Recombinant AAV Integration Is Not Associated With Hepatic Genotoxicity in Nonhuman Primates and Patients

Abstract

Recombinant adeno-associated viral vectors (rAAV) currently constitute a real therapeutic strategy for the sustained correction of diverse genetic conditions. Though a wealth of preclinical and clinical studies have been conducted with rAAV, the oncogenic potential of these vectors is still controversial, particularly when considering liver-directed gene therapy. Few preclinical studies and the recent discovery of incomplete wild-type AAV2 genomes integrated in human hepatocellular carcinoma biopsies have raised concerns on rAAV safety. In the present study, we have characterized the integration of both complete and partial rAAV2/5 genomes in nonhuman primate tissues and clinical liver biopsies from a trial aimed to treat acute intermittent porphyria. We applied a new multiplex linear amplification-mediated polymerase chain reaction (PCR) assay capable of detecting integration events that are originated throughout the rAAV genome. The integration rate was low both in nonhuman primates and patient's samples. Importantly, no integration clusters or events were found in genes previously reported to link rAAV integration with hepatocellular carcinoma development, thus showing the absence of genotoxicity of a systemically administered rAAV2/5 in a large animal model and in the clinical context.

Introduction

Recombinant adeno-associated viruses (rAAV) mainly persist episomally as concatemeric structures mediating stable long-term transgene expression in vivo.^1,2,3 Although they lack the Rep gene that mediates wild-type AAV (wt-AAV) integration, it has been extensively demonstrated that rAAV persistence is also mediated by random integration events occurring via nonhomologous recombination at low frequencies.^{4,5,6,7,8,9,10,11} Numerous groups have shown the safety of rAAV despite of vector integration sites (IS) and no genotoxic events have been reported in the >130 rAAV clinical trials conducted so far.^3,12,13,14 Nonetheless, the oncogenic potential of rAAV is still controversial due to preclinical studies suggesting their genotoxicity in mouse models of disease prone to hepatocellular carcinoma (HCC) development,^15,16,17 studies using neonatal administration^18,19 or a combination of both.^20,21 However, scarce evidences in wild-type mice,^11,18,20 as well as in large animal models, have raised concerns about the clinical relevance of these findings. Remarkably, a recent study has reported the presence of wt-AAV2 IS in 11/193 human HCC biopsies.²² Nevertheless, the wt-AAV2 potential to drive HCC development remains still undefined as, among the 11 AAV-positive biopsies, 5 biopsies presented other well-known HCC-driving conditions (such as alcohol consumption and hepatitis B or C virus infections) and 4 biopsies presented AAV-unrelated mutations which have been previously linked to liver cancer development (such as TP53 or AXIN1). Thus, only two biopsies remained in which no other potential liver cancer mutations were detected. In addition, the clinical significance of this study regarding rAAV gene therapy is unclear as this work was related to wt-AAV2, which has been shown to exhibit a different integration profile compared to recombinant AAV vectors.²³ Interestingly, exclusively incomplete wt-AAV genomes were found to be integrated.²⁴

Currently, analyses of vector IS in gene therapy studies are mainly addressed through polymerase chain reaction (PCR)-based approaches.^25,26,27 Like other PCR-based IS analysis methods, linear amplification-mediated (LAM)-PCR technology relies on serial amplifications performed with primers located at the vector termini, thus lacking the detection of internal rAAV genome regions. Under the hypothesis that rAAV breakage and subsequent integration may also occur from internal vector fragments, we developed a multiplex LAM (M-LAM)-PCR assay, performed aside of standard LAM-PCR, which identifies and quantifies IS and concatemeric forms derived also from internal vector breakage events. This novel approach was applied to the extensive analysis of the rAAV2/5 integration profile in nonhuman primate and human biopsies after the intravenous administration of clinical rAAV2/5-cohPBGD designed for the treatment of acute intermittent porphyria.²⁸ Here, we show low frequency integration events, as well as complex rearranged concatemeric structures, arising from internal rAAV regions derived from breakage events occurring throughout the vector sequence and the absence of integration events or clusters in genes previously reported to be associated with HCC development.

