Introduction

Familial hypercholesterolemia (FH) is an autosomal hereditary disorder that is linked to impaired low-density lipoprotein (LDL)-particle clearance that accelerates coronary artery disease (CAD) and leads to premature death.

FH remains underdiagnosed and undertreated globally. Most countries have a diagnostic rate of less than 1% [1]. Although homozygous FH presenting early clinical symptoms is rare (≤ 1:1000,000), heterozygous FH is estimated to be frequent (~1:300). However, it is often unnoticed until individuals reach their forties [2, 3]. At the time of the registry, about 40% of patients with FH were not on lipid-lowering therapy [3]. A criteria-based diagnosis like the Dutch Lipid Clinical Network (DLCN) uses a scoring system comprising multiple elements such as untreated LDL cholesterol (LDL-C) levels and family history. However, these data are not always available. Also, physical sign evaluation may be affected by selection bias. These factors increase the uncertainty of FH diagnosis.

Identification of pathogenic mutations provides a definite FH diagnosis. More than 2,100 pathogenic (or likely pathogenic) mutations in the LDL receptor gene (LDLR) have been collected in the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar). Mutations in proprotein convertase subtilisin/kexin type 9 (PCSK9) and apolipoprotein B (APOB) may also cause FH, but their presence is not as common as LDLR.

FH mutations have been considered independent CAD risk factors [4]. Among patients with severe hypercholesterolemia (LDL-C ≥ 190 mg/dL), mutation carriers have three times the risk of developing CAD [5], partly due to the tendency of a lifelong increase in the arterial deposits of LDL-C.

FH genetic testing has relied on amplification-based methods, such as next-generation sequencing (NGS) or PCR, for a long time. Due to the efficiency limits of polymerase and ligase enzymes, NGS generates short reads (usually ~150–250 bps). Because of this, it is commonly used to detect small variants in protein-coding regions [6]. Among patients classified as having "possible FH" by the Dutch lipid clinical network criteria, NGS exon sequencing reported that the FH-mutation detection yield was about 15% [7]. FH mutations are more diverse than expected. Mutations located in non-protein-coding regions, such as promoters, splicing regions, or deep introns, can also give rise to FH [8–10].

Moreover, it is difficult for PCR methods to detect large insertions/deletions (Indels). Long-range PCR enzymes can amplify up to 40 kb of genomic DNA (gDNA) [11], but primer design and reaction remain challenging for highly repetitive and GC-rich regions, such as the promoter region and exons 4, 7–9, and 13 of LDLR. Each amplicon requires specific reaction conditions for the PCR, increasing the labor requirements and level of bias. Furthermore, it is known that epigenetic factors, such as the methylation of regulatory elements, also frequently alter the expression of genes [12]. However, these features are entirely lost during amplification processes.

Long-read sequencing technologies have brought new solutions to the blind points of traditional genetic testing in recent years. The Oxford Nanopore Technology (ONT) sequencing system can generate long reads of dozens of kilobases, and detecting mutations is enabled for long targets. The platform also has sequencing protocols for PCR amplicons, but this does not take full advantage of long-read sequencing.

The CRISPR-Cas9 system, directed by a customized CRISPR ribonucleic acid (crRNA), can bind and cut double-strand DNA in a region of interest (ROI) [13, 14], which inspires an amplification-free target sequencing. The attachment of nanopore sequencing adapters requires 5’-phosphates of DNA molecules. First, the existing ends of gDNA molecules are dephosphorylated. After that, Cas9-guided cleavage on flanking sites of ROIs exposes new phosphate groups at the cutting sites. Sequencing adapters can now be added. As a result, ROIs are targeted and sequenced directly without amplification steps. This strategy can bypass the genomic complexity within the ROIs. The approach is officially suggested to enrich a single ROI < 20 kb, partly because long molecules are more easily broken during the experiment. The resulting coverage has a decreasing tendency toward the middle of long ROIs. This strategy has been used in analyzing cancer susceptibility and optimizing the genomic alignment in plants [15–17]. There are no related workflows to identify FH.

Our study successfully applied the strategy to FH diagnosis. We successfully enriched the entire LDLR (46.2 kb) and PCSK9 (26.5 kb) and detected FH mutations in patients with hypercholesterolemia. We also validated a false-positive deletion caused by a long-range PCR.

