Evaluation of cytosine base editing and adenine base editing as a potential treatment for alpha-1 antitrypsin deficiency

Abstract

Alpha-1 antitrypsin deficiency (AATD) is a rare autosomal codominant disease caused by mutations within the SERPINA1 gene. The most prevalent variant in patients is PiZ SERPINA1, containing a single G > A transition mutation. PiZ alpha-1 antitrypsin (AAT) is prone to misfolding, leading to the accumulation of toxic aggregates within hepatocytes. In addition, the abnormally low level of AAT secreted into circulation provides insufficient inhibition of neutrophil elastase within the lungs, eventually causing emphysema. Cytosine and adenine base editors enable the programmable conversion of C⋅G to T⋅A and A⋅T to G⋅C base pairs, respectively. In this study, two different base editing approaches were developed: use of a cytosine base editor to install a compensatory mutation (p.Met374Ile) and use of an adenine base editor to mediate the correction of the pathogenic PiZ mutation. After treatment with lipid nanoparticles formulated with base editing reagents, PiZ-transgenic mice exhibited durable editing of SERPINA1 in the liver, increased serum AAT, and improved liver histology. These results indicate that base editing has the potential to address both lung and liver disease in AATD.

Graphical abstract

Base editing is a technique that introduces single-base substitutions in genomic DNA. In this study, base editing reagents formulated in lipid nanoparticles mediate durable gene editing in a transgenic mouse model of alpha-1 antitrypsin deficiency. This treatment resulted in a therapeutically relevant increase in serum alpha-1 antitrypsin and improved liver histopathology.

Introduction

Alpha-1 antitrypsin deficiency (AATD) is an inherited autosomal codominant disorder that can lead to lung disease and liver disease.¹ AATD is caused by mutations in the SERPINA1 gene that encodes AAT, a potent and highly abundant neutrophil elastase inhibitor.² The PiZ SERPINA1 variant (p.Glu342Lys) is the most common severe variant associated with liver and lung disease, with 1 in 25 people of European ancestry being a carrier and the vast majority of AATD diagnoses in individuals homozygous for PiZ SERPINA1.³ Due to a lysine substitution for glutamic acid at the 342nd amino acid position, the conformation of AAT changes, allowing the sheet-loop polymerization of AAT. Polymerized aggregates of AAT retained in the rough endoplasmic reticulum (ER) of hepatocytes are responsible for causing cellular stress and liver disease.⁴ The inability of hepatocytes to properly secrete the polymerized AAT contributes to a deficiency in circulating AAT and dysregulated neutrophil elastase. This protease-antiprotease imbalance leads to the loss of lung extracellular matrix, manifesting clinically as pulmonary emphysema.⁵ Although RNA interference therapeutics for AATD liver disease have generated promising clinical results,⁶ there is currently no US Food and Drug Administration (FDA)-approved therapy for liver toxicity associated with the PiZ SERPINA1. Meanwhile, intravenous augmentation therapy with plasma-purified AAT remains the only licensed therapy for treating AATD lung disease.⁷

Given the substantial unmet need, there have been numerous studies aimed at developing genetic medicines that could permanently alter the course of AATD. These include the early applications of the intramuscular delivery of adeno-associated viral vectors to the currently advancing vectors that simultaneously deliver a wild-type SERPINA1 transgene along with a PiZ-silencing miRNA to hepatocytes.^8,9 Unlike gene therapies, gene editing holds the promise of directly converting the chromosomal PiZ allele back to a functioning SERPINA1 gene. However, attempts to accomplish this goal using CRISPR/Cas9 nuclease have been hampered by the relative inefficiency of homologous recombination in the liver.^10,11 This challenge has prompted the evaluation of additional editing modalities for AATD, such as prime editing and base editing.^12,13

Base editors are protein complexes capable of efficiently introducing single base conversions at a precise location within the genome without requiring DNA double-strand break intermediates.¹⁴ The first class of these molecular machines, the cytosine base editor (CBE), comprises a cytidine deaminase fused with catalytically altered Cas9 nickase. The Cas9 subunit is responsible for binding to the target site in a guide RNA (gRNA)-dependent manner and unwinding a stretch of single-stranded DNA. This single-stranded DNA serves as a substrate for the cytidine deaminase, resulting in the conversion of cytosine to uracil and thus the permanent conversion of C⋅G to T⋅A base pairs.¹⁵ In contrast to CBEs, adenine base editors (ABEs) use a laboratory-evolved deoxyadenosine deaminase (TadA^∗) to convert adenines to inosines, resulting in the permanent conversion of A⋅T to G⋅C base pairs.^16,17 Both base editors enable a variety of potential therapeutic gene editing strategies.^18,19

