AAV immuno-gene therapy platform delivering vectorized cytokines defines a new modality for high-grade glioma treatment

Abstract

High-grade gliomas are among the most treatment-refractory solid tumors, with profound immunosuppression, diffuse infiltration, and resistance to existing immunotherapies. Interferons (IFNs) have potent anti-tumor and immunostimulatory activities, but systemic IFN therapies have failed in solid tumors due to poor pharmacokinetics and dose-limiting toxicities. Here, we introduce vectorized IFNs (recombinant adeno-associated virus [AAV] gene therapies enabling sustained, localized IFN expression within tumors) as a new class of biologics for high-grade glioma treatment. We engineered AAV vectors expressing human IFNα1, IFNβ, IFNγ, or combinations thereof and demonstrated potent, selective anti-tumor activity in patient-derived glioblastoma organoids while sparing healthy brain cells. Transcriptomic profiling revealed a durable, IFN-specific response in treated tumor cells. AAV-hIFNβ was advanced to orthotopic patient-derived and cell line-derived xenograft models, where intratumoral convection-enhanced delivery produced complete tumor clearance and significantly prolonged survival. Spatial transcriptomic and histologic analyses showed tumor-localized IFN expression, immune activation, and apoptosis-mediated tumor regression. To enable clinical translation, we developed SRN-101, a clinical-grade AAV-hIFNβ construct with optimized potency and manufacturability, which achieved superior expression and in vivo efficacy vs. the research-grade vector. These findings establish AAV-vectorized cytokines as a durable, locally delivered immuno-gene therapy platform for high-grade gliomas.

Graphical abstract

AAV-vectorized interferon-β converts immunologically cold high-grade glioma into an immune-reactive tumor microenvironment. Local cytokine expression induces tumor apoptosis, microenvironment remodeling, and durable tumor clearance in orthotopic models, establishing vectorized cytokines as a new immuno-gene therapy strategy for high-grade gliomas.

Introduction

High-grade gliomas (HGGs) are the most common and lethal primary malignant brain tumors. Glioblastoma (GBM), the most prevalent HGG subtype, accounts for over 13,000 new diagnoses in the United States and approximately 225,000 deaths globally each year.¹ Despite maximal safe resection followed by conformal radiotherapy with concurrent/adjuvant temozolomide (TMZ) chemotherapy, median overall survival (mOS) remains just 14.6 months and 5-year survival rates are only 6.9%.¹ No effective standard of care exists for recurrent disease, underscoring the urgent need for new therapeutic modalities capable of improving outcomes in this uniformly fatal cancer.

Interferons (IFNs) are pleiotropic immunomodulatory cytokines with potent intrinsic anti-tumor properties that act broadly and independently of mutational context for tumor origin.² They are central to both basal cancer immunosurveillance and therapy-induced immunogenic responses. Indeed, many standard and experimental cancer therapies, ranging from radiotherapy to agents designed to stimulate pattern recognition receptors and pro-inflammatory signaling pathway members (e.g., TRAIL, STING, RIG-I, cGAS, etc.), achieve much of their efficacy through the induction of local IFN signaling. Endogenous IFNs orchestrate diverse anti-tumor functions, including suppression of angiogenesis,³ upregulation of tumor MHC-I,⁴ promotion of apoptosis,⁵ attenuation of immunosuppressive myeloid populations,⁶ activation of dendritic cells to mature and cross-prime T cells against tumor antigens,⁷ and increase the proliferation of activated macrophages and natural killer (NK) cells⁸ (Figure 1A). However, despite their potent biology, recombinant IFNs have demonstrated limited clinical efficacy when administered systemically due to short plasma half-lives and dose-limiting toxicities,⁹ highlighting the need for localized, durable, and tumor-confined delivery systems.

Figure 1.

Vectorized interferons drive potent, durable, and selective cytotoxicity in primary human glioblastoma organoids ex vivo

(A) Schematic overview of the immunomodulatory and anti-neoplastic mechanisms of interferons. (B) Time course of interferon secretion in HEK293T cells treated with AAV9-hIFNα1 (top), AAV9-hIFNβ (middle), AAV9-hIFNγ (bottom), or media (untreated, blue), quantified by ELISA. MOI = 4.0E5 vg/cell. (C) Experimental design for co-culture of primary human glioblastoma (GBM) tumor organoids with healthy human cerebral organoids ex vivo. (D) Representative live confocal images of co-culture organoids treated with PBS, AAV9-GFP, TMZ (37 μM in dimethyl sulfoxide [DMSO]), DMSO controls, and a panel of AAV9-IFNs. Primary human GBM tumor cells, red; healthy human cerebral organoids, green. All AAV treatment groups were performed at MOI = 500k; scale bar, 300 μm. (E) Quantification of mean GBM tumor cell fluorescence intensity over time from (D). Mean (blue line) and 95% confidence intervals (light blue) are shown for 6 replicate wells for each condition and time point. (F) Quantified tumor cell fluorescence in co-culture organoids treated with PBS, DMSO, TMZ, AAV9-GFP, and AAV9-hIFNβ. All AAV treatment groups were performed at MOI = 500k. (G) Confocal imaging of human GBM organoids treated with PBS or AAV9-hIFNβ. All AAV treatment groups were performed at MOI = 500k; scale bar, 250 μm. (H) Quantification of GBM organoid growth relative to untreated controls from (G). ∗∗∗∗∗p < 0.0001, unpaired t-test. (I) Representative time course images of healthy cerebral organoids treated with PBS, TMZ (37 μM), or AAV9-hIFNβ. All AAV treatment groups were performed at MOI = 500k; scale bar, 300 μm. (J) Healthy cerebral organoid area quantification from (I). ∗∗∗∗∗p < 0.0001, unpaired t-test. (K) Time course quantification of tumor cell fluorescence in co-culture organoids treated with AAV9-GFP, AAV9-hIFNβ, or TMZ in combination with AAV9-GFP or AAV9-hIFNβ. All AAV treatment groups were performed at MOI = 500k.

Recombinant adeno-associated viruses (rAAVs) are attractive vectors for therapeutic gene delivery due to their proven safety profile, non-lytic and non-pathogenic nature,¹⁰ and primarily episomal persistence¹¹ that minimizes insertional mutagenesis risks. These features of rAAV set it apart from the oncolytic viruses commonly used in neurooncology, which have repeatedly failed to extend survival in HGG clinical trials, including HSV-1,¹² adenovirus,¹³ reovirus,¹⁴ paramyxovirus,¹⁵ H1 parvovirus,¹⁶ and poliovirus.¹⁷ Oncolytic virus genomes (recognized as foreign by pattern recognition receptors like toll-like receptors, RIG-I/MDA5, and cGAS) act as potent pathogen-associated molecular patterns, triggering innate immune cascades. These pathways involve the recruitment and activation of DNA methyltransferases, which have been shown to methylate both host and viral regulatory regions, silencing antiviral genes and genomes, likely leading to direct methylation and silencing of oncolytic viral genomes, which may explain their repeated clinical failures in HGG. In comparison, rAAV is neither pathogenic nor is its genome silenced; indeed, these are two of the requirements for efficient gene transfer gene therapy for monogenic disorders. Moreover, rAAVs are replication incompetent, incapable of cell-to-cell spread, and cannot survive cell growth/division,¹⁸ thus providing naturally limiting expression of gene therapy payloads and making them ideally suited for localized, controllable delivery of potent immunotherapies.

The field of gene therapy has robust precedence for AAV delivery to the central nervous system (CNS) via both diffusion-based (intrathecal, intracerebroventricular, and intracisternal) and convection-enhanced delivery (CED)¹⁹ routes. CED, which employs positive pressure gradients to achieve bulk flow through brain parenchyma, enables homogenous vector distribution across clinically meaningful volumes²⁰ and can be performed with real-time MRI guidance in the clinic.²¹ This combination of spatial precision and safety makes CED an optimal strategy for both focal and infiltrative brain tumors.

Here, we combine the potent immunostimulatory activity of IFNs, the safety and durability of rAAV vectors, and the precision of CED to create a novel AAV immuno-gene therapy platform for HGG. We demonstrate that a single intratumoral administration of our lead therapeutic, SRN-101, drives local and sustained IFN expression, reprograms the tumor microenvironment toward immune activation, and induces potent anti-tumor responses in multiple HGG models. These results demonstrate a generalizable strategy for converting non-oncolytic viral gene therapy platforms into first-in-class precision immunotherapies, redefining how localized cytokine delivery can be used to treat solid tumors of the CNS and beyond.

Results

AAV immuno-gene therapies mediate durable, selective anti-tumor activity in primary human GBM organoids

To evaluate the therapeutic potential of AAV-based immuno-gene therapy for solid tumors, we generated a panel of recombinant AAV vectors encoding mono-, bi-, or tri-cistronic cassettes expressing human IFNα1, IFNβ, or IFNγ, and fluorescent reporter controls. All constructs were packaged in the clinically validated AAV9 capsid, chosen for its broad transduction profile and established safety record in human gene therapy. Vector potency and cytokine expression were confirmed in HEK293T cells (Figure 1B), after which candidate vectors were screened for anti-tumor activity in primary human GBM organoids.

We and others have previously established robust organoid co-culture models that recapitulate the invasive growth and cellular heterogeneity characteristic of human GBM.^22,23,24 To establish this ex vivo model of human GBM, we co-cultured freshly resected human GBM cells labeled with mScarlet (red) together with healthy cerebral organoids labeled with green fluorescent protein (GFP, green) (Figure 1C). Live confocal microscopy revealed that the primary GBM cells formed spheres that then invaded the healthy cerebral organoid, faithfully mirroring the infiltrative pattern of patient tumors (Video S1).

To benchmark the therapeutic efficacy of our AAV immuno-gene therapies, co-cultures were treated with vehicle controls (PBS, DMSO), AAV9-GFP, the standard of care chemotherapeutic temozolomide (TMZ), or one of the vectorized IFN constructs. Live imaging over 15 days demonstrated that control treatments did not affect tumor growth, while TMZ caused marked cytotoxicity in healthy cerebral organoids and only transiently suppressed tumor proliferation (Figures 1D and 1E). In contrast, all AAV-delivered IFNs selectively reduced GBM tumor burden without detectable toxicity to healthy cerebral cells (Figures 1D and 1E). AAV9-hIFNβ was prioritized for further study given the extensive clinical experience with recombinant IFNβ protein and its well-characterized toxicity profile when administered systemically or locally^25,26,27 (Table S1), and because AAV9-hIFNβ produced a pronounced anti-tumor response in deeper replicate organoid studies (Figure 1F).

To assess tolerability, AAV9-hIFNβ was tested in isolated GBM and healthy cerebral organoids separately. Cytocidal activity after AAV9-hIFNβ treatment was observed exclusively in GBM organoids (Figure 1G,H, p < 0.0001 by unpaired t-test) and not healthy organoids (Figures 1I and 1J), indicating tumor-selective cytotoxicity. In contrast, TMZ significantly impaired the growth of healthy organoids (Figures 1I and 1J; p < 0.0001 by unpaired t-test). Finally, to determine whether combination therapy could enhance efficacy, GBM organoids were co-treated with TMZ and AAV9-hIFNβ. The combination did not outperform AAV9-hIFNβ monotherapy but was more effective than TMZ plus AAV9-GFP (Figure 1K), suggesting that AAV-mediated IFNβ expression alone is sufficient to achieve maximal anti-tumor activity ex vivo.

Vectorized IFNβ drives sustained cytokine expression and prolonged transcriptional activity in human GBM cells

To compare the durability and transcriptional effects of vectorized vs. recombinant IFNβ, we treated human GBM6 cells—a well-characterized, EGFR-amplified, EGFRvIII-mutant line that faithfully models invasive GBM²⁸—with AAV9-hIFNβ, recombinant hIFNβ protein (r-hIFNβ), AAV9-GFP, or vehicle control. Before functional studies, vector integrity was confirmed by demonstrating full-length genome packaging, absence of rearrangements, and minimal host-cell DNA impurities (Figure S1A).

