English:Aging – Human causal inference for neural protein targets
aiMOOCAging – Human causal inference for neural protein targets
QR-Code
Introduction
This expert colloquium asks you to answer a deliberately difficult translational question: does changing a neural protein cause a change in an aging-related neurodegenerative outcome, and does the human genetic evidence identify the same intervention that a drug would make? The worked target is progranulin (PGRN), the secreted protein encoded by GRN. The course uses Alzheimer’s disease (AD) as the main common-disease GWAS outcome and frontotemporal dementia caused by GRN loss of function (FTD-GRN) as a high-penetrance mechanistic and intervention anchor.
The evidence snapshot is current to 29 September 2026. You should treat every numerical result as a property of a specific assay, tissue, ancestry, cohort, and estimand rather than as a context-free fact. The goal is not to “prove” a target from one Mendelian-randomization estimate. Your job is to triangulate public pQTL, eQTL, GWAS, rare-variant, biomarker, and intervention evidence and to stop when causal direction is not identifiable.
Learning goals
By the end of the colloquium, you should be able to distinguish a circulating biomarker from a CNS target exposure; audit cis and trans pQTLs; connect eQTL and pQTL evidence to disease GWAS through fine-mapping and colocalization; state the assumptions behind Mendelian randomization; design pleiotropy and reverse-causality checks; specify a target trial and its observational emulation; interpret a biomarker-positive but clinically negative trial; and write a stopping rule that prevents an attractive target story from outrunning identification.
Why progranulin is an instructive target
PGRN is a lysosomal and neuroimmune-relevant protein expressed in multiple tissues and cell types. Heterozygous loss-of-function variants in GRN cause progranulin haploinsufficiency and are an established cause of FTD. Common regulatory variation at the GRN locus also changes PGRN abundance and has been associated with AD risk. This creates an unusually rich causal chain from gene to RNA to protein to disease. It also creates traps: plasma and CSF PGRN are not interchangeable; trans-pQTLs such as SORT1 can alter PGRN through biologically real pathways that may also affect other traits; and a therapy that raises PGRN can still fail to improve a clinical endpoint.
A strong target dossier must therefore answer four different questions separately: target biology, causal direction, compartment-specific exposure, and intervention transportability. A positive answer to one does not automatically answer the others.
Causal Question and Estimands
The target question
The primary genetic question is: among people represented by the source GWAS populations, what is the effect of a genetically induced increase in CNS-relevant PGRN abundance on the risk of Alzheimer’s disease? The translational question is different: what would be the effect of initiating a defined PGRN-elevating intervention at a defined age and disease stage on a defined clinical outcome over a defined follow-up period?
MR usually estimates a contrast generated by lifelong genetic perturbation. A trial estimates a contrast generated by a finite intervention delivered at a particular dose, age, disease stage, route, and duration. You must not silently treat these estimands as identical.
Working causal diagram
Use the following verbal directed acyclic graph as your starting model:
GRN cis genotype → GRN expression in relevant brain cells → CNS PGRN → lysosomal and neuroimmune biology → neurodegenerative outcome.
Add competing paths explicitly:
GRN cis genotype → another local transcript or molecular phenotype → outcome is a horizontal-pleiotropy path.
SORT1 genotype → sortilin biology → PGRN is a useful trans-regulatory path, but SORT1 genotype → lipid or other biology → outcome would violate the exclusion restriction for an MR instrument if that path does not operate through PGRN.
Preclinical neurodegeneration → inflammation or cell loss → measured PGRN is a reverse-causality path.
Assay-binding variant → measured PGRN without changing biologically active protein is a measurement path.
Target Dossier: Progranulin
Molecular identity and intervention handle
Target protein: progranulin, encoded by GRN on chromosome 17.
Direct genetic perturbation: GRN loss-of-function establishes that large reductions in functional progranulin can cause FTD through haploinsufficiency.
Pharmacologic handle: sortilin, encoded by SORT1, binds progranulin and contributes to its cellular uptake and degradation. Latozinemab (AL001) is an anti-sortilin monoclonal antibody designed to reduce cell-surface sortilin and elevate PGRN.