Materials and methods

Patient samples

One family case, four unrelated cases, and one control case are included in this study (Table 1). They are all Japanese. The participants were recruited through the Department of Cardiovascular Medicine of Kyoto University Hospital. The proband in the family case was a 53-year-old mother with an LDL-C level of 227 mg/dL before medication was administered. Statin treatment lowers the LDL-C level to around 170 mg/dL. Her LDL-C level dropped to 68 mg/dL after receiving a PCSK9 inhibitor. Two of her children also exhibited hypercholesterolemia. Her father and elder brother had a history of myocardial infarction. The four independent patients had high initial LDL-C levels (>190 mg/dL) before medication. The Ethics Committee of Kyoto University approved the study. All members provided written informed consent.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Lipid profile of family cases, unrelated cases and control sample.

https://doi.org/10.1371/journal.pone.0297231.t001

Results

LDLR and PCSK9 were successfully enriched

In the benchmark experiment, we used 5 μg of gDNA for library preparation, followed by 36 hours of sequencing using the crRNAs panel (refer to S2 Table). To assess the enrichment performance of LDLR/PCSK9, we defined Regions of Interest (ROIs) as the regions between the innermost crRNA cutting sites of the two genes (-623 to LDLR +1230 bp and -1231 to PCSK9 +26 bp). Among the total 64,142 aligned reads in the dataset, 245 aligned to LDLR, and 667 were assigned to PCSK9 (Fig 1A). The on-ROI ratio was calculated as 0.38% (245/64,142) for LDLR and 1.04% (667/64,142) for PCSK9. The average coverage for LDLR and PCSK9 was 106× and 420×, respectively, compared to 1.2× for the entire genome (Fig 1B), indicating adequate enrichments.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Cas9-mediated enrichment performance on ROIs.

(A) The coverage plot at LDLR/PCSK9 in the benchmark experiment. The ROI areas were referred to each gene’s innermost crRNAs cutting sites. (B) On-ROI percent was calculated using the number of aligned reads on ROIs and the number of reads of the whole genome (64,142).

https://doi.org/10.1371/journal.pone.0297231.g001

Additionally, we observed 13 defined "off-target" pileups across the dataset (refer to S4 Table). These off-target coverages ranged from 25× to 41×, significantly lower than the ROI coverages, indicating that the on-ROI enrichment was adequate.

We detected annotated pathogenic (or likely pathogenic) mutations in other samples, which will be reported in the following results. We also calculated their LDLR/PCSK9 coverage (see S5 Table and S2 Fig).

Continuous reads rule out a false-positive 670 bp deletion due to long-range PCR error

Amongst the on-ROI reads obtained in the benchmark experiment, continuous reads spanning LDLR and PCSK9 were observed in 21% (52/245) and 48% (307/667), respectively (Fig 2A). These uninterrupted reads are highly useful for detecting SVs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Continuous long reads generated by this strategy rule out a false-positive 670 bp deletion due to PCR error.

(A) Coverage on ROIs of full-length reads spanning the whole LDLR and PCSK9. (B) The continuous long read sequences proved that the 670 bp deletion within the LDLR gene was a false positive result caused by PCR amplification error. We used the 2.5 kb DNA ladder (Takara, Cat # 3413A) as a DNA molecular size marker in agarose gel electrophoresis. The band of biased amplicons (~8.9 kb) of part three was challenging to distinguish from the band of complete amplicons (~9.5 kb). It was only noticeable by shorter amplicons and a smaller-scale DNA ladder. With the continuous long reads generated by Cas9-guided nanopore sequencing, it was easy to see no deletion in the relevant region.

https://doi.org/10.1371/journal.pone.0297231.g002

Before establishing this Cas9-guided sequencing workflow, we attempted to detect genetic variants in LDLR through long-range PCR. We divided LDLR into five segments and amplified each part using long-range PCR. Subsequently, we sequenced the amplicons of these five segments to identify mutations (PCR primers and reaction conditions can be found in the S6 Table).

