Genome-wide Scans of Myopia in Pennsylvania Amish Families Reveal Significant Linkage to 12q15, 8q21.3 and 5p15.33

Abstract

Myopia is one of the most common ocular disorders in the world, yet the genetic etiology of the disease remains poorly understood. Specialized founder populations, such as the Pennsylvania Amish, provide the opportunity to utilize exclusive genomic architecture, like unique haplotypes, to better understand the genetic causes of myopia. We perform genetic linkage analysis on Pennsylvania Amish families that have a strong familial history of myopia to map any potential causal variants and genes for the disease. 293 individuals from 25 extended families were genotyped on the Illumina ExomePlus array and merged with previous microsatellite data. We coded myopia affection as a binary phenotype; myopia was defined as having a mean spherical equivalent (MSE) of less than or equal to −1 D (diopters). Two-point and multipoint parametric linkage analysis was performed under an autosomal dominant model. When allowing for locus heterogeneity, we identified two novel genome-wide significantly linked variants at 12q15 (heterogeneity LOD, HLOD = 3.77) in PTPRB and at 8q21.3 (HLOD = 3.35) in CNGB3. We identified a further three genome-wide significant variants within a single family. These three variants were located in exons of SLC6A18 at 5p15.33 (LODs ranged from 3.51 – 3.37). Multipoint analysis confirmed the significant signal at 5p15.33 with six genome-wide significant variants (LODs ranged from 3.6-3.3). Further suggestive evidence of linkage was observed in several other regions of the genome. All three novel linked regions contain strong candidate genes, especially CNGB3 on 8q21.3, which has been shown to affect photoreceptors and cause complete colorblindness. Whole genome sequencing on these regions is planned to conclusively elucidate the causal variants.

INTRODUCTION

Myopia, commonly called nearsightedness, is one of the most widespread eye disorders in the world. In myopic individuals, light focuses in front of the retina instead of directly on it. This causes distant images to be blurry and distorted. Myopia affects one quarter of Americans and prevalence rates have been steadily rising since the 1970s (Vitale et al. 2009).

Myopia has been shown to be a complex phenotype with multiple genetic and environmental underpinnings (Stambolian 2013; Wojciechowski and Hysi 2013). Near work hours and education level have been shown to have deleterious effects, while outdoor activity has been shown to have protective effects (Stambolian 2013). On the genetic side, population-based genome-wide association studies (GWAS) have identified multiple risk loci for a negative refractive error (Fan et al. 2016; Kiefer et al. 2013; Miyake et al. 2015; Simpson et al. 2014; Simpson et al. 2013; Stambolian et al. 2013; Verhoeven et al. 2012; Verhoeven et al. 2013), although the vast majority of these associations are to common variants with small to moderate effect. Family-based linkage studies have identified several linked regions that appear to harbor risk loci as well (Ciner et al. 2009; Guo et al. 2015; Musolf et al. 2017; Stambolian et al. 2004; Wojciechowski et al. 2006; Zhou et al. 2015). although the actual causal variant(s) at the linked and/or associated loci mostly remain unknown.

Most complex traits are still analyzed using some form of population-based GWAS; the older linkage studies that were a hallmark of the pre-1990’s genetic epidemiology studies are considerably less frequently used at present. Both the advent of inexpensive commercially available genotyping microarrays combined with the ease and low cost of collecting population-based samples when compared to collecting family-based samples are the main reasons GWAS have proliferated compared to linkage studies. However, the availability of genotype arrays with large numbers of rare variants and whole exome and whole genome sequencing have recently led to a resurgence in linkage studies since these approaches can be powerful for detecting rare causal variants of moderate to large effect.

Both types of studies have advantages and limitations. GWAS are more effective at identifying common risk variants (minor allele frequency (MAF) > 0.05) but are severely underpowered for locating rare, highly penetrant variants. On the other hand, family-based linkage studies are better equipped to find these rare, highly penetrant variants than population-based studies. While a variant may be rare in a population at large, if a rare variant is present in a given family it will be much more common in family members than in the general population. Further, when analyzing genotype data that does not cover every variant in the genome, such as from genotyping chips, family-based linkage studies can use genomic architecture to their advantage better than population-based data for discovery of large regions harboring rare, causal variants. Individuals in a population are distantly related, and an extremely large number of meioses have broken up haplotypes into small blocks consisting of the variants that only exhibit the strongest linkage disequilibrium (LD). If an ungenotyped rare or common causal variant is not in high LD with the genotyped variants on the SNP genotyping chip, then association methods will not exhibit high power to detect such a variant and when only 1 or 2 individuals in an entire GWAS study carry a rare variant, power to detect its effect on the trait is minimal even if it is actually genotyped in the study. By contrast, in family-based studies the haplotypes are determined by the haplotypes of the observed founders of each family; these haplotypes will be passed on to each subsequent generation. Since the majority of linkage studies are small (with families ranging from 2-4 generations), few recombinations will occur between variants that are moderately far apart on a chromosome and the end product is longer shared haplotypes across a genetic region among relatively close relatives. These shared haplotypes will boost power to find linkage between a trait and a causal variant along the haplotype even if the variant is not genotyped. In addition, founder populations, such as the Amish, may exhibit less genetic heterogeneity for complex traits than in the general population.