Results

Multiplex LAM-PCR retrieves IS derived from internal vector regions in nonhuman primate and patients biopsies

We hypothesized that rare internal breakage events may also occur throughout the rAAV genome potentially leading to concatemerization and/or integration of internal vector fragments into the host genome. As current PCR-based techniques only allow the identification of IS containing inverted terminal repeats (ITR) sequences, we developed a multiplex variant of LAM-PCR (M-LAM-PCR) that simultaneously employs five primer sets covering the complete vector sequence (Figure 1a). First, we established M-LAM-PCR on rAAV2/5-cohPBGD plasmid limiting dilutions and defined the efficiency of each individual primer set (Supplementary Figure S1). Standard and M-LAM-PCR were performed on different tissues from Macaca fascicularis (liver, spleen, adrenal gland) and human liver biopsies obtained 1 month and 1 year, respectively, after the administration of clinical rAAV2/5-cohPBGD. This vector, aimed to treat acute intermittent porphyria, expresses human hydroxymethylbilane synthase under the control of a liver-specific promoter and was intravenously administered to nonhuman primates (NHP)²⁸ and acute intermittent porphyria patients (manuscript in preparation).

Figure 1.

Standard and multiplex (M)-LAM-PCR comparison analysis and retrieved vector integration sites (IS). (a) DNA was isolated from tissue samples and analyzed by standard LAM-PCR, that uses one primer set located within the inverted terminal repeats proximity, and M-LAM-PCR, where a total of five primer sets covering the whole vector are employed. Black arrows denote location of the individual LAM-PCR primer sets on the vector genome. (b) Vector IS identified in nonhuman primates tissues ( liver, adrenal gland, spleen) and human liver (▪) were mapped to their respective genome. Integration clusters found (◯) and the hepatocellular carcinoma-related genes recently described () are shown. LAM, linear amplification-mediated; PCR, polymerase chain reaction.

We characterized >5,700,000 sequence reads derived from the standard (1,394,612 reads) and M-LAM-PCR (4,312,209 reads) approaches performed on human and NHP samples, from which >100,000 sequences were identified as vector-genome junctions (IS). Standard LAM-PCR performed on NHP samples yielded 353 uniquely mappable IS (Table 1). M-LAM-PCR, that includes the primer set 1 used in standard LAM-PCR together with four additional primer sets, retrieved a comparable number of 365 unique IS, including IS that were originated from internal vector parts (Table 1; Supplementary Figure S2). The apparently lower efficiency of standard LAM-PCR primer set 1, when used in combination with primer sets 2–5 rather than solely, mirrors individual primer efficiencies (Supplementary Figure S1). Similarly, LAM-PCR analysis performed on liver biopsies from three subjects administered with rAAV2/5-hcoPBGD identified a total of 134 uniquely mappable IS (10 and 124 IS retrieved by standard and M-LAM-PCR, respectively) again with integration events derived from internal vector regions (Table 1).

Table 1. Overview of raw sequence read numbers and integration sites retrieved by standard and multiplex linear amplification-mediated polymerase chain reaction from nonhuman primates and patients receiving the rAAV2/5-cohPBGD vector.

A lower percentage of the rAAV amplicon sequences (0.04–9.24%) corresponded to integration events in both human and NHP tissues when compared to the percentage of vector-vector sequences (90.76–99.96%). Thus, supporting the rAAV persistence mainly as episomal forms, rather than integrating into the host genome. In NHP tissues, vector integration frequencies were found to range between 7.44 × 10⁻⁵ and 1.00 × 10⁻⁴ IS per cell and only slight variations were observed within the different tissues (Table 1). Average integration frequencies were low in both NHP and human livers (2.00 × 10⁻⁴ and 1.17 × 10⁻³ IS per cell on average, respectively), underlining the nonactively integrating nature of the vector that persists mostly as episomes within the human and NHP hepatic tissue.

rAAV IS are genome wide distributed in nonhuman primate tissues and patient liver biopsies