To simulate physiological clearance kinetics in treated GBM6 cells, media in the r-hIFNβ-treated cultures was replaced every 5 h during the first 20 h post-treatment and in all groups every 24 h thereafter. Media and cells were collected at 24, 48, 72, and 96 h for hIFNβ protein quantification and transcriptomic profiling. As expected, extracellular hIFNβ levels in r-hIFNβ-treated cells fell below the detection threshold (<23 pg/mL) within 24 h due to rapid cytokine decay (Figure 2A). In contrast, AAV9-hIFNβ treatment induced detectable hIFNβ secretion within 24 h, reaching >150 pg/mL by 72 h and remaining elevated throughout the 96 h time course despite daily media exchanges. These findings demonstrate that vectorization effectively circumvents the short half-life limitation of exogenous cytokine therapy, producing sustained supraphysiological IFNβ levels in vitro.

Figure 2.

Vectorized IFNβ drives durable signaling and complete tumor regression in human glioblastoma models in vivo

(A) Sustained hIFNβ secretion in human GBM6 cells treated with AAV9-hIFNβ (red, MOI = 4E5 vg/cell) or recombinant hIFNβ cytokine (r-hIFNβ, purple, 47 IU/mL, equivalent to 114 pg/mL), measured by ELISA at indicated time points. 50% media washouts every 5 h for the first 20 h in the r-hIFNβ condition mimic in vivo cytokine clearance (half-life = 4–5 h). Full media exchanges were performed at 24, 48, 72, and 96 h post-treatment. (B) Number of differentially expressed genes (DEGs, p-Adj<0.01) in GBM6 cells 24–96 h post-treatment with AAV9-hIFNβ or r-hIFNβ vs. media controls. (C) Enrichment scores for type I IFN and TNFα response pathways across treatments and time points. (D) Heatmap of the top 10 IFN and TNFα response genes (Log2FC vs. media controls) in GBM6 cells treated as in (A). (E) Schematic of orthotopic PDX (SF11411) and cell line-derived xenograft ([CDX], GBM6-FLuc) studies in athymic nu/nu mice treated intratumorally with saline, AAV9-GFP, or AAV9-hIFNβ via CED. (F) Kaplan-Meier survival curves for PDX mice treated as in (E). Saline = black, AAV9-GFP (2E11 vg/brain) = blue, AAV9-hIFNβ (2E11 vg/brain) = red. Vertical dashed line = day of treatment (day 9). p < 0.04 by log-rank (Mantel-Cox) test. n = 30 (10 per treatment arm). (G) Longitudinal BLI of GBM6-FLuc tumor growth in CDX mice treated as in (E). Saline = black, AAV9-GFP (2E11 vg/brain) = blue, AAV9-hIFNβ (2E11 vg/brain) = red. Thin lines = individual mice, thick lines = geometric mean. Vertical dashed line = day of treatment (day 9). ∗p < 0.04 by Kruskal-Wallis test with Dunn’s multiple comparisons correction on day 22. n = 30 (10 per treatment arm). (G′) Representative BLI images from each treatment group 11 days post-treatment. (H) Kaplan-Meier survival curves for CDX mice. p < 0.001 by log-rank (Mantel-Cox) test. (I) Distribution of treatment responses in CDX by BLI flux (photons/second) at day 27. Tumor free = BLI flux <2.5 × 10⁵ p/s, tumor reduction = ≥30% decrease from assignment on day 9, no change = between 30% decrease and 20% increase from assignment on day 9, tumor growth = ≥20% increase from assignment on day 9, death = mice that died before day 27. (J) Dose-response analysis of AAV9-hIFNβ efficacy in CDX mice. AAV9-GFP (2E11 vg/brain) = blue, AAV9-hIFNβ hi (2E11 vg/brain) = solid red, and AAV9-hIFNβ lo (1E11 vg/brain) = dashed red. Thin lines = individual mice, thick lines = geometric mean. Vertical dashed line = day of treatment (day 9). ∗∗p < 0.02 by Kruskal-Wallis test with Dunn’s multiple comparisons correction on day 20. n = 45 (15 per treatment arm). For data interpretation, tumor burden threshold = 2.5 × 10⁵. (J′) Representative BLI images of tumors 11 days post-treatment. (K) Kaplan-Meier survival curves from (J). p < 0.002 (AAV9-hIFNβ hi), p < 0.005 (AAV9-hIFNβ lo) by log-rank (Mantel-Cox) test compared to AAV9-GFP. (I) Distribution of treatment responses in CDX mice at day 27 by BLI flux as in (I).

Sustained cytokine expression corresponded closely with the transcriptional landscape. The number of differentially expressed genes (DESeq2, p-Adj<0.01 vs. media) over time paralleled cytokine kinetics, with transient induction of transcriptional response following r-hIFNβ and prolonged activation following AAV9-hIFNβ (Figure 2B; Table S2). Gene set enrichment analysis revealed type I IFN and TNFα signaling as the top upregulated pathways in both treatments, but these signatures waned rapidly in r-hIFNβ-treated cells, while persisting or intensifying in the AAV9-hIFNβ group (Figures 2C and 2D). No enrichment of these pathways was observed in AAV9-GFP controls, confirming a specific IFNβ-mediated response.

Together, these data establish that vectorized delivery of IFNβ induces a durable, tumor-intrinsic proinflammatory transcriptional program that cannot be achieved with recombinant cytokine administration. Based on these robust and sustained effects, AAV9-hIFNβ was selected for further efficacy evaluations in multiple orthotopic GBM models in vivo.

AAV9-hIFNβ induces durable tumor regression and complete responses in orthotopic GBM xenograft models

To assess therapeutic efficacy in vivo, we tested AAV9-hIFNβ in human orthotopic patient-derived xenograft (PDX) and cell line-derived xenograft (CDX) models of GBM. For the PDX model, freshly resected primary GBM cells from an adult patient with IDH-wild-type, CNS WHO grade 4 disease were transplanted into the brains of athymic nu/nu mice. These mice retain functional innate immune compartments (NK cells, dendritic cells, macrophages, B cells, and naive T cell precursors) but lack mature cytotoxic T cells, allowing human tumor engraftment while preserving innate immune responses.²⁹

To recapitulate the intended clinical route of administration, we used a murine-compatible CED device for localized intratumoral infusion. Before treating tumor-bearing animals, CED infusion of AAV9-hIFNβ was evaluated in naive mice and confirmed to be well tolerated with no acute neurotoxicity (Figure S1B). Thirty PDX-bearing mice (n = 10 per arm) were treated intratumorally via CED with saline, AAV9-GFP (2 × 10¹¹ vector genomes [vg]/brain), or AAV9-hIFNβ (2 × 10¹¹ vg/brain) (Figure 2E). Kaplan-Meier analysis revealed a significant survival benefit in the AAV9-hIFNβ group (mOS: 32 days, p < 0.04, log-rank [Mantel-Cox] test), with 30% of animals surviving beyond 65 days, compared with mOS of 28 days (saline) and 27 days (AAV9-GFP) and 0% survival to day 65 in controls (Figure 2F). No difference was observed between saline and AAV9-GFP groups, confirming that the survival advantage resulted from IFNβ expression rather than vector or infusion effects.

To extend these findings, we next evaluated AAV9-hIFNβ in the orthotopic GBM6-FL CDX model, a reproducible and widely used benchmark for GBM therapeutics. Thirty tumor-bearing mice (n = 10 per arm) were treated intratumorally via CED with saline, AAV9-GFP (2 × 10¹¹ vg/brain), or AAV9-hIFNβ (2 × 10¹¹ vg/brain) and monitored longitudinally by bioluminescent imaging (BLI). Tumors in saline- and AAV9-GFP-treated controls grew rapidly, whereas AAV9-hIFNβ induced significant tumor regression by day 22 (Figure 2G; p < 0.04, AAV9-hIFNβ vs. AAV9-GFP, Kruskal-Wallis test with Dunn’s multiple comparisons correction). Correspondingly, AAV9-hIFNβ-treated animals exhibited a striking improvement in survival, with 60% of mice surviving beyond day 65 compared to 0% in both control groups (Figure 2H; mOS 20 and 21 days, respectively; p < 0.001, log-rank [Mantel-Cox] test). At 2.5 weeks post-treatment, responses included 22% tumor free, 11% tumor reduction, 22% tumor growth, and 44% death, whereas all saline and AAV9-GFP controls succumbed to disease (Figure 2I).

To evaluate dose dependence and long-term durability, we performed a subsequent 45-mouse study (n = 15 per arm) to compare high-dose (2 × 10¹¹ vg/brain) and low-dose (1 × 10¹¹ vg/brain) AAV9-hIFNβ to AAV9-GFP (2 × 10¹¹ vg/brain) controls. BLI again confirmed rapid tumor progression in controls and marked regression in AAV9-hIFNβ high-dose-treated animals (Figure 2J; p < 0.02, AAV9-hIFNβ high vs. AAV9-hIFNβ low, Kruskal-Wallis test with Dunn’s multiple comparisons correction). Remarkably, 33% of mice receiving the high dose remained tumor-free for the entire 9-month study duration (295 days) with no evidence of recurrence (Figure 2K; p < 0.002, AAV9-hIFNβ high vs. AAV9-GFP, log-rank [Mantel-Cox] test). Even the lower dose conferred a significant survival benefit (p < 0.005, log-rank [Mantel-Cox] test) compared to controls. Treatment response at 2.5 weeks in the high-dose group included 31% tumor-free, 19% tumor reduction, 6% no change, 6% tumor growth, and 38% death, vs. 100% mortality in AAV9-GFP-treated animals (Figure 2L).

Collectively, these studies demonstrate that intratumoral AAV9-hIFNβ delivery elicits robust, dose-dependent, and durable anti-tumor effects across orthogonal orthotopic GBM models. Complete, long-term responses in a subset of animals support the potential for durable disease control following localized AAV-mediated cytokine therapy.

Species-specific interferon biology limits the interpretability of syngeneic GBM models

To evaluate our vectorized cytokine therapy in an immunocompetent setting, we next tested AAV9-mediated IFNβ delivery in the murine GL261 syngeneic allograft model—the most widely used immunocompetent model of GBM.³⁰ Because GL261 tumors are of mouse origin, we employed a species-matched murine IFNβ (mIFNβ) payload (AAV9-mIFNβ). Model establishment and vector validation are shown in Figure S2. To evaluate the efficacy of AAV9-mIFNβ against GL261 tumors, 45 wild-type C57BL/6 mice (n = 15 per arm) were implanted orthotopically with luciferase-expressing GL261 cells (GL261-FL) and treated intratumorally with saline, AAV9-GFP (2 × 10¹¹ vg/brain), or AAV9-mIFNβ (2 × 10¹¹ vg/brain) (Figure 3A).

Figure 3.

