Whole Exome Sequencing Identification of Novel Candidate Genes in Patients with Proliferative Diabetic Retinopathy

Introduction

Diabetic retinopathy (DR) is the leading cause of blindness among U.S. working-aged adults aged 20–74 years.¹ The Diabetes Control and Complications Trial (DCCT) showed the strongest factors (duration of diabetes and hemoglobin A1c) explained 11% of the risk of developing retinopathy.² Similarly, the Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR),³ a large population based study, showed that hemoglobin A1c, cholesterol and blood pressure only accounted for 10% of the risk of developing retinopathy, which suggests that other factors may influence the variation of DR. Twin studies and family studies have implicated strong genetic components in DR with heritability scores ranging from 25% to 52% for proliferative diabetic retinopathy (PDR) in either type 1 or type 2 diabetes (DM).⁴ However, previous analyses, including candidate gene and genome wide association studies (GWAS), have failed to identify genes that are reproducibly associated with DR.^5–16 This failure has been attributed to small sample size, incomplete phenotyping of patients, and lack of data on rare variants in such studies.

Whole exome sequencing (WES) of individuals at the phenotypic extremes of disease has previously been successful in identifying genetic factors in conditions for which genetic analyses of common variants have failed.¹⁷ In contrast to GWAS, which employs SNPs mostly in non-coding regions to identify common markers that are in linkage disequilibrium with the functional or causal variants, studies in individuals with extreme phenotypes have often detected rare variants in coding regions with large functional effects. WES is a highly effective approach in discovering genes underlying multifactorial diseases. Only one study employing this approach has been published thus far. It found three genes that were associated with protection from DR using a gene-based approach.¹¹

WES of extreme phenotypes is a study design conceived and used successfully by Emond et al. in identifying DCTN4 gene as a modifier of chronic Pseudomonas aeruginosa infection in patients with cystic fibrosis.¹⁸ In their study, they performed WES on 41 patients and 48 controls. The same strategy was successfully applied by the same group in discovering another two genes, CAV2 and TMC6, as modifiers of cystic fibrosis.¹⁹

Here, we hypothesized that rare or novel variants, especially the null alleles, are enriched in patients with PDR and may be involved in the pathogenesis of diabetic retinopathy. We used WES to identify rare variants of large effect in individuals at the extremes of the phenotypic spectrum of diabetic retinopathy: no DR with at least 10 years of diabetes mellitus (controls) and PDR (cases). We focused especially on frameshift, nonsense, and spicing variants at the canonical splice sites as these are expected to have severe consequences on gene expression and would have a larger functional impact on the pathogenesis of DR. After candidate genes were identified, we performed functional validation studies and investigated the RNA expression of these genes in human retinal endothelial cells (hRECs) under high glucose conditions.

Methods

Study population

The Massachusetts Eye and Ear Infirmary (MEEI), University of Mississippi Medical Center (UMMC), Dean McGee Eye Institute and the University of Oklahoma (DMEI) Institutional Review Boards approved all research involving human subjects. Written informed consent was obtained from all patients. Patients were recruited from two sources. The first is the African American Proliferative Diabetic Retinopathy Study (AAPDR), which has been previously described.^20–22 All patients from the AAPDR study had a known diagnosis of type 2 diabetes mellitus by the 2003 American Diabetes Association criteria and/or by being on anti-diabetic medication. All patients had bilateral, dilated wide-field fundus photography. Level of retinopathy was scored using the Early Treatment Diabetic Retinopathy Study (ETDRS) adaptation of the modified Airlie House classification²³ and determined by masked ophthalmologist graders. The second source of patients was from MEEI and DMEI. These patients had type 1 or type 2 diabetes mellitus and consisted of mixed ethnicities. All patients had PDR and surgical treatment with vitrectomy. For the analyses, cases were defined as patients with PDR in at least one eye. Controls were patients with no diabetic retinopathy in either eye and with at least 10 years of diagnosed diabetes.²⁰ All controls were from the AAPDR study and were all African American.

After consent was obtained, blood samples were obtained from PDR patients (n=57) and patients with no DR (n=13). DNA was extracted from whole blood and stored at −80 degrees Celsius until the sequencing was ready to be performed. Thirty-one out of the 57 cases and all 13 controls were from the AAPDR Study.²⁰ The 26 remaining cases recruited from the MEEI and DMEI were of different ethnicities. For the analyses, cases were divided into two groups: one group consisting of individuals from the AAPDR Study (AA cases)²⁰ and one group with participants from MEEI and DMEI of mixed ethnicities (ME cases). Analyses included AA cases versus controls (AA group) and ME cases versus controls (AA group). Demographic and clinical information was obtained directly as part of the AAPDR study and from the electronic medical record for MEEI and DMEI patients.

Whole-exome sequencing

Exome capture was performed using Agilent SureSelect Human All Exon V5 kit (Agilent Technologies, Santa Clara, CA) as per the manufacturer’s instructions. Paired-end sequencing (2 × 101 base pair) was performed on an Illumina (San Diego, CA) HiSeq 2000 Next-Generation Sequencing system using v3.0 SBS chemistry with flow cell lane cluster densities of ~700 – 800 K/mm² on average. One sample was loaded per flow cell lane to obtain a minimum 10× read depth across ~96% of the target regions.

