表位与 PAV 调查

4X · ★唯一否决点 · 图 1 张 · 发表版表格 4 个　（流程见 v5 定稿）

复制报警之后才做的调查。⛔ 否决的是「调查结论」，不是「复制结果」。

---

图

Figure F4X. Epitope and protein-altering-variant investigation of the 6 associations flagged by external replication

完整图注（英文，与投稿版一致）

Figure F4X. Epitope and protein-altering-variant investigation of the 6 associations flagged by external replication.

Why this step exists. A cis-pQTL can look causal simply because the variant changes the protein's amino-acid sequence and so changes how well the assay's antibody or aptamer binds it. The measured 'level' then differs without the actual abundance differing. Only this step can rule an association out on that basis; every earlier step either triages or annotates.

(a) The evidence examined. Y = present, - = absent, n/a = the question does not apply. 5 of the 6 flagged associations carry a missense variant in the target gene in near-perfect linkage disequilibrium with the instrument (r2 >= 0.99), yet only one was ruled out. Presence of a protein-altering variant is therefore treated as NECESSARY BUT NOT SUFFICIENT: on its own it cannot distinguish a binding artifact from a genuine effect of the amino-acid change on protein function or clearance.

(b) Verdicts. APOL1 for diabetic maculopathy is the single association vetoed anywhere in this workflow. What decided it was not the presence of the variant but independent evidence that the variant does not alter APOL1 abundance (Wang et al. 2026, iScience, PMID 41767276): if abundance is unchanged while the assay reports a change, the assay is reading the amino-acid substitution rather than the protein level.

The other verdicts. AOC1: a missense variant is present but a genuine effect on abundance remains plausible, so the association is retained with the caveat stated. MICB_MICA: two aptamers on the same platform disagree with each other, which points to ambiguity about which protein the probe is actually binding rather than to a resolved artifact; MICB and MICA are adjacent, highly polymorphic MHC genes. TNXB: the two platforms used different instruments that tag distinct cis signals, so they are not measuring the same thing and their disagreement is expected.

TRIM40 is included because it was flagged, but its alarm came from the alternative-outcome check, not from an alternative platform. Under the pre-specified workflow an alarm from that source is a question about the outcome cohorts, not about the assay, so the epitope questions are marked not applicable. The discovery and replication cohorts give effects of opposite sign (-0.34 versus +0.23) while their allele frequencies for the same allele differ by only 0.06, so the reversal is not explained by an allele mix-up. The association is retained and the heterogeneity reported.

Completeness. External replication raised 6 alarms in total: 4 from the alternative-platform check, 1 from the alternative-outcome check and 1 from the both-sides check. All 6 were investigated here; none was dropped. Sign reversals that did not reach significance in the alternative data were not investigated, on the grounds that the sign of a non-significant estimate is not evidence of anything; there were five such reversals in the alternative-outcome check and four in the both-sides check, and all of those associations are retained.

Linked variants. Across the four proteins taken down the epitope route, 25 protein-altering variants lie in the target gene in linkage disequilibrium with the instrument (r2 from 0.91 to 1.00); the strongest for each protein is the one shown in panel (a).

Method and thresholds. Protein-altering variants are variants in the target gene annotated as missense, stop gained, stop lost, start lost, or splice acceptor or donor. Linkage disequilibrium with the instrument is computed in the 1000 Genomes European panel within 500 kb; a variant counts as linked at r2 > 0.8, the threshold used by Zhao et al. 2023. Two instruments for the same protein are treated as tagging DISTINCT cis signals when r2 < 0.2 between them; that second threshold is defined by this study rather than taken from the literature, and is declared as such. Role in the analysis pipeline: this is the ONLY step that can veto an association on replication evidence. An association vetoed here is removed; an association retained here carries forward whatever caveat its verdict states.

Data sources and sample sizes. Exposure: UKB-PPP plasma proteome (Olink Explore 3072), 34,557 European participants, 1,954 proteins with a cis-pQTL. Outcomes: FinnGen R9 -- Diabetic retinopathy 10,413 cases / 308,633 controls; Diabetic maculopathy 3,572 cases / 308,547 controls; Diabetic nephropathy 4,111 cases / 308,539 controls; Diabetic neuropathy 2,843 cases / 271,817 controls.

---

发表版表格

文件	下载
artifact_aric_cis_lookup.csv	下载
artifact_investigation_4x.csv	下载
artifact_outcome_heterogeneity.csv	下载
artifact_pav_ld_annotation.csv	下载

本页图为网页版 PNG（最宽 1600 px）。投稿用的矢量 PDF 与 600 dpi TIFF 体积较大，留在仓库 results/external_replication/figures/，不随文档站分发。