The standard GL261 syngeneic GBM model fails to recapitulate interferon responses due to species-specific differences

(A) Schematic of the syngeneic allograft study in immunocompetent C57BL/6J mice orthotopically engrafted with GL261-FL cells and treated intratumorally via CED with saline, AAV9-GFP (2E11 vg/brain), or AAV9-mIFNβ (2E11 vg/brain). (B) Longitudinal BLI of FLuc+ tumor growth in mice treated as in (A). Saline = black, AAV9-GFP = green, AAV9-mIFNβ = orange. Thin lines = individual mice, thick lines = geometric mean. Vertical dashed line = day of treatment (day 5). n = 45 (15 per arm). (C) Kaplan-Meier survival curves for treated mice from (B). p = 0.05 by log-rank (Mantel-Cox) test for AAV9-mIFNβ vs. AAV9-GFP. (D) UMAP of single-cell RNA-seq profiles from brains of mice treated with saline, AAV9-GFP, or AAV9-mIFNβ, colored by cell type (left) and treatment type (right). (E) Differential gene expression analysis (DESeq2) of tumor cells from AAV9-mIFNβ-treated vs. AAV9-GFP-treated mice. Red = p-Adj<0.01, gray = p-Adj>0.01. (F) Differentially expressed genes in mouse PBMCs treated with mIFNβ vs. media control (left), human PBMCs treated with hIFNβ vs. media control (middle), and mouse PBMCs treated with hIFNβ vs. media control (right). Significantly differentially expressed genes (p-Adj<0.01) = red, with select genes labeled. (G) Venn diagrams showing minimal overlap of significantly upregulated (left) and downregulated (right) genes between mouse and human PBMCs treated with species-matched IFNβ. (H) Concordance plot comparing fold change responses to species-matched IFNβ in mouse vs. human PBMCs reveals no correlation (R² = 0.012), indicating that interferon responses are strongly species specific.

Longitudinal BLI revealed no significant difference in tumor progression across treatment groups (Figure 3B), and no complete responses were observed. Nevertheless, AAV9-mIFNβ treatment produced a modest improvement in survival compared to controls (Figure 3C; p = 0.05, AAV9-mIFNβ vs. AAV9-GFP, log-rank [Mantel-Cox] test). To explore the mechanistic basis of this partial response, we performed single-cell RNA sequencing (RNA-seq) of brain tumors from each treatment cohort. Few differentially expressed genes were detected between AAV9-mIFNβ-treated and AAV9-GFP-treated tumors (Figures 3D and 3E; Table S3; DESeq2, p-Adj<0.01, AAV9-mIFNβ vs. AAV9-GFP), suggesting that AAV9-mIFNβ failed to induce a robust IFN-stimulated gene signature in vivo. In vitro experiments confirmed that GL261 cells are susceptible to AAV9 transduction and express the mIFNβ payload at higher levels than GBM6 cells transduced with AAV9-hIFNβ at the same MOI (Figure S5), supporting that the reduced tumor responses to IFN treatment in mouse GL261 tumors compared to human GBM6 tumors was not driven by differences in AAV9 transduction or payload expression efficiency.

This apparent paradox prompted a closer examination of species-specific IFN biology. Prior studies have shown that murine and human cells exhibit markedly divergent transcriptional responses to IFNβ,^31,32 but these differences are often overlooked in preclinical immunotherapy studies. To directly test this, we profiled the transcriptomic response of mouse and human peripheral blood mononuclear cells (PBMCs) following treatment with species-matched recombinant IFNβ. Human PBMCs exhibited a dramatically stronger transcriptional response than mouse PBMCs to species-matched IFNβ (Figure 3F; Table S4). Only 8% of upregulated genes and 2% of downregulated genes in human PBMCs treated with hIFNβ were shared in mouse PBMCs treated with mIFNβ (Figure 3G), and no global correlation was observed between the human and mouse transcriptomic changes (Figure 3H).

Together, these data indicate that species-level differences in IFNβ signaling underlie the weak efficacy observed in the GL261 model. Consequently, syngeneic murine allografts are likely unsuitable for evaluating IFN-based immunotherapies, including AAV9-hIFNβ, due to their poor predictive value for human responses. For subsequent in vivo studies, we therefore focused on the human GBM6 cell line-derived xenograft (CDX) model, which provides superior pharmacological relevance and reproducibility compared to both syngeneic and PDX systems.

Vectorized hIFNβ induces tumor apoptosis and reactive astrogliosis, leading to complete tumor clearance in vivo

To investigate the mechanism and temporal dynamics underlying the potent tumor regression observed in orthotopic CDX mice in vivo, we performed a detailed time course analysis in GBM6-FL CDX mice treated with AAV9-hIFNβ (2 × 10¹¹ vg/brain). Tumor-bearing animals were collected before treatment (0 h) and at 2, 24, and 48 h and 7 days post-infusion for histological and molecular characterization (Figure 4A).

Figure 4.

Vectorized hIFNβ induces rapid tumor apoptosis and reactive astrogliosis in orthotopic human GBM xenografts

(A) Schematic of the early-treatment kinetics in human GBM6-FLuc cell line-derived xenograft (CDX) mice treated intratumorally via CED with AAV9-hIFNβ (2E11 vg/brain). (B) Time course brain histopathology and immunohistochemistry (H&E, human IFNβ, Ki67, cleaved caspase-3, IBA1, and GFAP) of coronal brain sections collected pre-treatment (0 h) and at 2, 24, and 48 h and 7 days post-treatment. All images were acquired under identical imaging conditions. Black/white dashed lines indicate regions shown in inset panels. Progressive loss of tumor tissue (H&E, Ki67) coincides with transient hIFNβ expression, induction of apoptosis (cCaspase-3), and activation of resident microglia/macrophages (IBA1⁺) and astrocytes (GFAP⁺), consistent with a localized apoptotic and neuroinflammatory response during tumor clearance. (C) Luminex analysis of serum cytokines at the indicated time points following treatment. Only hIFNβ was detected, with transient expression peaking within 48 h and returning to baseline by day 7. Values represent fold-change (pg/mL) relative to the untreated (0 h) samples; color scale denotes relative upregulation (red) or no change (gray).

Immunohistochemistry (IHC) revealed rapid and dynamic changes within the tumor following AAV9-hIFNβ administration (Figures 4B and S3). Human IFNβ protein was detectable as early as 2 h post-treatment, remained elevated through 48 h, and declined by day 7, coinciding with tumor clearance. RNAscope analysis confirmed the presence of hIFNβ transcripts at 2 h and absence at baseline, verifying that protein staining reflected vector-mediated expression rather than endogenous production by GBM6-FL cells (Figure S4).

At baseline (0 h), tumors displayed dense proliferation (Ki67⁺), minimal microglial or macrophage activation (IBA1⁺), and limited reactive astrogliosis (GFAP⁺) confined to the tumor border. Following AAV9-hIFNβ treatment, we observed the emergence of intratumoral apoptotic foci (cleaved caspase-3⁺ [cCASPASE-3]) that paralleled the kinetics of hIFNβ expression. By day 7, both proliferative and apoptotic signals were absent, and H&E confirmed complete histologic tumor clearance. Notably, reactive astrogliosis and microglial/macrophage activation expanded into both the region that had contained the tumor and the surrounding parenchyma, consistent with localized immune activation and tissue remodeling in response to tumor cell death.

These findings suggest a mechanism in which AAV9-hIFNβ initiates a rapid, tumor-targeted apoptotic cascade that is subsequently amplified by brain-resident immune cells, culminating in complete tumor elimination and localized inflammatory repair. Supporting this interpretation, multiplex Luminex analysis of serum cytokines across the same time points detected only human IFNβ—with no evidence of human IFNα1 or IFNγ—confirming selective vector-driven cytokine expression (Figure 4C). Further, circulating hIFNβ levels peaked within 48 h and declined to baseline by day 7, temporally aligned with tumor regression.

Together, these data reveal that intratumoral AAV9-hIFNβ acts through a dual mechanism of tumor-intrinsic apoptosis and immune-mediated clearance, resulting in rapid and durable elimination of GBM tumor cells.

Vectorized hIFNβ triggers a rapid and spatially confined interferon response within the tumor microenvironment

To dissect the spatial and transcriptional dynamics underlying AAV9-hIFNβ-induced tumor clearance, we performed spatial transcriptomic profiling using the Visium platform (10× Genomics) on GBM6-FL CDX mouse brains collected before (0 h) and 48 h after AAV9-hIFNβ treatment (2 × 10¹¹ vg/brain)—the time point corresponding to peak hIFNβ activity (Figure 4B). Spatial transcriptomics captures spatially resolved gene expression by sequencing mRNA from tissue sections overlaid onto barcoded capture spots. Following alignment and normalization, shared nearest neighbor clustering was used to distinguish transcriptionally unique regions. These regions were identified as tumor, peri-tumor, and normal brain tissue based on spatial position and known marker genes. Canonical correlation analysis (CCA) was used to integrate datasets across time points and visualize treatment-associated transcriptional changes. Spatial transcriptomics was performed on representative tumors at baseline and at 48 h post-treatment to capture the spatial dynamics of the IFN response. Expression data were aligned to the human genome to capture tumor-specific changes, supplemented by a custom probe set of mouse immune-response genes (Table S5) to assess local host responses.

Shared nearest neighbor clustering defined distinct tumor regions at both time points, characterized by canonical GBM marker expression (CD44, VIM, TOP2A, NOTCH1³³; Figures 5A–5C). A peri-tumoral cluster expressing GBM-associated transcripts but located beyond the H&E-defined tumor border was also identified, consistent with the infiltrative nature of human glioma. Consistent with the local delivery route, vector-derived hIFNβ transcripts were detected primarily within tumor and peri-tumor regions 48 h post-treatment (Figure 5D). Tumor and peri-tumor cells at 48 h exhibited pronounced upregulation of IFN-stimulated genes,³⁴ including CXCL10, IFIT1, IFIT2, MX1, OASL, and RTP4, confirming a robust IFNβ-specific transcriptional response (Figures 5B–5D). Although only modest changes were observed in the surrounding mouse brain, several inflammatory and immune activation genes (Gfap, Ifitm3, and Irf7) were upregulated within or adjacent to the tumor, consistent with the reactive gliosis and microglial activation observed by IHC (Figures 4 and 5E).

Figure 5.

Spatial transcriptomics reveals rapid, localized transcriptional remodeling of the tumor microenvironment following vectorized hIFNβ treatment

(A) Coronal brain sections from representative human GBM6-FLuc CDX mice collected pre-treatment (0 h, n = 1) or 48 h (n = 1) after intratumoral AAV9-hIFNβ infusion (2E11 vg/brain), stained with H&E (left) and subjected to Visium Spatial Gene Expression profiling (right). Annotated clusters were assigned based on anatomical localization and marker gene expression. Dashed lines denote tumor borders. Scale bars, 1 mm. (B) Top 10 marker genes for each spatially resolved cluster identified across 0 h and 48 h datasets. Values are shown as log-normalized expression centered at 0 (Seurat “scale.data”). (C) Spatial expression of canonical human GBM tumor markers (CD44, VIM, TOP2A, and NOTCH1) delineating tumor and peri-tumor regions before (0 h) (top) and after (bottom) (48 h) AAV9-hIFNβ treatment. (D) Expression maps of the human IFNβ payload and hallmark IFN-response genes (CXCL10, IFIT1, and IFIT2), demonstrating tumor-restricted transgene expression and induction of an IFN-specific transcriptional program within 48 h. (E) Spatial expression of host mouse immune-response genes (Gfap, Ifitm3, and Irf7) showing localized activation of astroglial and innate immune pathways proximal to the tumor. (F) Integrated datasets (0 and 48 h) visualized using canonical correlation analysis (CCA), showing distinct clustering of tumor and peri-tumor regions (left) and enrichment of IFN-response gene module expression (right). (G) Volcano plot depicting differential gene expression between 0- and 48-h tumor clusters. Red, IFN-response genes; gray, other significantly upregulated genes (p-Adj <0.01); blue, non-significant. (H) Top enriched Gene Ontology (GO) terms among upregulated genes in 48-h tumor cells, highlighting interferon and inflammatory response pathways (∗∗p-Adj <0.01; ∗p-Adj <0.05).

Canonical correlation analysis integrating all spatial transcriptomes distinguished tumor tissue from surrounding brain parenchyma while clearly separating 0 h and 48 h tumor clusters (Figure 5F), reflecting extensive treatment-induced transcriptional remodeling. Differential expression analysis in the 48 h tumor cluster compared to 0 h baseline identified 7,850 upregulated and 1,763 downregulated genes (Figure 5G; Table S6; Log₂FC > 1, p-Adj<0.01; Seurat non-parametric Wilcoxon rank-sum test). Gene set enrichment analysis revealed IFN and inflammatory response pathways as among the most significantly upregulated terms (Figure 5H; Table S7), indicating coordinated activation of innate immune signaling in response to AAV9-hIFNβ treatment.

