Codon optimization of therapeutic transgenes
Codon usage controls translation elongation rate and co-translational protein folding processes, and therefore plays an important role in determining protein expression levels[53–55]. Genes that encode highly expressed proteins are strongly enriched for preferred codons, and CO has been shown to increase endogenous and heterologous gene expression in diverse eukaryotes[55–57]. Many commercially available CO algorithms are based on empirical indices including codon adaptation index (CAI), frequency of relative synonymous codon usage, codon bias index, optimal codon usage and effective codon number[58]. The CAI is the primary index used to predict gene expression level because it indicates the extent to which the codon sequence represents the usage of codons in a particular organism. In this study, we employed 5 different CO tools to avoid bias of any one algorithm and applied manual optimization to the algorithmically optimized CO sequences to further refine the coding sequence, including elimination of cleavage sites, cryptic splice donor/acceptor sites, transcriptional factor binding sites, restriction endonuclease sites, repeats and high GC contents.
10 CO sequences of human FHL-1 coding DNA (cDNA) were first compared to the WT sequence (RC001) for FHL-1 protein expression in human retinal pigment epithelial (ARPE19) cells. pAAV-COFHL-1 constructs carrying the FHL-1 cDNA with a N-terminal CFH signal peptide were placed under the transcriptional control of a CAG promoter in conjunction with the woodchuck post-transcriptional element (WPRE), both elements have previously been included in clinical gene therapy vectors without any safety concerns[59,60] (Fig. 2A). The 10 pAAV-COFHL-1 constructs (RC138-147) were transiently transfected into ARPE19 cells at equal concentrations and the amount of FHL-1 protein in the cell supernatant was detected by Western blot (Fig. 2B). Densitometric analyses showed that RC141, RC144, RC145 and RC146 were amongst the highest expressing constructs compared to the WT FHL-1 sequence of RC001.
The top performing pAAV-COFHL-1 expression cassettes were then packaged into rAAV2 vectors, and the level of FHL-1 protein in cell supernatant was determined by both ELISA and Western blot 72 hr post transduction of ARPE19 cells. Only a proportion of CO sequences elicited higher FHL-1 protein expression than the WT sequence (RC001), however both data sets showed that the CO sequence of RC146 induced the highest increase in protein expression and secretion compared to WT FHL-1 sequence (RC001) (Fig. 2C and 2D).
Similarly, 10 CO aflibercept coding sequences (RC290-299) were subcloned into the same expression cassette as that of RC146 (Fig. 2A) and packaged into rAAV8 vectors. Aflibercept expression in the cell supernatant was determined by both ELISA and Western blot 72 hr post transduction of human embryonic kidney (HEK293) cells. Aflibercept ELISA showed that protein levels in the cell supernatant was highest for RC298 compared to the non-CO sequence RC288 (Fig. 2E). Western blot analysis showed a dominant protein band at approximately 150 kDa, which corresponds with the aflibercept homodimer molecular weight under non-reducing conditions. However, aberrant protein species were detected in RC299, possibly indicating that the substitution of synonymous codons had negatively impacted protein processing within this particular sequence. Densitometric analysis corroborated with the ELISA data showing that RC298 was the highest expressing CO sequence (Fig. 2F). Based on these expression data, the CO sequences of RC146 (FHL-1) and RC298 (aflibercept) were used in further experiments to evaluate the bicistronic vector.
Bicistronic vector design and in vitro characterization of secreted proteins
The use of rAAV vectors for efficient expression of two genes has previously been described in multiple preclinical studies and has been used for the clinical delivery of recombinant heavy- and light-chain Fab fragments[61,62]. Conventional bicistronic expression cassettes have employed either two tandem promoters or bidirectional promoters to drive the expression of two genes independently. However, the limited packaging capacity of rAAV vectors often render these options non-viable. Alternatively, a single promoter driving expression of two genes simultaneously linked by translational control elements such as an internal ribosome binding site (IRES) or 2A linkers can provide a partial solution. The IRES sequence permits the production of multiple proteins from a single mRNA transcript but suffers from two main limitations. First, the IRES-dependent downstream second gene can be expressed at significantly lower level within the vector, and secondly, the size of the IRES element is often in excess of 500 base pairs[63]. These issues can be mitigated by using 2A peptide sequences derived from a large group of viral families including the Foot and Mouth Disease Virus (F2A), equine rhinitis A virus (E2A), porcine teschovirus-1 (P2A) and Thosea Asigna virus (T2A)[64]. 2A peptides are 18–25 amino acid viral oligopeptides that have been shown to mediate efficient bicistronic expression of two gene products from a single promoter through ribosomal skipping (often referred to as self-cleavage) during protein translation[65,66]. Theoretically, the two gene products are expressed at 1:1 molar ratio and each gene product is expressed at high levels. The 2A and 2A-like bicistronic systems have been shown to be highly effective in the CNS where high expression levels of both genes have been demonstrated without any cytotoxic effects[67].