Exome Data Analyses

WES data was analyzed with the MEEI Bioinformatics Center standard pipeline (based on human reference genome GRCh37), as previously described,²⁴ and updated using BWA (version 0.6.2), Samtools (version 0.1.16 and 0.1.18) and latest version of ENSEMBL, (http://www.ensembl.org/index.html), 1000 Genomes Project (http://www.1000genomes.org), Exome Variant Server (EVS) (http://evs.gs.washington.edu/EVS/), SIFT (http://sift.jcvi.org/), PolyPhen-2 (http://genetics.bwh.harvard.edu/pph2/), UK10K Project (http://www.uk10k.org) and Exome Aggregation Consortium (ExAC) (http://exac.broadinstitute.org; release 0.3). A coverage depth cutoff of 10× was then applied. Heterozygous was defined as a fraction of a variant base between 0.25–0.75 and homozygous was defined as above 0.75.

Only variants likely to alter protein function, such as missense and loss-of-function mutations, were kept for subsequent analysis. Annotations such as the phastCons score and the GERP scores were extracted from batch downloaded data files for human reference genome build hg19 from the UCSC Genome Browser. Variants were filtered further to include only those with a minor allele frequency less than 0.1% in data from the 1000 Genomes Project, EVS, and ExAC.

Cell culture and experimental protocol

hRECs (Cell Systems) were grown in endothelial cell growth medium EBM-2 and singlequots (Lonza), antibiotics (penicillin and streptomycin- Lonza), and 4% FBS (Atlanta Biologicals). Media was changed every 48 hours until cells reached 80 to 90% confluency. Cells were divided and treated under normal glucose (5 mM of D-glucose – Sigma-Aldrich) and high glucose (30 mM of D-glucose) conditions for 72 hours. After 72 hours, cells were harvested for RNA extraction.

RNA Extraction and Quantitative (Real-Time) Reverse-Transcription Polymerase Chain Reaction

Total RNA was extracted from samples using the RNeasy Mini Kit (Qiagen) following the manufacture instructions. cDNA was prepared with 900 ng RNA using the iScript cDNA synthesis kit (Bio-Rad Laboratories, Inc., Hercules, CA). This was then probed for quantitative (real-time) reverse-transcription polymerase chain reaction using Faststart Universal SYBR Green Master (Hoffmann-La Roche, Basel, Switzerland). Primers were designed using either the NCBI Gene website, or Ensembl Genome website and were purchased from Integrated DNA Technologies. Fold changes were normalized using two different housekeeping genes: HPRT1 and B2M. Forty-four candidate genes and two calibrator genes were screened under high glucose conditions. Genes regulated under high glucose conditions are listed in figure 1.

Figure 1. Expression levels of candidate genes in human retinal endothelial cells (hRECs) after 72 hours of high glucose exposure.

All candidate genes were screened under high glucose conditions and genes regulated under high glucose conditions are listed. Gene expression of VPS13B, PHF21A, NAT1, ZNF600, PKHD1L1, VEGFB were measured in hRECs following 72 hours of treatment with 30mM D-glucose. Fold changes were calculated by 2^−ΔΔC_T method using B2M and HPRT1 as housekeeping genes and cells grown in 4% FBS media with 5 mM D-glucose as control. Error bars, ±1 standard deviation. **p ≤ 0.05 significantly different than control.

Results

Table 1 shows the characteristics of study participants with PDR compared to those with no DR. Participants with PDR in both AA and ME case groups were more likely to have a shorter duration of diabetes (p < 0.001) and higher HbA1c (p = 0.001) than controls.

Table 1.

Demographics and clinical characteristics

	Cases		Controls	AA cohort	ME cohort
	AA cohort (n=31)	ME cohort (n=26)	(n=13)	p-value	p-value
Gender
Male (%)	15 (48.4)	13 (50)	5 (38.5)	0.4478	0.7342
Female (%)	16 (51.6)	13 (50)	8 (61.5)
Race
African American (%)	31 (100)	7 (26.9)	13 (100)
Caucasian (%)	0	11 (42.3)	0
Hispanic (%)	0	4 (15.4)	0
Other (%)	0	4 (15.4)	0
Diabetes
Type 1	0	8 (30.8)	0
Type 2	31 (100)	18 (69.2)	13 (100)
Age at study entry (mean, years)	54.8	50.2	62.1	0.145	0.08516
Duration of diabetes	6.4	10.4	23.9	<0.001	<0.001
Hb A1c (%)	8.6	8.3	6.8	<0.001	<0.001

WES identified 721 candidate genes with rare or novel variants in cases with PDR, which were not present in the control samples. These genes were filtered further to include only genes with greater than 2 cases in the ME cohort or greater than 3 cases in the AA cohort. Table 2 lists the variants in 16 candidate genes identified in our AA cohort using these criteria. All variants had an allele frequency of less than 1% in the 1000 genomes project, UK10K Project, and ExAC.

Table 2.