Collectively, these findings demonstrate that AAV9-hIFNβ induces rapid, spatially restricted transcriptional reprogramming of the glioma microenvironment characterized by strong IFN signaling within tumor and peri-tumor regions and localized inflammatory activation of adjacent glial tissue.

SRN-101 is a clinically optimized vectorized cytokine demonstrating enhanced potency and durable efficacy in vivo

Building on our preclinical data, we next developed a translational AAV immuno-gene therapy candidate suitable for clinical and commercial advancement. The resulting vector, SRN-101, was engineered to incorporate genomic and regulatory elements compatible with clinical safety and manufacturing standards, while retaining immunostimulatory properties of the research-grade (RG) AAV9-hIFNβ construct. The following studies were designed to evaluate whether vector optimization for clinical translation impacts potency or in vivo efficacy, rather than to dissect the function of individual genomic elements or report Good Manufacturing Practice (GMP) release characteristics. To enable regulatory advancement, SRN-101 was produced under current GMP-like (cGMP-like) conditions with an industry Contract Development and Manufacturing Organization (CDMO) partner, yielding purified material with exceptional potency and critical quality attributes. In vitro, SRN-101 produced under both RG and cGMP-like conditions demonstrated significantly greater potency than the prior AAV9-hIFNβ vector (Figure 6A; p < 0.005 at MOI 8E5 vg/cell, unpaired t-test).

Figure 6.

SRN-101 demonstrates enhanced potency, in vivo efficacy, and translational readiness for clinical advancement

(A) In vitro potency of AAV9-hIFNβ (red) and SRN-101 vectors produced under RG (pink) or cGMP-like (maroon) conditions, compared to media (black), quantified by hIFNβ secretion via ELISA. ∗∗∗P < 0.005 SRN-101 RG vs. AAV9-hIFNβ RG by unpaired t-test at MOI 8E5 vg/cell. HEK293T cells were used for all conditions. Media-only controls were measured at a single baseline time point (0 h) and used as a background reference and, therefore, are shown as a single data point rather than a kinetic curve. (B) Tumor growth kinetics, and (C) Kaplan-Meier survival curve in the historical academic GBM6-FLuc CDX model treated via intratumoral CED (saline, AAV9-hIFNβ). Dashed lines mark day of treatment; n = 10 per arm. (C′) Representative H&E image at treatment, dashed outline = tumor border. (D) Tumor growth and (E) survival in the translational industry-grade human GBM-SRN01 CDX model. Dashed lines mark day of treatment; n = 5 per arm. (E′) Representative H&E section showing reproducible tumor morphology at infusion. (F) Time course BLI of FLuc⁺ tumor burden in treated industry-grade CDX mice. Treatment arms: formulation buffer (black), AAV9-mCardinal high-dose RG (blue), SRN-101 high-dose RG (2E11 vg/brain; pink), SRN-101 high-dose cGMP-like (maroon solid), SRN-101 low-dose cGMP-like (maroon dashed), and historical AAV9-hIFNβ RG (red). Vertical dashed line = day 45 (treatment). n = 90 (15/arm). ∗∗∗p < 0.005, ∗∗∗∗p < 0.002 by Kruskal-Wallis with Dunn’s correction on day 105. (G) Kaplan-Meier survival analysis of treated industry-grade CDX mice. Treatment groups as in (F). p < 0.004 by log-rank (Mantel-Cox) test, SRN-101 high RG vs. AAV9-mCardinal. (H) Distribution of treatment responses at day 91 based on BLI flux (photons/second). Tumor-free, <2.5 × 10⁵ p/s; tumor reduction, ≥30% decrease vs. baseline day 43; no change, −30% to +20%; tumor growth, ≥20% increase; and death, mice deceased before day 91.

To evaluate SRN-101 efficacy in vivo, we established a scalable, industry-compatible GBM CDX model (GBM-SRN01) designed for stereotactic infusion with the ClearPoint Neuro CED system, mirroring the intended clinical procedure. This model generated reproducible, 100% lethal human GBM tumors with a progressive disease course and uniform lethality, consistent with our prior academic CDX model, although with slower tumor growth kinetics (Figures 6B–6E).

In these orthotopic CDX GBM mice, used for all of Figure 6, local CED infusion of SRN-101 (2 × 10¹¹ or 5 × 10¹⁰ vg/brain) produced durable anti-tumor responses compared to vehicle or AAV9-mCardinal controls (n = 15 per arm). BLI revealed marked tumor regression across all SRN-101 treatment arms (Figure 6F; p < 0.002, SRN-101 vs. controls at day 105, Kruskal-Wallis test with Dunn’s multiple comparisons correction), with significantly greater efficacy than the historical AAV9-hIFNβ vector (p < 0.005 at day 105, historical AAV9-hIFNβ vs. SRN-101, Kruskal-Wallis test with Dunn’s multiple comparisons correction). Kaplan-Meier survival analysis demonstrated profound survival benefit, with >85% of SRN-101-treated animals surviving to study completion (Figure 6G; mOS undefined; p < 0.004 by log-rank (Mantel-Cox) test, SRN-101 RG vs. AAV9-mCard control). Response assessment at 2.5 weeks post-treatment revealed tumor-free responses in 46%–80% of SRN-101-treated mice, compared to 0% in controls (Figure 6H). Equivalent efficacy between RG and cGMP-like SRN-101, and across both dose levels, underscored the robustness of vector design and manufacturing reproducibility. Notably, SRN-101 achieved greater therapeutic benefit at lower doses than its AAV9-hIFNβ predecessor (Figures 2J–2L and 6F–6H).

Collectively, these results establish SRN-101 as a potent, manufacturable, and clinically translatable AAV immuno-gene therapy with robust efficacy in rigorous orthotopic GBM models. By combining the durable expression profile of AAV with the pleiotropic anti-tumor activity of IFNβ, SRN-101 represents a new therapeutic class capable of overcoming key limitations of cytokine protein therapies and oncolytic viruses for HGG treatment.

Discussion

Despite the transformative success of immunotherapies in many cancers, their efficacy in solid tumors remains limited by intrinsic barriers such as immunosuppressive tumor microenvironments, antigen heterogeneity, antigen loss, and poor immune infiltration. These challenges are particularly pronounced in HGGs, which remain among the most treatment-refractory cancers. HGGs are characterized by an immunologically “cold” microenvironment, scarce neoantigen expression, and the added obstacle of the blood-brain barrier, which restricts systemic drug access to the tumor. Even the most advanced recent cell therapies face limited durability in glioma^35,36,37 due to heterogeneous antigen expression, immune evasion, and profound antigen loss. Consequently, nearly all patients experience recurrence and 5-year survival remains below 7%.

Here, we introduce a non-targeted, broad-acting immunotherapy platform that combines the safety and durability of rAAV gene therapy with the pleiotropic, tumor-suppressive properties of IFN cytokines to effectively treat solid tumors such as HGGs. Across multiple preclinical models, AAV-vectorized human IFNβ produced durable payload expression, selective tumor cell killing, and long-term survival benefit. Mechanistically, we show that treatment elicits rapid, tumor-specific apoptosis followed by reactive astrogliosis and local immune activation, culminating in complete tumor clearance in a substantial fraction of orthotopic xenograft animals. Transcriptomic and spatial profiling confirmed a sustained IFN response within the tumor microenvironment, underscoring the potency and specificity of this platform.

Approved medicines for HGG³⁸ (TMZ, bevacizumab, and nitrosourea-based chemotherapies) extend median survival by only a few months and often compromise quality of life. These interventions are further limited by hematologic, gastrointestinal, and neurologic toxicities, highlighting the unmet need for durable, well-tolerated treatments. IFN cytokines have long been recognized for their therapeutic potential, but short systemic half-lives, dose-limiting toxicities, and inefficient tumor delivery have hindered clinical translation. Vectorization of IFNs via AAV overcomes these barriers by enabling continuous, localized expression of cytokines within the tumor milieu, achieving pharmacodynamic exposure that cannot be matched by recombinant protein administration.

AAV vectors also possess an unparalleled safety record in the CNS: they are non-pathogenic, replication incompetent, and largely non-integrating and have been used safely in thousands of patients across approved and investigational therapies.^39,40 Pairing AAV delivery with CED further improves precision delivery by bypassing the blood-brain barrier, reducing systemic exposure, and allowing real-time MRI-guided control of vector distribution in patients.⁴¹ Direct intratumoral infusion also reduces the total viral dose required compared to systemic administration, dramatically lowering manufacturing costs and greatly minimizing the risk of potential immunogenicity—significant advantages for scalable, accessible, and safe CNS gene therapy.

AAV can transduce multiple CNS cell types following parenchymal delivery, raising the question of payload expression in non-tumor cells. In the therapeutic context described here, vectors are administered locally using CED, which enables controlled spatial distribution within the brain. Clinically, this approach is paired with real-time MRI guidance, allowing infusion parameters and vector spread to be monitored and adjusted in real time during administration. As a result, vector exposure is confined to the intended target region and payload expression is spatially restricted. Consistent with this, IFN signaling and downstream transcriptional responses in our studies were concentrated within the tumor and peri-tumor regions and diminished rapidly following tumor clearance. Together, these considerations support the feasibility of localized AAV-mediated IFN delivery for glioma treatment, while minimizing exposure to surrounding brain tissue.

Comparison with existing cytokine gene therapy approaches^42,43,44,45 provides additional context for the translational impact of SRN-101. Several others are also advancing AAV platforms delivering cytokine payloads (e.g., IL-12) to stimulate immune infiltration in GBM. While IL-12 and related cytokines are potent immune activators, IFNβ offers several mechanistic and translational advantages. IFNβ exerts both direct anti-proliferative and pro-apoptotic effects on tumor cells and broad immune modulation, converting “cold” gliomas into inflamed, immunoreactive microenvironments. In contrast, interleukin-based strategies depend on sufficient immune cell presence and antigen recognition—features notably lacking in most gliomas. Moreover, IFNβ induces a diverse IFN-stimulated gene program, enhances antigen presentation, and limits immunosuppressive myeloid activity, yielding a more comprehensive anti-tumor response.⁴⁶ Finally, our data demonstrate complete and durable tumor regression in multiple orthotopic GBM models at clinically feasible doses, positioning SRN-101 as a mechanistically superior approach.

The findings presented here support translation of SRN-101 as a focal immuno-gene therapy for HGGs. Clinically, this approach is envisioned for patients with intact tumor tissue, with vector infusion administered only into tumor tissue using MRI-guided CED. This is particularly pertinent in recurrent disease, which has no effective standard of care.⁴⁷ Due to the risks and limited potential for benefit, HGG patients typically do not undergo re-resection following recurrence. Given that their tumor tissue remains intact, the recurrent setting can avoid the challenges associated with infusion into post-resection cavities. MRI-guided CED enables real-time visualization and control of AAV to the exact site of infusate placement,⁴⁸ allowing precise tissue targeting. By confining vector exposure to the intended anatomical space, MRI-guided CED provides a clinically feasible means to achieve localized, sustained IFN expression within an otherwise immunologically cold tumor microenvironment. Importantly, there is clinical precedence for both local delivery of AAV vectors, including AAV9, to the CNS⁴⁹ and for the use of MRI-guided CED for intratumoral infusion of therapeutics in HGG.^21,50 Together, these considerations define a realistic and neurosurgically aligned path for clinical translation of AAV-based IFN therapy in recurrent HGGs.

