Genome-wide Significant Associations for Variants With Minor Allele Frequency of 5% or Less—An Overview: A HuGE Review

Abstract

The authors survey uncommon variants (minor allele frequency, ≤5%) that have reached genome-wide significance (P ≤ 10⁻⁷) in genome-wide association study(ies) (GWAS). They examine the typical effect sizes of these associations; whether they have arisen in multiple GWAS on the same phenotype; and whether they pertain to genetic loci that have other variants discovered through GWAS, perceived biologic plausibility from the candidate gene era, or known mutations associated with related phenotypes. Forty-three associations with minor allele frequency of 5% or less and P ≤ 10⁻⁷ were studied, 12 of which involved nonsynonymous variants. Per-allele odds ratios ranged from 1.03 to 22.11. Thirty-two associations had P ≤ 10⁻⁸. Eight uncommon variants were identified in multiple GWAS. For 14 associations, also other common polymorphisms with genome-wide significance were identified in the same loci. Thirteen associations pertained to genetic loci considered to have biologic plausibility for association in the candidate gene era, and mutations with related phenotypic effects were identified for 11 associations. Twenty-five uncommon variants are common in at least 1 of the 4 different ancestry samples of the International HapMap Project. Although the number of uncommon variants with genome-wide significance is still limited, these data suggest a possible confluence of rare/uncommon and common genetic variation on the same genetic loci.

The large majority of discoveries in human genome epidemiology in the last 5 years pertain to associations of common genetic variants with diverse phenotypes (1, 2). In particular, genome-wide association study(ies) (GWAS) have dramatically increased the yield of associations with very high levels of statistical significance (3–6). GWAS conducted to date have used common genetic markers and have found mostly low penetrance variants with small effects (7, 8). Their genotyping platforms offer very good coverage across the genome for variants with minor allele frequency (MAF) of greater than 5% (8, 9). However, variants with lower MAF are either excluded routinely from commercial platforms or inadequately covered (8, 10). For most diseases, the associations identified to date through GWAS account for only a small portion of the estimated total heritability (11–13). There are many speculations about the reasons underlying the residual unknown component of the genetic architecture—also described as the “genetic dark matter” (13, 14). One explanation is the presence of associations involving uncommon (MAF, ≤5%) and rare (MAF, <0.5%) variants (8, 13). Associations with uncommon/rare variants may even have substantial genetic effects, but they have been difficult to discover to date, presumably because of inadequate coverage in most GWAS, very large sample size requirements, or inefficient analytical methods (8, 15).

Newer genotyping platforms (including exome and full-genome sequencing) (16–18) and analysis methods (15, 19, 20) are already being explored in the pursuit of associations involving uncommon and rare variants. Nevertheless, even traditional GWAS occasionally have discovered associations that pertain to such single- nucleotide polymorphisms (SNPs). Given that over 400 GWAS have been published to date (21, 22), an overview of this literature can already assemble a substantial corpus of associations with uncommon variants. Such an overview could yield some preliminary insights about these associations and their respective genetic loci. The following questions may be asked: What are the typical effect sizes of these associations, and how robustly are they replicated? Do they arise in single or multiple GWAS on the same phenotype? Are common variants also identified in the same loci? Have these genetic loci been considered to have biologic plausibility for association in the candidate gene era? Are any mutations with related phenotypes already known for these same loci? Are uncommon variants common in populations of different ancestry?

Here, we systematically evaluated these questions by perusing all associations for single-nucleotide variants with MAF of 5% or less that have been discovered in GWAS with strong statistical support.

MATERIALS AND METHODS

Search strategy and eligibility criteria

We screened A Catalog of Published Genome-Wide Association Studies (22) hosted by the National Human Genome Research Institute, Office of Population Genetics. The catalog is an online, regularly updated database of SNP–trait associations extracted from published GWAS, which attempt to assay at least 100,000 SNPs. It lists associations with P < 10⁻⁵ (21, 22). We identified all GWAS reporting at least 1 genome-wide significant association (P ≤ 10⁻⁷), regardless of the minor allele frequency of the involved SNP. Because the catalog reports only 1 SNP per gene locus for each association, we also searched all genome-wide significant studies (main articles and supplements) to identify additional associations involving rare/uncommon polymorphisms, regardless of whether they were mentioned in the GWAS Catalog or not. The last search was conducted on December 8, 2009.

Eligible associations for this overview were those involving variants with a risk allele frequency of 5% or less or 95% or greater (i.e., MAF, ≤5%) and that had attained genome-wide significance by using a threshold of P ≤ 10⁻⁷ in at least 1 GWAS when both the discovery and replication data were combined (23). The risk allele frequency criterion pertained to the control group for case-control designs and to the whole population for other designs. We focused on single-nucleotide variants and excluded genetic associations based on haplotypes or structural variants. If the same variant was found in more than 1 GWAS on the same phenotype, we counted this as 1 association but recorded all pertinent GWAS.

Data extraction

For each association, we extracted the following data: first author; publication date; journal; title; disease/phenotype; gene; variant (rs number); chromosome region; race/ethnicity of study populations; discovery and replication sample sizes; effect estimates (odds ratios per copy of risk allele for binary outcomes, standardized mean differences for continuous outcomes); and P value of the effect estimates including all data (discovery and replication).

Data extraction was conducted independently by 2 of the authors, and disagreements were discussed and resolved with a third investigator. Data extraction was performed directly from the respective GWAS articles and their supplements, because we have noted some discrepancies in the information already extracted in the GWAS Catalog and we required increased accuracy and additional information besides what was listed in the Catalog.

Evaluation of the eligible associations

We summarized descriptively the phenotypes involved in the eligible associations, the distribution of the risk allele frequencies, P values, and effect estimates. Whenever the effect estimates were not given and could not be calculated from the published information, we contacted the authors. To express all effect estimates on the same scale, we converted standardized mean differences to odds ratio equivalents multiplying the respective standardized mean difference by 1.81 to obtain the natural logarithm of the odds ratio (24). This method transforms a standardized mean difference of a quantitative trait into an odds ratio for the dichotomized version of that trait and uses a normality assumption for the effects.

Additionally, we estimated the average sample size of the eligible GWAS. For this typical sample size, we performed calculations to estimate the power to detect associations with various MAFs and odds ratio values at α = 1 × 10⁻⁷ under a multiplicative (log-additive) genetic model and under the optimal scenario where there is no loss of power due to multistage process in SNP selection. We used the QUANTO software (25). We categorized associations according to quartiles of odds ratio and according to MAF 1–2%, 3%–4%, and 5%. For each of the resulting 12 categories, we estimated the power G of a typical GWAS (average sample size of the analyzed GWAS) to detect an association of that odds ratio and MAF at the GWAS level. We used the median value of odds ratio and the midvalue of MAF in each category for these calculations. For each category of odds ratio and MAF values, one can calculate the total number of variants (those that have been discovered plus those that have not been discovered because of limited power), by multiplying the number of discovered variants by 1/G.

Using WGAViewer (26, 27), the University of California, Santa Cruz, Human Genome Browser (http://genome.ucsc.edu/cgi-bin/hgGateway) (28), the Single-Nucleotide Polymorphism Database (dbSNP) Build 130 (http://www.ncbi.nlm.nih.gov/projects/SNP/), and the Ensembl (www.ensembl.org) Database, we identified the functional position of the eligible uncommon/rare variants within the respective genes, that is, whether they are located in exons, introns, or promoter regions and whether they cause nonsynonymous changes or frameshift changes.