List of 16 candidate genes in African American cases of severe proliferative diabetic retinopathy

Gene	chr	pos	ref	alt	Nature	Transcript	Protein	dbSNP
						HGVSc	HGVSp
AKR1C3	10	5141639	C	T	stop_gained,splice_region_variant	NM_003739.5:c.568C>T	NP_003730.4:p.Gln190Ter	rs140580498
KIAA1751	1	1888057	A	G	splice_donor_variant	NM_001080484.1:c.2016+2T>C	-	rs191790164
CD96	3	111342600	G	A	splice_acceptor_variant	NM_005816.4:c.1181−1G>A	-	rs77738677,COSM4917117
CRIPAK	4	1388651	T	-CA	frameshift_variant	NM_175918.3:c.352_353delTC	NP_787114.2:p.Ser118ThrfsTer289	Novel
CRIPAK	4	1389433	C	A	stop_gained	NM_175918.3:c.1134C>A	NP_787114.2:p.Cys378Ter	rs145208075
RGMA	15	93616946	A	G	splice_donor_variant	NM_001166283.1:c.38+2T>C	-	rs3942115
ZNF77	19	2933649	G	C	stop_gained	NM_021217.2:c.1476C>G	NP_067040.1:p.Tyr492Ter	rs34789013,COSM4076533
MPZL3	11	118104210	T	-AC	frameshift_variant	NM_198275.1:c.645_646delTA	NP_938016.1:p.Met216GlyfsTer9	rs144871575
NLRP12	19	54299165	G	A	stop_gained	NM_001277126.1:c.3049C>T	NP_001264055.1:p.Arg1017Ter	rs35064500
FAM92A1	8	94740519	T	-AAGTA	frameshift_variant	NM_145269.3:c.866_870delAGTAA	NP_660312.2:p.Lys289AsnfsTer9	Novel
EFCAB3	17	60472551	T	C	splice_donor_variant	NM_173503.3:c.488+2T>C	-	rs73329490
HNRNPCL1	1	12907683	G	A	stop_gained	NM_001013631.1:c.460C>T	NP_001013653.1:p.Arg154Ter	rs142211889
HNRNPCL1	1	12907352	T	-C	frameshift_variant	NM_001013631.1:c.791delA	NP_001013653.1:p.Asp264ValfsTer6	rs545031916
SIGLEC11	19	50463539	A	-G	frameshift_variant	NM_001135163.1:c.600delT	NP_001128635.1:p.Arg201GlufsTer37	rs547387871
ATP12A	13	25266666	T	-CGGA	frameshift_variant	NM_001185085.1:c.1186_1189delTCGG	NP_001172014.1:p.Ser396ThrfsTer6	rs557563746
TMEM217	6	37182972	A	G	splice_donor_variant	NM_145316.3:c.*2T>C	-	rs116076202
FAM132A	1	1178848	G	A	stop_gained	NM_001014980.2:c.616C>T	NP_001014980.1:p.Gln206Ter	rs115005664
SLC5A9	1	48694627	G	A	splice_donor_variant	NM_001135181.1:c.339+1G>A	-	rs149485404
SLC5A9	1	48703520	G	A	splice_donor_variant	NM_001135181.1:c.1536+1G>A	-	rs775853981
SLC5A9	1	48697766	G	A	stop_gained	NM_001135181.1:c.915G>A	NP_001128653.1:p.Trp305Ter	rs61997217

Variant Frequency					Conservation scores
HGVSc	1KG	UK10K	ExAC	EVS	phastCons	GERP (range: −12.3–6.17)	Max	Samples
NM_003739.5:c.568C>T	0.002396	0.000132	0.000608	T=2/C=8598;T=29/C=4377;T=31/C=12975	1	2.67	73.2392	3
NM_001080484.1:c.2016+2T>C	0.000998	0	0.00043	G=0/A=8454;G=22/A=4134;G=22/A=12588	0.747	3.2	-	3
NM_005816.4:c.1181−1G>A	0.003594	0	0.001302	A=1/G=8597;A=61/G=4345;A=62/G=12942	0.94	4.02	-	3
NM_175918.3:c.352_353delTC	0	0	0.001209	-	0	−1.9	1.5807	2
NM_175918.3:c.1134C>A	0.003594	0	0.00075	A=0/C=8600;A=27/C=4379;A=27/C=12979	0.646	0.757	1.5807	1
NM_001166283.1:c.38+2T>C	0.004593	0	0.001184	G=0/A=3178;G=13/A=1359;G=13/A=4537	0.23	1.42	-	3
NM_021217.2:c.1476C>G	0.003994	0	0.001326	C=0/G=8600;C=62/G=4344;C=62/G=12944	0.039	−1.5	2.2823	3
NM_198275.1:c.645_646delTA	0.004593	0	0.001279	-	0.988	1.94	20.2794	3
NM_001277126.1:c.3049C>T	0.004193	0	0.001665	A=0/G=8600;A=79/G=4327;A=79/G=12927	0.117	−1.31	0.1836	3
NM_145269.3:c.866_870delAGTAA	0	0	0	-	0.13	3.04	18.0953	3
NM_173503.3:c.488+2T>C	0.003594	0	0.001634	C=0/T=8600;C=68/T=4338;C=68/T=12938	1	5.99	-	3
NM_001013631.1:c.460C>T	0.004593	0	0.001035	A=0/G=8594;A=17/G=4387;A=17/G=12981	0.926	−2.02	0.5411	1
NM_001013631.1:c.791delA	0.004393	0	0.000837	-	0.011	1.09	0.5411	2
NM_001135163.1:c.600delT	0.001997	0	0.000169	-	0	−6.23	0.5998	3
NM_001185085.1:c.1186_1189delTCGG	0.003794	0	0.000908	-	1	5.63	0.5259	3
NM_145316.3:c.*2T>C	0.003794	0	0.001279	G=0/A=8600;G=65/A=4341;G=65/A=12941	0.083	1.73	-	3
NM_001014980.2:c.616C>T	0.003195	0	0.000832	A=1/G=8423;A=56/G=4222;A=57/G=12645	1	3.8	0.2202	3
NM_001135181.1:c.339+1G>A	0.002796	0	0.000474	A=0/G=8600;A=20/G=4386;A=20/G=12986	1	5.51	0.8064	2
NM_001135181.1:c.1536+1G>A	0	0	3.16E-05	-	1	5.02	-	1
NM_001135181.1:c.915G>A	0.003395	0	0.001073	A=0/G=8600;A=55/G=4351;A=55/G=12951	1	6.04	0.6774	2

