Using CRISPR to activate inherited retinal disease genes in human cells for diagnosis

Even in developed countries with modern molecular diagnostic equipment, up to 50% of patients with genetic diseases receive no or insufficient genetic diagnosis (). This excludes patients from established treatments and participation in clinical trials for novel therapies. ¹

High-throughput next-generation sequencing (NGS) technologies, such as whole-genome sequencing (WGS) or whole-exome sequencing (WES), have facilitated the diagnostics of genetic diseases. Both WGS and WES, however, have key limitations. WES does not cover the noncoding regions (e.g., intronic sequences apart from exon-intron junctions, promoters, or other regulatory transcriptional elements), which are crucial for mRNA stability and processing. WGS is more expensive and requires extensive data storage capacities and additional bioinformatic tools for data interpretation (,). Even if potentially disease-causing mutations can be identified in coding or noncoding regions of candidate genes using WGS, experimental validation of how their impact on the mRNA level is inevitable but laborious using currently available techniques, e.g., minigene assays () or patient-derived induced pluripotent stem cell (iPSC) models. Single-nucleotide variants can affect mRNA processing via different mechanisms. Firstly, they may alter pre-mRNA splicing (). The potential impact of variants on splicing is usually assessed with standard splice prediction software (,), but these predictions are often not reliable and may be false positive or false negative. Many false-negative splicing mutations are classified as missense variants or as silent exonic variants, especially if distant from splice junctions (,). However, such point variations may affect regulatory splice elements and disrupt splicing, and should therefore be routinely tested at the transcript level, preferably in patient cells (). ^{2 3 4 5 6 7 8 9 10}

There is an unmet need for techniques enabling straightforward detection and investigation of disease-associated mutations at the transcript level. The most convenient method to analyze the transcripts of the corresponding genes is to use patients' tissue samples. Lack of expression in tissues that can routinely be obtained from the patients (leukocytes from peripheral blood, fibroblasts from skin biopsies) represents a frequent hurdle, in particular in case of inherited retinal dystrophy (IRD) genes. The expression of some IRD genes is largely restricted to the retina, a tissue that cannot be considered for diagnostic biopsies. We therefore used CRISPR activation (CRISPRa) in acutely isolated cells (CATALYTEC) to analyze single or multiple genes associated with IRDs.

To provide proof of principle for CATALYTEC, we initially focused on the IRD genes,,, andfor several reasons: (a) These genes are frequently mutated in the corresponding diseases:, Stargardt disease (STGD1);, Leber congenital amaurosis (LCA) and retinitis pigmentosa (RP);, RP and Usher syndrome (USH). (b) Three of these genes (,, and) are very large, complicating the identification of potentially pathogenic mutations, particularly those located in noncoding regions. (c) For-associated IRD, an approved AAV gene supplementation therapy (voretigene neparvovec) is available. Improved genetic testing could help to assess pathogenicity of novelvariants and thus qualify affected patients for treatment. ABCA4 RPE65 MYO7A USH2A ABCA4 RPE65 USH2A USH2A MYO7A ABCA4 RPE65 RPE65

Our initial gene activation experiments were performed and optimized in HEK293T cells transfected with cassettes expressing nuclease-deficient dCas9-VPR, one of the most effective CRISPRa systems (,), in combination with single guide (sg) RNAs targeting the transcriptional start site (TSS) region of the respective genes (). We were able to identify sgRNAs for efficient activation of each of these genes. Multiplexing experiments demonstrated that the combination of multiple sgRNA cassettes resulted in simultaneous activation of all genes without any substantial loss of activation efficiency (). Using RT-PCR, we could amplify all fragments covering the entire coding region and parts of the untranslated regions (, and; supplemental material available online with this article;). The identity of all bands was confirmed by Sanger sequencing. ^{11 12} Figure 1A Figure 1D Figure 1, B and C Supplemental Figure 1 https://doi.org/10.1172/jci.insight.189615DS1↗

In the first attempts to establish a robust and simple protocol for the transfection of PBMCs and skin fibroblasts from healthy individuals, we tested various commonly used transfection techniques (e.g., calcium phosphate, lipofection, and electroporation). For most of these approaches, however, we observed no or weak expression of the transfected GFP reporter gene.

