosine A1 receptor, will abolish the reduction in neuronal death and excitability that microglia-conditioned medium produces after oxygen-glucose deprivation.

Attempt type

wet-lab

Design

Stage one repeats the paper's conditioned-medium transfer with added arms. Primary cortical neurons at the published density are given 5 minutes of oxygen-glucose deprivation, then neurobasal medium mixed one to one with conditioned medium taken from microglia, astrocytes or mixed glia 24 hours after 30 minutes of oxygen-glucose deprivation. Arms: medium alone; microglia-conditioned medium; microglia-conditioned medium pre-treated with adenosine deaminase; microglia-conditioned medium with an adenosine A1 receptor antagonist on the neurons; conditioned medium from microglia pre-treated with an inhibitor of the ectonucleotidase CD39; and astrocyte-conditioned medium. Readouts are the paper's own terminal deoxynucleotidyl transferase nick end labelling count and lactate dehydrogenase release, plus two new ones: adenosine and adenosine triphosphate concentrations in each medium, and spontaneous and glutamate-evoked activity on a multi-electrode array. Stage two is an in vivo two by two: PLX3397 or control chow crossed with an adenosine A1 receptor agonist or vehicle, with 60 minute middle cerebral artery occlusion and 7 tesla infarct volume at 24 hours, testing whether A1 agonism rescues the depletion penalty.

Controls

• no-treatment Primary cortical neurons with no oxygen-glucose deprivation and no conditioned medium, the paper's own control arm in Figure 6.
  Rules out: That handling, medium exchange or culture age rather than the insult produces the death and activity readings.
• internal-positive The astrocyte-conditioned medium arm, which the paper reports increases both nick end labelling counts and lactate dehydrogenase release.
  Rules out: That the assay was insensitive on the day. This is the designated positive control for the in vitro stage, and a null in the microglia arms counts as evidence of absence only if this arm moves.
• vehicle The solvent for the A1 receptor antagonist, the CD39 inhibitor and the agonist, at matched concentration, with heat-inactivated adenosine deaminase as the matched control for enzyme treatment.
  Rules out: That dimethyl sulfoxide, carrier protein or enzyme protein load rather than the targeted activity changed neuronal survival.
• designated In stage two, the flow cytometry check for roughly 90 percent loss of CD11b positive CD45 intermediate microglia, and in stage one the published culture purity staining for ionised calcium-binding adapter molecule 1 and glial fibrillary acidic protein.
  Rules out: That the microglia-conditioned medium was contaminated with astrocytes, which would make the comparison between the two media meaningless, and that an in vivo null came from failed depletion.
• within-subject Baseline activity recorded on the same electrode array before the medium is applied.
  Rules out: That between-well differences in neuronal density or electrode coupling, rather than the medium, account for activity differences.
  Vocabulary note: Applies only to the multi-electrode array arm, where each well is recorded before and after medium transfer.

Sensitivity

For the in vitro stage this cannot be derived. The paper gives no sample size at all for Figure 6 and no values for nick end labelling counts or lactate dehydrogenase release, so neither a variance nor an effect size is available. What is needed is a pilot of six independent culture preparations giving the standard deviation of the lactate dehydrogenase release fraction, after which a minimum detectable difference can be stated; saying so is the honest output here. For the in vivo stage two the earlier arithmetic applies: about 1.45 standard deviations for the interaction at 15 animals per cell, 32 per cell to reach 1.0 standard deviations.

Would count as

Supported → positive · opposed → negative-no-effect
Inconclusive if: Microglial cultures are not verified pure, or the astrocyte-conditioned medium arm fails to increase neuronal death, or the adenosine measurement sits below the assay's detection floor so that enzyme treatment cannot be confirmed to have worked. Note that the paper's own in vitro result is that microglia-conditioned medium alone had no significant effect on neuronal death, so the protective signal this design chases rests on the lactate dehydrogenase arm and on the mixed glia comparison, not on a clean microglia-alone benefit. A null here licenses only the claim that adenosine through the A1 receptor is not the carrier; it does not license the claim that no microglia-derived factor is protective.

