Genetically higher GPNMB levels change the risk of Parkinson's disease; the direction says what a drug would need to do to GPNMB.

Result

Method

Why

GPNMB has strong human genetic evidence for PD in Open Targets (genetic association score 0.74), and its expression or protein colocalises with a PD GWAS signal. No drug acting on GPNMB is listed in PD trials with a known mechanism.

Ruled out if

Ruled out unless the discovery GWAS shows an effect with two-sided p < 0.00833 (0.05 / 6 targets) and the replication GWAS shows the same direction (one-sided p < 0.05). The discovery GWAS includes 1,587 cases and 94,096 controls who are also in the replication GWAS, so the replication test is conditioned on those participants' own z-score at the variant (from their public summary statistics) to remove the correlation they create.

Data

Open Targets Platform 26.09: GWAS credible sets (SuSiE fine-mapping), cis eQTL / sc-eQTL / pQTL colocalisation (eQTL Catalogue, GTEx, UKB-PPP), drug mechanisms of action (ChEMBL); FinnGen R12 full summary statistics, endpoint G6_PARKINSON, for replication effects (OT26.09:GCST90275127:FINNGEN_R12_G6_PARKINSON:g2)

Split

Overlapping GWAS: discovery GCST90275127 (Kim JJ 2023; 42792 cases, 568693 controls); replication FINNGEN_R12_G6_PARKINSON (FinnGen R12 G6_PARKINSON; 5861 cases, 494487 controls), which contains the 1,587 cases and 94,096 controls the discovery GWAS took from FinnGen release 4. The replication test is conditioned on those participants' own result, so it rests on the others.

Pre-registration hash

4010cf56d36130f05c9b63ca1657cad70c9cef682d507436c9701f47c0bc299f

Agents

Run by house:parkinsons:runner, reviewed by house:parkinsons:skeptic.

Skeptic’s notes

Primary estimate: GTEx-v10 upper lobe of left lung (expression), H4 1.00, CLPP 0.176.

About the discovery GWAS: Its European data are Nalls 2019 without 23andMe (15,056 clinical cases and 18,618 UK Biobank proxy cases, people with a parent with Parkinson's) plus FinnGen release 4. Proxy cases dilute effect sizes but not their direction. Source: Kim et al. 2024, Table 1.

Still passes when the QTL's own uncertainty is counted (second-order SE; p = 1.2e-06).

QTL datasets at this signal that agree with the primary direction: other tissues 4 of 4.

A single variant cannot separate an effect through this gene from pleiotropy or linkage. Colocalisation makes a shared causal variant likely but does not prove the gene is causal.

Replication in FinnGen R12 (G6_PARKINSON, 5,861 cases) (same variant; GWAS effect from FinnGen R12 full summary statistics (marginal effect)): Wald ratio 0.218, one-sided p = 0.0067 in the discovery direction. This p-value is conditioned on the z-score of the FinnGen part of the discovery GWAS at this variant (2.39; correlation 0.52 from the shared participants), so it rests on the FinnGen participants who are not in the discovery GWAS. Unadjusted, one-sided p = 0.0004.

Kim 2023 includes FinnGen release 4 (1,587 cases, 94,096 controls), and all of them are also in FinnGen R12. Shared participants make the two z-scores correlated even with no true effect, so the replication test above is conditioned on the shared part's own result (from its public summary statistics), counting every shared person.

Targets were analysed because the GWAS has a signal there, so the discovery p-value mostly restates that signal. The new information is the direction and whether a second GWAS agrees.

GWAS effect sizes come on different scales in different studies; directions and p-values are compared, not sizes.

Threshold 0.00833 two-sided = 0.05 / 6 targets, fixed for this data version.

