Novel strategy to target glioblastoma-initiating cells using a braintropic adeno-associated virus carrying a miR-dependent genome-editing system

Abstract

Background

Since glioblastoma (GBM)-initiating cells (GICs) were identified as the cells-of-origin for GBM, various GIC factors have been analyzed as potential therapeutic targets. However, these targets are also present in normal cells outside of the brain, raising concerns about potential side effects when directly targeted. The aim of this study is to develop a novel method that specifically eradicates GICs with reducing side effects.

Methods

We selected micoRNAs (miRs) that are significantly decreased in GICs compared to normal cells and developed a genome-editing (GE) system that knocks out a functional GIC factor in a miR-dependent manner (miR-dependent GE). Additionally, we developed mosaic-capsids that consist of braintropic and universal capsids, which deliver genes into GIC brain tumors.

Results

Systemic administration of the mosaic-capsids Adeno-associated virus (AAV) carrying a miR-dependent GFP expression cassette selectively expressed GFP in GICs transplanted into the brains of immunodeficient mice, without expression in either mouse brain cells or non-brain tissues. The mosaic-capsids AAV carrying a miR-dependent GE prevented GIC tumorigenesis in the brain and extended the survival time of tumor-bearing mice.

Conclusion

These data indicate that the mosaic-capsids AAV containing a miR-dependent GE represents a novel therapeutic virus for GBM with fewer side effects.

Background

Glioblastoma (GBM) is a highly aggressive brain cancer with a median survival of approximately 15 months [1]. Despite advancements in multimodal treatments, including surgery, chemotherapy, and radiotherapy, the overall survival rate has not improved over the past few decades. The discovery of GBM-initiating cells (GICs) has revealed their central role in tumor development and resistance to conventional therapies [2–4]. Therefore, targeting and eliminating GICs is an important goal in GBM research.

In previous studies, we and others identified various therapeutic candidates, including membrane proteins, transcriptional regulators, and microRNAs (miRs) through comparative analyses of gene expression profiles in neural stem cells (NSCs), oligodendrocyte precursor cells, and GICs [5–10]. However, it became apparent that these factors are also expressed in non-brain cells. For example, CD133 is found in several normal adult stem cells and photoreceptor cells [11]. Ceacam1, another membrane protein expressed on GICs, is also present in various non-neural cells, including blood vessels and immune cells [8, 12]. Dihydroorotate dehydrogenase (DHODH), a key component of the de novo pyrimidine synthetic pathway, is a potential therapeutic target in GICs [13]. DHODH is important for the proliferation of blast cells, which play a critical role in eliminating infectious diseases and cancer [14]. These findings raise concerns about potential side effects when targeting such GIC factors directly, especially in the absence of method to selectively target GICs, while sparing normal cells. Therefore, we have developed an innovative approach to specifically eradicate GICs without harming normal cells.

Gene therapy enables the development of methods tailored to target GIC-specific factors, compared to antibody-based methods or the identification of chemicals specific to GICs. Naldini and colleagues elegantly demonstrated that endogenously expressed miRs, such as miR-142-3p specific to hematopoietic cells and miR-302a specific to human embryonic stem cells, can suppress the expression of an exogenous gene containing four tandem complementary sequences of these miRs (miRCS) in vitro and in vivo [15]. Sayeg et al. showed that miRCS derived from miR-128 and miR-221, both of which are expressed in excitatory neurons but not in inhibitory neurons, selectively induced the expression of an exogenous gene in inhibitory neurons of the brain following the delivery of miRCS via a lentivirus [16]. These findings suggest that post-transcriptional gene silencing by endogenous miR can selectively activate the expression of any exogenous gene with miRCS in miR-negative, but not miR-positive cells. However, the miRNA Tissue Atlas (https://ccb-web.cs.uni-saarland.de/tissueatlas2/) indicates that miR-142-3p has the highest expression in the lung and lower levels in the brain, whereas both miR-128 and miR-221 exhibit very low expression in non-brain tissues. This suggests that miRCS alone may not be sufficient to selectively regulate the expression of an exogenous gene in the target cells.

The GIC factor-targeting method and miR-dependent gene expression system alone cannot specifically eradicate GICs; however, when combined, they can effectively, selectively eliminate GICs in brain tumors. Many methods have been developed for delivering genes to brain tumors [17–19]. Among these, recombinant viruses, such as adeno-associated virus (AAV), Zika virus, and Herpes simplex virus, are promising candidates as they can infect a variety of neural cells in vivo [19–21]. In particular, AAV stands out due to its low immune response, high yield, long-term expression in non-proliferating cells, and histotropic effect [20, 22]. Recent studies have introduced several new AAV capsids with brain-targeting activity (braintropic AAV capsids like PHP.eB, CAP-B10, and CAP-B22), which facillitate the analysis of both endogenous and exogenous gene functions in the brains of mice and non-human primates. However, these capsids have also been shown to deliver a small quantity of the virus to the liver [23, 24]. AAV may only transiently express an exogenous gene in rapidly dividing cells, such as cancer cells, due to its low genome-integration efficiency [22]; however, this limitation can be addressed by using a Cas9-dependent genome-editing (GE) system as its transient expression is sufficient to knock out the target gene.