1KG, 1000 genomes; UK10K, UK10K Project; ExAC, Exome Aggregation Consortium; EVS, exome variant server

The 1000 Genomes database included large cohorts from various ethnicities: African, Latino, East Asian, European and South Asian. The ExAC database included sequences from African American, Latino, East Asian, Finnish, Non-Finnish European, and South Asian ethnicities. The EVS exome database included two ethnically distinct U.S. populations: African-Americans and European-Americans. The UK10K database had a majority of sequences from samples of European-descent.

One candidate gene in particular, SLC5A9, a sodium-dependent glucose transporter, had null allele variants in 5 cases in the AA cohort: two samples with rs149485404 (chr1:48694627 G>A) resulting in a +1bp 5' splice site, two samples with rs61997217 (chr1:48697766 G>A) resulting in a p.W305 protein change and one with rs775853981 (chr1:48703520 G>A) causing a +1bp 5' splice site mutation. All three variants occurred at evolutionally conserved positions (phastCons scores of 1, 1, and 1, respectively). The phastCons score, ranging from 0–1, represents the posterior probability that the base position is at its most conserved state according to phastCons's phylogenetic hidden Markov model.²⁵

In our ME cohort, we identified 28 candidate genes on the basis of their carrying null alleles in at least two subjects with PDR and none in the controls (Table 3). Twenty-three variants from 17 genes were novel or have been reported at very low frequencies by the EVS, 1000 genomes project, UK10K Project or ExAC. One variant was found in 4 cases in the ME cohort. This variant was a frameshift deletion in the ZNF600 gene (c.1934_1935del), resulting in a NP_940859.2:p.Lys645AsnfsTer9 protein change. This variant had a low allele frequency in the commonly referenced exome databases: 0.0018 (1000 genomes), 0 (UK10K), and 0.00068 (ExAC).

Table 3.

List of 28 candidate genes in Mixed Ethnicity cases of severe proliferative diabetic retinopathy