In this study, two different base editing approaches were explored in the context of PiZ AATD. The first approach used a CBE to introduce a compensatory mutation (p.Met374Ile), conferring increased thermostability to alleviate AAT polymerization.²⁰ This approach was compared to the use of an ABE to correct the pathogenic PiZ mutation. Both approaches yielded significant editing in vitro, producing alleles that correspond with improved secretion and elastase inhibition properties relative to PiZ AAT. Subsequent studies in PiZ-transgenic mice resulted in the durable editing of SERPINA1 in the liver, increased serum AAT, and improved liver histology. These results demonstrate the therapeutic potential of base editors to alleviate lung and liver pathologies caused by the PiZ mutation.

Results

Validation of two SERPINA1 editing strategies in vitro

Two distinct base editing strategies that could address AATD were explored in parallel. The first strategy would introduce a compensatory mutation in PiZ SERPINA1. Previous reports indicated that various single-point mutations may restore secreted PiZ AAT.²⁰ Of these mutations, it appeared most feasible to introduce p.Met374Ile using the prototypical CBE BE4²¹ and a gRNA that places an ‘NGG’ protospacer-adjacent motif (PAM) at positions 21–23, the target cytosine in the protospacer at position 6, and a neighboring (bystander) cytosine at protospacer position 2 (Figure 1A). The capability of these reagents to introduce p.Met374Ile was tested in primary human hepatocytes, transfected with gRNA and mRNA coding for BE4. Up to 31% of sequenced alleles exhibited editing of only the target cytosine (6T: E342K + M374I), and <2.5% of alleles contained the conversion of both the target cytosine and the bystander cytosine (2T+6T: E342K + M374I + E376K) (Figure 1C).

Figure 1.

Validation of SERPINA1 base editing strategies

Target site sequences for (A) compensatory editing strategy with cytosine base editor and (B) correction editing strategy with adenine base editor. Allele frequencies from targeted amplicon sequencing of SERPINA1 in (C) primary human hepatocytes treated with compensatory editing reagents and (D) patient-derived fibroblasts treated with correction editing reagents. (E) Secreted AAT levels in culture supernatants from HEK293T transfected with expression vectors for mutant AAT proteins. Labels indicating base editor-induced nucleotide change and corresponding amino acid substitution. ND indicates a measurement below the limit of detection. (F) Normalized elastase activity in the presence of purified AAT protein variants. All of the midpoint values reflect averages, and all of the error bars reflect standard deviations. Statistical significance was assessed by 1-way ANOVA.

The second strategy relies upon ABE to directly correct the pathogenic mutation, thus reverting the 342nd amino acid from lysine (PiZ) to glutamate (wild type [WT]). In this case, the gRNA was designed with a non-canonical spCas9 ‘NGC’ protospacer-adjacent motif at positions 21–23, target adenine at protospacer position 7, and neighboring (bystander) adenines at positions 5, 8, and 10 (Figure 1B). Previous studies with this gRNA have yielded correction of the PiZ mutation at varying levels, depending on the spCas9 variant incorporated into the ABE, with ngcABEvar9 being the most promising published editor for this target.¹² In transfections of patient-derived fibroblasts, up to 28.98% of sequenced alleles contained only conversion of the target adenine (7G: WT). An additional 16.10% of sequencing reads exhibited base editing of both the target adenine and a single bystander adenine at position 5 (5G+7G: D341G), while 2.8% of reads contained the bystander edit alone (5G: D341G + E342K) (Figure 1D). Further chemical modification of the correction gRNA with locked nucleic acids yielded improved gene editing efficiencies; this modified gRNA was used in subsequent experiments (Figure S1).

Of the editing outcomes from both base editing strategies, only one (7G: WT) is naturally occurring and known to be benign. Although SERPINA1 (p.Asp341Gly) is not observed in human genomic databases, Asp341 appears to be tolerant to substitution, with various polymorphisms listed as benign, likely benign, or of unknown significance.²² Given this uncertainty, the AAT proteins corresponding to frequently observed editing outcomes were assayed for secretion in HEK293T cultures, transiently transfected with plasmids encoding each variant (Figure 1E). As expected, there was significantly less PiZ AAT in the culture supernatant as compared with WT. The AAT protein that would result from the conversion of only the target cytosine in the compensatory editing strategy (6T: E342K + M374I) exhibited increased secretion relative to E342K AAT, indicating the potential of this editing strategy. However, the protein corresponding to the editing of both target and bystander cytosines (2T+6T: E342K + M374I + E376K) was secreted at levels similar to PiZ AAT. In light of this observation, it was determined that only the single nucleotide base editing outcome (6T: E342K + M374I) could function as a compensatory edit.

