Bioengineered protein nanocarrier facilitating siRNA escape from lysosomes for targeted RNAi therapy in glioblastoma

Abstract

RNA interference (RNAi) represents a promising gene-specific therapy against tumors. However, its clinical translation is impeded by poor performance of lysosomal escape and tumor targeting. This challenge is especially prominent in glioblastoma (GBM) therapy, necessitating the penetration of the blood-brain barrier (BBB). Leveraging the intrinsic tumor-targeting and BBB traversing capability of human H-ferritin, we designed a series of ferritin variants with positively charged cavity and truncated carboxyl terminus, termed tHFn(+). These nanocarriers respond to weak acid and disassemble in endosomal compartments, exposing the internal positive charges to facilitate the lysosomal escape of loaded small interfering RNA (siRNA). Functioning as universal siRNA nanocarriers, tHFn(+) significantly enhanced the uptake of different siRNAs and suppressed gene expressions associated with GBM progression. Furthermore, tHFn(+) traversed the BBB and targeted glioma in vivo by binding to its receptors (e.g., transferrin receptor 1). tHFn(+)-delivered siRNAs exhibited exceptional therapeutic effects against glioma in vivo, advancing RNAi therapeutics beyond GBM for the treatment of various diseases.

Ferritin nanocages with inner surface and interface rationally designed to deliver siRNA for glioblastoma therapy.

INTRODUCTION

The discovery of RNA interference (RNAi) presents a promising strategy for gene-specific therapy (1, 2). Small interfering RNA (siRNA), consisting of double-stranded RNA with 21 to 23 base pairs, works by binding to mRNA in the cytoplasm and suppressing the expression of sequence-specific genes (3). By aiming at pivotal genes implicated in tumor occurrence and progression, siRNA holds great promise as a compelling therapeutic tool to combat cancer (4, 5). However, siRNA molecules are hydrophilic and unable to passively cross cell membranes (6). Furthermore, siRNAs have a length of 7~8 nm with a molecular weight of 13.3 kDa, making them susceptible to elimination by the kidneys (7). In addition, once internalized by cells, siRNA molecules are typically trapped within endosomal/lysosomal compartments, resulting in degradation before exerting any effects (8). Furthermore, siRNA molecules cannot target the tumor site efficiently, thereby limiting their applications in vivo. To surmount these challenges, researchers exploited various delivery systems including lipid nanoparticles (LNPs) (9, 10), N-acetylgalactosamine–based nanocarriers (11), polymers (12), micelles (13), exosomes (14, 15), etc. (16, 17). These carriers have addressed the issues of siRNA stability and delivery efficiency; however, the cancer treatment efficacy is often unsatisfactory mainly due to poor targeting ability. Nevertheless, the tumor-targeting difficulties of nanocarriers are particularly acute in the treatment of glioblastoma (GBM), where siRNA drugs need to overcome a biological hurdle—the blood-brain barrier (BBB) (18). Therefore, to achieve optimal targeted therapy, there is an urgent need for a carrier that can effectively traverse the BBB and target GBM.

Ferritin is a naturally occurring protein discovered in numerous organisms. Mammalian ferritin features a hollow spherical structure and the reversible disassembly/assembly process mediated by pH, rendering it an optimal nanocarrier for drug loading (19, 20). Furthermore, human recombinant heavy chain ferritin (HFn) is capable of specifically targeting many tumor cells that express high levels of transferrin receptor 1 (TfR1) (21, 22). Notably, ferritin drug carriers can traverse the BBB via binding to its receptors and deliver drugs to glioma specifically (23). Owing to its uniform size, exceptional biocompatibility, high stability, minimal toxicity, ease of modification, and facile production, ferritin has been extensively studied as a promising drug carrier in cancer theranostics (24–26). Specific modifications can be engineered to the native HFn to improve its nucleic acid loading capacity. In our previous work, we constructed ferritin variants by genetic manipulation, which endowed ferritin with high nucleic acid affinity and effective loading ability for different nucleic acid agonists of Toll-like receptors (TLRs), suggesting the potential of ferritin as nucleic acid carriers (27). In contrast to TLR-activating nucleic acids that act in the endosomes and lysosomes, siRNAs must enter the cytosol to be effective. Consequently, for the development of RNAi therapeutics, ferritin not only needs to be engineered for effective loading but also is required to be capable of lysosomal escape.

Given the environment-responsive disassembly properties of ferritin, it is of great potential to use its self-assembly behavior to reasonably design ferritin-based siRNA carriers with lysosomal escape function. According to the structure analysis of ferritin, the subunit of HFn consists of four α helices (A to D) and a short α helix (E), with a total of 24 subunits self-assembling into a nanocage. Ferritin nanocages boast with 6 fourfold channels (C₄ interfaces), 8 threefold channels (C₃ interfaces), 12 twofold channels (C₂ interfaces), and 24 C₃-C₄ interfaces (i.e., B-channels) (28). Among them, C₂ and C₃-C₄ interfaces mainly promote the assembly of ferritin nanocages (29). Thus, the overall architecture and oligomerization state of ferritin can be manipulated by removing, introducing, or substituting amino acids at crucial interfaces (30).

In light of these findings, we aimed to affect the interactions between the C₃-C₄ interfaces through truncating the C terminus. We hoped to construct ferritin carriers with weak acid–responsive properties that can escape from endosomes/lysosomes. Regarding the issue of siRNA loading and delivery, we developed a series of tHFn(+) variants based on native HFn through genetic engineering with an enlarged internal cavity and a positively charged surface. The ferritin was modified by truncating the protein sequence, taking advantage of the effect of the C₃-C₄ interface on assembly (29). It has been demonstrated that tHFn(+) nanocarriers expose the internal positive charges under weak acid conditions of endosomes. In addition, this modification also improved siRNA loading. These variants were demonstrated to display a spherical architecture, self-assembly ability, and efficient encapsulation of siRNA. Furthermore, tHFn(+) was found to deliver siRNA against different targets to suppress the expression of various genes. In vivo experiments proved that tHFn(+) was able to traverse the BBB and deliver siRNA to the tumor sites in the brain. Animal experiments demonstrated that siRNA@tHFn(+) effectively inhibited glioma growth and knocked down target genes in vivo, outperforming siRNA encapsulated in the LNPs. This finding provides a potential platform for ferritin as a universal siRNA carrier, which may have potential therapeutic applications for multiple diseases (Fig. 1).

Fig. 1. Schematic illustration of siRNA@tHFn(+) for glioma-targeted therapy.

The process of tHFn(+) delivering siRNA into a cell occurs as follows: tHFn(+) binds to the cell membrane of glioma through TfR1. Subsequently, the glioma cell engulfs siRNA@tHFn(+) through endocytosis, forming vesicles known as endosomes. tHFn(+), along with the encapsulated siRNA payloads, is internalized into the cell. tHFn(+) undergoes conformational changes in response to the acidic environment of the endosomes and reveals the positive charges, allowing the lysosomal escape. At the same time, tHFn(+) nanocarrier releases the siRNA cargos, allowing it to bind to mRNA, which can be recognized and cleaved by the RNA-induced silencing complex (RISC). As a result, the target mRNA is degraded, leading to suppression of protein expression. This process effectively silences the expression of target genes involved in tumor progression, ultimately inhibiting GBM growth and proliferation.

RESULTS

Design and characterization of tHFn(+) variants

To enhance the loading efficiency of siRNA molecules, we initially analyzed the internal structure of ferritin, which revealed a number of negatively charged groups with high density. Subsequently, the mutation sites of the negatively charged amino acids were screened through rational design (Fig. 2A). There are two types of negatively charged amino acids, including glutamic acids [Glu, isoelectric point (pI) = 3.22] and aspartic acids (Asp, pI = 2.77). Each HFn subunit contains 16 Glu sites and 15 Asp sites. On the basis of the distribution of negatively charged amino acids, those located at the inner surface were selected as the initial candidates for mutation. Then, amino acids with side chains facing outward or hidden between subunits were excluded. Last, seven positions of amino acids were selected as mutation sites (Glu⁶¹, Glu⁶⁴, Glu¹⁴⁰, Glu¹⁴⁷, Asp¹⁷⁷, Asp¹⁷⁹, and Glu¹⁸¹). Arginine (Arg, pI = 10.76) and lysine (Lys, pI = 9.74) are positively charged amino acids, which can be used to substitute Glu or Asp. Nevertheless, because of steric hindrance, Arg cannot be widely used for substitution in the ferritin structure; otherwise, it may result in the instability of the protein structure. The substitutions were designed based on the spatial environment surrounding the selection sites. As the number of mutation sites increases, the inner cavity tends to become positively charged (Fig. 2B and table S1). Last, the mutation scheme was determined to be E61K+E64R+E140K+E147K+D177K+D179K+E181K, termed as HFn(+).