However, even with high-fidelity PCR enzymes claimed to have high amplification efficiency, the high GC% or low complexity regions within LDLR still cause amplification troubles. Previous PCR experiments involving other samples reported a 670 bp deletion (chr19:11107881–11108550). This variation appeared in multiple unrelated samples, making it most likely a false positive finding. In this study, we performed the same PCR amplification on the control sample used in the benchmark experiment. It also exhibited this 670 bp deletion. The deletion could not be distinguished from the 9.5 kb PCR amplicon but was noticeable through agarose gel electrophoresis in a shorter amplicon. The problematic site is located in a 295 bp CpG island in intron six of LDLR. Those original continuous reads bypass the influence of difficult regions within the gene on PCR amplification, clearly indicating that the 670 bp deletion does not exist at that reported site (Fig 2B).

A heritable FH mutation was detected in a family case

We performed sequencing and variant screening of LDLR and PCSK9 in a 53-year-old female patient with hypercholesterolemia (Fig 3A). A reported mutation [24] was identified. It was c.1297G>C: p. (Asp433His) in LDLR exon nine (Fig 3B). We screened her three children (Fig 3A) using Sanger Sequencing to validate the inheritance of that mutation. The proband’s 19-year-old son and 12-year-old daughter were found with the same mutation. The 16-year-old son, without hypercholesterolemia, was free of this mutation. The mutation was confirmed via Sanger sequencing (Fig 3C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. FH-causative mutations were detected in the LDLR.

(A) Pedigree of a family with hereditary hypercholesterolemia. The proband case (mother) and two of her three children (19-year-old son, 12-year-old daughter) exhibited hypercholesterolemia. The proband’s father and elder brother had a myocardial infarction history. (B) The mutation c.1297G>C: p. (Asp433His) is located in exon 9 of the LDLR gene. Visualization tool IGV showed that both strands supported the mutation. (C) Sanger sequencing confirmed this mutation happened in the proband patient and the two children with hypercholesterolemia. The other 16-year-old son, without hypercholesterolemia, was free of this mutation. (D) Via the same variant filtering strategy, three heterozygous SNVs were detected in other unrelated cases. Those variants were supported by both strands and were confirmed by Sanger sequencing.

https://doi.org/10.1371/journal.pone.0297231.g003

Pathogenic SNVs were found in another three cases

FH-causative SNVs were found in three unrelated patients with hypercholesterolemia (Table 1). All the mutations were previously reported in other studies and classified as pathogenic or likely pathogenic (Table 2). One was a missense mutation located in LDLR exon 7. Mutations were found in introns 12 and 14 of the LDLR (Fig 3D). Both of them were located at splice donor sites.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. FH-causative SNVs were detected in three unrelated cases with hypercholesterolemia.

https://doi.org/10.1371/journal.pone.0297231.t002

An 11,029 bp novel deletion was identified

A heterozygous novel deletion was found in the LDLR of patient #04. The continuous long reads showed that the deletion was located in chromosome 19: 11093597–11104625, skipping exons 2–3 (Fig 4A). We confirmed the deletion through an in-fusion cloning experiment (Fig 4B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. An 11,029 bp deletion in LDLR was identified.

(A) Continuous long reads clearly show the deletion on IGV. By amplifying the flanking region of the deletion, a 624 bp short amplicon containing the breaking point was confirmed by gel electrophoresis. (B) The deletion was further confirmed on the sequence level.

https://doi.org/10.1371/journal.pone.0297231.g004

Discussion

Identification of causative variants is crucial for precision medicine in FH patients and their families. This study integrated the CRISPR-Cas9 system with nanopore sequencing technologies to address clinical demands. We developed novel crRNA panels (S1 and S2 Tables) and successfully achieved targeted enrichment of LDLR and PCSK9 without amplification (Fig 1). Through this approach, we identified multiple mutations of various types and sizes. The workflow can be easily implemented in a small laboratory using a portable sequencer and laptop computer, allowing completion by one person within 1.5 to 2 days (including 2 hours for DNA extraction, 3 hours for library preparation, and 24 to 36 hours for sequencing). Please refer to the S7 Table for details.