In the correction editing approach, the protein that would result from the conversion of both the target and bystander adenines (5G+7G: D341G) appears to be secreted comparably to the WT, whereas the protein corresponding to the 5G bystander edit alone (5G: D341G + E342K) was secreted at levels similar to E342K AAT. These results indicate that p.Asp341Gly appears to have little impact on AAT secretion. Additional low-frequency editing outcomes from the correction strategy were assayed, the majority of which were not significantly different from PiZ AAT.

A subset of these AAT variants was purified from culture supernatants and assayed for elastase inhibition (Figure 1F). It was found that PiZ AAT purified in this manner was only weakly inhibitory against elastase. However, the AAT protein resulting from compensatory editing (6T: E342K + M374I) was purified and found to have increased elastase inhibitory capacity compared to the PiZ AAT, albeit less than WT AAT. Conversely, D341G AAT (5G+7G) achieved an elastase inhibitory capacity similar to the WT protein. Given the properties of D341G AAT in vitro, correction editing is hereafter calculated as the sum of alleles containing target adenine base editing with or without bystander editing at position 5 (5G+7G: D341G and 7G: WT, respectively).

Durable in vivo base editing of SERPINA1 in NSG-PiZ mice

After successfully editing the target bases in SERPINA1 in vitro, the base editing constructs were evaluated in a human PiZ-transgenic mouse model on the NOD SCID gamma (NSG) background (NSG-PiZ).^23,24 In vivo delivery of the base editors to hepatocytes was achieved via the use of lipid nanoparticles (LNPs) co-formulated with the mRNA coding for a base editor and gRNA as previously shown.^25,26 LNPs have shown encouraging results as gene and RNAi delivery systems in vitro and in vivo, and in some cases, are approved in pharmaceutical preparation for human use.²⁷ Furthermore, when applied to in vivo base editing, LNPs have numerous beneficial properties, including the ability to deliver full-length base editor mRNA, transient expression, low toxicity, low immunogenicity, long-term stability, scalable manufacturing, and lower cost of goods relative to viral-based delivery.28, 29, 30

BE4 mRNA and compensatory gRNA were formulated in compensatory LNPs. Likewise, ngcABEvar9 mRNA and correction gRNA were formulated in correction LNPs. A previously reported gRNA targeting mouse PCSK9 and BE4 mRNA were also formulated into PCSK9 LNPs to serve as a positive control for base editing and as a negative control for efficacy in AATD.³¹ These 3 LNP formulations were administered at 1.5 mg/kg in 4- to 6-week-old NSG-PiZ mice via tail vein injection. Liver tissues and serum were harvested 1, 12, and 32 after treatment (Figure 2A). Study groups going out to week 32 were bled intermittently at week 0, 1, and every four weeks until the endpoint.

Figure 2.

Comparison of compensatory and correction editing in NSG-PiZ mice

(A) Experimental design with mRNA and gRNA formulated in LNP and dosed intravenously in adolescent mice. Serum collection occurred monthly, and terminal liver tissue collection occurred at the indicated time points. (B) Amplicon sequencing results from livers of mice treated with compensatory LNP. (C) Amplicon sequencing results from livers of mice treated with correction LNP. (D) Representative example PAS-D-stained liver sections 1 week after treatment. (E) Threshold analysis of PAS-D stained area across the entire liver section. (F) Human AAT assessed by immunoassay in serial serum samples from 32-week cohorts. (G) Neutrophil elastase inhibition capacity in serial serum samples from 32-week cohorts treated with the correction or compensatory LNP, with values normalized to average values in age-matched PCSK9 LNP-treated cohort. All midpoint values reflect averages of 5–10 subjects per treatment group at the indicated time point. All error bars reflect standard deviations. Statistical significance was assessed by 1-way ANOVA. Schematics generated with BioRender.com.