Fig. 2. Design, purification, and characterization of HFn, HFn(+), and tHFn(+) variants.

(A) Schematic diagram of the screening process of the mutation sites in HFn. (B) Surface charge distribution of HFn and positively charged variants calculated by PyMOL. (C) Schematic illustration of HFn, HFn(+), and tHFn(+) variants. (D) SEC analysis of HFn, HFn(+), and tHFn(+) variants. mAu, milli-absorbance unit. (E) Size distribution and representative TEM images of HFn, HFn(+), and tHFn(+) variants. Scale bars, 50 nm. (F and G) SDS-PAGE (F) and Native-PAGE (G) analysis of HFn, HFn(+), and tHFn(+) variants. (H) Zeta potential of HFn, HFn(+), and tHFn(+) variants. Data are shown as means ± SEM (n = 3 technical replicates).

To further expand the internal cavity and modify the interfacial interactions, we rationally designed a series of truncated HFn(+) variants, namely, tHFn(+)^1-174, tHFn(+)^1-162, tHFn(+)^1-159, tHFn(+)^1-156, tHFn(+)^1-153, and tHFn(+)^1-150 (Fig. 2C and table S2). It has been reported that the truncation of ferritin enables it to disassemble under weak acid conditions (31). Therefore, these variants have the potential to disassemble in the endosome/lysosome and expose internal positively charged groups, allowing for escape. Moreover, C terminus–truncated ferritin has a greater internal space and can accommodate more siRNA molecules.

The HFn and tHFn(+) variants were successfully expressed in Escherichia coli and purified by anion ion-exchange chromatography (AIEX), hydrophobic interaction chromatography (HIC), and size exclusion chromatography (SEC). According to the SEC profile, the oligomerization state of HFn(+), tHFn(+)^1-174, tHFn(+)^1-162, and tHFn(+)^1-159 were consistent with the native HFn (Fig. 2D). After purifying all proteins, we analyzed their hydrodynamic diameter and structure integrity. Dynamic light scattering (DLS) and transmission electron microscopy (TEM) revealed the dispersibility and structural morphology of these proteins (Fig. 2E, fig. S1, and table S3). DLS and TEM results show that tHFn(+)^1-156, tHFn(+)^1-153, and tHFn(+)^1-150 were unable to self-assemble into uniform nanocages, and late eluting peaks were observed. Combined together, tHFn(+)^1-156, tHFn(+)^1-153, and tHFn(+)^1-150 exhibited abnormal assembly, possibly due to the incomplete D helix, which is known to play a role in ferritin assembly (29). Consequently, these variants will not be used for subsequent studies. As determined by DLS, HFn(+), tHFn(+)^1-174, tHFn(+)^1-162, and tHFn(+)^1-159 exhibited an average particle size of 14.65 ± 0.55, 12.35 ± 0.37, 12.08 ± 0.54, and 13.19 ± 0.11 nm, respectively, similar to the 13.33 ± 0.64 nm of HFn. TEM images demonstrated the cage-like structure of these tHFn(+) variants, indicating that their overall structures are not affected by the substitutions and truncations. SDS–polyacrylamide gel electrophoresis (SDS-PAGE) showed that the bands of the HFn and tHFn(+) variants were located at 18 to 21 kDa, corresponding to their theoretical molecular weight values, as shown in Fig. 2F. Moreover, the major bands of these variants were close to HFn in Native-PAGE analysis, suggesting that tHFn(+) variants can self-assemble as the native HFn (Fig. 2G). We also measured the zeta potentials of HFn and tHFn(+) variants, which demonstrated that mutations inside ferritin and truncations did not influence the surface charges (Fig. 2H). In summary, four of tHFn(+) variants were successfully obtained and can be used for further research.

Screening of tHFn(+) variants and mechanisms of lysosomal escape

After characterizing the tHFn(+) variants, we began to screen the optimal siRNA nanocarriers. Lysosomal escape remains a bottleneck in the development of RNAi. siRNA requires lysosomal escape to avoid degradation and enter the cytoplasm, where it binds to its target mRNA and carries out its function (32). Unstable protein carriers such as tHFn(+)^1-156, tHFn(+)^1-153, and tHFn(+)^1-150 have been excluded as described previously. The evaluation of the optimal nanocarrier was based on its abilities to escape lysosomes (Fig. 3A). We first screened suitable nanocarriers by studying the colocalization of siRNA and lysosomes. U87MG cells were treated with Cy5.5-labeled HFn, HFn(+), and tHFn(+) variants for 3 hours, and confocal laser scanning microscopy (CLSM) was used to observe the colocalization profile. The Pearson’s correlation coefficient (PCC) was calculated to reflect the level of colocalization between protein and lysosomes (33). Higher values of PCC indicate stronger correlation. As the length of truncation increases, the PCC of protein decreases (Fig. 3, B and C). The PCC of tHFn(+)^1-159 is less than 0.5, demonstrating notable lysosomal escape. Therefore, we chose tHFn(+)^1-159 [hereafter referred to as tHFn(+)] as the optimal siRNA carrier and investigated the mechanism of lysosomal escape.

Fig. 3. Protein screening and mechanisms of lysosomal escape of tHFn(+) variants.

(A) Schematic illustration of protein screening by lysosomal escape. (B and C) Representative CLSM images (B) and the corresponding PCC analysis (C) (n = 5 biological replicates) of Cy5.5-labeled HFn and tHFn(+) variants with lysosomes after incubation with U87MG cells for 3 hours. Scale bar, 10 μm. (D) Zeta potential of tHFn(+) at pH 4.0, 5.0, and 6.0. Data are shown as means ± SEM (n = 3 technical replicates). (E) Native-PAGE of HFn and tHFn(+) at different pH values. (F) TEM images of tHFn(+) at pH 7 to 4. Scale bars, 100 nm (enlarged area: 25 nm). The arrow indicated the incomplete ferritin nanocages. (G) Cryo-EM structure of tHFn(+) (colored in orange, residues 5 to 159) superposed with HFn (PDB 8F49; colored in green for residues 5 to 159 and colored in gray for residues 160 to 176). The figure on the right is an enlargement of the residues 160 to 176 from HFn. (H and I) Schematic representation of the C₃-C₄ interface and the distance of hydrogen bonds within the interface measured by ChimeraX. X, Y, and Z represent different subunits in the C₃-C₄ interface. (J and K) Merged CLSM images (J) and the corresponding PCC analysis (K) of Cy5.5-HFn and Cy5.5-tHFn(+) with endosomes after incubation with U87MG cells for 15 and 30 min. Scale bar, 10 μm. Data are shown as means ± SEM (n = 5 biological replicates). (L and M) Merged CLSM images (L) and the corresponding PCC analysis (M) of Cy5.5-HFn and Cy5.5-tHFn(+) with lysosomes after incubation with U87MG cells for 15, 30, 60, and 120 min. Scale bar, 10 μm. Data are shown as means ± SEM (n = 5 biological replicates).

First, Native-PAGE was conducted to investigate the acid responsiveness of HFn and tHFn(+), which revealed a difference in the assembly state mediated by pH (Fig. 3E). Specifically, HFn began to disassemble at pH 4 (34), whereas tHFn(+) began to disassemble at pH 6. Similarly, TEM images indicated that tHFn(+) retained its spherical structure at pH 7 and exhibited a tendency to disassemble at pH 6 and completely disassembled at pH 4 (Fig. 3F). These protein subunits that disassembled at varying pH values were capable of reassembling to complete nanocages at pH 8, suggesting that tHFn(+) exhibits a similar reversible self-assembly ability to that of HFn (figs. S2 and S3). In addition, We performed zeta potential analysis to evaluate the charges on the external surface of tHFn(+) (Fig. 3D). As the pH decreased from 6 to 5, the outer surface charges of the tHFn(+) underwent a flip from negative to positive, suggesting that the structure of tHFn(+) altered.