Important distinction: GRN variation, SORT1 inhibition, gene replacement, and exogenous PGRN delivery are different interventions. They can converge on PGRN abundance while producing different spatial, temporal, cellular, and off-target effects.
Public pQTL audit
A useful plasma source is the UK Biobank Pharma Proteomics Project, which assayed 2,923 proteins with Olink in about 54,000 participants and released pQTL summary statistics. The resource is accessible through UKB-PPP. In a 2026 GRN analysis using these data, rs5848 at the GRN locus showed a very strong cis association with plasma PGRN; the T allele was associated with lower measured PGRN. The study reported an average decrease of about 4.9% in plasma PGRN per T allele in its analysis.
For CNS-proximal protein evidence, a large CSF proteogenomic resource measured proteins in 4,968 individuals across eight neurodegenerative cohorts using SomaScan platforms, with data available through NIAGADS dataset NG00130 and GWAS Catalog accessions GCST90421033–GCST90428040. A separate ADNI analysis used a PGRN-specific CSF assay in 818 participants. In the 2026 GRN analysis, rs5848 T was associated with lower CSF PGRN, with an estimated average decrease of about 10.8% per T allele in ADNI.
Audit questions you must answer before MR: Is the pQTL cis to GRN or trans? Is the assay affinity based, and could a protein-altering variant change binding rather than abundance? Was the association conditional on multiple signals? Which ancestry and genome build were used? Is the effect expressed in NPX, rank-normalized units, SD units, or log2 protein? Does the same allele have the same directional effect in an orthogonal assay?
Plasma is not brain exposure
An older paired-compartment study found only a weak correlation between plasma and CSF PGRN after accounting for age and sex, and genetic determinants differed between compartments. In that study, rs5848 was strongly associated with CSF PGRN whereas a SORT1-region signal had a prominent plasma association. This is exactly why a plasma pQTL cannot be assumed to instrument brain exposure.
For this dossier, use an explicit evidence hierarchy. CSF PGRN is CNS-proximal but is still not identical to parenchymal or cell-type-specific brain exposure. Brain eQTL and single-cell or single-nucleus QTL evidence can improve localization, but RNA abundance is not protein abundance. Plasma PGRN is useful as a systemic pharmacodynamic marker, but it should not be used as a surrogate for CNS target engagement unless the bridging evidence is demonstrated.
Public eQTL audit
Use GTEx and MetaBrain to audit GRN cis-eQTLs across brain regions. The 2026 integrative GRN study reported that the AD-risk rs5848 T allele was associated with lower GRN mRNA across brain datasets. Recent single-nucleus work also illustrates that eQTL effects can be cell-type specific, so bulk brain results should be treated as averages over changing cellular composition.
For each brain eQTL result, record tissue or cell type, sample size, ancestry, RNA processing pipeline, covariates, fine-mapped credible set, effect allele, effect direction, and whether the signal is shared with the pQTL and GWAS signal. Do not equate “same lead SNP” with “same causal variant.”
Disease GWAS audit
Use large AD GWAS summary statistics, including the Bellenguez et al. 2022 meta-analysis and an independent resource such as FinnGen. The 2026 GRN analysis reported that rs5848 T was associated with increased AD risk in both a large AD GWAS and FinnGen release 9, while the same allele decreased PGRN. The direction is therefore compatible with the hypothesis that higher PGRN is protective for AD.
However, ancestry portability is not guaranteed. A 2026 ancestry-focused study reported population-specific patterns around rs5848. Your dossier must therefore identify the ancestry of the QTL and GWAS samples, use ancestry-matched LD when fine-mapping, and avoid presenting a European-ancestry estimate as universal.
Rare-variant evidence
Rare GRN loss-of-function variants provide a qualitatively different perturbation from common cis regulation. In FTD they establish a causal role for severe progranulin deficiency. In AD/dementia analyses, aggregated rare-variant results can add directionally informative evidence but do not automatically calibrate the effect of a modest protein increase. A rare loss-of-function syndrome and a common-variant MR estimate occupy different parts of the dose-response curve and may act at different developmental and aging stages.