Collectively, these data establish AAV-vectorized IFN therapy as a powerful new modality for treating solid tumors such as HGG. By combining precision local delivery, sustained cytokine expression, and intrinsic tumor-killing activity, SRN-101 addresses key limitations of both recombinant cytokine therapy and oncolytic virotherapy. Ongoing and future studies in large animal tumor models with cross-reactivity to human IFNs will further characterize their safety, biodistribution, and translational readiness. Ultimately, this platform may enable a new generation of gene-encoded, locally delivered immunotherapies that are agnostic to mutational status and applicable across a spectrum of solid tumors.

Materials and methods

AAV productions

Recombinant single-stranded AAV vector productions were manufactured at research grade (RG) using standard transient triple transfection protocols in adherent HEK293 cells, followed by double cesium chloride density gradient purification, desalting, and filter sterilization. Plasmids for triple transfection included a pAAV adenoviral helper, our custom transfer vectors expressing various engineered IFN cytokines or GFP synthesized at GenScript, and an AAV-Rep2Cap9 packaging plasmid. All vectors were confirmed endotoxin-free using a limulus amebocyte lysate assay before use. cGMP-like productions of SRN-101 and control vectors were manufactured by Catalent in suspension HEK293 cells using scalable manufacturing workflows; detailed process parameters are proprietary and, therefore, not disclosed.

Fast-Seq validation of packaged viral genomes

To validate packaged genome identities and integrity, we performed whole-genome next-generation sequencing via Fast-Seq.⁵¹ Total packaged genomic DNA (gDNA) was extracted from 1 × 10¹¹ full AAV particles for each AAV vector. AAV libraries were prepared following the Fast-Seq Tn5 tagmentation-based protocol. The following indexed adapters compatible with Illumina were obtained from IDT:

Index 1 (i7):

CAAGCAGAAGACGGCATACGAGAT NNNNNNNNNNNN GT CTCGTGGGCTCGG

Index 2 (i5):

AATGATACGGCGACCACCGAGATCTACAC NNNNNNNNNNNN TCGTCGGCAGCGTC

Bold = P5/P7 adapter sequence

Italics = unique 12-bp barcode index

Underline = primer for mosaic ends added during tagmentation.

Each adapter contained a 12-nucleotide unique barcode to identify samples after multiplexing. The resulting library was diluted to 10 pM in 600 mL of HT1 hybridization buffer (Illumina Nextera XT kit Cat#FC-131-1024), and 10 mL was loaded onto a 300-cycle MiSeq Nano v.2 flow cell (Illumina Cat#MS-102-2002) for paired-end PE75 sequencing. Resultant reads were demultiplexed using Illumina’s bcl2fastq v.2.19.0.316. Data were returned in fastq format and filtered using Trimmomatic to remove adapter sequences, low-quality reads (PHRED score <30, or length <50 bp), and unmapped and unpaired reads. Trimmed reads were then aligned to the AAV transfer vector plasmid reference sequence with BWA v.0.7.17, using the MEM algorithm. Alignments were saved as BAM files, which were then used to generate VCF files using the GATK HaplotypeCaller algorithm. SNPs and indels identified in VCF files were filtered using the BCFtools filter algorithm, with a 15X depth threshold and a 90% allele fraction requirement. A consensus sequence was generated using the BCFtools consensus algorithm. Alignment and fragment distribution statistics were obtained with Picard tools. Coverage spanned 100% of the AAV transfer vector genome reference sequence, including the ITRs.

Vector potency confirmation by IFN ELISA

To validate vector potency (both infectivity and expression) before use in mice, HEK293T cells were transduced with AAV9-IFNα1, AAV9-IFNβ, or AAV9-IFNγ or treated with media alone and cell supernatants were evaluated by ELISA for IFNα1, IFNβ, or IFNγ. HEK293T cells were maintained in DMEM (Corning Cat#10-013-CV) supplemented with 10% FBS (Bio-Techne Cat#S12450H), 1X GlutaMAX (Gibco Cat#35050061), and 1X antibiotic/antimycotic (HyClone Cat#SV30079). 10,000 cells/condition were plated in 48-well TC-treated plates. 24 h after plating, cells were transduced with AAV9-IFNα1, AAV9-IFNβ, AAV9-IFNγ, or media control at an MOI of 4E5 vg/cell. 24, 48, 72, and 96 h after treatment, cell supernatants were collected and IFN variant levels were measured using IFN ELISA kits following the manufacturer’s instructions (human IFNα [PBL Cat# 41135-1], human IFNβ [PBL Cat#41410], and human IFNγ [R&D Systems Cat#: DIF50C]). Experiments were performed in technical quadruplicate.

For potency described in Figure 6A, HEK293T cells were transduced with SRN-101 or AAV9-hFINβ at MOI 5E4, 2E5, 4E5, and 8E5 vg/cell, and cell supernatants were evaluated by ELISA for hIFNβ (PBL Cat#:41410) 48 h after treatment. Experiments were performed in technical triplicate.

Human GBM tissue

De-identified tumor tissue SF11411 was obtained fresh from the UCSF Brain Tumor Center Tissue Bank IRB#10-01318 from an adult female patient with a confirmed CNS WHO grade 4 IDH-wild-type GBM. These studies were in accordance with the ethical standards of the Institutional Research Committee and with the 1964 Helsinki declaration and its later amendments.

Primary human glioblastoma cell culture

SF11411 cells were maintained at 37°C in a 5% CO₂ atmosphere with 21% oxygen and grown in a 1:1 ratio of DMEM/F12 (Life Technologies) and Neurobasal medium (Life Technologies) supplemented with 5% FBS (Life Technologies), B-27 supplement without vitamin A (Life Technologies), N-2 supplement (Life Technologies), 1X GlutaMAX (Life Technologies), 1 mM NEAA (Life Technologies), 100 U/mL antibiotic-antimycotic (Life Technologies), 20 ng/mL hEGF (R&D Systems), and 20 ng/mL hFGF2 (PeproTech). Cells were validated using short-tandem repeat profiling at the UCSF Clinical Cancer Genomics Laboratory.

Development of healthy human cerebral organoid cultures

Healthy human cerebral organoids were created from astroglia induced from pluripotent human stem cells,⁵² maintained at 37°C in a 5% CO₂ atmosphere with 21% oxygen. In brief, free-floating neuroepithelial aggregates from WTC11 cells were induced by dual SAMD inhibition with SB431542 and DMH-1 (2 μM each, Tocris Cat#HY-10431 and #4126, respectively). On day 5, aggregates were seeded to 100 μg/mL on Matrigel-coated (Corning Cat#356231) plates and cultured in DMEM/F12 (Life Technologies Cat#11320), supplemented with 2 μg/mL heparin (Sigma Cat#H3149), B-27 supplement without vitamin A (Life Technologies Cat#12587), N-2 supplement (Life Technologies Cat#17502), and 100 U/mL antibiotic-antimycotic (Life Technologies Cat#15240), referred to as NPC medium. Between 5 and 7 days after attachment, rosettes were mechanically isolated and transferred to a culture flask and maintained in NPC medium for 7 days. Thereafter, rosette aggregates were further cultured in 10 ng/mL recombinant human EGF (R&D Systems Cat#236-EG) and 10 ng/mL recombinant human FGF2 (PeproTech Cat#100-18C) to induce astroglial differentiation over the course of 5 months. Before primary human GBM organoid co-culture experiments, the healthy cerebral organoids were transduced with lentivirus expressing a membrane-bound GFP (Addgene Cat#22479) to label all healthy cells green and primary human GBM tumor cells were transduced with lentivirus expressing mScarlet (Addgene Cat#159172) to label all tumor cells red.

Human glioblastoma organoid co-culture conditions ex vivo

For co-culture experiments, premature 5-month-old healthy cerebral organoids were dissociated with StemPro Accutase (Thermo Fisher Scientific Cat#A1110501). Five thousand single-cell suspension astroglial cells were mixed with 2,000 primary GBM cells and were seeded into each well of a PrimeSurface ultra-low attachment V-shaped 96-well plate (S-Bio Cat#MS-9096VZ). On day 3, astroglia/GBM spheroids were transferred to an imaging microplate (Corning Cat#4515). Co-culture spheroids were maintained for 14 days in NPC medium supplemented with 10 ng/mL hEGF and 10 ng/mL hFGF2.

AAV transduction in human glioblastoma organoids ex vivo

On the day of treatment, the media was changed before adding each respective drug treatment. Negative controls included no treatment, DMSO diluent (control for TMZ), and AAV9-GFP; positive control was TMZ (37 μM) in DMSO. All AAV treatment groups were performed at MOI 500k. There were ∼7,000 cells/organoid in 300 μL media. Half of the media was changed every other day (150 μL of old media was exchanged for 150 μL of fresh media). Live confocal imaging was performed on the indicated days post-treatment to both assess cellular health and measure organoid volume and intensity.

Live confocal fluorescent imaging of human glioblastoma organoids ex vivo

A Cell Observer spinning disc confocal microscope (Zeiss, Oberkochen, Germany) fitted with a temperature and carbon dioxide-controlled chamber was used to live image GBM cells co-cultured with cerebral organoids. Organoids were imaged at the stated times using a 10× objective with 0.4 numerical aperture.

Confocal organoid image analysis

To estimate organoid and tumor cell volume and intensity, the 3D confocal stacks were first convolved with a 3D Laplacian of Gaussian filter with sigma of Gaussian = 2 to reduce the uniform blur and identify the volume occupied by the cells. The filtered image was intensity thresholded with Otsu’s method to obtain the binary images of cells. Volumes of organoid and tumor were estimated by counting the voxels of cells in each channel. Mean fluorescence intensity was computed by averaging voxel intensity over the segmented volumes. Due to scattering by the organoids, confocal imaging could only resolve the cells up to ∼20 microns into the organoids. Image analysis was performed in Python 3.7 using SciPy 1.5.2 and scikit-image 0.17.2. Intensity plots of the mean with the 95% confidence interval were generated using Seaborn 0.11.1.

GBM6 cell culture and treatment conditions for transcriptomic profiling by RNA-seq

GBM6 cells were maintained at 37°C in a 5% CO₂ atmosphere and cultured in 1X DMEM (Corning #10-013-CV), 1X NEAA (Corning #25-025-CI), 1X anti-anti (penicillin G/streptomycin/amphotericin B, Cytiva #SV30079.01), 10% FBS (Bio-Techne #S12450H). To prepare GBM6 cells for treatment and RNA-seq, cells were plated on day 1 in a 48-well plate at 20,000 cells per well and allowed to adhere overnight. Additional wells were seeded to facilitate counting for MOI calculations. On day 2, cells from the extra seeded wells were lifted with TrypLE (Gibco Cat# 12604013) and spun at 300 × g for 5 min to collect the cell pellet. Cells were then resuspended and counted using trypan blue on the Countess 3 automated cell counter. Cell count values were averaged, and this value was used for MOI calculations. Cells were treated on day 2 with the following: media as a negative control, AAV9-hIFNβ (MOI 4E5), or r-hIFNβ (47 IU/mL, equivalent to 114 pg/mL,PBL Assay Science Cat#11415-1, Lot:7600). Supernatants were collected from r-hIFNβ, AAV9-hIFNβ, and media groups at 5, 10, 15, and 24 h post-treatment. During these same time points, 50% media exchanges were performed in the r-hIFNβ and media groups to simulate the in vivo clearance of IFNβ protein. Every 24 h, a full media exchange occurred in all treatment conditions. Supernatants and cell lysates were collected in all treatment conditions at 24, 48, 72, and 96 h post-treatment. Cell lysates were collected by removing media in the wells and lysing cells with PureLink Genomic Lysis/Binding Buffer (Invitrogen, Cat #K182302). Cell lysates and supernatants were transferred to −80°C for storage before protein and/or RNA analysis. ELISA for hIFNβ protein in supernatants was performed using a commercial kit (PBL Cat#41410) according to the manufacturer’s instructions. RNA-seq of cell lysates is described below.