Since lentiviral vectors (LVs) have been shown to efficiently transduce nondividing cells (), we investigated the efficacy of gene activation upon transduction with LVs. In our experiments, using dual LVs, each expressing GFP or dsRed, we obtained high cotransduction efficiencies for human skin fibroblasts and PBMCs (, A and B). Surprisingly, the dual LV approach for gene transcriptional activation in transduced HEK293T, fibroblasts, and PBMCs led to a relatively weak transcriptional activation of, despite numerous optimization steps (e.g., use of high multiplicity of infections [MOIs], different promoters etc.,, C–F). These results suggest that dCas9-VPR delivered via LVs is substantially less efficient in activating target genes compared with delivery by plasmid transfection. To elucidate the causes for the low activity of dCas9-VPR in cells transduced with the corresponding LVs, we generated a HEK293T cell line stably expressing dCas9-VPR through genomic integration of the corresponding cassette (). After transduction of these cells with LVs expressing only the sgRNAs targeting the individual genes, we detected a robust transcriptional activation of,, and. A similarly high activation was also achieved in a multiplexing approach, in which all 3 genes were targeted simultaneously (). Together with the results shown in, this indicates that the efficiency of transcriptional activation is not compromised by LV-mediated delivery of sgRNAs, excluding a general impact of LVs on gene activation. A possible explanation for our observation of low-level gene activation in all cell types upon LV-based administration of dCas9-VPR is that dCas9-VPR is not well tolerated by LV vectors due to its size (5.8 kb). To test this hypothesis, we used a shorter variant of this CRISPRa system, dCas9-VPRmini () (). However, this approach also did not lead to increased activation efficiency compared with full-length dCas9-VPR (). ¹³ Supplemental Figure 2 Supplemental Figure 2 Supplemental Figure 3A Supplemental Figure 3B Supplemental Figure 2 Supplemental Figure 3C ¹⁴ Supplemental Figure 3D ABCA4 ABCA4 MYO7A USH2A

From this set of unfavorable results, we concluded that the LV-based approach in combination with dCas9-VPR–mediated transcriptional activation of genes is not suitable for diagnostic purposes.

Achieving reasonable levels of transcriptional activation for genes of interest in PBMCs or skin fibroblasts from healthy individuals is a prerequisite for establishing a robust protocol for diagnostic purposes in patient cells. So far, although we initially obtained high activation efficiencies for the individual IRD genes in transfected HEK293T cells, none of the previously described approaches has reached this goal. In another attempt, we used nucleofection for transferring sgRNA and transgene plasmids into our primary target cells. Nucleofection of a GFP-encoding cassette resulted in robust expression of this reporter in both cell types (). Next, we nucleofected the dCas9-VPR expression cassette together with sgRNAs targeting,,, or. Using RT-qPCR, we observed high activation efficiencies forand. Additionally, we were able to detect the entire coding and untranslated regions using RT-PCR (, and). By comparison, we found thatis already endogenously expressed in human skin fibroblasts and in PBMCs, respectively, in sufficient amounts to be detected by RT-PCR. Nevertheless, after transcriptional activation a moderate increase inexpression was detectable via RT-qPCR (, B and C). For, we could only detect RNA corresponding to exons 1–7, 45–55, and the 3′-UTR (, D and E). To optimize transcriptional activation ofin PBMCs, we replaced dCas9-VPR with p65-HSF1-Suntag, another effective CRISPRa module (). However, similar to dCas9-VPR, p65-HSF1-Suntag led to an efficientactivation in HEK293T cells, but no expression of relevant transcripts was detectable in PBMCs (). Supplemental Figure 4A Figure 2, A and B Supplemental Figure 5 Supplemental Figure 4 Supplemental Figure 4 ¹² Supplemental Figure 6C MYO7A USH2A ABCA4 RPE65 ABCA4 RPE65 MYO7A MYO7A USH2A USH2A USH2A

As shown previously, inhibition of nonsense-mediated mRNA decay (NMD) could increase the expression of activated transcripts in patient-derived cells, such as skin fibroblasts (,). Indeed, using cycloheximide to inhibit NMD in PBMCs, we were able to detect larger parts of thetranscript covering exons 1–46 and the 3′-UTR (, A and B). ^{15 16} Supplemental Figure 6 USH2A

