Main
Adeno-associated virus (AAV) gene therapies directed to the liver have been approved for hemophilia A and B, and there are many AAV treatments in late-stage clinical development1,2. However, concerns regarding the durability of AAV gene therapies in the liver, along with the challenge of re-administration, have raised questions about its ultimate utility3,4,5,6,7,8. Clinical studies of AAV-based liver gene therapy have demonstrated reductions in efficacy within the first 2 months of treatment9,10. Correlations between the appearance of capsid-directed T cells, serum transaminase elevations and reductions of factor IX in hemophilia B trials implicate immune responses in the lack of durability9. An immediate reduction in serum bilirubin was observed following AAV gene therapy in an individual with Crigler–Najjar syndrome. However, serum bilirubin returned to pre-treatment levels within 2 months without apparent vector immunity or liver inflammation, suggesting that a non-immune mechanism may lead to loss of efficacy10. Following an initial period of expression instability lasting several months, expression appears to be remarkably consistent, albeit at low levels, in both nonhuman primates (NHPs) and humans, which is surprising because the constant turnover of hepatocytes should dilute the episomal AAV genome11.
Results
AAV gene therapy is efficient but does not persist at high levels in NHP liver
The goal of our study was to define the mechanism(s) that limit efficient and durable transgene expression following liver gene therapy with AAV vectors. Previous studies with complete preclinical and clinical datasets suggest that NHPs are better suited for evaluating key aspects of vector performance than other animal models9,10,12,13,14,15,16. We conducted initial studies in rhesus macaques using macaque-derived β-choriogonadotropic hormone (rh-β-CG; CGB) as the transgene. rh-β-CG is secreted and has a short serum half-life, meaning it can provide a longitudinal, real-time readout of transgene transcription. As rh-β-CG should be viewed as a self-protein in macaques, confounding adaptive immune responses to the transgene product or to transgene-expressing cells are unlikely. We evaluated two clade E capsids that have been used in multiple clinical trials, AAV8 and AAVrh10 (n = 6 NHPs per vector). As an essential and unique aspect of our studies, we analyzed three sequential liver tissue samples from each animal by biopsy at days 14 and 98 and at necropsy at day 182. These analyses included assessment of transgene DNA and RNA by quantitative PCR (qPCR) and cellular distribution of DNA and rh-β-CG protein expression by in situ hybridization (ISH) and immunohistochemistry (IHC), respectively.
We observed similar levels and profiles of rh-β-CG protein expression in serum for each capsid; peak levels were achieved by day 7, followed by a gradual decline to stable levels three- to sixfold lower than the peak (Fig. 1a) without transaminase elevations (Fig. 1b). Statistically significant reductions in total vector DNA and RNA occurred over time, although the magnitude of reduction was smaller for DNA than for RNA. Although DNA levels decreased further over the two later time points, RNA levels were stable between days 98 and 182 after the initial decline relative to day 14 (Fig. 1c,d).
Fig. 1: Initial peak followed by a decline to lower stable levels of self-transgene after i.v. administration of AAV vectors to NHPs.
a,b, NHPs received i.v. injections of 1013 genome copies per kg (body weight) of AAV8 or AAVrh10 vectors expressing the self-transgene rh-β-CG (n = 6 per group). Serum rh-β-CG levels were evaluated throughout the in-life phase for transgene expression (a) as well as alanine aminotransferase (ALT) levels (b). Liver tissue was collected during a liver biopsy procedure (14 or 98 days after vector administration) or at the time of necropsy (182 days after vector administration). c,d, DNA (c) and RNA (d) were extracted from liver samples to evaluate the number of vector genome copies (GC) and transgene RNA levels, respectively. e–j, IHC for rh-β-CG protein was performed on liver samples (brown staining) from animals administered AAV8 (e–g) or AAVrh10 (h–j). ISH was performed on liver samples using the RNAscope Multiplex Assay. Hybridized probes were imaged with a fluorescence microscope. k,l, Vector DNA was quantified by ISH, and transgene expression was determined by IHC as the percentage of AAV+ cells for NHPs administered AAV8 (k) or AAVrh10 (l). Values are presented as mean ± s.e.m.; *P 3,000-fold, and RNA diminished>55,000-fold) than for the transgene encoding rhLDLR (DNA diminished 6-fold, and RNA diminished 29-fold). The hLDLR vector results were more similar to the data for the rhLDLR vector, showing DNA and RNA reductions of 9-fold and 227-fold, respectively.