The IS retrieved by both standard and M-LAM-PCR in the NHP tissues and human liver biopsies were mapped to the corresponding genome (Figure 1b). The integration profile of the rAAV2/5-hcoPBGD vector in the NHP tissues and human livers was characterized in terms of chromosome distribution by comparison with in silico generated human or NHP random datasets of 10,000 and 9,228 IS, respectively (Figure 2a,b). No significant differences were found among the three NHP tissues, indicating that vector integration characteristics seem not to be influenced by the cell type. Accordingly, IS detected in human liver biopsies showed no targeting of specific genomic regions comparable to the NHP integration profile. Also no significant variations were found for rAAV integration frequency within gene coding or 10 kb surrounding genomic regions for the NHP (average: 9.7 and 13.8%, respectively) and human (average: 49.3 and 58.2%, respectively) samples when compared to their respective random datasets (Supplementary Figure S3). Analysis of common integration sites (CIS), considered as vector integration hotspots, yielded no significant clusters of IS in NHP or patients samples. Ten low order CIS (≤3) were identified within the NHP tissues but no CIS could be retrieved in human liver biopsies (Figure 1b; Supplementary Table S1). Most importantly, none of the rAAV integration events or CIS retrieved was located in genomic regions associated with HCC development in previous rAAV integration studies.

Figure 2.

Chromosomal distribution of the retrieved integration sites (IS). The percentage of IS identified on each chromosome was determined in the different nonhuman primates tissues (a) and human liver biopsies (b). Chromosomal distributions from synthetic data sets of random IS are shown for comparison.

Internal vector breakage gives rise to complex rearranged concatemeric structures

As vector persists mainly as nonintegrated structures within the host cell, we next screened vector-vector sequences (concatemers) to determine the frequency and structure of rearrangements along the vector genome. We developed a bioinformatical approach, based on the diagram method,²⁹ in order to dissect rearranged vector sequences. In most of the sequences (with a frequency of 0.93 in human liver and 0.65, on average, in the different NHP tissues) only nonrearranged vector fragments were identified. The most frequent rearrangements found corresponded to structures bearing two vector fragments that occurred in an average frequency of 0.31 and 0.07 in NHP and patient samples, respectively (Figure 3). Uniquely, we also identified multiple rearranged structures bearing up to eight different vector fragments with frequencies, ranging from 4.0 × 10⁻² to 2.0 × 10⁻⁶, which progressively decreased as the complexity of the rearrangements increased. Notably, all the detected multiple rearrangements were pure vector rearrangements and none of them was related to rAAV integrated forms. We next mapped the vector fragments to the rAAV genome and found that these recombination events took place even between distant vector regions and did not always involve the ITR region (Supplementary Figure S4).

Figure 3.

Analysis of vector rearrangements. Among reads bearing a single vector fragment (1 Fragment), we also detected concatemeric structures containing up to 8 different rearranged vector fragments (2 to 8 Fragments).

Discussion

This study shows that rare breakage events occurring along the rAAV genome give rise to integrated internal vector fragments in the host genome. In addition, we report that the rAAV2/5-hcoPBGD vector IS are distributed genome wide in the liver of NHP and patients, including IS derived from the complete vector sequence, with no integration events or clusters within genes previously associated with HCC development.^{15,16,17,18,19,20,21} Importantly, our findings were comparable in samples from patients and nonhuman primates, indicating the potential clinical relevance of this large animal model in the study of rAAV genotoxicity.

Although complete integration studies upon rAAV systemic delivery in large animal models are still missing, the genome-wide distribution of the NHP IS here retrieved is in line with the majority of investigations reporting rAAV integration events occurring at low frequencies with random and safe distributions. Nonetheless, some studies performed in mice attributed oncogenic potential to clustered rAAV integrations into the RNA imprinted and accumulated in nucleus (Rian) gene that were identified in liver tumor samples.^{17,18,19,20,21} When we analyzed our liver biopsies, showing no signs of malignant transformation, we did not find any integration event into the delta-like homolog 1-deiodinase type 3 gene (the human homolog for the mouse Rian locus). Although species-related differences could underlie distinct integration patterns, the developmental stage at the time of vector administration may also play a role. Integrations within the Rian locus have been reported most exclusively upon neonatal administration and the data here presented proceeds from adult individuals where the Rian homolog might be located in a chromatin region not accessible for vector integration. Importantly, we also did not find other rAAV IS within genes proposed as HCC drivers in adult mice prone to liver cancer development.¹⁵ The analysis performed on NHP tissues showed a slightly higher number of IS retrieved from the liver when compared to the other organs. Given the similar integration frequencies found in our study and previously reported variations in the vector copy number among these organs,²⁸ this is most likely due to differences in the viral load present on each organ, supporting the passive nature of rAAV2/5 integration. Although the number of IS retrieved in NHP was low in order to establish the statistical randomness of the rAAV integration, studies performed with lentivirus and HIV IS datasets with a comparable size already revealed preferential viral integration into specific subgenomic regions.^30,31 As only low order CIS were retrieved in NHP samples, this suggests the absence of specific integration patterns for the rAAV.

