Reference

Glossary

Every technical term used on this site, in one place. Dotted-underlined words across the site open these same definitions in a small card; this page holds the full versions. New to the subject entirely? The reader’s guide builds the concepts up in order.

Bulk tissue: Most expression datasets measure RNA from a homogenized piece of tissue, so every value is an average over neurons, glia, and all other cell types in the sample. An effect confined to one cell type is diluted by the rest, which is one proposed reason risk variants fail to match bulk eQTLs. Single-cell eQTL data address this directly by measuring expression per cell type.See also: Single-cell eQTL · eQTL (expression quantitative trait locus)
Colocalization: When a GWAS signal and a molecular QTL sit in the same genomic window, colocalization asks whether the two signals point at the same underlying causal variant or at two different variants that happen to be neighbors. The coloc method compares five hypotheses, from no association at all (H0) through two independent signals (H3) to one shared causal variant (H4), and reports a posterior probability for each. This site uses PP4, the probability of the shared-variant hypothesis, with a threshold of 0.8. Colocalization is evidence of a shared cause, not proof that the molecular trait transmits the disease effect.See also: PP4 · eQTL (expression quantitative trait locus) · Linkage disequilibrium (LD)
CpG site and DNA methylation: DNA methylation is a chemical mark, a methyl group added to a cytosine base, occurring mainly at CpG dinucleotides. Methylation levels vary between people, are partly controlled by nearby genetic variants, and correlate with the regulatory state of the surrounding DNA, since promoters and enhancers change methylation as they switch on and off. The Brain-mMeta dataset used here measured methylation in roughly 1,160 human brains. A methylation association marks regulatory activity but does not by itself say which gene is affected or in which direction.See also: mQTL (methylation quantitative trait locus) · Regulatory chromatin · Methylation chain
Credible set: Fine-mapping cannot usually name a single causal variant, but it can shrink the candidate list. A 95% credible set is the smallest group of variants whose combined posterior probability reaches 95% for one independent signal; a locus with several independent signals has several credible sets, and 40% of the loci analyzed here carry two or more. On the locus pages, credible-set mass in regulatory chromatin is the fraction of a set's probability that falls on variants inside enhancer or promoter states.See also: Fine-mapping · SuSiE · Regulatory chromatin
eQTL (expression quantitative trait locus): An expression quantitative trait locus is a variant whose alleles are associated with higher or lower expression of a nearby gene, measured across many donors. If a schizophrenia risk variant is also an eQTL for a gene in brain tissue, and colocalization confirms that the two signals share a causal variant, that gene becomes the leading candidate for the locus. This project tested 13 GTEx brain regions plus the much larger MetaBrain cortex meta-analysis.See also: Gene expression · Colocalization · Bulk tissue
Expression-orphan locus: A risk locus is an expression orphan when no expression-based test links it to a gene: no bulk eQTL, no fine-mapped multi-signal pairing, no single-cell or fetal eQTL, and no splicing QTL reaches PP4 above 0.8. These loci are where the methylation chain matters most, since it can nominate a gene by routing through a shared methylation signal; 44 orphan loci received a high-tier nomination this way.See also: Methylation chain · Nomination tier
Fine-mapping: GWAS associations arrive as broad peaks because linkage disequilibrium correlates neighboring variants. Fine-mapping models the association pattern together with LD to estimate, for each variant, the probability that it is the causal one, and to split a region into independent signals when more than one exists. This project fine-mapped every locus with SuSiE, using LD computed from 503 European-ancestry 1000 Genomes samples. The results feed the multi-signal colocalization channel and the credible-set annotations on each locus page.See also: SuSiE · Credible set · Linkage disequilibrium (LD)
Gene expression: Genes act by being transcribed into RNA and, for protein-coding genes, translated into protein. Expression level, the amount of RNA a gene produces, differs between tissues, cell types, developmental stages, and people. Regulatory DNA variants change expression rather than the protein itself, which is why expression data sit at the center of interpreting GWAS loci.See also: eQTL (expression quantitative trait locus) · Non-coding (regulatory) variant