We performed ISH to characterize the number of cells harboring intranuclear vector DNA compared to those with cytoplasmic transgene-derived RNA (Supplementary Fig. 4). The pattern observed for the rhLDLR vector was similar to the rh-β-CG vector results, with high numbers of DNA-containing and RNA-expressing cells at day 14 (65% and 15%, respectively), followed by a 7-fold reduction of DNA over 2 years and a 12-fold reduction in RNA-expressing cells by day 77. The number of RNA-expressing cells then remained stable through day 760, reaching a steady state of ~1% RNA-expressing cells (Fig. 2i and Supplementary Fig. 4). The pattern for hLDLR was essentially the same, although with greater reductions in DNA (28-fold) and RNA (~3,000-fold), with RNA-expressing cells falling below the 0.1% threshold of detection at day 760 (Fig. 2j and Supplementary Fig. 4). Animals that received the vector encoding GFP exhibited the same high level of gene transfer and transgene expression based on analyses for day 14, although the expression dropped to undetectable levels by day 760 for DNA and by day 77 for RNA (Fig. 2k and Supplementary Fig. 4).
DNA-specific ISH revealed two patterns of intranuclear hybridization at day 14: a diffuse granular pattern and a single bright circular structure (Fig. 3a). Evaluations at later time points demonstrated retention of the circular structures, with a loss of the background granular staining (Fig. 3b). Confocal imaging illustrated the spherical shape of these structures with a diameter between 0.8 and 1.6 µm (Fig. 3c). We co-stained sections for vector DNA by ISH and for fibrillarin as a nucleolus marker by IHC. There was no overlap between these two structures, indicating that the single nuclear domain that harbors vector genomes is not related to the nucleolus (Fig. 3d).
Fig. 3: Cell type and lack of substructure association of vector DNA and RNA following i.v. administration of AAV vectors to NHPs.
a–e, NHPs received i.v. injections of 1013 genome copies per kg (body weight) of AAV8 vectors encoding rhLDLR, hLDLR, or GFP (n = 2 per group). Liver tissue was collected during a liver biopsy (14 days after vector administration) or necropsy (760 days after vector administration). ISH was performed on liver samples using a DNA-specific probe (binding to the antisense strand). a,b, DNA ISH images from NHPs administered rhLDLR at days 14 (a) and 760 (b) after vector administration. Arrows mark cells with DNA ISH signal that are not hepatocytes; green, vector DNA; blue, DAPI (nuclear counterstain). c,d, Hybridized probes were imaged with a confocal microscope (c) and co-stained with a nucleolus marker (fibrillarin antibody) shown in red (d). e, Nuclei were extracted from necropsy samples (day 760), and cDNA libraries were created from single nuclei. Nuclei from similar cell types cluster together, and the total percentage of nuclei expressing transgene RNA was evaluated. To enable analysis of transduction at an early time point, we used samples from two additional animals previously treated with 7.5 × 1012 genome copies per kg (body weight) of AAV8.TBG.GFP and necropsied at day 7 (ref. 17). A representative uniform manifold approximation and projection (UMAP) is shown for each group; NK cell, natural killer cell; HSC, hepatic stellate cell.
Source data
To better understand the global picture of liver transduction with respect to cell type, we performed single-nucleus RNA sequencing (snRNA-seq; Fig. 3e and Supplementary Fig. 5). To enable analysis of transduction at an early time point, we used samples from two additional animals previously administered AAV8.TBG.GFP and necropsied at day 7 (ref. 17). snRNA-seq analysis of these liver tissues at day 7 revealed an ~20% transduction of multiple cell types, including hepatocytes, liver sinusoidal endothelial cells (LSECs) and Kupffer cells, indicating that the thyroxine-binding globulin (TBG) promoter is not hepatocyte specific. The same analysis of day 760 tissues (Fig. 3e and Supplementary Fig. 5) showed reduced transduction similar to that for other measurements of transgene expression in hepatocytes (that is, RNA PCR and RNA ISH), with transduction of 2.5% for rhLDLR, 0.2% for hLDLR and 0.02% for GFP (Fig. 3e). Only rhLDLR-treated animals had significant detectable transgene expression in non-hepatocytes (that is, LSECs and Kupffer cells) at day 760.