For each eligible genotype–phenotype association, we identified also all other GWAS listed in the GWAS catalog (22) that had evaluated the same phenotype. We examined if the eligible uncommon/rare variant had been reported by any other GWAS on the same phenotype, regardless of whether it had reached genome-wide significance or not. Moreover, we evaluated whether any other GWAS on the same phenotype reported on associations with any other variants in the same gene locus as the eligible uncommon variant. We use the term “locus” here to denote either a single gene or several genes, if the authors of the GWAS could not pinpoint which gene among the several listed was most likely to harbor the functional causative variants (e.g., when genes overlapped or when associated SNPs were located in an area lying between 2 genes). Whenever such other variants were reported, we recorded their effect estimates and P values. Then, we examined whether the uncommon variants were in high linkage disequilibrium (r ² ≥ 0.8) with the other variants in the same gene locus, using the Web-based tool, SNP Annotation and Proxy Search (SNAP), version 2.1 (29), selecting data based on the International HapMap Project, Phase 3, Release 2, for the HapMap panel with similar ancestry as the population where the uncommon variant was discovered. Upon unavailability of results, we used HapMap, Release 22.

Furthermore, we searched on Human Genome Epidemiology (HuGE) Navigator, a continuously updated database in human genome epidemiology (2), whether the gene loci containing the eligible uncommon/rare variants had been investigated by candidate-gene association studies conducted prior to the discovery of these loci in a GWAS. We recorded the number of studies on gene–phenotype associations involving the same gene locus and phenotype published until the end of the year before the first GWAS proposing the association with the uncommon variant gene locus, as well as the total number of studies published to date. We also recorded any comments made in the eligible GWAS that had identified the uncommon variant regarding prior evidence on the proposed gene locus, for example, if it had been proposed by previous linkage or candidate-gene studies or GWAS. Additionally, for each gene locus, we recorded whether any Mendelian mutations have been previously reported in association with the same, similar/related, or unrelated phenotype(s), using the Online Mendelian Inheritance in Man (OMIM) Database (http://www.ncbi.nlm.nih.gov/omim/).

Finally, we recorded the minor allele frequencies of the eligible uncommon variants in the populations genotyped in the International HapMap Project (30, 31), using data from HapMap, Phases 1 and 2 (31, 32), on people of European, African, and Asian (Chinese and Japanese) ancestry. We then examined whether the eligible SNPs had estimated MAFs of 5% or less in all of these populations or only in some of them.

RESULTS

Description of the eligible associations

We screened 440 GWAS with a total of 2,497 entries in the GWAS catalog. Of those, 74 GWAS were excluded because they reported no genome-wide significant (P ≤ 10⁻⁷) SNP–disease association. Of the remaining 366 studies listed in the catalog, we identified 91 entries with associations that had MAFs of 5% or less. We excluded 61 entries because they were not significant at the P ≤ 10⁻⁷ level, 4 because they had MAFs of greater than 5% upon scrutinizing the respective article, and another 3 because the respective associations were based on haplotypes. Of the remaining 23 associations, 1 (rs16901979 and prostate cancer) had been identified by 2 different GWAS and, thus, we regarded as eligible the one published earlier (33) and the subsequent study (34) as a replication. Thus, 22 different associations discovered in 18 different GWAS were eligible through the catalog search.

The main articles and the supplements of those 366 GWAS reporting at least 1 association significant at the P ≤ 10⁻⁷ level were further scrutinized for uncommon/rare variants with genome-wide significance. Hence, we identified 23 additional SNP–disease associations with genome-wide significance (P ≤ 10⁻⁷) implicating uncommon/rare SNPs (MAF, ≤5%), of which 1 (rs1800562 and mean corpuscular volume) had been reported by 2 GWAS published at the same time; thus, we regarded as eligible one of them (35), and the other study (36) was recorded as concurrent. Hence, a total of 44 associations were identified by combing the catalog-based and the full text-based searches. Of those associations, 1 (rs2066847 and Crohn's disease) had been discovered by 2 different GWAS, of which the 1 published earlier (37) was included in our analysis and the subsequent was recorded as a replication (38). Finally, 43 different genome-wide significant associations implicating 40 uncommon/rare SNPs discovered in 28 GWAS (33, 35–61) were eligible (Table 1). One uncommon SNP was implicated in 2 different phenotypes and another in 3 different phenotypes. Among these 40 SNPs, the authors of the respective GWAS implicated a single gene for 31 cases; for 4 SNPs, they implicated more than 1 gene; for 1 SNP, they implicated a single gene in 1 GWAS and more genes in another; and 4 SNPs were not allocated to any specific gene. Overall, 30 different locus–phenotype pairs were implicated (some had been implicated for ≥1 SNP).

Table 1.

Eligible Associations With a Minor Allele Frequency of 5% or Less and P ≤ 10^-7

Disease/Trait	Region	Reported Gene(s)	Variant-Risk Allele	Position/Function	Risk Allele Frequency	Population Descent	P Value	Odds Ratio	95% Confidence Interval	Reference
ALL	12q24.22	KRTHB5	rs2089222-A	Intronic	0.03	European	8.0 × 10-8	2.26	1.60, 3.00	39
AIDS progression	6p21.33	HCP5, MICB, MCCD1, BAT1, LTB, TNF	rs2395029-G	Nonsynonymous coding (missense)	0.03	European	3.0 × 10-19	3.47	2.39, 5.04	40
	6p21.3	C6orf48	rs9368699-C	5′-UTR	0.03	European	2.0 × 10-11	NR	NR
Blue vs. green eyes	15q13.1	OCA2	rs1667394-A	Intronic	0.97	European	2.0 × 10-53	3.67	2.67, 5.05	41
Freckles	16q24.3	MC1R	rs1805007-T	Nonsynonymous coding (missense)	0.05	European	1.0 × 10-96	3.33	2.92, 3.80
BMD (lumbar spine)	13q14	AKAP11	rs180851-C (I)	Intergenica	0.95	European	2.0 × 10-12	1.46b	1.31, 1.63	42
	13q14	AKAP11	rs7326472-A	Intergenica	0.95	European	1.0 × 10-10	1.39b	1.24, 1.54
	13q14	AKAP11	rs12854504-T (I)	Intergenica	0.95	European	1.0 × 10-10	1.39b	1.24, 1.54
	13q14	AKAP11	rs7998154-T (I)	Intergenica	0.02	European	2.0 × 10-8	1.75b	1.46, 2.10
	13q14	TNFSF11	rs6561055-G (I)	Intergenica	0.95	European	3.0 × 10-10	1.39b	1.24, 1.54
	13q14	TNFSF11	rs17639156-T (I)	Intergenica	0.95	European	5.0 × 10-10	1.39b	1.24, 1.54
Cognitive performance	Xp22.2	HCCS	rs5934953-C	Intronic	0.02	European	1.0 × 10-7	NR	NR	43
Crohn's disease	16q12.1	NOD2	rs2066844-T	Nonsynonymous coding (missense)	0.05	European	1.0 × 10-18	2.48	1.98, 3.10	37
	16q12.1	NOD2	rs2066845-C	Nonsynonymous coding (missense)	0.01	European	8.0 × 10-10	3.04	2.09, 4.42
	16q12.1	NOD2	rs2066847-C	Frameshift coding	0.04	European	3.0 × 10-49	4.30	3.42, 5.42
	12q12	LRRK2, MUC19	rs11175593-T (I)	Intronic	0.02	European	3.0 × 10-10	1.54	1.34, 1.76	38
HDL cholesterol	20q13.12	HNF4A	rs1800961-C (I)	Nonsynonymous coding (missense)	0.97	European	8.0 × 10-10	1.41b	1.27, 1.57	44
	9q31.1	ABCA1	rs9282541-T	Nonsynonymous coding (missense)	0.03	European, Mexicans, Asian Indians	5.0 × 10-8	1.33b c	1.21, 1.45	45
Hematocrit	6p22.1	HFE	rs1800562-A (I)	Nonsynonymous coding (missense)	0.04d	European	2.0 × 10-9	1.74	1.45, 2.09	36
Hemoglobin	6p22.1	HFE	rs1800562-A (I)	Nonsynonymous coding (missense)	0.04d	European	6.0 × 10-19	1.33	1.25, 1.42
LDL cholesterol	1p32.3	PCSK9	rs11591147-G	Nonsynonymous coding (missense)	0.99	European	2.0 × 10-44	2.34b	2.07, 2.64	46
MCH	6p22.2	SLC17A3	rs1408272-G (I)	Unknown	0.03d	European	4.0 × 10-39	1.03	1.02, 1.04	36
MCV	6p22.1	HFE	rs1800562-A (I)	Nonsynonymous coding (missense)	0.04d	European	1.0 × 10-23	12.83	7.73, 20.92	35
NCP	3p22.2	ITGA9	rs189897-A	Intronic	0.03	Asian	7.0 × 10-8	3.18	1.94, 5.21	48
	3p22.2	ITGA9	rs197757-T	Intronic	0.03	Asian	1.0 × 10-7	3.09	1.89, 5.05
NSCL	18q22.3	Intergenic	rs17085106-T	Intergenic	0.02	European	4.0 × 10-8	4.07	2.37, 7.00	47
Panic disorder	12p13.31	TMEM16B	rs12579350-A	Intronic	0.01	Asian	4.0 × 10-9	22.11	5.30, 92.14	49
	1q32.1	PKP1	rs860554-T	Intronic	0.05	Asian	5.0 × 10-8	4.03	2.40, 6.76
Prostate cancer	8q24.21	Intergenic	rs16901979-A	Intergenic	0.03	European	1.1 × 10-12	1.79	1.53, 2.11	33
Primary biliary cirrhosis	6p21.3	C6orf10	rs2395148-A	Intronic	0.02	European	4.0 × 10-14	2.87	2.16, 3.82	50
ALP	12q12	PDZRN4, CNTN1	rs1880887-C	Intronic	0.03	European	1.0 × 10-10	NR	NR	51
fT3	17p12	HS3ST3B1	rs3848445-C	Unknown	0.05	European	8.4 × 10-9	NR	NR
Psoriasis	6p21.33	HLA-C	rs2395029-C	Nonsynonymous coding (missense)	0.03	European	2.1 × 10-26	4.10	3.10, 5.30	52
Response to treatment for ALL	10p12.33	ST8SIA6	rs359312-T	Intronic	0.04	European, African, other	9.0 × 10-8	3.91	1.52, 10.10	53
Response to antipsychotic therapy	2p12	Intergenic	rs17022444-G	Intergenic	0.03	European, African, other	1.0 × 10-10	NR	NR	54
	4q24	Intergenic	rs7669317-C	Intergenic	0.04	European, African, other	8.0 × 10-8	NR	NR
SLE	6q23.3	TNFAIP3	rs5029939-G	Intronic	0.03	European	3.0 × 10-12	2.28	1.80, 2.88	55
	6q23.3	TNFAIP3	rs2230926-C	Nonsynonymous coding (missense)	0.04	Asian	1.0 × 10-17	1.72	1.52, 1.94	56
Tanning	5p13.3	MATP	rs35391-C	Intronic	0.97	European	3.0 × 10-10	2.22	1.72, 2.86	57
Triglycerides	11q23.3	APOA1, APOC3, APOA4, APOA5	rs662799-G (I)	Upstream	0.05	European	2.0 × 10-15	1.31b,c	1.22, 1.40	58
	11q23.3	APOA1, APOC3, APOA4, APOA5, DSCAML1	rs10892151-A	Intronic	0.03	European	3.0 × 10-29	NR	NR	59
Type 1 diabetes	7p12.1	COBL	rs4948088-C (I)	Unknown	0.95	European	4.0 × 10-8	1.30	1.11, 1.49	60
Type 2 diabetes	10q25.2	TCF7L2	rs7903146-T	Intronic	0.04	Asian	8.0 × 10-12	1.54	1.36, 1.74	61

