Laboratory evolution of virus-like nucleocapsids from nonviral protein cages

Significance

Viruses consist of a protective proteinaceous shell that packages a DNA or RNA genome. The critical step in the evolution of primitive viruses was presumably the emergence of protein cages that could load, protect, and transfer their own genetic information. Here, we show that a nonviral protein cage formed by Aquifex aeolicus lumazine synthase and its encoding mRNA can be engineered and evolved into virus-like nucleocapsids. The optimized proteins specifically recognize designed motifs on cognate mRNAs, and these RNA tags can also be employed for the encapsulation of other RNA molecules. These model systems have great potential for the investigation of nucleocapsid assembly and evolution and as tailored protein compartments for protective packaging and delivery of nucleic acids.

Abstract

Viruses are remarkable nanomachines that efficiently hijack cellular functions to replicate and self-assemble their components within a complex biological environment. As all steps of the viral life cycle depend on formation of a protective proteinaceous shell that packages the DNA or RNA genome, bottom-up construction of virus-like nucleocapsids from nonviral materials could provide valuable insights into virion assembly and evolution. Such constructs could also serve as safe alternatives to natural viruses for diverse nano- and biotechnological applications. Here we show that artificial virus-like nucleocapsids can be generated—rapidly and surprisingly easily—by engineering and laboratory evolution of a nonviral protein cage formed by Aquifex aeolicus lumazine synthase (AaLS) and its encoding mRNA. Cationic peptides were appended to the engineered capsid proteins to enable specific recognition of packaging signals on cognate mRNAs, and subsequent evolutionary optimization afforded nucleocapsids with expanded spherical structures that encapsulate their own full-length RNA genome in vivo and protect the cargo molecules from nucleases. These findings provide strong experimental support for the hypothesis that subcellular protein-bounded compartments may have facilitated the emergence of ancient viruses.

Viral particles, also called virions, consist of a protective proteinaceous shell that packages a DNA or RNA genome (1). As all aspects of the viral life cycle, including genome storage, replication, and infection, depend on virion formation, the emergence of protein cages capable of encapsulating genomic information was presumably the critical step in the evolution of primitive viruses (2). Although the precise evolutionary origins of viruses are still debated, sequence and structural analyses suggest that the enormous number of structurally diverse capsid proteins found in today’s virosphere evolved from ancestral cellular proteins on multiple, independent occasions (2–4). Therefore, bottom-up construction of virus-like structures from nonviral components could shed light on how their natural counterparts arose.

Aquifex aeolicus lumazine synthase (AaLS) is an enzyme that shares no homology with natural viruses but spontaneously self-assembles into capsid-like 60-subunit protein shells (5). AaLS cages have been engineered to package various guest molecules, including nucleic acids, via complementary electrostatic interactions (6–8). Although positively supercharged variants have been shown to capture short, endogenous RNA molecules in vivo (7, 8), sequence selectivity was not observed. Moreover, the interior volume of the parent AaLS capsid (270 nm³) is too small to encapsulate the full-length mRNA encoding its own 154-aa shell protein. For comparison, the icosahedral capsid of satellite tobacco mosaic virus (STMV), one of the smallest natural ssRNA viruses, has an inner volume of ∼1,050 nm³ (9).

Here we show that engineering and coevolution of AaLS and its encoding mRNA can be employed to produce virus-like nucleocapsids in the laboratory. The resulting particles adopt expanded spherical structures that selectively encapsulate their own full-length RNA genome in vivo and protect guest molecules from nucleases. Similar to natural viruses, terminal peptides of the engineered capsid proteins specifically recognize designed packaging signals on cognate mRNAs. As such, these constructs have great potential for investigating evolutionary assembly pathways and as evolvable protein compartments for therapeutic nucleic acid delivery and vaccine applications (10–14).

Results

Engineering of AaLS Protein and Its mRNA.

Several ssRNA viruses employ genome packaging strategies that rely on direct interactions between arginine-rich motifs (ARMs) on the viral capsid and specific RNA structures in their genome (15–17). To equip AaLS capsids with ARMs, we relocated the N and C termini of the individual subunits from the exterior to the interior of the cage, between residues 84 and 85, by circular permutation (SI Appendix, Fig. S1). Unlike another recently reported AaLS circular permutant with a different ring-opening site (18), the resulting protein, cpAaLS(84), forms homogeneous spherical structures similar in size to WT AaLS assemblies (SI Appendix, Fig. S2A). To enable RNA binding, the cationic λN+ peptide, engineered from the λ-phage antitermination complex, was appended to the new N terminus to give λcpAaLS (Fig. 1A) (18, 19). The 5′- and 3′-UTRs of the corresponding gene were equipped with artificial BoxBr packaging signals, consisting of a BoxB stem-loop structure that binds λN+ peptides (19), plus a fluorogenic Broccoli aptamer (20) for detection purposes (SI Appendix, Fig. S3).

Fig. 1.