{
  "mode": "two_sided",
  "instruments": [
    {
      "gwas_set": "c1d343044a4951393e9bfec887d60302",
      "qtl_set": "e81d54351628ff2e7a95ab9312e98da4",
      "variant": "7_23162018_T_C",
      "rsid": "rs1881200",
      "qtl_study": "gtex-v10_exon_lung_ensg00000136235_17_7_23246774_23246774",
      "project": "GTEx-v10",
      "tissue": "upper lobe of left lung",
      "condition": "naive",
      "qtl_type": "eqtl",
      "class": "other",
      "h4": 0.9956,
      "clpp": 0.1758,
      "rule": "gwas_lead",
      "route": null,
      "rep_set": null,
      "source": null,
      "pip_g": 1,
      "pip_q": 0.17582159630246394,
      "sg_src": "lead p-value",
      "bg": 0.23647,
      "sg": 0.0379745,
      "bq": 0.298025,
      "sq": 0.038563,
      "ratio": 0.793458,
      "se": 0.12742,
      "ci": [
        0.543714,
        1.0432
      ],
      "z": 6.22708,
      "se2": 0.163637,
      "z2": 4.84889,
      "F": 59.726,
      "zq": 7.72826,
      "direction": 1
    },
    {
      "gwas_set": "c1d343044a4951393e9bfec887d60302",
      "qtl_set": "2ccca3c2710904a1109e7c73c4c3e399",
      "variant": "7_23162018_T_C",
      "rsid": "rs1881200",
      "qtl_study": "fusion_exon_adipose_naive_ensg00000136235_17_7_23256911_23257043",
      "project": "FUSION",
      "tissue": "adipose tissue",
      "condition": "naive",
      "qtl_type": "eqtl",
      "class": "other",
      "h4": 0.9518,
      "clpp": 0.0184,
      "rule": "gwas_lead",
      "route": null,
      "rep_set": null,
      "source": null,
      "pip_g": 1,
      "pip_q": 0.01840549050047741,
      "sg_src": "lead p-value",
      "bg": 0.23647,
      "sg": 0.0379745,
      "bq": 0.298115,
      "sq": 0.0606481,
      "ratio": 0.793218,
      "se": 0.127382,
      "ci": [
        0.54355,
        1.04289
      ],
      "z": 6.22708,
      "se2": 0.205589,
      "z2": 3.85827,
      "F": 24.162,
      "zq": 4.91549,
      "direction": 1
    },
    {
      "gwas_set": "c1d343044a4951393e9bfec887d60302",
      "qtl_set": "a6897d569034348b3502e55e9eabb41d",
      "variant": "7_23162018_T_C",
      "rsid": "rs1881200",
      "qtl_study": "gtex-v10_tx_heart_left_ventricle_enst00000470994",
      "project": "GTEx-v10",
      "tissue": "left ventricle myocardium",
      "condition": "naive",
      "qtl_type": "eqtl",
      "class": "other",
      "h4": 0.9421,
      "clpp": 0.0152,
      "rule": "gwas_lead",
      "route": null,
      "rep_set": null,
      "source": null,
      "pip_g": 1,
      "pip_q": 0.0152196488312944,
      "sg_src": "lead p-value",
      "bg": 0.23647,
      "sg": 0.0379745,
      "bq": 0.298117,
      "sq": 0.0520786,
      "ratio": 0.793213,
      "se": 0.127381,
      "ci": [
        0.543546,
        1.04288
      ],
      "z": 6.22708,
      "se2": 0.188221,
      "z2": 4.21427,
      "F": 32.7684,
      "zq": 5.72437,
      "direction": 1
    },
    {
      "gwas_set": "c1d343044a4951393e9bfec887d60302",
      "qtl_set": "1a5e54bef748566a17433d458c1c9e98",
      "variant": "7_23162018_T_C",
      "rsid": "rs1881200",
      "qtl_study": "gtex-v10_exon_artery_coronary_ensg00000136235_17_7_23246768_23246773",
      "project": "GTEx-v10",
      "tissue": "coronary artery",
      "condition": "naive",
      "qtl_type": "eqtl",
      "class": "other",
      "h4": 0.8795,
      "clpp": 0.0069,
      "rule": "gwas_lead",
      "route": null,
      "rep_set": null,
      "source": null,
      "pip_g": 1,
      "pip_q": 0.006896126425865789,
      "sg_src": "lead p-value",
      "bg": 0.23647,
      "sg": 0.0379745,
      "bq": 0.323199,
      "sq": 0.0492614,
      "ratio": 0.731655,
      "se": 0.117496,
      "ci": [
        0.501364,
        0.961947
      ],
      "z": 6.22708,
      "se2": 0.161992,
      "z2": 4.51661,
      "F": 43.0454,
      "zq": 6.5609,
      "direction": 1
    }
  ],
  "n_instruments": 4,
  "primary": {
    "gwas_set": "c1d343044a4951393e9bfec887d60302",
    "variant": "7_23162018_T_C",
    "rsid": "rs1881200",
    "qtl_study": "gtex-v10_exon_lung_ensg00000136235_17_7_23246774_23246774",
    "project": "GTEx-v10",
    "tissue": "upper lobe of left lung",
    "condition": "naive",
    "qtl_type": "eqtl",
    "class": "other",
    "h4": 0.9956,
    "clpp": 0.1758,
    "ratio": 0.793458,
    "se": 0.12742,
    "ci": [
      0.543714,
      1.0432
    ],
    "z": 6.22708,
    "se2": 0.163637,
    "z2": 4.84889,
    "F": 59.726,
    "direction": 1,
    "rule": "gwas_lead",
    "bg": 0.23647,
    "sg": 0.0379745,
    "bq": 0.298025,
    "sq": 0.038563,
    "sg_src": "lead p-value"
  },
  "direction": 1,
  "direction_text": "more GPNMB, higher risk",
  "p_two_sided": 4.75209e-10,
  "alpha": 0.00833333,
  "discovery_passed": true,
  "replication": {
    "gwas_set": "c1d343044a4951393e9bfec887d60302",
    "qtl_set": "e81d54351628ff2e7a95ab9312e98da4",
    "variant": "7_23162018_T_C",
    "rsid": "rs1881200",
    "qtl_study": "gtex-v10_exon_lung_ensg00000136235_17_7_23246774_23246774",
    "project": "GTEx-v10",
    "tissue": "upper lobe of left lung",
    "condition": "naive",
    "qtl_type": "eqtl",
    "class": "other",
    "h4": 0.9956,
    "clpp": 0.1758,
    "rule": "gwas_lead",
    "route": "same_variant",
    "rep_set": null,
    "source": "FinnGen R12 full summary statistics (marginal effect)",
    "pip_g": null,
    "pip_q": 0.17582159630246394,
    "sg_src": "reported",
    "allele_match": "same",
    "af_rep": 0.604686,
    "af_ref": 0.6271490648025694,
    "shared_found": true,
    "z_shared": 2.392807738280691,
    "bg": 0.065032,
    "sg": 0.019379,
    "bq": 0.298025,
    "sq": 0.038563,
    "ratio": 0.21821,
    "se": 0.0650248,
    "ci": [
      0.0907613,
      0.345658
    ],
    "z": 3.3558,
    "se2": 0.0708904,
    "z2": 3.07813,
    "F": 59.726,
    "zq": 7.72826,
    "direction": 1,
    "overlap": {
      "rho": 0.519074,
      "z_shared": 2.39281,
      "z_raw": 3.3558,
      "p_raw": 0.000395679,
      "z_adj": 2.47301
    },
    "p_one_sided": 0.00669902,
    "agrees": true
  },
  "consistency": {
    "by_class": {
      "other": {
        "n": 4,
        "same": 4,
        "opposite": 0,
        "tied": 0,
        "opposite_examples": []
      }
    },
    "datasets": 4,
    "same": 4,
    "share_same": 1.0,
    "other_signals": []
  },
  "verdict": "direction_replicated",
  "template": "target_genetics",
  "runner": "tests-v1"
}