GWAS (genome-wide association study): A genome-wide association study genotypes a large sample of cases and controls and tests each variant, typically millions of them, for a difference in frequency between the two groups. Significance thresholds are strict (p below 5 per hundred million) because so many tests are run. The schizophrenia GWAS used here, PGC3, compared 67,390 cases with 94,015 controls and reported 287 significant regions. A GWAS says where risk-associated variation lies; it does not say which variant is causal or which gene it acts through.See also: PGC3 · Locus · Summary statistics
Index variant: Each locus is labeled by its index variant, also called the lead SNP: the variant with the strongest association in the region. Because of linkage disequilibrium, the index variant is often just the best-measured proxy for a causal variant nearby. Locus identifiers on this site combine the chromosome with the index rsID, and the regional plots color every other variant by its correlation with the lead.See also: Linkage disequilibrium (LD) · Locus · Fine-mapping
Linkage disequilibrium (LD): Variants that sit close together on a chromosome tend to be inherited as a block, so their genotypes are correlated across a population. This correlation is linkage disequilibrium, quantified by r squared. LD is why a single causal variant produces a broad GWAS peak, why colocalization must compare whole signals rather than single variants, and why fine-mapping is needed at all. The LD reference used here is 503 European-ancestry genomes from the 1000 Genomes Project.See also: Index variant · Fine-mapping · Colocalization
Locus: A risk locus is the unit of a GWAS result: a genomic region, often hundreds of kilobases wide, containing an association signal and everything correlated with it. A locus typically spans several genes, and the gene nearest the strongest variant is frequently not the true target, which is exactly why the variant-to-gene problem exists. PGC3 reported 287 schizophrenia loci; the 281 autosomal loci outside the extended MHC region are analyzed here.See also: Index variant · GWAS (genome-wide association study) · Linkage disequilibrium (LD)
Methylation chain: Where no direct expression colocalization exists, a gene can still be nominated transitively. Step one: the GWAS signal colocalizes with a methylation QTL at some CpG site. Step two: the same CpG's methylation QTL colocalizes with the expression of a gene in independent expression data. The chain score is the minimum of the two PP4 values, so a chain is only as strong as its weaker link. Chained nominations are ranked hypotheses for follow-up, not demonstrations of mechanism; the methylation signal may mediate the effect, respond to it, or simply mark the same regulatory element.See also: CpG site and DNA methylation · mQTL (methylation quantitative trait locus) · Nomination tier · Expression-orphan locus
mQTL (methylation quantitative trait locus): A methylation quantitative trait locus is a variant whose alleles track higher or lower methylation at a particular CpG site. Brain mQTLs from the Brain-mMeta meta-analysis colocalized with 193 of the 281 schizophrenia loci, more than twice the count of the bulk expression channel and six times its rate on a per-test basis. Because methylation marks regulatory DNA without naming a target gene, an mQTL colocalization places the risk variant on active chromatin but leaves gene identity to the methylation chain.See also: CpG site and DNA methylation · Methylation chain · Colocalization
Nomination tier: Gene nominations are banded by their chain score, the weaker of the chain's two colocalization links. High tier requires both links at PP4 of 0.8 or more, the same threshold used for direct colocalization everywhere else on this site. The suggestive tier (0.5 to 0.8) is provided for completeness and filtering; treat it as a weaker hypothesis, not as a finding.See also: Methylation chain · PP4
Non-coding (regulatory) variant: The large majority of GWAS associations land outside protein-coding sequence. Such variants cannot change a protein's composition, so the prevailing model holds that they alter regulatory elements, promoters and enhancers, and thereby shift the expression of nearby genes. This regulatory hypothesis is the premise the whole project stress-tests: if it holds, risk loci should colocalize with QTLs for expression or for other regulatory readouts such as methylation.See also: Gene expression · Regulatory chromatin · eQTL (expression quantitative trait locus)
PGC3: PGC3 is shorthand for the third and largest schizophrenia genome-wide association study from the Psychiatric Genomics Consortium, published by Trubetskoy and colleagues in 2022. Its core meta-analysis of 67,390 cases and 94,015 controls identified 287 genome-wide significant loci. This project excludes the extended MHC region, whose long-range LD defeats locus-level analysis, and five chromosome X loci, leaving 281.See also: GWAS (genome-wide association study) · Locus