Strategy to generate virus-like nucleocapsids from nonviral protein shells. (A) Design of a circularly permutated AaLS variant fused to a λN+ peptide (λcpAaLS). The original N and C termini were connected by a linker containing a hexahistidine sequence (green dashed line), a new cleavage site was introduced between residues 84 and 85, and the λN+ peptide (cyan) was appended to the newly created N terminus via an 8-aa linker (cyan dashed line). (B) λcpAaLS forms capsids that are similar in size to tetrahedrally symmetric AaLS-neg cages, which self-assemble from 180 subunits organized as 36 pentameric capsomers (21). Large keyhole-shaped pores in the shell wall enable facile cargo transport. (C) Directed evolution strategy. In the first evolutionary round, BoxBr tags that bind to λN+ peptides were introduced in the 5′- and 3′-UTRs of HIV protease mRNA. Capsid variants were initially selected from large gene libraries for their ability to inhibit production of the toxic protease by sequestration of its BoxBr-tagged mRNA (Upper). For further optimization (Lower), the BoxBr tags were transferred to the UTRs of mRNA encoding the selected λcpAaLS variants, and the encoded cage proteins were produced in E. coli cells. After capsid purification and nuclease treatment, encapsulated mRNAs were extracted and converted to cDNA by RT-PCR. The best variant, λcpAaLS-β16, was further diversified by epPCR, and the resulting library was subjected to the procedures shown (Lower). Protein Data Bank ID codes: AaLS, 1HQK; λN+, 1QFQ; AaLS-neg, 5MQ3.

Introduction of the λN+ peptides induced expansion of the 16-nm WT cages to give spherical structures with diameters in the range of 27–29 nm (Fig. 2A). These positively supercharged structures are comparable in size to AaLS-neg, a negatively supercharged AaLS variant that forms unusual 180-subunit cages with tetrahedral symmetry and large keyhole-shaped pores in the shell wall (Fig. 1B) (21). The lumenal volume of AaLS-neg (4,200 nm³) is nearly four times larger than that of STMV, which should provide ample space to harbor the shorter RNA genome of λcpAaLS (863 nt vs. 1,058 nt).

Fig. 2.

Characterization of evolved λcpAaLS variants. (A) Backbone traces of capsid subunits (Upper) and negatively stained EM images of assembled cages (Lower) are shown for λcpAaLS, λcpAaLS-β16, λcpAaLS-α2, and λcpAaLS-α9. Residues mutated in λcpAaLS-β16 are shown as gray spheres (I78F, V80D, E95V, I111V, G114D, and V115I); additional mutations in λcpAaLS-α2 (E2G, A25T, P54T, L81P, and E145G) and λcpAaLS-α9 (M1I, A25T, G64D, I92N, and F150Y) are shown as magenta spheres. Residues are numbered as for AaLS-WT. (Scale bars: 50 nm.) Black arrows indicate the line of evolutionary descent. (B) The amounts of total RNA packaged in λcpAaLS capsids expressed from mRNAs possessing two BoxBr tags (white bars) were determined from the 260/280-nm absorbance ratio, whereas full-length capsid-encoding mRNA (black bars) was determined by quantitative RT-PCR. The data are reported as the mean ± SD from three independent experiments. ND, not detectable. (C and D) RNA guests in the λcpAaLS capsids were analyzed by 8% (C) or 6% (D) denaturing PAGE. “mRNA” indicates in vitro-transcribed full-length mRNA (863 nt).

Substantial amounts of host RNA associated with the isolated λcpAaLS particles, judging by the A260/A280 ratio and the susceptibility of the cargo to RNase A digestion (Fig. 2B and SI Appendix, Figs. S2 E and I and S4A). The purified nucleocapsids were treated with Benzonase (22), a promiscuous 60-kDa endonuclease that is too large to enter the cage, to destroy incompletely encapsulated DNA and RNA as well as any nucleic acids simply bound to the shell exterior (SI Appendix, Fig. S4A). Subsequent extraction of the remaining nucleic acids from the particles and analysis by denaturing PAGE confirmed encapsulation of RNA fragments smaller than 200 nt (Fig. 2C). Even though it was highly overexpressed from a T7 promoter, full-length mRNA encoding λcpAaLS was not detected by RT-PCR (Fig. 2B and SI Appendix, Fig. S5A). Simply introducing viral packaging signals is apparently insufficient to afford virus-like nucleocapsids that protect encapsulated nucleic acid molecules.

Directed Evolution of λcpAaLS Nucleocapsid.

To improve the RNA packaging capabilities of the λcpAaLS capsids, we turned to directed evolution by using a two-step approach (Fig. 1C and SI Appendix, Fig. S6). First, we envisaged that encapsulation of the mRNA for HIV protease, a toxic enzyme when produced cytoplasmically, would provide a growth advantage to its bacterial host by inhibiting translation of this enzyme (6). Indeed, moderate induction of a protease gene equipped with flanking BoxBr packaging signals with 10 ng/mL of tetracycline greatly impaired cell viability. In contrast, when this construct was coexpressed with λcpAaLS, cell growth was partially restored (SI Appendix, Fig. S5B), consistent with successful sequestration of the toxic gene. In contrast, restoration was not observed when the HIV protease mRNA did not possess packaging signals (SI Appendix, Fig. S5C). A λcpAaLS library was subsequently generated by error-prone PCR (epPCR) using a gene lacking the packaging signals as a template, and variants were selected for their ability to inhibit protease production upon induction with 10-fold higher tetracycline concentrations (100 ng/mL).