We have previously reported the safe integration profile of the intramuscularly delivered Glybera medicinal drug in patient's muscle biopsies and we even identified rAAV integration events within the mitochondrial genome.¹³ The genome wide distribution of rAAV IS in patient's livers suggests that these vectors may exhibit similar integration patterns in the human muscle and liver. However, in the present study we did not identify any integration event within the mitochondrial genome. Most likely, this is due to the fact that rAAV mitochondrial integration is a rare event and, therefore, can only be detected in organs with a high mitochondrial content such as the muscle. In line with our findings, no severe side effects or cancer development has been described in any treated patient upon rAAV gene therapy. Nonetheless, the recent study reporting the presence of partial wt-AAV2 sequences integrated in HCC biopsies²² has again raised concerns in rAAV gene therapy. The authors report an almost exclusively integration of incomplete AAV genomes lacking complete Rep sequences in all cases and even an integrated sequence completely lacking the ITR region was retrieved. Our findings in rAAV-administered patients, also suggest that integration of internal vector fragments may arise from vector breakage events. However, we found no integration events in the human biopsies within any of the reported HCC-related genes. A detailed report of wt-AAV2 IS in matched non-tumor biopsies and samples from healthy individuals will probably contribute to the understanding of these differences. Notably, we could not identify any integration cluster within the patient samples. As the number of rAAV integration events retrieved from patient liver biopsies was limited, integration studies performed on larger amounts of material or in biopsies with higher vector loads may give a further statistical confirmation of the absence of rAAV integration hotspots in the clinical context.

Complementary to the retrieval of IS, the multiplex LAM-PCR approach presents the advantage over other PCR-based methods of allowing the study of how internal vector regions are rearranged within the host cell. We show a novel analysis of the structure of internal vector rearrangements that provides new insights into the analysis of vector integrity. Most of the rearrangements found involved the ITR or nearby regions. However, also rare structures were identified exclusively affecting internal vector sequences, thus suggesting the possible engagement of small homologous regions along the rAAV genome. Therefore, studies focusing on the effect of internal homologies over the frequency of rearranged vector structures will allow even further optimization of efficient and safe rAAV vectors.

In conclusion, our study shows the genome wide distribution of rAAV2/5 IS in both human and nonhuman primate livers lacking integration clusters or events in the previously reported HCC-associated genes, thus showing the absence of genotoxicity of the rAAV2/5 in the clinical context. In addition, here we show that M-LAM-PCR is capable of detecting rare rAAV integration events that are originated throughout the rAAV genome and provides information about the integrity of vector genomes persisting within the host cells.

Materials and Methods

Nonhuman primate and human biopsies. Adult Macaca fascicularis (R.C. Hartelust, The Netherlands) were treated as described in Pañeda et al.²⁸ Briefly, macaques were intravenously injected with 1 × 10¹³ or 5 × 10¹³ vector genomes (vg)/kg body weight of the rAAV2/5-cohPBGD vector. One month after vector administration, animals were sacrificed and liver, spleen, and adrenal gland biopsies were collected. Cohorts comprising three animals (two females, one male) for each viral dose were included. The phase 1 rAAV2/5-cohPBGD clinical trial is a phase 1, multicentre, open label, single dose, and dose escalation clinical trial for the treatment of acute intermittent porphyria that, received authorization from the Agencia Española de Medicamentos y Productos Sanitarios on 25 September 2012 and was performed in accordance to good clinical practices (CPMP/ICH/135/95) and the Declaration of Helsinki. All subjects provided written informed consent. The ClinicalTrials.gov identifier is NCT02082860. We analyzed three hepatic biopsies obtained from three acute intermittent porphyria patients, 52 weeks after vector administration, who received 6 × 10¹² vg/kg (Liver 1, standard LAM-PCR analysis) and 5 × 10¹¹ vg/kg (Liver 2 and 3, multiplex LAM-PCR analysis). DNA was isolated from the different biopsies with the AllPrep RNA/DNA Isolation Kit (Qiagen) and used for LAM-PCR analysis.