As expected, mice treated with PCSK9 LNPs exhibited high rates of base editing at all tested time points (48–56%) (Figure S2A). One week after LNP administration, an average compensatory editing rate (6T: E342K + M374I) of 28.3% was detected in the liver of mice that received compensatory LNPs. Cohorts of subjects assessed at 12 and 32 weeks after treatment demonstrated similar rates of compensatory editing—34.3% and 27.2%, respectively (Figure 2B). However, mice treated with correction LNPs exhibited correction editing rates of 12% in the liver at 1 week, but higher rates of correction editing (29% and 35.7%) were observed in subjects assessed at 12 and 32 weeks after treatment (Figure 2C). These high levels of PiZ correction in the liver genomic DNA at 12 weeks were further confirmed at the mRNA level using in situ hybridization (ISH) with probes specific to either the PiZ or WT SERPINA1 transcripts. Mice treated with correction LNP showed robust staining of WT transcript in most hepatocytes (Figure S2B). It has been previously shown that genetic correction of the PiZ mutation in this model should confer a proliferative advantage to the corrected hepatocytes.²³ The observed increase over time in correction editing but not compensatory editing suggests that the correction editing approach more effectively addresses the toxic gain-of-function associated with the PiZ mutation. Furthermore, both individual correction editing outcomes (7G and 5G+7G) exhibit a similar rate of increase in frequency over time, suggesting that they each individually correct PiZ AATD liver toxicity (Figure S2C).

To directly assess the impact of base editing treatment on liver pathology, liver sections were stained with periodic acid-Schiff stain with diastase digestion (PAS-D) to visualize non-glycogenic PiZ AAT aggregates in hepatocytes.²³ Despite the differences in editing rates between compensatory and correction treatments, similar declines in AAT aggregates were observed relative to subjects treated with PCSK9 LNP at 1 week (Figures 2D, S2E, and S3). By 32 weeks after LNP administration, there were fewer PAS-D-stained AAT aggregates in control liver sections, indicating that AAT aggregates tend to clear with age in this disease model (Figure S4).³² Unlike PAS-D-stained AAT aggregates, hepatic collagen is expected to increase with age as AATD liver disease develops. To this end, picrosirius red and Masson’s trichrome stains were used to detect collagen and assess the development of liver fibrosis and cirrhosis (Figure S5A). Although collagen staining was observed at 32 weeks, all of the control subjects received low fibrosis grading (Batts-Ludwig) by a clinical pathologist with no significant difference from grades in the correction-treated subjects (Figure S5B).³³ This low degree of fibrosis in NSG-PiZ mice differs from previously published reports of significant fibrosis at 32 weeks of age in C57BL6-PiZ mice.²³

It is well understood that PiZ-transgenic mice do not develop pulmonary pathology owing to the intact functional mouse SERPINA1 genes. However, the effects of gene editing treatments on AAT deficiency were assessed indirectly through measurements of total human AAT in serum and serum anti-elastase capacity, a composite measure of both human AAT and endogenous mouse serum factors. One week after LNP administration, human AAT serum levels were elevated by 1.7-fold and 1.6-fold in mice treated with compensatory and correction LNPs, respectively, although the correction LNP-treated group displayed a sustained increase for 32 weeks (Figure 2F).

In addition, serum anti-elastase activity in mice treated with correction LNPs increased to roughly 2-fold the levels of the PCSK9 control group at each time point assessed over 32 weeks (Figure 2G). However, the compensatory LNP-treated group initially exhibited an increase of 36% in serum anti-elastase activity relative to controls, but declined to the same level as controls by 32 weeks. These results indicate that the correction editing approach is superior to the compensatory editing approach in addressing deficiency of circulating AAT and may confer better protection of pulmonary function in AATD patients.

In vivo base editing in newborn PiZ mice

In light of the results in NSG-PiZ mice, the correction editing approach was selected for further validation in a second model of AATD, the PiZ-transgenic mouse with C57BL6 background.^34,35 Unlike the NSG-PiZ mouse, this model enabled the relevant assessment of liver fibrosis progression in an immunocompetent background. Correction LNPs and vehicle control were injected in neonatal (P1) mice, thus evaluating the long-term proliferative advantage of corrected hepatocytes. The terminal collection of liver tissue 1 week after treatment included males and females, whereas 13- and 21-week time points were separated by gender (females and males, respectively) and included periodic serum collections (Figure 3A). One week after treatment, correction edits observed in the liver were 1.69% and 3.47% in the 0.75-mg/kg and 1.5-mg/kg dose levels, respectively (Figure 3B). As predicted, correction editing rates increased to 9% at week 13 and 10% at week 21 in mice treated with the 0.75-mg/kg dose.

Figure 3.