In the second step, capsid genes from cells that survived these more stringent selection conditions were amplified by PCR, provided with flanking BoxBr tags, and expressed in Escherichia coli (Fig. 1C, Lower). The resulting protein cages were isolated, purified, and treated with Benzonase. RT-PCR of RNA extracted from the particles yielded the full-length λcpAaLS gene, confirming that some variants in the population successfully packaged their own mRNA (SI Appendix, Fig. S5A). To enrich this population, the entire procedure was repeated twice before producing and characterizing 18 representative clones. One of these cages, λcpAaLS-β16, was found to contain some full-length capsid mRNA (Fig. 2D and SI Appendix, Fig. S5D). Nevertheless, most of the packaged nucleic acid consisted of mRNA fragments, including many that contained the Broccoli aptamer, which could be selectively visualized by using the DFHBI-1T fluorophore (20) (SI Appendix, Fig. S4 B and C), suggesting that the guests are degraded before packaging or that the cage does not completely shield its cargo from small cellular RNases. For this reason, we further diversified λcpAaLS-β16 by epPCR and performed three more cycles of nucleocapsid screening, this time in the presence of RNase A (14 kDa) instead of Benzonase (Fig. 1C). Under these more demanding conditions, the size of the encapsulated RNA steadily increased over each cycle, indicating improved protection of the packaged guests (SI Appendix, Fig. S7A). Characterization of representative purified clones from the final round afforded two variants, λcpAaLS-α2 and λcpAaLS-α9, which contained the full-length mRNA and safeguarded it against digestion by Benzonase and RNase A (Fig. 2 C and D and SI Appendix, Figs. S4 and S7B) (2–4).

Properties of Evolved λcpAaLS Nucleocapsid Variants.

Like the original λcpAaLS variant, λcpAaLS-β16, λcpAaLS-α2, and λcpAaLS-α9 form spherical assemblies with average diameters in the range of 28–31 nm (Fig. 2A). Detailed structures are not yet available, but these diameters also fall in the range previously seen for AaLS-neg (∼28 nm; Fig. 1B) (6, 21). Consequently, the ability of the evolved λcpAaLS variants to package and protect their cognate mRNA cannot be ascribed to significant changes in container volume. Instead, the accumulated mutations presumably induced structural changes that restrict nuclease access to the capsid lumen, perhaps by reorienting the λN+ peptides to close off the pores in the shell wall or through more dramatic changes in assembly form.

The λcpAaLS-β16 variant has six amino acid changes compared with λcpAaLS, whereas its α2 and α9 descendants each have an additional five substitutions. These mutations are distributed over the entire protein scaffold (Fig. 2A and SI Appendix, Fig. S8A), and may alter subunit structure or affect higher-order monomer–monomer and pentamer–pentamer interactions in the quaternary assemblies. Unlike the negatively supercharged AaLS variants that were previously evolved to sequester positively charged proteins, however, these changes did not increase the net positive charge on the cage. In fact, the introduction of three aspartates had the opposite effect, potentially optimizing specific interactions of the capsid with the positively charged λN+ peptide and/or the overall electrostatics of the system.

Mutagenesis experiments showed that the λN+ peptides are crucial for RNA recognition by the λcpAaLS variants. Deletion of these segments abrogated RNA binding (SI Appendix, Fig. S2). Although all of the nucleocapsids have the same λN+ peptide in common, the evolved variants encapsulate two to four times more total RNA than the starting circular permutant and the RNA sequences are longer (Fig. 2 B–D). Based on quantitative RT-PCR, the fraction of particles encapsulating full-length capsid mRNA increased from undetectable for λcpAaLS to ∼2% for λcpAaLS-β16 and ∼10% for λcpAaLS-α2 and λcpAaLS-α9 (Fig. 2B). The most evolved variants exhibit packaging efficiencies comparable to those of recombinant natural viruses (23).

Because RNA–capsid protein interactions are known to enhance the assembly of natural virions (24), we investigated whether coevolution of mRNA and protein sequences, as opposed to simple gene sequestration, might have contributed to the improved packaging efficiency. In support of this hypothesis, reversion of the four silent mutations present in λcpAaLS-β16 mRNA abolished encapsulation of the full-length gene without affecting capsid structure (SI Appendix, Figs. S8B, S9A, S10A, and S11A). To evaluate structural changes caused by silent mutations, we calculated maximum ladder distances (MLDs) for the respective mRNAs (25). MLD is a measure of the length of the longest direct path across an RNA secondary structure, and hence an estimate of compactness. This analysis suggests that the silent mutations alter mRNA compactness substantially (SI Appendix, Fig. S12). In contrast, reverting all of the silent mutations in λcpAaLS-α2 and λcpAaLS-α9, or just the four or five introduced in the last round of mutagenesis, had no effect on capsid structure or the distribution of packaged RNAs (Fig. 3 A and B and SI Appendix, Figs. S9 C and D, S10 C and D, and S11 B and C). For these variants, silent mutations may be less important for RNA–capsid interactions than those that altered amino acid sequence.

Fig. 3.