To gain insight into the mechanism of tHFn(+) disassembly in a weak acidic environment, we used cryo–electron microscopy (cryo-EM) to resolve the protein structure (fig. S4, A to D). A total of 5362 micrograph stacks were recorded, and 1,725,802 particles were selected for two-dimensional (2D) classification. The final structure was solved to an overall resolution of 1.73 Å with 382,703 particles for homogeneous refinement and contrast transfer function (CTF) refinement procedures. Similarly to HFn, tHFn(+) presents as a spherical structure composed of 24 subunits. The root mean square deviation of the main chain between tHFn(+) and wild-type HFn [Protein Data Bank (PDB) ID: 8F49] is 0.20 Å, indicating that the two share an almost identical structure. Apart from the minor shift among subunits, the region that differs most within the subunit is located at the C terminus, where tHFn(+) exhibits larger fourfold pores due to the truncation of the fifth helix (Fig. 3G). The fourfold hydrophobic core, constructed from residues of the adjacent four subunits, forms a tighter link among subunits whose absence would undoubtedly weaken the interface interactions of tHFn(+). Moreover, the overall interfacial hydrogen bonding of tHFn(+) is weaker than that of HFn, which also contributes to tHFn(+) dissociation, endowing tHFn(+) with weak acid–responsive structural transformation properties.

It was reported in the previous literature that C₃-C₄ interfaces of ferritin play a crucial role in protein assembly (35–37). Therefore, we analyzed the hydrogen bonds between subunits in the C₃-C₄ interfaces. As shown in Fig. 3 (H and I), the overall interfacial hydrogen bonding of tHFn(+) is weaker than that of native HFn. These findings provided a possible mechanism for the endosomal/lysosomal escape of siRNA@tHFn(+). In the acidic environment of endosomes (pH 6.0~6.5), tHFn(+) dissociated and exposed the internal positive charges, which could then cause damage to the endosome compartment, leading to the release of encapsulated siRNA (38).

To verify the mechanism of endosomal/lysosomal escape of tHFn(+), U87MG cells were treated with Cy5.5-labeled HFn and tHFn(+) for 15 and 30 min to examine the colocalization of protein with endosomes. CLSM images demonstrated that these nanocages entered into endosomes after 15 min (Fig. 3, J and K, and fig. S5). We subsequently investigated the colocalization of protein with lysosomes. Figure 3 (L and M) and fig. S6 illustrated that HFn enriched in the lysosome from 15 to 120 min. In contrast, tHFn(+) did not colocalize with lysosomes, implying that tHFn(+) escaped from lysosomes. Both HFn and tHFn(+) entered the tumor cell via TfR1-mediated endocytosis. However, the lysosomal colocalization was different due to the weak acid–responsive disassembly property of tHFn(+). It has been demonstrated that tHFn(+) begins to disassemble at pH 6 (close to the pH of endosome environment), whereas wild-type ferritin starts to disassemble at pH 4, which is able to remain intact in the endosomes. It is indicated that tHFn(+) disassembled and exposed its positive charges of internal surface in the acidic environment of endosomes to facilitate escape, whereas HFn tended to further enter and accumulate in the lysosome via TfR1-mediated endocytosis. All these results verified that tHFn(+) achieved lysosomal escape.

siRNA encapsulation and characterization of siRNA@tHFn(+)

After screening the tHFn(+) variants and verifying the mechanism of lysosomal escape, we further studied the encapsulation ability of tHFn(+). siRNA@tHFn(+) was prepared through the pH-mediated disassembly/reassembly strategy, following a previously reported procedure (27). We initially assessed the loading capacity and protein recovery of HFn and tHFn(+) variants. As shown in Fig. 4A, tHFn(+) series exhibited higher encapsulation of siRNA molecules. Specifically, tHFn(+)^1-159 showed the highest loading efficiency among the protein variants (~1.49 siRNA molecules per ferritin nanocage). In addition, the protein recovery of the tHFn(+) series was comparable to that of the HFn (fig. S7). Subsequently, we conducted the characterization of siRNA@tHFn(+). DLS and TEM were applied to assess the structural stability of these nanocages. As presented in Fig. 4B, siRNA@tHFn(+) had a diameter of 13.73 ± 0.42 nm and formed into a spherical cage similar to HFn and tHFn(+). Furthermore, zeta potential analysis suggested that siRNA was encapsulated inside tHFn(+) as its charges were consistent before and after siRNA encapsulation (Fig. 4C). Agarose gel electrophoresis (AGE) demonstrated that the band of encapsulated siRNA appeared close to the loading well due to the electrophoretic mobility shift, which means that protein–nucleic acid complexes migrate more slowly than free nucleic acids (Fig. 4D). Free siRNA migrated further and was susceptible to digestion by ribonuclease (RNase) A, whereas tHFn(+)-encapsulated siRNA remained stable in the presence of RNase A. The enlarged pores of the fourfold channel (~19 Å) in the tHFn(+) permit the embedment of a certain amount of siRNA molecules; however, the loading quantity is smaller than that of the pH method. A proportion of siRNA molecules may be exposed on the external surface of tHFn(+) nanocages, which were susceptible to digestion by RNase A. These results confirmed that siRNA was loaded within ferritin nanocages during the pH-mediated reassembly process rather than adhering to the outer surface. Furthermore, the in vitro stability of siRNA@tHFn(+) was assessed through coincubation with Dulbecco’s modified Eagle’s medium (DMEM) containing 10% serum at 37°C for 1, 4, and 24 hours. As presented in fig. S8, the siRNA band exhibited no noticeable changes, indicating that siRNA@tHFn(+) remained stable.

Fig. 4. Evaluation of siRNA encapsulation and characterization of siRNA@tHFn(+).

(A) Molar ratio of siRNA to HFn and tHFn(+) variants following siRNA encapsulation. Data are shown as means ± SEM (n = 3 biological replicates). (B) DLS analysis and TEM images of siRNA@tHFn(+). Scale bar, 40 nm. (C) Zeta potential of HFn, tHFn(+), siRNA, and siRNA@tHFn(+). Data are shown as means ± SEM (n = 3 technical replicates). (D) AGE patterns of siRNA encapsulation. (E) Surface charge distribution of HFn and tHFn(+) calculated by PyMOL. (F) Representative CLSM images of Cy5-siRNA@tHFn(+) after incubation with U87MG cells for 15 and 30 min. Scale bar, 10 μm. (G) Representative CLSM images of Cy5-siRNA@tHFn(+) after incubation with U87MG cells for 15, 30, 60, and 120 min. Scale bar, 10 μm. (H and I) Corresponding PCC analysis of Cy5-siRNA with endosomes (F) and lysosomes (G). Data are shown as means ± SEM (n = 5 biological replicates).

We further analyzed the charge distribution of the interior surface of both HFn and tHFn(+). According to the electrostatic potential analyzed by PyMOL, the inner surface of tHFn(+) had an overall positive charge, in contrast to the negatively charged interior surface of HFn, which explains the ideal tHFn(+) nanocarrier to encapsulate siRNA molecules within its cavity (Fig. 4E). The positively charged inner surface contributed to siRNA encapsulation, and the loading capacity of tHFn(+) increased with the truncation length.

To confirm the lysosomal escape of siRNA@tHFn(+), Cy5-siRNA was loaded into tHFn(+). The colocalization of siRNA@tHFn(+) with endosomes suggests that tHFn(+) nanocages deliver siRNA to the endosome and the siRNA molecules escaped at 30 min, thereby avoiding the enzyme degradation in the lysosomes (Fig. 4, F and H). It appeared that siRNA molecules achieved lysosomal escape at a faster rate than tHFn(+) due to its smaller size. As shown in Fig. 4 (G and I), most of siRNA@tHFn(+) was found outside of the lysosomes, indicating the successful release of siRNA into the cytoplasm, which is consistent with the result of tHFn(+). In conclusion, the tHFn(+) nanocages demonstrated exceptional performance in encapsulating siRNA molecules. Furthermore, siRNA@tHFn(+) was capable of escaping lysosomes, which will be further investigated both in vitro and in vivo functional experiments.