Detection of rAAV2/5-cohPBGD vector sequences by LAM-PCR and next-generation sequencing. LAM-PCR assays were performed for the 5′ end of the vector on 1 µg of each non-human primate sample and 100–200 ng of the human liver biopsies. Standard LAM-PCR assay was performed as previously described.^27,32 Briefly, LAM-PCR was initiated with two 50-cycles linear PCRs and restriction digest was performed using the MseI restriction enzyme. Two 35-cycles exponential PCRs were performed and LAM-PCR amplicons were visualized on Spreadex (Elchrom Scientific) gels to evaluate PCR efficiency. An additional PCR step was performed with a barcoded primer for library preparation. Multiplex LAM-PCR was performed according to the standard protocol but including primers for 5 different vector regions at a final concentration of 0.167 and 16.67 µM in the linear and exponential amplifications, respectively. Primer sequences are detailed in Supplementary Table S2 and all PCR steps were performed using an annealing temperature of 58 °C. Sequencing of LAM-PCR derived amplicons was conducted on Illumina MiSeq V2 PE 250.

Bioinformatical analysis of IS. LAM-PCR derived sequence reads were aligned to the vector sequence using a locally installed BLAST+ instance configured to gain maximum alignment sensitivity (word size: 4, e value: 10, identity: 75%). This alignment was used to remove (i) sequence reads lacking the vector-specific sequence included in the primer used for library preparation (minimum identity: 95%) and (ii) sequence reads corresponding to a full length vector fragment. Remaining sequences were then analyzed in terms of their 3′ end and divided into: (i) potential IS, sequence reads not aligning to the vector sequence at their 3′ end and (ii) concatemeric sequences, reads discontinuously aligning but in full length to the vector, suggesting vector recombination events exclusively involving the vector genome. In reads corresponding to potential IS, vector fragment was trimmed and the remaining sequence was mapped to the target genome (human: NCBI37 HG19, NHP: BGI CR1.0/rheMac3) using UCSC BLAT to derive chromosomal integration coordinates. Finally, IS were analyzed by automated bioinformatical data mining tools, including identification of nearby genes and other integration features as previously described.³³ Sequences corresponding to concatemeric forms were analyzed in terms of vector fragments identification and determination of the vector coordinates describing the transitions between different vector fragments.

Vector rearrangements reconstruction and analysis. For each primer set s = 1,2,3,4,5 sequence reads were parsed with the aim of detecting signs of vectors fragments in any orientation. Only vector fragments with a length of 16 bases or more are used as rearrangement blocks. The structure of the blocks is then reconstructed on any single read in order to capture the topology of the rearrangements. The recurrence rate of each rearrangement containing r = 1,2,...,n blocks (rearrangement order; n=8 in the most complex case) is then taken into account in order to retrieve the most frequent block topology. For each primer set and rearrangement order the read topology and vector topography for the most common rearrangements is reported and exhibited graphically.

Common IS analysis. Vector integration hotspots were analyzed using the definition of CIS previously established.^34,35 Briefly, a CIS of second order is defined by two IS within 30 kb, a CIS of third order by three IS within 50 kb and a CIS of fourth order by four IS within 100 kb. CIS of fifth, sixth or higher orders are defined by five, six or more IS within a window of 200 kb.

SUPPLEMENTARY MATERIAL Figure S1. Primer efficiency test. Figure S2. Detailed comparison of standard and M-LAM-PCR in NHP tissues. Figure S3. IS distribution within gene-coding and surrounding regions. Figure S4. Structure of vector rearrangements. Table S1. Detailed Common Integration Site (CIS) identified in NHP tissues. Table S2. Primer sequences used for standard and multiplex LAM-PCR.