Analysis of RNAs packaged by λcpAaLS-α2 and λcpAaLS-α9 capsids. (A and B) Denaturing PAGE analysis of RNA extracted from λcpAaLS-α2 (A) and λcpAaLS-α9 (B) cages. mRNA indicates in vitro-transcribed mRNA possessing two BoxBr tags. dB, capsids expressed from mRNAs with two BoxBr tags; m3′B, mutations in the 3′-BoxBr tag; m5′B, mutations in the 5′-BoxBr tag; mdB, mutations in the both 3′/5′-BoxBr tags; S1, reversions of silent mutations that were introduced in the first round of evolution; S2 and S2*, reversions of silent mutations that were introduced into λcpAaLS-α2 and λcpAaLS-α9 in the second round of evolution, respectively. Mutation sites are reported in SI Appendix, Figs. S3, S8, and S9. (C) Bar graph showing relative RNA populations packaged in the λcpAaLS-α2 and λcpAaLS-α9 nucleocapsids. RNA was extracted from particles expressed from mRNA possessing (+) or lacking (−) 5′- and 3′-BoxBr tags. Miscellaneous RNAs from the E. coli host genome are shown in gray. Further details are provided in SI Appendix, Figs. S13–S15.

Analysis of Encapsulated RNA Cargo.

In addition to capsid-encoding mRNA, the evolved nucleocapsids harbor considerable amounts of cellular RNA from the host (Fig. 3 A and B). Natural viruses also encapsulate host RNAs, especially in the absence of specific packaging signals in their genome (26). For instance, ∼50% of the RNA packaged by MLV derives from the host cell, a fraction that increased to 99% when packaging signals were deleted from the viral genome (26). Flock house virus (FHV) is another relevant example. Although host RNA accounts for only 1% of total encapsulated RNA for natural FHV, virus-like FHV particles expressed from baculoviral expression vectors package the FHV genome (7.5%), baculoviral RNA (13.6%), and host RNA (78.9%) (27).

We observed similar trends for the optimized λcpAaLS capsids. Long-read single-molecule sequencing of extracted RNA revealed that 46% of the RNAs packaged in λcpAaLS-α2, and 86% in the case of λcpAaLS-α9, are derived from capsomer-encoding mRNA (Fig. 3C). The next most abundant cargo molecules are derived from plasmid-encoded RNAs (nahR, RNA I and II, ampR, and lacI), followed by housekeeping noncoding RNAs from the Escherichia coli genome (tmRNA, M1 RNA, 16S rRNA) and miscellaneous other RNAs (SI Appendix, Fig. S13). Most of the noncapsomer RNAs are short fragments with an average length in the range of 200–300 nt (SI Appendix, Figs. S14 and S15). Although denaturing PAGE shows that the amounts of 16S rRNA and λcpAaLS mRNA produced in the cell are comparable under our experimental conditions (SI Appendix, Fig. S16), far less rRNA becomes encapsulated, presumably as a result of its sequestration in ribosomes.

Deletion or mutation of the recognition sequence for the λN+ peptides in the BoxBr tags decreased loading of full-length capsid mRNA while increasing the total amount of noncognate RNA that was packaged (Fig. 3 and SI Appendix, Figs. S9 B–D and S10 B–D). These effects are particularly pronounced for λcpAaLS-α9, which is inherently more selective for its own mRNA than λcpAaLS-α2. Interestingly, although short fragments again dominate the population of encapsulated RNAs, small amounts of full-length 23S rRNA (2,904 nt) were detected in this sample (SI Appendix, Fig. S15I), suggesting that these capsids could conceivably be further evolved to package considerably longer RNA sequences. Furthermore, because the BoxBr tags function as packaging signals but are not essential for λcpAaLS nucleocapsid formation, they can be used to direct preferential packaging of other genes such as the BoxBr-tagged mRNA coding for HIV protease (SI Appendix, Figs. S9E and S10E).

Discussion

The evolutionary origins of viruses have been widely discussed, and the emergence of nucleocapsids able to package their own genomic information was presumably a critical step in the development of functional viral particles (2–4). Efficient virion assembly depends on many factors, including complementary protein–protein and protein–genome interactions, genomic structure, coordination of replication and translation, and subcellular localization of viral components in the host cell (1, 28–31). Natural viruses have optimized each of these features over eons of evolution. The demonstration that nonviral proteins and their encoding RNA can be easily and rapidly evolved into virion-like particles is surprising and supports the “escape” hypothesis regarding virus origins (2–4).

Inspired by natural ssRNA viruses (1, 15–17), we exploited engineered interactions between cationic peptides and RNA packaging signals to create AaLS variants capable of capturing their own mRNA. Although such interactions did not ensure efficient packaging of the full-length capsid genome on their own, several rounds of mutagenesis and screening afforded nucleocapsids with substantially higher loading capacity, greater specificity, and enhanced cargo protection. Consistent with suggestions that RNA architecture plays important roles in controlling the assembly of natural viruses (24, 25, 32), protein–protein and protein–RNA interactions were simultaneously optimized by coevolution of the circularly permuted capsid subunit with its encoding RNA. The resulting constructs are reminiscent of plant satellite viruses like satellite tobacco necrosis virus, which also have RNA genomes that encode only capsid protein and depend on cognate helper viruses to complement other aspects of the viral life cycle (33).