Cellular uptake and gene knockdown efficiency of siRNA@tHFn(+)

After studying the mechanism of lysosomal escape of tHFn(+), we conducted in vitro experiments to evaluate its functions. To investigate the cell uptake of siRNA, U87MG cells were treated with Cy5-labeled siRNA and siRNA@tHFn(+) for 10 min, 2 and 4 hours, respectively. The uptake of free siRNA and siRNA@tHFn(+) were visualized through CLSM (Fig. 5A). Flow cytometry demonstrated that tHFn(+) significantly enhanced the cellular uptake of siRNA (Fig. 5, B and C). At 10 min, the naked siRNA and siRNA@tHFn(+) were both located at the membrane. However, after 2 or 4 hours, the naked siRNA molecules were unable to traverse the cytoplasmic membrane due to their hydrophilic nature and were progressively degraded owing to their instability. Instead, most siRNA@tHFn(+) entered the cytosol after 2 hours, and by the fourth hour, it was fully taken up into the cell. These findings indicate that tHFn(+) enables the efficient uptake of siRNA, addressing the instability challenges of naked siRNA.

Fig. 5. Cellular uptake and gene silencing efficiency of siRNA@tHFn(+) in U87MG cells.

(A) Cellular localization of Cy5-siRNA and Cy5-siRNA@tHFn(+) at different times observed by CLSM (blue: nuclei stained with DAPI; red: Cy5-labeled siRNA). Scale bars, 40 μm (enlarged area: 10 μm). h, hours. (B) Flow cytometry analysis of cellular uptake of free Cy5-siRNA and Cy5-siRNA@tHFn(+) in U87MG cells. (C) Quantitative analysis of Cy5 fluorescence of cell uptake. Data are shown as means ± SEM (n = 3 biological replicates). a.u., arbitrary units. (D and E) WB assays (D) and corresponding grayscale values (E) of TERT expression in U87MG cells treated with indicated formulations for 48 hours. Data are shown as means ± SEM (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparisons test. ****P < 0.0001. (F) RT-qPCR analysis of relative TERT transcriptional levels with different treatments for 24 hours. Data are shown as means ± SEM (n = 3 technical replicates). Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparisons test. *P < 0.05; ****P < 0.0001. (G and H) WB analysis (G) and corresponding grayscale values (H) of EGFR protein levels treated with control or siEGFR-loaded nanocarriers for 48 hours. Data are shown as means ± SEM (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparisons test. ***P < 0.001; ****P < 0.0001; ns, no significant difference. (I) RT-qPCR analysis of relative EGFR mRNA levels treated with different formulations for 24 hours. Data are shown as means ± SEM (n = 3 technical replicates). Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparisons test. ***P < 0.001; ****P < 0.0001; ns, no significant difference.

To examine the gene silencing efficacy of siRNA@tHFn(+), we initially selected telomerase reverse transcriptase (TERT) as the therapeutic target for glioma treatment. Mutations in TERT lead to an increase in telomerase, which plays an important role in maintaining the activity in malignant cells (39). TERT mutations are frequently found in patients with glioma, making it a potential therapeutic target (40–42). siRNA targeting TERT was synthesized and encapsulated within tHFn(+) nanocages, named as siTERT@tHFn(+). To compare the delivery efficiency of tHFn(+) carriers with commonly used LNPs, we also prepared an siRNA-loaded LNP system with reference to the previous literature (43). U87MG cells were incubated with control and siTERT-loaded formulations for 48 hours. As depicted in Fig. 5 (D and E), Western blot (WB) results indicated that free siTERT and tHFn(+) had minimal impact on the expression levels of TERT, whereas siTERT@tHFn(+) knocked down TERT in a dose-dependent manner. Moreover, siTERT@LNP also showed effective knockdown of TERT. In addition, real-time quantitative polymerase chain reaction (RT-qPCR) was used to assess the mRNA levels of TERT. The transcriptional levels of TERT were also noticeably suppressed, consistent with the WB analysis (Fig. 5F). TERT mRNA levels declined with the treatment of siTERT delivered by LNP and tHFn(+), whereas tHFn(+), siTERT, or siNC@tHFn(+) had no impact on gene transcription. We further explored the universal suitability of tHFn(+) in siRNA delivery. Epidermal growth factor receptor (EGFR) plays a crucial role in receptor tyrosine kinase signaling, affecting the proliferation and progression of glioma (44, 45). EGFR amplification is a common occurrence in GBM, which has been identified as an effective target for glioma therapy (46, 47). As previously, siEGFR@tHFn(+) and siEGFR@LNP were fabricated and used in U87MG cells. WB and qPCR results showed that siEGFR-loaded tHFn(+) and LNP down-regulated EGFR protein levels efficiently in comparison with controls (Fig. 5, G to I). Furthermore, siEGFR@tHFn(+) demonstrated a concentration-dependent decrease in the expression of EGFR. To summarize, tHFn(+) serves as a versatile carrier, capable of accommodating various siRNA molecules for gene silencing. In vitro studies have revealed the immense potential of tHFn(+) in siRNA encapsulation and delivery, which can be further used in glioma therapy.

The ability of tHFn(+) to traverse the BBB

It has been reported that HFn enters tumor cells by binding to TfR1, which is overexpressed in many tumor cells (21, 22). Our group has also demonstrated that HFn binds to its receptor in brain endothelial cells and crosses the BBB via receptor-mediated transcytosis (23). Before in vivo verification, we compared the binding ability of genetically modified tHFn(+) for HFn receptor with that of HFn in vitro. Enzyme-linked immunosorbent assay (ELISA) was applied to determine the binding ability of HFn and tHFn(+) toward their receptor, showing that their receptor binding affinities were comparable (fig. S9). Moreover, structural alignment of HFn and tHFn(+) revealed that the outer surfaces of the two proteins are almost identical, further indicating similar receptor binding capabilities (Fig. 6A).

Fig. 6. Evaluation of the ability of siRNA@tHFn(+) to cross the BBB in vitro and in vivo.

(A) Superposition of tHFn(+) with HFn. HFn is colored in green, whereas tHFn(+) is colored in orange. (B) Schematic illustration of the transwell model. (C) Transcytosis efficiency of indicated protein or formulations in the transwell model. Data are shown as means ± SEM (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparisons test compared with the mean of the control (column LFn). ***P < 0.001; ****P < 0.0001; ns, no significant difference. (D) Native-PAGE of HFn and tHFn(+) before and after crossing the in vitro BBB models. HFn (blank) indicated that endothelial cells were not seeded on the microporous membrane of the transwell plate. (E) Molar ratio of siRNA to tHFn(+) of siRNA@tHFn(+) before and after crossing the in vitro BBB models. Data are shown as means ± SEM (n = 3 biological replicates). Statistical significance was determined by two-tailed unpaired Student’s t test. ns, no significant difference. (F) Ex vivo NIRF images of main organs in orthotopic glioma-bearing mice after intravenous injection of Cy5-siRNA or Cy5-siRNA@tHFn(+). B, brain; H, heart; LI, liver; S, spleen; LU, lung; K, kidney. (G) Quantitative Cy5 fluorescence intensity of different siRNA-loaded NPs in the brain. Data are shown as means ± SEM (n = 3 biological replicates). Statistical significance was determined by two-way ANOVA with Tukey’s multiple comparisons test. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. (H) In vivo NIRF images of orthotopic glioma-bearing mice after intravenous injection of Cy5-siRNA, Cy5-siRNA@LNP, or Cy5-siRNA@tHFn(+). The yellow circle indicates the brain area.

We used an in vitro transwell model to simulate the BBB and study the transcytosis ability of tHFn(+) nanocarriers (Fig. 6B). It was demonstrated that tHFn(+) and siRNA@tHFn(+) effectively cross the BBB as HFn (Fig. 6C). However, few siRNA@LNP NPs managed to traverse the BBB. Cytotoxicity assay of bEnd.3 cells demonstrated that tHFn(+) is not toxic to brain endothelial cells, thus preventing the disruption of the BBB (fig. S10). It was revealed that tHFn(+) exhibited a comparable affinity for its receptor compared to HFn and achieved transcytosis in the brain endothelial cells, which opens up the possibility of using tHFn(+) carriers in brain tumor therapy.