The DNA ISH analyses from rhLDLR tissues were re-evaluated for evidence of vector genomes in non-parenchymal cells in light of the snRNA-seq results that showed expression across multiple cell types. Nuclear staining was indeed demonstrated at day 14 time points in non-hepatocytes, although this was substantially reduced by day 77 (Fig. 3a,b). The relevance of the wide distribution and expression of the AAV genome throughout different liver cells is unclear to the issue of durability being studied in this manuscript, although it could be important in understanding toxicity of systemic AAVs.
Vector integrates at high frequency throughout chromosomal DNA as complex concatemeric structures
The presence of many non-expressing hepatocytes that continued to harbor vector DNA compelled us to more fully characterize the structure and location of the vector genomes detected by ISH. We and others have shown integration of AAV genome sequences into chromosomal DNA in the setting of DNA repair following double-stranded breaks18,19,20,21,22,23,24,25. We evaluated liver DNA from rh-β-CG- and LDLR/GFP-treated animals for integrated vector sequences using inverted terminal repeat sequencing (ITR-seq), an anchored multiplexed PCR-based next-generation sequencing method developed by our group to detect AAV ITR sequences that have integrated into the genome18. This method captures the chromosomal sequences directly adjacent to insertion sites by ligating specific unique molecular identifier (UMI)-containing adapters to sheared DNA. This is followed by PCR amplification using primers specific to the AAV ITR and ligated adapter sequences. AAV integration sites are determined from chimeric sequencing reads that contain both AAV ITR DNA and the adjacent host chromosomal DNA. The numbers and locations of unique genome–AAV junctions were determined for each sample, and the adjacent genomic DNA sequences in each unique AAV integration were further characterized and annotated.
Analysis of tissues at day 182 from rh-β-CG vector-treated animals showed integration events at frequencies of 1.6 per 100 genomes to 1 per 1,000 genomes (Fig. 4a). A time course of integrations performed in the GFP/LDLR study showed similar levels of integration events, which declined between days 14 and 77 and subsequently stabilized to levels of 0.1–0.7 AAV integration events per 100 genomes (Fig. 4a). These integrations occurred across the genome and mostly followed a widely dispersed distribution pattern, with the exception of an increase in integrations in and around genes that are highly expressed in the liver (Fig. 4b and Supplementary Fig. 6). None of the identified insertion sites from any of the evaluated NHP samples were located within genic regions frequently mutated in human hepatocellular carcinomas (HCCs; including TP53, TERT, CTNNB1 and so on26,27,28) nor in genic regions identified in the development of mouse HCC following AAV treatment (Dlk1, Tax1bp1, Meg8 (the mammalian ortholog of the Rian locus) and so on)29,30,31,32.
Fig. 4: Localization of integrated vector DNA following i.v. administration of AAV vectors to NHPs.
a–c, NHPs received i.v. injections of 1013 genome copies per kg (body weight) AAV8 vectors encoding rhLDLR, hLDLR or GFP (n = 2 per group). Liver tissue was collected during a liver biopsy procedure (14, 77 and 224 days after vector administration) or at the time of necropsy (760 days after vector administration). a, DNA was extracted from liver samples, and the number of AAV integration loci in all injected NHPs was determined by ITR-seq. The number of unique integration loci was normalized to 100 genomes based on input DNA. Each NHP had two samples run at each time point, except for at day 760 when several replicates of each NHP were performed. b, The genomic locations of AAV integration loci were determined by ITR-seq and annotated as being within a gene-coding region (genic) or outside of a gene-coding region (intergenic) according to the rhesus macaque RefSeqGene annotation. Integrations within genic regions were further annotated by the distribution of RNA expression in the human liver (data were taken from the Human Genome Atlas at www.proteinatlas.org). Expression levels were determined by normalized expression (nx) in the human liver for each annotated gene. The following categories were determined: genes not expressed in liver (1 20). The CCS reads were then analyzed for the presence of vector and host genome sequences. This approach yielded individual reads ranging from 51 to 50,419 base pairs (bp) in length, with average sequence sizes ranging from 4,216 to 6,113 bp. The number of reads that contained vector sequences was proportional to the total amount of vector DNA originally detected by qPCR, with rhLDLR > hLDLR > GFP (Table 1).