Family-based linkage studies also have the advantage of observing linkage both within an individual family unit and across multiple families. In parametric linkage analyses, each family is analyzed as its own discrete unit (not as a group of cases and controls like population-based association studies) and each family receives its own LOD score. LOD scores are cumulative and can be added across families to get an overall LOD score. The cumulative LOD scores allow for more power to identify variants that are increasing disease risk in multiple families, whereas the individual family LOD scores allow for detection of a unique linkage only present in a single large family (genetic heterogeneity). This permits identification of any LOD scores that may be relatively large in a single family but are washed out or overwhelmed by multiple negative LOD scores in other families in the cumulative LODs. This is especially important in complex phenotypes that exhibit locus heterogeneity, where multiple families can reasonably be expected to have different causal variants and possibly different causal genes. This ability to identify individual risk variants within a family will become more and more important as we drive toward the age of personalized medicine. Greater emphasize will be placed upon our ability to quickly and accurately detect what is going on with a single patient and their family instead of at the population level.

Here, we present results from our family-based linkage study of myopia in the Myopia Family Study (MFS), using an exome-based array. The families studied here are Pennsylvania Amish, a founder population comprised primarily of German and Swiss immigrants. The Amish began settling in the Pennsylvania colony in the 18^th century seeking religious tolerance and have become famous for their rural lifestyle and reluctance to adopt many aspects of modern technology. We have previously performed several genetic studies on Pennsylvania Amish families and refractive error/myopia, both of which have been been shown to be heritable and strongly aggregate in these Amish families (Peet et al. 2007). All of our previous studies used microsatellite data. A previously published genome-wide linkage scan of a sparse map of microsatelite markers in these families found several suggestive linkages to myopia, including 3q26 (HLOD=1.84), 11p13 (LOD=1.54) and 8p23 with a LOD=2.04 (Stambolian et al. 2005), which all independently confirmed Hammond et al.’s prior study’s genome-wide significant evidence of linkage of refractive error to these regions in an English population (Hammond et al. 2004). A further fine-mapping study of the 1p34-36 region confirmed linkage to refractive error in our MFS Amish and Ashkenazi Jewish families (Wojciechowski et al. 2009) and a family-based association study found two significant variants on chromosomes 11 and 16 with both variants being located in matrix metalloproteinase genes(Ciner et al. 2009). Here, we present for the first time genome-wide linkage results of these Pennsylvania Amish families using a dense exome microarray panel of marker genotypes, which has the advantage that a large number of moderately rare exonic variants as well as a backbone of common variants are genotyped.

MATERIALS AND METHODS

Study Design and Patient Recruitment

As part of the Myopia Family Study, 293 individuals from 25 extended Pennsylvania Amish families were recruited for this study. Details of the study recruitment are given in Stambolian et al (Stambolian et al. 2005). Briefly, these families were examined and enrolled in the study at the Amish Eye Clinic in Strasburg, Pennsylvania from January 1999 through August 2002. Eligibility in the study was limited to families that had at least three participating members and probands that had only one myopic parent and at least two myopic siblings. Any children under five years old were not permitted to participate. Both DNA samples and phenotype measurements were collected from all participants. The phenotype collection consisted of a comprehensive, dilated eye examination including refraction measurements. The specific measurement used in this study was mean spherical equivalent (MSE) in diopters (D). This value is calculated by adding the spherical refraction component to one-half of the cylindrical refraction component and averaging over both eyes. This study was approved by the institutional review boards of both the National Human Genome Research Institute and the University of Pennsylvania. All subjects provided informed consent and all procedures adhered to the Declaration of Helsinki.

Genotyping and Quality Control

Previously extracted (Stambolian et al. 2005), stored DNA samples were genotyped at the Center for Inherited Disease Research (CIDR) at Johns Hopkins University with the Illumina HumanExome-12v1-PlusCustomContent array. This array provided coverage of the exome and limited coverage of introns and intergenic regions. It is also highly enriched for rare variants; 6,000 of the final cleaned variants had a MAF < 0.01 and 28,000 had a MAF < 0.2, making it an effective array for finding rare, highly penetrant variants of large effect on the phenotype. Individual families were genotyped on the same plate. Blind duplicates and HapMap controls were distributed across the plates for concordance checks. Gender mismatched samples or samples with unusual X and Y patterns (such as XXY individuals) were removed prior to release. SNPs with a genotype quality score of less than 0.15 were recoded as missing; SNPs with cluster separation of less than 0.3 or heterozygote rate greater than 80% were dropped. SNPs were filtered to a call rate of 95% which resulted in a mean call rate of 99% per variant.

Seventy-two individuals who did not have genotype data were added to connect any disjointed pedigrees and insure proper relationships for the linkage analyses. These connecting individuals were known to exist from family history but for various reasons did not fully participate. In some cases, phenotype data was collected on these individuals. All these individuals were coded as having unknown genotypes at all variants and those without phenotype data were also coded as having an unknown phenotype. Identity-by-descent (IBD) values were calculated by PLINK (Purcell et al. 2007) to verify all familial relationships. Any variant with a single Mendelian inconsistency was dropped in that family; variants with multiple Mendelian errors were removed from the entire data set. PLINK was also used to detect any sex discrepancies and to remove all monomorphic markers.

