Patient enrollment
The primary analysis was conducted with a case-control cohort from Iowa and confirmatory studies were conducted using cohorts from Massachusetts, New York, and Australia described below.
Iowa cohort
This study was approved by the Institutional Review Board at the University of Iowa and adhered to the tenets of the Declaration of Helsinki and the ARVO statement on human subjects. All study subjects provided written informed consent and were enrolled at the University of Iowa Department of Ophthalmology and Visual Sciences. Study subjects were evaluated for PDS with a complete eye examination including slit lamp examination, gonioscopy, and ophthalmoscopy. In some cases, infra-red iris transillumination was performed to search for subtle signs of iris transillumination defects. Patients were diagnosed with PDS if they had two of four principal features of this condition: radial iris transillumination defects, Krukenberg spindle, Scheie stripe, and moderate to heavy pigmentation of the trabecular meshwork (≥2+ pigmentation on gonioscopy) as we have previously described [23]. Control subjects were enrolled from the same ophthalmology clinics at the University of Iowa and were examined by a board-certified ophthalmologist and judged not to have glaucoma or ocular hypertension. Although control subjects were not specifically examined for the presence of PDS and they were not examined with infra-red iris transillumination, they were also not noted to have classic signs of disease (Krukenberg’s spindle, iris transillumination defects, or heavily pigmented trabecular meshwork) [26].
A total of 210 cases of PDS and 362 controls were enrolled. Self-reported race/ethnicity was available from 204 (97%) of 210 cases and from 167 (46%) of controls (Table 1). However, 9 of the 210 PDS patients in the Iowa cohort were excluded from analysis due to incomplete clinical records (7 subjects), presence of iris cysts (1 subject), or presence of a sulcus intraocular lens (1 subject). Thus, 201 PDS patients and 362 controls were available for genetic study. DNA was obtained from blood samples using standard techniques as we have previously described [27].
New York eye and ear Infirmary (NYEEI) cohort
A second cohort of patients with PDS (n = 88) from NYEEI was used to replicate results from analysis of Iowa patients and controls. Patients were judged to have PDS based on identification of Krukenberg’s spindle, iris transillumination defects, and heavily pigmented trabecular meshwork on clinical exam.
Massachusetts eye and ear (MEE) cohort
A third cohort of patients with PDS (n = 150) and control subjects from MEE (n = 1500) was also used to replicate our analysis of Iowa patients and controls. The diagnostic criteria for the MEE cohorts and their clinical features have been previously described [12].
Australian cohort
A fourth cohort of patients with PDS (n = 177) and control subjects (n = 145) from the Flinders Medical Center in Adelaide, Australia was also used for replication studies. Enrollment criteria for this cohort has also been previously described [12].
DNA sequencing
Whole exome sequencing of the Iowa cohort was performed on PDS patient and control subject DNA at the same time using the same capture system as previously described [28] in collaboration with Regeneron, Inc. DNA was fragmented using sonication. Library preparation and sample bar-coding was accomplished with KAPA reagents (KAPA Biosystems) and exome capture was conducted using SeqCap VCRome probes (Nimblegen). Paired-end sequencing was performed on an Illumina HiSeq 2500. DNA sequence reads were aligned to the human reference genome GRCh37 using the Burrows-Wheeler aligner [29]. Sequence variants were identified using the Genome Analysis Tool Kit (GATK) [30] and a custom sequence analysis and annotation pipeline (Institute for Vision Research, Iowa City, IA). MEE cases and Australian cohorts were analyzed with exome sequencing as previously described [12, 31]. The MEE control data was extracted from a deidentified exome repository derived from individuals who had an eye exam but did not have a diagnosis of glaucoma. The cohort of PDS patients from the NYEEI was tested for mutations in the MRAP gene using standard Sanger sequencing as previously described [32].
Variant filtering and mutation analysis strategy
We excluded variants from our analysis of the Iowa cohorts if they had a variant quality of less than 50, or if fewer than 20% of the reads supported the variation. Such variations are a common source of false positives that do not validate upon Sanger sequencing. A total of 17,253 genes met criteria for whole exome analyses. We judged variants with a minor allele frequency > 2.5% in gnomAD populations (Non-Finnish European, African, and South Asian) or present in > 2.5% of our control patients to be too common to cause PDS and these variants were excluded from the analysis.
The Iowa cohort of 201 PDS patients and 362 controls in the genetic study was further filtered based on their exome sequences. Two subjects (1 PDS patient and 1 control) were excluded from analysis due to relatedness with another study subject. We also conducted a principal components analysis of the remaining of 200 PDS patients and 361 controls that had complete clinical records and exome sequence available as previously described [33]. We identified two PDS patients and 2 control patients as outliers for the first two principal components (Supplemental Figure 1). Three of the outliers were of African ancestry (two PDS patients and one control patient), while one outlier was a non-Hispanic white control patient. These four outliers were excluded from our study and our subsequent genetic studies were conducted on the exomes from the remaining 198 PDS patients and 359 controls from Iowa.