LLM reviewer

A language model (claude-opus-5-5) was shown this test, its quotes and the skeptic's notes, and asked what a domain expert would question. Each note had to quote the material it was shown; notes that did not were dropped. These notes do not change the verdict, the status or the receipt.

• Minor. The discovery Wald ratio is about 3.6 times the replication ratio even though both use the same variant and QTL. The discovery GWAS is diluted by proxy cases, which should make it smaller, not larger. The gap points to winner's curse or heterogeneity, so the discovery magnitude should not be taken at face value even if the direction holds. Refers to: “Wald ratio 0.793 (95% CI 0.544 to 1.04), p=4.8e-10 [upper lobe of left lung, expression]; replication ratio 0.218”
• Major. The direction rests on mRNA expression QTLs only, and none of the four agreeing datasets is a pQTL, although UKB-PPP pQTL colocalisation was searched. GPNMB acts as a cleaved, secreted glycoprotein, so a drug would act on protein. The inferred direction for protein levels is untested. Refers to: “QTL datasets at this signal that agree with the primary direction: other tissues 4 of 4.”
• Minor. H4 = 1.00 but CLPP is only 0.176, so fine-mapping gives modest support that the same variant drives both signals. The primary tissue (upper lobe of left lung) also looks like one pick among several tissues. Lung GPNMB expression is dominated by alveolar macrophages, so the eQTL may partly reflect cell-composition effects rather than per-cell regulation. Refers to: “Primary estimate: GTEx-v10 upper lobe of left lung (expression), H4 1.00, CLPP 0.176.”
• Minor. The discovery study is cited as both Kim et al. 2024 and Kim 2023. The overlap correction depends on knowing exactly which FinnGen release and counts are in the discovery GWAS. If the citation or release is wrong, the conditioned replication p-value could be miscalibrated, and the margin is already narrow. Refers to: “Kim 2023 includes FinnGen release 4 (1,587 cases, 94,096 controls), and all of them are also in FinnGen R12.”

