The study was conducted in accordance with the Declaration of Helsinki. Informed consent for diagnostic testing was obtained from the proband in clinic. Genomic DNA was extracted from a peripheral blood sample by Bristol Genetics Laboratory and tested against the retinal dystrophy panel of 176 genes in the NHS accredited Genomic Diagnostics Laboratory at Manchester Centre for Genomic Medicine, United Kingdom.

We used Human Splicing Finder (Desmet et al., 2009) to identify and predict the effect of variants on splicing motifs, including the acceptor and donor splice sites, branch point and auxiliary sequences known to enhance or repress splicing. This program uses 12 different algorithms to make a comprehensive prediction of the effect of variants on splicing.

PyMol (Schrodinger Ltd.) program was used to characterize the effect of missense variants in human PRPF31 protein. Missense variants were modeled on PRPF31 in the pre-catalytic spliceosome primed for activation (PDB file 5O9Z) (Bertram et al., 2017).

Full-length, sequence-validated PRPF31 ORF clone with C-terminal myc tag was obtained from OriGene. c.341T > A or c.581C > A variant was introduced using NEB Q5 site-directed mutagenesis kit. The entire wild-type and mutant clone sequence was verified by Sanger sequencing (Source Bioscience).

HEK293 cells and 661W cells were cultured in DMEM high glucose + 10% FCS at 37°C, 5% CO₂, and split at a ratio of 1:8 once per week. hTERT-RPE1 cells (ATCC CRL-4000) were cultured in DMEM/F12 (50:50 mix) + 10% FCS at 37°C, 5% CO₂, and split at a ratio of 1:8 once per week.

The construct was transfected into HEK293 cells using PEI, and into hTERT-RPE1 and 661W cells using the Lonza Nucleofector.

Cells were grown for 72 h, and treated with 30 μg/ml cycloheximide in DMSO. Untreated cells were treated with the equivalent volume of DMSO.

Total protein was extracted from cells using 1% NP40 lysis buffer and scraping. Insoluble material was pelleted by centrifugation at 10,000 ×g. Cell fractionation was carried out by scraping cells into fractionation buffer containing 1 mM DTT, and passed through a syringe 10 times. Nuclei were pelleted at 720 ×g for 5 min and separated from the cytoplasmic supernatant. Insoluble cytoplasmic material was pelleted using centrifugation at 10,000 ×g for 5 min. Nuclei were washed, and lysed with 0.1% SDS and sonication. Insoluble nuclear material was pelleted using centrifugation at 10,000 ×g for 5 min.

20 μg of total protein per sample with 2 × SDS loading buffer was loaded onto pre-cast 4–12% Bis-Tris gels (Life Technologies) alongside Spectra Multicolor Broad range Protein ladder (Thermo Fisher). Samples were separated by electrophoresis. Protein was transferred to PVDF membrane. Membranes were incubated with blocking solution [5% (w/v) non-fat milk/PBS], and incubated with primary antibody overnight at 4°C. After washing, membranes were incubated with secondary antibody for 1 h at room temperature and exposed using 680 nm and/or 780 nm laser, or incubated with SuperSignal West Femto reagent (Pierce) and exposed using Chemiluminescence settings on Li-Cor Odyssey imaging system (Li-Cor).

Mouse anti β-actin clone AC-15. 1:4000. Sigma-Aldrich A1978.

Goat anti-PRPF31 primary antibody 1:1000 (Abnova).

Mouse anti-c myc 1:5000 (Sigma).

Mouse anti PCNA-HRP conjugated 1:1000 (Bio-Rad).

Donkey anti mouse 680 1:20,000 (Li-Cor).

Donkey anti goat 800 1:20,000 (Li-Cor).

Cells were fixed 24, 48, and 72 h after nucleofection. Cells were fixed in ice-cold methanol at -20°C for 5 min, immediately washed with PBS, and incubated with blocking solution (1% w/v non-fat milk powder/PBS). Coverslips were incubated with primary antibodies at 4°C overnight and with secondary antibodies and DAPI for 1 h at room temperature. Cells were mounted onto slides with Mowiol.