Table 1 Summary of liver-directed gene therapy outcomes at 2 years after i.v. administration of AAV vectors to NHPs
Sequence-based analysis of these vector-containing reads demonstrated remarkable heterogeneity in the structure of AAV sequences present in vivo (Fig. 5), which often were present in complex concatemers containing mixtures of rearranged and truncated genomes (examples of these complex concatemers are presented in Fig. 5a–d). Mapping the reads against the input vector sequence illustrated the extensive nature of the rearrangements/truncations that could not have been explained by erroneous input vector and instead must have occurred after administration (compare Supplementary Fig. 7a and 7b–d).
Fig. 5: Structure of integrated vector DNA following i.v. administration of AAV vectors to NHPs.
a–d, NHPs received i.v. injections of 1013 genome copies per kg (body weight) of AAV8 vectors encoding rhLDLR, hLDLR or GFP (n = 2 per group). Liver tissue was collected at necropsy (760 days after vector administration) for each NHP, and high-molecular-weight DNA was extracted from liver samples and enriched by hybridization to probes that tiled the vector sequence. HiFi long-read sequencing of liver DNA was performed using high consensus accuracy CCS on a PacBio Sequel II instrument (>99% accuracy and Q > 20). CCS reads were mapped to both the vector and host genomes. For CCS reads containing host genomic sequence, the host genomic sequence is indicated below the scale bars as red boxes and the location of integrated vector DNA was determined as indicated by the red lines marked on each chromosomal picture. Blue boxes indicate vector genome sequences, which were further annotated by component (5′ ITR; enhancer, E; promoter, TBG; intron, IVS2; transgene, rhLDLR; PolyA, BGH; 3′ ITR) and by orientation (arrows above scale bar). a, A complete integration would contain two chromosomal junctions flanking the entire integration. b–d, Most integrations were complex concatemers and were incompletely captured, as represented by only one chromosomal junction. Vector integrations in head-to-tail (b), head-to-tail and head-to-head (c) and head-to-tail and tail-to-tail (d) complex concatemeric configurations were seen.
To help determine the likelihood that these rearranged genomes retained the ability to express transgene, we evaluated the individual long-read sequences for the presence of intact transgene cDNA. The number of intact cDNA sequences was very low in the GFP group and increased in the LDLR groups, with rhLDLR > hLDLR; in each group, the number of reads with cDNA sequences was 0-44% of those that showed any vector sequence.
A subset of cDNA containing sequences (that is, ~12%) were shown to contain flanking rhesus macaque genomic sequence in at least one end, suggesting that they were derived from integrated genomes. By simultaneously quantifying the presence of genomic and AAV DNA, we estimated that for rhLDLR samples (where the most integration was seen), between 33% and 39% of integrated sequences contained a fully intact transgene cDNA. However, this likely represents an underestimate because of shearing of the DNA during isolation and premature termination of the sequencing run.
Of all reads containing at least one flanking sequence, 92.9% contained at least one ITR, and 85.2% contained two or more ITRs. Only 10.2% of the ITRs were intact, with the majority showing heterogeneous truncations near a common region of the B–B′ palindromic loop, resulting in an average ITR length of 102.6 bp (Supplementary Fig. 8)35,36. There was remarkable homogeneity in ITR sequences within reads that harbored more than one ITR and substantial variation between ITR sequences of different reads. We broadly assessed variations within the genomic flanking DNA to provide insight into the consequences of integration on genome integrity. Based on all reads containing at least one flanking sequence, the genomic DNA was largely intact (96.8%), with only 0.7% insertions and 2.6% deletions in the flanking genomic DNA. We captured at least four long reads with flanking genomic sequence on both ends of a vector sequence. In these examples, flanking host genomic sequences demonstrated a low percentage of mismatches (94–100% identity); in a few cases, some of the flanking sequences were flipped in orientation and/or repeated, but otherwise, the rhesus macaque genome was largely unperturbed.