All SNP genotypes were then merged with a previously published set of 367 genotyped sequence tagged sites (STS) or microsatellite markers (Stambolian et al. 2005). The final genotype data set consisted of 365 individuals (293 genotyped) from 25 extended families and 52, 035 genotyped markers (51,668 SNPs and 367 STS). All markers were mapped onto the Rutgers Genetic Map v3 using GRCh37 physical positions (Matise et al. 2007).

For the linkage analyses, allele frequencies were calculated for the data set using sib-pair (Duffy 2008). Estimating allele frequencies from the data set to be analyzed has been shown to properly control type I error rates (Mandal et al. 2006; Mandal et al. 2001; Mandal et al. 2000) since the allele frequencies in the founders of the families to be studied may differ substantially from allele frequencies in databases, especially for families from ethnic groups that are not well-represented in these databases. This is especially important for a founder population such as the Amish, because their allele frequencies at certain variants may deviate from those in the general European population.

Myopia Affection Classification

We used a binary coding scheme (i.e. an individual was either affected with myopia or not) based on MSE thresholds. All participants with a refraction of −1.00 D were coded as being affected with myopia. The refraction score was based on an actual eye examination or if the person was not available for the exam, medical records or glasses prescriptions were used instead. Adult participants that had a refractive score of 0.00 D or greater were coded as unaffected and those with a refractive score in the range of −1.00 D to 0.00 D were coded as having a missing or unknown phenotype. Extra precaution was given to unaffected children, as normal developmental changes during childhood could lead to potential misclassification. Therefore, stricter thresholds were used for unaffected status. Children in the 5-10 years old range were considered unaffected if their MSE was 2.00 D or higher and unknown if their refractive score was between −1.00 D and 2.00 D. Children in the 11-20 age group were considered unaffected if their refractive score was 1.50 D or higher and unknown if the refractive score was between −1.00 D and 1.50 D. This is a conservative approach based on lack of a good segregation model of age-dependent penetrance and appropriate genotype probabilities for young unaffected subjects. The final data set contained 165 affected individuals, 116 unaffected individuals, and 84 individuals with unknown phenotype. The average MSE was −1.60 D with a standard deviation of 2.52. There were 25 pedigrees – 12 two generation pedigrees, 12 three generation pedigrees, and 1 four generation pedigree. 14 of the pedigrees had 7-10 individuals, 6 had 11-17 individuals, 3 had 20-22 individuals, 1 had 30 individuals and 1 had 73 individuals (the four generation pedigree). All pedigrees had at least three affected individuals.

Parametric Linkage Analysis

Parametric two-point linkage analysis was performed on these families using the TwoPointLods software (Thomas). We assumed an autosomal dominant mode of inheritance for our model. This was clearly evident from examination of the pedigrees, as the phenotype appeared in all generations and did not affect males or females at a significantly different rate. The disease allele frequency (DAF) was 0.0133 and penetrance was set at 90% for carriers with no phenocopies. We selected a rare disease allele frequency even though myopia itself is a common trait since our objective in this study was to find a rare, highly penetrant variant with a large effect on myopia. Also, these families were selected for extreme aggregation of myopia in an apparent autosomal dominant pattern. Since such families are not common in the general population, it is reasonable to believe that a high penetrant myopia risk variant would be rare in the general population. The allele frequencies, penetrances, and autosomal dominant model were further confirmed by previous sensitivity analysis of this data set (Ibay et al. 2004; Stambolian et al. 2005; Wojciechowski et al. 2009) using microsatellites that confirmed these parameters as the best fit for this data. Further sensitivity analyses on these data using DAF = 0.01 and 0.02 showed no significant difference in LOD scores from the preferred model of DAF = 0.0133. Linkage analysis was performed on all families individually and LOD scores were added across families to form a cumulative LOD score at each variant. Heterogeneity LOD (HLOD) scores were also calculated across families to account for any heterogeneity across families at a given variant.

Parametric multipoint analysis was performed using SimWalk2(Sobel and Lange 1996; Sobel et al. 2002; Sobel et al. 2001). To prevent type I error inflation caused by inter-marker linkage disequilibrium in a dense marker map (such as in this exome chip), the variants were collapsed into 1 cM bins. The SNP with the highest MAF (and thus highest information content) in the bin was chosen to represent the bin in the multipoint analysis. Further pruning removed any SNPs with a r² value greater than 0.2. 3,285 variants were left or analysis after the pruning process. The multipoint analysis assumed the identical parameters as the two-point analysis - autosomal dominant mode of inheritance, DAF = 0.0133, and penetrance matrix of 90% for carriers and 0% for non-carriers.

Functional Annotation of Variants

All variants were annotated using wANNOVAR(Chang and Wang 2012; Wang et al. 2010), which provides information such as allele frequencies from 1000Genomes. It also provides protein predictions (i.e. whether a variant is predicted damaging) from multiple programs, including SIFT(Kumar et al. 2009; Ng and Henikoff 2003; Sim et al. 2012; Vaser et al. 2016; Zhao et al. 2013), PolyPhen2(Adzhubei et al. 2013), MutationTaster(Schwarz et al. 2014), and REVEL (Ioannidis et al. 2016).