By using similar evolutionary approaches, Butterfield et al. (12) recently optimized computationally designed two-component icosahedral protein assemblies (I53-50) that also package their own full-length genomes. In contrast to the λcpAaLS nucleocapsids, however, the I53-50 constructs rely on nonspecific electrostatic interactions rather than designed protein–RNA recognition motifs for nucleic acid encapsulation and were able to bind small amounts of full-length capsid mRNA even before directed evolution. During optimization of I53-50, the positions of charged residues on the capsid surface were reconfigured, particularly around the pores in the protein shell, potentially restricting nuclease access to the lumen. Understanding the differences between I53-50 and the λcpAaLS cages, which likely have structural origins, will require future investigation. Nevertheless, like λcpAaLS, the optimized two-component capsids package approximately one full-length mRNA for every 10 capsids, with loading efficiency correlating with mRNA overexpression levels.

The high selectivity of our λcpAaLS-α9 capsids for target RNA over cellular competitors is particularly notable (Fig. 3). It derives from specific interactions between lumenal RNA binding peptides on the cage and specific RNA packaging signals present in the cargo. As such, this strategy could be used to package any RNA molecule simply by equipping the target with the appropriate packaging tag, as shown by successful encapsulation of the HIV protease gene. Nevertheless, additional rounds of evolution will clearly be needed for both families of artificial nucleocapsids to achieve the packaging specificity and efficiency of natural viruses (27). In addition to optimizing capsid–RNA interactions for specific cargo molecules of interest, harnessing additional cellular factors to promote target recognition and minimize RNA degradation before nucleocapsid assembly may prove necessary.

AaLS cages have been widely engineered by circular permutation (18), modular assembly (8), and surface modification (34). Integration of this programmability with the “evolvability” of AaLS-based nucleocapsids may allow them to be endowed with additional useful capabilities, such as cell recognition, infectability, uncoating, and self-replication. The resulting assemblies may serve as practical alternatives to natural viruses for diverse applications in biotechnology and synthetic biology.

Materials and Methods

Molecular Cloning.

To express AaLS cages from the T7 promoter, a pMG plasmid was used; a pAC-Ptet plasmid was employed for the expression of HIV-protease under control of the tetracycline promoter (6). Detailed procedures are described in SI Appendix, Supplementary Methods.

Expression and Purification of Capsids.

λcpAaLS capsids were produced in E. coli BL21(DE3)-gold transformed with a pMG or pMG-dB vector encoding λcpAaLS genes. The cells were grown at 37 °C in Luria Bertani (LB) medium until the OD600 reached 0.4–0.6, at which point protein production was induced by adding isopropyl β-d-1-thiogalactopyranoside (IPTG) to a final concentration of 1 mM. Then, the cells were cultured at 37 °C for 3 h. The cells were harvested by centrifugation at 5,000 × g and 4 °C for 10 min. The cell pellet was frozen by liquid nitrogen and stored at −20 °C until purification. The cell pellet from a 1-L culture was resuspended in 20 mL lysis buffer [50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 25 mM imidazole] supplemented with lysozyme (A3711, 1 mg/mL; AppliChem), DNase I (A3778, 10 μg/mL; AppliChem), and a protease inhibitor mixture (P8849; Sigma), and incubated at room temperature for 1 h. After lysis by sonication and clearance by centrifugation at 9,500 × g and 25 °C for 25 min, the supernatant was loaded onto 3 mL of Ni(II)-NTA agarose resin (Qiagen) in a gravity flow column. After washing with lysis buffer containing 20 mM imidazole, protein capsid was eluted with lysis buffer containing 500 mM imidazole. The eluted solution was dialyzed with 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA by using Snakeskin Dialysis Tubing [10K molecular weight cutoff (MWCO); Thermo Scientific]. Protein capsids were further purified by size-exclusion chromatography [column, Superose 6 increase 10/300 GL (GE Healthcare); eluent, 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA; flow rate, 1 mL/min; room temperature]. Protein and RNA concentrations were determined by UV absorbance following a previously reported protocol (35). Extinction coefficients for proteins were calculated using the ExPASy ProtParam tool (https://web.expasy.org/protparam/).

RNA Extraction from Purified Capsids.

After size-exclusion chromatography, the protein capsids containing ∼2 μg RNA were mixed with 2.5 U/μL Benzonase (101654; Merck Millipore) or 10 μg/mL RNase A (R4875; Sigma-Aldrich) at 37 °C for 1 h in 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl, 5 mM EDTA, and 5 mM MgCl₂. After Benzonase treatment, RNAs were extracted with TRIzol reagent (15596026; Invitrogen) and dissolved in water.

Growth Assay.