In the benchmark experiment, the on-ROI read ratios of LDLR and PCSK9 were 0.38% and 1.04%, which might appear low. The on-ROI ratios were calculated with the entire genomic DNA background, while the background DNA molecules were not removed from the targets during the purification step. Commonly, the coverage required for variant calling from Nanopore sequencing is several tens-fold [39]. The average coverages at LDLR and PCSK9 were 106x and 420x (Fig 2B). As for the continuous reads, the coverages were 52x and 307x, respectively (Fig 2A). Thus, we considered the enrichment performance to be generally satisfactory. As for PCSK9, the enrichment performance even exceeded our expectations. We need to sample down reads to save unnecessary computational power during variant calling.

The coverage of LDLR was lower than PCSK9 (Fig 1, S2 Fig). This observation is consistent with the benchmark experiment results. There are several possible reasons for this. ONT recommends targeting lengths not exceeding 20 kb due to the propensity of long DNA molecules to break. Otherwise, it will result in a decreasing read count towards the center of the target region. LDLR length surpasses the recommendation by more than two-fold. Considering the inherent advantages of long-read sequencing in detecting SVs, we were challenged to cut the LDLR and PCSK9 only at both sides for uninterrupted gene enrichment. As expected, long targets have decreased coverage toward the middle of the ROIs (Fig 1A, S1, S2 Figs). Another possible reason is that the repetitive regions upstream of LDLR restrict crRNA selection. After the benchmark experiment, we conducted sequencing for Family-mother and Individual 1 and discovered that pre-mixing the 12 crRNAs could improve experimental efficiency. Subsequently, we purchased the same crRNAs and divided them into separate portions to avoid repeated freeze-thaw cycles. This approach effectively enhanced enrichment efficiency for subsequent samples. Similar improvements can be observed in PCSK9 data as well. More notably, we found that some Cas9 complexes for LDLR targeted more reads of opposite directions than expected, causing a sharp decrease in coverage between the innermost cutting sites. The Cas9 complexes may have caused unintended cleavage at nearby off-target locations, resulting in the retention of reads from opposite directions. Structural complexities in the LDLR, such as hairpin loops or secondary structures, can also hinder efficient cleavage by the Cas9-complexes. This is a reminder that although all candidate crRNAs for LDLR (with one exception) scored > 50 for both on- and off-target performance, the actual implementation can still vary. CrRNAs for PCSK9 generally scored higher for on-target predictions and had better enrichment performance, indicating that a high on-target score is more important in enrichment success.

Our study defined "off-targets" as "pileups occurring outside of ROIs with a distance greater than 1,000 bp and coverage exceeding 25x". We identified 13 such "off-targets" in the benchmark experiment. The purpose of presenting their locations and coverage alongside the enrichment results of on-ROI samples was to facilitate comparison. However, since our research does not involve gene editing, off-target reads outside the ROIs are not particularly problematic for mutation detection. Analysis tools can easily exclude them.

We successfully ruled out a tricky PCR error (Fig 2B). Previously, we sequenced LDLR by long-range PCR amplicons with five primer pairs. In long-range PCR, the Takara PrimeSTAR GXL DNA Polymerase was considered highly accurate in high GC% regions [11]. However, designing long-range PCR primers and ensuring amplification accuracy for LDLR remains a formidable challenge. The 670 bp deletion reported in our study serves as such an example. The deletion occurred in many samples under the same PCR conditions. Thus, it is considered highly possible to be a false positive.

The control sample used in the benchmark experiment also showed this PCR error under the same conditions mentioned in the S6 Table. The continuous long reads generated by our strategy were directly sequenced without any amplification. It is quickly determined that the deletion does not exist.

Regarding mutation identification, we successfully identified a hereditary pathogenic SNV (c.1297G>C) within a family case. This mutation occurs in the epidermal growth factor (EGF) precursor homology domain of LDLR and induces an alteration in amino acids [24]. The mutation leads to impaired maturation and accelerated degradation of the mature LDLR protein. The LDLR EGF-precursor homology domain, spanning exons 7–14 and consisting of 411 amino acids, plays a crucial role in lipoprotein release and stabilization of the LDL receptor [25–28]. Despite inadequate response to conventional statin therapy, the proband patient exhibited sensitivity towards PCSK9 inhibitors which partially mitigated PCSK9-induced degradation of LDLR.