Mutation analysis and statistics
We conducted one primary analysis and three secondary analyses on the mutation data available from our cohorts of PDS patients (n = 198) and control subjects (n = 359) from Iowa.
Primary analysis
The frequency of loss-of-function mutations in each of 5 candidate genes was compared between the PDS and control groups using Fisher’s exact test. Loss-of-function mutations were defined as premature termination mutations, frameshift mutations, and canonical splice site mutations. The five top candidate genes for causing human PDS were genes that have been previously shown to cause pigment dispersion in mice (TYRP1 [10], GPNMB [10], LYST [21], DCT [19], and MITF [19]). The frequency of loss-of-function mutations in each gene was analyzed separately and a Bonferroni corrected p-value of 0.05 / 5 genes examined = 0.01 was used as a threshold for significance.
Secondary analysis
The frequency of loss-of-function mutations in two additional sets of genes were compared between the PDS and control groups in additional, secondary analyses. As secondary analyses, these investigations were conducted for hypothesis generation rather than establishing statistical significance. Consequently, we did not employ multiple measures corrections for these analyses.
Secondary analysis− 1
We identified loss-of-function mutations in a set of 21 additional genes (Supplemental Table 1). We compared the frequency of these mutations in the case cohort with the frequency in the control cohort. The frequency of loss-of-function mutations in each of the 21 genes were analyzed separately and uncorrected p-values were calculated using Fisher’s exact test. Uncorrected p-values were calculated to prioritize candidate genes identified with this hypothesis generation set of experiments.
Secondary analysis-2
We identified all non-silent mutations in all of the genes in the exome that passed quality control filtering (n = 17,253). Non-silent mutations include: missense, nonsense, frameshift, canonical spice site, and in-frame deletion/insertion mutations. After manual inspection of in-frame deletions and insertions, we determined that variants of this nature with less than 35% of the total overlapping reads should be removed from the analysis due to a high number of false positive calls with less frequently observed variations. Loss-of-function mutations with a minor allele frequency reported to be greater than 2.5% in gnomAD [34] were removed. We calculated the non-silent mutation burden in each of the 17,253 genes for the PDS cohort using the SKAT-O software package and the SKATBinary algorithm [35].
Immunohistochemical analysis of human eyes
Human donor eyes were obtained from the Iowa Lions Eye Bank (Iowa City, IA). Consent for research was obtained from the donor’s next of kin in all cases, and all experiments were performed in accordance with the Declaration of Helsinki. Eyes were fixed for two hours in 4% paraformaldehyde in 10 mM PBS (pH 7.4) within 6–8 h after death. Anterior segments from all eyes were cryoprotected by passing through a sucrose gradient before being embedded in 20% sucrose in Optimal Cutting Temperature compound (Ted Pella, Redding, CA) [36]. Sections of 7 μm thickness were collected from each sample using a Microm H505E cryostat (Waldorf, Germany) and mounted on Superfrost plus slides (Ted Pella, Redding, CA). Immunofluorescence procedures were performed as previously described [37]. Sections were blocked for 15 min using a PBS solution with 1 mg/mL bovine serum albumin. Sections were then incubated in the primary antibody solution for 1 h, followed by rinsing three times with PBS and incubation in the appropriate Alexa 488– and Alexa 546-conjugated secondary antibodies (Invitrogen, Eugene, OR) for 30 min. Sections were counterstained with 4′,6-diamidino-2-phenylindole (DAPI), washed three times for five minutes in PBS, and coverslipped with Aquamount. The polyclonal rabbit anti-human MRAP antibody (OAAB06581, Aviva Systems Biology, San Diego, CA) was used at a concentration of 9μg/mL and a mouse anti-human collagen IV monoclonal antibody (M3F7, Development Studies Hybridoma Bank, University of Iowa, Iowa City, IA) was used at a concentration of 1μg/mL. Sections were photographed using an Olympus BX41 fluorescence microscope with a SPOT RT camera.
Statistical analyses
Power calculations were conducted to evaluate our ability to detect an enrichment of loss-of-function mutations in PDS patients when compared with control subjects in five candidate genes (TYRP1, GPNMB, LYST, DCT, and MITF). With the cohort in our study (198 PDS patients and 359 controls), we had 80% power to detect a statistically significant skew in loss-of-function mutations at significance level 1% if they occur at a frequency of 3.2% or greater in the PDS population and 0% in the control population. Fisher’s exact test was used with a threshold of p < 0.05 / 5 (p < 0.01) for significance in the primary analysis. Secondary analyses were not corrected for multiple measures as they were hypothesis generating experiments. MRAP mutations were analyzed in a second cohort of 88 PDS patients from NYEEI; 150 PDS patients and 1500 controls from MEEI; and 177 PDS patients and 145 controls from Australia. This cohort had greater than 80% power to detect a statistically significant increase in the frequency of non-synonymous mutations at significance level 5% if they occurred at a frequency of 3% or greater in the PDS population and 0% in the control population. The combined data from these cohorts were analyzed using Cochran-Mantel-Haenszel test with continuity correction [38, 39].