PP4: The coloc framework weighs five hypotheses for a pair of association signals in one region: no signal (H0), a signal in only one trait (H1, H2), two distinct causal variants (H3), and one shared causal variant (H4). PP4 is the posterior probability of H4. Values above 0.8 are treated as colocalized throughout this site; the central methylation results also hold at 0.9 and 0.95. A low PP4 alongside a high PP3 is informative too: it says both signals exist but favor different variants.See also: Colocalization
Regulatory chromatin: Chromatin, DNA together with its packaging proteins, carries chemical marks that distinguish active regulatory elements from silent DNA. The Roadmap Epigenomics project segmented the genomes of many tissues, including fetal and adult brain, into states such as active promoter and enhancer using the chromHMM model. On the locus pages, credible-set mass in regulatory chromatin is the fraction of fine-mapping probability that falls in those states; a high fetal value beside a low adult value hints that a variant acts during development.See also: Credible set · CpG site and DNA methylation
Sequential attribution: The bookkeeping rule behind the waterfall figure. The six evidence channels are applied in a fixed order (bulk expression, multi-signal rescue, single-cell eQTL, fetal eQTL, splicing, methylation), and each locus is credited to the first channel that produces a colocalization at PP4 above 0.8. A locus explained by both splicing and methylation therefore counts under splicing. The order encodes an evidentiary preference, from the most direct and best-replicated evidence toward the most indirect, and later channels receive only the loci that earlier ones missed.See also: Colocalization · PP4 · Methylation chain
Single-cell eQTL: Single-cell RNA sequencing assigns each measured cell to a type before testing variants against expression, producing per-cell-type eQTLs. This recovers effects confined to one population, such as a variant active only in excitatory neurons, that bulk tissue averages away. The dataset used here covers eight brain cell types from 192 donors; its modest sample size limits power, and it added six loci beyond the bulk channels.See also: Bulk tissue · eQTL (expression quantitative trait locus)
sQTL (splicing quantitative trait locus): Most human genes produce several transcript forms through alternative splicing. A splicing QTL shifts the balance between forms, sometimes without changing the gene's total expression, so an eQTL analysis can miss it entirely. Seventeen splicing datasets were tested here, adding nine loci not explained by any expression channel.See also: eQTL (expression quantitative trait locus) · Gene expression
Summary statistics: Summary statistics are the published per-variant results of an association study: effect size, standard error, p-value, and allele frequencies, with no individual genotypes attached. Modern methods can run colocalization and fine-mapping from these tables alone, given a public LD reference. Every analysis in this project, and every number on this site, derives from publicly released summary statistics; no controlled-access data were used.See also: GWAS (genome-wide association study) · Linkage disequilibrium (LD)
SuSiE: SuSiE models an associated region as a sum of single-variant effects, which lets it separate up to ten independent signals here and produce one credible set per signal. This project ran SuSiE on the GWAS summary statistics with a 1000 Genomes LD reference, then paired each GWAS signal with published per-gene SuSiE results from the eQTL Catalogue. That pairing is the multi-signal rescue channel, the waterfall's second bar; separating signals reclassified about a quarter of single-variant colocalization calls in each direction.See also: Fine-mapping · Credible set
TSS (transcription start site): The transcription start site anchors a gene's promoter. Regulatory contacts tend to weaken with distance, so a chained nomination whose CpG lies within a few kilobases of the gene's TSS is mechanistically easier to believe than one half a megabase away. The nominations table reports this distance and offers a 100 kb filter.See also: Methylation chain · CpG site and DNA methylation
Variant (SNP): A genetic variant is any place the genome differs between individuals; the variants tested in GWAS are almost all single-nucleotide polymorphisms, single-letter differences carried by some fraction of the population. Variants are named by rs identifiers, such as rs4702. The risk variants studied here are common and individually of small effect; none is deterministic for schizophrenia.See also: GWAS (genome-wide association study) · Index variant