Abbreviations: AIDS, acquired immunodeficiency syndrome; ALL, acute lymphoblastic leukemia; ALP, alkaline phosphatase; BMD, bone mineral density; fT3, free triiodothyronine; GWAS, genome-wide association study(ies); HDL, high density lipoprotein; I, risk variants imputed rather than directly genotyped; LDL, low density lipoprotein; MAF, minor allele frequency; MCH, mean corpuscular hemoglobin; MCV, mean corpuscular volume; NCP, nasopharyngeal carcinoma; NHANES, National Health and Nutrition Examination Survey; NR, not reported and data not adequate for computing the missing values; NSCL, nonsyndromic cleft lip with or without cleft palate; SLE, systemic lupus erythematosus; SNP, single-nucleotide polymorphism; 5′-UTR, 5′-untranslated region.

^aFor these SNPs, AKAP11 and TNFSF11 were reported as the closest genes in the GWAS, but WGA Viewer and the Ensembl characterized them as “intergenic.”

^bOdds ratio equivalent was calculated from the standardized mean difference.

^cOdds ratio equivalent was computed from the mean difference using also the population standard deviation from NHANES data on HDL cholesterol and triglyceride levels, because the population standard deviation was not given in the GWAS.

^dMAFs reported in the original GWAS were based on the International HapMap Project frequencies.

The phenotypes for these 43 associations were acquired immunodeficiency syndrome (AIDS) progression (n = 2 associations), bone mineral density (n = 6 associations in 2 loci), Crohn's disease (n = 4), high density lipoprotein (HDL) cholesterol (n = 2), nasopharyngeal carcinoma (n = 2 associations in the same locus), panic disorder (n = 2), response to antipsychotic therapy (n = 2), systemic lupus erythematosus (n = 2 associations in the same locus), triglyceride levels (n = 2 associations in the same locus), acute lymphoblastic leukemia in children, eye color, cognitive performance, freckles, low density lipoprotein (LDL) cholesterol, hematocrit levels, hemoglobin levels, mean corpuscular hemoglobin, mean corpuscular volume, nonsyndromic cleft lip with or without cleft palate, prostate cancer, alkaline phosphatase, free triiodothyronine, primary billiary cirrhosis, psoriasis, response to treatment for childhood acute lymphoblastic leukemia, tanning, type 1 diabetes, and type 2 diabetes.

Location and function of gene variants

Nine of the 40 uncommon variants (22.5%) constituted nonsynonymous coding SNPs, whereas 15 (37.5%) were intronic, 10 (25%) were intergenic (although 6 of them were related to specific genes by the authors of the GWAS), 1 was located in the 5′-untranslated region (5′-UTR), 1 was found upstream of the respective gene, 1 interfered with the function of the frameshift, and for 3 SNPs the function/location was unknown.

Frequency and effect sizes

All 40 variants had MAFs that would characterize them as uncommon rather than rare. Of the 22 associations pertaining to diseases rather than quantitative traits or nondisease- related phenotypes, 21 had risk variants with a risk allele frequency of 5% or less, and only 1 association had a risk allele frequency of 95%. The latter was actually the association with the smallest odds ratio estimate. Thirty-three associations had been discovered and replicated exclusively in populations of European ancestry, whereas 10 were discovered and/or replicated in non-European or mixed populations.

Eleven of the 43 associations had P values between 10⁻⁷ and 10⁻⁸, and 32 had greater statistical significance. Odds ratios were extracted, obtained from the authors, or calculated in 36 associations (no data were retrievable for 7 associations). Per-allele odds ratios ranged from 1.03 (for rs1408272 contributing to mean corpuscular hemoglobin levels) to 22.11 (for rs12579350 in panic disorder). The median was 2.24 (interquartile range, 1.40–3.40).

Power calculations and observed and expected distributions of uncommon variants

The average sample size utilized in the 28 identified GWAS was 7,637 individuals for case-control studies and 10,647 individuals for all studies (case-control and cohort). Table 2 shows the number of discovered associations implicating uncommon variants split according to odds ratio quartiles and according to MAFs = 1%–2%, 3%–4%, and 5% categories. As shown, no variant with an odds ratio of less than 1.40 and a MAF = 1%–2% is included, because the power to detect such variants with the typical sample size used in these GWAS in minimal (0.37%). Power calculations suggest that only 11% and 23% of the variants with similar odds ratio and a MAF = 3%–4% or 5%, respectively, would have been discovered with the average sample size of the GWAS that we considered. Variants with an odds ratio = 1.40–2.24 and a MAF = 1%–2% had a 56% chance to be discovered. In all other categories of odds ratio and MAF combinations, the power is greater than 99%. This means that, with a sample size of 10,647, it should be possible to discover almost all variants with an odds ratio greater than 1.40 and MAF = 3%–5% and those with an odds ratio greater than 3 and a MAF greater than 1%. Consideration of the power calculations suggests that the number of variants with an odds ratio less than 1.40 and a MAF = 3%–5% may be 3-fold larger than that with an odds ratio greater than 1.40 and a similar MAF, but the latter variants are far easier to discover with the typical sample size used in these GWAS.