In this study, we combined braintropic AAV, a miR-dependent gene expression system specific to GICs, and GE of the GIC functional factor DHODH to generate novel braintropic AAVs carrying the GE system with miRCS (miR-dependent GE system). We demonstrate that these AAVs can selectively inhibit the proliferation of GICs and induce their death, thereby preventing GIC tumorigenesis when systemically administered to brain tumor-bearing immunodeficient mice. Furthermore, these AAVs exhibit minimal expression of the exogenous gene in neural cells and non-targeted tissues, such as the liver, lung, kidney, and spleen, confirming their specificity for GIC tumor cells. These results indicate that these AAVs represent novel therapeutic viruses against GBM with minimal side effects.

Methods

Chemical reagents

Chemicals and growth factors were purchased from ThermoFisher, SIGMA, PeproTech, and other suppliers.

Cell culture

Five primary human glioma samples, including three GBMs, one anaplastic astrocytoma (AA), and one diffuse oligodendroglioma (DO), were obtained from Ehime University Hospital and used to prepare glioma lines, GICs (E3, E6, E16), AA, and DO. The glioma lines were maintained in a 50% medium containing DMEM/F12 (SIGMA, D8062) supplemented with chemicals, heparin (5 μg/ml SIGMA, 9041-08-1), bFGF (20 ng/mL, PeproTech, 100-18B) and EGF (20 ng/mL, PeproTech, AF-100-15), referred to as NSC medium, along with 50% DMEM (Nacalai Tesque, 16971-55) containing 10% fetal calf serum (FCS), as previously described [6, 9]. Due to contamination in the E3 culture, we used E6 and E16 for further experiments. The expression profiles of the GICs revealed that E6 and E16 correspond to pro-neural and mesenchymal types, respectively (GSE61078 and GSE61079 for E6, unpublished data for E16). After expanding the glioma cell lines, we prepared large stocks and used the cells up to passage 50. The human neural stem cell (NSC) line ReN was cultured on a laminin-coated culture dish or a chamber slide (Nunc) in DMEM/F12 supplemented with B27 supplement (ThermoFisher, 17504001), GlutaMax supplement (ThermoFisher, 35050061), penicillin-streptomycin (Nacalai Tesque, 26253-84), heparin, bFGF (10 ng/mL), and EGF (20 ng/mL). LentiX-293T (TAKARA, 632180) was cultured in DMEM with 10% FCS. All the cells were treated with anti- mycoplasma removal agent (Wakenbtech, MC-210) every half year. To measure cell proliferation, the MTT assay was performed as described previously [6].

Vectors

To monitor the miR-dependent repression of gene expression, enhanced green fluorescence protein (GFP) and KeimaRed cDNAs were inserted into the pVITRO1-hygro vector (InvivoGen, pvitro1-mcs), resulting pVITRO1-keimaRed-GFP. Four copies of miR complementary sequences (4xmiRCS) were then inserted immediately after the gfp gene to obtain the pVITRO1-keimaRed-GFP-4xmiRCS vector. The sequences of the miRCS are provided as supplementary information. The fluorescence intensity of both GFP and KeimaRed was measured using Image-J and the ratio of GFP intensity to Keima-Red one was calculated.

To prepare miR-dependent GE vectors, three candidate sgRNAs (sgRNA1–3) targeting human DHODH were designed using the Benchling website (https://www.benchling.com) and inserted into the pX601-AAV-CMV NLS-SaCas9-NLS-3xHA-bGHpA;U6 BsaI-sgRNA vector (obtained from Dr. Feng Zhang through Addgene, pAAV-SaCas9-sgRNA, 12844). This resulted in a set of pAAV-SaCas9-sgDHODH vectors. The 4xmiRCS sequence was then inserted immediately after SaCas9 to create the pAAV-SaCas9-miRCS-sgDHODH vector. The oligonucleotides for human DHODH sgRNA were synthesized as follows: for sgDHODH1, the forward sequence is 5’-CACCGCCATAAATTCCGAAATCCAGT-3’ and the reverse sequence is 5’-AAACACTGGATTTCGGAATTTATGGC-3’; for sgDHODH2, the forward sequence is 5’-CACCGCGAAATCCAGTAGGAATTGCTG-3’ and the reverse sequence is 5’-AAACCAGCAATTCCTACTGGATTTCGC-3’; sgDHODH3, the forward sequence is 5’-CACCGAAATCCAGTAGGAATTGCTG-3’ and the reverse sequence is 5’-AAACCAGCAATTCCTACTGGATTTC-3’.

The pAAV-GFP control vector was constructed by inserting the SV40 enhancer/promoter-GFP-bGH poly(A) signal fragment, amplified from the pCMS-EGFP vector (Clontech, VT1142), into the pAAV-Guide-it Up vector (TAKARA, 632608).

To prepare the CAP-B10 and CAP-B22 expression vectors, AAV2 Cap in pRC2-mi342 (TAKARA, 632608) was first replaced with the PHP.eB fragment from pUCmini-iCAP-PHP.eB (Addgene) to obtain the pRC2-mi342-PHP.eB vector. Subsequently, using a standard PCR method, the sequence (aacggttctggacagaatcaa) from PHP.eB was replaced with either CAP-B10 (gacggcgccgccaccaaaaac) or CAP-B22 (gacggccaaagcagcaaaagc), resulting in the pRC2-mi342-CAP-B10 or pRC2-mi342-CAP-B22 vectors, respectively.

The nucleotide sequences of the inserted oligonucleotides were verified using the BigDye Terminator Kit version 3.1 (ThermoFisher, 4337455) and an ABI sequencer model 3130xl (Applied Biosystems).