Intervention evidence
Phase 1 latozinemab studies showed proof of mechanism. A first-in-human phase 1 report found that a single infusion in healthy volunteers approximately tripled plasma PGRN and doubled CSF PGRN, and a multiple-dose phase 1 study in symptomatic FTD-GRN participants raised plasma and CSF PGRN toward levels observed in healthy volunteers. These results establish pharmacodynamic target engagement, not clinical benefit.
The decisive translational update is the phase 3 INFRONT-3 trial, ClinicalTrials.gov NCT04374136. In October 2025, the sponsor reported that the 96-week randomized, double-blind trial did not meet its clinical co-primary endpoint of slowing FTD-GRN progression on the CDR plus NACC FTLD Sum of Boxes. Plasma PGRN changed as intended, but the reported secondary and exploratory fluid-biomarker and volumetric-MRI outcomes did not show treatment-related effects. The trial record was later marked terminated. See the ClinicalTrials.gov record and the October 2025 sponsor filing.
This is not a clean falsification of the AD MR hypothesis because the trial tested a different disease, population, intervention, age distribution, disease stage, duration, and outcome. It is, however, strong evidence against the claim that raising PGRN with latozinemab under the INFRONT-3 protocol produces measurable clinical benefit in FTD-GRN over 96 weeks. Any dossier written after 2025 must include this negative intervention evidence.
Other PGRN-restoring strategies, including GRN gene therapy such as AVB-101, are mechanistically distinct and should not inherit either the MR estimate or the latozinemab trial result without a transportability argument.
Triangulated evidence table
| Evidence layer | Public source or study | What it can support | Main threat |
|---|---|---|---|
| Plasma pQTL | UKB-PPP Olink pQTL | Genetic control of circulating PGRN | Peripheral biology, assay effects, ancestry |
| CSF pQTL | NIAGADS neurodegeneration cohorts and ADNI | Genetic control of CNS-proximal PGRN | CSF is not cell-specific brain protein |
| Brain eQTL | GTEx and MetaBrain | Direction of GRN transcriptional regulation | Tissue averaging and RNA-protein discordance |
| AD GWAS | Bellenguez meta-analysis and FinnGen | Disease association at the GRN locus | LD, ancestry, phenotype heterogeneity |
| Rare variants | GRN loss-of-function analyses | Large perturbation and biological necessity in FTD | Nonlinear dose response and phenotype differences |
| Phase 1 intervention | Latozinemab studies | Pharmacodynamic increase in plasma and CSF PGRN | Small samples and no efficacy conclusion |
| Phase 3 intervention | INFRONT-3 NCT04374136 | Direct test of one PGRN-elevating strategy in FTD-GRN | Different estimand from AD genetic analyses |
Colocalization Plan
Why colocalization comes before causal storytelling
Two traits can have significant associations in the same region because different causal variants are correlated through linkage disequilibrium. MR using a lead pQTL can then produce a persuasive but non-causal estimate. Colocalization asks whether the molecular trait and disease outcome are compatible with a shared causal variant.
Analysis sequence
- Variant harmonization: Put pQTL, eQTL, and GWAS data on the same genome build; align effect alleles; remove strand-ambiguous variants when allele frequency cannot resolve them; and verify rsIDs against positions.
- Fine-mapping: Fine-map each trait in a pre-specified GRN region, using ancestry-matched LD and a method that permits multiple causal variants such as SuSiE.
- Colocalization: Run pairwise analyses for plasma PGRN versus AD, CSF PGRN versus AD, brain GRN expression versus AD, and GRN expression versus PGRN; use a multiple-signal method such as coloc with SuSiE where warranted.
- Prior sensitivity: Report posterior probabilities across defensible prior settings rather than treating one PP4 value as absolute truth.
- Cross-ancestry replication: Repeat where adequately powered summary statistics and matched LD exist; interpret failure to replicate as uncertainty unless power is sufficient.