GBM6 and GL261 cell culture and treatment conditions for measurement of supernatant hIFNβ and mIFNβ levels by ELISA

GBM6 cells cultured in 1X DMEM (Corning #10-013-CV), 1X NEAA (Corning #25-025-CI), 1X anti-anti (penicillin G/streptomycin/amphotericin B, Cytiva #SV30079.01), and 10% FBS (Bio-Techne #S12450H) and GL261 cells cultured in 1X DMEM (Corning #10-013-CV), 1X GlutaMAX (Gibco #35050061), 1X anti-anti (penicillin G/streptomycin/amphotericin B, Cytiva #SV30079.01), and 10% FBS (Bio-Techne #S12450H) were both maintained at 37°C in a 5% CO₂ atmosphere. To prepare GBM6 and GL261 cells for treatment, cells were plated on day 1 in a 48-well plate and allowed to adhere and grow to confluence overnight. Additional wells were seeded to facilitate counting for MOI calculations. On day 2, cells from the extra seeded wells were lifted with TrypLE (Gibco Cat# 12604013) and spun at 300 × g for 5 min to collect the cell pellet. Cells were then resuspended and counted using trypan blue on a Countess 3 automated cell counter. Cell count values were averaged, and this value was used for MOI calculations. Cells were treated on day 2 with the following: media as a negative control, AAV9-mCard (MOI 4E5), AAV9-hIFNβ (MOI 4E5), or AAV9-mIFNβ (MOI 4E5). Supernatants were collected from all groups at 24, 48, and 72 h post-treatment. ELISA for hIFNβ protein (PBL Cat# 41415) and mIFNβ protein (PBL Cat# 42410) in supernatants was performed using a commercial kit according to the manufacturer’s instructions.

Human and mouse PBMC culture and treatment conditions for transcriptomic profiling by RNA-seq

PBMCs were acquired from iQ Biosciences as either individual donors (human) or donor pools (mouse). Human donor 1 (Donor 9279: LotP23C0900: IQB-PBMC102) had the following demographics: age: 21, female, Hispanic, 57 kg, height: 160 cm, blood type: O+, smoker: no. Human donor 2 (Donor 7165: LotP23C0970: IQB-PBMC102) had the following demographics: age: 34, female, African American, 65 kg, height: 173 cm, blood type: B+, smoker: no. Human donor 3 (Donor 7925, Lot #P23H1800, IQB-PBMC102) had the following demographics: age: 42, female, African American, 96 kg, height: 168 cm, blood type: O+, smoker: no.

PBMCs were plated in TC-treated 48-well plates at a density of 500k viable cells per well or non-TC-treated 24-well plates at a density of 1 million viable cells per well in 1 mL of media per well (1X DMEM, 1X NEAA, 1X anti-anti [penicillin G/streptomycin/amphotericin B], 10% FBS). Human PBMCs were treated with either recombinant human IFNβ (PBL Cat#11415-1) or media control and placed in an incubator (37°C, 5% CO₂, with saturating humidity) for 24 ± 1 h. Mouse PBMCs were treated with recombinant mouse IFNβ (PBL Cat#12405-1 Lot:7573R), recombinant human IFNβ, or media control and placed in an incubator (37°C, 5% CO₂, with saturating humidity) for 24 ± 1 h. At the time of collection, plates were spun in a swinging bucket rotor centrifuge (350 × g, 5 min). Supernatant was aspirated, and RNA extractions from cells were performed using the PureLink Pro 96 RNA Purification Kit (Invitrogen Cat#12173011A).

Bulk RNA-seq

Sequencing was performed at Novogene (Sacramento, CA). Briefly, polyA-enriched RNA-seq libraries were sequenced on an Illumina NovaSeq 6000, with PE150 reads, targeting 6 GB raw data per sample. Reads were aligned to a custom genome made using Cell Ranger v.7.0.0 from GRCh38.p14 to include the fluorescent reporters and the IFNβ sequences used in our AAV constructs. Gene-wise inclusion criteria were set at >25 reads in >3 samples in any treatment group. Normalization for sequencing depth, as well as differential expression analysis, was carried out in R using the DESeq2 package v.1.42.0. Computational analysis methods are available on GitHub: https://github.com/SirenBio/aav_immuno_gene_therapy_for_HGG.

Enrichment scores for treatment conditions were calculated compared to their time-matched media-only-treated controls using IFN⍺ and TNF⍺ response Gene Ontology terms.³⁴

Animals for academic studies

Adult 5- to 6-week-old athymic nude mice (homozygous nu/nu) were purchased from Harlan Laboratories (Cat#490) as recipients for both CDX and PDX xenografts. Adult 5- to 6-week-old C57BL/6 mice (C57BL/6NCrl) were purchased from Harlan Laboratories (Cat#027) as recipients for allografts. All mice were housed under specific-pathogen-free housing conditions and were given continual access to food and water ad libitum. The Institutional Animal Care and Use Committee of UCSF approved all mouse procedures. Before treatment, all mice were randomized based on BLI tumor sizing to ensure similar average tumor sizes across groups. There was no blinding during the in vivo experiments. Animals were euthanized when pre-determined Institutional Animal Care and Use Committee (IACUC)-approved endpoints were reached.

Animals for industry studies

Adult 7- to 9-week-old athymic nu/nu mice (Jackson Laboratories, strain #002019) were used as recipients for CDX xenografts. All mice were housed under specific-pathogen-free housing conditions and were given continual access to food and water ad libitum. The Institutional Animal Care and Use Committee of Crown Bioscience and Siren Biotechnology approved all mouse procedures. Before treatment, all mice were randomized based on BLI tumor sizing to ensure similar average tumor sizes across groups. There was no blinding during the in vivo experiments. Animals were euthanized when pre-determined IACUC-approved endpoints were reached.

Intracranial orthotopic tumor establishment in mice

The following parameters were controlled for in all mouse experiments: (1) age-matched mice were used to prevent confounding effects of age-related AAV transduction,⁵³ (2) AAV vectors were produced at the same time by the same method and thawed simultaneously to control for vector deterioration effects,⁵⁴ and (3) all mice were housed by treatment groups to prevent cross-contamination via vector shedding. For the animal studies described in Figures 2, 3, and 4, all procedures were carried out under sterile surgical conditions by neurosurgeons at the UCSF Brain Tumor Center. Briefly, mice were anesthetized by intraperitoneal injection of a mixture containing ketamine (100 mg/kg) and xylazine (10 mg/kg). The scalp was swabbed with 2% chlorhexidine, 20–30 μL of 0.25% bupivacaine was injected into the intra-cutaneous space of the scalp, and a skin incision ∼15 mm in length was made over the middle frontal to parietal bone. The surface of the skull was exposed so that a small hole could be made with a 25G needle 3 mm to the right of bregma on top of the coronal suture. A 26G needle attached to a Hamilton syringe was inserted into the hole in the skull. The needle was covered with a sleeve that limits the depth of the injection to 3.5 mm. In all cases (see below for specifics to each cell type), 300k cells in a 3-μL suspension were injected very slowly (∼3 μL/min) by hand, and then the needle was removed. The skull surface was swabbed with hydrogen peroxide before the hole was sealed with bone wax to prevent reflux. The scalp was closed with surgical staples. In all cases, treatment and vehicle negative controls included 10–15 similarly transplanted mice treated with AAV9-GFP and 10–15 similarly transplanted mice treated with vehicle (saline). All mice with FLuc^pos tumors were imaged 1–2 times/week and monitored for survival. Mice with FLuc^neg tumors were only observed for survival.

For orthotopic PDX studies, primary human adult female SF11411 GBM tumor cells were grown and passaged as described above, and 300k tumor cells were injected intracranially into anesthetized athymic recipients in a volume of 3 μL. Seven days post-transplant, once the tumor growth was in log phase, AAV9-hIFNβ (2 × 10¹¹ vg/brain), AAV9-GFP (2 × 10¹¹ vg/brain), or saline was administered via CED in a volume of 15 μL.

For orthotopic CDX studies, immortalized donor human GBM6-FL tumor cells from a 65-year-old male were a kind gift from Dr. David James at UCSF. Cells were grown and passaged as described above, and 300k tumor cells were injected intracranially into anesthetized athymic recipients in a volume of 3 μL. Nine days post-transplant, once the tumor growth was in log phase, AAV9-hIFNβ (2 × 10¹¹ vg/brain), AAV9-GFP (2 × 10¹¹ vg/brain), or saline was administered via CED in a volume of 15 μL. In the long-term survival study, 4/15 mice were euthanized due to observations of rectal prolapse. These were considered in the professional judgment of the animal care staff as unrelated to treatment and were excluded from Figure 2K. These mice had no evidence of tumor burden before reaching humane endpoints and were included in BLI and treatment response assessments in Figures 2J and 2L.

For orthotopic syngeneic allograft mouse studies, immortalized male GL261-FL mouse glioma tumor cells were acquired from the NCI Tumor Repository (Frederick, MD). Cells were maintained at 37°C in a 5% CO₂ atmosphere and cultured in 1X DMEM, 1X GlutaMAX supplement, 1X anti-anti, and 10% FBS. A total of 300k GL261-FL tumor cells were injected intracranially into anesthetized C57BL/6 recipients in a volume of 3 μL. Five days post-transplant, once the tumor growth was in log-phase, AAV9-mIFNβ (2 × 10¹¹ vg/brain), AAV9-GFP (2 × 10¹¹ vg/brain), or saline were administered via CED in a volume of 15 μL.

For the animal studies described in Figures 5 and 6, all procedures were carried out under sterile surgical conditions at Crown Bioscience. Briefly, mice received 1 mg/kg of 1 mg/mL slow-release buprenorphine subcutaneously 30 min before anesthetization with 1.5%–2.0% isoflurane at a flow rate of 0.4–0.8 L/min. Mice were positioned in a stereotaxic head frame, and eye lubricant was applied. The skin on the head of each mouse was disinfected with betadine, followed by 70% alcohol, and a 2- to 3-mm-long incision was made at the cranial midline. A hydrogen peroxide solution was used to clean away tissue from the skull and better visualize cranial suture lines. Origin coordinates (x, y, and z) were taken at the bregma region of the brain, where the needle touches the surface of the skull. The drill was then moved to the following coordinates toward the right frontal lobe: x = 2.0 mm right lateral and y = 0.5 mm anterior to bregma, and a ∼0.3-mm-diameter hole was drilled in the skull. Mice were inoculated intracranially with 300k viable GBM-SRN01 tumor cells in 3 μL of PBS at the right frontal lobe (see below for cell culture conditions). Chilled cells were drawn up into a Hamilton syringe with a 33G needle, which was already mounted onto the stereotaxic frame. At the burr hole, the needle was inserted z = 3.5 mm into the brain and 3 μL of cell suspension was slowly injected at a rate of 1 μL/min. The needle was slowly withdrawn and the mounted Hamilton syringe was flushed with sterile cell culture-grade water, followed by an ethanol wipe between each animal. Bone wax was used for the closure of the hole in the skull. The skin incision was closed with tissue glue, and the mice were placed in a warmed recovery cage. In all cases, treatment and vehicle negative controls included 10–15 similarly transplanted mice treated with AAV9-mCardinal and 10–15 similarly transplanted mice treated with vehicle (saline). All mice were imaged 1–2 times/week and monitored for survival.

GBM-SRN01 cells were grown and passaged in DMEM (Corning #10-13-CV) supplemented with 10% FBS, 1X anti-anti, and 1X MEM NEAA (Gibco #11140050) in a humidified atmosphere of 5% CO₂. 300k tumor cells were injected intracranially into anesthetized athymic recipients in a volume of 3 μL. 45 days post-transplant, once the tumor was established, SRN-101 produced under cGMP-like conditions (2 × 10¹¹ or 5 × 10¹⁰ vg/brain), SRN-101 produced at RG (2 × 10¹¹ vg/brain), AAV9-hIFNβ (2 × 10¹¹ vg/brain), AAV9-mCard (2 × 10¹¹ vg/brain), or formulation buffer was administered via CED in a volume of 15 μL as described below.