Goat anti-PRPF31 primary antibody 1:200 (Abnova).

Mouse anti-c myc 1:1000 (Sigma).

Rabbit anti-caspase 3 1:500 (Abcam).

Donkey anti mouse IgG Alexa Fluor 488 1:500.

Donkey anti goat IgG Alexa Fluor 633 1:500.

Goat anti rabbit IgG Alexa Fluor 488 1:1000.

Goat anti mouse IgG Alexa Fluor 568 1:1000.

Confocal images were obtained at the Centre for Research in Biosciences Imaging Facility at UWE Bristol, using a HC PL APO 63×/1.40 oil objective CS2 lens on a Leica DMi8 inverted epifluorescence microscope, attached to a Leica SP8 AOBS laser scanning confocal microscope, or in the Bioimaging Unit of University of Southampton, using a HC PL APO 63×/1.30 glycerol objective lens on a Leica DMi8 inverted epifluorescence microscope, attached to a Leica SP5 laser scanning confocal microscope. For publication, images were captured using LASX or LAS Af software, assembled in Adobe Photoshop, and figures prepared using Adobe Illustrator. For image analysis, images were captured using LASX software, assembled in Adobe Photoshop and randomized for analysis by an independent researcher blinded to the identity of samples.

A 39 years old female presented to the Genetic Eye clinic at Bristol Eye Hospital in 2013 complaining of some difficulty with dark adaptation, driving at night and a reduction in her field of vision (having to turn her head to see her children). Her general health was otherwise good. Over a 4 years period her best corrected visual acuity remained good at 6/6-3 right eye and 6/7.5 left eye (Snellen equivalent using a LogMar chart) whilst her peripheral vision deteriorated from an isolated mid-peripheral scotoma to tunnel vision by 2017 (Figure 1A). Fundoscopy showed widespread bilateral bone spicule pigmentation, attenuated retinal vessels and pale optic nerves typical of RP (Figure 1B). There was no evidence of lens opacities or macula oedema in either eye.

The patient described other family members having similar symptoms and losing their sight at a relatively young age (Figure 2A). A heterozygous PRPF31 change, c.341T > A p.Ile114Asn was identified in the patient and her asymptomatic father, which was confirmed by bidirectional Sanger sequencing. All affected members of the family were on the father’s side. We were not able to contact any other affected relatives for testing. Pathogenic variants in PRPF31 are associated with a form of RP which shows incomplete penetrance, consistent with the pattern of inheritance seen in this family. The PRPF31 c.341T > A p.Ile114Asn variant is not present in the heterozygous or homozygous state in any individuals within the gnomAD database, nor are any other variants affecting Ile114, suggesting that this is a highly conserved residue. Analysis by PolyPhen2 suggested this change was probably damaging, with a score of 0.963 (Figure 2B) and SIFT concurred with this prediction with a score of 0.0. Comparative genomic alignment shows the residue to be conserved from humans to amphibia, within a highly conserved region, conserved across diverse metazoa including sponges (Figure 2B). The Grantham score (Grantham, 1974) is 149, where 0–50 is conservative, 51–100 is moderately conservative, 101–150 is moderately radical and >151 is radical (Li et al., 1984).

We undertook in silico splicing analysis of our novel variant of interest c.341T > A p. Ile114Asn and found that it was not predicted to affect splicing. We also studied the nine published variants in PRPF31 annotated as missense, and interestingly, five were predicted to potentially alter splicing, and one [c.1373A > T, p. Gln458Leu (Xiao et al., 2017)] was predicted to be highly likely to affect splicing (Table 2). This suggests that either this splice predictor should be used with caution, or that p.Gln458Leu may be mis-annotated as a missense variant, when it actually affects splicing. We suggest that this variant should be a priority for further functional characterization in vitro.