Table 2.

Number of Discovered^a and Expected Associations Implicating Uncommon Variants Split According to Odds Ratio Quartiles and According to Minor Allele Frequency Categories

	Odds Ratio Range in Quartiles
	<1.40 (Median, 1.33)		1.40–<2.24 (Median, 1.72)		2.24–3.40 (Median, 2.87)		>3.40 (Median, 4.07)
Minor Allele Frequency, %	Discovered Associations, no.	Expected Associations, no.	Discovered Associations, no.	Expected Associations, no.	Discovered Associations, no.	Expected Associations, no.	Discovered Associations, no.	Expected Associations, no.
1–2	0	—b	2	4	3	3	2	2
3–4	3	28	6	6	4	4	6	6
5	6	26	1	1	2	2	1	1

^aThe total number of observed associations in Table 2 is 36 and not 43 as expected from Table 1, because for 7 associations the effect estimates were not retrievable.

^bNot possible to calculate.

Variants in the same loci in other GWAS

For 37 of the 43 associations, we identified at least 1 other GWAS on the same phenotype (Web Table 1). (This information is described in a supplementary table posted on the Journal’s website (http://aje.oxfordjournals.org/).) No other GWAS was found for 6 associations (freckles, panic disorder (n = 2 associations), primary biliary cirrhosis, free triiodothyronine, and response to treatment for acute lymphoblastic leukemia).

For 15 associations, additional GWAS had presented data on the same uncommon SNP (n = 16) (34–36, 38, 44, 62–71) and/or other SNPs in the same locus (n = 74 associations) (44–46, 58, 62–65, 67, 68, 72–88) (Table 3). For 1 association (prostate cancer and rs16901979), no other polymorphisms except the same uncommon variant were identified; hence, for 14 uncommon variant–phenotype associations (corresponding to 10 gene locus–phenotype associations), other GWAS discovered 1 or more common SNPs at the same locus with the uncommon variant. For 4 associations (eye color, LDL cholesterol, triglycerides, type 2 diabetes), the same additional GWAS had presented data on both the same uncommon/rare SNP and 1 or more other SNPs (44, 62–65, 67, 68).

Table 3.

Variants in the Same Gene Loci as the Uncommon Variants, Described in Other Genome-wide Association Study(ies) on the Same Phenotype

Disease/Trait	Uncommon Variant(s)	Gene Locus	Other GWAS That Found Variants in Same Locus, reference	Timing of Other GWAS	Variant	Risk Allele Frequency	P Value	Per-Allele Odds Ratio
Blue vs. green eyes	rs1667394-A	OCA2	62	Subsequent	(Same)	0.13a	3 × 10-87	1.82
					rs7495174	0.05a	0.018	1.60
			63	Subsequent	(Same)b	0.15	8.50 × 10-31	NRc
					rs11855019	0.19	8.60 × 10-25	NR
					rs6497268	0.18	3.70 × 10-19	NR
					rs7495174	0.10	2.00 × 10-22	NR
BMD (lumbar spine)	rs6561055-A, rs17639156-G	TNFSF11	72	Previous	rs9533093	0.80	5.40 × 10-11	1.22
					rs9594738	0.42a	4.00 × 10-23	1.34
					rs9594759	0.62	1.50 × 10-17	1.24
			73	Previous	rs9594759	0.49a	NR	NR
					rs9594738	0.42a	NR	NR
			74	Previous	rs9594759	0.63	1.10 × 10-16	1.27
					rs10507507	0.82	1.60 × 10-5	1.26
					rs7992970	0.78	8.50 × 10-7	1.27
					rs9594738	0.56	2.00 × 10-21	1.36
Crohn's disease	rs2066844-T, rs2066845-C, rs2066847-C	NOD2	75	Previous	rs2076756	0.35a	5.10 × 10-10	NR
					rs2066843	0.36a	2.90 × 10-9	NR
			76	Previous	rs2076756	0.24	7.00 × 10-14	NR
			77	Previous	rs5743289	0.17	3.80 × 10-10	1.45
					rs17221417	0.29	9.40 × 10-12	1.29
			38	Subsequent	(Same: rs2066847)	0.02	3.00 × 10-24	3.99
			78	Subsequent	rs2076756	0.26	9.70 × 10-8	1.33
			79	Subsequent	rs2076756	0.35a	1.00 × 10-9	NR
HDL cholesterol	rs9282541-T	ABCA1	46	Concurrent	rs3890182	0.87	3.00 × 10-10	1.19d
			58	Concurrent	rs4149268	0.35	1.20 × 10-10	1.10d,e
					rs4149274	0.69	7.40 × 10-8	1.20d,e
			80	Concurrent	rs3890182	0.12	2.00 × 10-6	5.58d,e
			81	Subsequent	rs3905000	0.86	8.60 × 10-13	1.24d
					rs3847303	0.88	3.40 × 10-12	1.25d
			44	Subsequent	rs1883025	0.26	1.00 × 10-9	1.16d
			64	Subsequent	rs4149268	0.27a	0.69	1.20d
			82	Subsequent	rs2740491	0.36	3.10 × 10-4	1.06d
					rs3847303	0.13	3.20 × 10-3	1.06d
Hemoglobin	rs1800562-A	HFE	35	Concurrent	(Same)	0.04	1.60 × 10-4	1.25d
			83	Concurrent	rs198833	0.08a	1.40 × 10-8	NR
					rs129128	0.08a	3.30 × 10-8	NR
					rs198851	0.08a	3.40 × 10-8	NR
					rs1799945	0.14a	4.30 × 10-8	NR
					rs198846	0.11a	8.70 × 10-6	1.23d
LDL cholesterol	rs11591147-G	PCSK9	44	Subsequent	(Same)	0.02	9.00 × 10-6	2.65d
					rs11206510	0.19	4.00 × 10-8	1.17d
			64	Subsequent	(Same)	0.02	1.60 × 10-7	1.88d
			58	Concurrent	rs11206510	0.81	3.50 × 10-11	1.15d,e
			82	Subsequent	rs11206510	0.01	2.00 × 10-12	1.33d
MCV	rs1800562-A	HFE	36	Concurrent	(Same)	0.04	1.00 × 10-46	1.02d
			83	Concurrent	rs198846	0.11	8.60 × 10-13	4.91d
Prostate cancer	rs16901979-A	Intergenic	34	Subsequent	(Same)	0.04	2.50 × 10-14	1.80
Psoriasis	rs2395029-C	HLA-C	84	Previous	rs3134792	0.15a	1.00 × 10-9	NR
			85	Subsequent	rs12191877	0.15	<1.00 × 10-100	2.64
Triglycerides	rs662799-G	APOA1, APOC3, APOA4, APOA5	86	Previous	rs481843	0.11	3.30 × 10-5	NR
			46	Concurrent	rs28927680	0.07	2.00 × 10-17	1.60d
			45	Concurrent	rs2075292	0.16	5.30 × 10-8	1.10d,e
					rs7124741	0.17	8.60 × 10-7	1.10d,e
					rs17120139	0.17	2.30 × 10-6	1.09d,e
			80	Concurrent	rs6589566	0.06	3.00 × 10-11	10.90d,e
			64	Subsequent	(Same)	0.06	2.90 × 10-15	1.60d
					rs3135506	0.06	5.50 × 10-12	1.57d
			81	Subsequent	rs12272004	0.93	5.40 × 10-13	1.39d
					rs480878	0.86	8.00 × 10-9	1.19d
					rs28927680	0.93	3.90 × 10-9	1.64d
					rs12292921	0.07	9.06 × 10-13	1.39d
					rs35120633	0.93	2.30 × 10-10	1.75d
					rs3135506	0.06	7.40 × 10-10	1.74d
					rs2075292	0.13	5.70 × 10-12	1.23d
					rs588918	0.87	4.90 × 10-8	1.19d
					rs1351452	0.86	7.40 × 10-10	1.23d
			44	Subsequent	rs964184	0.14	4.00 × 10-62	1.72d
			82	Subsequent	rs12292921	0.06	1.40 × 10-3	1.20d
Triglycerides	rs10892151-A	APOA1, APOC3, APOA4, APOA5, DSCAML1	44	Previous	rs964184	0.14	4.00 × 10-62	1.72d
			46	Previous	rs28927680	0.07	2.00 × 10-17	1.60d
			64	Previous	rs3135506	0.06	5.50 × 10-12	1.57d
					rs662799	0.06	3.00 × 10-15	1.60d
			58	Previous	rs12286037	0.94	1.00 × 10-26	1.51d,e
			45	Previous	rs2075292	0.16	5.30 × 10-8	1.10d,e
					rs7124741	0.17	8.60 × 10-7	1.10d,e
					rs17120139	0.17	2.30 × 10-6	1.09d,e
			86	Previous	rs481843	0.11	3.30 × 10-5	NR
			81	Previous	rs12272004	0.93	5.40 × 10-13	1.39d
					rs480878	0.86	8.00 × 10-9	1.19d
					rs28927680	0.93	3.90 × 10-9	1.64d
					rs12292921	0.07	9.00 × 10-13	1.39d
					rs35120633	0.93	2.30 × 10-10	1.75d
					rs3135506	0.06	7.40 × 10-10	1.74d
					rs2075292	0.13	5.70 × 10-12	1.23d
					rs588918	0.87	4.90 × 10-8	1.19d
					rs1351452	0.86	7.40 × 10-10	1.23d
			80	Previous	rs6589566	0.06	3.00 × 10-11	10.90d,e
			82	Concurrent	rs12292921	0.06	1.40 × 10-3	1.20d
Type 2 diabetes	rs7903146-T	TCF7L2	65	Previous	(Same)	0.25a	4.20 × 10-15	1.43
					rs7901695	0.28a	8.30 × 10-13	1.37
			66	Previous	(Same)	0.25a	3.00 × 10-23	1.37
			67	Previous	(Same)	0.25a	5.50 × 10-8	1.71
					rs7901695	0.28a	3.40 × 10-7	1.66
					rs12255372	0.22a	5.30 × 10-7	1.64
			69	Previous	(Same)	0.18	1.00 × 10-48	1.37
			71	Previous	(Same)	0.29	1.50 × 10-34	1.65
			68	Previous	(Same)	0.49	0.005f	1.28f
					rs7100927	0.49	0.007f	1.56f
			70	Subsequent	(Same)	0.27	1.20 × 10-30	1.48
			77	Previous	rs4506565	0.32	5.70 × 10-13	NR
			87	Previous	rs7901695	0.28a	1.00 × 10-48	1.37
			88	Previous	rs7100927	0.40a	0.007f	1.56f
					rs10509966	0.25a	0.64f	1.09f
					rs10509969	0.20a	0.60f	1.12f
					rs290483	0.42a	0.93f	1.01f
					rs7917983	0.39a	0.17f	1.22f
					rs10509970	0.23a	0.51f	1.14f
					rs10509967	0.26a	0.82f	1.04f