Intratumoral CED administration

For the RG studies described in Figures 2, 3, and 4, all procedures were carried out under sterile surgical conditions at the UCSF Brain Tumor Center, as described previously. Before treatment initiation, animals were randomized to treatment groups. Mice were anesthetized with an intraperitoneal injection of ketamine/xylazine as described above. The scalp was cleaned with 2% chlorhexidine, and a skin re-incision ∼10 mm in length was made over the middle frontal to parietal bone. The surface of the skull was exposed so that the hole made for tumor implantation was exposed. The CED brain infusion cannula was lowered through this hole into the tumor for injection. First, a glass Hamilton syringe (World Precision Instruments Cat#Nanofil-10μL) was loaded with the sample and then attached to an Ultra MicroPump with Micro4 Syringe Pump Controller (World Precision Instruments Cat#UMP3). The pump was used to drive fluid slowly (1 μL/min) into the GBM tumor. The syringe was attached to Polymicro Flexible Fused Silica Capillary Tubing of inner diameter 320 μm (Molex Polymicro Technologies, Cat#1068150027) and then inner diameter 100 μm (Molex Polymicro Technologies, Cat#1068150020). The syringe was then fit to an internal cannula (InVivo1, a division of Plastics One Cat#C313IS-4/PK/SPC) and then a guide cannula (InVivo1 Cat#C313GS-4/PK/SPC) to guide the internal cannula to the specific injection site. Samples (saline or AAV) were infused at the rate of 1 μL/min until the desired dose (AAVs at 1−2e11 vg/brain in a volume of 10–15 μL) had been delivered. The brain infusion cannula was removed 2 min after infusion completion. The skull was swabbed with hydrogen peroxide, and the hole was covered with bone wax before closing the scalp with staples.

For the industry-grade studies described in Figures 5 and 6, all procedures were carried out under sterile surgical conditions at Crown Bioscience (San Diego, CA). Before treatment initiation, animals were randomized to treatment groups with equivalent distribution of tumor sizes across groups. Mice received slow-release buprenorphine subcutaneously, followed by anesthetization with isoflurane as described above. Mice were positioned in a stereotaxic head frame, disinfected with betadine followed by 70% alcohol, and their incision reopened using a 2- to 3-mm-long incision at the cranial midline. A hydrogen peroxide solution was used to clean away tissue from the skull and better visualize cranial suture lines. Any remaining bone wax was removed, and the previous burr hole was re-drilled if needed as described above. A sterile glass Hamilton syringe (Gastight 1700 model series with PTFE Luer Lock) was loaded with each test article and affixed to a needle-mounted rodent CED device composed of a 27G Luer Lock blunt tip needle with 100-μm inner diameter fused silica capillary tubing (Molex Cat#106815-0020) affixed within the needle with a heat-stable cyanoacrylate glue seal (Permabond 825). The 100-μm-inner-diameter tubing (170 μm outer diameter) was ensheathed in 450-μm outer diameter tubing (320 μm inner diameter) (Molex Cat#106815-0020) to form a sealed 1 mm step at the injection tip. The loaded rig was mounted to a rodent stereotaxic arm configured with a microsyringe pump. The pump was engaged to prime the CED device to remove any air bubbles and confirm patency, followed by a sterile alcohol wipe. The CED rig was then gently positioned at bregma so the stereotaxic syringe arm could be zeroed for dorsal/ventral z-positioning. The syringe stereotaxic arm was then lifted, relocated above the burr hole, and slowly lowered to 3.5 mm ventral to the skull. The microsyringe pump was used to infuse the test article at a rate of 1 μL/min for 15 min, followed by a brief (∼1 min) hold period before slowly lifting the CED using the stereotaxic arm until it could clear the cranium. Bone wax was used to close the hole in the skull, and the skin incision was closed with tissue glue.

In vivo firefly luciferase imaging of tumor volume and quantitation in mice

For studies described in Figures 2, 3, and 4, all mice with FLuc⁺ tumors were imaged non-invasively 1–2 times/week on a Xenogen VivoVision IVIS Spectrum imaging system (Caliper Life Sciences). D-luciferin substrate (Biosynth Cat#L-8220) was administered at 120 mg/kg in saline by intraperitoneal injection with a 1 cc insulin syringe. Images were acquired 10 min after luciferin administration under inhalation isoflurane anesthesia. Living Image v.4.3.1 software (Caliper LS, US) was used for image analysis, and average radiance was quantified in p/s/cm²/sr. All mice in each experiment are shown on the same non-individualized radiance scale to enable accurate comparisons of bioluminescent intensity.

For studies described in Figure 6, all mice were imaged non-invasively 1–2 times/week on a Xenogen VivoVision IVIS Spectrum imaging system (Caliper Life Sciences). D-luciferin substrate (Biosynth Cat#L-8220) was administered at 150 mg/kg 15 min before imaging. Ten minutes following administration of D-Luciferin, mice were anesthetized, placed into the imaging chamber, and imaged for luminescence in dorsal view. Duration and binning of the image acquisition were set through default and auto-exposure features of the Living Image software 4.3.1 (Caliper LS, US). Primary imaging analysis was quantified in total flux (p/s) for regions of interest and average radiance (p/s/cm²/sr) for comparisons.

Treatment response based on BLI was determined by calculating the percent change in BLI flux (p/s) per mouse as follows:

$$\%\mathit{C}\mathit{h}\mathit{a}\mathit{n}\mathit{g}\mathit{e}=\frac{\left({\mathit{B}\mathit{L}\mathit{I}\mathit{F}\mathit{l}\mathit{u}\mathit{x}\left(\mathit{p}/\mathit{s}\right)}_{\mathit{D}\mathit{a}\mathit{y}\mathbf{27}\mathit{o}\mathit{r}\mathit{D}\mathit{a}\mathit{y}\mathbf{91}}-{\mathit{B}\mathit{L}\mathit{I}\mathit{F}\mathit{l}\mathit{u}\mathit{x}\left(\mathit{p}/\mathit{s}\right)}_{\mathit{A}\mathit{s}\mathit{s}\mathit{i}\mathit{g}\mathit{n}\mathit{m}\mathit{e}\mathit{n}\mathit{t}}\right)}{\left({\mathit{B}\mathit{L}\mathit{I}\mathit{F}\mathit{l}\mathit{u}\mathit{x}\left(\mathit{p}/\mathit{s}\right)}_{\mathit{A}\mathit{s}\mathit{s}\mathit{i}\mathit{g}\mathit{n}\mathit{m}\mathit{e}\mathit{n}\mathit{t}}\right)}\ast \mathbf{100}$$ (Equation 1)

Mice were then classified by outcome based on the calculated percent change as follows. Tumor free: BLI flux signal <2.5 × 10⁵ p/s at day 27 (Figure 2) or day 91 (Figure 6), tumor reduction: ≥30% decrease from assignment on day 9 (Figure 2) or day 43 (Figure 6); no change: between 30% decrease and 20% increase from assignment on day 9 (Figure 2) or day 43 (Figure 6); tumor growth: ≥20% increase from assignment on day 9 (Figure 2) or day 43 (Figure 6); death: mice that died before day 27 (Figure 2) or day 91 (Figure 6).

Treated brain processing for library generation and scRNA-seq

Before brain harvest at 48 h, mice were transcardially perfused with cold PBS to remove potentially contaminating peripheral blood in the brain microvasculature during harvest. Following perfusion, treated brains were harvested and placed immediately in ice-cold Hibernate-E medium (Gibco Cat#A1247601) for transport. In the laboratory, each treated brain hemisphere was processed into a single-cell suspension using DMEM (Lonza Cat#12-614F) supplemented with 10% FBS. The resultant single-cell suspension was validated for viable cells and placed on ice. 6,000 viable cells were loaded for single-cell emulsion, and libraries were prepared using the Chromium Next GEM Single Cell 3ʹ Reagent Kit v.3.1 (10× Genomics). Library quality control analysis was performed by both Qubit fluorometer (using the Qubit BR kit Cat#Q32850) and Bioanalyzer analysis to determine both the concentration and average size (bp).

scRNA-seq data pre-processing

Libraries were sequenced on an Illumina HiSeq 4000 using PE150 reads. Reads were trimmed and filtered with TrimGalore v.0.6.5 (parameter: −q = 30) and Cutadapt v.3.4. The quality control-passed reads were mapped to the human GRCh38 genome assembly with BWA, and only uniquely matched paired reads were used for analysis. PicardTools and the GATK toolkit carried out quality score recalibration, duplicate removal, and re-alignment around indels. CellRanger v.3.0.2 was used for alignment and gene expression quantification, following the guidelines from the CellRanger website. Cells with >2.5% mitochondrial read counts and <200 expressed genes were removed. DoubletFinder v.2.0.2 was used to remove doublets and was run using the first 10 principal components and default parameters.

Dimensionality reduction, clustering, cell-type classification, and marker analysis for scRNA-seq

The scRNA-seq data were processed with Seurat v.3. Data were normalized via the LogNormalize method with scale.factor = 10,000 using the NormalizeData function. Highly variable genes were identified via Seurat using the mean.var.plot method with default parameters. Based on these genes, a principal-component analysis was performed, and the first 15 principal components were retained for clustering and visualization via uniform manifold approximation and projection (UMAP). Neoplastic cells were separated from non-neoplastic cells based on the presence of CNVs. For neoplastic cells, mesenchymal (MES) and proneural (PN) cell-type labels were assigned via ELSA, an ensemble-learning approach that has been trained on historical data. For non-neoplastic cells, cell clusters were identified via the “FindClusters” function via the Louvain algorithm with the resolution parameter = 0.5. Cluster-specific genes were identified via the FindAllMarkers function in Seurat v.3 under R 3.6.0 via a MAST test, and used to assign cell-type labels. For identification of genes for AAV9-mIFNβ- vs. AAV9-GFP-treated tumor cells, and AAV9-GFP- vs. saline-treated tumor cells, the FindMarkers function in Seurat v.3 under R3.6.0 was used to directly compare AAV9-mIFNβ to AAV9-GFP (ident.1 = “AAV9-mIFNβ”, ident.2 = “AAV9-GFP”) and AAV9-GFP to saline (ident.1 = “AAV9-GFP”, ident.2 = “saline”). Differential gene expression was performed using DESeq2. Computational analysis methods are available in GitHub: https://github.com/SirenBio/aav_immuno_gene_therapy_for_HGG.

Brain histopathology

The UCSF Histology and Biomarker Core and the UCSF Brain Tumor Research Center Tissue Core conducted H&E and immunohistochemical stains. 4–5 micron formalin-fixed paraffin-embedded (FFPE) brain sections were mounted on positively charged glass slides (Fisher Scientific Cat#12-550-15) following laboratory standard procedures for the sectioning of paraffin blocks. Slides were air-dried overnight for 8–24 h and then baked and deparaffinized. For H&E staining, slides were baked at 60°C for 15 min before staining; for IHC staining, slides were baked overnight at 60°C before staining. H&E staining was performed on the Leica XL Autostainer using hematoxylin (Thermo Scientific Cat#6765015) for 7 min and alcoholic eosin (Thermo Scientific Cat#6765040) for 20 s. IHC was done on the Roche Ventana Discovery Ultra Autostainer using antibodies and staining conditions as outlined below. Positive control tissue staining was performed for all antibodies (Figure S3). A Zeiss Axio scanner was employed for digital slide scanning with bright-field images collected at 20× magnification using Zeiss Zen software.

The following antibodies were used: Ki67 (Abcam clone SP6 Cat#ab16667 at 1:50), CC3 (CST Cat#9661S at 1:100), human IFNβ (Abcam Cat#ab216475 at 1:100), Iba1 (Wako Cat#019-19741 at 1:1,300), and GFAP (DAKO Cat#Z0334 at 1:50,000).