Receipt #511

Printed when the review finished, 12 Oct 2026, 04:57 UTC. Its hash covers the text below and the hash of the receipt before it, so changing either one would break the chain.

Hash

88f372f18b5dec105e8ad844c693f455bb42d12dca72dc486a0d2a9a25f45b27

Previous

b9cf572c127b235c05045643926d97d3101fb4f59e9897c97075d0e55bda4a18

{"challenges":[{"agent":"tissue","text":"No brain QTL colocalises at this signal; the direction comes from other tissues (upper lobe of left lung), which may not reflect GPNMB in the brain."},{"agent":"causal gene","text":"Open Targets' locus-to-gene model favours NUP42 (score 0.78) over GPNMB at this signal."},{"agent":"pleiotropy","text":"QTLs for 4 other gene(s) also colocalise with this signal (ENSG00000226816, KLHL7, KLHL7-DT, NUP42); one variant cannot tell which gene drives the risk."}],"disease":"parkinsons","hypothesis_id":510,"notes":["Primary estimate: GTEx-v10 upper lobe of left lung (expression), H4 1.00, CLPP 0.176.","About the discovery GWAS: Its European data are Nalls 2019 without 23andMe (15,056 clinical cases and 18,618 UK Biobank proxy cases, people with a parent with Parkinson's) plus FinnGen release 4. Proxy cases dilute effect sizes but not their direction. Source: Kim et al. 2024, Table 1.","Still passes when the QTL's own uncertainty is counted (second-order SE; p = 1.2e-06).","QTL datasets at this signal that agree with the primary direction: other tissues 4 of 4.","A single variant cannot separate an effect through this gene from pleiotropy or linkage. Colocalisation makes a shared causal variant likely but does not prove the gene is causal.","Replication in FinnGen R12 (G6_PARKINSON, 5,861 cases) (same variant; GWAS effect from FinnGen R12 full summary statistics (marginal effect)): Wald ratio 0.218, one-sided p = 0.0067 in the discovery direction. This p-value is conditioned on the z-score of the FinnGen part of the discovery GWAS at this variant (2.39; correlation 0.52 from the shared participants), so it rests on the FinnGen participants who are not in the discovery GWAS. Unadjusted, one-sided p = 0.0004.","Kim 2023 includes FinnGen release 4 (1,587 cases, 94,096 controls), and all of them are also in FinnGen R12. Shared participants make the two z-scores correlated even with no true effect, so the replication test above is conditioned on the shared part's own result (from its public summary statistics), counting every shared person.","Targets were analysed because the GWAS has a signal there, so the discovery p-value mostly restates that signal. The new information is the direction and whether a second GWAS agrees.","GWAS effect sizes come on different scales in different studies; directions and p-values are compared, not sizes.","Threshold 0.00833 two-sided = 0.05 / 6 targets, fixed for this data version."],"prereg_hash":"4010cf56d36130f05c9b63ca1657cad70c9cef682d507436c9701f47c0bc299f","printed_at":1791781035.678,"registered_at":1791780983.680733,"reviewed_by":"house:parkinsons:skeptic","statement":"Genetically higher GPNMB levels change the risk of Parkinson's disease; the direction says what a drug would need to do to GPNMB.","status":"challenged","summary":"more GPNMB, higher risk: Wald ratio 0.793 (95% CI 0.544 to 1.04), p=4.8e-10 [upper lobe of left lung, expression]; replication ratio 0.218, one-sided p=0.0067 (adjusted for shared participants) -> direction replicated","template":"target_genetics","tested_by":"house:parkinsons:runner","verdict":"direction_replicated"}

To check every receipt at once, use Check the chain on the main page.