Our previous research demonstrated that the potential mechanism of ferritin traversing the BBB is receptor-mediated transcytosis (23). We further studied the mechanism of traversing the BBB. As demonstrated in fig. S11, both HFn and tHFn(+) were observed within endosomes rather than lysosomes in brain endothelial cells, consistent with our previous findings. Furthermore, in vitro BBB models were constructed to assess protein integrity after traversing the BBB. According to the Native-PAGE results, both HFn and tHFn(+) were capable of maintaining structural stability (Fig. 6D). Moreover, the siRNA loading amount of siRNA@tHFn(+) remained consistent before and after traversing the BBB, suggesting that the transcytosis of endothelial cells did not disrupt the structure of siRNA@tHFn(+) (Fig. 6E).

We then examined the plasma pharmacokinetics of free siRNA, tHFn(+), and siRNA@tHFn(+). The Cy5-siRNA, Cy5.5-tHFn(+), and Cy5-siRNA@tHFn(+) were injected intravenously, and then plasma samples were collected from the tail at different times. As illustrated in fig. S12, tHFn(+) and siRNA@tHFn(+) exhibited a half-life of 41.29 and 44.94 min, respectively, which was eightfold greater than that of free siRNA (5.03 min). It was found that tHFn(+) remarkably prolonged the retention time of siEGFR.

To investigate the targeting ability and metabolism of free siRNA and siRNA-loaded NPs, an in vivo biodistribution study was conducted using Balb/c-nu mice with orthotopic glioma. The mice were intravenously injected with Cy5-labeled siRNA, siRNA@LNP, or siRNA@tHFn(+), and their major organs along with brain tissues were collected at intervals of 1, 3, 6, 12, and 24 hours. Ex vivo near-infrared fluorescence (NIRF) imaging showed that siRNA@tHFn(+) accumulated in the tumor area of the brain, indicating its excellent targeting ability and capability of traversing the BBB (Fig. 6, F and G, and fig. S13). On the other hand, the free siRNA group exhibited minimal signals in the brain, which may be attributed to its instability in plasma. The signal of siRNA@LNP was largely concentrated in the liver and spleen, whereas little signal was observed in the brain possibly due to its low efficiency in crossing the BBB. Tissue biodistribution analysis revealed stronger fluorescence of free siRNA and siRNA@tHFn(+) in the liver and kidneys, with little signal detected in the heart, spleen, and lung, suggesting that these NPs were metabolized by the liver and kidneys. Another study was performed in the same orthotopic glioma models, which were subjected to intravenous injection and monitored at 6, 12, and 24 hours postinjection. As depicted in Fig. 6H and fig. S13, free siRNAs were predominantly detected in the liver and kidneys, and they were rapidly metabolized, whereas siRNA@LNP was accumulated in the liver and spleen in the first 6 hours and gradually declined thereafter. Notably, it is demonstrated that siRNA@tHFn(+) was concentrated in the brain for up to 24 hours postinjection. In conclusion, tHFn(+) was able to overcome the BBB effectively and reach the tumor, proving to be an exceptional carrier of siRNA in vivo.

In vivo therapeutic efficacy of siRNA@tHFn(+) in orthotopic glioma models

On the basis of our research, it has been proven that tHFn(+) is able to traverse the BBB and selectively target glioma in vivo. To further explore the therapeutic potential of siTERT@tHFn(+) and siEGFR@tHFn(+), in vivo experiments were conducted on Balb/c-nu mice with orthotopic glioma. As illustrated in Fig. 7A, orthotopic glioma models were established by injecting U87MG-luc-mCherry cells into the mouse brain, and the mice were randomly divided into six groups. Subsequently, phosphate-buffered saline (PBS), tHFn(+), siTERT, siNC@tHFn(+), siTERT@LNP, and siTERT@tHFn(+) were administered intravenously every other day over 2 weeks at an siRNA dose of 1 mg/kg. Glioma volume was monitored through bioluminescence imaging (Fig. 7C). The results demonstrated that siTERT@tHFn(+) effectively inhibited the growth of glioma, whereas tumors in the other groups grew rapidly. Quantitative analysis indicated a tumor inhibition rate of 78.9% (Fig. 7D). Although siTERT delivered by LNPs showed excellent knockdown effects at the cellular level, no therapeutic effects were observed in the in vivo animal experiments. This may be due to the inability of LNPs to cross the BBB and reach the tumor site. Hematoxylin and eosin (H&E) staining of brain sections showed that the tumor area of the siTERT@tHFn(+)-treated group was smaller than in the other groups, demonstrating visible tumor regression effects (Fig. 7E). As shown in Fig. 7B, no notable changes in body weight were observed in any group, highlighting the low toxicity of siTERT@tHFn(+). To explore the gene silencing efficiency of siTERT@tHFn(+), WB and immunofluorescence were used to detect the expression of TERT in vivo. WB revealed a reduced protein level of TERT in the treatment group (Fig. 7F). Immunofluorescence imaging also demonstrated that TERT expression was down-regulated as evidenced by its weak signals in tumor slices of siTERT@tHFn(+) group (Fig. 7G). Therefore, siTERT@tHFn(+) effectively inhibited the growth of glioma in vivo via knocking down TERT expression. The above data suggest that tHFn(+) nanocarriers effectively deliver siTERT to the tumor site in vivo and exhibit significant therapeutic outcomes. However, siTERT@LNP failed to suppress the growth of glioma because of poor BBB traversing and tumor uptake.

Fig. 7. In vivo anti-glioma therapeutic effect of siTERT@tHFn(+) in U87MG-luc-mCherry orthotopic mouse models.

(A) Schematic illustration of glioma establishment and treatment. iv, intravenously. (B) Body weight of mice in different treatment groups. Data are shown as means ± SEM (n = 6 to 7 biological replicates). (C) In vivo bioluminescence images of glioma-bearing mice treated with PBS, tHFn(+), siTERT, siNC@tHFn(+), siTERT@LNP, or siTERT@tHFn(+). An equivalent dose of siTERT/siNC (1 mg/kg) was applied. (D) Quantitative analysis of tumor growth by bioluminescence signals of (C). Data are shown as means ± SEM (n = 6 to 7 biological replicates). Statistical significance was determined by two-way ANOVA with Tukey’s multiple comparisons test. *P < 0.05; **P < 0.01. (E) Representative images of H&E staining of brain tissues collected at day 21. (F) WB of TERT expression in glioma of Balb/c-nu mice with different treatments. (G) Representative immunofluorescence images of TERT expression in glioma with different formulations. Scale bars, 40 μm.

To verify the universality of tHFn(+), we further investigated the in vivo performance of siEGFR@tHFn(+). The tumor model was established the same as previously mentioned. PBS, siEGFR, tHFn(+), and siEGFR@tHFn(+) were administered intravenously every other day over 2 weeks at an siEGFR dose of 1 mg/kg, starting on day 7 (Fig. 8A). Bioluminescence imaging demonstrated that siEGFR@tHFn(+) achieved an inhibition rate of 63.3%, whereas tumors in the free siEGFR-treated and tHFn(+)-treated group grew rapidly (Fig. 8, C and D). During the treatment course, there were no distinct changes of body weight (Fig. 8B). H&E staining results indicated the siEGFR@tHFn(+)-treated group exhibited the most substantial suppression of glioma (Fig. 8E). Flow cytometry analysis verified that the EGFR protein was down-regulated in vivo (Fig. 8F). As before, immunofluorescence was applied to evaluate EGFR levels in glioma. Figure 8G shows that the signal of EGFR was much weaker in the siEGFR@tHFn(+) group than in the other groups. In summary, siTERT/siEGFR@tHFn(+) treatment successfully inhibited tumor growth in vivo by regulating the expression of relative protein. These findings highlighted the potential of siRNA@tHFn(+) in glioma therapy, contributing to its superior therapeutic efficacy and minimal toxicity.

Fig. 8. In vivo antitumor efficacy of siEGFR@tHFn(+) in U87MG-luc-mCherry orthotopic models.

(A) Schematic diagram of glioma modeling and therapy. (B) Changes in body weight of mice in different treatment groups. Data are shown as means ± SEM (n = 6 to 7 biological replicates). (C) In vivo bioluminescence images of tumor-bearing mice treated with PBS, siEGFR, tHFn(+), or siEGFR@tHFn(+). An equivalent dose of siEGFR (1 mg/kg) was applied. (D) Quantitative analysis of tumor growth by bioluminescence signals of (C). Data are shown as means ± SEM (n = 6 to 7 biological replicates). Statistical significance was determined by two-way ANOVA with Tukey’s multiple comparisons test. *P < 0.05. (E) Representative images of H&E staining of brain tissues collected at day 21. (F) Flow cytometry and quantitative analysis of EGFR expression in glioma of Balb/c-nu in different treatment groups. MFI, mean fluorescence intensity. Data are shown as means ± SEM (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparisons test. ****P < 0.0001. (G) Immunofluorescence imaging of EGFR expression in glioma with different treatments. Scale bars, 40 μm.