Gene	chr	pos	ref	alt	Nature	Transcript HGVSc	Protein HGVSp	dbSNP
ABCA7	19	1049426	A	-C	frameshift_variant	NM_019112.3:c.2544delC	NP_061985.2:p.Thr849ProfsTer6	Novel
ABCA7	19	1044707	G	-GGGCACCTGGT	frameshift_variant	NM_019112.3:c.1179_1189delGGGGCACCTGG	NP_061985.2:p.Leu396AlafsTer45	Novel
ABCA7	19	1063762	G	T	stop_gained	NM_019112.3:c.5851G>T	NP_061985.2:p.Glu1951Ter	Novel
ABHD17A	19	1881528	A	-AG	frameshift_variant	NM_031213.3:c.39_40delCT	NP_112490.3:p.Phe13LeufsTer27	Novel
ANO2	12	5941617	C	T	splice_donor_variant	NM_001278596.1:c.788+1G>A	-	rs200258741
ANO2	12	5842031	C	G	splice_donor_variant	NM_001278596.1:c.1449+1G>C	-	rs200963904
BPIFB6	20	31631146	C	A	stop_gained	NM_174897.2:c.1302C>A	NP_777557.1:p.Tyr434Ter	rs140595029
BPIFB6	20	31619550	C	T	stop_gained, splice_region_variant	NM_174897.2:c.97C>T	NP_777557.1:p.Gln33Ter	rs774063482
C15orf32	15	93015466	A	T	stop_gained	NM_153040.2:c.88A>T	NP_694585.1:p.Lys30Ter	rs115999940
CCDC105	19	15121689	G	-CC	frameshift_variant	NM_173482.2:c.55_56delCC	NP_775753.2:p.Pro19SerfsTer129	Novel
CDKL1	14	50799034	T	-GTTG	frameshift_variant	NM_004196.3:c.912_915delAACA	NP_004187.2:p.Thr305GlyfsTer4	Novel
CDKL1	14	50862418	C	G	splice_donor_variant	NM_004196.3:c.171+1G>C	-	rs200559651
CEP192	18	13056620	C	T	missense_variant	NM_032142.3:c.4031C>T	NP_115518.3:p.Thr1344Ile	rs144622986
COL6A5	3	130095079	G	A	splice_acceptor_variant	NM_153264.6:c.68−1G>A	-	rs142846354
CRIPAK	4	1389424	T	-CA	frameshift_variant	NM_175918.3:c.1125_1126delTC	NP_787114.2:p.Thr377ValfsTer30	Novel
CRIPAK	4	1388466	T	-CA	frameshift_variant	NM_175918.3:c.167_168delTC	NP_787114.2:p.Leu56HisfsTer351	Novel
DNHD1	11	6592054	C	T	stop_gained	NM_144666.2:c.13312C>T	NP_653267.2:p.Arg4438Ter	rs80197979
DNHD1	11	6566679	C	T	stop_gained	NM_144666.2:c.4510C>T	NP_653267.2:p.Gln1504Ter	rs536843662
DNHD1	11	6587805	T	-GTTACCCCAA	splice_region_variant,intron_variant	NM_144666.2:c.11207−12_11207−3delTGTTACCCCA	-	Novel
DNHD1	11	6579307	C	T	stop_gained	NM_144666.2:c.8782C>T	NP_653267.2:p.Arg2928Ter	rs199752008
GPATCH1	19	33579035	T	-TTA	splice_region_variant,intron_variant	NM_018025.2:c.74−5_74−3delTTT	-	Novel
HMCN1	1	186060001	G	A	stop_gained	NM_031935.2:c.9839G>A	NP_114141.2:p.Trp3280Ter	rs375502689
KIF24	9	34306386	A	-CT	frameshift_variant	NM_194313.2:c.676_677delGT	NP_919289.2:p.Val226LeufsTer18	Novel
KIF24	9	34269293	T	-TGTC	frameshift_variant	NM_194313.2:c.1402_1405delACAA	NP_919289.2:p.Thr468ArgfsTer7	Novel
LRBA	4	151936623	A	T	splice_donor_variant	NM_006726.4:c.−220+2T>A	-	Novel
LRBA	4	151765814	G	A	missense_variant	NM_006726.4:c.4457C>T	NP_006717.2:p.Ala1486Val	rs149639181
LRP8	1	53793510	C	-TG	frameshift_variant	NM_001018054.2:c.75_76delGC	NP_001018064.1:p.Gln25HisfsTer10	Novel
MSH2	2	47641558	G	-TAAA	splice_donor_variant,intron_variant	NM_000251.2:c.942+1_942+4delGTAA	-	Novel
NAT1	8	18080115	C	T	stop_gained	NM_001160175.1:c.745C>T	NP_001153647.1:p.Arg249Ter	rs5030839,CM981375
PHF21A	11	45957289	T	-G	frameshift_variant,splice_region_variant	NM_016621.3:c.1545delA	NP_057705.3:p.Glu517LysfsTer14	Novel
PKHD1L1	8	110477066	C	T	stop_gained	NM_177531.4:c.8005C>T	NP_803875.2:p.Gln2669Ter	rs72687034
PKHD1L1	8	110460558	C	G	stop_gained	NM_177531.4:c.5963C>G	NP_803875.2:p.Ser1988Ter	Novel
SLC6A13	12	368999	C	-A	frameshift_variant	NM_001243392.1:c.221delT	NP_001230321.1:p.Val74GlyfsTer32	Novel
SLURP1	8	143822563	A	G	stop_lost	NM_020427.2:c.310T>C	NP_065160.1:p.Ter104ArgextTer16	rs62636565
TTC22	1	55247197	C	A	stop_gained	NM_001114108.1:c.1429G>T	NP_001107580.1:p.Glu477Ter	rs61733131
TTC22	1	55251251	G	A	stop_gained	NM_017904.3:c.1087C>T	NP_060374.2:p.Arg363Ter	rs112886857
UPK3A	22	45683240	C	A	stop_gained	NM_006953.3:c.396C>A	NP_008884.1:p.Tyr132Ter	rs138640270
UPK3A	22	45685025	G	A	splice_donor_variant	NM_006953.3:c.571+1G>A	-	rs145723454,COSM5859557
VPS13B	8	100844595	A	-G	splice_acceptor_variant	NM_017890.4:c.9406−1delG	-	Novel
ZDHHC11B	5	710709	C	-TG	3_prime_UTR_variant	XM_003118532.5:c.*1695_*1696delAG	-	Novel
ZDHHC11	5	840725	C	-AG	frameshift_variant	NM_024786.2:c.668_669delTG	NP_079062.1:p.Phe224ProfsTer46	Novel
ZNF600	19	53269073	C	-TT	frameshift_variant	NM_198457.2:c.1935_1936delAG	NP_940859.2:p.Lys645AsnfsTer9	Novel