A practical preregistered criterion could call colocalization supportive when the posterior probability for a shared causal signal is at least 0.80 and remains high under prior sensitivity analyses, while the probability of two distinct causal variants is low. This threshold is a decision rule, not a law of nature.
What would invalidate the naive coloc result
Standard single-causal-variant coloc can be misleading in loci with multiple signals. Poor LD reference matching can distort fine-mapping. A shared association can also colocalize with several nearby genes, so a high PP4 for GRN does not by itself prove GRN is the only causal gene. You should inspect neighboring eQTLs, splice QTLs, chromatin QTLs, and protein-altering variants and ask whether the same credible set supports an alternative molecular mechanism.
Mendelian Randomization Plan
Core MR assumptions
For a genetic instrument Z, protein exposure X, and outcome Y, state the assumptions explicitly.
Relevance: Z changes the exposure of interest. For a neural target, relevance must be demonstrated for the compartment you claim to instrument.
Independence: Z is independent of causes of the outcome, conditional on design assumptions. Population structure, assortative mating, dynastic effects, and selection can undermine this.
Exclusion restriction: Z affects the outcome only through the target exposure. Horizontal pleiotropy, a neighboring causal gene, or a protein-independent effect of the variant violates this assumption.
To interpret an MR estimate as a drug-target effect, add gene-intervention equivalence or an explicit transportability assumption: the biological consequences of genetically higher PGRN must be sufficiently similar to those of the intervention. Effect estimation may additionally require homogeneity or monotonicity assumptions, plus correct model specification, allele harmonization, and appropriate handling of sample overlap.
Primary MR estimand
If rs5848 is the only defensible cis instrument after fine-mapping and colocalization, use a Wald ratio for the effect of genetically predicted PGRN on AD. Report the exposure scale clearly. A log2 protein scale permits interpretation per genetically predicted doubling, but do not extrapolate far beyond the observed genetic perturbation without emphasizing nonlinearity and time-scale assumptions.
A single-variant MR has a major limitation: you cannot estimate heterogeneity across instruments, and methods such as MR-Egger cannot rescue pleiotropy identification. The credibility must therefore come from molecular colocalization, functional annotation, neighboring-gene checks, and independent compartments rather than from a long menu of underpowered sensitivity tests.
Multi-instrument sensitivity analysis
If conditional fine-mapping yields several approximately independent cis instruments that all affect the same molecular exposure, compare inverse-variance weighted, weighted-median, and robust methods as appropriate. Use MR-Egger only when instrument count and strength make its assumptions remotely plausible. Report conditional F statistics or an equivalent strength metric. Do not add trans instruments merely to increase power if they introduce distinct biological pathways.
Pleiotropy Audit
Cis pleiotropy
At the GRN locus, test whether the instrument colocalizes with expression or splicing of neighboring genes as well as GRN. Examine coding consequences and regulatory annotations. A shared causal variant that changes several molecular traits creates an identification problem: statistical colocalization can show a shared variant, but it cannot by itself identify which molecular mediator carries the disease effect.
Trans pleiotropy and SORT1
SORT1 is biologically attractive because manipulating sortilin changes PGRN. It is also a warning example. A SORT1-region instrument has established effects on additional systemic phenotypes, including lipid-related biology. Therefore, a trans-SORT1 MR estimate for PGRN can violate the exclusion restriction even if the PGRN association is real. Treat SORT1 genetics as mechanistic triangulation and as an intervention-proxy analysis only after testing alternative pathways; do not automatically combine it with a GRN cis instrument.
Practical checks
Query PheWAS and OpenGWAS resources for instrument associations; test colocalization of the same signal with plausible alternative mediators; repeat analyses after excluding instruments with clear horizontal pathways; compare cis-only and cis-plus-trans estimates without averaging away disagreement; and perform negative-control outcome analyses where a biologically unrelated outcome can expose broad pleiotropy.
Reverse Causality and Directionality
Why reverse causality remains possible in protein epidemiology
Neurodegeneration changes cell composition, lysosomal stress, inflammation, vascular integrity, and protein clearance. A measured PGRN difference in cases can therefore be a consequence of disease. MR reduces ordinary reverse causality because germline genotype precedes disease, but interpretation can still fail if the genetic instrument changes another upstream trait or if disease liability alters the measured exposure through selection or index-event mechanisms.