Biosafety evaluation

To evaluate the biosafety and potential side effects of siRNA@tHFn(+), healthy Balb/c mice were injected intravenously with PBS, tHFn(+), siTERT@tHFn(+), and siEGFR@tHFn(+) at the same therapeutic dosage. As demonstrated in fig. S14, major organs were examined to visualize histological changes. H&E staining confirmed the biosafety of siTERT/EGFR@tHFn(+) with no notable pathological changes in the heart, liver, spleen, lung, kidneys, or brains. In addition, the blood routine test identified that there were no abnormal changes in the mice (table S4). Liver function was assessed by measuring serum levels of alanine transaminase (ALT), aspartate aminotransferase (AST), and alkaline phosphatase (ALP), whereas renal function was evaluated through blood urea nitrogen (BUN) and creatinine (CREA) (fig. S15). There was no notable difference between the treatment and control groups. The above data showed that tHFn(+) functioned as a biocompatible carrier with minimal systemic toxicity.

DISCUSSION

Here, we successfully constructed a promising siRNA carrier [tHFn(+)] based on ferritin, which is favored with lysosome-escapable ability, featuring a positively charged inner cavity and a truncated C terminus. tHFn(+) carriers disassemble and release siRNA molecules in endosomes, which do not contain any digestive enzymes unlike lysosomes, making them a safer environment for siRNA. Furthermore, the structural basis for tHFn(+) nanocarriers to achieve lysosomal escape has been verified through structural determination. The modified ferritin was effectively loaded with siRNA through pH-mediated disassembly/reassembly and acted as a protective shield against enzymatic degradation. We compared the function of siRNA@LNP with siRNA@tHFn(+) at in vitro and in vivo levels. In vitro experiments confirmed that tHFn(+) successfully delivered siRNA to the intracellular compartment and escaped from lysosomes through weak acid–mediated disassembly. In vivo studies revealed that tHFn(+) extended the plasma half-life of siRNA, penetrated the BBB, and accumulated at brain tumor sites. Notably, siTERT/siEGFR@tHFn(+) displayed remarkable knockdown efficiency and therapeutic efficacy against glioma in orthotopic mouse models with low toxicity and minimal impact on major organs. siRNA@LNP demonstrated excellent knockdown abilities in vitro as expected; however, it was unable to restrain the growth of glioma because it cannot traverse the BBB to reach the tumor sites in the brain. In contrast to existing studies, we have rationally designed siRNA nanocarriers that can disassemble in the weak acidic environment of endosomes, which enables earlier escape from lysosomes, contributing to the stability of siRNA and thus enhancing the delivery efficiency.

Among the diverse siRNA carriers, tHFn(+) nanocages boast distinctive advantages. First of all, tHFn(+) nanocarriers, derived from natural sources, exhibit high biocompatibility and biosafety. LNPs and cationic polymers are noted for their high encapsulation efficiency and lysosomal escape capability. Nevertheless, high doses of cationic lipids are prone to causing cellular toxicity (48). Furthermore, it has been demonstrated that tHFn(+) nanocarriers are favored with great tumor targeting ability, which is attributed to their ability to bind to TfR1. TfR1 is overexpressed in plenty of tumor cells, thereby expanding the potential applications of tHFn(+) in cancer treatment beyond GBM (49). Apart from the intrinsic tumor targeting, tHFn(+) nanocarriers can be functionalized with specific targeting ligands for the treatment of other diseases.

As a safe and efficient siRNA delivery vehicle, the tHFn(+) nanocarrier has favorable properties such as excellent biocompatibility, tumor targeting, biosafety, and biodegradability. Despite variations in sequence, siRNA molecules aimed at diverse targets share structural similarities. We have successfully loaded different siRNAs into tHFn(+) nanocarriers, which establishes a solid foundation for efficient siRNA delivery in RNAi-based treatments for various tumors and hereditary diseases, expecting to be translated into clinic applications in the biomedical field.

MATERIALS AND METHODS

Materials

RNase A, DMEM, Opti-MEM, fetal bovine serum (FBS), trypsin, and penicillin-streptomycin were purchased from Thermo Fisher Scientific Inc. (USA). Beetle luciferin was acquired from Promega (USA). Kanamycin sulfate and isopropyl-β-d-1-thiogalactopyranoside were acquired from Inalco (Italy). Agar, prestained marker, 5× SDS-PAGE sample loading buffer, and Albumin Bovine V were purchased from LABLEAD (China). Omni-Easy one-step color PAGE gel rapid preparation kit, protein-free rapid blocking buffer, and universal antibody dilution buffer were provided by Epizyme Biotech (China). Native marker was obtained from Cytiva (USA). Collagenase NB 4 was purchased from SERVA (Germany). DNA I was obtained from Beyotime (China). Other reagents were obtained from Sigma-Aldrich (USA) unless stated otherwise. Cy5-labeled siRNAs were synthesized by Suzhou GenePharma Co. Ltd. (China).

Construction, expression, and purification of recombinant tHFn(+) variants

The plasmids of HFn and tHFn(+) variants were synthesized by GenScript (Nanjing) and then were expressed in E. coli BL21 (TransGen, Beijing). E. coli bacterial suspension was disrupted through a high-pressure homogenizer and centrifuged at 10,000g for 30 min. The supernatant of HFn, HFn(+), tHFn(+)^1-174, and tHFn(+)^1-162 was heated at 72°C for 20 min and purified by AIEX and SEC using a Superdex 200 pg column (GE Healthcare). The supernatant of tHFn(+)^1-159 was heated at 60°C for 10 min followed by HIC and SEC, whereas tHFn(+)^1-156, tHFn(+)^1-153, and tHFn(+)^1-150 were purified by HIC and SEC only.

Characterization of recombinant tHFn(+)

The protein samples were diluted to 0.05 mg/ml, and TEM images were obtained from an FEI Tecnai Spirit electron microscope (100 kV). The particle size of the protein was measured through DLS (Wyatt DynaPro NanoStar) at a concentration of 0.15 mg/ml. SDS-PAGE and Native-PAGE were conducted to evaluate the purity and protein assembly of HFn and tHFn(+) variants, respectively. Zeta potential of the protein samples was measured using phase analysis light scattering (PALS) of Brookhaven NanoBrook 90Plus.

Lysosomal/endosomal colocalization

U87MG cells (1.2 × 10⁵) were seeded on glass-bottom dishes (Nest). Cy5.5-labeled HFn and tHFn(+) variants were added to cells. After incubation for 3 hours, the treated cells were washed three times with PBS. The cells were stained with Lysotracker Green (final concentration: 75 nM) (Beyotime, C1047S) for 1 hour at 37°C and Hoechst 33342 (final concentration: 10 μg/ml) (Sangon Biotech, E607328) for 10 min at room temperature. The confocal images were obtained from Olympus FV1200 (Japan), and PCC analysis was performed using ImageJ.

To further investigate the colocalization of different formulations with lysosomes over time, U87MG cells were treated with Cy5.5-tHFn(+), Cy5-siRNA, or Cy5-siRNA@tHFn(+) at the indicated time. Then, the cells were fixed using 4% paraformaldehyde (PFA) for 10 min and incubated with PBS containing 0.2% Triton X-100 for 10 min at room temperature. To block unspecific binding, the cells were incubated with 5% bovine serum albumin (BSA) for 30 min. For lysosome staining, the cells were incubated with LAMP-1 antibody (Santa Cruz Biotechnology, sc-20011) for 1 hour at 37°C, followed by incubating with Alexa Fluor 568 goat anti-mouse IgG (H+L) for 1 hour at 37°C. Last, the nucleus was stained with 4′,6-diamidino-2-phenylindole (DAPI) (Beyotime, C1006) for 10 min at room temperature. Fluorescence imaging and data processing were performed as described above.