Variant Frequency					Conservation scores
HGVSc	1KG	UK10K	ExAC	EVS	phastCons	GERP (range: −12.3–6.17)	Max	Samples
NM_019112.3:c.2544delC	0	0	5.62E-05	0	1	2.99	25.0107	1
NM_019112.3:c.1179_1189delGGGGCACCTGG	0.002196	0	0.000437	0	0.043	0.601	47.8482	1
NM_019112.3:c.5851G>T	0	0	0	0	1	3.52	45.96	1
NM_031213.3:c.39_40delCT	0	0	4.09E-05	0	1	3.69	6.9166	2
NM_001278596.1:c.788+1G>A	0.000799	0	0.000223	T=0/C=8366;T=13/C=3889;T=13/C=12255	1	5.72	-	1
NM_001278596.1:c.1449+1G>C	0.0002	0	0.00012	G=0/C=8394;G=8/C=4088;G=8/C=12482	1	4.76	-	1
NM_174897.2:c.1302C>A	0.002396	0	0.000852	A=1/C=8599;A=36/C=4370;A=37/C=12969	0.923	−1.93	0	1
NM_174897.2:c.97C>T	0	0	2.37E-05	0	0.199	4.33	0	1
NM_153040.2:c.88A>T	0.001997	0	0.000616	T=0/A=8596;T=25/A=4371;T=25/A=12967	0.004	−0.477	0	2
NM_173482.2:c.55_56delCC	0.000799	0.001851	0.000449	0	0.053	2.77	0.0052	2
NM_004196.3:c.912_915delAACA	0	0.000264	8.68E-05	0	0	−2.4	6.1617	1
NM_004196.3:c.171+1G>C	0.0002	0.000264	0.000245	G=2/C=8598;G=1/C=4405;G=3/C=13003	1	4.21	-	1
NM_032142.3:c.4031C>T	0.000998	0	0.000395	T=0/C=8600;T=17/C=4389;T=17/C=12989	0.292	3.22	4.438	2
NM_153264.6:c.68−1G>A	0.001398	0	0.000385	0	0.017	4.13	-	1
NM_175918.3:c.1125_1126delTC	0.004593	0	0	0	0.185	−1.1	1.5807	1
NM_175918.3:c.167_168delTC	0	0	0.000428	0	0.001	−2.22	1.5807	1
NM_014675.3:c.3618C>T	0.004393	0	0.001302	T=0/C=8552;T=60/C=4284;T=60/C=12836	0.97	−5.31	22.5101	1
NM_144666.2:c.4510C>T	0.0002	0	6.26E-05	0	0	−3.18	0.6055	1
NM_144666.2:c.11207−12_11207−3delTGTTACCCCA	0	0	0.000088	0	0.001	3.07	1.0815	1
NM_144666.2:c.8782C>T	0.000799	0.000661	0.000655	T=2/C=3180;T=1/C=1383;T=3/C=4563	0.015	−2.28	0.7669	1
NM_018025.2:c.74−5_74−3delTTT	0	0	0	0	0.814	0.368	-	2
NM_031935.2:c.9839G>A	0	0	2.37E-05	A=1/G=8597;A=0/G=4406;A=1/G=13003	1	5.98	2.3114	2
NM_194313.2:c.676_677delGT	0	0	0.000032	0	0.996	4.58	0.5141	1
NM_194313.2:c.1402_1405delACAA	0.000799	0	0.001889	0	1	5.42	1.2178	1
NM_006726.4:c.−220+2T>A	0	0	0	0	1	3.77	-	1
NM_006726.4:c.4457C>T	0.000399	0	8.68E-05	A=1/G=8599;A=0/G=4406;A=1/G=13005	1	5.19	19.9326	1
NM_001018054.2:c.75_76delGC	0	0	0.003842	0	0.298	2.53	0.257	2
NM_000251.2:c.942+1_942+4delGTAA	0	0	0	0	1	4.65	-	2
NM_001160175.1:c.745C>T	0.002796	0.002116	0.00266	T=27/C=8573;T=9/C=4397;T=36/C=12970	0	−0.582	1.1163	2
NM_016621.3:c.1545delA	0	0	0	0	0.973	−4.14	6.0199	2
NM_177531.4:c.8005C>T	0.000799	0.001984	0.001353	T=17/C=8201;T=0/C=3728;T=17/C=11929	1	5.78	0.21	1
NM_177531.4:c.5963C>G	0	0	0	0	0.995	2.49	0.1821	1
NM_001243392.1:c.221delT	0.001797	0.003174	0.003709	0	0.136	4.33	6.1782	2
NM_020427.2:c.310T>C	0.001198	0	0.000499	G=0/A=8598;G=22/A=4384;G=22/A=12982	0	−1.35	1.4982	3
NM_001114108.1:c.1429G>T	0.001597	0	0.00024	0	0.995	3.93	5.2207	1
NM_017904.3:c.1087C>T	0.002796	0.000132	0.00027	0	0.016	−0.397	0.3699	1
NM_006953.3:c.396C>A	0	0	7.90E-06	A=0/C=8600;A=1/C=4405;A=1/C=13005	1	2.26	0	1
NM_006953.3:c.571+1G>A	0.003594	0	0.001294	A=1/G=8599;A=50/G=4356;A=51/G=12955	0.999	4.68	-	1
NM_017890.4:c.9406−1delG	0	0	0	0	0.998	5.55	-	2
XM_003118532.5:c.*1695_*1696delAG	0	0	0	0	0.061	0	7.7216	1
NM_024786.2:c.668_669delTG	0	0	0.000111	0	0	−1.99	5.5912	1
NM_198457.2:c.1935_1936delAG	0.001797	0	0.000679	0	0.008	−2.23	1.9314	4