Directionality checks
Run bidirectional MR using independent AD-liability instruments as the exposure and PGRN as the outcome, while recognizing that a null reverse MR does not prove absence of feedback. Apply Steiger-type directionality checks only as a supporting diagnostic because measurement error can reverse the variance comparison. Compare pre-symptomatic and symptomatic cohorts longitudinally. Examine whether genotype-protein associations are stable across disease stages. Prioritize instruments whose molecular effect is demonstrated in unaffected participants as well as disease cohorts.
Separating Plasma from Brain Exposure
Compartment ladder
Use a five-level exposure ladder: plasma protein → CSF protein → bulk brain RNA → cell-type-specific brain RNA or protein → functional intracellular or extracellular target engagement. Evidence does not automatically move upward on this ladder.
For PGRN, plasma and CSF can both respond to genotype and therapy, but their effect sizes and regulators differ. The intervention may also change PGRN in blood more strongly or earlier than in brain. Your dossier should therefore present plasma and CSF estimates in separate columns and should never pool them as if they were repeated measures of the same exposure.
Compartment-specific decision rule
Call a genetic instrument CNS-relevant only if it has a reproducible effect on CSF PGRN or a credible brain molecular proxy and the same causal signal is compatible with the disease association. Plasma-only instruments may be used to study systemic PGRN but must not be used to claim a brain PGRN effect without a validated bridge.
Target-Trial Emulation
The hypothetical target trial
A target-trial protocol for a future PGRN-elevating intervention should define: eligible adults by genotype, disease stage, biomarker state, and contraindications; a precise intervention strategy including molecule, dose, route, and persistence; an active or usual-care comparator; assignment at a common time zero; follow-up long enough to observe clinical progression; a clinical outcome and key safety outcomes; a causal contrast such as treatment-policy or per-protocol effect; and a pre-specified analysis plan.
Do not define treatment as “high PGRN” versus “low PGRN.” Protein concentration is a post-baseline state affected by many causes and does not describe a manipulable intervention.
Emulation assumptions
Consistency: each observed treatment history corresponds to the well-defined strategy being emulated.
Conditional exchangeability: after adjustment for measured baseline and, for per-protocol effects, time-varying confounders, treatment groups are comparable with respect to potential outcomes.
Positivity: every covariate pattern in the target population has a non-zero probability of receiving each treatment strategy under comparison.
Correct time zero: eligibility, treatment assignment, and start of follow-up are aligned to avoid immortal-time and selection bias.
No informative loss to follow-up after adjustment: censoring can be treated as conditionally independent, or appropriately weighted.
Measurement validity: treatment, adherence, confounders, outcomes, and competing events are measured sufficiently well.
No interference or a justified interference model: one participant’s treatment does not meaningfully change another participant’s outcome.
At present, latozinemab is not an established routine therapy that supplies a broad real-world initiator cohort, and INFRONT-3 was clinically negative. Therefore, a claims-based target-trial emulation of latozinemab effectiveness is not currently identified merely because EHR data exist. The correct response to absent treatment variation is to stop, not to substitute a biomarker comparison.
Triangulation Logic
Evidence that points in the same direction
The strongest directionally coherent chain is: GRN loss-of-function causes progranulin deficiency and FTD; the rs5848 T allele lowers GRN expression and PGRN and is associated with higher AD risk in major datasets; plasma and CSF cis-pQTL signals at GRN have been reported to colocalize with the AD signal; and pharmacologic sortilin blockade can raise plasma and CSF PGRN.
Evidence that limits the claim
Compartment differences mean that plasma cannot stand in for brain. Common-variant effects may not mimic large therapeutic changes. The main common instrument can be effectively single-variant, limiting empirical pleiotropy tests. Ancestry-specific results limit transportability. Most importantly, the phase 3 FTD-GRN trial showed that strong biomarker engagement did not yield the expected clinical benefit under that protocol. This forces a separation between target engagement and clinical efficacy.