Plots of cell density vs. time were generated to assess the ability of individual capsid variants to protect cells from the harmful effects of HIV protease (6). E. coli BL21(DE3)-gold cells were transformed with plasmids pMG-AaLS-WT or pMG-λcpAaLS and pAC-Ptet-HIV or pAC-Ptet-dB-HIV. Single colonies were used to inoculate precultures in LB medium (5 mL) supplemented with ampicillin (50 μg/mL) and chloramphenicol (10 μg/mL). These precultures were shaken overnight at 30 °C and 230 rpm and then used to inoculate a second batch of precultures in fresh LB medium (5 mL) containing ampicillin (50 μg/mL), chloramphenicol (10 μg/mL), and salicylate (100 μM), which induces production of the AaLS capsid at a starting OD600 of 0.01. These second precultures were grown for 1 h at 30 °C and 230 rpm. This 5-mL culture was split into 300-μL portions, and production of HIV protease was induced with different tetracycline concentrations. The cultures were shaken at 30 °C and 240 rpm for 11 h, and the OD600 was recorded every 30 min.

epPCR Library Construction.

epPCR was carried out by using the JBS Error-Prone Kit (PP-102; Jena Bioscience) according to the manufacturer’s protocols. Primers Lcp-F1 and Lcp-R2 and template pMG-λcpAaLS were used for the library in round 1, and primers Lcp-F3 and Lcp-R3 and template pMG-dB-λcpAaLS-β16 were used for the library in round 2. Purified epPCR products and the acceptor vector were digested with XbaI/XhoI for pMG-λcpAaLS and NdeI/XhoI for pMG-dB-λcpAaLS. The fragments were ligated with T4 DNA ligase (M0202; NEB) and purified. Electrocompetent E. coli XL1-Blue cells were transformed with the capsid libraries (approximately 1 μg ligation product) by electroporation and then incubated in 50 mL LB medium for 1 h at 37 °C. The library size was determined by plating of serial dilutions of these cells onto LB agar containing ampicillin (50 μg/mL), and a 1–5 × 10⁶ library size was confirmed for each round of directed evolution. To the remaining cells, LB medium and ampicillin (50 μg/mL) were added to a final volume of 50 mL and then cultured overnight at 37 °C and 230 rpm, and plasmids were prepared.

Round 1 of Directed Evolution by Inhibiting Translation of HIV Protease mRNA.

Plasmids encoding capsid variants with enhanced HIV protease mRNA encapsulation abilities were selected by growing pools of cells containing individual members of the plasmid libraries under conditions of high protease production (6). Electrocompetent E. coli XL1-Blue cells containing pAC-Ptet-dBHIV encoding double BoxBr-tagged mRNA for HIV protease were transformed with the pMG-λcpAaLS libraries (approximately 1 μg plasmid) by electroporation and then incubated in 50 mL LB medium for 1 h at 30 °C and 230 rpm. Ampicillin (50 μg/mL), chloramphenicol (10 μg/mL), and salicylate (100 µM) were added to a final volume of 50 mL, and cells were cultured for 1 h at 30 °C. To begin the selection, the HIV-protease gene was induced with 100 ng/mL tetracycline. The cells were then grown at 30 °C and 230 rpm for 24 h, and plasmid library (pMG-λcpAaLS-Lib1.1) was purified.

Round 2-1 of Directed Evolution.

The ORF region of pMG-λcpAaLS-Lib1.1 was amplified by PCR by using the primers Lcp-F1 and Lcp-R4. Purified PCR product and the acceptor vector pMG-dB-λcpAaLS were digested with NdeI/XhoI. The digested PCR product and the acceptor vector were purified and ligated, and the plasmid library was prepared as described in epPCR Library Construction. Electrocompetent E. coli BL21(DE3)-gold cells were transformed with the resulting library (pMG-dB-λcpAaLS-Lib1.1; approximately 1 μg ligation product) by electroporation and then incubated in 5 mL LB medium for 1 h at 37 °C. After ampicillin (final concentration, 50 μg/mL) was added, the cells were cultured at 37 °C and 230 rpm until the OD600 reached 0.4–0.6, at which point protein production was induced by adding IPTG to a final concentration of 1 mM. Then, the cells were cultured at 37 °C and 230 rpm for 3 h. The cells were harvested by centrifugation at 15,000 × g and 4 °C for 10 min. The cell pellet was resuspended in 400 μL lysis buffer (50 mM sodium phosphate buffer, pH 7.4. containing 1 M NaCl and 20 mM imidazole) supplemented with 1× BugBuster protein extraction reagent (70921; Merck Millipore), 30 U/μL diluted rLysozyme (71110; Merck Millipore), 10 μg/mL DNase I, and a protease inhibitor mixture and incubated at room temperature for 1 h. After clearance by centrifugation at 15,000 × g and 25 °C for 25 min, the supernatant was loaded onto 0.5 mL of Ni(II)-NTA agarose resin (Qiagen) in a spin column. After washing with lysis buffer, protein capsid was eluted with 50 mM sodium phosphate buffer (pH 7.4) containing 0.5 mM EDTA, 1 M NaCl, and 500 mM imidazole. The buffer was exchanged to 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA using an Amicon Ultra-15 centrifugal filter unit (30K MWCO; Merck Millipore). The protein capsids containing ∼2 μg RNA were treated with 2.5 U/μL Benzonase at 37 °C for 1 h in 250 μL 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl, 5 mM EDTA, 5 mM MgCl₂. After Benzonase treatment, RNAs were extracted with TRIzol reagent (15596026; Invitrogen) and dissolved in water. The resulting RNA sample was incubated with RQ1 DNase (M6101; Promega) in 1× reaction buffer at 37 °C for 30 min, and RNA was prepared by phenol-chloroform extraction and ethanol precipitation. cDNA was reverse-transcribed from RNA by SuperScript III reverse transcriptase (18080044; Invitrogen) with the primer λcpAaLS-R3 according to the manufacturer’s protocol. The resulting cDNA was amplified by 30 cycles of PCR with Phusion High-Fidelity DNA polymerase (M0530; NEB) with the primers λcpAaLS-F3 and λcpAaLS-R3. Purified DNA was digested and ligated with pMG-dB vector. After three repeats of these steps, 18 clones were picked and the expression of the capsid was observed in four clones (dB-λcpAaLS-β3, -β5, -β10, -β16). These four variants were expressed and purified by Ni-NTA resin as the same way and further purified by size-exclusion chromatography [column, Superose 6 increase 10/300 GL (GE Healthcare); eluent, 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA; flow rate, 1 mL/min; room temperature]. After size-exclusion chromatography, Benzonase treatment, RNA extraction, and RT-PCR were carried out as the same way as round 1–2 of directed evolution.

