In Vivo Base Editing of PCSK9 as a Therapeutic Alternative to Genome Editing

Abstract

Graphical abstract

INTRODUCTION

Naturally occurring nonsense variants in PCSK9 in African Americans result in substantially reduced blood cholesterol levels and 88% reduced risk of coronary heart disease,¹ the leading cause of death worldwide. We and others have previously used clustered regularly interspaced short palindromic repeats (CRISPR)-CRISPR-associated 9 (CRISPR-Cas9) to disrupt PCSK9 orthologs in mouse hepatocytes^2,3 and human hepatocytes⁴ in vivo, resulting in substantially reduced plasma PCSK9 protein and cholesterol levels. Standard genome editing with CRISPR-Cas9 introduces a double-strand break into DNA in a sequence-specific fashion, followed by repair of the break with either non-homologous end-joining (NHEJ)—which is error-prone and can introduce disruptive insertions or deletions (indel mutations) of varying sizes at the target site, as was done in the aforementioned PCSK9 studies—or homology-directed repair (HDR)—which can use a custom-made repair template to perform precise knock-in of a desired alteration at the target site, but only operates in proliferating cells.⁵

The most discussed safety concern with respect to CRISPR-Cas9 genome editing is the possibility of unintended mutagenesis at off-target sites with sequence similarity to the desired target site. In the aforementioned PCSK9 studies, no mutagenesis at potential off-target sites was discerned, although rare mutagenesis below the detection limit of next-generation DNA sequencing could not be ruled out.^2–4 A second safety concern arises from the unpredictable nature of the on-target indel mutations resulting from NHEJ, which can have unintended and even deleterious consequences. Furthermore, precision genome editing in adult animals has been attempted via homology-directed repair (HDR) but is relatively inefficient (a few percent of alleles at most) due to HDR being inactive in non-proliferating cells, resulting in a large burden of NHEJ-induced indel mutations instead.^6,7 In an instructive example, in vivo genome editing to precisely correct a mutant Otc gene in a mouse model of ornithine transcarbamylase deficiency paradoxically resulted in worse outcomes in adult mice.⁶ The amount of HDR-mediated gene correction was minimal, whereas NHEJ-mediated mutagenesis (which occurred in ~50% of alleles) resulted in a substantial burden of large indels that evidently disrupted the residual activity of the mutant Otc gene product.⁶

The use of base editors comprising CRISPR-Cas9 fused to a cytosine deaminase domain can result in the alteration of cytosine bases to thymine bases (C-to-T changes) at precise locations in the genome without the need for double-strand DNA breaks or DNA replication (i.e., the cells do not need to be proliferating, unlike with HDR) (Figure 1).^8–11 The most widely used base editor, “BE3” (also known as “base editor 3”),⁸ contains at its core the Streptococcus pyogenes Cas9 protein with a mutation that renders it into a “nickase” that can only make a single-strand DNA break on the strand that does not contain the standard S. pyogenes NGG protospacer-adjacent motif (PAM) sequence that anchors Cas9. The N-terminal end of Cas9 is attached to the cytosine deaminase domain from the RNA-editing enzyme APOBEC1, which ordinarily functions to edit a specific cytosine base to a uracil base in the messenger RNA encoding apolipoprotein B. By virtue of being tethered in the vicinity of DNA by CRISPR-Cas9, the cytosine deaminase domain can convert cytosine bases in the DNA molecule into uracil bases. Specifically, BE3 can edit any cytosine base that lies within a “window” 13 to 17 nucleotides upstream of the NGG PAM sequence and is contained in the same strand as the PAM sequence (but not the opposite strand) into uracil. If there are multiple cytosine bases within the window, any combination of the cytosine bases can potentially be edited.