Correction editing in newborn C57BL6-PiZ mice

(A) Experimental design with mRNA and gRNA formulated in LNP and dosed intravenously in newborn mice. Serum collection occurred monthly and terminal liver tissue collection occurred at indicated time points. (B) Amplicon sequencing results from livers of mice treated with correction LNP. (C) Human AAT assessed by immunoassay in serial serum samples from the 21-week cohort. (D) Correlation of correction editing and human AAT in terminal serum collection from 13-week cohort. (E) Correlation of correction editing and human AAT in terminal serum collection from the 21-week cohort. (F) Mass spectrometry quantification of AAT proteins containing amino acid substitutions corresponding to predominant base editing outcomes. (G) Correlation of both D341G allele frequencies with serum D341G AAT protein abundance as well as WT allele frequency and serum WT AAT protein abundance. All midpoint values reflect averages of 10–18 subjects per treatment group at the indicated time point. All error bars reflect standard deviations. Schematics generated with BioRender.com.

Furthermore, mice treated with the 1.5-mg/kg dose displayed editing rates of up to 22% at week 13 and 25% at week 21 (Figure 3B). Similar to the results in NSG-PiZ mice, precise correction (7G) was the predominant outcome, and both types of correction edits (5G+7G and 7G) increased in frequency over time (Figures S6A and S6B). Correction editing was further confirmed by robust staining of WT SERPINA1 transcripts by ISH in liver sections of treated mice at week 21 (Figure S6C).

Over the course of the experiment, a dose-dependent augmentation in the total AAT serum levels was observed (Figure 3C). Furthermore, AAT concentration in terminal serum collections at 13 and 21 weeks was strongly correlated with the frequency of correction edits (Figures 3D and 3E). Serum AAT protein was also quantified by mass spectrometry of trypsin-digested sera. This method enables the quantification of peptides corresponding to total human AAT and peptides unique to AAT isoforms harboring amino acid substitutions resulting from base editing. As expected, WT AAT was most abundant in the serum of treated subjects, followed by the D341G AAT corresponding to target and bystander adenine base editing (5G+7G) (Figure 3F). The abundance of these 2 AAT isoforms (WT and D341G) correlated well with their respective allele frequencies, suggesting no difference in relative expression between these 2 corrected AAT variants (Figure 3G).

Given that liver gene editing took place just after birth, liver pathology was expected to diverge between control and correction cohorts over time. Therefore, liver sections from subjects at 13 weeks were stained with PAS-D to detect AAT aggregates. The high-dose correction-treated animals exhibited less PAS-D staining on average, but this difference was not significant due to variability among control animals and insufficient statistical power (Figures 4A, 4B, and S7). Consequently, liver sections from mice harvested at the end of 21 weeks were stained with picrosirius red and Masson’s trichrome stain to evaluate the extent of fibrosis. A qualitative difference in the spatial organization of stained collagen between treated and control animals was observed, with controls exhibiting far more fibrous septa bridging between portal areas (Figures 4C and S8). For quantitative analysis, images were randomized and independently graded (Batts-Ludwig scoring) by a clinical pathologist.³³ In agreement with qualitative differences, the Batts-Ludwig scores were significantly lower in animals treated with 1.5 mg/kg correction LNP than controls (Figure 4D). This result indicates that correction of the PiZ mutation through base editing can prevent the development of AATD liver pathology.

Figure 4.

Liver histopathology in correction LNP treated C57BL6-PiZ

(A) Representative example of PAS-D-stained liver sections at 13 weeks after treatment. (B) Threshold analysis of PAS-D-stained area across the entire liver section. Midpoint values reflect averages, and statistical significance was assessed by 1-way ANOVA. (C) Representative example of picrosirius red and Masson’s trichrome-stained liver sections at 21 weeks after treatment. (D) Batts-Ludwig fibrosis staging assessed by a blinded independent pathologist. Midpoint values reflect medians, and statistical significance was assessed by the Student’s t-test.

Discussion

AATD is a severe chronic condition with insufficient treatment options for the associated liver and lung diseases. Base editing has the potential to permanently address both disease processes at the genetic source of dysfunction. The first step in realizing this potential was the identification of two distinct sets of base editing reagents: (1) an ABE mRNA and gRNA capable of correcting the disease-causing p.Glu342Lys mutation, and (2) a CBE mRNA and gRNA capable of introducing pMet374Ile, an amino acid substitution that stabilizes AAT protein. These two sets of base editing reagents were each formulated in LNPs to mediate delivery and thus base editing in the livers of a transgenic mouse model of AATD. As early as 7 days after treatment, both editing strategies resulted in the elevation of serum AAT levels and the clearance of PiZ aggregates in the liver. However, only mice that received correction LNP demonstrated a durable increase in serum neutrophil elastase inhibition, a measure of functional AAT.