Round 2-2 of Directed Evolution.

An epPCR library was prepared from pMG-dB-λcpAaLS-β16 as described in epPCR Library Construction. Electrocompetent E. coli BL21(DE3)-gold cells were transformed with the resulting library (pMG-dB-λcpAaLS-Lib2; approximately 1 μg ligation product) by electroporation and then incubated in 5 mL LB medium for 1 h at 37 °C. Ampicillin (50 μg/mL) was added to a final volume of 5 mL and cultured at 37 °C and 230 rpm for 6 h. These 5-mL cultures were inoculated to 1.2 L LB medium containing 50 μg/mL ampicillin and cultured at 37 °C and 230 rpm until the OD600 reached 0.4–0.6, at which point protein production was induced by adding IPTG at final concentration 1 mM. Then, the cells were cultured at 37 °C and 230 rpm for 3 h. The cells were harvested by centrifugation at 5,000 × g and 4 °C for 10 min. The cell pellet was stored at −20 °C until purification. The cell pellet was resuspended in 20 mL lysis buffer (50 mM sodium phosphate buffer, pH 7.4, containing 1 M NaCl and 25 mM imidazole) supplemented with lysozyme (1 mg/mL), DNaseI (10 μg/mL), and a protease inhibitor mixture and incubated at room temperature for 1 h. After lysis by sonication and clearance by centrifugation at 9,500 × g and 25 °C for 25 min, the supernatant was loaded onto 3 mL of Ni(II)-NTA agarose resin (Qiagen) in a gravity flow column. After washing with lysis buffer containing 20 mM imidazole, protein was eluted with lysis buffer containing 500 mM imidazole. The buffer was exchanged for 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA by using an Amicon Ultra-15 centrifugal filter unit (30K). Capsids were further purified by size-exclusion chromatography [column, HiPrep 16/60 Sephacryl-S400 (GE Healthcare); eluent, 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA; flow rate, 1 mL/min; room temperature]. The capsids containing 2 μg RNA were treated with 10 μg/mL RNase A at 37 °C for 1 h in 250 μL 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl, 5 mM EDTA, and 5 mM MgCl₂. After RNase treatment, RNAs were extracted with TRIzol reagent and full-length mRNAs were purified by denaturing PAGE [8% acrylamide, 8 M urea, 1× Tris-borate-EDTA (TBE)] and ethanol precipitation. Contaminated DNA was removed from purified RNAs by DNase I treatment, and RT-PCR was carried out by using Lcp-F6 and Lcp-R6 primers under the same conditions as in round 1–2. Purified DNA was digested and ligated with pMG-dB vector. After three repeats of these steps, 12 clones were picked and the expression of the capsid was observed in eight clones (dB-λcpAaLS-α1, -α2, -α6, -α8, -α9, -α10, -α11, -α12). These eight variants were expressed and purified by Ni-NTA and size-exclusion chromatography [column, Superose 6 increase 10/300 GL (GE Healthcare); eluent, 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl and 5 mM EDTA; flow rate, 1 mL/min; room temperature], and packaged RNAs were extracted with TRIzol reagent after Benzonase or RNase A treatment. Extracted RNAs were analyzed by denaturing PAGE (8% acrylamide, 8 M urea, 1× TBE) and GelRed staining.

RNA Extraction from Purified Capsids.

After size-exclusion chromatography, the protein capsids containing ∼2 μg RNA were mixed with 2.5 U/μL diluted Benzonase at 37 °C for 1 h in 50 mM sodium phosphate buffer (pH 7.4) containing 1 M NaCl, 5 mM EDTA, and 5 mM MgCl₂. After Benzonase treatment, RNAs were extracted with TRIzol reagent (15596026; Invitrogen) and dissolved in water.

Denaturing PAGE Analysis.