We transfected GICs and ReN with the vectors using a Nucleofector device according to the manufacturer’s instructions (Lonza), and transfected LentiX-293T cells using polyethylenimine (PEI, Polyscience, 02371) as previously described [25]. One day after transfection, the cells were harvested and re-cultured on chamber slides for gene expression analysis.

AAV production

To generate recombinant AAVs, Lenti-X 293 T cells were transfected with an equal ratio of the following plasmids: capsid, and non-structural Rep expression vector (pRC2-mi342, pRC2-mi342-CAP-B10, pRC2-mi342-CAP-B22, or their combination), pHelper (TAKARA, 632608), and a transfer vector (pAAV-GFP, pAAV-SaCas9-sgDHODH, or pAAV-SaCas9-4xmiRCS-sgDHODH), using PEI. Brief, PEI (36 μl, 1 mg/ml) was suspended in serum-free DMEM (464 μl) and mixed with 6 μg of DNA. The PEI-DNA mixture was then added to the cells cultured in a 100 mm dish. Four days post-transfection, the cells and culture supernatant were collected and the viruses were enriched using chloroform and polyethylene glycol as described previously [26]. The virus titer was determined using the AAVpro Titration Kit (TAKARA, 6233) according to the manufacturer’s protocol. For the MTT and cell death assays, up to 10³ virus genome (vg)/cell and 5 × 10² vg/cell were used, respectively. For the in vivo experiments, approximately 10¹¹ vg/mouse were injected intravenously.

Immunocytochemistry

Immunostaining was performed as previously described [6, 7]. Briefly, cells were fixed with 2% paraformaldehyde-PBS for 10 min at room temperature (RT), treated with blocking buffer (50% FCS and 0.3% Triton X100 in PBS) for 30 min at RT, and incubated with primary antibodies (Abs) suspended in blocking buffer for 2 h at RT. After washing three times with PBS, the samples were incubated with secondary Abs and Hoechst 33342 (1 μg/ml, ThermoFisher, H3570) diluted in blocking buffer for 1 h at RT and mounted using the Fluoromount (SIGMA, F4680). The following antibodies were used: mouse monoclonal anti-Nestin (1:200, Beckton Dickinson, 611658), chicken monoclonal anti-Glial fibrillary acidic protein (GFAP, 1:2000, Novus Bio, NBP1-05198), chicken monoclonal anti-βIII tubulin (1:500, Novus Bio, NB100-1612), chick monoclonal anti-GFP (1:2000, Abcam, ab300643), rabbit polyclonal anti-active Caspase 3 (1:1000, Cell Signaling Technology, 9661), rabbit polyclonal Ki67 (1:250, ThermoFisher, MA5-14520), and mouse anti-SaCas9 (1:2000, MBL, D366-3). The antibodies were detected with Alexa594-conjugated goat anti-mouse IgG (1:500, ThermoFisher, A-11032), Alexa488-conjugated goat anti-chicken IgY (1:500, Jackson ImmunoResearch, 103-547-008), or Cy5-conjugated goat anti-rabbit (1:500, ThermoFisher, A10523). Fluorescence images were obtained using an AxioImager A1 microscope (Carl Zeiss).

Quantitative reverse-transcription PCR for miRs

Total RNA containing the miR was extracted from cells using the mirVana miRNA isolation kit (ThermoFisher, AM1560), following the manufacturer’s instructions. Real-time PCR was conducted using the StepOnePlus system (ThermoFisher) with the miRCURY LNA SYBR Green PCR Kit (QIAGEN, 339345), in according with the manufacturer’s protocol. The miRCURY LNA miRNA PCR Assays (QIAGEN, 339306) were utilized to detect U6 snRNA (YP02119464), miR219a-2-3p (YP00204674), and miR340-5p (YP00206068). The PCR conditions were as follows: 2 min at 95 °C; followed by 40 cycles of 95 °C for 10 s, and 56 °C for 1 min. All miR expression data were normalized to the U6 snRNA using the ∆∆Ct method.

Animals

The animal experiments were conducted in accordance with protocols approved by the Animal Care and Use Committees of Hokkaido University. NOG (NOD/Shi-scid, IL-2RγKO Jic) and C57BL/6 mice, both 10- to 12-week-old, were obtained from In Vivo Science, Inc. and CLEA Japan, respectively. All mice were housed with free access to food and water under a standard 12:12 light-dark cycle. Animals were randomly assigned to treatment groups.

Brain tumorigenesis and therapeutic effect of AAV

Brain tumorigenesis was performed as described previously [6]. Briefly, 1 ×10⁵ hGICs were suspended in 5 µl of culture medium and injected into the brains of NOG mice (n = 3 mice per group for miR-dependent AAV-GFP experiments, n = 5 mice per group for anti-tumorigenic activity experiments of miR-dependent GE AAV with mosaic capsids) under anesthesia with 10% pentobarbital. The stereotactic injection coordinates were 2 mm anterior from the lambda, 2 mm lateral from the sagittal suture, and 4 mm deep.

Injected GICs form tumors in the brain and kill mice approximately 30 days post-transplantation as shown previously [8–10], while systemically injected AAVs begin producing an exogenous gene product in the infected cells two weeks after inoculation. Considering the time required for tumor formation in the transplanted GICs, we injected the AAVs intravenously on days 10 and 12, when the tumorigenicity in each mouse was still clarified, to enhance AAV activity following GIC transplantation.