Discussion
Our studies reveal a complex interplay between adaptive immunity and DNA repair that affects the performance of AAV gene therapy in the primate liver. An important finding was the formation of large discrete nuclear domains of vector DNA that persist despite concomitant reductions in transgene expression. The presence of these nuclear domains is reminiscent of the replication centers observed during viral infection with adenovirus and herpes simplex virus. One hypothesis is that transcriptionally active genomes located within these structures, or elsewhere in the nucleus, undergo epigenetic silencing in a similar manner to the inactivation of intranuclear genomes of viruses (such as herpesviruses) by innate host responses. Another potential explanation is that genomes that are transcriptionally active soon after delivery are unstable37, such as the transiently appearing vector DNA that is widely dispersed in the nucleus, while the vector genomes within the persisting discrete nuclear domains were never transcriptionally active. Transgene-specific biology (for example, circulating half-life, transgene activity and biodistribution) could also contribute to variability in the pattern of expression observed for several AAV gene therapies over time9,16. The persistence of these non-expressing nuclear vector domains helps explain the long-standing enigma in primate liver gene therapy studies of high vector DNA in the setting of low vector RNA38,39.
The demonstration of integrated vector sequences at levels of 1 per 100 cells informs the consideration of safety and efficacy of liver gene therapy in several ways. The abundant and broad distribution of integrations has the potential to lead to malignant transformation. However, none of these integrations were within genes known to be associated with HCC, nor was there evidence of marked clonal expansion during the 2 years of follow-up. The fact that integration events are far less common (about 1%) than the number of cells that express at day 14 (that is, ~20%) and the number of cells that retain vector DNA within the nuclear domain (that is, ~10%) indicates that episomal forms of the vector genome must be responsible for early transcription and the formation of discrete nuclear domains. Similarities in the frequency of integration sites to the number of long-term expressing cells as measured by ISH/IHC and snRNA-seq (~1%) suggests that integration may be responsible for durable expression (summarized in Table 1). Long-read sequencing revealed the presence of vector genomes in large complex concatemers harboring many truncations and rearrangements. Evidence to support integration of these concatemers was provided in a subset of reads that contained flanking macaque chromosomal DNA. Limitations in the capture of complete concatemeric structures by long-read sequencing precluded a quantitative assessment of integration frequency, although the presence of integrated sequences with intact vector genomes supports the hypothesis that integrated vectors may be responsible for long-term expression. The finding of high vector DNA in the setting of low vector RNA was shown in our studies with two capsids and three transgenes and was similar to that described by others using different promoters, suggesting that this is a class effect of AAV in primate liver38,39. However, we cannot rule out attributes of the vector, such as promoter sequence, in influencing expression profiles.
Our studies also demonstrate the impact that adaptive immune responses can have on the frequency and molecular state of the vector DNA (for example, episomal versus integrated DNA). In the case of a non-immunogenic transgene, ~10 genome copies per cell are distributed across nuclear domains present in ~10% of cells, with unique integrations detectable in ~1 out of 100 cells at later time points. Activation of a T cell response to the transgene as observed with GFP substantially reduced the overall vector DNA content to 0.02 genome copies per cell and eliminated most of the nuclear domains to only 0.04% of cells without impacting the number of detected unique integration sites (~1 integration locus per 100 cells). Immune-mediated elimination of episomal viral DNA genomes is a well-known host defense mechanism40,41,42. It is interesting that animals treated with immunogenic transgenes demonstrated more truncations/rearrangements, thereby suggesting that cells expressing functional vector genomes are eliminated over time. It has been shown in clinical trials that steroids may diminish the T cell-mediated reduction in transgene expression9,43,44, although we recently showed in NHPs that steroids do not prevent the acute reduction in expression observed in rhLDLR-treated animals45.
