Recombinant Haplotypes Narrow the ARMS2/HTRA1 Association Signal for Age-Related Macular Degeneration

Age-related macular degeneration (AMD) is the leading cause of blindness in ageing societies, triggered by both environmental and genetic factors. The strongest genetic signal for AMD with odds ratios of up to 2.8 per adverse allele was found previously over a chromosomal region in 10q26 harboring two genes, ARMS2 and HTRA1, although with little knowledge as to which gene or genetic variation is functionally relevant to AMD pathology. In this study, we analyzed rare recombinant haplotypes in 16,144 AMD cases and 17,832 controls from the International AMD Genomics Consortium and identified variants in ARMS2 but not HTRA1 to exclusively carry the AMD risk with P-values between 1.0 × 10−773 and 6.7 × 10−5. This now allows prioritization of the gene of interest for subsequent functional studies.

2010; Kanda et al. 2010;Friedrich et al. 2011Friedrich et al. , 2015. In addition, as both ARMS2 and 88 HTRA1 harbor functional variants which can be related to relevant disease processes, 89 so far, it is unclear to which gene the observed disease association can functionally be 90 attributed (Dewan et al. 2006;Fritsche et al. 2008;Cheng et al. 2013;Friedrich et al. 91 2015). 92 Despite the strong LD in the ARMS2/HTRA1 interval, the region still exhibits some level 93 of recombination resulting in rare recombinant haplotypes. Similar to gene mapping in 94 monogenic diseases, recombinant haplotypes can be helpful in dissecting a disease-95 associated genomic region. To this end, we used the currently largest dataset on AMD 96 genetics including over 33,000 genotyped individuals (Fritsche et al. 2016) and analyzed the rare, but informative recombinant haplotypes on 10q26 to define a minimal 98 set of variants associated with AMD.

101
The study followed the tenets of the Declaration of Helsinki and was approved by the 102 local Ethics Review Board at participating sites, as previously described (Fritsche et al.  best-guess imputed haplotypes, which can be outputted using the -haps command. We 121 used these haplotypes for further haplotype analyses. In addition, the imputed genotypes were coded as dosage data ranging from 0 to 2 for single variant association 123 testing as well as mediation analyses. In total, 3.446 variants were either genotyped or 124 could be imputed reliably in the ARMS2/HTRA1 region.

125
Several DNA specimens had been genotyped after whole genome amplification (WGA).

126
To account for possible confounding effects of WGA, a categorical variable "WGA" was

136
We also aimed to exclude the possibility that the main signal represented by rs3750846 137 was explained by two causal variants where the risk carrying alleles are inherited 138 together with one of the rs3750846 alleles. In such a scenario the two causal variants 139 would represent two or more haplotypes, which would be tagged by one of the alleles of capturing the variants with the strongest evidence for association. The best-guess 163 haplotypes defined by these variants (i.e. haplotypes that carry alleles of these variants) 164 were investigated for their association with AMD using multivariable logistic regression 165 models including all haplotypes with reasonable counts (≥ 34) in the study sample 166 except the non-risk allele carrying haplotype (H0), which served as reference. We To determine haplotype phase, the calculation of several plausible haplotypes for each 176 individual is required. This process is repeated many times and eventually returns the 177 best guess haplotypes for each individual. It thus is possible that the estimated best 178 guess haplotypes are only slightly more likely than other haplotypes and that random 179 effects may come into play. In order to account for this uncertainty, we repeated the  Table 1). This resulted in 13 207 haplotypes that were counted at least 34 times in the study sample and thus resulted in 208 a haplotype frequency ≥ 0.05% (Figure 1). Twelve haplotypes were included in a 209 haplotype association analysis by logistic regression modeling with the common non-210 risk allele representing haplotype H0 as reference (Figure 1). As expected, the common 211 haplotype carrying all risk increasing alleles (H12) was strongly associated with AMD (P 212 < 10 -50 ).

213
Two additional findings were of particular interest when focusing on the rare 214 recombinant haplotypes: (1) haplotypes H1 to H4 with risk increasing alleles exclusively 215 at the variants downstream of rs3750846 were not associated with AMD (P > 0.05) 216 (Figure 1), while (2) haplotypes H7 to H11 with the non-risk alleles at the variants 217 downstream of rs3750846 were highly significantly associated with AMD (P from <10 -50 218 to 6.68x10 -5 ). In addition, the first two variants upstream of ARMS2 (rs61871744 and 219 rs11200630) were also not associated with AMD, since haplotypes without the risk 220 increasing alleles at the two variants were associated with increased disease risk.

221
Together, these findings reveal that variants downstream of the lead variant rs3750846 222 do not contribute to AMD risk and that the most likely candidates driving the AMD 223 association are 13 variants in and immediatley upstream of the ARMS2 gene. A further 224 narrowing of the refined AMD-associated interval would require an even larger data set 225 due to the low rate of recombination between the local variants at and around ARMS2. in the currently largest available dataset of AMD patients and controls has now refined 242 the associated interval pointing to associated variants close to ARMS2 but excluding 243 variants near the HTRA1 locus from disease association. In particular, the two 244 synonymous variants in the first exon of HTRA1 as well as the HTRA1 promoter variant 245 rs11200638 were excluded from AMD association making it rather unlikely that HTRA1 246 plays a significant causative role in AMD pathogenesis. This is also true for AMD-247 associated HTRA1 variants rs1049331 and rs2293870, previously reported to strongly 248 influence gene transcription (Yang et al. 2006) and more importantly its ability to bind 249 insulin-like growth factor 1 or to regulate TGFß signalling (Friedrich et al. 2015). 250 Interestingly, the complex variant evs2663177 (del443ins54), which is located within the 251 3'-untranslated region of ARMS2 and which has been shown to influence stability of the 252 ARMS2 transcript (Fritsche et al. 2008), was also excluded by our analysis as being 253 AMD-associated. Therefore, a mechanism other than haploinsufficiency needs to be 254 considered as disease related involving the ARMS2 gene product. In line with this is an 255 earlier notion emphasizing that variant ARMS2:rs2736911 resulting in a truncated 256 ARMS2 protein (R38X) was never found to be associated with AMD (Friedrich et al.  In concluding, we demonstrate that genetic variants in or close to ARMS2 but not 268 HTRA1 are responsible for disease susceptibility at the 10q26 locus. This finding will 269 help to focus the functional analysis on ARMS2 and its role in AMD pathogenesis.    the risk for the disease, we concluded that this variant is not associated with AMD risk.

357
The minimal set of risk-associated variants (grey box) includes 13 variants not excluded 358 by the haplotype analysis and located exclusively at the immediate ARMS2 locus. Thin