Abbreviations: BMD, bone mineral density; GWAS, genome-wide association study(ies); HDL, high density lipoprotein; LDL, low density lipoprotein; MCV, mean corpuscular volume; NHANES, National Health and Nutrition Examination Survey; NR, not reported and data not adequate for computing the missing values.

^aRisk allele frequency was retrieved from the International HapMap Project phases 1 + 2 data on the same ancestry populations as the eligible variants, because it was not reported in the GWAS.

^brs1667394 was reported to be located in the HERC2 gene locus.

^cThe odds ratio was not retrievable based on the data given, but based on the P value and sample size of the eligible GWAS and of the GWAS reporting the same uncommon variant, we compared the 2 effect estimates and determined that the missing odds ratio is probably smaller than that of the eligible uncommon variant.

^dThe odds ratio equivalent was computed from the standardized mean difference.

^eThe odds ratio equivalent was computed from the mean difference by using also the population standard deviation from NHANES data on LDL cholesterol, HDL cholesterol, and triglyceride levels, because the population standard deviation was not given in the GWAS.

^f P values and hazard ratios from Cox survival analysis.

Whenever the same uncommon SNPs were identified by additional GWAS (8 uncommon SNPs in 16 additional GWAS), the odds ratio estimates were larger than those proposed by the first study with genome-wide significance in 6 cases and smaller in 10 cases. Twelve of the 16 estimates were genome-wide significant. All 16 were nominally significant (P < 0.05).

When other GWAS had presented other SNPs in the same locus, almost all (72/74) of the additional SNPs were common (MAF, >5%). The odds ratio per risk allele was smaller than the effect size of the index uncommon variant with 14 exceptions. Fifty of these 74 additional associations had reached levels of genome-wide significance, and 65 were nominally significant (P < 0.05), whereas for 2 associations the exact P value was not reported.

Evaluation of these variants in SNAP showed that the 2 uncommon variants in TNFAIP3 that were associated with systemic lupus erythematosus in 2 different GWAS were in high linkage disequilibrium (r ² = 1 and D′ = 1 in both Europeans and Asians). Furthermore, the bone mineral density-associated uncommon SNP rs180851 was in high linkage disequilibrium with the uncommon SNPs rs7326472 and rs12854504 (r ² = 0.82 and D′ = 1 for pairwise comparison), which were discovered in the same GWAS. Also in the same GWAS, the uncommon SNP-pair rs7326472 and rs12854504, as well as the SNP-pair rs6561055 and rs17639156, were in high linkage disequilibrium (r ² = 1 and D′ = 1). Moreover, the type 2 diabetes susceptibility uncommon variant rs7903146 located in TCF7L2 was in linkage disequilibrium with rs7901695 (r ² = 1 and D′ = 1) and rs4506565 (r ² = 1 and D′ = 1), which have been highlighted by 3 and 1 previous GWAS, respectively. Both rs7901695 and rs4506565 are common in the European populations used in these GWAS but not in Japanese populations where rs7903146 reached genome-wide significance. None of the other SNPs in the same genetic loci as the uncommon variants had high linkage disequilibrium with them based on the r ². Besides these associations that had both D′ = 1 and r ² = 1, another 41 pairs of uncommon-other SNPs had D′ = 1 but not r ² = 1.

Prior literature

In HuGE Navigator, we identified 2 prior studies for the association between AIDS progression and the HCP-TNF gene locus; 3 studies for the association between TNFSF11 and bone mineral density; 176 studies for the association between Crohn's disease and NOD2; 1 study for the association between Crohn's disease and MUC19; 3 studies for HDL cholesterol levels and HNF4A; 34 studies for the association between ABCA1 and HDL cholesterol; 3 studies for the association between HFE and hematocrit; 17 studies for the association between HFE and hemoglobin; 10 studies for LDL cholesterol levels and PCSK9; 3 studies for the association between HFE and mean corpuscular volume; 57 studies for psoriasis and HLA-C; 118 studies for triglyceride levels and any gene in the APOA1-APOC3-APOA4-APOA5 complex, and 14 studies for type 2 diabetes and TCF7L2. Results are summarized in Table 4 along with the total number of studies on each locus published to date.

Table 4.