Histopathological analysis

Mouse brains were dissected, frozen in Tissue-Tek OCT compound (SAKURA Finetek, 4583), and stored at −80 °C overnight. Coronal sections (12-µm thick) were prepared from the cerebral cortex and immunolabeled for antigens after retrieval using HistoVT One, following the manufacturer’s instructions (Nacalai Tesque, 06380-05). The following antibodies were used to detect antigens: rabbit polyclonal anti-GFAP (1:500, DakoCytomation, N1506), rabbit polyclonal anti-MAP2 (x500, ProteinTech, 17490-1-AP), mouse monoclonal anti-Galactocerebroside (GalC) (x10, hybridoma sup), chicken monoclonal anti-GFP, mouse monoclonal anti-human mitochondria (1:200, Novus Bio, NBP2-32980-0.1 mg), rabbit polyclonal anti-active Caspase 3 (1:1000, Cell Signaling Technology, 9661), rabbit polyclonal anti-DHODH (1: 100, SIGMA, HPA010123), and chicken monoclonal anti-SaCas9 (1:2000, Novus Bio, NBP3-05547-50 µl). Antibodies were detected with Alexa594-conjugated goat anti-mouse or -rabbit (1:500, ThermoFisher, A-11012) IgG, Alexa488-conjugated goat anti-chicken IgY (1:500, Jackson ImmunoResearch, 103-547-008), and Cy5-conjugated goat anti-rabbit (1:500, ThermoFisher, A-10523). The ratio of tumor size per brain was determined using Image J software. Images were obtained using an AxioImager A1 microscope (Carl Zeiss) and BZ-X800 microscope and BZ-X800 Analyzer software (Keyence).

Statistical analysis

Statistical analyses were performed using GraphPad Prism version 4 software (GraphPad). Data were analyzed using a one-way ANOVA with Tukey’s post-hoc test as indicated in the figure legends. Survival data were assessed for significance using Kaplan-Meier methods, with P values calculated from the Log-rank (Mantel-Cox) test. All results are expressed as mean ± SD values. Sample size of each experiment was determined by our previous experiments. All experiments in culture were conducted more than three times with similar results.

Results

Identification of functional miRs with lower expression in GICs compared with human neural stem cells (NSCs)

To develop a new expression system based on endogenous miRs that are decreased in GICs compared with those in human NSC ReN (ReN), we reanalyzed the expression profiles of miRs in hNSCs, GICs (E3, E6, E16), anaplastic astrocytoma (AA), and diffuse oligodendroglioma (DO), which we previously described [9]. We selected six miRs (miR-124-3p, miR-153-3p, miR-219a-5p, miR-219a-2-3p, miR-301b-3p, and miR-340-5p) (fold change <−2) as candidates (Fig. 1a). Further analysis using the Small Noncoding RNA Tissue Atlas (https://ccb-web.cs.uni-saarland.de/tissueatlas2/) revealed that these candidate miRs are expressed in various human tissues with the highest expression detected in the brain (Supplementary Fig. 1). Previous studies have shown that miR-124-3p, miR-153-3p, miR-219a-5p, and miR-340-5p act as tumor suppressors in GBM by inducing cell-cycle arrest, cell death, and senescence [9, 27, 28]. In contrast, the roles of miR-219a-2-3p and miR-301b-3p in GBM are unclear, however, miR-219a-2-3p has been shown to inhibit the proliferation of thyroid cancer cells, while miR-301b-3p promotes cancer proliferation in several types, including breast and colorectal cancers [29, 30].

Fig. 1. Identification of miR-219a-2-3p and miR-340-5p as candidate miRs for the miR-dependent gene expression system.

a Fold changes in the expression of the six selected miRs, miR-124-3p, -153-3p, -219a-5p, -219a-2-3p, -301b-3p, and -340-5p, in GICs (E3, E6, and E16), anaplastic astrocytoma (AA), and diffuse oligodendroglioma (DO), compared with ReN. b A miR-function monitoring vector that contains both a GFP expression cassette with 4xmiRCS and a KeimaRed expression cassette. c Proportion of GFP-positive cells in KeimaRed-positive cells in ReN and GICs, E6 and E16. d RT-qPCR measurement of miR-219a-2-3p and miR-340-5p in ReN and GICs. Statistical significance was determined using a one-way ANOVA. * p < 0.05, *** p < 0.001, **** p < 0.0001. Error bars indicate ±SD.

To determine the function of these candidate miRs in ReN and GICs, we designed monitoring vectors that express both GFP with four tandem complementary sequences of one of the six candidate miRs (miRCS) and KeimaRed as a transfection marker (Fig. 1b). These vectors were transfected into ReN and GICs E6 and E16. Three days after transfection, KeimaRed-positive cells were observed in all transfected cells (Supplementary Fig. 2). GFP expression was also detected in all transfected cells, whereas the ratio of GFP fluorescence intensity to KeimaRed one in ReN transfected with vectors containing miR-219a-2-3pCS, miR-301b-3pCS, and miR-340-5pCS was significantly low (Fig. 1c and Supplementary Fig. 2). These results suggest that miR-219a-2-3pCS, miR-301b-3pCS, and miR-340-5pCS may be used to selectively express an exogenous gene in GICs, but not in ReN.

Since miR-301b-3pCS is expressed at low levels in all tissues (Supplementary Fig. 1) and has been shown to induce cancer cell proliferation [29], we excluded it from further experiments. RT-qPCR was used to verify the expression of miR-219a-2-3p and miR-340-5p in ReN and GICs. As shown in Fig. 1d, the expression of both miRs was significantly lower in GICs compared with ReN. Based on these findings, we used both miR-219a-2-3pCS and miR-340-5pCS to construct an expression system with miRCS.