Our hypothesis is that episomal vector genomes are inactivated via intracellular host defense mechanisms that the primate liver has evolved to neutralize the onslaught of viruses from the gut virome40,41,46. Insertion of vector genomes into transcriptionally active domains of chromosomal DNA should escape this inactivation. Vector-mediated transgene expression is indeed stable in the livers of lower organisms such as rodents, where host defense pathways are very different47,48. Although the data are limited, we believe a similar situation exists in some models of human liver gene therapy. Analysis of liver biopsies of human gene therapy recipients demonstrated the presence of rearranged concatemers, variation in transgene RNA/DNA ratios, and low levels of integrated vector genomes, consistent with our studies in NHPs20,49. Indirect measures of transgene expression, such as serum biomarkers, are complicated. A biphasic response to gene therapy was observed using the short half-life biomarker bilirubin in a human CN1 clinical trial, whereas a more uniform kinetic response has been observed in hemophilia A and hemophilia B clinical trials using the longer half-life readout of serum clotting factors3,4,5,10. The durability of AAV transgene expression observed in extrahepatic tissues of all species speaks to the unique behavior of primate liver in modulating AAV transgene expression.
Our findings have implications for improving the efficacy of liver gene therapy. One strategy is to enable the retained episomal AAV genomes located in the discrete nuclear domains to be more transcriptionally active by incorporating cis elements, such as different promoters and insulators, or by using drugs that prevent epigenetic silencing50. This approach may be effective regardless of whether vector genomes are active and then silenced or are simply never active. Another strategy, based on the proposal that integrated genomes are more likely to confer durable expression, would be to increase the number of vector integrations into safe harbor sites, as is the goal of some editing strategies for treating loss-of-function diseases. In this scenario, we can view editing-based gene insertions as an extension of gene therapy, in which the editing-based insertions are directed to a specific chromosomal location with higher frequency than the broadly distributed background of insertions observed following gene therapy.
Methods
AAV vector production
All AAV vectors were produced by the Penn Vector Core at the University of Pennsylvania, as previously described29, and visualized using SnapGene (version 6.2.1). Briefly, plasmids expressing rh-β-CG, codon-optimized rhLDLR, codon-optimized hLDLR and enhanced GFP from the TBG promoter were packaged within the AAV8 capsid. A vector expressing rh-β-CG from the TBG promoter was also packaged with the AAVrh10 capsid. Unique biological materials are available upon request, pending permission by the authors and/or patent holder(s).
Animal studies
The Institutional Animal Care and Use Committee of the University of Pennsylvania approved all animal procedures in this study. We obtained wild-type adult cynomolgus and rhesus macaques aged 3–6 years (n = 20) from Covance Research Products. We conducted NHP studies at the University of Pennsylvania within a facility that is registered with the U.S. Department of Agriculture, accredited by the American Association for Accreditation of Laboratory Animal Care, and assured by the Public Health Service. As previously described51, we housed animals in stainless steel squeeze-back cages with perches. All cage sizes and housing conditions were compliant with the Guide for the Care and Use of Laboratory Animals. A 12-h light/12-h dark cycle was maintained and controlled via an Edstrom Watchdog system. Animals were fed Certified Primate Diet 5048 (PMI Feeds) two times per day (morning and evening). We also provided an additional variety of food treats that were fit for human consumption, including fruits, vegetables, nuts and cereals, daily as part of the standard enrichment process. Manipulanda, such as kongs, mirrors, a puzzle feeder and raisin balls, were provided daily. Animals also received visual enrichment and daily human interaction. All interventions were performed during the light cycle, and animals were fasted overnight before being anesthetized.
On study day 0, rhesus macaques received 1013 genome copies per kg (body weight) AAV8.TBG.rh-β-CG (n = 6), AAVrh10.TBG.rh-β-CG (n = 6), AAV8.TBG.rhLDLR (n = 2), AAV8.TBG.hLDLR (n = 2) or AAV8.TBG.GFP (n = 2) as a 10-ml infusion of vector into the saphenous vein at a rate of 1 ml min–1 via an infusion pump (Harvard Apparatus). Macaques were anesthetized, and blood was collected on selected days via the femoral vein for analysis of complete blood counts, clinical chemistries and coagulation panels by Antech. We determined neutralizing antibody titers using serum samples taken before the initiation of the study, as previously described52. All animals had neutralizing antibody titers of
>>> Read full article>>>
Copyright for syndicated content belongs to the linked Source : Nature.com – https://www.nature.com/articles/s41587-023-01974-7