Open in a separate window

Figure 1

Schematic of the mechanism of action of the BE3 base editor. The N-terminal end of Cas9 is attached to the cytosine deaminase domain from the RNA-editing enzyme APOBEC1. By virtue of being tethered in the vicinity of DNA by CRISPR-Cas9, the cytosine deaminase domain can convert cytosine bases in the DNA molecule into uracil bases. Specifically, BE3 can edit any cytosine base that lies within a “window” 13 to 17 nucleotides upstream of the NGG PAM sequence and is contained in the same strand as the PAM sequence (but not the opposite strand) into uracil. The C-terminal end of the Cas9 portion of BE3 is attached to a domain that inhibits the activity of uracil-DNA glycosylase. By being tethered in the vicinity of the edited cytosine bases, this inhibitor prevents uracil repair. In parallel with the cytosine deaminase activity, the Cas9 portion of BE3 nicks the opposite strand of DNA. The process of nick repair entails the removal of some bases on the DNA strand around the nick, followed by filling in of the gap using the bases on the edited strand as the complementary template. Because the edited strand has uracil bases, the opposite strand will be filled in with adenine bases rather than the guanine bases that were originally in the strand. Thus, C-G basepairs are converted to U-A basepairs. After BE3 releases from the DNA, each uracil base can be removed by uracil-DNA glycosylase, followed by filling in with a thymine base to match the complementary adenine base now on the opposite strand, rendering the base editing permanent.

Ordinarily, the cell would restore each edited cytosine base via base-excision repair of the uracil base through the action of the enzyme uracil-DNA glycosylase, followed by filling in of the gap using the guanine base on the opposite strand as the complementary template for the correct cytosine base. To prevent this, the C-terminal end of the Cas9 portion of BE3 is attached to a domain that inhibits the activity of uracil-DNA glycosylase. By being tethered in the vicinity of the edited cytosine bases, this inhibitor prevents uracil repair. In parallel with the cytosine deaminase activity, the Cas9 portion of BE3 nicks the opposite strand of DNA. The process of nick repair entails the removal of some bases on the DNA strand around the nick, followed by filling in of the gap using the bases on the edited strand as the complementary template. Because the edited strand has uracil bases, the opposite strand will be filled in with adenine bases rather than the guanine bases that were originally in the strand. Thus, C-G basepairs are converted to U-A basepairs. After BE3 releases from the DNA, each uracil base can be removed by uracil-DNA glycosylase, followed by filling in with a thymine base to match the complementary adenine base now on the opposite strand, rendering the base editing permanent.

As a proof of principle of a potentially safer alternative to NHEJ-mediated disruption of the PCSK9 gene in vivo, we sought to use base editing to specifically introduce nonsense variants into the murine Pcsk9 ortholog, analogous to the naturally occurring beneficial PCSK9 variants found in humans, in adult mice with a goal of prolonged reduction of blood cholesterol levels.

MATERIALS AND METHODS

Materials and Methods are available in the online-only Data Supplement.

RESULTS

C-to-T changes can produce nonsense codons in several ways: if targeted to the sense strand, glutamine to stop (CAG to TAG, CAA to TAA) or arginine to stop (CGA to TGA); if targeted to the antisense strand, tryptophan to stop (TGG to TAG, TGG to TGA, TGG to TAA) (Figure 2A). We made use of BE3, which recognizes the NGG PAM sequence, as well as versions of BE3 that carry Cas9 variants that alter PAM recognition (“VQR” = NGA PAM; “VRER” = PAM NGCG).¹² Each of these BE3 versions has the potential to edit any C bases within a window spanning 13 to 17 nucleotides upstream of the PAM.

Open in a separate window

Figure 2

Base editing of human PCSK9 in vitro. (A) Codons encoding glutamine, arginine, or tryptophan residues that could potentially be changed to stop codons with C-to-T or G-to-A edits. (B) On-target base editing was assessed by Sanger sequencing for various codons in human PCSK9 in HEK 293T cells using BE3, BE3-VQR, or BE3-VRER. The protospacer and PAM sequences for each target site are listed, with the bold letters corresponding to the editing “window” (13 to 17 bases upstream of the PAM) and the underline indicating the target codon (either sense or antisense), in which a C-to-T change would result in a stop codon on the sense strand. “–” indicates no base editing detected, “+” indicates modest base editing, and “++” indicates substantial base editing. (C) Top left panel: CEL-I nuclease assays were performed with genomic DNA from HEK 293T cells transfected with BE3 and a guide RNA targeting the codon encoding Q302. Arrows show the cleavage products resulting from the CEL-I nuclease assays; the intensity of the cleavage product bands relative to the uncleaved product band corresponds to the mutagenesis rate. MW = molecular weight. Top right panel and bottom panels: Sanger sequencing confirmed base editing (C-to-T) of Q302 (codon CAG) in both HEK 293T cells and induced pluripotent stem cells.

