MNEMEChoose a mission ↗

RESEARCH NOTE R-003 · 11 OCTOBER 2026 · VERSION 1.0

One target.
Different outcomes.

When does TREM2 help, show no demonstrated benefit, or cause harm? Ten primary publications, compared by intervention, pathology, timing and endpoint.

AI-assisted targeted literature comparison; not peer reviewed · A follow-up to R-002
Literature analysis completed. Proposed experiments have not been performed.

Explore the comparison ↓Download reportEvidence CSVSource data

RESEARCH DECISION

Find the conditions for protection.

Prioritize a state- and pathology-matched test of neuronal protection. Broad TREM2 activation has no established clinical benefit in the reviewed trial; no effective human subgroup or optimal dose is identified here.

This is a research priority, not a treatment recommendation or a claim that a cure has been found.

AT A GLANCE / READ WITH THE FULL EVIDENCE

A lead to test.
A reason to question it.

01 / WHY INVESTIGATE

The supporting signal

The comparison separates beneficial, null, harmful and mixed findings by model, intervention and endpoint.

Inspect the comparisons ↓
02 / WHAT LIMITS THE CLAIM

The counterevidence

Tau-seeding experiments and the reviewed clinical result challenge broad activation-as-benefit claims. No human treatment window is established.

Check the boundary ↓
03 / WHAT WOULD HELP NEXT

The next check

Compare cell state, pathology and agonist mechanism under matched conditions, with separate safety checks.

Read the proposed plan ↓

Summary of this dated literature report. Findings come from the cited studies; MNEME has not performed the proposed experiments. These categories are not evidence-quality scores.

What changes the interpretation?

Model & pathology

Amyloid, seeded tau, vascular injury and myelin repair ask different biological questions.

Intervention & cell state

Knockout, metabolic rescue, overexpression and distinct agonist antibodies are not interchangeable.

Outcome

Cellular energy, plaque burden, neuronal injury and human cognition must remain separate.

Amyloid alone versus added tau

Observed: The same murine agonist appears in a beneficial amyloid experiment and a harmful tau-seeding experiment. S9 · S6

Our interpretation: Tau context is a priority moderator to test, but age, dose and other design differences prevent a causal explanation from this pair alone.

Timing is not a universal early-good / late-bad rule

Observed: The harmful seeded experiment began before tau injection; a different antibody had favorable pathology findings when started in older amyloid mice. S6 · S10

Our interpretation: Stage must be defined by pathology and cell state, not age alone. The papers do not locate a human treatment window.

Activation mode deserves a controlled comparison

Observed: 03O05 increased sTREM2. The human AL002 trial lowered CSF sTREM2; acute AL002a experiments also reported increases. S10 · S6 · S8

Our interpretation: Do not collapse antibodies or soluble and membrane TREM2 into a simple good/bad rule. Shedding, signaling duration and brain exposure need matched measurements; higher sTREM2 is not itself proof of protection.

A cellular response is not a clinical success

Observed: Functional rescue and expression-dependent metabolic responses coexist with a negative human efficacy trial. S1 · S2 · S3 · S8

Our interpretation: Select interventions by neuronal outcomes and safety, alongside target engagement, rather than treating a biomarker change as a therapeutic result.

11 CONTEXTS · 10 PUBLICATIONS · NO VOTE COUNTING

Inspect benefit, null results and harm.

Filter by the reported outcome. “No demonstrated benefit” does not mean the true effect is exactly zero. The mixed category is not proof of harm. Two rows from one paper remain one source.

Showing 11 of 11 contexts.

C1 · Mouse / mechanism

Benefit in tested context

TREM2-deficient 5XFAD; impaired cellular energy handling

Intervention
Cyclocreatine metabolic support; not direct receptor agonism
Timing
Rescue experiment in a deficient amyloid model; regimen not fully extracted
Comparator
Deficient mice without metabolic rescue
Reported result
Improved microglial clustering and reduced plaque-adjacent neuritic dystrophy.
Interpretation limit
Rescuing a deficiency does not establish a benefit from increasing normal TREM2. No human cognitive outcome.
Replication / sample
Endpoint-specific biological replicate counts not extracted

S1 · Metabolic rescue results

C2 · Human-derived cells

Benefit in tested context

iPSC-derived microglia carrying the R47H risk variant

Intervention
Manipulation of PPARγ / p38MAPK-related signaling
Timing
Cell-culture intervention; no clinical disease stage
Comparator
Variant and comparator cell conditions
Reported result
Metabolic switching and phagocytic function could be rescued.
Interpretation limit
A cellular functional result, not neuronal protection or patient benefit. Donor replication needs full-text verification.
Replication / sample
Donor and experiment counts not extracted

S2 · Indexed abstract

C3 · Mouse / behavior

Benefit in tested context

Male 5XFAD amyloid model without added human tau seeds

Intervention
AL002a, 50 mg/kg intraperitoneally each week; MOPC control
Timing
Start at 4 months; 14 weeks of treatment; behavior during the final 2 weeks
Comparator
Isotype-matched MOPC; additional wild-type groups
Reported result
Less amyloid deposition and improved spatial learning and novel-object recognition.
Interpretation limit
One aggressive amyloid model; comparison with C6 also changes age, dose, sex composition and endpoints.
Replication / sample
5XFAD: 12 per antibody group; wild type: 8 AL002a, 7 control