One possible explanation for the discrepancy in efficiency ofactivation between HEK293T cells and PBMCs would be that sgRNA target sites contain some nonpathogenic variants, such as single-nucleotide polymorphisms (SNPs), or small insertion and deletion (INDEL) sites. However, we found no INDELs or SNPs that overlapped with the sgRNA binding sites of theandgenes. Nevertheless, multiple SNPs were identified in binding regions of sgRNAs targetingand(). USH2A USH2A RPE65 ABCA4 MYO7A Supplemental Table 6

So far, we have achieved efficient activation in PBMCs for 3 of the 4 genes examined, enabling detection of the entire transcript. The fact thatcould not be efficiently activated may be due to the sheer size of the gene/transcript (>800 kb/>18.9 kb). Another explanation could be that our CATALYTEC approach is less efficient in PBMCs than in HEK293T cells. To address this, we activated 2 additional genes associated with the Usher 2 subtype:and, with the latter gene being comparable towith respect to its gene/transcript size (>600 kb/>19.5 kb). The corresponding RT-qPCR data from PBMCs show a substantial increase in gene expression for both genes. Moreover, the activation efficiency was stronger in PBMCs compared with HEK293T cells (), suggesting that neither the gene's size nor the inherent inefficiency of CATALYTEC in PBMCs was the main cause of the low and incomplete activation ofin these cells. USH2A PDZD7 ADGRV1 USH2A USH2A Supplemental Figure 7

Forand, we detected several splice variants besides the expected RT-PCR bands encoding the full-length proteins, most of which resulted in frameshifts and premature termination codons (PTCs) (). The corresponding bands could represent alternative splicing that might differ in PBMCs and in the retina. RPE65 ABCA4 Supplemental Tables 1 and 2

To test how the splice pattern observed in PBMCs compared to that of human retinal cells, we performed RT-PCR of RNA samples isolated from human donor retinas and human retinal organoids using the same primer pairs. We observed a very similar band pattern compared to the RT-PCR results from skin fibroblasts and PBMCs, suggesting no major differences in mRNA splicing ofandbetween the different cell and tissue types (). ABCA4 RPE65 Figure 2A

Although RT-PCR is a robust method for the qualitative detection of mRNA splicing defects, it is less suitable for the reliable quantification of the results. In addition, particularly for low-abundance transcripts, it is susceptible to artificial variations, manifesting during cDNA synthesis and subsequent PCR amplification. To overcome the drawbacks of RT-PCR and to validate our results with additional methods, we analyzed RNA from PBMCs of several healthy individuals using 2 next-generation RNA sequencing (RNA-Seq) techniques: short-read RNA-Seq (Illumina) and long-read RNA-Seq (PacBio Revio) (). Additionally, to compare the transcript pattern of activated genes in PBMCs to their counterparts from a more native environment, we performed long-read sequencing of RNA isolated from human retinal organoids. Figure 3A

Short-read RNA-Seq of samples with transcriptionally activatedandshowed a robust increase in gene expression for both genes in comparison to nontreated control samples (). Splice junction analysis ofdemonstrated that the level of transcriptional activation achieved with CATALYTEC is in principle sufficient to analyze the different transcript isoforms (). Simultaneous transactivation ofandand single transactivation ofled to full transcript coverage by long-read RNA-Seq, corroborating the increase in gene expression as previously determined via RT-qPCR and short-read RNA-Seq (). The respective consensus sequences, constructed by clustering similar reads and eliminating potential artifacts, were used to identify isoforms of biological relevance. Forand, several full-length isoforms were identified as consensus sequences (). In contrast, only 1 consensus sequence was obtained for, as multiple transcript reads were excluded in the filtering process. ABCA4 RPE65 ABCA4 ABCA4 RPE65 MYO7A RPE65 MYO7A ABCA4 Figure 3B Supplemental Figure 4F Supplemental Figure 8A Supplemental Figure 8B

Long-read RNA-Seq of human retinal organoids resulted in the identification of more than 500 sequence reads within thegene, of which more than 95% were classified as incomplete splice matches or novel in catalog. Only a few reads were sequenced for(= 5) and(= 19;). Sequencing of transcriptionally activated PBMCs resulted in only 1 read within thegene. Forand, long-read sequencing showed similar or higher transcript abundance in PBMCs compared with human retinal organoids (). ABCA4 RPE65 n MYO7A n ABCA4 MYO7A RPE65 Figure 3C Figure 3C