RESULTS

Overall (Across Family) Two-point Linkage Results

In analyses of all families combined, there were two genome-wide significantly linked variants located on chromosomes 12 and 8 and an additional 64 variants that showed suggestive linkage to myopia (Figure 1) using two-point linkage. Here, we define genome-wide significant as a (H)LOD >= 3.3 and genome-wide suggestive as (H)LOD >= 1.9, as recommended by Lander and Kruglyak (Lander and Kruglyak 1995). The highest HLOD score (3.77) was located at rs2584021, an exonic SNP in the PTPRB gene at 12q15. The variant is nonsynonymous and predicted benign by SIFT and PolyPhen2 but possibly damaging by MutationTaster. It is moderately rare in Europeans (MAF = 0.055 in 1000Genomes) and slightly enriched in our data set (MAF = 0.078). There was one other suggestive variant at 12q, rs1025016, an exonic nonsynonymous SNP in BEST3. This variant was predicted damaging by multiple prediction algorithms including SIFT, PolyPhen2, and MutationTaster. It has a MAF of 0.08 in 1000Genomes and 0.05 in our data set. A LocusZoom style plot of the 12q15 signal with gene locations can be found in Supplemental Figure 1 (Pruim et al. 2010).

Fig. 1.

Two-point Genome-wide HLOD Scores for All Families. The genome-wide plot for the HLOD scores across all 25 families using two-point linkage analysis. The lines at 3.3 and 1.9 represent the thresholds for genome-wide significant and suggestive HLOD scores

The other significant variant was rs6471482 (HLOD = 3.35) a nonsynonymous exonic SNP in CNGB3 at 8q21.3. It is predicted benign by SIFT and PolyPhen2 but possibly damaging by MutationTaster. In 1000Genomes Europeans it has a MAF of 0.14 but is rare in other populations and only has an overall 1000Genomes MAF of 0.04. In our data set, it has a MAF of 0.08. Three other additional SNPs were suggestive at 8q21.3, including one in an intron of CNGB3. A zoomed plot of the 8q21.3 signal with gene locations can be found in Supplemental Figure 2.

There was another SNP that exhibited linkage evidence that was just below genome-wide significance (HLOD = 3.26). This was rs2881194, an intronic SNP located in RPS6KA2 at 6q27. The additional suggestive variants were spread among 10 different chromosomes. There was a high concentration of suggestive linkage signals (28 variants) in the 6p21.32 region; however, this region is home to the human leukocyte antigen (HLA) complex of proteins. A list of significant and suggestive variants, along with selected annotations, can be found in Table 1.

Table 1:

All Significant and Suggestive SNPs across all Families

CH	SNP	HLOD	GENE	FUNC	EXON	FRQ	SIFT	POLPH	MUTR
12	rs2584021	3.77	PTPRB	exonic	nonsyn	0.06	T	B	P
8	rs6471482	3.35	CNGB3	exonic	nonsyn	0.14	T	B	P
6	rs2881194	3.26	RPS6KA2	intronic	.	0.08	.	.	.
7	rs7792939	2.85	TMEM225B	intronic	.	0.13	.	.	.
5	rs74581452	2.73	SLC6A18	exonic	syn	0.03	.	.	.
5	rs13188259	2.57	FYB	intronic	.	0.17	.	.	.
5	rs7705355	2.57	SLC6A18	exonic	nonsyn	0.04	T	B	P
2	rs76644468	2.56	GREB1	exonic	nonsyn	0.05	T	B	N
19	rs17628	2.56	RPS16	exonic	syn	0.38	.	.	.
5	rs10058728	2.55	CSNK1A1	intronic	.	0.47	.	.	.
8	rs10504826	2.52	CNGB3	intronic	.	0.28	.	.	.
7	rs2572023	2.48	OR2AE1	exonic	nonsyn	0.49	T	B	P
6	rs2239804	2.44	HLA-DRA	intronic	.	0.46	.	.	.
12	rs1025016	2.43	BEST3	exonic	nonsyn	0.08	D	D	D
5	rs7728667	2.36	SLC6A18	exonic	nonsyn	0.21	T	B	P
7	rs17822236	2.33	BMPER	intronic	.	0.17	.	.	.
10	rs4750568	2.30	MEIG1	exonic	nonsyn	0.33	T	P	P
19	rs10775583	2.29	SBSN	exonic	nonsyn	0.36	T	B	P
8	rs2882217	2.26	LOC100130298, CLVS1	intergenic	.	0.40	.	.	.
19	rs2304220	2.24	TIMM50	exonic	nonsyn	0.17	D	B	P
5	rs17586674	2.16	LOC100506858, LSINCT5	intergenic	.	0.44	.	.	.
6	rs2894239	2.13	NOTCH4, LOC101929163	intergenic	.	0.41	.	.	.
6	rs440169	2.13	NOTCH4, LOC101929163	intergenic	.	0.41	.	.	.
6	rs3115573	2.13	NOTCH4, LOC101929163	intergenic	.	0.41	.	.	.
7	rs148981077	2.13	LRWD1	exonic	nonsyn	0.04	T	B	N
15	rs2279482	2.13	FAM189A1	exonic	nonsyn	0.08	T	B	P
15	rs2306933	2.13	FAM189A1	exonic	nonsyn	0.07	T	B	P
19	rs8102258	2.10	ZNF844	exonic	nonsyn	0.08	D	B	P
6	rs537757	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs4713518	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs742582	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs2143465	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs505274	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs1033500	2.10	C6orf10	exonic	nonsyn	0.37	T	B	P
6	rs2073046	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs1559876	2.10	NOTCH4, LOC101929163	intergenic	.	0.36	.	.	.
6	rs531094	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs9268368	2.10	C6orf10	exonic	nonsyn	0.37	T	B	P
6	rs9405090	2.10	C6orf10	exonic	nonsyn	0.37	T	.	P
6	rs547261	2.10	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs477005	2.09	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs2076540	2.09	LOC101929163	ncRNA	.	0.37	.	.	.
6	rs560505	2.09	C6orf10	exonic	nonsyn	0.37	T	B	P
6	rs2395110	2.09	NOTCH4, LOC101929163	intergenic	.	0.42	.	.	.
6	rs6929776	2.09	LOC101929163	ncRNA	.	0.38	.	.	.
6	rs547077	2.09	LOC101929163	ncRNA	.	0.38	.	.	.
4	rs10009336	2.08	KCTD8, YIPF7	intergenic	.	0.18	.	.	.
1	rs10489962	2.06	FHAD1	exonic	nonsyn	0.06	T	B	P
2	rs2241883	2.06	FABP1	exonic	nonsyn	0.35	T	B	P
4	rs1497574	2.04	GABRG1	intronic	.	0.45	.	.	.
4	rs976156	2.04	GABRG1	exonic	syn	0.45	.	.	.
4	rs1391174	2.04	GABRG1	intronic	.	0.45	.	.	.
7	rs3800939	2.03	LRRC17	exonic	nonsyn	0.04	T	P	P
15	rs6495867	1.96	C15orf41	intronic	.	0.14	.	.	.
15	rs16963938	1.96	C15orf41	intronic	.	0.13	.	.	.
15	rs16963949	1.96	C15orf41	intronic	.	0.13	.	.	.
6	rs2187818	1.95	LOC101929163, HLA-DRA	intergenic	.	0.38	.	.	.
6	rs614549	1.95	SLC44A4	intronic	.	0.40	.	.	.
4	rs34540355	1.94	NEK1	exonic	nonsyn	0.05	T	B	N
19	rs73058047	1.94	PRR12	exonic	nonsyn	0.004	D	D	N
19	rs8110889	1.94	CHST8	intronic	.	0.25	.	.	.
6	rs6941112	1.93	STK19	intronic	.	0.33	.	.	.
6	rs4151657	1.93	CFB	intronic	.	0.36	.	.	.
6	rs181997	1.92	LOC100294145, HLA-DMB	intergenic	.	0.34	.	.	.
8	rs2976189	1.92	SLC7A13	exonic	nonsyn	0.15	D	D	P
6	rs3130320	1.90	LOC101929163	ncRNA	.	0.35	.	.	.

Legend: Table of all significant and suggestive SNPs from two-point linkage analysis. Here, the headers represent: CH = chromosome, SNP = rsID of SNP, HLOD = heterogeneity LOD score, GENE = gene location of SNP, FUNC = SNP function (e.g. exonic, intronic), EXON = exonic function (nonsyn = nonsynonymous, syn = synonymous), FRQ = 1000Genomes minor allele frequency for Europeans, SIFT = SIFT prediction (D = damaging, T = tolerated), POLPH = PolyPhen2 prediction (B = benign, D = damaging, P = possibly damaging), MUTR = MutationTaster prediction (N = neutral, D = damaging, P = possibly damaging)

Individual Two-point Linkage Family Results

An advantage to having the large, multiplex families that comprised this study is that there is often sufficient power to observe significant linkage signals in an individual family. Indeed, we observed one family (Family 3316) that contained 3 genome-wide significant LOD scores (Figure 2). Family 3316 (the family with the significant linkage peak on chromosome 5p) is a three generational pedigree with 30 people (10 affecteds, 9 unaffecteds, and 11 unknowns) (Supplemental Figure 3). All three variants were located within the SLC6A18 gene at 5p15.33. rs74581452, a synonymous exonic SNP, had the highest LOD score of 3.5. The SNP is moderately rare in 1000Genomes Europeans (MAF = 0.03) and slightly enriched in our data set (MAF = 0.065). The other two variants were both nonsynonymous exonic variants, rs7705355 (LOD = 3.47) and rs7728667 (LOD = 3.37). rs7705355 is moderately rare in Europeans (MAF = 0.04) and enriched in our data set (MAF = 0.08). rs7728667 is relatively common both in our data set (MAF = 0.14) and Europeans in general (MAF = 0.21). Both SNPs were predicted to be possibly damaging by MutationTaster. The three genome-wide significant SNPs are part of a larger linkage peak at 5p15.33 that contained an additional 6 suggestive SNPs (Figure 2, Table 2, Supplemental Figure 4).

Fig. 2.