S9 · Systemic administration; behavioral and chronic-treatment results

C4 · Mouse / pathology

Benefit in tested context

5xFAD amyloid model; ligand-mimetic antibody 03O05

Intervention
30 mg/kg intraperitoneally weekly for 12 weeks
Timing
Start at 7 months; tissue 2 days after the final dose
Comparator
Vehicle-treated 5xFAD; age-matched wild type
Reported result
Higher soluble TREM2, fewer filamentous plaques and reduced neuronal dystrophy.
Interpretation limit
The disease experiment used vehicle rather than an isotype antibody control. No matched comparison proves superiority over other agonists or benefit in people.
Replication / sample
14–15 mice per group in the chronic study

S10 · In vivo efficacy methods; Figures 3–6

C5 · Mouse / cell state

Mixed / context-dependent

APP SAA amyloid reporter mice; low, mid and high TREM2-expressing microglia

Intervention
Brain-penetrant ATV:4D9 versus isotype; 1 mg/kg monthly
Timing
Four doses at ages 8–11 months; collection at 12 months
Comparator
Isotype within expression strata
Reported result
Mid-expression cells showed increased metabolic indicators; low cells changed little; high cells showed attenuated or opposite responses.
Interpretation limit
These are metabolic profiles, not established benefit or injury. A high-expression response must not be labeled neuronal harm. Delayed collection limits acute-response interpretation.
Replication / sample
Assay-specific biological replicate counts not fully extracted

S3 · Figure 7 and chronic antibody-treatment methods

C6 · Mouse / amyloid + tau

Harm in tested context

5XFAD with human Alzheimer-derived tau injected at 6 months

Intervention
AL002a 80 mg/kg intraperitoneally weekly versus IgG
Timing
Start 1 week BEFORE tau injection; continue to 9 months
Comparator
Control IgG in the same seeded model
Reported result
More tau seeding/spreading and neuritic dystrophy; total amyloid burden did not fall.
Interpretation limit
Harm was not restricted to treatment after established tau pathology. Artificial seeding and other design differences prevent attribution to tau alone.
Replication / sample
Chronic groups: 27 antibody (13 female/14 male), 26 control (14 female/12 male); endpoint subsets vary

S6 · Figure 1; chronic treatment and tau pathology results

C7 · Mouse / AD models

No demonstrated benefit

PS2APP and TauPS2APP; multiple agonist antibodies

Intervention
Para.09, mAb5 and mAb7 across separate designs
Timing
Early, middle and later treatment designs; 12- or 16-week regimens
Comparator
Matched experimental controls in each design
Reported result
Limited or no improvement in the tested Alzheimer-related pathological and neuronal endpoints.
Interpretation limit
A null finding is not proof of equivalence. CNS exposure, antibody format and timing remain possible explanations, not resolved causes.
Replication / sample
No single pooled n; full endpoint-level extraction pending

S7 · AD-model results and discussion

C8 · Mouse / demyelination

Harm in tested context

Toxin-induced demyelination and repair models

Intervention
TREM2 agonist antibodies tested during injury/recovery
Timing
Repair-stage experiments; not a dementia clinical stage
Comparator
Experimental control treatment
Reported result
Recovery from demyelination was impaired in the tested settings.
Interpretation limit
A safety-context finding, not direct evidence of harm in human dementia. C7 and C8 are one publication, not independent replications.
Replication / sample
Endpoint-specific n and regimens not fully extracted

S7 · Demyelination results

C9 · Mouse / ischemic injury

Benefit in tested context

Young adult male mice after transient bilateral carotid occlusion

Intervention
Lentiviral TREM2 overexpression
Timing
After 24-hour recovery; spatial learning assessed about 4 weeks later
Comparator
Injury/control-vector and sham conditions
Reported result
Better spatial-learning performance and hippocampal neuronal preservation.
Interpretation limit
Acute ischemia differs from aged vascular amyloid disease. Gene overexpression is not an antibody treatment.
Replication / sample
Three groups of 10; individual endpoint counts require verification

S4 · Animal groups, treatment and behavioral results

C10 · Mouse / vascular amyloid

Mixed / context-dependent

CAA-prone Tg-SwDI mice; Trem2 wild type, heterozygous and knockout

Intervention
Lifelong genetic loss, not a timed treatment
Timing
Assessed at 16 months
Comparator
Different Trem2 genotypes in the same model
Reported result
Knockout reduced vascular amyloid but increased overall brain amyloid.
Interpretation limit
Opposing compartment outcomes are a trade-off. Knockout cannot be read backwards as proof that an agonist causes human vascular harm.
Replication / sample
Figure 1 total amyloid: WT 7, heterozygous 7, knockout 6; not all endpoints

