Integration of Gene Therapy Vectors: A Risk Factor for Tumorigenesis or Another Commensal Property of Adeno-Associated Viruses That Benefits Long-Term Transgene Expression?

Since their beginnings as conceptually viable gene delivery vectors, adeno-associated viruses (AAVs) have been heavily studied and engineered for their application in gene therapy. From this work, over 200 gene therapy clinical trials using recombinant (r)AAVs have been carried out, and seven gene therapies have been approved in select countries for commercial use. ^{1 –4} Despite the exponential growth in the field within the past 5 years, there still persists concerns over whether AAV-based gene therapies are safe.

Aside from adverse effects brought on by transgene toxicity and immune responses, concerns related to vector genome integration remain hotly debated. With current technologies, the frequency of integration, whether they harbor the full transgene cassettes (promoters + cDNAs + polyAs), and whether they cause insertional mutagenesis and/or genotoxicity are still not completely known. Commercially available treatments are relatively new and have only been used in the clinic and in trials within the past 10 years. The long-term outcomes, stability of treatments, and potential risks related to specific vector designs will only be revealed over time.

Most of the research in the AAV field have conventionally fell under the pretexts that they are nonpathogenic, nonintegrating viruses that are ideal as gene delivery vectors. As a consequence, the main scientific advancements have been in enhancing production and transduction of vectors, leaving the scrutiny into AAVs as potential disease risk factors limited.

The first seminal papers regarding integration of AAV revealed site-specific insertion into human chromosome 19, within intron 1 of the protein phosphatase 1 regulatory subunit 12C (PPP1R12C) gene in vitro, ⁵ and the Rian locus in mice. ⁶ These articles were crucial because the former led to the discovery of AAVS1 as a “safe harbor” for gene insertions, while the latter showed that AAV integration was the driver for hepatocellular carcinoma (HCC) in mice. As a result of these contrasting findings, a long-running debate within the community has been related to whether AAV vectors are a risk factor for HCC, with both sides supported by studies using next-generation sequencing (NGS) methods to acquire an incomplete survey of rAAVs integrated into the genomes of target cells. The consensus for some time seemed to be that integration occurred at very low frequencies and do not correlate with tumorigenesis outside of mice. ^7,8 The debate nevertheless continued, since the potential risk for HCC is further elevated when considering that (1) all natural serotypes can efficiently target the liver when delivered systemically, and (2) the liver has been a proposed bio-pump for secreted proteins for many gene therapy modalities. These features established the liver as the proverbial battle ground for determining AAV-based gene therapy safety.

Within the past 10 years, the controversy over whether AAV genome integration is a risk factor has intensified, in part, due to three studies that were categorically paradigm-shifting.

The first was a report by Nault et al., which showed that some patients with HCC had integrations of wild-type (wt)AAV within cell cycle-related genes, with evidence of clonal expansion. This finding offered a potential mechanistic link between AAV insertional mutagenesis and HCC. The recent finding that wtAAV2 harbors a sequence proximal to the 3′-ITR that exhibits liver-specific enhancer function, ⁹ supports the hypothesis that wtAAVs naturally infect hepatocytes, integrate, and drive proximally located genes. In turn, these actions may promote clonal expansion, increase viral copy number during latency, and support viral fitness.

The second paradigm-shifting study by Nguyen et al. described a 10-year follow-up study in rAAV-treated canines, in which gene therapy waned quickly after treatment, followed by stabilized and even increased in expression overtime. ¹⁰ The study suggested that long-term transgene expression in the liver is in part supported by integrated genomes that expanded with the proliferation of transduced cells. Despite the fact that the dogs did not exhibit HCC, the results offered further lines of evidence that vector genome integrations can occur with unintended consequences.

The third study was by Dalwadi et al., ¹¹ which revealed using single molecule, real-time (SMRT) sequencing that integration of rAAVs in liver-humanized mice occurred at frequencies of 1–3%, much higher than previously thought. Although these latter two studies were performed in nonprimate models, the findings have drastically changed our understanding of AAV's safety profile.

These new insights into wtAAV biology and rAAV gene therapy are now crossing paths with the continued advancements that gene therapy treatments are making in the clinic. To date, only one patient receiving AAV-based gene therapy has ever been diagnosed with HCC. However, recent deep molecular analyses of the patient's tumor found that there was no correlation between rAAV genome integration events and tumorigenesis. ¹² Nevertheless, the fact that rAAVs can integrate into the host cell genome cannot be ignored, and their consequences toward gene therapy safety and efficacy is more important than ever.