Furthermore, the frequency of corrected alleles increased over the course of the study, while compensatory editing remained relatively constant. Unlike compensatory editing, corrective editing may resolve hepatoxicity at a cellular level, thus conferring a proliferative advantage to edited hepatocytes. To further substantiate this hypothesis, newborn mice were also treated with LNPs containing correction base editing reagents. In this experiment, rates of correction edits in the liver also increased over time.

The PiZ-transgenic mice used in these studies are an imperfect model of AATD. Due to the many transgene copies, the load of toxic PiZ aggregates in the liver of these mice is significantly higher than it is in a typical AATD patient. Despite this supraphysiological burden, treatment with correction LNP prevented the development of fibrosis. It remains to be seen whether base editing could improve liver pathology with an intervention later in the development of AATD liver disease, as would likely be the case in a clinical setting.

The expression of mouse orthologs of SERPINA1 ensures that this mouse model does not develop AATD-related pulmonary disease. Although serum AAT can serve as a meaningful biomarker for AATD lung disease, the level of AAT in PiZ mouse serum declines with age, and at certain times may be higher than what is seen in ZZ patients. Consequently, the physiological relevance of treatment-induced increases in serum AAT can only be assessed relative to an age-matched control cohort. Correction LNP-treated NSG-PiZ mice exhibited averages of 1.65- and 4.70-fold increases in serum AAT relative to control mice at weeks 1 and 16, respectively. Treated C57BL6-PiZ mice exhibited an even greater average increase in serum AAT up to 16.6-fold relative to control mice at week 16. If these fold change values were applied to the typical ZZ patient (4 μM), then their AAT levels would certainly approach or cross the 11-μM pulmonary protective threshold.³⁶

The LNP doses used in these proof-of-concept studies (1.5 mg/kg) are higher than have been reported recently for other liver-targeted LNP therapeutics currently being used and studied in the clinic.^28,30 Therefore, further optimization of these base editing reagents and the LNP delivery may be required to achieve an appropriate therapeutic index. In addition to the acute tolerability of the LNP, an essential safety attribute of any gene editing therapeutic is its off-target editing profile. Base editors are typified by both guide-dependent and guide-independent off-targets.¹⁵ The ABE and gRNA used to correct the PiZ mutation in this study have previously been shown to generate iPSC lines with no detected off-targets of either class.¹² However, this does not preclude the presence of lower frequency off-target effects that could appear in more sensitive assays.^15,37,38 Despite these limitations, this study demonstrates that a single systemic administration of an LNP carrying gRNA and mRNA coding for an ABE can permanently correct the PiZ mutation in the liver, increase functional serum AAT levels, and prevent the development of liver pathology. These results highlight the therapeutic potential of base editing as a one-time treatment for AATD lung and liver diseases.