We first screened a number of potential sites of base-edited nonsense codons in the human PCSK9 gene for appropriate positioning vis-à-vis one of the three PAMs (Figure 2B). We tested for base editing activity at the qualifying sites with the appropriate BE3 version in HEK 293T cells. We observed especially efficient base editing of the C base in the first position of the CAG codon for Q302, resulting in TAG nonsense codons encoding Q302X in almost half of the PCSK9 alleles, as judged by Sanger sequencing (Figure 2C). Using the same vectors as used in the HEK 293T cells, we assessed the Q302 site in human induced pluripotent stem cells and also observed significant base editing, albeit to a lesser degree (Figure 2C). This might reflect more efficient delivery of BE3 plasmid into HEK 293T cells compared to human induced pluripotent stem cells, or it might reflect differing efficiency of BE3 in different cell types. Of note, C bases outside of the Q302 codon were edited—1, 4, and 5 bases upstream of the codon, within the expected window of base editing—but these changes are expected to be of no functional consequence in the context of gene knockout. There was differing efficiency of editing of the 4 C bases—the C in the codon and the C 1 base upstream of the codon were edited substantially more than the ones 4 and 5 bases upstream of the codon—reflecting that the cytosine deaminase domain has varied preferences for editing within the targeting window. This might be due to sequence context, i.e., which bases lie immediately upstream and downstream of each C.

We next performed the same gene-wide analysis for the mouse Pcsk9 gene in Neuro-2a (N2a) cells (Figure 3A). We observed the most base editing activity at the G bases in the second and third positions of the TGG codon for W159, with alteration of either or both of the G bases into A bases resulting in W159X alleles (Figure 3B). Of note, the wide variability of base-editing efficiency we observed at sites throughout the human and mouse PCSK9 genes is reminiscent of the considerable site-to-site variability observed with standard CRISPR-Cas9 genome editing. This highlights the need to test numerous sites in a gene if one wishes to identify those with the highest on-target efficiency.

Open in a separate window

Figure 3

Base editing of mouse Pcsk9 in vitro. (A) On-target base editing was assessed by Sanger sequencing for various codons in mouse Pcsk9 in Neuro-2a (N2a) cells using BE3, BE3-VQR, or BE3-VRER. The protospacer and PAM sequences for each target site are listed, with the bold letters corresponding to the editing “window” (13 to 17 bases upstream of the PAM) and the underline indicating the target codon (either sense of antisense), in which a C-to-T change would result in a stop codon on the coding strand. “–” indicates no base editing detected, and “+” indicates modest base editing. (B) Top panels: CEL-I nuclease assays were performed with genomic DNA from N2a cells transfected with BE3 and a guide RNA targeting various codons. Bottom panels: Sanger sequencing confirmed base editing (G-to-A and C-to-T) of two target codons (codon TGG encoding W159 and codon CAG encoding Q347) in N2a cells.

We then assessed whether efficient base editing could be undertaken in vivo at the W159 codon in Pcsk9 in adult mouse liver, the predominant site of PCSK9 production and secretion. Due to the large size of BE3, which cannot be accommodated in a single adeno-associated viral vector, we used a single adenoviral vector to deliver BE3 and the accompanying guide RNA (BE3-Pcsk9). We initially performed a pilot experiment with two 8-week-old female C57BL/6J mice and, 5 days after administration of the virus, observed substantial base editing activity in the liver but not in the lungs or heart (Figure 4A). We performed a larger experiment with 5-week-old male C57BL/6J mice that received either BE3-Pcsk9 or a control adenoviral vector (n = 5 in each group) (Figure 4B) and found that, on average, base editing of Pcsk9 resulted in 56% reduction of plasma PCSK9 protein levels and 28% reduction of plasma cholesterol levels (Figure 4C–E), as well as substantially increased hepatic low-density lipoprotein receptor levels (Figure 4G). ALT levels were within the normal range (Figure 4F), and no gross morphological differences were observed between liver sections from mice receiving the BE3-Pcsk9 or control viruses (Figure 4H).