RNA samples were mixed with equal volume of 2× RNA loading buffer (10 M urea, 2 mM EDTA, and 2 mM Tris⋅HCl, pH 7.5) and incubated at 95 °C for 5 min and then chilled on ice. Denatured RNA samples were loaded onto denaturing PAGE gel (8% or 6% acrylamide, 8 M urea, 1× TBE). The gels were stained with DFBHI-1T (410; Lucerna) and subsequently with GelRed (41002; Biotium) following previously reported protocols (16). DFHBI-1T–stained gels were visualized by blue LED transillumination (Pearl Biotech) and GelRed stained gels by UV transillumination.

Quantification of Encapsulated RNA.

The total nucleic acid content of λcpAaLS capsids was quantified by UV absorbance following a previous report (35). The number of full-length mRNAs in the extracted RNA was quantified by quantitative RT-PCR. RNAs extracted from capsids were treated with RQ1 DNase (M6101; Promega) following the manufacturer’s protocol. DNase-treated RNAs were purified by phenol-chloroform extraction, precipitated with ethanol, dissolved in water, and quantified by Qubit RNA HS Assay Kit (Q32852; Invitrogen). First-strand cDNAs were reverse-transcribed from 30-pg RNAs by SuperScript III (18080093; Invitrogen) with the Lcp-RT-R1 primer following the manufacturer’s protocol, and 1/10 volume samples were mixed with PowerUp SYBR Green Master Mix (A25779; Applied Biosystems) and 500 nM forward (Lcp-qPCR-F1) and reverse primers (Lcp-qPCR-R1). Quantitative PCR was performed by using a StepOnePlus Real-Time PCR System (4376600; Applied Biosystems). Standard curves were prepared by using in vitro transcribed mRNAs. The number of full-length mRNAs per one capsid was calculated from the total nucleic acid content in λcpAaLS and the number of full-length mRNAs in the extracted RNA.

Negative Stain-Transmission EM.

Negative-stain transmission EM experiments were performed as reported previously (36) by using capsid (3 μM monomer) purified by size-exclusion chromatography, 50 mM sodium phosphate (pH 7.4) containing 1 M NaCl and 5 mM EDTA as washing solution, and 2% (wt/vol) aqueous uranyl acetate (pH 4) as staining solution.

In Vitro Transcription.

The RNAs were prepared by using runoff in vitro transcription with T7 RNA polymerase (EP0111; Thermo Scientific). DNA templates were prepared by PCR with the primers T7-F1 and BspEI-R1 by using Phusion High-Fidelity DNA polymerase from plasmids bearing desired sequences. After the in vitro transcription, templated DNA was digested by RQ1 DNase and RNA was precipitated by using isopropanol. Desired RNAs were purified by Tris-acetate-EDTA-agarose gel and Zymoclean Gel RNA Recovery Kit (R1011; Zymo Research) and quantified by Qubit RNA HS Assay Kit.

Long-Read Single-Molecule Sequencing.

The RNAs extracted from the capsids were polyadenylated by E. coli Poly(A) polymerase (M0276; NEB) following the manufacturer’s protocol. Then, cDNA libraries were prepared with the Direct cDNA Sequencing Kit (SQK-DCS108; Oxford Nanopore Technologies) and Native Barcoding Kit 1D (EXP-NBD103; Oxford Nanopore Technologies) following the manufacturer’s protocol. Sequencing was carried out in the flow cell (FLO-MIN106) using the 48-h 1D protocol, and acquisition was stopped after approximately 36 h. After the sequencing run finished, base calling and demultiplexing were performed with Oxford Nanopore Technology’s Albacore command line tool (v1.2.6). We then trimmed adapter sequences from the reads by using Porechop (v0.2.2; https://github.com/rrwick/Porechop). To reduce the risk of cross-barcode contamination, we ran Porechop with barcode demultiplexing and kept only reads for which Albacore and Porechop agreed on the barcode bin (37). The resulting demultiplexed, barcode-trimmed read sets were mapped by LAST (v869), following the protocol for aligning long DNA and RNA reads to a genome (https://github.com/mcfrith/last-rna/blob/master/last-long-reads.md), onto the E. coli-K12 genome (NC000913) and pMG or pMG-dB vectors encoding the corresponding λcpAaLS genes.

Calculation of RNA MLD.

One thousand suboptimal RNA structures for each sequence were generated by RNAsubopt, a program in the Vienna RNA software package version 2.2 (38). MLD values were calculated by using MATLAB software (MathWorks).

Coexpression of a BoxBr-Tagged mRNA Coding HIV Protease and λcpAaLSs.

λcpAaLS capsids and HIV protease mRNA with BoxBr tags were coproduced in E. coli BL21(DE3)-gold cells that were transformed with pMG-ΔdB-λcpAaLS-α2 or pMG-ΔdB-λcpAaLS-α9 and pAC-Ptet-dBHIV. The cells were grown at 37 °C in LB medium until the OD600 reached 0.4–0.6, at which point protein production was induced by adding IPTG to a final concentration of 1 mM and tetracycline at a final concentration of 500 ng/mL. Then, the cells were cultured at 37 °C for 3 h. λcpAaLS capsids were purified following the same procedure described in Expression and Purification of Capsids.

Graphic Software.

The graphics of protein structures used for the figures were generated by using University of California, San Francisco, Chimera (version 1.12) (39) or PyMoL (version 1.7; Schrödinger).