Fig. 4: Associations of CSF to plasma ratios of peripherally derived proteins with cognitive impairment.

a, Volcano plot showing associations between CSF to plasma ratios of peripherally derived proteins and MMSE scores in GNPC cohort Q (n = 1,280), assessed by linear regression of ratio by MMSE score with age and sex as covariates. The x axis shows MMSE coefficient β as a fraction of the mean CSF to plasma ratio. Although lower MMSE scores indicate more severe cognitive impairment, the plotted effect size was flipped such that positive effect sizes indicate that the ratio increases with cognitive impairment. The y axis shows the Benjamini–Hochberg-corrected q value. b, Bar plot showing the number of CSF to plasma ratios, the up- or downregulation of which was associated with cognitive (cog.) impairment (q < 0.05), by protein source. c, Overlap among peripheral proteins with ratios significantly (q < 0.05) associated with both cognitive impairment and healthy aging. For the association with age, significant associations were required in both the GNPC aging analysis and Knight-ADRC/Stanford aging meta-analysis. N/A, no overlap. d, Box plot showing the association of the fibrinogen CSF to plasma ratio with cognitive impairment (MMSE) in GNPC cohort Q (n = 1,280). Cutoffs for cognitive impairment visualization are: none: MMSE > 25; mild: MMSE 21–25; moderate (Mod): MMSE 11–20; and severe (Sev): MMSE < 11. The P value refers to the significance of the MMSE coefficient from linear regression of the fibrinogen ratio by MMSE score with age and sex as covariates and is two sided. e, Scatterplot showing the association of the fibrinogen CSF to plasma ratio with age in healthy participants in the GNPC cohort (n = 551). The P value refers to the significance of the age coefficient from linear regression of the fibrinogen ratio by age with sex and contributor code as covariates, and is two sided. f,h,j, Box plots showing the associations of CSF to plasma ratios of DCUN1D1 (f), MFGE8 (h) and VEGFA (j) with cognitive impairment in the indicated cohorts. Cognitive impairment was assessed by MMSE score in GNPC cohort Q (n = 1,280), MoCA score in GNPC cohort N (n = 240) and CDR-global score in Stanford (n = 238) and Knight-ADRC (n = 243) cohorts. The P values refer to the significance of the cognitive score coefficient from linear regression of ratio by cognitive score with age and sex as covariates, and are two sided. MoCA cutoffs for cognitive impairment visualization are: none: MoCA > 25; mild: MoCA = 18–25; moderate: MoCA = 10–17; and severe: MoCA < 10. The center line represents the median, the box limits the upper and lower quartiles and the whiskers the largest or smallest value not exceeding 1.5× the IQR from the box limits. g,i,k, Forest plots showing the associations of CSF to plasma ratios of DCUN1D1 (g), MFGE8 (i) and VEGFA (k) with cognitive impairment in the indicated cohorts. Cognitive test scores were z-score normalized before analysis. GNPC cohort Q: n = 1,280; GNPC cohort N: n = 240; Stanford: n = 238; and Knight-ADRC: n = 243. The red squares represent the cognitive test score coefficient β from linear regression as a fraction of the mean CSF to plasma ratio. The bars represent 95% confidence intervals (CIs). The diamonds represent the mean normalized coefficient estimate (center) and 95% CI (edges) from a crosscohort, random-effects meta-analysis, with two-sided P values displayed.