Family 3316 Two-point Genome-wide and Chromosome 5 LOD scores. a The genome-wide two-point LOD score plot for family 3316. b The chromosome 5 two-point LOD score plot for family 3316, showing a clearer look at the significant linkage signal at 5p15. The lines at 3.3 and 1.9 represent the thresholds for genome-wide significant and suggestive LOD scores

Table 2:

Significant and Suggestive SNPs at 5p15.33 in Family 3316

CH	SNP	LOD	GENE	FUNC	EXON	FRQ	SIFT	POLPH	MUTR
5	rs74581452	3.51	SLC6A18	exonic	syn	0.03	.	.	.
5	rs7705355	3.47	SLC6A18	exonic	nonsyn	0.04	T	B	P
5	rs7728667	3.37	SLC6A18	exonic	nonsyn	0.21	T	B	P
5	rs17586674	3.07	LOC100506858, IRX2	intergenic	.	0.44	.	.	.
5	rs31489	2.98	CLPTM1L	intronic	.	0.41	.	.	.
5	rs401681	2.97	CLPTM1L	intronic	.	0.44	.	.	.
5	rs4975616	2.97	MIR4457, CLPTM1L	intergenic	.	0.42	.	.	.
5	rs4073918	2.59	SLC6A18	exonic	nonsyn	0.33	T	B	P
5	rs402710	2.45	CLPTM1L	intronic	.	0.33	.	.	.

Legend: Table of all significant and suggestive SNPs from two-point linkage analysis. Here the headers represent: CHR = chromosome, SNP = rsID of SNP, LOD = LOD score in family 3316, GENE = gene location of SNP, FUNC = SNP function (e.g. exonic, intronic), EXON = exonic function (nonsyn = nonsynonymous, syn = synonymous), FRQ = 1000Genomes minor allele frequency for Europeans, SIFT = SIFT prediction (D = damaging, T = tolerated), POLPH = PolyPhen2 prediction (B = benign, D = damaging, P = possibly damaging), MUTR = MutationTaster prediction (N = neutral, D = damaging, P = possibly damaging)

No other families contained any genome-wide significant variants, though many contained multiple genome-wide suggestive LOD scores. Two signals deserve closer examination. First is the 1p signal in Family 3333, a three generation family with 13 people (5 affecteds and 8 unaffecteds) (Supplemental Figure 5). This signal consists of a long linked haplotype across 1p36.22-32.3 (approximately 57 cM) consisting of 249 variants with LOD scores of 1.9 (Figure 3, Supplemental Figure 6). The figure also clearly shows that there are several other haplotypes of lower magnitudes (e.g. haplotypes at LOD = 1.6 and 1.3) and little to no negative linkage signal in the region. This suggests a lower probability of this particular signal being a false positive.

Fig. 3.

Family 3333 Two-point Genome-wide and Chromosome 1 LOD scores. a The genome-wide two-point LOD score plot for family 3333. b The chromosome 1 two-point LOD score plot for family 3333, showing a clearer look at the suggestive long linked haplotype at 1p36.22-32.22. The line at 1.9 represents the threshold for genome-wide suggestive LOD scores

The other interesting individual family signal was in Family 3074, a two generation pedigree with 13 individuals (6 affecteds, 4 unaffecteds, and three unknowns) (Supplemental Figure 7). This signal was a very small linked haplotype of 5 variants at 3q25.1 (Figure 4, Supplemental Figure 8). All variants had a LOD score of 2.22, with four of the five variants being located in or near antisense RNA genes. Like the larger linked haplotype in Family 3333, there are many lower magnitude linked variants underneath and almost no negative signal.

Fig. 4.

Family 3074 Two-point Genome-wide and Chromosome 3 LOD scores. a The genome-wide two-point LOD score plot for family 3074. b The chromosome 3 two-point LOD score plot for family 3074, showing a clearer look at the suggestive long linked haplotype at 3p25.1. The line at 1.9 represents the threshold for genome-wide suggestive LOD scores

Multipoint Linkage Results

The multipoint linkage results on the pruned marker set did not produce any genome-wide significant HLOD scores (Figure 5). Eight genome-wide suggestive markers were identified, all located in the 5p15.33 region identified by the two-point analysis. Almost all the genome-wide significant and suggestive signals identified in the two-point analyses were severely depressed here in the multipoint analysis, likely do to the large loss of information in the pruning process and some stronger negative signal in a few families. For instance, the signal at 8q21.3 is still clearly present but has been diminished to a maximum of 0.87. Performing the multipoint analysis with the 20, 15, and 10 most informative families gradually restores the signal to genome-wide significance at 3.3 (Supplemental Figure 9a). This is also true of the signal at 12q15, though even using the most highly linked families here only recovers the signal to 1.7; there is more information loss here even at the most highly linked families (Supplemental Figure 9b).

Fig. 5.

Multipoint Genome-wide HLOD Scores for All Families. The genome-wide plot for the HLOD scores across all 25 families using multipoint linkage analysis. The line at 1.9 represents the threshold for genome-wide suggestive HLOD scores

Unlike the across family results, the individual family multipoint results closely mirrored the two-point results. Family 3316 displayed six genome-wide significant LOD scores at 5p15.33, ranging from LOD = 3.62-3.33 (Figure 6, Table 3) and was the driving force in this signal being the only genome-wide suggestive signal across families. An additional two genome-suggestive results were found at the same region; no other genome-wide suggestive results were found anywhere else in the family. Again, the highest variant was rs74581452 in SLC6A18. All other significant or suggestive variants were intronic or intergenic.

Fig. 6.