Already done

• Microglia suppress neuronal activity through a negative feedback loop: they sense extracellular adenosine triphosphate released on neuronal activation, convert it with CD39 to adenosine monophosphate and then with CD73 to adenosine, and adenosine acts on the A1 receptor. Ablating microglia amplifies and synchronises neuronal activity and causes seizures. (Badimon et al. 2020, Nature, doi:10.1038/s41586-020-2777-8)
  So: This answers the general form of the authors' question three years after they asked it and names the specific factor, so the remaining gap is narrow and specific: whether the same loop operates on ischemic neurons. The design is therefore the ischemia test of a known mechanism, not a search.
• Somatic microglia-neuron junctions have a nanoarchitecture optimised for purinergic signalling, and brain injury induced changes at these junctions trigger P2Y12 receptor dependent microglial neuroprotection that regulates neuronal calcium load and functional connectivity. (Cserep et al. 2020, Science, doi:10.1126/science.aax6752)
  So: Protection may be delivered at a direct contact site rather than by a soluble factor, which is a real alternative to the conditioned-medium premise the authors' in vitro design rests on and the main reason stage two exists.
• Selective elimination of microglia increased infarct size by 60 percent, reversed by microglial repopulation, and the absence of microglia led to dysregulated neuronal calcium responses, calcium overload, increased neuronal death and a markedly reduced incidence of spreading depolarisation. (Szalay et al. 2016, Nature Communications, doi:10.1038/ncomms11499)
  So: This is reference 39 of the paper under design and it had already measured neuronal activity in vivo before Jin published, so the untested part is only the identity of the factor, and the in vivo activity phenotype should not be reclaimed as new.
• Microglia modulate cerebrovascular reactivity through the ectonucleotidase CD39. (Fu et al. 2025, Nature Communications, doi:10.1038/s41467-025-56093-5)
  So: CD39 on microglia also acts on blood vessels, so an A1 receptor rescue in vivo could work through perfusion rather than through neuronal excitability, and cerebral blood flow must be monitored in stage two.
• Surveilling microglia dampen neuronal activity through a purinergically mediated negative feedback loop. (Mo et al. 2021, Signal Transduction and Targeted Therapy, doi:10.1038/s41392-021-00586-4)
  So: Independent commentary confirming the loop is now accepted, which strengthens the case that a null in this design would be informative about ischemia specifically rather than about the mechanism's existence.

Searched: Searched Europe PMC and OpenAlex for microglial control of neuronal activity through adenosine, adenosine triphosphate, CD39 and the A1 receptor, for somatic purinergic junctions, and for microglia-derived factors acting on ischemic or excitotoxic neurons. The general mechanism is settled by Badimon 2020 and the in vivo stroke activity phenotype by Szalay 2016. Searched specifically for an adenosine A1 receptor rescue of CSF1R inhibitor induced infarct growth and found nothing, so stage two appears untried.

Feasibility

A laboratory with primary neuron and glia culture, a multi-electrode array or calcium imaging rig, and access to the filament occlusion model; the culture protocols in the paper are sufficient for stage one on their own. · 4 to 8 months for stage one, a further 6 months for stage two · moderate for stage one, which is mostly culture work on an existing protocol, and high for stage two with its animal numbers and imaging
Blockers: No sample size or variance published for Figure 6, so stage one starts with a pilot; Adenosine is rapidly degraded, so medium handling and timing have to be worked out before the enzyme arms mean anything; Doses for the A1 antagonist, A1 agonist and CD39 inhibitor are not derivable from this record

Strongest objection

A specialist would say that conditioned medium is the wrong preparation for a mechanism that Cserep and colleagues showed operates at a direct microglia-neuron contact site, and that adenosine's half life in medium is short enough that anything protective in a 24 hour transfer is unlikely to be adenosine. They would add that an A1 receptor agonist given systemically lowers heart rate and blood pressure, so an in vivo rescue could be haemodynamic rather than neuronal.

Not derivable from what is published

• Sample size and variance for any Figure 6 measurement
• Concentrations and schedules for the A1 receptor antagonist, the A1 agonist, the CD39 inhibitor and adenosine deaminase
• Whether the paper's conditioned medium was collected with or without a medium change before the 24 hour window, which determines adenosine concentration
• Whether the microglia-conditioned medium benefit the authors infer is on lactate dehydrogenase release only or also on nick end labelling, since the record reports no significant effect of microglia medium alone on neuronal death

This is the authors' own proposal and the honest limitation of the whole paper: PLX3397 inhibits both colony stimulating factor 1 receptor and c-Kit, so they cannot say the worse strokes came from losing microglia rather than from the drug doing something else, and they suggest a more selective inhibitor as the fix.

Question

Does a genetic model that lacks microglia without any drug reproduce the infarct growth and neurodeficit seen with PLX3397 after sixty minutes of middle cerebral artery occlusion?

Hypothesis

Csf1r delta FIRE homozygous mice, which lack brain microglia but keep circulating monocytes, will show the same aggravation of infarct volume after occlusion as PLX3397-fed wild type mice run in the same cohort.