1KG, 1000 genomes; UK10K, UK10K Project; ExAC, Exome Aggregation Consortium; EVS, exome variant server

We also sought to determine if there were rare or novel variants in two well-studied genes believed to be important in the pathogenesis of diabetes or DR: vascular endothelial growth factor (VEGFB) and apolipoprotein B (apoB). VEGF has several members including VEGF-A, VEGF-B, VEGF-C, VEGF-D and placental growth factor (PlGF). Decreased VEGF-B signaling in rodent models of type 2 diabetes has been found to restore insulin sensitivity. In addition, VEGF-B levels have been found to be significantly correlated with diabetic retinopathy.^26–28 Three nonsynonymous variants in VEGFB were detected in 4 PDR patients (3 AA cases, 1 ME case) as shown in Table 4. One variant was novel, (chr11:64002973 G>A). ApoB is the main component of low-density lipoprotein cholesterol (LDL-C) and apoB100 has been found to be associated with diabetic retinopathy.^{29, 30} Eight non-synonymous variants were identified with four variants that were novel (table 5).

Table 4.

VEGFB variants in severe PDR cases

Gene	chr	pos	ref	alt	RS ID	Amino acid (protein)
VEGFB	11	64004924	G	A	rs138325963	p.R148H
	11	64005108	C	T	rs61384522	p.L176F
	11	64002973	G	A	novel	+1bp 5' splice site

Variant Frequency				Samples
1KG	UK10K	ExAC	EVS
0.00479233	0	0.001383	A=1/G=8585;A=61/G=4337;A=62/G=12922	2
0.00359425		0.0006746	T=0/C=8226;T=27/C=4185;T=27/C=12411	1
0	0.000793	1.58E-05	none	1

Table 5.

APOB variants in severe PDR cases

Gene	Variant	Amino acid (protein)
APOB	rs12713540	p.S3801T
	rs61741974	p.F3753L
	Novel	p.E3382Q
	rs72653099	p.L3076M
	Novel	p.G2540V
	Novel	p.K2110fs
	Novel	p.L1568M
	rs140877474	p.S1459G

To determine whether the candidate genes may play a role in proliferative diabetic retinopathy, we assessed 44 candidate genes for changes in expression in human retinal endothelial cells (hRECs) cultured in high or normal glucose conditions using RT-PCR. We found that the expression of six of our candidate genes including VEGFB, VPS13B, PHF21A, NAT1, ZNF600, PKHD1L1 was reduced in hRECs cultured in high glucose (Figure 1).

Discussion

We performed a WES study using an extreme phenotype design on 57 PDR patients and 13 controls without retinopathy to detect genes contributing to disease risk. We selected for rare or novel protein-truncating variants and identified a total of 44 candidate genes: 16 genes in our AA case group and 28 genes in ME case group. The final variants identified were not observed or were observed at very low frequencies in the 1000 genomes, UK10K, ExAC and EVS exome databases. In both combined cohorts, 25 novel variants in 19 genes were identified. To our knowledge, these genes have not been previously identified in other candidate gene studies or from genome-wide association studies in population cohorts in DR. Furthermore, our functional study suggests a potential role of six candidate genes in the pathogenesis of PDR.

Prior studies have identified several candidate genes in association with diabetic retinopathy.^{31, 32} In our AA cases with type 2 diabetes, one gene related to angiogenesis, TMEM217 on chromosome 6, was identified as a candidate gene. TMEM217 is a transmembrane protein involved in the MAPK pathway. One GWAS study suggested an association of two polymorphisms with DR, but none of the variants were statistically significant after Bonferroni correction.³³ One study found a 2-fold increase in gene expression of TMEM217 in vascular endothelial cells treated with DMU-212, a derivative of resveratrol that possesses potent pro-apoptotic and anti-angiogenic effects.³⁴ This suggests a potential role for TMEM217 in the signal transduction of inflammation and apoptosis pathways, which are involved in the pathogenesis of DR.³⁵

We also discovered a 2 bp frameshift mutation in ZNF600 with an allele frequency of 4/26 or 0.154 within our ME cohort. This mutation was absent in the controls and its allele frequency was very low in the commonly referenced exome databases: 0.0018 (1000 genomes), 0 (UK10K), and 0.00068 (ExAC). It is statistically unlikely (p < 0.01), therefore, to see 4/26 unrelated PDR patients to have this particular variant even after applying Bonferroni multiple testing correction for the number of candidate variants after filtering, as listed in Table 3. ZNF600 is a zinc finger protein that may be involved in transcriptional regulation. In a recent GWAS study, ZNF600 was found to be significantly associated with novel phospholipid loci.³⁶ Phospholipids are key regulators of intracellular processes and have been implicated in the pathology of type 2 diabetes.³⁷ Disruption of their metabolism has diverse metabolic consequences and may be associated with diabetic retinopathy.^37–39