Multipoint Genome-wide and Chromosome 5 LOD scores for Family 3316. a The genome-wide multipoint LOD score plot for family 3316. b The chromosome 5 multipoint point LOD score plot for family 3316, showing a clearer look at the significant linkage signal at 5p15. The lines at 3.3 and 1.9 represent the thresholds for genome-wide significant and suggestive LOD scores

Table 3:

Significant and Suggestive SNPs at 5p15.33 in Family 3316 for Multipoint Linkage Analysis

CH	SNP	LOD	GENE	FUNC	EXON	FRQ
5	rs74581452	3.62	SLC6A18	exonic	syn	0.03
5	rs27047	3.62	SLC6A3	intronic	.	0.31
5	rs2354124	3.61	MIR4277, MRPL36	intergenic	.	0.45
5	rs4527649	3.60	MIR548BA, LOC100506858	intergenic	.	0.48
5	rs17586674	3.59	LOC100506858, IRX2	intergenic	.	0.44
5	rs2962595	3.33	LOC100506858, IRX2	intergenic	.	0.45
5	rs1651478	2.28	IRX1, LOC10192915	intergenic	.	0.46
5	rs492478	2.11	IRX1, LOC10192915	intergenic	.	0.075
5	rs12110273	1.93	IRX1, LOC10192915	intergenic	.	0.47

Legend: Table of all significant and suggestive SNPs from multipoint linkage analysis. Here the headers represent: CHR = chromosome, SNP = rsID of SNP, LOD = LOD score in family 3316, GENE = gene location of SNP, FUNC = SNP function (e.g. exonic, intronic), EXON = exonic function (nonsyn = nonsynonymous, syn = synonymous), FRQ = 1000Genomes minor allele frequency for Europeans

The variants at 5p15.33 in Family 3316 were the only genome-wide significant variants in any of the individual families; though there were additional genome-wide suggestive variants. Specifically, the suggestive signal at 1p36.22 in Family 3333 was recapitulated in the multipoint (Supplemental Figure 10). Here, the signal started still started at 1p36.22 but ended slightly earlier at 1p34.2. All variants within this haplotype had LOD scores of ~2. The linked haplotype on 3q25.1 was also replicated in the multipoint, again with five tightly clustered variants with LOD scores ranging from 2.23-1.8 (Supplemental Figure 11).

DISCUSSION

In this parametric linkage study using multiplex Pennsylvania Amish families, we have identified three genome-wide significant linkage signals to myopia at 12q15, 8q21.3, and 5p15.33. The 12q and 8q signals were each due to a single variant that exhibited a genome-wide significant HLOD score across families while the 5p15 linkage consisted of three genome-wide significant SNPs in a single family. Multipoint analysis confirmed the final signal with six genome-wide significant variants at 5p15.33 in the same single family.

The 12q15 signal had the highest HLOD score (3.77) in the two-point analysis. Several previous studies had found linkage upstream of this region at 12q21-24 (Farbrother et al. 2004; Hawthorne et al. 2013; Klein et al. 2011; Young et al. 1998) and downstream at 12q13.11 (Metlapally et al. 2009) but neither overlaps with this signal on 12q15. The signal centers on the protein tyrosine phosphatase receptor type B (PTPRB). The linked variant and gene do not have any prior evidence of involvement in causation of myopia or eye disease in general. However other protein tyrosine phosphatase receptors have been associated or linked with myopia including PTPRR, PTPRF, and PTPRJ (Haw thorne et al. 2013; Musolf et al. 2017). Both PTPRR and PTPRF have been put forth as possible candidate genes for the MYP3 locus at 12q21-24. A suggestive rare nonsynonymous variant that was predicted highly damaging by multiple databases was found at BEST3, also at 12q15. While BEST3 has not been implicated in eye disease, another member of the bestrophin family, BEST1, is known to be the causal gene for vitelliform macular dystrophy (also known as Best’s macular dystrophy or Best’s disease) (Marchant et al. 2001; Marquardt et al. 1998; Petrukhin et al. 1998). Given that multiple independent linkage studies have implicated this region, the 12q13.1 to 12q24 region should be considered an excellent candidate region for one or more genes involved in myopia causation.

The significant linkage signal at 8q21.3 is novel and centered on the cyclic nucleotide gated channel beta 3 gene (CNGB3). A previous suggestive linkage had been located on chromosome 8 using microsatellites in these Amish families, however this was located on the opposite arm at 8p23 (Stambolian et al. 2005). CNGB3 has been found to affect photoreceptors and has been found to be causal in eye diseases, especially achromatopsia (Kohl et al. 2000; Kohl et al. 2005) and progressive cone dystrophy (Michaelides et al. 2004). Achromatopsia symptoms include total colorblindness (i.e. only seeing in shades of gray), reduced visual acuity, and light sensitivity. Progressive cone dystrophy results in reduced color vision and visual acuity with symptoms worsening over time. This is the first time CNGB3 has been reported to be linked or associated with myopia.