Materials and methods

• 1.mRNA synthesis: Production of mRNA by in vitro transcription was conducted as previously described.¹⁷ Template plasmids encoding a T7 promoter, a 3′ UTR, a base editor open reading frame (ORF), a 5′ UTR, and a 120 polyA tail followed by a type II restriction enzyme site (mRNA sequences are described in Table S1). Plasmids were prepared by ZymoPURE II plasmid kits, linearized by restriction digestion, and purified by phenol-chloroform extraction. Transcription reactions were performed using the NEB HiScribe T7 High-Yield RNA synthesis kit (NEB) with CleanCap AG reagent (Trilink) and complete substitution N1-methylpseudouridine triphosphate for uridine-5′-O-triphosphate (UTP). Transcripts were purified by lithium chloride precipitation.
• 2.Cell culture: Primary human hepatocytes (BioIVT) were seeded in CP medium (BioIVT) onto a collagen type I coated 24-well plate (Corning) at a density of 350,000 cells/well and incubated at 37°C, 5% CO₂ for several hours to generate an adherent cell monolayer. Upon monolayer formation, hepatocytes were transfected using the MessengerMAX transfection reagent (Thermo Fisher Scientific) co-formulated with base editor mRNA (600 ng) and gRNA (200 ng) (gRNA sequences are provided in Table S2). Hepatocytes were lysed and harvested for genomic DNA 48 h post-transfection. Primary fibroblasts derived from a PiZZ individual (GM11423, Coriell) were cultured according to supplier instructions. Fibroblasts were transfected using the Neon Transfection System (Thermo Fisher Scientific). Each electroporation was conducted using 70,000 cells, 100 ng mRNA, and 50 ng gRNA in a 10-μL electroporation tip with a single 40-ms pulse of 1,000 V. Transfected cells were cultured in a 24-well plate for 48 h post-electroporation, at which point genomic DNA was extracted with QuickExtract DNA Extraction Solution (Lucigen).
• 3.AAT secretion assay in HEK293T: On day 0, HEK293T (American Type Culture Collection [ATCC]) cells were plated in 48-well tissue culture-treated plates (60,000 cells per well). The following day, cells were transfected with 125 ng of plasmid cytomegalovirus (pCMV) expression vectors encoding AAT mutants using TransIT-293 (Mirus). Cell culture supernatants were collected 24 h after transfection, and AAT levels were measured using ELISA, as previously published.³⁹
• 4.AAT purification and elastase inhibition assay: HEK293T cells were passaged at a 1:4 ratio into 75 cm² tissue culture-treated flasks, and on the following day, transfected with 20 μg pCMV expression vector encoding AAT mutants using 60 μL of TransIT-293. Culture supernatants were collected at 48 h and 72 h after transfection and pooled together. Culture supernatants were diluted 2-fold into the binding buffer (20 mM Tris, pH 8.0, and 100 mM NaCl), and gravity flowed through a 0.5-mL bed volume of Alpha-1 Antitrypsin Select Resin (Cytiva). The column was washed with 20 column volumes of bind buffer and eluted in 4 column volumes of 20 mM Tris pH 8.0 and 2 M MgCl_2. Eluate was concentrated and buffer exchanged into the binding buffer using a 15-kDa cutoff centrifugal device (Amicon). Concentrations were assessed by absorbance at 280 nm (NanoDrop, Thermo Fisher Scientific). Elastase inhibition was assessed using the EnzChek Elastase Assay (Thermo Fisher Scientific) at 0.5 U elastase/mL final concentration.
• 5.Lipid nanoparticle formulations: The base editor mRNA and gRNA were co-encapsulated at a 1:1 weight ratio in LNPs. The LNPs were generated by rapidly mixing an aqueous solution of the RNA at a pH of 4.0 with an ethanol solution containing four lipid components: an ionizable lipid, distearoylphosphatidylcholine (DSPC), cholesterol, and a lipid-anchored polyethylene glycol (PEG) (all of the lipids were obtained from Avanti Polar Lipids). The two solutions were mixed using the benchtop microfluidics device from Precision Nanosystems. Post-mixing, the formulations were dialyzed overnight at 4°C against 1× Tris-buffered saline (TBS) (Millipore Sigma). They were subsequently concentrated down using 100,000 Da molecular weight cutoff (MWCO) Amicon Ultra centrifugation tubes (Millipore Sigma) and filtered with 0.2-μm filters (Pall Corporation). The total RNA concentration was determined using Quant-iT Ribogreen (Thermo Fisher Scientific); the particle size was determined using the Malvern Panalytical Zetasizer. All in vivo studies described in the main text and main figures were conducted with a proprietary ionizable lipid and LNP composition. Similar results were also obtained with TT3 lipid-like nanoparticle (TT3-LLN) formulated as previously described (Figure S9).^40,41
• 6.Targeted amplicon sequencing and data analysis: Amplicon sequencing was performed as previously described.¹⁷ Genomic loci were amplified in 25-μL PCR reactions using Q5 2× Hot Start Master Mix (NEB), 0.5 mM forward and reverse primer, and 2 μL of genomic DNA template. Primers oBTx222 and oBTx223 were used to assess rates of E342K correction, whereas primers oBTx628 and oBTX629 were used to measure rates of M374I compensatory base editing, and primers oBTx136 and oBTX137 were used to measure PCSK9 editing (sequences in Table S3). Barcoded amplicons were generated in 25-μL PCR reactions using Q5 2× Hot Start Master Mix, 0.5 mM barcode primers, and 2 μL of the prior PCR reaction. Barcoded amplicons were combined and purified via DNA agarose gel extraction (Zymo Research) or by SPRIselect bead cleanup (Beckman Coulter). Final clean libraries were then quantified using a Qubit 4 Fluorometer (Thermo Fisher Scientific) and sequenced on an Illumina MiSeq Instrument. The data analysis was performed as previously described.¹⁷ Briefly, this process can be summarized in four steps: (1) demultiplexing of sequencing reads into fastq files, (2) read trimming and quality score filtering, (3) alignment of all reads to the expected amplicon sequence, and (4) quantification of editing rates and allele frequencies. Allele frequency outputs and sample details are listed in Tables S4–S11.