Open in a separate window

Figure 4

Base editing of mouse Pcsk9 in vivo. (A, B) CEL-I nuclease assays performed with genomic DNA from organ samples taken from mice 5 days after receiving the BE3-Pcsk9 adenoviral vector, the control vector, or no vector. Arrows show the cleavage products resulting from the CEL-I nuclease assays. (C–F) Data from the mice in Figure 4B (n = 5 mice in each group). The fraction of base-edited alleles was determined by deep sequencing of the target site (W159 codon). The Mann–Whitney U test was performed to compare plasma analyte levels in the two groups. The bars indicate median values within each group. (G) Results of Western blot analysis of liver samples taken from the mice in Figure 4B. LDLR indicates low-density lipoprotein receptor. (H) Hematoxylin/eosin staining of liver sections from representative mice, 10× magnification. (I) Sanger sequencing of the on-target site in liver genomic DNA from the mouse with the greatest degree of base editing (the fourth BE3-Pcsk9 mouse in Figure 4B). (J) Consequences and frequencies of base-edited alleles and indel-bearing alleles in liver genomic DNA from the fourth BE3-Pcsk9 mouse in Figure 4B, determined by deep sequencing. The underlines indicate the target codon. (K) Base-editing and indel rates at on-target and off-target sites determined by deep sequencing of liver samples from 3 individual BE3-Pcsk9 mice (the values are separated by slashes) and 1 control mouse. The underlines indicate the base-editing windows within the protospacer sequences. In some cases, no base-editing rates are shown because the base-editing window contains no C bases.

With deep sequencing of genomic DNA samples from the 5-week-old mice sacrificed 5 days after treatment, we observed as much as 34% base editing of alleles in the liver, with a median rate of 25% (average rate of 24%), whereas the indel burden was far lower (~1%) (Figure 4C, I–K). The low indel mutagenesis rate with base editing contrasts with the ~40% indel rate observed in prior in vivo PCSK9 genome editing studies.^3,4 Of note, we did observe low rates of C-to-A and C-to-G changes (instead of C-to-T changes), resulting in missense variants rather than nonsense variants in some alleles (Figure 4J); on average, 22% of alleles were specifically edited to W159X. We also used deep sequencing to assess for base editing at nine potential off-target sites in BE3-Pcsk9-treated (n = 3) and control (n = 1) mice and did not observe any alteration of alleles (either C-to-T edits or indels) at rates appreciably greater than control background rates at any of the sites (Figure 4K).

In a parallel experiment, we maintained BE3-Pcsk9-treated, 5-week-old male C57BL/6J mice for 4 weeks post-treatment. Upon sacrificing mice at 4 weeks, we observed base editing of Pcsk9 at levels comparable to those observed at 5 days in the preceding experiments (Figure 5A), with a median rate of 28% by deep sequencing. We measured plasma PCSK9 and cholesterol levels every two weeks and found that every individual mouse displayed reductions of both analytes at 2 weeks and at 4 weeks compared to pre-treatment levels (Figure 5B, C). On average, the mice had 54% reduced PCSK9 levels and 28% reduced cholesterol at 4 weeks compared to pre-treatment levels. Thus, Pcsk9 disruption appeared stable at least out to 4 weeks.

Open in a separate window

Figure 5

Effects of base editing on plasma PCSK9 and cholesterol levels over time. (A) CEL-I nuclease assays performed with genomic DNA from liver samples from mice taken 4 weeks after receiving the BE3-Pcsk9 adenoviral vector. Arrows show the cleavage products resulting from the CEL-I nuclease assays. (B, C) Plasma analytes measured immediately prior to treatment, 2 weeks after treatment, and 4 weeks after treatment. The data are from 6 male mice treated with the BE3-Pcsk9 adenoviral vector at 5 weeks of age. * indicates P < 0.05 when compared to the pre-treatment levels as calculated by the Wilcoxon signed-rank test.