Attempt type

wet-lab

Design

One cohort, four arms, run and read together. Arm one is Csf1r delta FIRE homozygous mice; arm two is their Csf1r wild type littermates; arm three is wild type C57BL/6 on PLX3397 at 40 mg per kg per day in AIN-76A chow for 21 days; arm four is the same wild type mice on AIN-76A chow alone. Every arm gets sham or 60 minutes of occlusion under the published laser Doppler criteria. Endpoints are deliberately identical to the original paper so the comparison is like for like: modified Neurological Severity Score, corner turning, forelimb placing, 7 tesla T2 infarct volume at day 1 and day 3, CD11b positive CD45 intermediate microglia counts, the luminol reactive oxygen species signal at 200 mg per kg, and the astrocyte intracellular cytokine panel. Primary endpoint is day 1 infarct volume. The authors' proposal is amended on purpose: the design does not swap PLX3397 for PLX5622, because PLX5622 has since been shown to be no more microglia-specific; it swaps the drug for genetics and keeps the drug arm as the bridging comparator. Analysis is pre-registered as an equivalence test between arms one and three, not a difference test.

Controls

• designated The PLX3397 arm run in the same cohort, which must reproduce the published aggravation of infarct volume, together with the roughly 90 percent microglia reduction check in both depleted arms.
  Rules out: That the whole cohort failed. This is the pivot of the design: if the drug arm does not reproduce the original result, nothing can be concluded about the genetic arm, and the record must say inconclusive-no-positive-control.
• background Csf1r wild type littermates from the same breeding, bred and housed alongside the homozygotes.
  Rules out: That the mixed or drifting genetic background of the mutant line, rather than absence of microglia, changes infarct size. Note that the original study back-crossed its Rag2 and gamma c double mutants for 12 generations, so the same standard applies here.
  Vocabulary note: A littermate wild type genotype is a genetic background control, which the vocabulary has no dedicated kind for; background is the closest and is being used here for strain background rather than for assay background signal.
• vehicle AIN-76A standard chow alone for the wild type drug comparison.
  Rules out: That the chow base or the 21 day feeding regime rather than CSF1R inhibition drives the effect.
• no-treatment Sham surgery in each of the four arms.
  Rules out: That lifelong absence of microglia produces a baseline neurological deficit that inflates the post-stroke score. The original paper did exactly this check for PLX3397 and found no sham deficit; the genetic model needs its own because its microglia are absent from development rather than removed in adulthood.

Sensitivity

This design turns on an equivalence test, so the margin matters more than the power. With a two one-sided test at alpha 0.05 and power 0.8, the sample size per group is about 12.4 divided by the square of the margin in standard deviations. The paper's 15 animals per group can therefore support an equivalence margin of about 1.0 standard deviations but not 0.5, which would need about 50 per group. The one published effect size that bears on the margin is the 60 percent increase in infarct size reported by Szalay and colleagues for microglia elimination in a comparable ischemia model; a margin of half that increase is the defensible choice and the standard deviation needed to express it in those units is not published. So the honest statement is: the margin can be set in standard deviation units from the paper's own power assumptions, but not in cubic millimetres or percent of hemisphere, and a pilot of 8 control-chow occlusion animals is needed to bridge the two.

Would count as

Supported → positive · opposed → negative-not-replicated
Inconclusive if: The PLX3397 arm in the same cohort fails to reproduce the published aggravation, or microglia counts in the genetic arm do not reach the roughly 90 percent reduction the drug achieves so the two arms are not comparable in depletion depth, or the sham genetic arm shows a baseline neurological deficit that confounds the post-stroke score. A null in the genetic arm alongside a working drug arm is the interesting case and would additionally license the stronger class refuted for the reading that the drug effect is CSF1R-specific and microglia-mediated. A null in both arms is negative-not-reproducible and says nothing about selectivity. What no outcome here licenses is the claim that microglia are irrelevant to stroke outcome, because the direction of the depletion effect is already known to flip with model and with timing.