RNAscope for IFNβ

Human FFPE brain sections were evaluated by RNAscope chromogenic in situ hybridization assay for the expression of hIFNβ using Advanced Cell Diagnostics probes specific for hIFNβ (RNAscope 2.5vs Probe #Hs-IFNB1) following manufacturer protocols. RNAscope was done on the Roche Ventana Discovery Ultra Autostainer by the UCSF Brain Tumor Research Center Tissue Core.

Luminex IFN cytokine assay

Serum was harvested from the blood of treated mice and aliquoted in protein LoBind polypropylene tubes for storage at −80°C until use in the assay. Serum from each mouse was run in technical duplicate. The assay was performed at the Human Immune Monitoring Center at Stanford University. A custom human kit containing antibodies for human IFNα2, IFNβ, and IFNγ was purchased from eBiosciences and used according to the manufacturer’s recommendations with modifications described below. Briefly, beads were added to a 96-well plate and washed in a BioTek ELx405 washer. Thawed samples were added to the plate containing the mixed antibody-linked beads and incubated at 25°C for 1 h, followed by overnight incubation at 4°C with shaking. Cold and 25°C incubation steps were performed on an orbital shaker at 500–600 rpm. Following the overnight incubation, plates were washed in a BioTek ELx405 washer, and then biotinylated detection antibody was added for 75 min at 25°C with shaking. The plate was washed as above, and streptavidin-PE was added. After incubation for 30 min at 25°C, a wash was performed as above, and reading buffer was added to each well. Plates were read using a Luminex 200 instrument with a lower bound of 50 beads per sample per cytokine. Custom assay control beads by Radix BioSolutions were added to all wells. Luminex heat maps were generated using PRISM v10.6.1.

Spatial transcriptomics

Tissue slides for the CytAssist FFPE assay were prepared following the guidelines provided by 10× Genomics in “CG000518 Demonstrated Protocol Visium CytAssist Spatial Gene Expression for FFPE - Tissue Preparation Guide.” Briefly, 5 μm sections were cut from FFPE blocks on a standard microtome and mounted on regular microscope slides. Slides were placed in a rack and positioned on their side in a drying oven at 42°C and then stored in a desiccator. The next day, the slides were deparaffinized, H&E stained, and imaged as per “CG000520 Visium CytAssist Spatial Gene Expression for FFPE - Deparaffinization, H&E Staining, Imaging and Decrosslinking.” The H&E-stained slides were imaged on a Zeiss Axio Observer Z1 automated microscope with a 10× Neofluar objective and a Zeiss AxioCam 506 camera. Image frames were tiled and stitched with Zeiss Zen v.3.2 software. The slides were then destained and decrosslinked for probe hybridization.

A custom probe set of 113 mouse probes (Table S5) was combined with Human Probe Set v2 (Cat#PN-1000466) and hybridized to the decrosslinked slides following the protocols given in “CG000495 Visium CytAssist Spatial Gene Expression Reagent Kits for Formalin Fixed & Paraffin Embedded (FFPE).” Briefly, probes were hybridized overnight in the CytAssist cassette at 55°C in an Eppendorf MasterCycler Pro thermal cycler. The next day, the probes were ligated and released by RNA digestion and tissue removal. The cDNA probes were extended and pre-amplified for 10 cycles in an Eppendorf MasterCycler Pro using the cycle parameters specified in CG000495. After size selection with SPRI beads, the cDNAs were quantified by qPCR and libraries were constructed as per CG000495. Libraries were sequenced on an Illumina NovaSeq 6000 with the PE150 cycle kit.

The resultant fastq sequence files were mapped to the human genome, and QC analysis was performed using SpaceRanger 2.0.1 pipeline software (see 10× Genomics site):

https://www.10xgenomics.com/support/software/space-ranger/latest/analysis/running-pipelines/probe-based-assay-count-cytassist-gex. Briefly, the low-resolution TIFF image captured on the CytAssist instrument was aligned to the high-resolution JPEG image captured on the Zeiss Axio Observer using the manual alignment protocol in Loupe Browser v.6.0 software (10× Genomics) and the resultant JSON file specified in the SpaceRanger command. A combined human-mouse reference genome was used for mapping.

https://cf.10xgenomics.com/supp/cell-exp/refdata-gex-GRCh38-and-mm10-2020-A.tar.gz.

Differential expression testing was conducted using the standard Seurat implementation of the Wilcoxon test. Computational analysis methods are available on GitHub:

https://github.com/SirenBio/aav_immuno_gene_therapy_for_HGG.

Statistical analyses

Statistical analyses were conducted with Prism v.10 software. Experimental differences were evaluated as follows: Kaplan-Meier percent survival used a log-rank (Mantel-Cox) test, tumor BLI used a Kruskal-Wallis test with Dunn’s multiple comparisons correction, and the Luminex assay used a one-way ANOVA with Tukey’s multiple comparisons test; p values < 0.05 were considered statistically significant (∗p < 0.04, ∗∗p < 0.02, ∗∗∗p < 0.005, ∗∗∗∗p < 0.002, ∗∗∗∗∗p < 0.0001).

Data and code availability

All data are available in the main text or the supplementary materials.

Acknowledgments

The authors wish to first thank the patients at UCSF for donating their tumor tissues for our study; Scott VandenBerg, Jennifer Bolen, Kristine Wong, and Mohammed Naser of the UCSF Helen Diller Family Comprehensive Cancer Center Tissue Core and Anny Shai and Yunita Lim from the UCSF Brain Tumor Center Tissue Core for supportive help with histopathology and immunohistochemistry; Raquel Santos, Kyounghee Seo, Tomoko Ozawa, and David Raleigh from the UCSF Preclinical Therapeutics Core for help with the early animal studies; the UCSF Center for Advanced Technology for help with some of the single-cell sequencing; David James from UCSF for the original gift of the GBM6 and GL261 cells; Iris Herschmann of Stanford University Human Immune Monitoring Core for help with the Luminex assay; Shalin Mehta from the CZ Biohub for assistance with image analysis; and all of the physicians and scientists in the UCSF Glioblastoma Precision Medicine Program for guidance and discussions. The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of the funding bodies, associated universities, or organizations. Funding was provided by the following: Siren Biotechnology, Inc. (N.K.P.), the UCSF Glioblastoma Precision Medicine Program (N.K.P.), UCSF Catalyst Program grant 7030179 (N.K.P.), US Department of Defense Translational Research Award W81XWH-19-1-0348 (A.A.D.), National Institute of Diabetes and Digestive and Kidney Diseases grant K01-DK107607 (N.K.P.), and National Institute of Environmental Health Sciences grant U01-ES032673 (N.K.P.). Additional resources were provided by the UCSF Brain Tumor SPORE Biorepository (National Cancer Institute P50-CA097257).

Author contributions

Conceptualization, N.K.P.; data acquisition, A.K., E.M.F., C.L., T.L., L.W., M.F., J.J., S.-M.G., S.B.M., L.H.M., Y.R.-H., J.L.C., J.P., L.K., N.C., and N.K.P.; data analysis, A.K., E.M.F., C.L., T.L., M.F., L.H.M., Y.R.-H., A.A.D., J.J.P., L.K., N.C., and N.K.P.; visualization, A.K., E.M.F., J.J., S.-M.G., S.B.M., Y.R.-H., J.P., L.K., and N.K.P.; funding acquisition, N.K.P. and A.A.D.; project administration, A.K., A.P.M., J.L.C., A.A.D., J.J.P., L.K., N.C., and N.K.P.; writing – review and editing, all authors.

Declaration of interests

Corporate funding from Siren Biotechnology was used for this study. Siren employee authors held equity in Siren Biotechnology during the study and are inventors of various patents related to the contents of this study.

Supplemental information

Table S2. Time course of differentially expressed genes in hIFNβ-treated GBM6 cells vs. media controls

Differentially expressed genes (R package “DESeq2″) in AAV9-hIFNβ (MOI 4E5 vg/cell)- or recombinant hIFNβ protein (rIFNβ, 47 IU/mL)-treated GBM6 cells at 24, 48, 72, and 96 h post-treatment vs. time-matched media controls. Column definitions: baseMean = mean expression, log2FoldChange = log2 fold change in expression of treatment condition (“treat_time_exp”) vs. control condition (“treat_time_control”), lfcSE = log2 fold change standard error; stat = Wald statistic, pvalue = p value, padj = adjusted p value, Gene.name = gene name, treat_time_exp = treatment condition and time point, treat_time_control = control treatment and time point, cell_line = cell line tested, pvalue.filt = filtered p value to exclude low read genes, and padj_BH = Benjamini-Hochberg adjusted p value.

Table S3. Differentially expressed genes in syngeneic mouse allograft GBM tumors

Differentially expressed genes in single-cell sequencing data from mouse GL261 GBM tumors treated with AAV9-mIFNβ (2 × 10¹¹ vg/brain) vs. AAV9-GFP (2 × 10¹¹ vg/brain). Column names: gene_name = gene name, p_val = p value, avg_log2FC = average log2 FC AAV9-mIFNβ treated vs. AAV9-GFP treated, pct.1 = percentage of cells where the feature is detected in AAV9-mIFNβ-treated tumors, pct.2 = percentage of cells where the feature is detected in AAV9-GFP-treated tumors, and p_val_adj = adjusted p value across all genes.

Table S4. Differentially expressed genes in human and mouse PBMCs treated with IFNβ

Differentially expressed genes in bulk RNA sequencing data from human or mouse PBMCs treated with either species-matched IFNβ identified using DESeq2. Column names: gene_name = gene name, baseMean = mean expression, log2FoldChange = log2 fold change of expression between experimental treatment (“treat_exp”) vs. control (“treat_control”), lfcSE = standard error of log2 fold change, treat_exp = IFN treatment condition, treat_control = control treatment condition, species_key = PBMC species, and padj_BH = Benjamini-Hochberg adjusted p value.

Table S5. Mouse genes included in spatial transcriptomic sequencing

Mouse gene list used to generate custom probes for spatial transcriptomic profiling of CDX GBM mouse brains

Table S6. Differentially expressed genes in tumor clusters 48 h vs. 0 h post-AAV9-hIFNβ treatment

Differential expression analysis (R package “DESeq2″) performed on spatial transcriptomic data in human GBM6-FLuc CDX mice collected pre-treatment (0 h, n = 1) or 48 h (n = 1) after intratumoral AAV9-hIFNβ infusion (2 × 10¹¹ vg/brain). Column definitions: p_val = p value, avg_log2FC = average log2 fold change in expression in 48 h vs. 0 h AAV9-hIFNβ treatment tumors, pct.1 = percentage of cells where the feature is detected in 48 h post-treatment tumor, pct.2 = percentage of cells where the feature is detected in pre-treatment (0 h) tumor, p_val_adj = adjusted p value across all genes, control_group = control group used in analysis (pre-treatment tumor), experimental_group = experimental group used in analysis (48 h post-treatment), gene_species = species of differentially expressed gene (hs = human, ms = mouse), gene_name = gene name, and padj_BH = Benjamini-Hochberg adjusted P value.

Table S7. Gene set enrichment analysis of tumor clusters 48 h vs. 0 h post-AAV9-hIFNβ treatment

Gene set enrichment analysis (R package “fgsea”) performed on spatial transcriptomic data in human GBM6-FLuc CDX mice collected pre-treatment (0 h, n = 1) or 48 h (n = 1) after intratumoral AAV9-hIFNβ infusion (2 × 10¹¹ vg/brain). Column definitions: pathway = MSigDB pathway name, pval = p value, padj = adjusted p value, log2err = expected error for the standard deviation of the p value logarithm; ES = enrichment score; NES = normalized enrichment score, size = number of genes in pathway, and leadingEdge = genes that contribute most to ES.