S5 · Indexed abstract and Figure 1 descriptions

C11 · Human / randomized trial

No demonstrated benefit

381 participants with early Alzheimer’s disease

Intervention
AL002 15, 40 or 60 mg/kg IV every 4 weeks versus placebo
Timing
48–96 weeks; primary estimate at week 96
Comparator
Randomized, double-blind placebo control
Reported result
Target engagement without a significant primary CDR-SB benefit. MRI changes resembling ARIA were the most frequent reported treatment-emergent adverse events.
Interpretation limit
No established clinical efficacy for this regimen. Confidence intervals allow benefit and harm; adverse-event rates were not extracted from the subscription full text.
Replication / sample
381 randomized across four groups

S8 · Public abstract; NCT04592874

HUMAN EVIDENCE · S8

The clinical result sets a boundary.

Reported week-96 CDR-SB treatment–placebo differences. Negative estimates favor AL002; every interval crosses zero. These are published estimates, not a new analysis. S8

Compare columns · scroll sideways on smaller screens.

AL002 phase 2: primary endpoint not met
DoseDifference95% confidence interval
15 mg/kg-0.31-1.61 to 0.98
40 mg/kg0.13-1.18 to 1.43
60 mg/kg-0.17-1.49 to 1.15

Biological activity is insufficient to claim clinical efficacy. The MRI safety signal also needs a full quantitative assessment before translational decisions.

CONTINUE THE SOURCE SEARCH

Scout can retrieve papers in the relevant broad track. A source scan does not carry out the validation plan below.

Open Scout’s TREM2 track ↗Bring a source or challenge ↗

PRIORITIZED QUESTIONS · PROPOSED, NOT EXECUTED

Where we go next.

01 · State-specific protection with a tau safety check

Best next mechanistic question: when does functional microglial rescue coincide with less neuronal injury?

Test: Compare matched amyloid-only and amyloid-plus-tau conditions, stratify baseline TREM2 expression and functional impairment, and measure actual CNS exposure. Use neuronal injury as the prespecified primary endpoint; metabolism, tau spread and vascular injury are explanatory or safety endpoints.

Decision gate: Advance only if neuronal protection replicates independently without a prespecified unacceptable increase in tau or vascular injury. A metabolic improvement alone does not pass.

Priority hypothesis and proposed design; no experiment executed

02 · Compare agonist mechanisms under matched conditions

Binding epitope, shedding and signal duration could help explain conflicting outcomes; current comparisons are confounded.

Test: Compare ligand-mimetic and other agonist approaches with matched exposure, appropriate isotype controls, Fc characterization, membrane/soluble TREM2 measures and the same injury endpoints.

Decision gate: Deprioritize the proposed mechanism if its predicted advantage disappears with matched controls or if soluble TREM2 shifts without neuronal protection.

Exploratory mechanism; no validated superiority claim

03 · Keep vascular and myelin outcomes as decision gates

A favorable result in one tissue compartment may hide a worse result in another.

Test: Extract vessel-wall versus parenchymal amyloid, barrier injury and repair outcomes separately. First complete endpoint-level sample sizes, blinding, exclusions and uncertainty estimates.

Decision gate: Do not interpret lower total amyloid or a repair biomarker as sufficient evidence of overall benefit.

Required evidence work before selecting a translational candidate

Deprioritize: a generic “more TREM2 is better” claim, a single biomarker as proof of success, and selection of a human dose or subgroup from these heterogeneous experiments.

Trace every result.

Method & unfinished work.

scope: Ten purposively selected primary publications; eleven context rows. Eight publications carried forward from R-002 plus two direct agonist comparators. This is a targeted narrative comparison, not a systematic review or meta-analysis.

search Date: 2026-10-11

retrieval: Follow-up of R-002 references and primary publisher/PubMed pages. Targeted searches concerned AL002a chronic activation, neutral/detrimental agonists, TREM2 expression-dependent metabolism and ligand-mimetic agonism. No exhaustive retrieval or screening denominator was recorded.

classification: Benefit, no demonstrated benefit and harm refer only to the stated endpoints and experimental context. Mixed includes trade-offs and cellular responses with uncertain therapeutic meaning. Categories are not certainty grades. Papers and rows are not votes.

limitations: Incomplete endpoint-level extraction, no formal risk-of-bias assessment, no pooled effects, no independent expert review and no reanalysis of underlying datasets. Missing sample sizes and regimens are explicitly marked. The trial subscription full text was not reviewed; several inherited records remain abstract-level. Funding and conflicts were not systematically compared. Access depth is listed per source. No exhaustive correction/retraction check is claimed.

maintenance: Dated editorial report, separate from automatic publication scans. Future evidence changes require a versioned review. The website did not run new biological experiments or discover a validated therapeutic connection.

next Deliverable: A matched-comparison extraction sheet with all biological replicate counts, sex, dose/exposure, controls, blinding, exclusions and outcome uncertainties; followed by an independently reviewed protocol. No immediate human treatment experiment is proposed.

Version 1.0 adds two primary agonist studies to R-002 and separates eleven experimental contexts. R-002 remains available as the earlier report.