AAV for gene therapy drives a nephrotoxic response via NFκB in kidney organoids

Dear Editor,

Genome editing holds potential to cure human diseases. The first CRISPR/Cas9 gene therapy, Casgevy, may offer a definitive treatment for hemoglobinopathies through autologous transplantation of ex vivo genome-corrected hematopoietic stem cells. However, in vivo editing of solid tissues has been marred by organ failures and deaths, poorly predicted by animal studies.¹ The lack of reliable preclinical tools for human-specific therapies, such as genome editing, has remained a critical unmet need. Preclinical safety testing of genome editing could benefit from complex tissue models that bear the human genome and reconstitute human responses. Given the kidney’s susceptibility to toxic reactions from investigational drugs, studies using human stem cell-derived kidney organoids may inform clinical trials and improve patient outcome.² Here, kidney organoids serve as a preclinical testing platform for CRISPR/Cas9 genome editing via adeno-associated virus (AAV) delivery, a common strategy in clinical trials that have raised safety concerns due to viral toxicity.¹

AAV2 was chosen to deliver staphylococcus aureus Cas9 (saCas9) and guide RNA (gRNA) to kidney organoids, via a commonly used dual vector strategy that overcomes size limitations of AAV-based delivery.⁴ The DMD locus was targeted with two gRNAs to coordinate a deletion spanning introns 44 to 55 that restores the reading frame, a theoretical treatment for 65% of Duchenne’s Muscular Dystrophy cases. Fluorescent-labeling of the gRNA vector with mCherry, combined with saCas9 immunostaining, reflected co-delivery to 47.2 ± 5.7% of LTL⁺ proximal tubules as compared to modest amounts of CDH1⁺ and PODXL⁺ cells. Given the co-delivery of saCas9 and gRNA to nearly half of the proximal tubules in CRISPR samples, a T7 endonuclease assay was used to detect indel-related DNA mismatches in kidney organoids from both female embryonic stem cells (H9) and male-induced pluripotent stem cells (BJFF). This assay failed to detect on-target DNA mismatches, possibly related to the inaccessibility of the DMD locus, while HK-2 immortalized proximal tubules demonstrated ~50% on-target editing (Fig. 1b). As the T7 endonuclease assay was designed to detect indels in genomic DNA, but not our desired DMD deletion, RNA analysis via spatial transcriptomics (ST) was conducted with specific pobes including DMD. Additionally, CRISPR samples were standardized to empty-vector AAV, and empty-vector AAV was standardized to control, for safety profiling of genome editing and AAV transduction, respectively.

STARmap (spatially-resolved transcript amplicon readout mapping), an ST method coupling hydrogel technology with mRNA rolling amplification and in situ sequencing and in situ hybridization,⁵ included a probe for wild-type DMD designed to target the junction of exons 44 and 45. Although wild-type DMD was poorly expressed at 0.02%-0.03% across cell types, a modest tubular reduction in CRISPR compared to empty-vector AAV samples suggested a quite limited and non-significant degree of on-target editing. Regarding safety profiling, 1486 customized STARmap probes targeting 496 genes included differentially expressed genes (DEGs) in injured proximal tubules and mesenchymal cells in organoids (Zenodo: 10.5281/zenodo.15742418). STARmap resolved a similar profibrotic phenotype in proximal tubules of both CRISPR and empty-vector AAV samples, with DEGs by Model-based Analysis of Single-cell Transcriptomics indicating that AAV alone drove partial epithelial-to-mesenchymal transition (*VIM), interstitial fibrosis (*COL5A1), apoptosis (*CASP8), and NF-κB-related inflammation (*NFKBIA), as standardized to vehicle control. Similarities in the gene expression profiles of CRISPR “loaded” compared to empty-vector AAV “unloaded” samples confirm that AAV infection itself, as opposed to genome editing, drives transcriptional differences between samples. A profibrotic and senescent (*CDKN1A) phenotype was common to “loaded” and “unloaded” samples by whole RNA sequencing (GEO GSE286037). Next, immunostaining confirmed γH2AX and CDKN1A positivity localized to LTL⁺ proximal tubules, which manifest RELA of canonical NF-κB signaling as a potential druggable therapeutic pathway. The extensive DNA damage in these highly transduced tubules likely stems from the DNA release step of AAV infection, where viral uncoating triggers nuclear stress, genotoxicity, and a strong DNA damage response. IL-1β and IL-6, biomarkers of tubular injury and activators of the NF-κB cascade, were also detected in organoid supernatants to establish AAV transduction as a primary instigator of a nephrotoxic response (Fig. 1c). As the genomic differences amongst AAV serotypes are often limited to the tropism-defining variable region of the capsid sequence, namely viral protein 3 (VP3), the nephrotoxic results of AAV2 may extend beyond AAV2 to any naturally-occurring or synthetic serotype with tubular tropism.

Mechanistic testing was conducted using bardoxolone, an FDA safety-approved agent and potent inhibitor of the IKK kinase complex central to canonical NF-κB signaling. Following concurrent IKK inhibition (IKKi) during AAV transduction (day 21 to 28), longitudinal monitoring to day 42 found that IKKi significantly rescued AAV-induced COL1A1⁺ interstitial fibrosis in kidney organoids from both male and female, embryonic and induced pluripotent stem cells. Reduced fibrosis correlated with IL-1β and IL-6 expression falling to control levels, whereas senescence in injured tubules reduced from 69.2 ± 6.6% to 33.4 ± 5.5%. Although continued treatment to day 42 may have been more beneficial, IKKi did not mediate its significant nephroprotective effect by limiting AAV transduction, which would be counterintuitive to genome editing, as evidenced by the preserved transgenic delivery of GFP to tubules (Fig. 1d).

The translational utility of kidney organoids may be limited by their immaturity, which may over- or under-estimate the tropism and toxicity of AAVs in the kidney. Additionally, the lack of physiologic vasculature in organoids, which prevents the assessment of different AAV administration routes (e.g., intravenous, intra-arterial, retrograde ureteral), and the inability to mimic glomerular filtration dynamics, poses further challenges. Although AAVs are below the size threshold of glomerular sieving at ~25 nm, systemically delivered viral particles would largely traverse the glomerulus and reach the peritubular capillary bed since the glomerular filtration fraction is typically 20%. Here kidney organoids were bathed in viral particles to bring AAV in contact with the basolateral membrane of tubules, a region typically supplied by the peritubular capillary bed. Despite these limitations, the model still offers valuable insights into AAV tropism and toxicity, especially regarding proximal tubule effects. Future improvements, such as incorporating vascularized structures, may further enhance the physiological relevance of these models.

Genome editing holds great promise to revolutionize medicine through curative treatment, however, early efforts have raised unanticipated safety concerns in clinical trials. A review of 255 AAV-based gene therapy trials reported 11 patient deaths, independent of the AAV serotype used or route of administration.¹ Clinical translation of genome editing may benefit from therapeutic strategies delineated in preclinical studies. Using kidney organoids, we find that AAV transduction induces genotoxicity and IL-1β secretion that activates the canonical NFκB pathway, whose inhibition halts interstitial fibrosis by preventing tubular senescence. When targeting alternative organs, IKKi may prevent AAV nephrotoxicity due to off-target kidney tropism. Beyond safety testing, kidney organoids may be an optimal preclinical platform for on-target efficacy testing, as tubulopathies largely comprise the >600 monogenic kidney diseases which constitute 50% of CKD in children and 30% of non-diabetic CKD in adults. Our findings demonstrate the utility of kidney organoids as a preclinical testing platform for genome editing, which may inform clinical trials of strategies that limit organ failures and death.