DHODH GE AAVs encoding miRCS inhibit GIC proliferation and induce cell death in culture

For the specific eradication of GICs in vivo, we constructed AAVs carrying a miR-dependent GE system that targets the GIC factor. Due to the limited insert size of recombinant AAVs [31], we utilized SaCas9, a smaller variant of Cas9, to generate a single packaging virus containing the GE system along with a single guide RNA (sgRNA) for the GIC factor and the candidate miRCS (Fig. 2a). Among the many factors essential for maintaining the characteristics of GICs, we selected DHODH as the target for the following reasons: First, highly proliferative cells, including cancer cells, require both the DHODH-dependent de novo pathway and the salvage pathway to produce a large amount of pyrimidine, whereas differentiated cells rely solely on the salvage pathway for proliferation and maintenance [32, 33]. Second, our previous study demonstrated that DHODH indirectly regulates the nuclear localization of SOX2, an essential stemness factor in GICs, through O-GlcNAcylation [13]. Third, DHODH inhibitors, either alone or in combination with conventional anti-cancer drugs effectively kill various types of cancer cells in vivo [13, 34–36].

Fig. 2. GE AAV encoding miRCS induces cell-cycle arrest and cell death in GICs.

a Schematic illustration of DHODH GE AAV vector encoding miRCS. b Proportion of proliferating/survival cells infected with various AAVs, sgDHODH1 (closed circle), sgDHODH2 (closed triangle) and sgDHODH3 (closed square), their combination, sgDHODH1 and 2 (open circle), sgDHODH1 and 3 (open triangle), and sgDHODH2 and 3 (open square), or control AAV (cross, dashed line). c Proportion of Ki67-negative cells expressing Cas9 alone, Cas9 and sgDHODH, or Cas9 with miRCS, 219a-2-3pCS or 340-5pCS, and sgDHODH. d Proportion of cleaved Casp3-positive cells expressing Cas9 alone, Cas9 and sgDHODH, or Cas9 with miRCS and sgDHODH. Statistical significance was determined using a one-way ANOVA. * p < 0.05, ** p < 0.01, *** p < 0.001. Error bars indicate ±SD.

Next, we generated a set of DHODH GE AAVs encoding miRCS, infected ReN and GICs (E6 and E16) with either a single virus or a combination of them, and assessed cell survival and proliferation three days post-transduction using an MTT assay. All the GE AAVs reduced GIC proliferation and survival compared to the control AAV (which contained SaCas9 alone). In contrast, the viruses exhibited lower cytotoxicity in ReN although cell viability decreased with higher concentration of the viruses (Fig. 2b). Among the viruses, we found that the DHODH GE AAV carrying sgRNA1 was the most effective at inhibiting GIC proliferation and survival (Fig. 2b). Additionally, since the sequence of sgRNA1 is identical to the mouse sequence (Supplementary Fig. 3c), we used sgRNA1 to verify if our system could selectively eliminate GICs without affecting mouse brain cells.

We further conducted an experiment to determine whether the DHODH GE AAV inhibited GIC proliferation and induced cell death, as previously demonstrated using DHODH knockdown vectors [13]. We used 5 ×10² vg/cell under conditions in which over 20% of the GIC survived according to the MTT assay. Three days after transduction with the control AAV and the GE AAV, the cells were fixed and immunolabeling was performed for SaCas9, proliferation marker Ki67, cleaved Caspase 3 (Casp3), and differentiation markers (βIII tubulin for neurons and GFAP for astrocytes). Figure 2c and Supplementary Fig. 4 show that approximately 5% of the E6 cells and 10% of the E16 cells infected with control AAV were negative for Ki67, while 34% and 42% of the cells infected with the GE AAV were negative, respectively. Less than 6% of ReN infected with the viruses were Ki67 negative.

On the other hand, 2% of E6 cells and none of the E16 cells infected with control AAV were positive for Casp3, while approximately 20% of both E6 and E16 cells infected with GE AAV were positive (Fig. 2d and Supplementary Fig. 5). Less than 8% of ReN infected with the viruses were Casp3 positive, although the percentage of positive cells slightly increased at higher doses. Notably, GE AAV did not induce neuronal or glial differentiation of the GICs (Supplementary Figs. 4 and 5).

Next, we infected ReN and GICs with a GE AAV encoding either miR-219a-2-3pCS or miR-340-5pCS, and assessed cytotoxicity. Similar to the GE AAVs, both miRCS-encoding GE AAVs inhibited GIC proliferation and induced cell death without affecting GIC differentiation, whereas their effect on ReN was minimal (Fig. 2c, d and Supplementary Figs. 4 and 5).

Both CAP-B10 and CAP-B22 transduce an exogenous gene into the brain of NOG mice

Numerous studies have demonstrated the efficacy of braintropic AAV capsid variants, however it is unclear which capsids effectively target the brains of immunodeficient mice, such as NOD/Shi-scid and IL-2RγKO (NOG) mice. Among these capsids, we selected CAP-B10, CAP-B22, and 9P33 based on their ability to selectively deliver a gene into the brains of marmosets and C57BL/6 mice, while avoiding liver delivery [24, 37].