What the dossier may conclude
A defensible dossier can conclude that human genetics supports a causal role for GRN/PGRN biology in neurodegeneration and supports the hypothesis that lower PGRN increases AD risk, provided the reported colocalization and molecular-direction findings reproduce under a pre-specified audit. It should not conclude that any PGRN-raising drug will prevent AD, nor that plasma PGRN is a validated surrogate endpoint, nor that the negative FTD-GRN trial directly disproves the AD genetic association.
Stopping Rule for Causal Direction
Pre-register this rule before looking at the final MR effect estimate:
Stop and report “causal direction unresolved” if no instrument survives all three gates: Gate 1, the variant changes the claimed exposure in the relevant CNS compartment or a justified CNS proxy; Gate 2, the molecular QTL and disease GWAS are compatible with a shared fine-mapped causal signal under sensitivity analyses; Gate 3, there is no equally plausible horizontal pathway supported by colocalized molecular traits, phenome-wide associations, or known biology.
Also stop if forward and reverse-direction analyses remain similarly compatible with the data; if allele harmonization or ancestry-matched LD cannot be secured; if a plasma-only signal is being used to infer brain exposure; or if the result depends on one prior setting or one assay platform. When the stopping rule fires, do not average conflicting evidence into a “best estimate.” The scientific result is non-identification.
A separate translation stopping rule applies after causal direction is supported: do not claim that a specific therapy is effective when its randomized clinical endpoint is null. Genetic support can justify a new intervention design or disease-stage hypothesis, but it cannot overwrite the trial.
Reproducible Audit Workflow
Minimum dossier fields
Create a machine-readable table with one row per evidence item and the fields: source; accession; phenotype; compartment; assay; ancestry; sample size; genome build; variant; effect allele; other allele; beta or log odds ratio; standard error; p value; allele frequency; LD reference; conditioning status; fine-mapping method; credible set; colocalization posterior; exposure scale; outcome definition; recruitment period; and access date.
Suggested public resources
Use UKB-PPP for plasma pQTLs; NIAGADS NG00130 and GWAS Catalog for CSF proteogenomic summary statistics; GTEx and MetaBrain for brain eQTLs; FinnGen and major AD GWAS publications for disease outcomes; and ClinicalTrials.gov NCT04374136 for the intervention record.
Evidence register
The following sources anchor the worked example and should be re-checked at the time of analysis:
- Progranulin: Robins and colleagues, 2026, Genetic evidence supports therapeutic modulation of progranulin levels in Alzheimer’s Disease, Molecular Neurodegeneration Advances, DOI 10.1186/s44477-025-00011-y.
- Proteomics: Sun and colleagues, 2023, Plasma proteomic associations with genetics and health in the UK Biobank, Nature, DOI 10.1038/s41586-023-06592-6.
- Cerebrospinal fluid: public CSF pQTL resources linked through NIAGADS NG00130 and GWAS Catalog accessions GCST90421033–GCST90428040.
- Frontotemporal dementia: phase 1 latozinemab reports showing plasma and CSF PGRN elevation and ClinicalTrials.gov NCT04374136 for INFRONT-3.
- Mendelian randomization: STROBE-MR for relevance, independence, exclusion restriction, and transparent reporting.
- Target trial: the TARGET Statement and the Hernán-Robins target-trial framework for explicit protocol specification and causal estimands.
- Colocalization: the Bayesian colocalization framework, supplemented by multiple-signal fine-mapping approaches when loci contain more than one causal variant.