• 7.Animal studies: All of the animal procedures were performed according to the guidelines of the Institutional Animal Care and Use Committee of the University of Massachusetts Medical School and CRADL in Cambridge, Massachusetts. The two mouse models, NSG-PiZ and C57BL6-PiZ, used in this study have been described previously and were purchased from Jackson ImmunoResearch Laboratories or bred in-house.^23,24,34 LNPs were administered in the adult mouse by tail vein injection at various doses of total RNA to subject body weight (mg/kg) indicated in the text, at a volume of 10 mL/kg body weight. LNPs were administered in neonatal (P1) mice by submandibular vein injection at different concentrations in a volume of 50 μL. Blood was collected in serum tubes (BD) from the submandibular vein of adult mice, pre- and post-injection. The serum was separated by centrifugation at 1000 × g for 15 min, transferred to a new tube, and stored at −80°C until use.
• 8.Liver tissue processing for NGS and gDNA extraction: Liver samples were homogenized in phosphate-buffered saline using an OMNI Prep homogenizer (Omni International). Two aliquots of 50 μL per homogenized liver sample were transferred to a 96-well plate for proteinase K treatment. 20 μL of proteinase K and 20 μL of enhancer (Applied Biosystems) was added per well, and samples were incubated for 30 min at 65°C. gDNA extraction was performed using the MagMAX DNA Multi-Sample Ultra 2.0 kit (Applied Biosystems) following the manufacturer’s instructions.
• 9.Liver histology: For the determination of histological changes, liver samples were fixed in 10% neutral-buffered formalin (Fisher Scientific) and embedded in paraffin. Fixed sections (4 μm) were stained with PAS with diastase digestion to remove glycogen. Fixed sections were stained using standard protocols for either picrosirius red stain or Masson’s trichrome stain.
• 10.BaseScope Duplex ISH: ISH was performed as directed in the BaseScope Duplex Reagent Kit (ACD) using custom made-to-order probes specifically validated by ACD for E342K (PiZ) SERPINA1 and WT SERPINA1 transcripts.
• 11.Histology image analysis: Whole digital slide images of PAS-D-stained sections were created using an Aperio CS ScanScope (Leica Biosystems) and analyzed using ImageJ (National Institutes of Health) color thresholding.⁴² Masson’s trichrome-stained digital slide images were randomized in order and transferred for Batts-Ludwig grading by an independent and blinded clinical pathologist.
• 12.Electrochemiluminescence immunoassay for human AAT in serum samples: The human SERPINA1 antibody set (Meso Scale Discovery) was used in conjunction with MSD GOLD Small Spot Streptavidin plate (Meso Scale Discovery) per the manufacturer’s protocol. Measurements were taken on a MESO QuickPlex SQ 120.
• 13.Elastase inhibition capacity of mouse serum: Elastase inhibition was measured using the Neutrophil Elastase Inhibitor Screening Kit (Millipore Sigma), with minor modifications. Serum samples were diluted 1:10,000 with assay buffer or PBST (phosphate-buffered saline with Tween 20) (Thermo Fisher Scientific). Inhibitor control and neutrophil elastase were prepared following the manufacturer’s instructions. AAT from human plasma (Millipore Sigma) was used to generate a standard curve for absolute quantification. The diluted samples, standards, inhibitor control, or blank were added into respective wells within a black 96-well plate with a clear flat bottom (Corning). Neutrophil elastase and substrate were then added following the manufacturer’s instructions. Fluorescence was measured every minute at ex = 400/em = 505 nm for 20 min using Infinite 200 Pro (Tecan). Finally, relative inhibition was calculated following the manufacturer’s instructions.
• 14.ELISA: Human AAT (hAAT) protein levels were detected by ELISA as described previously.³⁵ Nunc-Immuno MicroWell 96-well solid plates (MilliporeSigma) were coated with 100 μL of goat anti-human AAT antibody (1:500 diluted; Bethyl Laboratories, A80-122A) in Voller’s buffer overnight at 4°C. After blocking with 3% BSA (Millipore Sigma, B4287), duplicate standard curves (hAAT; Millipore Sigma, cat. no. A-9024) and serially diluted unknown samples were incubated in the plate at 37°C for 1 h. A second antibody, goat anti-human AAT antibody (1:5,000 diluted; Bethyl Laboratories, A80-122P), was reacted with the captured antigen at 37°C for 1 h. The plate was washed with PBST between reactions. After reaction with 3,3′,5,5′-tetramethylbenzidine (TMB) peroxidase (KPL Kit, 50-76-00) substrate, development was stopped by adding 1 M H₃PO₄ (Thermo Fisher Scientific, A242–500). Plates were read at 450 nm on a microplate reader.

Data availability

Sequencing data are available through the NCBI Sequencing Read Archive under PRJNA793592.