Previous reports have compared the cleavage efficiencies of different 2A motifs in a multi-cistron setting and demonstrated mixed results with a dependency on the genes being expressed[68,69]. The F2A peptide has been previously used in a clinical rAAV8-antiVEGFfab vector[62]. In addition, a furin recognition site was placed upstream of the 2A to enable the removal of the remaining 2A residues after self-cleavage. To further guide optimal bicistronic design, the effect of gene position on protein expression was assessed whereby four bicistronic vectors (RC304, RC312, RC318 and RC319) with alternating positions of FHL-1 and aflibercept genes were evaluated (Fig. 2A). Two combinations of promoter and post-transcriptional elements (CMV-WPRE and CAG-WPRE3) were also tested.
rAAV8 bicistronic vectors were transduced into HEK293 cells and protein expression was assessed in the cell supernatant by ELISA and Western blot at 72 hr post-transduction. FHL-1 expression was detected from all bicistronic vectors but at relatively lower levels compared to monocistronic vector (RC146), irrespective of gene position or promoter-WPRE combination, with RC312 showing the highest level of FHL-1 expression (Fig. 3A). Aflibercept expression from RC304 and RC312 were similar to the monocistronic vector (RC298), while RC318 and RC319 showed reduced expression (Fig. 3A). Overall, the CMV-WPRE configuration was observed to drive higher expression levels of both transgenes. The position of the transgenes within each promoter/post-transcriptional element combination appeared to have affected expression differently. Interestingly, it was observed across all bicistronic configurations that the overall molar ratio of FHL-1 expression was higher than that of aflibercept, irrespective of promoter and post-transcriptional element types. Western blots showed that secreted FHL-1 and aflibercept proteins were of the correct expected band size with no observation of uncleaved or incorrectly processed protein products (Fig. 3B).
The biological activity of vector-derived proteins in the cell supernatant of RC304 and RC312 transduced cells was then assessed by respective in vitro VEGF- and complement-binding assays. To determine the binding affinity of aflibercept for VEGF, an equilibrium binding assay was performed in which different concentrations of vector-derived aflibercept in the cell supernatant of transduced cells were incubated with VEGF-A165 and the amount of unbound VEGF-A165 was measured. Figure 3C shows that vector-derived aflibercept had a similar VEGF-binding affinity to the recombinant aflibercept control. Furthermore, to assess the ability of vector-derived aflibercept to block VEGF-stimulated human umbilical vein endothelial cell (HUVEC) proliferation, both VEGF and vector-derived aflibercept were added to cultured HUVEC cells. In agreement with the VEGF-binding affinity assay, all vector-derived aflibercept demonstrated an equal inhibitory effect on VEGF-dependent HUVEC proliferation compared to recombinant aflibercept control (Fig. 3D). The activity of vector-derived FHL-1 was demonstrated through an in vitro C3b binding assay showing the detection of high levels of iC3b (breakdown product of C3b cleavage) from both RC304 and RC312, similar to the recombinant FHL-1 protein control (Fig. 3E). These data clearly indicate that all bicistronic vectors express and secrete biologically active aflibercept and FHL-1 proteins.
In vivo dose escalation expression study
A dose escalation expression study was first carried out using monocistronic RC298 in vivo to determine a safe dose range for subsequent efficacy assessment of the bicistronic vectors in the mouse laser-induced CNV model. Monocistronic rAAV8-aflibercept vectors (RC298) were subretinally injected into WT mice in escalating doses ranging from 5e7 to 1e10 vg/eye (Fig. 4A). At 4-weeks post-injection, vector genome copy numbers were quantified in mouse eyecups and a dose-dependent increase in genome copies was observed up to 5e9 vg/eye (Fig. 4B). Similarly, a linear dose-dependent increase in aflibercept protein expression was observed, as well as a ceiling effect at the higher doses of 5e9 and 1e10 vg/eye (Fig. 4C). Based on the observed dose to expression relationship, the dose threshold was established to be between 5e8 and 5e9 vg/eye. To avoid the potential toxic effects that are often associated with higher doses, 5e7 vg/eye was selected as the representative dose for evaluating the therapeutic effect of bicistronic vectors.
In vivo bicistronic expression study
To assess dual expression of FHL-1 and aflibercept in vivo, both RC304 and RC312 were subretinally injected into mouse eyes at 5e7 and 5e8 vg/eye (Fig. 5A). At 4 weeks post-injection, vector genome copy numbers and protein expression levels were measured by qPCR and ELISA in mouse eyecups and ocular fluids respectively. For quantitative comparison of vector genome copies and protein expression levels between monocistronic and bicistronic vectors, we extrapolated relative expression data from monocistronic aflibercept vectors in Fig. 4. qPCR analyses in mouse eyecups showed a clear dose-response effect in genome copies, but it was observed that the concentration of rAAV8 bicistronic vectors were 1.2 to 3.6-fold lower than that of monocistronic aflibercept vectors at both respective doses (Fig. 5B).
Assessment of protein expression in mouse ocular fluids also showed a linear dose-response effect, with RC304 and RC312 showing similar levels of aflibercept expression across both doses (Fig. 5C). However, it was noted that aflibercept protein expression from both bicistronic vectors was approximately 3–10 fold lower than the monocistronic vector (Fig. 5C). FHL-1 protein expression was observed to be moderately higher in RC304 across both doses compared to RC312 (Fig. 5D). In addition, it was observed that the overall molar ratio of aflibercept expression was higher than FHL-1 for both configurations, contrasting the in vitro data observed in Fig. 3A.
The data here showed that the bicistronic vectors were capable of driving the expression of two gene products in rodent retinas in a dose-dependent manner, despite lower expression levels of aflibercept compared to the monocistronic counterpart. In the ocular fluids, both RC304 and RC312 showed similar aflibercept expression levels but RC304 was observed to deliver higher FHL-1 protein levels. Based on this observation, RC304 was investigated for its therapeutic effect in the laser-induced CNV mouse model.