Interactive Tasks
Quiz: Test Your Knowledge
Which evidence most directly supports a CNS-relevant genetic effect on progranulin abundance? (A cis pQTL for progranulin in cerebrospinal fluid) (!A plasma biomarker association after diagnosis) (!A trans pQTL with unknown pleiotropy) (!A case control difference in serum protein)
What is the main purpose of colocalization in this dossier? (To test whether molecular and disease traits can share a causal variant) (!To prove that a protein is druggable) (!To eliminate all population stratification) (!To estimate trial adherence)
Which MR assumption is violated by a variant that affects disease through a pathway independent of progranulin? (Exclusion restriction) (!Relevance) (!Positivity) (!Consistency)
Why should plasma and CSF progranulin be analyzed separately? (They can have different genetic and biological determinants) (!They are measured in identical tissues) (!Plasma always reflects brain concentration) (!CSF is a randomized exposure)
What did the INFRONT-3 result demonstrate most directly? (Latozinemab raised a biomarker but did not meet the clinical co-primary endpoint) (!All progranulin biology is noncausal) (!Plasma progranulin is a validated surrogate endpoint) (!Alzheimer disease risk increases when progranulin rises)
What is the safest use of a SORT1 trans instrument in this setting? (As mechanistic triangulation after explicit pleiotropy checks) (!As an automatic replacement for a GRN cis instrument) (!As proof that plasma and brain exposure are identical) (!As a way to avoid colocalization)
Which design feature prevents immortal-time bias in a target-trial emulation? (Aligning eligibility treatment assignment and follow up at time zero) (!Selecting only long term survivors) (!Defining treatment from future biomarker values) (!Starting follow up after outcome assessment)
What should happen if the molecular QTL and disease GWAS favor distinct causal variants? (The causal direction claim should stop pending better identification) (!The MR estimate should be reported as definitive) (!A trans instrument should be added automatically) (!The plasma result should replace the brain result)
Which statement best describes a single variant Wald-ratio MR? (It can estimate a ratio but offers limited empirical pleiotropy diagnostics) (!It automatically proves the exclusion restriction) (!It requires no allele harmonization) (!It is immune to linkage disequilibrium)
What does a genetically predicted protein effect estimate represent most directly? (A lifelong genetic perturbation under MR assumptions) (!A guaranteed short term drug effect) (!A randomized treatment policy effect) (!A direct measure of brain drug concentration)
Memory Game
| cis-pQTL | Variant near the encoding gene associated with protein abundance |
| Colocalization | Test of whether two association signals are compatible with a shared causal variant |
| Pleiotropy | Effect of a genetic variant on multiple biological pathways |
| CSF | CNS-proximal fluid compartment used for protein measurement |
| Relevance | MR assumption that the instrument changes the exposure |
| Exchangeability | Target-trial assumption that adjusted treatment groups are comparable |
| Timezero | Common start of eligibility assignment and follow-up |
| Progranulin | Secreted protein encoded by GRN and used as the worked target |
Drag and Drop
| Match the correct terms. | Topic |
|---|---|
| GRN cis instrument | Protein perturbation close to the encoding gene |
| CSF pQTL | CNS-proximal protein genetic association |
| Colocalization | Shared-signal assessment |
| Reverse MR | Disease liability tested as the exposure |
| Target trial | Explicit hypothetical randomized protocol |
...
Crossword Puzzle
| Progranulin | Which GRN-encoded protein is the worked target? |
| Sortilin | Which receptor is blocked by latozinemab? |
| Pleiotropy | What term describes one variant acting through multiple pathways? |
| Colocalization | What analysis tests whether two traits can share a causal variant? |
| Exchangeability | What target-trial assumption concerns comparability after adjustment? |
| Harmonization | What process aligns alleles and variant coding across datasets? |
LearningApps
Cloze Text
Open-Ended Tasks
Easy
- Evidence map: Draw a one-page causal map linking GRN genotype, brain expression, CSF PGRN, plasma PGRN, neurodegeneration, and intervention; label every arrow as measured, assumed, or experimentally supported.
- Compartment audit: Create a two-column table comparing plasma and CSF PGRN evidence and explain why the columns must not be pooled.
- Variant card: Produce a data card for rs5848 containing alleles, molecular direction, disease direction, ancestry, assay, and the exact public source you used.
- Trial summary: Record a three-minute video explaining why biomarker target engagement in INFRONT-3 is not equivalent to clinical efficacy.
Standard
- pQTL replication: Download public GRN-region pQTL summary statistics from one plasma and one CSF source, harmonize effect alleles, and compare effect directions and scales.