Number of Candidate–Gene Association Studies on Each Gene Locus–Disease Association (per HuGE Navigator) and Comments Regarding Previous Knowledge on the Loci Containing the Uncommon Variants as They Appear in the Eligible GWAS

Disease/Trait	Reported Gene(s)	Uncommon Variant(s)	No. of Studies Until December of the Year Before the First GWAS Proposal	Total No. of Studies on Gene–Phenotype to Date	Comments on Gene Locus in Text
ALL	KRTHB5	rs2089222-A	0	0	No comment
AIDS progression	HCP5, MICB, MCCD1, BAT1, LTB, TNF	rs2395029-G	1 for HCP, 1 for TNF, 0 for the rest	1	HCP5 was previously identified by the GWAS-based Euro-CHAVI cohort and also proposed by candidate–gene association studies
	C6orf48	rs9368699-C	0	0	No comment
Blue vs. green eyes	OCA2	rs1667394-A	0	1	Previously reported to be associated with albinism, eye color, hair color, and skin pigmentation; OCA2 mutations are known to be a major cause for albinism; OCA2 has been discovered in linkage studies.
Freckles	MC1R	rs1805007-T	0	3	It was known by previous reports; previously documented mutations in MC1R
BMD (lumbar spine)	AKAP11	rs180851-G, rs7326472-G, rs12854504-G, rs7998154-T	0	0	No comment
	TNFSF11	rs6561055-A, rs17639156-G	3	11	No comment
Cognitive performance	HCCS	rs5934953-C	0	0	No comment
Crohn's disease	NOD2	rs2066844-T, rs2066845-C, rs2066847-C	176	301	NOD2 is a previously known Crohn's disease locus.
	LRRK2, MUC19	rs11175593-T	1 for MUC19, 0 for LRRK2	1 for MUC19, 0 for LRRK2	LRRK2: evidence from a previous cell study; MUC19: evidence from a previous animal study
HDL cholesterol	HNF4A	rs1800961-C	3	5	Function in humans has previously been studied; although mice lacking either Hnf4a or Hnf1a have altered plasma cholesterol levels, there has been only modest evidence to date connecting these genes to either HDL or LDL cholesterol concentrations in humans.
	ABCA1	rs9282541-T	34	53	It is a well-recognized association.
Hematocrit	HFE	rs1800562-A	3	4	Mutations in the HFE gene are already known to underlie hereditary hemochromatosis. The HFE gene induces expression of the iron-regulatory hormone hepcidin.
Hemoglobin	HFE	rs1800562-A	17	21	Mutations in the HFE gene are already known to underlie hereditary hemochromatosis. The HFE gene induces expression of the iron-regulatory hormone hepcidin.
LDL cholesterol	PCSK9	rs11591147-G	10	26	Prior evidence for association with LDL cholesterol concentrations; has also been shown to cause Mendelian syndromes or to harbor multiple rare alleles that contribute to trait variation
MCH	SLC17A3	rs1408272-G	0	0	No comment
MCV	HFE	rs1800562-A	3	5	HFE is known to be associated with iron homeostasis.
NSCL	Intergenic	rs17085106-T	N/A	N/A
Nasopharyngeal carcinoma	ITGA9	rs189897-A	0	1	The gene is located at the chromosomal 3p22-21.3 segment, which is known to be commonly deleted in various types of carcinoma including NPC. A linkage study also mapped an NPC susceptibility locus to chromosome 3p21.31-21.2, indicating that the genes in this region are crucial for the formation of NPC.
		rs197757-T	0	1
Panic disorder	TMEM16B a	rs12579350-A	0	1	No comment
	PKP1	rs860554-T	0	1	The gene has an important role in the cytoskeleton–cell membrane interaction. The protein of PKP1, plackoglobin, acts as linker molecules at adherence junctions and desmosome at the plasma membrane.
Prostate cancer	Intergenic	rs16901979-A	N/A	N/A
Primary biliary cirrhosis	C6orf10	rs2395148-A	0	0	No comment
ALP	PDZRN4, CNTN1	rs1880887-C	0	1	No comment (locus found only in supplement)
fT3	HS3ST3B1	rs3848445-C	0	1	No comment (locus found only in supplement)
Psoriasis	HLA-C	rs2395029-C	57	65	Strongest association with this region is consistent with previous results from our group and others.
Response to treatment for ALL	ST8SIA6	rs359312-T	0	0	No comment
Response to antipsychotic therapy	Intergenic	rs17022444-G	N/A	N/A
	Intergenic	rs7669317-C	N/A	N/A
SLE	TNFAIP3	rs5029939-G	0	10	Previously unreported for SLE susceptibility; recent reports for influencing rheumatoid arthritis risk. This GWAS identifies TNFAIP3 as a new susceptibility locus in SLE.
SLE	TNFAIP3	rs2230926-C	0	10	Reported by previous GWAS
Tanning	MATP	rs35391-T	0	0	SNPs in MATP were previously evaluated in the GWAS of natural hair color by our group. Three SNPs in the MATP gene have been associated with human pigmentation.
Triglycerides	APOA1, APOC3, APOA4, APOA5	rs662799-G	118 for APOA1, APOC3, APOA4, APOA5	165 combined for APOA1, APOC3, APOA4, APOA5	These loci have been previously implicated in lipid metabolism.
Triglycerides	APOA1, APOC3, APOA4, APOA5, DSCAML1	rs10892151-A	118 for APOA1, APOC3, APOA4, APOA5	165 combined for APOA1, APOC3, APOA4, APOA5	APOA1, APOC3, APOA4, APOA5 is a cluster of more likely candidate genes, given the established key roles of their products in lipid metabolism.
Type 1 diabetes	COBL	rs4948088-C	0	0	No comment
Type 2 diabetes	TCF7L2	rs7903146-T	14	140	It was reported by previous studies.

Abbreviations: AIDS, acquired immunodeficiency syndrome; ALL, acute lymphoblastic leukemia; ALP, alkaline phosphatase; BMD, bone mineral density; fT3, free triiodothyronine; GWAS, genome-wide association study(ies); HDL, high density lipoprotein; HuGE, Human Genome Epidemiology; LDL, low density lipoprotein; MCH, mean corpuscular hemoglobin; MCV, mean corpuscular volume; N/A, nonapplicable because the variants are in intergenic regions; NPC, nasopharyngeal carcinoma; NSCL, nonsyndromic cleft lip with or without cleft palate; SLE, systemic lupus erythematosus; SNP, single-nucleotide polymorphism.

^a TMEM16B was found as ANO2 in HuGE Navigator.

On the basis of the comments of the GWAS authors (Table 4), several of the loci of discovered uncommon variants had some evidence support from prior studies, although not necessarily gene–disease association studies on human populations.

Known mutations in the same gene loci

According to the Online Mendelian Inheritance in Man Database, for 11 gene loci (implicated in a total of 13 gene locus–phenotype associations) where uncommon variants had been identified by GWAS (OCA2 and eye color; TNFSF11 and bone mineral density; NOD2 and Crohn's disease; MC1R and freckles; HNF4A and HDL cholesterol levels; ABCA1 and HDL cholesterol; HFE and hematocrit, hemoglobin, and mean corpuscular volume; PCSK9 and LDL cholesterol levels; MATP and tanning; HLA-C and psoriasis; and APOA1/C3/A4/A5 and triglycerides), there were known mutations conferring the same or related phenotypic effects (Table 5).

Table 5.