We mapped all published missense variants onto the PRPF31 protein structure in the pre-catalytic spliceosome. For simplicity, we only show PRPF31 in complex with U4 snRNA and 15.5K (SNU13) protein (Figure 3) and (in complex with PRPF6 in Supplementary Figure S1; in complex with PRPF8 in Supplementary Figure S2). This showed that variants are located throughout the protein, but concentrated in several key domains. Three variants (Arg288Trp, Ala291Pro, and Cys299Arg), are located in α-helix 12 of the protein, in the Nop domain which interacts with RNA and the 15.5K (SNU13) protein. Three variants are in α-helix 6 of the coiled-coil domain (Ala194Glu, Leu197Pro, and Ala216Pro) and one variant is in α-helix 3 of the protein in the coiled-coil tip (Thr138Lys). Gly261Arg is within the flexible loop between the Nop and coiled-coil domains and Arg408Trp alone is in the C-terminal domain.

Analysis of interactions within 4 Å of each amino acid show that in most cases (Thr138Lys, Ala194Glu, Gly261Arg, Arg288Trp, Ala291Pro, and Cys299Arg), these substitutions are predicted to affect hydrogen (H) bonding in PRPF31. H bonds with donor-acceptor distances of 2.2–2.5 Å are strong and mostly covalent; 2.5–3.2 Å are moderate mostly electrostatic and 3.2–4 are weak electrostatic interactions and can be predicted to be affecting protein folding and solubility (Jeffrey, 1997). In the case of Arg408Trp, the substitution does not affect H bonding within PRPF31, but does introduce a new interaction with neighboring PRPF6 (Figure 4A and Supplementary Figures S1, S2). Gly261Arg also introduces a new interaction with neighboring PRPF8 (Figure 4B and Supplementary Figure S2). Of the three small substitutions which do not affect H bonding, we discovered that in all cases the variant amino acid was proline, which introduces a new kink in the amino acid chain. Each of these substitutions also resulted in the loss of a polar contact (Figure 4C–E).

We next mapped the variant found in our patient attending the genetics eye clinic at Bristol Eye Hospital; Ile114Asn (Figure 5A). Ile114Asn is in the coiled-coil domain of the protein, in close proximity to published pathogenic variants Thr138Lys and Ala194Glu (Figure 5B). The substitution introduces new H bonds between this residue and Ala190 of an adjacent α-helix, and is predicted to affect protein folding and solubility, and be pathogenic (Figure 5C).

The effect of missense variants is summarized in Table 3.

To test the accuracy of our predictions, we took on c.341T > A p.Ile114Asn for further in vitro characterization.