We administered a GFP-expressing AAV (AAV-GFP) coated with each capsid to both C57BL/6 and NOG mice. Two weeks after transduction, brain sections were prepared, and immunolabeling for GFP and neural markers was performed, including the neuronal marker MAP2, the astrocyte marker GFAP, and the oligodendrocyte marker GalC. As shown in Supplementary Fig. 6, all of capsids transduced the GFP gene into neurons, astrocytes, and oligodendrocytes in the brains of both NOG and C57BL/6 mice. Nobably, CAP-B22 exhibited slightly higher transduction efficiency in astrocytes compared to CAP-B10 and 9P33, consistent with previous findings in C57/BL6 mice [24, 37]. Given that GICs are predominantly GFAP-positive, CAP-B22 is likely the best capsid for delivering the GE system with miRCS into GICs in vivo.

Mosaic capsids-coated AAV-GFP encoding miRCS selectively expresses GFP in GIC tumor cells in the brain

There was no evidence that CAP-B22 could transduce an exogenous gene into human brain cells. Therefore, we examined the infection efficiency of CAP-B22 in GICs. The transduction efficiency of CAP-B22-coated AAV-GFP was very low, even when used an MOI of 10³. In contrast, the AAV2 capsid, which exhibits less brain specificity, was able to efficiently deliver a gene into the GICs. Furthermore, other capsids, PHP.eB, CAP-B10, or 9P33, also failed to deliver a gene into the GICs (Supplementary Fig. 7). Consequently, we investigated whether a combination of AAV2 and CAP-B22 capsids (mosaic capsids) could transduce a gene into GICs (Fig. 3a). To test this possibility, we generated a set of AAV-GFP particles coated with mosaic capsids at various ratios, infected GICs with the viruses, and examined GFP expression in the cells. The results indicated that the GFP signal correlated with the ratio of AAV2 capsid (Fig. 3b and Supplementary Fig. 8a, b). Considering both the brain tropism of AAV and its transduction efficiency into GICs, we concluded to use equal amounts of CAP-B22 and AAV2 capsids for in vivo experiments.

Fig. 3. miR-dependent AAV-GFP selectively express GFP in GIC brain tumor cells in NOG mice.

a Image of AAV with mosaic capsids, AAV2 (gray) and CAP-B22 (red). b Proportion of GFP-positive GICs, E6 (left panel) and E16 (right panel), infected with AAV-GFP with mosaic capsids consisting of different ratios of AAV2 and CAP-B22. Error bars indicate ±SD. Representative images of immunoreactivity for GFP (green) and hMitochondria (red) in E6 tumor-bearing brain of NOG mice (n = 3 mice per group) that were intravenously injected with either mosaic capsids-coated AAV-GFP c or mosaic capsids-coated AAV-GFP encoding either miR-219a-2-3pCS (left panels) or miR-340-5pCS (right panel) d. The higher magnification images are shown on the right of each figure surrounded by white dotted lines in the left images. Nuclei were counterstained with Hoechst 33342 (Blue). Scale bar, 0.1 mm.

Additionally, we examined whether the mosaic capsids affect virus production using qPCR analysis to detect the number of AAV-GFP particles with single or mosaic capsids secreted in culture. As shown in Supplementary Fig. 8d, both single and mosaic capsids generated viruses similarly.

Next, we determined whether the AAV-GFP virus coated with mosaic capsids could infect GIC tumor cells in the NOG brain. We intravenously injected the virus into E6 tumor-bearing mice at approximately 10¹¹ vg/mouse. Two weeks after injection, we immunostained brain sections for GFP and human mitochondria (hMitochondria) to detect infected cells and GICs, respectively. As shown in Fig. 3c, GFP was observed in both E6 cells and mouse brain cells, indicating that the mosaic capsids successfully transduced the GFP gene into E6 cells in the NOG brain.

To further explore the potential of mosaic capsids, we generated another AAV-GFP coated with AAV2 and 9P33 capsids. After validating its transduction efficiency in GICs in culture (Supplementary Fig. 9a), we injected the virus into GIC brain tumor-bearing NOG mice intravenously. We confirmed that AAV2/9P33 mosaic capsids also transduce the GFP gene into both GIC brain tumor cells and mouse brain cells, similar to the AAV2/CAP-B22 mosaic capsid-coated AAV (Supplementary Fig. 9b). This suggests that mosaic capsids are applicable for various types of in vivo Research.

We generated two types of mosaic capsids-coated AAV-GFP constructs encoding miRCS, injected these viruses intravenously into E6 tumor-bearing mice, and prepared sections of the brain and other tissues, including liver, kidney, spleen, and lung, followed by immunostaining for GFP. We found that GFP was selectively expressed in hMitochondria-positive E6 brain tumor cells, but not in mouse neural cells although blood vessels were likely GFP positive (Fig. 3d).

Immunohistochemical analysis of other tissues revealed that GFP+ cells were found in the liver, kidney, spleen, and lung when injected with mosaic capsids-coated AAV-GFP. This result indicated that AAV2 capsid, even co-existence with brain tropic B22 capsid, transduces GFP gene into various tissues in mice. In contrast, number of GFP-positive cells decreased in these tissues when injected with mosaic capsids-coated AAV-GFP encoding miR-219a-2-3pCS, while no GFP+ cells were detected in same tissues when injected with the AAV encoding miR-340-5pCS (Fig. 3d and Supplementary Fig. 10).