Already done

• CSF1R inhibition by PLX5622 is not restricted to microglia: it affects myeloid and lymphoid compartments, suppresses interleukin 1 beta, CD68 and phagocytosis in bone marrow derived macrophages long after treatment stops, and reduces resident and interstitial macrophages of peritoneum, lung and liver but not spleen. (Lei et al. 2020, Proceedings of the National Academy of Sciences, doi:10.1073/pnas.1922788117)
  So: This directly undercuts the authors' proposed fix. Swapping PLX3397 for the more selective PLX5622 would not settle the off-target question, which is why this design goes to genetics instead.
• Deletion of the fms-intronic regulatory element super-enhancer in the Csf1r locus ablates brain microglia and resident macrophages of skin, kidney, heart and peritoneum, while monocyte homeostasis is unaffected, and Csf1r delta FIRE homozygous mice are healthy and fertile without the abnormalities seen in Csf1r null rodents. (Rojo et al. 2019, Nature Communications, doi:10.1038/s41467-019-11053-8)
  So: This is the tool the design needs and the reason it can be run at all: microglia absent, monocytes intact, no drug. It also sets the caveat, since heart and peritoneal macrophages are lost too.
• PLX3397 chow at 660 mg per kg for seven days strongly depleted brain microglia with minimal effect on peripheral immune cells in spleen, bone marrow, kidney and lung, but reduced CD3 positive cells, monocytes and neutrophils in the heart, and under lipopolysaccharide challenge it lowered tumour necrosis factor alpha, interleukin 1 alpha, interferon gamma and interleukin 1 beta. (Okojie et al. 2023, Journal of Neuroinflammation, doi:10.1186/s12974-023-02924-5)
  So: Off-target burden depends on dose and duration, and this exposure is not the one used here, which was 40 mg per kg per day for at least 22 days. The cardiac finding matters for a stroke model, and the blunted cytokine response under challenge is a direct alternative explanation for the altered inflammation in the original paper.
• Specific genetic depletion of resident microglia using CX3CR1-CreER with an inducible diphtheria toxin receptor reduced infarct volume and improved motor ability three days after Rose Bengal photothrombotic stroke, and raised anti-inflammatory transforming growth factor beta 1, Arg1, interleukin 10, interleukin 4 and Ym1 messenger RNA. (Li et al. 2021, Journal of Neuroinflammation, doi:10.1186/s12974-021-02127-w)
  So: A genetic depletion model gave the opposite result in a different stroke model, so a null in the delta FIRE arm would be ambiguous between off-target drug action and model dependence. This is the strongest reason to keep the drug arm in the same cohort.
• Long-term CSF1R inhibitor treatment to deplete microglia increased brain neutrophil numbers and enlarged the ischemic lesion in permanent middle cerebral artery occlusion. (Otxoa-de-Amezaga et al. 2019, Acta Neuropathologica, doi:10.1007/s00401-018-1954-4)
  So: An independent replication of the direction of the Jin result with a different occlusion model, which raises the prior that the effect is real rather than off-target and makes an equivalence design the right framing.
• PLX3397, PLX5622 and GW2580 differ in their depletion profiles, and the effects of these strategies on neurons, astrocytes and oligodendrocytes remain poorly characterised. (Guenoun et al. 2025, Glia, doi:10.1002/glia.24664)
  So: A current review confirms that no pharmacological inhibitor is clean, which is the reason the authors' original proposal cannot be followed as written.

Searched: Searched Europe PMC and OpenAlex for the selectivity of PLX3397, PLX5622 and GW2580, for genetic microglia depletion models including CX3CR1-CreER with diphtheria toxin receptor and the Csf1r delta FIRE line, and specifically for any stroke or ischemia experiment performed in Csf1r delta FIRE mice. The selectivity question is answered and answered against the authors' proposed fix; the stroke experiment in delta FIRE mice appears not to have been done, so this design is still untried as of October 2026. Also searched for PLX5622 combined with middle cerebral artery occlusion and found two 2025 studies on the early depletion and repopulation phase rather than on acute infarct size.

Feasibility

A rodent stroke laboratory with access to the Csf1r delta FIRE line, which means either a collaboration with the Edinburgh group that made it or an import and a rederivation; the surgical and imaging capability is standard. · 12 to 18 months, dominated by importing, rederiving and breeding the line to produce homozygotes and littermate controls in cohorts large enough for an equivalence test · high, because an equivalence margin tighter than 1.0 standard deviations needs roughly 50 animals per group and the breeding has to deliver littermates in matched batches
Blockers: Mouse line access and rederivation time; Csf1r delta FIRE mice lack microglia from development, so compensation cannot be ruled out by this design alone; The line also lacks heart, kidney, skin and peritoneal macrophages, which is a new confound rather than the removal of an old one; Equivalence margin cannot be expressed in absolute infarct units without a pilot

Strongest objection

A specialist would say that comparing a lifelong developmental absence of microglia with a 21 day drug depletion in an adult is not a clean test of off-target pharmacology, because any difference between the arms is equally explainable by developmental compensation, and any similarity could be a coincidence of two different biologies converging on infarct size. They would add that Csf1r delta FIRE mice also lose cardiac and peritoneal macrophages, which in a model whose outcome depends on systemic inflammation and on cardiac function during anaesthesia is a substitution of one off-target problem for another. The cleanest answer they would push for is an adult inducible microglia-specific deletion, not a constitutive enhancer deletion.

Not derivable from what is published

• Infarct volume standard deviation in this model, needed to express the equivalence margin in absolute units
• Whether Csf1r delta FIRE homozygotes reach a depletion depth comparable to the roughly 90 percent achieved by PLX3397 in this strain and assay
• Baseline neurological scores for Csf1r delta FIRE mice on the modified Neurological Severity Score, corner turning and forelimb placing tests
• Whether the line tolerates 60 minutes of occlusion with the same surgical mortality as C57BL/6, since the original study lost 14 of 294 animals to post-surgical death
• Cerebral blood flow and collateral anatomy in the mutant line, which the laser Doppler inclusion criteria assume are normal