We applied the workflow to more cases and found FH-causative SNVs in another three unrelated hypercholesterolemic patients. Previous studies show that intronic mutations in the LDLR could also result in the FH phenotype. Most of these mutations are located within 20 bp of exon-intron boundaries, and extensive screening of deeper introns is not usually performed [29]. However, intron mutations that locate hundreds or even thousands of bps away from exon-intron boundaries have also been reported to cause disease phenotypes [8, 30, 31], which indicates the need to expand the sequencing for patients whose genetic cause of FH remains unknown. The length of LDLR is approximately 46.2 kb, which exceeds the amplification capacity of conventional long-range PCR polymerases (<40 kb) [11]. However, using our method, we were able to enrich reads covering the entire LDLR gene, including intronic regions that are far from the exons. This provides a convenient means for screening for the aforementioned intronic variations. Also, regardless of the complexity of the genomic region, our variant filtering strategy enables filtering for different variants, including SNPs, SNVs, Indels, SVs, etc. Of course, our research still has limitations.

Nanopore long reads are capable to identify mutations of various sizes [32–35], as shown in our results. However, due to the limited sample size, we reported only SNVs, one large deletion this time. For the annotated putative pathogenic genes, the recall rate was 100%. Unfortunately, we are currently unable to provide precise estimates for the method’s large-scale application. To address this limitation, we plan to expand our sample size and gather further evidence through ongoing research.

In this study, we aimed to establish the Cas9-guided amplification-free workflow and accumulating evidence from real-world cases, so only included LDLR and PCSK9 this time. Other relevant genes including APOB and APOE also contribute to the FH phenotype. By following the similar crRNA designing principle, more genes can be added to the enrichment panel and targeted simultaneously. We plan to expand the current gene panel in future work.

Currently, our primary focus is on identifying known pathogenic or likely pathogenic variations that are already documented in public databases. For unreported variations, we use genetic position annotation as a preliminary screening method, considering factors such as their presence in or proximity to protein-coding regions or regulatory regions. Experimental validation is then required to confirm the pathogenicity of these variations. In other words, our current analysis workflow is particularly efficient in confirming reported mutations and visually confirming SVs rather than identifying de novo small-scale variations.

Our method has low initial costs and minimal equipment and labor requirements, making it highly suitable for small-scale laboratories. However, the cost per sample is still higher compared to PCR or NGS. The current workflow uses one ONT flow cell per sample, which costs approximately $600, including consumables. The average cost of materials for a short-read whole-genome sequence (WGS) is approximately $1,000. ONT has already released 24 extended barcodes that can sequence multiple samples in a single experiment, but they mainly support amplification-based sequencings. We are currently working on barcoding sequencing using different patterns of DNA sequences as markers and attempting to separate the data of different samples based on this during the analysis stage. We expect to facilitate sequencing multiple samples in a single experiment, thereby reducing overall costs. Based on our experience so far, our approach is especially effective in rapidly screening suspected variations in LDLR/PCSK9 among patients displaying obvious FH phenotypes or with a family history of the condition rather than for large-scale population screening.

The long reads obtained from non-amplification DNA molecules hold great potential for uncovering new diagnostic clues for FH. For example, a study shows that FH-mutation-negative patients are characterized by the accumulation of differential DNA methylation throughout the genome [36]. Sequencing data generated through amplification-free nanopore sequencing can keep methylation information [37, 38]. Continuous reads also allow analyzing deep intronic regions, microRNA binding sites in the 3’-UTR, or exploring SNP distribution patterns.

Our current variant filtering strategy allows us to filter for variants already collected in the public databases. We plan to expand the sample size and incorporate more databases into the annotation step with regular updates. In addition, there has been a proliferation of analysis tools designed to screen for variations in long nanopore sequencing data. These include tools for calling small variants (e.g., PEPPER-Margin-Deep Variant, Medaka, Clair3) and identifying SVs (e.g., Sniffles, cuteSV). ONT also continuously updates its hardware to reduce random errors and improve sequencing accuracy in low-complexity regions. For example, the latest ONT R10.4 series flowcell has been reported to have significantly better accuracy in homopolymers [39].

In conclusion, we established a new workflow that combines the CRISPR-Cas9 cleavage system and nanopore sequencing for FH genetic testing. We sincerely hope this study serves as a starting point for future research and contributes to identifying more FH cases.

Supporting information