To study the colocalization of these formulations with endosomes, EEA1 antibody (Cell Signaling Technology, 48453S) was used to stain endosomes. To study the colocalization of HFn and tHFn(+) in endothelial cells, protein samples were added to bEnd.3 cells. Rab5 antibody (Abcam, ab218624) and LAMP-1 antibody were used to stain endosomes and lysosomes, respectively. Other steps were identical to the previous ones.

pH-responsive disassembly

HFn and tHFn(+) were incubated at different pH values for 30 min. Subsequently, Native-PAGE, TEM, and zeta potential analysis were applied to investigate the disassembly state of protein. Moreover, these proteins at different pH values were adjusted to pH 8 for reassembly studies, including Native-PAGE and TEM imaging.

Cryo-EM structure determination

Grid preparation and data collection

The cryo-EM grids containing tHFn(+) were prepared using a Thermo Fisher Scientific Vitrobot Mark IV, operating at 4°C and 100% humidity. A volume of 4 μl of the sample was applied to Quantifoil R1.2/1.3 Au 300-mesh grids that had undergone a glow discharge for 30 s. The grids were blotted for 2 s with a blot force of 4 using Whatman No.1 filter paper, followed by immediate plunge freezing in liquid ethane and subsequent transfer to sealed boxes for storage and data collection.

All cryo-EM data were collected in super-resolution mode using the Thermo Fisher Scientific Titan Krios G2, equipped with a Bioquantum GIF K3 camera operating at 300 kV. Micrographs were acquired at a magnification of ×165,000, corresponding to a pixel size of 0.44 Å and a defocus range from −0.3 to −0.7 μm. Each stack, with a total dose of 60 e^-/Ų, was exposed for 0.96 s with an exposure time of 0.03 s per frame, resulting in 32 frames per stack. The EPU software was used for the automated data collection.

Imaging processing and structure refinement

To achieve 3D reconstruction of tHFn(+), a total of 5362 micrograph stacks were recorded. All data processing was performed using the CryoSPARC software. The raw stack frames were subjected to motion correction using the Patch Motion Correction module. The CTF parameters of each summed image were estimated via the Patch CTF Estimation module. Particle selection was conducted using the Template Picker module. Ultimately, a total of 1,725,802 particles were selected and extracted, which were then subjected to multiple rounds of 2D classification. Of these, 966,873 particles were retained for the initial model generation and subsequent 3D classification using ab initio reconstruction techniques. Subsequently, a subset of 382,703 particles was chosen for homogeneous refinement using O symmetry and CTF refinement procedures, resulting in a reconstruction with an overall resolution of 1.73 Å based on the Fourier shell correlation criterion set at 0.143.

Preparation and characterization of siRNA@tHFn(+)

First, siRNA molecules were synthesized using the T7 RNAi Transcription Kit (Vazyme Biotech, Nanjing). Sequences of siRNA were listed in table S5. Subsequently, siRNA@tHFn(+) was prepared following a previously reported procedure (27). Briefly, the pH of tHFn(+) (10 mg/ml) was adjusted to 2 using HCl and incubated for 20 min at 4°C, and then carbonate buffer (pH 9.4) containing siRNA (ready to use) (40 μM) was added to increase the pH to 8 and incubated for 2 h at 4°C. Free siRNA molecules were removed using ultrafiltration tubes with a molecular weight cutoff of 100 kDa. Protein concentration was measured with BCA Protein Assay Kits (Thermo Fisher Scientific), whereas RNA concentration was quantified using the Qubit 4 Fluorometer (Thermo Fisher Scientific). DLS was conducted to measure the particle size, and the morphology of protein was visualized on TEM. To confirm the successful loading of siRNA within tHFn(+) variants, AGE and PALS were carried out. To assess the stability of siRNA@tHFn(+), RNase A was added at a final concentration of 20 μg/ml and incubated at 37°C for 30 min. The samples were mixed with 6×gel loading dye (NEB, B2075S) and loaded onto a 2% agarose gel. AGE was conducted at 120 V for 25 min and observed under ultraviolet light using Tanon 3500R. To evaluate the serum stability of siRNA@tHFn(+), samples were incubated in DMEM containing 10% FBS for 1, 4, and 24 hours.

Preparation of siRNA@LNP

The synthesis method of the LNP was derived from the previous literature (43). Briefly, The pre-LNP was prepared by mixing the key lipid, 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), cholesterol (Cho), and 1,2-dimyristoyl-rac-glycero-3-methoxypolyethylene glycol-2000 (DMG-PEG₂₀₀₀), which were provided by Huang’s group. The organic mixture in ethanol was mixed with citrate buffer (50 mM, pH 4.0) in a ratio of 1:3. Next, siRNA molecules containing 25% ethanol were added to the solution and incubated for 20 min at 50°C to form siRNA@LNP. Last, the mixture was dialyzed in PBS at 4°C overnight.

Cell culture

The U87MG cells and U87MG-luc-mCherry cells were cultured in DMEM supplemented with 10% FBS, whereas the bEnd.3 cells were cultured in DMEM containing 20% FBS.

Cellular uptake

U87MG cells were plated on glass-bottom dishes at a density of 1.2 × 10⁵ cells. Free Cy5-siRNA and Cy5-siRNA@tHFn(+) were added to the cells for 10 min, 2 hours, and 4 hours. The cells were then stained with DAPI for 10 min at room temperature. Cellular uptake was visualized on Olympus FV1200 (Japan). To measure the uptake of siRNA, flow cytometry was applied. U87MG cells were seeded in 6-well plates (Corning) and treated with free Cy5-siRNA or Cy5-siRNA@tHFn(+) for 10 min, 2 hours, and 4 hours. Then, the cells were harvested and flow cytometry (BD, FACSCalibur) was performed using Cy5 fluorescence. Data analysis was conducted using FlowJo 10.6.

In vitro gene knockdown

U87MG cells were seeded into 6-well plates at a density of 2 × 10⁵ cells per well. Once the cells reached 50% confluence, the indicated formulations were added into the wells. After 48-hour incubation, cells were washed with PBS and harvested using radioimmunoprecipitation assay (RIPA) lysis buffer containing protease inhibitor phenylmethylsulfonyl fluoride (PMSF) (Elabscience). The cells were then centrifuged for 10 min at 12,000g at 4°C. The supernatant was collected, and the protein concentrations were determined using BCA Protein Assay Kits. Protein samples of 20 μg were loaded onto a 10% SDS-PAGE gel and transferred to a polyvinylidene fluoride membrane. Subsequently, the membrane was incubated in blocking buffer for 15 min and then primary antibodies (EGFR: Abcam, ab32562, 1:1000; TERT: Abcam, ab32020, 1:1000; β-tubulin: ABclonal, AC021, 1:5000) were added and incubated overnight at 4°C. Secondary antibodies (rabbit: ABclonal, AS014; mouse: Abclonal, AS003) were diluted at 1:5000. The membrane was visualized using the Femto-Sig ECL Western Blotting Substrate (Tanon, Shanghai) and performed on Tanon 4600SF.

RT-qPCR was also performed to evaluate the knockdown efficiency in vitro. U87MG cells were plated in 6-well plates with 2.5 × 10⁵ cells per well. On the following day, the cells were incubated with different formulations for 24 hours. Total RNA was extracted from U87MG cells using the TRIzol reagent (Thermo Fisher Scientific), followed by reverse transcription using the All-In-One 5× RT MasterMix (Applied Biological Materials Inc.). RT-qPCR assays were conducted using the Taq Pro Universal SYBR qPCR Master Mix (Vazyme, Nanjing, China) and the ΔΔCT method was applied. β-Tubulin was used as the control gene, and untreated cells (blank) were used as the control sample. The primer sequences are listed in table S6.

Receptor binding ability

ELISA was applied to assess the binding ability of HFn and tHFn(+) to HFn receptor. First, the protein samples were diluted in a gradient with coating solution (carbonate buffer, pH 9.4) and added to the ELISA plates (LABLEAD) according to the experimental design. The plates were then incubated for 2 hours at 37°C. Afterward, the plates were incubated with blocking buffer (0.5% BSA) for 1 hour at 37°C. The plates underwent sequential incubation with TfR1, anti-TfR1 (Sino Biological, 1:1000), and anti-mouse IgG (H+L) (1:5000) for 1 hour at 37°C. Last, TMB substrate solution (Solarbio) was added to the plates, which was read kinetically at OD (optical density) 652 nm using a microplate reader (Molecular Devices, SpectraMax Plus384) until the desired color intensity was reached.