The 5p15.33 linkage was driven by a single family and the linked region contained three genome-wide significant variants in the solute carrier family 6 member 18 (SLC6A18) gene in the two-point analysis. Six genome-wide significant variants (one repeated from the two-point analysis) were found in the same region in the same family. One variant was again in SLC6A18, one was in solute carrier family 6 member 3 (SLC6A3), and the rest were in intergenic regions. SLC6A18 is a sodium cotransporter that is not known to have any implication in myopia or eye disease in general, though SLC transporters have been shown to be expressed in the retina (Nishimura and Naito 2008), including SLC6A3(Andre et al. 2012), identified as significant in the multipoint analysis. There are other good candidate genes in the region, including iroquois homeobox 2 (IRX2) and iroquois homeobox 1 (IRX1). Intergenic regions between both these genes were found to be significant in the multipoint analysis. The iroquois homeobox genes are critical transcription factors in development and have been shown to regulate retinogenesis (Choy et al. 2010).

We identified multiple suggestive signals as well, including an apparent long linked haplotype from 1p36.22-32.2 that was identified by both two-point and multipoint analysis within a single family. This region contains the known myopia risk locus MYP14 at 1p36, first identified in Ashkenazi Jewish subjects using quantitative measures of refraction (Wojciechowski et al. 2006) and later replicated in other, larger scale linkage studies (Abbott et al. 2012; Klein et al. 2011; Li et al. 2009). This is the first time the locus has been identified in the Amish population, or within a single family (all other linkage signals were cumulative across multiple families). Despite multiple replications, the causal gene of MYP14 remains unknown. The fact that our signal is confined to a single family gives us the unique opportunity to perform targeted sequencing of the 1p haplotype to try to identify the causal locus.

Another very small suggestive linked haplotype was found at 3q25.1, again identified in a single family by both two-point and multipoint analysis. This region does contain a few interesting genes, specifically muscleblind like splicing regulator 1 (MBNL1) and its antisense RNA MBNL1-AS1. MBNL1 is known to cause myotonic dystrophy and eye abnormalities (Kanadia et al. 2003).

The two-point and multipoint results were in clear agreement for the individual family results, both identified multiple genome-wide significant variants at 5p15.33 in Family 3316, and genome-wide suggestive signals at 1p36-34 in Family 3333 and 3q25.1 in Family 3074. There were some discrepancies in the across family HLOD scores; the multipoint cumulative HLOD scores were consistently lower than the two-point scores, including at 8q21.3 and 12q15. This is almost certainly due to the loss of information from the pruning process. The exome-based chip used here is not well-suited to multipoint analyses. It provided very good coverage of the exome but also created a very dense marker map in those regions with high LD and a very sparse in intergenic regions. Thus, pruning was required and we were left with only 3,285 variants for multipoint analysis; a very large loss of information. For example, in the 5 cM flanking regions around the 8q and 12q signals, 88 and 108 variants, respectively, were removed by pruning. Many of these SNPs had informative HLOD scores > 1 and led to a loss of power at these sites. Further, this exome array is not a SNP linkage panel (where the variants are chosen to be highly informative). It is enriched with rare variants, with over 6,000 of the variants having a MAF < 0.01 and over half the variants (approximately 29,000) had a MAF < 0.2. Even after pruning for the most informative markers at each 1 cM bin, there were over 1000 markers with MAF < 0.3. Pruning the markers resulted in situations where certain families were less informative or uninformative at a given marker, resulting in wholesale lower HLOD scores for all signals than the two-point analysis. This is why we observed a wholesale decrease in across family HLOD scores in the multipoint. The signals at 8q and 12q, which were of small, cumulative effect and centered on variants with a MAF < 0.15, are at particular risk for losing information added across the families. We were able to demonstrate that by using the most highly linked families for each of the regions, the signal could be gradually recovered; in the case of 8q21.3 it was recovered to being genome-wide significant (Supplemental Figure 9). The information problem was not as much of an effect on the individual family signals like 5p15.33, which are to a single family, is of larger effect, and is located on a longer, intra-family haplotype. Variants only needed to be informative in that family, not in cumulative fashion across the families.

In conclusion, we have found significant linkage for myopia in Amish families to 12q15, 8q21.3, and 5p15.33. Further suggestive signals were found, including a replication of the knownMYP14 locus at 1p36 (localized to a single family). The 5p15.33 signal is particularly promising, as it was identified in both the two-point and multipoint linkage analyses. Since it is confined to a single family, it is likely of large effect on myopia within that family. The 8p21.3 and 12q15 signals are cumulative across families and of small effect on the phenotype. They were not present in the multipoint analysis, likely due to loss of information in the pruning process. Many of the linked loci have very promising candidate genes, particularly CNGB3 at 8q21.3. SLC6A18 at 5p15.33 is also very interesting as even though it has not been previously implicated in any eye disease, multiple exonic variants were found to be genome-wide significant within a single family We plan on performing future functional analysis with SLC6A18 in an attempt to observe any possible eye phenotypes. The two IRX genes at 5p15.33, which are known are also be involved in retinogenesis, are also very good candidate genes. However, we note any implications of causality are still speculative at this point. We used a limited exome-based microarray in this study, thus it is possible that the causal variant was not genotyped on this chip and the significant variants here are simply tagging the true causal variants located along the same linked haplotype. To further elucidate the casual variants, targeted sequencing is planned. We plan to perform targeted sequencing on our most informative families at 12q15 and 8q21.3 as well as all of family 3316 at 5p15. Sequencing may also be performed at some of the suggestive linked haplotypes in individual families, particularly at 1p36.22-32.22 in family 3333.