Following long-read RNA-Seq of human retinal organoids, we were able to confirm alternative splicing ofandin PBMCs, previously identified after Sanger sequencing of RT-PCR amplicons (, and). ABCA4 RPE65 Supplemental Tables 1 and 2 Figure 3D

Finally, we applied the described CATALYTEC protocol to 1 patient with confirmed biallelic-associated LCA (P1), 2 patients with the clinical phenotype of rod-cone dystrophy (RP, P2 and P3) but unclear molecular genetic diagnosis, and 3 patients with the clinical diagnosis of-associated STGD (P4–P6) (and). RPE65 ABCA4 Table 1 Figure 4

All patients, except for P3, had previously undergone molecular genetic testing, of whom only P1 and P4 had a definite molecular genetic diagnosis (compound heterozygosity), which was confirmed by segregation analysis.

P1 is a 6-year-old male who has been severely visually impaired from birth and received a genetic diagnosis of-associated LCA at the age of 5 years. He was treated with voretigene neparvovec at the age of 6 years (). P2 and P3 are 2 female siblings who first experienced impaired vision (night blindness and, in the further course, visual field restriction) in their early 20s with the clinical diagnosis of rod-cone dystrophy (RP) lacking a definite genetic diagnosis. In 2013, panel sequencing of 42 IRD genes in P2 revealed 3 heterozygous variants of unknown significance in,, and, respectively (). WES performed in 2021 did not identify any additional candidate variants. P4 is a 49-year-old female who first experienced visual deterioration at the age of 33 years representing late-onset STGD with pronounced macular atrophy and foveal sparing (). P5 is a 28-year-old female and P6 is her 31-year-old brother, both diagnosed with macular dystrophy at the ages of 19 and 21, respectively. RPE65 RPE65 CRB1 RP1L1 ¹⁷ Table 1 Figure 4

Potentially pathogenic compound heterozygous mutations inandwere found in all STGD1 patients (P4–P6) and in P1, respectively (). ABCA4 RPE65 Table 1

We performed the CATALYTEC on PBMCs of these patients and analyzed the RNA by RT-PCR. P1 carries c.11+5G>A within the extended splice donor consensus sequence of exon 1 in thegene. According to RT-PCR analysis, this mutation does not severely affect mRNA splicing at that position (and). For the second mutation affecting the canonical consensus sequence of the splice acceptor site in intron 7 (c.726-2A>T,), RT-PCR analysis revealed a clear difference in the band pattern compared with the unaffected control sample, experimentally confirming the splicing divergence. Sanger sequencing of these bands revealed partial skipping of exon 8 and complete inclusion of intron 7 (). RPE65 Figure 5B Supplemental Figure 9 Figure 5A Figure 5, B and C

The c.1937+1G>A mutation of P4 disrupts the splice donor site of exon 13 in(). A previous case study has associated this mutation with severe phenotypes, including STGD and macular degeneration (). A recent publication reported that it causes aberrant splicing by activating cryptic splice donor sites within exon 13 and intron 13, ultimately resulting in a 132 bp in-frame deletion (). This conclusion was based on data obtained with a minigene assay. Potential consequences at the protein level, such as p.[(Y.603_S646del, F647*)], have not yet been experimentally validated. For this mutation, we visually detected an increased ratio of alternative exon 15 splicing in comparison with healthy control samples in RT-PCR (, and). As described in, this alternatively splicedvariant leads to an in-frame skipping of exon 15. However, the significance of this deletion for ABCA4 protein expression, stability, and function remains unclear and is outside the focus of this study. For all other patients carrying potentially pathogenic mutations in IRD genes analyzed herein, no obvious effects on splicing were detected (). ABCA4 ABCA4 Figure 5A ^{18 19} Figure 5, B and D Supplemental Figure 9 Table 1 Figure 5B

Taken together, these results suggest that our CATALYTEC is in principle suitable for the detection and quantification of splicing mutations in patients' PMBCs.