- Brain eQTL audit: Query GTEx and MetaBrain for GRN, document tissue-specific cis signals, and write a short report on whether the same variant is likely to regulate GRN across brain regions.
- Colocalization protocol: Write an analysis-ready protocol for fine-mapping and coloc.susie of GRN pQTL, brain eQTL, and AD GWAS data, including LD reference and prior sensitivity.
- Pleiotropy interview: Interview a statistical geneticist, neurologist, or pharmacologist about one plausible non-PGRN pathway for a GRN or SORT1 instrument and summarize how it would alter interpretation.
Advanced
- Triangulated dossier: Reproduce the full target dossier with a versioned evidence table, allele harmonization log, fine-mapping results, colocalization posteriors, MR estimate, and explicit uncertainty statement.
- Reverse causality stress test: Design and, where data permit, run bidirectional MR and directionality analyses, then explain which conclusions depend on protein measurement error or case ascertainment.
- Target trial protocol: Specify a future target trial and an observational emulation for a clearly defined PGRN-elevating intervention, including eligibility, strategies, time zero, outcomes, causal contrast, confounders, censoring, and positivity diagnostics.
- Stopping rule simulation: Create synthetic scenarios in which colocalization succeeds or fails, pleiotropy is present or absent, and plasma and CNS effects agree or disagree; apply the pre-registered stopping rule without changing it after seeing the simulated result.
Learning Assessment
- Causal identification memo: Given harmonized GRN-region pQTL and AD GWAS summary statistics, justify whether an MR estimate is identified, explicitly addressing compartment, colocalization, pleiotropy, and reverse causality.
- Compartment transfer critique: Evaluate a claim that a plasma PGRN increase proves brain target engagement and identify the extra evidence required for that inference.
- Trial versus genetics synthesis: Reconcile a supportive AD genetic analysis with the negative INFRONT-3 clinical result without treating either as automatically overriding the other.
- Instrument selection defense: Compare a GRN cis instrument with a SORT1 trans instrument and defend which question each can answer under stated assumptions.
- Target-trial emulation design: Build a protocol table that aligns eligibility, treatment assignment, time zero, follow-up, outcomes, estimand, and analysis, then identify at least three ways real-world data could fail to emulate it.
- Ancestry transportability analysis: Explain how ancestry-specific LD and allele-frequency differences can change fine-mapping, colocalization, and the transportability of an MR estimate.
- Stop or continue decision: Apply the course stopping rule to an evidence packet with conflicting QTL and GWAS signals and write a one-page decision that reports non-identification when appropriate.
Evidence of Learning
Strong evidence of learning includes a reproducible and versioned target dossier; correct separation of plasma, CSF, brain, and cell-specific evidence; an allele-harmonized QTL-GWAS dataset; fine-mapping and colocalization outputs with prior sensitivity; an MR analysis with explicit assumptions rather than only a point estimate; a documented pleiotropy and reverse-causality audit; a target-trial protocol with a defensible time zero and estimand; a synthesis of randomized and genetic evidence; and a pre-registered stopping rule that is actually obeyed when causal direction cannot be identified.
You should also demonstrate transfer: explain how the same workflow would change for a membrane receptor, an intracellular enzyme, or a protein with only trans-pQTL instruments. The key skill is not memorizing progranulin facts but knowing which causal question each data layer can and cannot identify.
OERs on the Topic
Open resources for deeper work include STROBE-MR, the TARGET Statement, GTEx, MetaBrain, GWAS Catalog, NIAGADS, and ClinicalTrials.gov.
Linked Learning Areas
aiMOOC Projects
MOOCwiki · Deutsch
Nach dem Lernen ist vor dem Lernen
Entdecke direkt den nächsten Lernkurs. Weitere Inhalte erscheinen, wenn Du weiter nach unten scrollst.
Zur MOOCwiki-HauptseiteMediathek
Mediathek wird aus dem Wiki geladen ...
Keine passenden Inhalte gefunden. Bitte ändere Suche oder Filter.