We investigated whether dose escalation increases transgene expression levels in the tissues. We examined GFP expression in the brain and liver after injecting the virus once (10¹¹ vg/mouse) or four times (4 × 10¹¹ vg/mouse in total). As shown in Supplementary Fig. 11, we observed an increase in GFP+ cells in both the brain and liver with four injections of mosaic capsids-coated AAV-GFP. Expression of GFP by the mosaic capsids-coated AAV-GFP encoding either miR-340-5pCS or miR-219a-2-3pCS was significantly inhibited at higher doses in the brain. In contrast, infection with either virus increased the number of GFP+ cells in the liver. These findings suggest that mosaic capsids effectively deliver the virus to the NOG brain, and that miRCS preferentially induces the expression of an exogenous gene in GIC brain tumors. However, higher doses of the virus encoding miRCS also lead to exogenous gene expression in the liver.

Mosaic capsids-coated AAV with miR-dependent DHODH GE system prevents GIC tumorigenesis in the NOG brain

Based on the findings regarding the gene transduction ability of mosaic capsids into GICs in the mouse brain, selective gene expression in GICs by miRCS, and GIC elimination by the DHODH GE, we tested whether AAVs integrated with these characteristics can selectively kill GIC brain tumor cells in the NOG brain (Fig. 4a). We intravenously injected mosaic capsids-coated AAV-GFP with or without a miR-dependent DHODH GE system into tumor-bearing mice at a dose of 10¹¹ vg/mouse on days 10 and 12 after GIC transplantation into NOG mouse brains and measured their survival. All mice injected with AAV-GFP died around 30 days following GIC transplantation, while the injection of AAV with either the DHODH GE system or miR-dependent DHODH GE system significantly increased the survival of GIC tumor-bearing mice (Fig. 4b). The median survival times of E6-transplanted mice were 27, 37, 47, and 36 days following infection with AAV-GFP, AAV with DHODH GE, and AAV with DHODH GE encoding either miR-219a-2-3pCS or miR-340-5pCS, respectively. Similarly, the median survival times of E16-transplanted mice were 27, 43, 42, and 44.5 days.

Fig. 4. Anti-tumorigenic activity of miR-dependent GE AAV with mosaic capsids in GIC brain tumors.

a Image of GIC specificity of mosaic capsids-coated GE AAV encoding miRCS. b Survival curve of the GIC tumor, E6 (upper panel) and E16 (lower panel), in mice (n = 5 mice per group) infected with AAV-GFP (Cross, dashed line), GE AAV (circle), GE AAV encoding miR-219a-2-3pCS (triangle), or GE AAV encoding miR-340-5pCS (square). Arrows indicate AAV injection. Statistical significance was determined using the Log-rank (Mantel–Cox) test. p = 0.0097 in both experiments. c Ratio of tumor area per brain infected with AAV-GFP (white column), GE AAV (black column), GE AAV encoding miR-219a-2-3pCS (vertical line column), or GE AAV encoding miR-340-5pCS (horizontal line column). d Quantitative data of Casp3+ cells per mm² in brain tumors that were infected with a set of viruses. Statistical significance in c and d was determined using a one-way ANOVA. * p < 0.05, *** p < 0.001. Error bars indicate ±SD.

Additionally, we examined whether intracranial injection of the virus enhances its therapeutic effects. We observed that intracranial AAV injections also prolonged the survival time of E6 brain tumor-bearing mice, similar to the results from the intravenous injection experiments (Supplementary Fig. 12). While the anti-tumorigenic activity of both injection methods was comparable, we concluded that intravenous AAV injection, being less invasive, is preferable for clinical application.

To confirm the anti-tumorigenic activity of the viruses, we stained sections of tumor-bearing brains with H&E two weeks after the second virus injection. The brain tumor sizes of mice injected with either AAV containing the DHODH GE system or AAV with miR-dependent DHODH GE system were significantly smaller compared to those of AAV-GFP injected mice (Fig. 4c and Supplementary Fig. 13a). The average ratios of E6 tumor size in the brains were 0.37, 0.1, 0.11, and 0.08 for control AAV, AAV with DHODH GE, and AAV with DHODH GE encoding either miR-219a-2-3pCS or miR-340-5pCS, respectively. Similarly, the average ratios of E16 tumor size in the brains were 0.24, 0.04, 0.05, and 0.04, respectively.

To further verify the anti-tumorigenic activity of the viruses, we immunostained the brain sections for SaCas9, a marker for AAV-infected miR-non-expressing cells, as well as hMitochondria, and either DHODH or Casp3. We confirmed that both the AAV with the DHODH GE system and the AAV with the miR-dependent DHODH GE system decreased DHODH levels in GICs (Supplementary Fig. 13b, c) and induced cell death (Fig. 4d and Supplementary Fig. 13d, e). Thus, these data indicate that systemic administration of mosaic capsid-coated AAV with the miR-dependent DHODH GE system selectively targets GICs in the NOG brain by depleting DHODH.

Discussion

Since the discovery of cancer stem cells (CSCs) in various tumor types, extensive studies have been undertaken to characterize these cells. This research has led to the identification of CSC markers, including CD133 and CD44, which represent promising therapeutic targets, as well as the development of novel therapeutic strategies, such as antibody-drug conjugates and chimeric antigen receptor-T cells [38–41]. However, these targets are not exclusive to CSCs and are also expressed in normal cells, such as tissue-specific stem cells [11, 12, 42], raising concerns about potential side effects if these factors are targeted directly. To address this issue and selectively target GICs, we developed a novel GE system that depletes DHODH in a miR expression-dependent manner. By incorporating this system into an AAV coated with brain-targeting mosaic capsids, we demonstrated that the systemic administration of the AAV selectively killed GICs in mouse brains, prevented tumorigenesis, and extended survival in the mice.