Mutations in the Same Gene Loci With the Uncommon Variants Causing Related Phenotypic Effects, as Found in the Online Mendelian Inheritance in Man Database

Reported Gene(s)	Region	Uncommon Variant(s)	Disease/Trait	Mutations With the Same or Related Phenotypic Effects	Phenotypic Effects of Mutations
OCA2	15q13.1	rs1667394-A	Blue vs. green eyes	2.7-kb del, ex7del	Oculocutaneous albinism type 2
				IVS17DS, G-T, +1
				Pro743Leu
				1-bp del
				Ala334Val
				122.5-kb del
				Trp679Cys
				Asn489Asp
				Met394Ile
TNFSF11	13q14	rs6561055-A, rs17639156-G	BMD	5-bp del, IVS7+4	Osteopetrosis, autosomal recessive type 2
				Met199Lys
				2-bp del, 828CG
NOD2	16q12.1	rs2066844-T, rs2066845-C, rs2066847-C	Crohn's disease	3020insC	Crohn's disease
				Gly881Arg
				Arg675Trp
				IVS8+158
MC1R	16q24.3	rs1805007-T	Freckles	3-bp del 439TTC	UV-induced skin damage
				Thr157Ile
				Pro159Thr
HNF4A	20q13.12	rs1800961-C	HDL cholesterol	Gln268Ter	Maturity-onset diabetes of the young, type 1
				Arg154Ter
				Arg127Trp
				1-bp del Phe75T
				IVS5, Del A, -2
				Met364Arg
				Val393Ile	Type 2 Diabetes
ABCA1	9q31.1	rs9282541-T	HDL cholesterol	Cys1417Arg	Tangier disease (HDL deficiency type 1)
				IVS24DS, G-C
				Gln537Arg
				110-bp in/14-bp del
				2-bp del, 3283TC
				1-bp del, 2665C
				Ser1446Leu
				int12-14 del, int16-31 del
				Arg1680Trp
				Asp1229Asn
				Arg2021Trp
				1-bp del, 1764G
				Asn875Ser
				Ala877Val
				Trp530Ser
				1-bp del, 1764G
				Tyr573Ter
				3-bp del	HDL deficiency type 2
HFE	6p22	rs1800562-A	Hematocrit, Hemoglobin, MCV	Cys282Tyr	Hemochromatosis
				His63Asp
				Arg330Met
				Gln283Pro
PCSK9	1p32.3	rs11591147-G	LDL cholesterol	Asp374Tyr	Familial hypercholesterolemia, type 3
				Tyr142Ter	LDL cholesterol level quantitative trait locus 1
				Cys679Ter
				3-bp del 290_292delGCC
HLA-C	6p21.33	rs2395029-C	Psoriasis	HLA-C, HLA-Cw6 allele	Psoriasis
MATP	5p13.3	rs35391-T	Tanning	IVS2, G-A, -1	Oculocutaneous albinism type 4
				1-bp del, 986C
				3-bp del
				Ala486Val
				Asp157Asn
				1-bp del, 1121T
APOA1, APOC3, APOA4, APOA5	11q23.3	rs662799-G, rs10892151-A	Triglycerides	Gln84Ter (APOA1/APOC3)	Apolipoprotein A-I deficiency
				Val156Glu (APOA1/APOC3)
				Gln-2Ter (APOA1/APOC3)	Analphalipoproteinemia
				1-bp ins (APOA1/APOC3)	Primary hypoalphalipoproteinemia
				Gln32Ter (APOA1/APOC3)	Periorbital xanthelasma
				Gln139Ter (APOA5)	Hyperlipoproteinemia type 4

Abbreviations: BMD, bone mineral density; HDL, high density lipoprotein; kb, kilobase(s); LDL, low density lipoprotein; MCV, mean corpuscular volume; SNP, single-nucleotide polymorphism; UV, ultraviolet.

In 5 loci (HCCS, LRRK2, PKP1, CNTN1, HLA-C), mutations had been described with phenotypic effects (syndromic micropthalmia, Parkinson's disease, ectodermal dysplasia/skin fragility syndrome, Compton-North myopathy, human immunodeficiency virus, type 1 (HIV-1), viremia, respectively) that were not similar to those implicated in the GWAS-identified uncommon variants.

Confluence of common SNPs, prior candidate variants, or mutations in loci with uncommon variants discovered in GWAS

Overall, GWAS have discovered 30 different gene locus–phenotype associations involving uncommon variants where a single or multiple genes have been implicated. Of those, for 16 associations other common SNPs have been described by GWAS (n = 10), variants have been proposed by candidate gene studies prior to the first GWAS proposing the respective locus (n = 13), or mutations conferring similar or related phenotypes have been described (n = 13). For 4 of the 16 locus–phenotype associations, 2 of the 3 statements hold true, and for another 8 all 3 statements hold true.

For the remaining 14 gene locus–phenotype associations (KRTHB5 and acute lymphoblastic leukemia, C6orf48 and AIDS progression, AKAP11 and bone mineral density, HCCS and cognitive performance, SLC17A3 and mean corpuscular hemoglobin, ITGA9 and nasopharyngeal carcinoma, TMEM16B and panic disorder, PKP1 and panic disorder, C6orf10 and primary biliary cirrhosis, PDZRN4/CNTN1 and alkaline phosphatase, HS3ST3B1 and free triiodothyronine, ST8SIA6 and response to treatment for acute lymphoblastic leukemia, TNFAIP3 and systemic lupus erythematosus, COBL and type 1 diabetes), we did not identify common SNPs in the same locus with the uncommon variants, prior candidate-gene association studies, or mutations with a similar/related phenotypic effect.

Allele frequencies in populations of different ancestry

Three variants (rs11591147-T, rs9282541-T, and rs2066847-C) were not found in any of the 4 HapMap samples. Another 12 variants were uncommon in all 4 HapMap samples (Table 6). Therefore, 25 of the 40 variants were common in at least 1 HapMap sample.

Table 6.

Minor Allele Frequencies of the Eligible Uncommon Variants in the 4 HapMap Phases 1 + 2 Populations

Strongest SNP-Risk Allele	Reported Gene(s)	Region	MAF in GWAS	GWAS Population	MAF in CEU HapMap 1 + 2	MAF in CHB HapMap 1 + 2	MAF in JPT HapMap 1 + 2	MAF in YRI HapMap 1 + 2
rs2089222-A	KRTHB5	12q24.22	0.03	European	0.04	0.26	0.33	0.19
rs2395029-G	HCP5, MICB, MCCD1, BAT1, LTB, TNF	6p21.33	0.03	European	0.05	0.01	0	0
	HLA-C		0.03	European
rs9368699-C	C6orf48	6p21.3	0.03	European	0.06	0.18	0.10	0
rs1667394-A	OCA2	15q13.1	0.02	European	0.13	0.20	0.14	0.05
rs1805007-T	MC1R	16q24.3	0.05	European	0.15	0	0	0
rs5934953-C	HCCS	Xp22.2	0.02	European	0.04	0	0	0
rs180851-G	AKAP11	13q14	0.05	European	0.05	0.08	0.03	0.14
rs7326472-G	AKAP11	13q14	0.05	European	0.04	0.11	0.09	0.16
rs12854504-G	AKAP11	13q14	0.05	European	0.04	0.10	0.07	0.03
rs7998154-T	AKAP11	13q14	0.02	European	0.02	0	0	0
rs6561055-A	TNFSF11	13q14	0.05	European	0.04	0	0	0
rs17639156-G	TNFSF11	13q14	0.05	European	0.04	0.07	0.09	0
rs2066844-T	NOD2	16q12.1	0.05	European	0.11	0	0	0
rs2066845-C	NOD2	16q12.1	0.01	European	0.02	0	0	0
rs2066847-C	NOD2	16q12.1	0.04	European	0	0	0	0
rs11175593-T	LRRK2, MUC19	12q12	0.02	European	0.02	0.03	0.01	0
rs1800961-C	HNF4A	20q13.12	0.03	European	0.05	0.01	0	0
rs9282541-T	ABCA1	9q31.1	0.03	Mixed	0	0	0	0
rs1800562-A	HFE	6p22.1	0.04	European	0.04	0	0	0
			0.04	European
			0.04	European
rs11591147-G	PCSK9	1p32.3	0.01	European	0	0	0	0
rs1408272-G	SLC17A3	6p22.1	0.03	European	0.03	0	0	0
rs17085106-T	Intergenic	18q22.3	0.02	European	0	0	0	0.16
rs189897-A	ITGA9	3p22.2	0.03	Asian	0.27	0.08	0.12	0.01
rs197757-T	ITGA9	3p22.2	0.03	Asian	0	0.07	0.17	0.02
rs12579350-A	TMEM16B	12p13.31	0.01	Asian	0	0.01	0	0
rs860554-T	PKP1	1q32.1	0.05	Asian	0.20	0.12	0.06	0.01
rs16901979-A	Intergenic	8q24.21	0.03	European	0.02	0.29	0.16	0.46
rs2395148-A	C6orf10	6p21.3	0.02	European	0.05	0.21	0.07	0.09
rs1880887-C	PDZRN4, CNTN1	12q12	0.03	European	0.03	0.14	0.08	0.38
rs3848445-C	HS3ST3B1	17p12	0.05	European	0.05	0.32	0.26	0.17
rs359312-T	ST8SIA6	10p12.33	0.04	European, African, other	0	0.47	0.42	0
rs17022444-G	Intergenic	2p12	0.03	European, African, other	0	0	0	0.07
rs7669317-C	Intergenic	4q24	0.04	European, African, other	0.05	0	0	0
rs5029939-G	TNFAIP3	6q23.3	0.03	European	0.04	0.09	0.20	0.50
rs2230926-C	TNFAIP3	6q23.3	0.04	Asian	0.01	0.09	0.18	0.48
rs35391-T	MATP	5p13.3	0.03	European	0	0.38	0.35	0.47
rs662799-G	APOA1, APOC3, APOA4, APOA5	11q23.3	0.05	European	0.02	0.27	0.29	0.13
rs10892151-A	APOA1, APOC3, APOA4, APOA5, DSCAML1	11q23.3	0.03	European	0.02	0.06	0	0.41
rs4948088-C	COBL	7p12.1	0.05	European	0.02	0	0	0.04
rs7903146-T	TCF7L2	10q25.2	0.04	Asian	0.25	0.02	0.02	0.29