Another interesting candidate gene was SLC5A9, where three rare variants were identified in five patients in the AA cohort. SLC5A9 (SGLT4) is found on chromosome 1 and is a sodium-dependent glucose transporter.⁴⁰ Hypoxia, growth factors and cytokines upregulate glucose transport in endothelial cells in diabetes.⁴¹ Although usually found in the intestine and kidney, this gene was also found to be expressed in hREC in our study (Table 2). Solute carrier family (SLC) proteins play a role in insulin secretion and several studies have looked at the association of SLC proteins with DR.^42–44 Some have found nominal associations⁴⁵ and others have found no association to DR.⁴²

One interesting candidate variant, rs115005664, found in three cases in the AA cohort was a non-synonymous mutation in the FAM132A gene (CTRP12/Adipolin). FAM132A is an adipose-derived insulin-sensitizing factor and functions as an adipokine, which are cytokines that are secreted by adipose tissue. Acute and chronic hyperinsulinemic states were found to have significantly increased circulating levels of adipolin.⁴⁶ Adipolin also has anti-inflammatory effects that exert beneficial actions on glucose metabolism.⁴⁷ Some adipokines are known to regulate (repress) endothelial angiogenesis, e.g. adiponectin⁴⁸, and may play a role in diabetic cardiovascular and metabolic complications such as DR.

We also examined two well-studied genes of biological relevance in diabetic retinopathy: VEGFB and apoB. VEGF is considered a primary promoter of the neovascularization in PDR. A number of polymorphisms in VEGFB have been analyzed either with DR or severe DR.^{16, 49–51} Our study identified three non-synonymous variants in 4 PDR patients: rs138325963, rs61384522 and one novel variant. These variants have not been associated in prior genetic studies and warrant further investigation. In addition, ApoB was also analyzed with eight non-synonymous variants identified including four novel ones. ApoB is the main component of LDL-C and is correlated with atherogenicity.^52,53

To date, only one other study has used extreme phenotype design with WES to identify candidate genes associated with diabetic retinopathy. In their study, Shtir et al. performed whole exome sequencing on 64 diabetics with diabetic retinopathy (controls) and 43 diabetics without retinopathy (cases), all of Saudi descent.¹¹ They identified three genes, NME3, LOC728699, and FASTK, whose increased rare variant burden appeared to protect against DR. We did not identify these genes in our study and this may be due to the difference in patient populations. While Shtir et al. described a cohort of Saudi descent, our study’s population were primarily Caucasian and African American.

In this study, we also examined the expression of 44 genes and found that six of these candidate genes (VEGFB, VPS13B, PHF21A, NAT1, ZNF600, PKHD1L1) are regulated by culture conditions resembling the diabetic milieu. We demonstrate that high glucose challenge significantly decreases the expression of all six genes in human retinal endothelial cells. These results suggest an important role of hyperglycemia-induced suppression of these genes in diabetic retinopathy. Although additional functional analyses are necessary, our data allow us to postulate a model whereby a diabetic milieu interacts with rare gene variation to modify the susceptibility to PDR.

We recognize several limitations of this study. Our cases of mixed ethnicities were compared to controls consisting of only African American patients. This may not provide optimal or adequately comparable results because of racial differences in allele frequencies and genetic risks for the disease. Non-Hispanic blacks, for example, are known to have significantly higher risk for diabetic retinopathy than the non-Hispanic whites.⁵⁴ A candidate rare variant might have different allele frequencies in different ethnic groups. However, we divided our cases into two cohorts in order to minimize this problem, with a threshold difference of greater than two cases between either cohort compared to controls. Many of the identified candidate rare variants this way do have different allele frequencies in the European Americans comparing to African Americans, as indicated by the EVS column in the tables. If such a candidate gene or variant is indeed associated with increased risk for diabetic retinopathy, it would therefore contribute to different amount of genetic risk in each ethnic group. It is also possible that these variants are associated with ethnicity and not disease itself. Further validation in other cohorts will be necessary to determine this. The number of controls was limited because we chose a very stringent definition of the extreme of phenotype – no diabetic retinopathy despite at least 10 years of type 2 diabetes. The limited sample size could have led to false positive results, underscoring the need for further follow-up functional characterization of these changes.

We have identified numerous candidate genes that warrant further investigation. This mode of analysis allows us to begin to understand the complex genomics of PDR, and may help to identify the pathways that contribute to the disease process in this population. The global burden of diabetic retinopathy is significant, and efforts in precision health have galvanized new ways to identify at-risk patients and spurred clinical discoveries in targeted treatments. Future in vitro and in vivo studies involving knockdown of gene targets are warranted. The variants we identified highlight the spectrum of defects potentially relevant in this population. Identifying genes that may be responsible for severity of disease may allow for the development of new therapies, alleviating significant morbidity worldwide.