Here, we established CATALYTEC, a robust and simple approach for CRISPRa-based activation of genes in human PBMCs and skin fibroblasts for routine diagnostic purposes.

CATALYTEC fulfills various criteria for use in diagnostics: (a) It is simple and easy to implement. (b) It may indicate mutations that have either escaped detection due to technical limitations. (c) It can be used to validate the proposed pathogenicity of detected mutations or adjust the categorization seemingly neutral changes (e.g., silent exonic variants). (d) It is without further modifications transferable to other genetic disorders where biopsy of affected tissues is not possible.

In recent studies, similar approaches were developed to activate the genes associated with various inherited diseases, including(RP) and, in human skin fibroblasts (,,). Compared with these studies, our method offers additional key advancements. We demonstrate that dCas9-VPR can effectively activate disease-associated genes in PBMCs readily isolated from patient blood samples. Skin fibroblasts have the disadvantage that their isolation requires skin punching, which is more elaborate and associated with reduced patient compliance. Therefore, genetic testing of human skin fibroblasts is less suitable for broad routine diagnostics compared to PBMCs. Additionally, the isolation and cultivation of skin fibroblasts is more time consuming. With our CATALYTEC approach, we achieved a turnaround time from blood collection to first RT-PCR results of 48–72 hours, ultimately increasing the convenience for patients. Under the conditions used herein, we could not achieve sufficient activation ofin PBMCs to cover all coding regions of the corresponding transcript. Given that this gene could be sufficiently activated in HEK293T cells in this study and recently also in human skin fibroblasts, we assume that the chromatin accessibility ofis cell type dependent. CRB1 USH2A USH2A USH2A ^{15 16 20}

We have applied CATALYTEC to several common genes associated with IRDs, a heterogeneous group of genetic diseases affecting the retina. In particular, we have proven the ability of our approach to activate large genes likeand. This is of great importance, as deep-intronic mutations in large genes are not covered by WES or, when identified by WGS, cannot be correctly interpreted in terms of their potential impact on mRNA splicing. ABCA4 MYO7A

Our data show that the level of transcriptional activation obtained with CATALYTEC is high enough to be combined with the most commonly used readout methods (RT-PCR, RT-qPCR, short- and long-read RNA-Seq). RT-PCR analysis provided the first evidence that CATALYTEC can be used to detect pathogenic splice mutations in PBMCs of IRD patients. Additionally, we show that short- and long-read RNA-Seq can be used to analyze the expression and alternative splicing of transcriptionally activated genes in PBMCs. However, a meaningful and substantial result for diagnostic purposes necessitates a more detailed analysis of isoforms, which could be achieved through higher coverage of the respective target genes.

We demonstrate that the splicing patterns of activated IRD genes in human PBMCs and fibroblasts are very similar to those observed in human retinas and retinal organoids. Yet, we cannot exclude the possibility that other activated genes in PBMCs might have different splicing patterns than in the cells in which these genes are naturally expressed. In addition, rare tissue-specific splicing events may have been missed by RT-PCR. More detailed investigations using short- and long-read sequencing of the human retina and organoid transcriptomes could provide valuable insights in this regard.

An attractive alternative to PBMCs would be an application of the CATALYTEC protocol in buccal epithelial cells, as their collection is technically even simpler and less invasive than blood sampling. It remains to be seen whether this goal can be achieved using CRISPRa or other methods for introducing DNA or RNA into these cells.

In summary, we provide a proof of concept for a CRISPRa-based approach to activate IRD genes in PBMCs in sufficient quantities. The resulting transcripts can be detected and quantified using standard methods and analyzed for structural variants and splicing defects. The CATALYTEC method is universally applicable, can be easily adapted to other target genes, and could help closing important gaps in the diagnosis of inherited (retinal) diseases.

VJW, KSH, SM, and EB designed the research studies. VJW, MJG, AR, KSH, HJB, DYO, SVDE, and ZG conducted the experiments and acquired data. TH and VJW analyzed data. FB and IM provided human retina samples. HJB and CB organized and supervised diagnostics of P2, P4, P5, and P6, and PacBio long-read sequencing experiments. VJW, MJG, TH, and EB wrote the manuscript. EB, SM, and MB acquired funding. SM and MJG supervised experiments with human patient samples. EB supervised the project. All authors contributed to the final manuscript.