DISCUSSION

Our results for the first time establish the feasibility of efficient therapeutic base editing in somatic tissues in vivo. Although the degree of overall mutagenesis of Pcsk9 with BE3 and the reduction of plasma cholesterol levels were less than that seen with standard CRISPR-Cas9 genome editing in previous studies (~30% cholesterol reduction with base editing compared to 35% to 40% cholesterol reduction with genome editing),^2,3 they were nonetheless substantial enough to significantly reduce the risk of coronary heart disease if translated to humans. By way of comparison, Pcsk9 germline knockout mice display 36% to 52% lower cholesterol levels compared to wild-type mice.¹³ As with genome editing, a single administration of a base-editing therapy would be expected to result in a prolonged and possibly lifelong reduction of cardiovascular risk. Furthermore, it should be feasible to efficiently base-edit multiple genes simultaneously with a single therapy, choosing among genes for which naturally occurring loss-of-function mutations in people are protective against cardiovascular disease and well-tolerated even in the complete absence of functional protein product, such as PCSK9,^1,14 ANGPTL3,^15,16 and APOC3.^17–19

It was reassuring that there was no evidence of base-editing or indel mutagenesis at a variety of predicted off-target sites, although we cannot rule out rare mutagenesis below the threshold represented by the baseline error rate of next-generation DNA sequencing. We and others did not previously find any evidence of indel mutagenesis at a variety of predicted off-target sites with standard CRISPR-Cas9 genome editing,^3,4,6 either, precluding a direct comparison of the relative safety of in vivo base editing and in vivo genome editing. However, in principle, in vivo base editing of PCSK9 to introduce nonsense variants should be substantially safer than in vivo genome editing of PCSK9 to disrupt the gene, given the lack of need for double-strand DNA breaks, the much reduced burden of on-target indels (with their possible unintended consequences), and the reduced potential for off-target indels (with any off-target effects that do occur being more likely to be relatively benign C-to-T changes). Indeed, early empirical evidence obtained in vitro demonstrated fewer sites of off-target mutagenesis with BE3 base editing compared to standard CRISPR-Cas9 genome editing.²⁰

In this study and previous studies of somatic genome editing, we observed no adverse consequences resulting from the use of adenovirus.^2,4 Nonetheless, the disadvantages of adenovirus for therapeutic applications, most notably the possibility of a deleterious immune response, militate against its use for genome-editing or base-editing therapies. However, the large size of BE3 (~5.1 kb) exceeds the cargo capacity of adeno-associated virus (~4.7 kb). Accordingly, the goal of future studies should be to establish an alternative means of employing base editing in vivo, such as the use of a base editor built upon a smaller Cas9 ortholog from another bacterial species such as Staphylococcus aureus²¹ or the use of lipid nanoparticles,⁷ without a loss in base-editing efficiency.

Our strategy to introduce nonsense variants could potentially be applied to any gene for which loss of function would be of therapeutic benefit. Even more enticing is the possibility that pathogenic C-to-T or G-to-A mutations could be modified with high efficiency in cells that are largely non-proliferative, such as in the adult liver, heart, and brain, since HDR-mediated genome editing is not a viable approach in those cells. One such example is Alzheimer disease, for which the ε4 alelle of APOE is a substantial genetic risk factor. BE3 has been demonstrated to be able to efficiently convert the ε4 alelle into the ε3r alelle in vitro, raising the prospect of in vivo base editing to reduce the risk of Alzheimer disease in ε4 alelle carriers.⁸ Thus, therapeutic base editing could prove to be invaluable in tackling some of the most prevalent human diseases, as well as some rare genetic disorders for which there are currently no cures or treatments.

Supplementary Material

Acknowledgments

ABBREVIATIONS

Footnotes

References