Abbreviations: CEU, Utah residents with Northern and Western European ancestry from the CEPH (Centre de'Etude du Polymorphism Humain) collection; CHB, Han Chinese in Beijing, China; GWAS, genome-wide association study(ies); HapMap, International HapMap Project; JPT, Japanese in Tokyo, Japan; MAF, minor allele frequency; SNP, single-nucleotide polymorphism; YRI, Yoruba in Ibadan, Nigeria.

DISCUSSION

Here, we systematically evaluated the characteristics of variants with a MAF of 5% or less that have reached levels of genome-wide significance (P ≤ 10⁻⁷) in GWAS. We identified 43 eligible SNP–disease associations, in 12 of which the implicated SNPs (9 in total) were exonic. Most were discovered and replicated in populations of European descent. The effect sizes were typically large. Some of these variants were identified in more than 1 GWAS on the same phenotype and, for 14 uncommon variant–phenotype associations (corresponding to 10 gene locus–phenotype associations), GWAS had also identified common variants for the same phenotype. Eleven loci implicated in 13 different locus–phenotype associations also had some evidence support from prior studies. Additionally, for 11 loci implicated in a total of 13 locus–phenotype associations, there was evidence for mutations conferring the same or related phenotypic effects. Most of the eligible uncommon SNPs would be common in at least 1 HapMap sample.

There are considerable debate and some preliminary evidence regarding the “rare variant–common disease” model of susceptibility to many complex diseases such as cancer, diabetes, and lupus (7, 89–95). According to this hypothesis, the multiplicative action of uncommon (13, 90) and rare (13) variants with modest and high odds ratios may explain a significant fraction of genetic variance in many common traits (89–91). In almost all the eligible associations that we overviewed that pertained to diseases, the risk allele had a frequency of 5% or less rather than 95% or greater. The only exception was a COBL variant apparently conferring susceptibility to type 1 diabetes, where the effect size was atypically small and the statistical support was among the weakest. Uncommon risk alleles may have an evolutionary disadvantage, and this does not allow them to become more prevalent in the population. They may also tend to be more recent, even if their effects are evolutionary neutral. Additionally, most of the associations in our study had odds ratios above 2, which is the usual odds ratio expected for associations involving uncommon variants (89–92, 94). However, odds ratios exceeding by far the small effect sizes typical of most GWAS-identified common variants (7) do not necessarily prove that uncommon variants routinely should always have such large effects. Because of power considerations, current studies are expected to identify predominantly those uncommon variants that have the largest effects (4, 13, 90). This is also supported by our analysis, which showed that the average sample sizes of most GWAS conducted to date are insufficient to detect the majority of uncommon SNPs with an odds ratio of less than 1.40. There are likely to be far more associations of uncommon variants with modest effects rather than large effects in the genetic architecture of complex traits. The majority of associations in the latter group have probably already been discovered, especially when large sample sizes have been amassed in GWAS.

Although uncommon and rare variants may constitute about 60% of variation in the human genome (90, 96), they are poorly covered in GWAS (8, 91, 97) and are often excluded from GWAS analyses by default, since a MAF threshold of 1% or greater or even 5% is often adopted as a quality control criterion by GWAS conducted to date. This may also explain the fact that all the SNPs that we identified were uncommon rather than rare; that is, they have a MAF = 0.5%–5%. Indeed, in our study, only a small minority of the variants indexed in the Catalog of Published Genome-Wide Association Studies had a MAF of 5% or less, and an even smaller minority were genome-wide significant. Detection of uncommon variants requires sample sizes (4, 98) much larger than those of most GWAS conducted to date (13). The situation may improve with much larger studies (99) or meta-analysis of multiple GWAS (100).

The finding that most of the uncommon variants in this overview were detected in populations of European ancestry simply reflects the fact that most GWAS have been conducted to date in these ethnic groups (101). As we have shown, relatively few of the identified uncommon variants are uncommon across all different ancestry groups. Conversely, several of the discovered common variants in GWAS are uncommon in other ancestry groups (102). Hence, investigating loci in other ethnicities that are statistically significantly associated with traits in 1 ethnicity may be a mechanism for discovering further associated rare variants.

Finally, we have identified several gene loci that contain both uncommon and common variants with genome-wide significance. The effect estimates of the uncommon variants were generally larger than the effects of the common variants. This supports the hypothesis that genes containing common variants with modest effects on common traits may also contain uncommon variants with much larger effects (13). Alternatively, uncommon and rare variants may create “synthetic associations” by occurring, stochastically, more often in association with one of the alleles at a common SNP site (103). However, we found few examples where common and uncommon variants had high linkage disequilibrium. Furthermore, some of these same loci carry known mutations causing related traits. Overall, this picture is more consistent with a confluence of rare, uncommon and common genetic variation on the same genetic loci, perhaps conferring independent effects in shaping complex traits (14).

Our study has some limitations. First, the number of the eligible associations is still limited. Second, the MAF of a specific allele may differ significantly between different studies, depending on the populations studied; thus, the same allele may be characterized as uncommon in 1 population and as common in another (104). The emergence of mature data from the 1,000 Genomes Project should give better accuracy in allele frequencies and a better characterization of rare/uncommon variants than is currently possible (105, 106). Third, we did not have data on the examined variants from all agnostic GWAS done on the same phenotype, since for some of them their effect estimate, P values, and MAFs were not retrievable. Effect sizes may be smaller than what we observed based on published data that may suffer to some extent from winner's curse (107–109).

The number of associations with uncommon/rare variants discovered in agnostic genotyping methods is expected to rise with new technologies for whole genome or exome sequencing (16–18). A current debate is whether focusing on exons rather than sequencing the whole genome may suffice for identifying a large share of the missing genetic dark matter. On the basis of our series, exons may include only a minority of these uncommon variants and, thus, full genome sequencing may be unavoidable for successful identification of most variants of interest. Moreover, given technical and power considerations, GWAS to date have not been able to tell us anything about the rare variants with a MAF less than 0.5%. Even with newer technologies, these will be captured only if they confer extremely large causal effects.

Glossary

Abbreviations

AIDS

acquired immunodeficiency syndrome

GWAS

genome-wide association study(ies)

HDL

high density lipoprotein

HuGE

Human Genome Epidemiology

LDL

low density lipoprotein

MAF

minor allele frequency

SNP

single-nucleotide polymorphism