A manuscript entitled, “Integrated vector genomes appear to be responsible for long-term transgene expression in primate liver following AAV” by Greig et al., which was recently accepted for publication in Nature Biotechnology, has added to this debate. In this study, the authors showed that treatment of AAV vectors led to transient and strong transgene expression. Overtime, vector genomes and transduced hepatocytes were significantly reduced. In the few animals investigated, there were indications of increased transgene expression and vector genome content in liver biopsies 760 days posttreatment. These results mirror those reported in the canine study by Nguyen et al. ¹⁰

Using NGS-based assays (ITR-seq), ¹³ the authors found that integrated vector genomes could explain the phenomena. Intriguingly, among biopsies taken 14 days posttreatment, 0.5–1.25 unique integrations events were observed per 100 genomes. The abundance of detected integrations were reduced overtime (between 14 and 760 days). Importantly, there was no significant evidence of clonal expansion throughout this period. Average clone sizes at unique integration loci were between 2 and 3 clones across all timepoints, and the largest clones among all treated tissue biopsies at day 760 was 14 (under four doublings). Although the time span for the investigation was limited to 2 years, the data suggest that AAV vector integration is not an immediate driver of clonal expansion and tumorigenesis.

A highlight to this investigation was the author's use of SMRT sequencing to capture integrated genomes. Similar to what was observed by Dalwadi et al., the integrated genomes exhibited complex rearrangements and were concatenated. One key finding from this study was the capture of ITR integrity. In past investigations, which relied on linear and/or exponential amplification of integrated vector genomes, only partial ITRs were observed at insertion junctions. It was hypothesized that because chain-termination and sequence-by-synthesis methods (e.g., Sanger- and Illumina-based platforms) have poor processivity across the ITRs, the representation of full ITR species was lost.

Using SMRT sequencing, which has high processivity through ITRs, ¹⁴ the authors showed that 10% of the integrated ITRs were full length at the junctions of insertion, suggesting that integrated genomes can retain full ITRs. Nevertheless, most were truncated or became so upon recombination into the host genome. Despite these intriguing results, the use of SMRT sequencing did not identify full-length cDNAs at integration sites in 4/6 animals. Reads containing full cDNA compositions from one of these animals indicated that 0.8% came from integrated genomes, lending support to the notion that integrated species can indeed contribute to transgene expression in treated tissues.

Although the work by Greig et al. is more than remarkable, it only represents a limited amount of animals, restricted to a 2-year study. It could therefore be argued that the study alone does not provide sufficient insight into whether vector genome integration can play a significant role into stable transgene expression with liver gene therapy. More importantly, the report does not definitively address the question whether these integration events can cause genotoxicity.

A companion paper entitled, “Prevalent and disseminated recombinant and wild-type AAV integration in macaques and humans” by Martins et al., published in this issue of Human Gene Therapy helps to elevate and substantiate the Greig et al. study. This Martins et al. study is to date, the largest investigation of wtAAV genome integrations in humans and nonhuman primates (NHPs) and is the most comprehensive analysis comparing wtAAV and rAAV genome integration in NHPs. In this work, 168 NHP (rhesus and cyno macaques) liver tissues were profiled, and 85 human liver biopsies were interrogated in total.

Using ITR-seq, the authors showed that unique integrations of either wtAAV or rAAV genomes did not appear to be sequence specific. Rather, they tended to be within open regions of the chromosomes (e.g., highly transcribed genes and promoters) and within regions that are susceptible to DNA damage. This data suggests that AAV genome integration is opportunistic. Although events were widespread, there was little evidence of clonal expansion with vector genome integration. When detected, these occurrences were sparse and mostly reflected low copy events. In all, these findings suggest that AAV genome integration is not a clear risk factor for tumorigenesis. These profiles hold true for rAAV integration and in the integration of wtAAV in NHPs naïve to vector treatment. The only difference observed was that integrations by rAAV occurred at higher frequencies than by those seen with wtAAV.

Intriguingly, these findings also propose that wtAAV integration can be used to inform on the behavior of rAAVs following treatment. Case in point, integration events of wtAAV in both human and NHPs were within genes that tended to be highly expressed in the liver, similar to the findings in Greig et al. Whether increases in transgene expression can be contributed by the native enhancers that are proximal to the integrated rAAV genomes, including those specific to liver expression, is indeed an intriguing prospect. Such a concept would validate the need for liver-detargeting design strategies for all nonliver-targeting gene therapy vectors.

Integration of rAAV gene therapy vectors into the host cell genome is a reality and a feature that is shared among wtAAVs and rAAVs. Whether these events are a causative factor for tumorigenesis is not supported by the data presented in the two reports by Martins and Greig et al. The hypothesis that rAAV integration can drive tumorigenesis in humans remains unsubstantiated, at least presently. Nevertheless, the field now has to contend with the fact that the mechanism of transgene stability and expression is not solely through nonintegrated, episomal forms. In fact, it may be that integrated transgene cassettes are what permit the lifetime expression of therapeutic transgenes in the liver. How the field now responds to these new paradigms and what engineering advances to make these gene therapy platforms more efficacious, predictable, and reliable will likely set the tone for future AAV studies.