To develop a miR-dependent GE system specifically targeting GICs, we selected miR-219a-2-3p and miR-340-5p, both of which are expressed in various cell types throughout the body, including tissue-specific stem cells, such as NSCs and mesenchymal stem cells, but are expressed at lower levels in GICs. Notably, miR-219a-2-3p in NSCs acts as an anti-inflammatory factor by inhibiting the NF-κB signaling pathway and it is also delivered to surrounding cells via exosomes. This suggests that miR-219a-2-3p functions as a neuroprotective factor for both NSCs and adjacent brain cells [43]. Conversely, decreased expression of miR-219a-2-3p is advantageous for GICs as inflammation has been shown to enhance the tumorigenicity of GBM [44]. In contrast, the function of miR-340-5p in normal cells has not been established. However, it has been demonstrated that many cancer cells exhibit reduced expression of miR-340-5p, while its overexpression prevents tumorigenicity by inhibiting the expression of oncogenes and stem cell factors including SOX2 [45]. This indicates that decreased expression of miR340-5p reinforces the stemness and tumorigenicity of GICs.

Since the AAV2 capsid can deliver the AAV genome into various types of cells in vivo, it is evident that mosaic capsids, though CAP-B22 specifically targets the brain, also transduce genes into many types of tissues. However, two additional mechanisms in our system not only restrict the expression of exogenous genes in a miR-dependent manner but also knockout the DHODH gene, which is crucial for the proliferation of GICs and blast cells without affecting normal cells in vivo. This enhances the safety profile of the virus.

There is increasing evidence that cancer comprises heterogeneous CSCs, their related cancer cells, and non-cancer cells including immune and vascular cells, contributing to therapy resistance [46–49]. Moreover, CSC plasticity has been demonstrated, as CSC-marker negative cells (non-CSC) can form tumor and acquire characteristics of CSCs [50]. Since there is no data indicating how many types of CSC exist within a single tumor, it is impossible to develop tailor-made therapeutic strategies targeting each CSC type. Therefore, targeting common CSC factors, such as SOX2 and Myc, is likely a more sustainable therapeutic approach, especially when combined with strategies aimed at minimizing off-target effects on normal cells. Our system, which combines tissue targeting, a cancer specific miR-dependent gene expression system, and the knockout of CSC factors, is precisely one of the strategies suitable for addressing this challenge [51].

We successfully inhibited GIC tumorigenesis in vivo using AAVs with mosaic capsids and the miR-dependent DHODH GE system. Although the therapeutic effect of current viruses may be insufficient for clinical application, this can be enhanced by using more effective sgRNA, smaller Cas9 protein (e.g. enAsCas12f) [52], a new single braintropic capsid (e.g. BR1N) [53], or a combination of these approaches. Increased expression of the GE system may further improve the therapeutic effects of the virus by switching a stronger promoter, such as the CAG promoter, or by enhancing the genome-integration activity of AAV with other genome-integration machinery, such as Sleeping Beauty [54]. On the other hand, GFP expression from the mosaic capsid-coated AAV-GFP encoding miRCS increased in the liver with higher doses, indicating a saturation point for miR inhibition. This finding is crucial for the future development of this system.

Our system is applicable to various types of research by replacing sgRNA and capsids, depending on the target gene, cells, and tissue. For instance, deletion of interleukin-17 in Th17 cells using braintropic AAV encoding a miR-dependent knockout system may prevent the progression of multiple sclerosis. Knocking out either TNFα or IL6 in fibroblasts in a miR-dependent manner at the disease site can potentially cure rheumatoid arthritis, similar to therapeutic antibodies targeting the pathways of these factors [55, 56]. Additionally, this modified system may be used to convert the phenotype of macrophages from immunosuppressive to pro-inflammatory in tumors, by overexpressing the constitutively active form of NF-κB p65 subunit RelA while knocking down STAT6 in a miR-dependent manner.

It is clear that identifying specific capsids, which target each human tissues, is essential for establishing AAV-based gene therapy in clinical setting. Human organoid models may serve as possible screening systems for selecting capsids specific to target tissues [57]. Another option is the use of chimeric animals, which replace host tissues with human ones, except for brain organoid due to ethical concerns [58, 59]. Alternatively, as we demonstrated in this manuscript, mosaic capsids can be utilized for clinical applications, by combining different capsids according to the intended purpose.

Author contributions

TK conceived the study and designed the experiments. ZW, PZ, ZC, JH, YLS, DY and TK performed the experiments and analyzed data. ZC, JH, YLS and TK have contributed to the interpretation of the results. TK wrote the manuscript. All authors have read and approved the final version.

Funding

This work was partly supported by AMED Practical Research for Innovative Cancer Control (17ck0106236h0002), JSPS KAKENHI Grant-in-Aid for Scientific Research (20H03559), and the Joint Research Program of the Institute for Genetic Medicine Hokkaido University (all to TK).

Data availability

Additional data supporting the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

Five primary human glioma samples, including three GBMs, one anaplastic astrocytoma (AA), and one diffuse oligodendroglioma (DO), were collected from Ehime University Hospital. Informed consent was obtained from all patients according to the Research Ethics Committee guidelines (approval number: 1208009) and the specimens were used to prepare glioma lines, GICs (E3, E6, E16), AA, and DO. The glioma lines were used in compliance with the research guidelines of the Ehime University Graduate School of Medicine and the Institute for Genetic Medicine of Hokkaido University.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-025-03007-3.