In vitro BBB models

Mouse endothelial cells (bEnd.3) were cultured at a density of 2 × 10⁴ cells per well on a Transwell plate (Corning, 3413) with a pore diameter of 0.4 μm. The transepithelial electrical resistance value was measured every other day until it reached 200 ohm·cm², indicating the formation of a simulated BBB. Cy5.5-labeled LFn, HFn, and tHFn(+) and Cy5-labeled siRNA@tHFn(+) and siRNA@LNP were added to the upper chamber, and 200 μl of culture media of the lower chamber was collected after 2 hours. Fluorescence concentrations were detected on the Enspire Multimode Plate Reader (PerkinElmer). The transcytosis efficiency of different formulations was calculated as follows: transcytosis efficiency (%) = (C_lower·V_lower)/(C_upper·V_upper) × 100%. To evaluate protein integrity after traversing the BBB, HFn, tHFn(+), and siRNA@tHFn(+) were added to the upper chamber, and the culture media of the lower chamber was collected after 2 hours, which were then subjected to Native-PAGE analysis and concentration determination of siRNA@tHFn(+).

Cell viability

The cell viability of HFn and tHFn(+) on bEnd.3 cells was determined using a Cell Counting Kit-8 (CCK-8, Dojindo). First, bEnd.3 cells were seeded into a 96-well plate at a density of 1.5 × 10⁴ cells per well. The cells were treated with HFn and tHFn(+) at the indicated concentration for 24 hours. After that, the reagents were replaced with a medium containing CCK-8. Last, the absorbance was measured at 450 nm using a microplate reader.

Animals

Balb/c-nu mice and Balb/c mice were obtained from SpePharm (Beijing) Biotechnology Co. Ltd. All animal experiments were approved by the Institutional Animal Care and Use Committee of the Institute of Biophysics, Chinese Academy of Sciences (SYXK2023155). All mice were housed in a pathogen-free environment on a 12-hour/12-hour light/dark cycle. Room temperature was maintained at 22°~26°C with a humidity of 40~70%.

The sample size of animal studies was determined based on the experimental requirement and published studies. We used n = 3 mice per group for in vivo distribution studies, n = 6 to 7 mice per group for therapeutic efficacy studies, n = 3 mice per group for in vivo gene knockdown evaluation, and n = 3 mice per group for biosafety analysis. All animal studies were conducted with randomization and blinding experimental procedures.

In vivo biodistribution and pharmacokinetics

U87MG-bearing mice were intravenously administered with 100 μl of Cy5-siRNA, Cy5-siRNA@LNP, and Cy5-siRNA@tHFn(+) at a concentration of 15 μM Cy5 equivalents. In vivo imaging was conducted, and the brain, heart, liver, spleen, lung, and kidneys were collected at the indicated time for ex vivo imaging using IVIS Lumina3 (PerkinElmer). The fluorescence intensity analysis of Cy5 was performed using Living Image.

To evaluate the in vivo pharmacokinetics, 100 μl of Cy5-siRNA, Cy5.5-tHFn(+), and Cy5-siRNA@tHFn(+) were injected into the healthy Balb/c mice at an equivalent fluorescence doses of 15 μM. The plasma samples were collected at various time points (5, 10, 20, 30, 60, 90, 120, 240, 480, 720, and 1440 min), and the drug concentrations were then determined by the Enspire Multimode Plate Reader (PerkinElmer).

In vivo antitumor efficacy

Male Balb/c-nu mice (6 weeks old) were used to establish animal models of U87MG-luc-mCherry GBM, following the previous literature (23). Briefly, tumor cells were intracranially injected into the right hemisphere of the mouse brain via a stereotaxic apparatus on day 0. For TERT-based treatment, mice were weighed and randomly divided into six groups (n = 6 to 7) and injected with PBS, tHFn(+), siTERT, siNC@tHFn(+), siTERT@LNP, and siTERT@tHFn(+). Mice were treated with various formulations at an siRNA dose of 1 mg/kg every other day for 2 weeks. Glioma volume was monitored using IVIS Lumina3 (PerkinElmer) to detect luciferin bioluminescence signals catalyzed by luciferase in tumor cells. Bioluminescence images were taken every 3 days and analyzed through Living Image. At the end of the treatment period, brain tissue was collected for paraffin sectioning and H&E staining (Wuhan Baiqiandu Technology Co. Ltd., China). For EGFR-based treatment, mice were intravenously injected with PBS, siEGFR, tHFn(+), and siEGFR@tHFn(+) at an equivalent siRNA dose of 1 mg/kg. The body weight and tumor volume were recorded according to the experimental design. On day 21, mice were euthanized, and their brains were harvested for paraffin sectioning and H&E staining (Wuhan Baiqiandu Technology, China).

In vivo gene knockdown

Various techniques have been used to evaluate gene knockdown levels in vivo including WB, flow cytometry, and immunofluorescence. To perform WB analysis, tumor issue was collected from the brain, cut into pieces, and mixed with ceramic beads (Welab Science and Technology, Beijing). RIPA lysis buffer with PMSF was added, and the mixture was homogenized with a tissue homogenizer (Bertin Technologies, Precellys 24). The protein concentrations of tissue samples were determined using BCA kits. The remaining steps were identical to those used in previous in vitro WB experiments.

For flow cytometry, glioma tissue was isolated from the brain and cut into pieces. The tissue was then digested using NB4 tissue digestion solution containing DNase I and incubated for 1 hour at 37°C and 100 rpm. Next, DMEM was added to terminate digestion and the resulting suspension was filtered through 70-μm cell strainers (Corning, 352350) to obtain a single-cell suspension. The cells were washed with PBS containing 2% BSA and stained with Alexa Fluor 647 Anti-EGFR antibody (Abcam, ab30403, 1:500) for 30 min on ice. Last, the cells were washed and analyzed using a flow cytometer (BD, FACSCalibur). Channel 2 was used to detect mCherry, which was used to select glioma cells and set gates accordingly. EGFR signals were detected on channel 4 to evaluate protein levels. Fluorescence analysis was performed using FlowJo 10.6.

For immunofluorescence experiment, the paraffin slides were dewaxed and hydrated. Next, the slides were placed in boiling antigen retrieval buffer (citric acid, pH 6.0) for 3 min. After natural cooling, the slides were washed with PBS. For TERT staining, PBS containing 0.5% Triton X-100 was added for permeabilization. The slides were blocked with goat serum at 37°C for 1 hour. Subsequently, primary antibodies (EGFR: Abcam, ab32077, 1:500; TERT: Thermo Fisher Scientific, MA5-16034, 1:100) were incubated with the slides at 4°C overnight, followed by secondary antibodies (rabbit: Abcam, ab150079, 1:1000; mouse: Abcam, ab175473, 1:1000) for 60 min at room temperature. Then, the slides were stained with DAPI for 10 min at room temperature and sealed with antifade mounting medium (Beyotime). Last, the slides were observed using CLSM (Olympus FV1200).

Biological safety

PBS, tHFn(+), siEGFR@tHFn(+), and siTERT@tHFn(+) were intravenously injected into healthy Balb/c mice every 2 days for 2 weeks. Mice were euthanized on day 7 after the last injection. To evaluate the in vivo safety of nanocarriers, hematology analysis and pathological detection were used. Blood was collected for routine testing, and the serum was separated for biochemical analysis (Wuhan Baiqiandu Technology, China), including ALT, AST, ALP, BUN, and CREA. In addition, the heart, liver, spleen, lung, kidneys, and brains were collected and fixed in 4% PFA. The tissue samples were embedded in paraffin and cut into 5-μm slices for H&E staining (Wuhan Baiqiandu Technology). The tissue slides were visualized using Aperio CS2 (Leica Biosystems).

Statistical analysis

All statistical analysis was performed using GraphPad Prism 9.0. Data are presented as means ± SEM. The statistical significance between two groups and among multiple groups was analyzed by two-tailed unpaired Student’s t test and one-way/two-way analysis of variance (ANOVA) with Tukey’s multiple comparisons test, respectively. Statistic differences were labeled as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. No samples were excluded from the analysis.

Data deposit

The atomic coordinates and EM maps have been deposited in the PDB (http://rcsb.org) under accession number 9KAY and in the EMDB (Electron Microscopy Data Bank) under accession number EMD-62216.

Supplementary Materials

This PDF file includes:

Figs. S1 to S15

Tables S1 to S6