To investigate whether c.341T > A p.Ile114Asn caused mislocalization of the protein, we transfected hTERT-RPE1 cells, an immortalized cell line derived from human retinal pigment epithelium, with plasmids expressing either wild-type (WT) PRPF31 or PRPF31 341T > A, both tagged with c-myc epitope tag. We used the Lonza nucleofector to ensure high transfection efficiency. We assayed the cells after 24, 48, and 72 h by immunofluorescence confocal microscopy using an anti-cmyc antibody. At 24 h we saw mid- to high-level expression of the WT protein exclusively in the nucleus of, on average, 52.5% of cells (Figure 6A). This gradually reduced at 48 h (44.0% of cells), and 72 h (34.4%) (Figure 6A). We did not observe the same pattern in cells expressing the mutant protein. In these cells, only 9.11% of cells showed nuclear c-myc staining at 24 h, and it was very intense, and also observed in the cytoplasm (Figure 6A). This was maintained at 48 h (11.9%), but dropped to 7.5% of cells at 72 h (Figure 6A). The difference in c-myc staining, in terms of % of cells with nuclear staining and % of cells with cytoplasmic staining, was statistically significant at all time points with the exception of cytoplasmic staining at 48 h (t-test, p < 0.05, n = 4 independent biological replicates) (Figure 6B). At each time point, many cells transfected with mutant PRPF31 showed abnormal nuclear morphology, with some micronuclei present (Figure 6C). There were statistically significantly more abnormal nuclei and micronuclei in the mutant cells compared to WT at 48 and 72 h (paired t-test, p < 0.05, n = 4 independent biological replicates) (Figure 6D). To confirm these findings, we compared these findings to cells transfected with PRPF31 c.581C > A p.Ala194Glu, and observed a similar pattern of staining and nuclear changes, although we were not able to calculate the statistical significance of these observations (n = 2) (Supplementary Figure S3). We also repeated these experiments in the 661W cell line, which is derived from mouse cone photoreceptor cells (Tan et al., 2004). Although we achieved lower transfection efficiency, we observed the same pattern of c-myc staining as we saw in the hTERT-RPE1 experiments. At 24 h we saw mid- to high-level expression of the WT protein exclusively in the nucleus of around 16.2% of cells (Figure 7A). We did not observe the same pattern in cells expressing the mutant protein. In these cells, at 24 h, only around 7.9% of cells showed c-myc staining in the nucleus, and it was very intense and also throughout the cytoplasm (Figure 7A). The difference in c-myc staining, in terms of % of cells with nuclear staining, was statistically significant at all time points (t-test, p < 0.05, n = 3 independent biological replicates) (Figure 7B). Again, we saw more abnormal nuclei and micronuclei in cells transfected with mutant PRPF31 compared to WT after 48 and 72 h (Figure 7C). The number of micronuclei was significantly higher in mutant cells at 48 and 72 h (t-test, p < 0.05, n = 3 independent biological replicates) (Figure 7D). In keeping with previously published studies (Yuan et al., 2005) we hypothesized that the mutant PRPF31 protein was aggregating in the nuclei and causing cell death by apoptosis. However, co-immunostaining of cells at each timepoint with caspase-3, a marker of apoptosis, did not confirm this. We consider two alternative possible hypotheses to explain the observation of nuclear abnormalities; that expression of mutant PRPF31 has an effect on centrosomal stability, affecting separation of nuclear material in mitosis, or that expression of mutant PRPF31 causes genome instability. The first hypothesis is consistent with recent findings that PRPF31 localizes to the primary cilium’s basal body, which is derived from the centrosome (Wheway et al., 2015). The second hypothesis is consistent with the recent findings that the splicing machinery plays an important role in DNA damage response sensing, in association with the transcription machinery (Tresini et al., 2015). This will require further investigation beyond the scope of this project.

In order to investigate whether c.341T > A p.Ile114Asn affected protein stability in a similar way, we transfected HEK293 cells, a human embryonic kidney cell line which is useful for expressing protein at high levels for protein extraction experiments, with plasmids expressing either wild-type PRPF31 or PRPF31 341T > A, both tagged with c-myc epitope tag. We treated the transfected cells with cycloheximide protein translation inhibitor over a time course of 6 h, and assayed protein concentration over this period via western blotting.

Following our usual method for total protein extraction from cells using 1% NP40 detergent, we had difficulty extracting any mutant protein from the transfected cells (Figure 8A). This was despite the fact that we could observe protein expression in both cell types via immunofluorescent staining with anti-PRPF31 and anti-cmyc antibodies. We proceeded to repeat the experiment using cell fractionation, to selectively extract protein from the nuclear fraction using 0.1% SDS. This yielded a small amount of mutant protein (Figure 8B). Based on our observations, we hypothesized that the mutant protein was in the insoluble nuclear fraction. Once again, we fractionated the cells and lysed the nuclei with 0.1% SDS, but this time we did not remove the insoluble material by centrifugation, instead loading both soluble and insoluble nuclear protein on the gel. This revealed mutant protein, and confirms that the mutant protein is expressed in cells, but is insoluble (Figure 8C). No difference in protein stability was observable in the course of cycloheximide treatment (Figure 8C). Once we had optimized protein extraction from these cells, we were able to confirm our finding from immunofluorescent imaging that both the WT and mutant protein localized to the nucleus, not the cytoplasm (Figure 8D).

In summary, our findings suggest that c.341T > A p.Ile114Asn variant in PRPF31 results in protein insolubility, with downstream effects on nuclear morphology, and is likely the pathogenic cause of RP in this individual. In silico structural analysis of this variant complemented existing techniques for predicting pathogenicity of this variant.