糖尿病对照臂

F-DM · 图 3 张 · 发表版表格 7 个　（流程见 v5 定稿）

T1D / T2D 两条对照臂，用来区分「并发症特异」与「糖尿病驱动」——这是本研究方法学上的主句。

图

Figure F-DM-b. How the 57 significant protein-complication associations divide once the effect on diabetes liability is taken into account

完整图注（英文，与投稿版一致）

Figure F-DM-b. How the 57 significant protein-complication associations divide once the effect on diabetes liability is taken into account.

Counts across all complications: Diabetes-driven 35; Complication-enhanced 11; Complication-specific 10; Discordant direction 1. 61% of the associations are diabetes-driven, that is, their effect on the complication is not distinguishable in magnitude from their effect on diabetes liability itself.

Definitions. Diabetes-driven: the complication-side effect is not significantly larger than the diabetes-side effect. Complication-enhanced: the complication-side effect is significantly larger. Complication-specific: the diabetes-side estimate has P >= 0.05, so there is no detectable effect on diabetes liability to compare against. Discordant direction: the two estimates are significantly different and point opposite ways.

These four labels are DESCRIPTIVE CATEGORIES, not a ranking and not a filter. No category is preferred over another and no association is removed on the basis of this classification; the colours are four distinct hues rather than a light-to-dark ramp precisely so that no ordering is implied.

Method: the classification compares the complication-side and diabetes-side Wald ratio estimates for the same cis sentinel variant by a z test on their difference; the diabetes-side threshold for calling an association complication-specific is p >= 0.05. Role in the analysis pipeline: as for Figure F-DM-a, this is a stratified reading of the step-2 results, not a step of the pre-specified workflow.

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. Diabetes control arms: Type 1 diabetes 20,355 cases / 797,363 controls; Type 2 diabetes 55,005 cases / 400,308 controls.

Figure F-DM-a. Effect on each diabetic complication plotted against the effect of the same protein on diabetes liability, for all 57 associations that passed FDR < 0.05 in a complication

完整图注（英文，与投稿版一致）

Figure F-DM-a. Effect on each diabetic complication plotted against the effect of the same protein on diabetes liability, for all 57 associations that passed FDR < 0.05 in a complication.

Both axes are log odds ratios per one standard deviation increase in genetically predicted plasma protein level, estimated from the SAME cis sentinel variant and signed relative to the SAME effect allele. The diabetes-side estimate is taken from whichever diabetes arm (type 1 or type 2) shows the stronger effect for that protein. The dashed diagonal marks equal effect on diabetes and on the complication: points above it act more strongly on the complication than on diabetes liability.

Shaded quadrants contain associations whose two estimates point in OPPOSITE directions (8 of 57). This is the reason the figure keeps the sign on both axes: taking absolute values would fold these into the concordant cloud and make the discordance invisible. A diabetes-side P >= 0.05 is classified as complication-specific whatever the sign of that near-null estimate, so the shaded quadrants hold both the 1 formally discordant association and 7 complication-specific ones whose diabetes-side estimate is indistinguishable from zero.

Triangles mark proteins encoded in the MHC region. Labels: the associations classified as discordant plus the proteins named in the text; a protein significant for several complications is labelled once, at its largest complication-side effect.

Method: both axes are Wald ratio estimates from the same cis sentinel variant; the two are compared by a z test on their difference. Variance of the difference between the two estimates is computed assuming independence. This is conservative rather than liberal: the two estimates share the exposure instrument and the FinnGen control pool, so they are positively correlated and the true variance is smaller. Assuming independence therefore makes it HARDER, not easier, to call an association complication-specific.

Role in the analysis pipeline: this comparison is NOT a step of the pre-specified workflow and NOT a filter. It is a stratified reading of the step-2 results and removes no candidate.

Figure F-DM-c. Whether the diabetes control-arm estimates hold up when the diabetes GWAS is swapped for an independent one

完整图注（英文，与投稿版一致）

Figure F-DM-c. Whether the diabetes control-arm estimates hold up when the diabetes GWAS is swapped for an independent one.

Why this matters: the main type 1 diabetes source (GCST90824163) includes UK Biobank participants and therefore overlaps the UKB-PPP exposure sample by roughly 5-6%. Sample overlap biases two-sample MR towards the observational association, so the control-arm comparison in Figures F-DM-a and F-DM-b rests on estimates that are not fully two-sample. The two zero-overlap sources test whether that matters.

The dashed line in each panel is the number of associations significant in the main analysis. Bars are nested subsets, read left to right: how many of those associations the alternative source even contains (Covered), how many of the covered ones point the same way, and how many of those also reach P < 0.05 and FDR < 0.05 in the alternative source.

Results. Against Crouch et al. (GCST90013791, zero overlap): 39 of 57 covered, 37 of the 39 same direction, 29 nominally significant, 22 at FDR < 0.05. Against MVP (GCST90475667, zero overlap): 27 of 28 covered, 27 of 27 same direction, 24 nominally significant, 22 at FDR < 0.05. Against the retired type 1 diabetes source (GCST90475661): 55 of 57 covered but only 36 same direction, which is consistent with the independent evidence that this source is type 2 diabetes dominated and supports the decision to retire it.

Associations that the alternative source does not contain CANNOT BE JUDGED and are not counted as replication failures. Coverage is lower for Crouch et al. because that dataset is smaller, which is the price of requiring zero sample overlap.

Method: each association is re-estimated in the alternative diabetes GWAS with the same Wald ratio estimator and the same cis sentinel variant, after harmonising to the same effect allele; the nested bars then count direction agreement, nominal p < 0.05 and FDR < 0.05 in that source. Role in the analysis pipeline: a sensitivity analysis supporting Figures F-DM-a and F-DM-b. It is not a step of the pre-specified workflow and removes no candidate.

发表版表格

文件	下载
`diabetes_contrast.csv`	下载
`diabetes_contrast_complication_specific.csv`	下载
`diabetes_contrast_diabetes_driven.csv`	下载
`source_compare_summary.csv`	下载
`source_compare_t1d_retired_source.csv`	下载
`source_compare_t1d_zero_overlap.csv`	下载
`source_compare_t2d_zero_overlap.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/diabetes_contrast/figures/，不随文档站分发。

糖尿病对照臂 ​

图 ​

Figure F-DM-b. How the 57 significant protein-complication associations divide once the effect on diabetes liability is taken into account ​

Figure F-DM-a. Effect on each diabetic complication plotted against the effect of the same protein on diabetes liability, for all 57 associations that passed FDR < 0.05 in a complication ​

Figure F-DM-c. Whether the diabetes control-arm estimates hold up when the diabetes GWAS is swapped for an independent one ​

发表版表格 ​

糖尿病对照臂

图

Figure F-DM-b. How the 57 significant protein-complication associations divide once the effect on diabetes liability is taken into account

Figure F-DM-a. Effect on each diabetic complication plotted against the effect of the same protein on diabetes liability, for all 57 associations that passed FDR < 0.05 in a complication

Figure F-DM-c. Whether the diabetes control-arm estimates hold up when the diabetes GWAS is swapped for an independent one

发表版表格