Stoichiometric expression of mRNA polycistrons by eukaryotic ribosomes (SEMPER) for compact, ratio-tunable multi-gene expression

SUMMARY

Here we present a method for expressing multiple open reading frames (ORFs) from single transcripts using the leaky scanning model of translation initiation. In this approach termed Stoichiometric Expression of mRNA Polycistrons by Eukaryotic Ribosomes (SEMPER), adjacent ORFs are translated from a single mRNA at tunable ratios determined by their order in the sequence and the strength of their translation initiation sites. We validate this approach by expressing up to three fluorescent proteins from one plasmid in two different cell lines. We then use it to encode a stoichiometrically tuned polycistronic construct encoding gas vesicle acoustic reporter genes that enables efficient formation of the multi-protein complex while minimizing cellular toxicity. We also demonstrate that SEMPER enables polycistronic expression of recombinant monoclonal antibodies from plasmid DNA and of two fluorescent proteins from single mRNAs made through in vitro transcription. Finally, we provide a probabilistic model to elucidate the mechanisms underlying SEMPER.

Graphical Abstract

eTOC blurb

Discover SEMPER, a novel method for precise, user-defined multi-protein expression from a single transcript. Utilizing the leaky scanning model, SEMPER efficiently controls protein ratios, proven in applications for fluorescent protein, gas vesicle, and antibody expression. Its design and probabilistic model optimize complex protein assemblies in eukaryotic systems.

INTRODUCTION

Mammalian cell engineering promises to enable treatment of age-related degeneration, reversal of genetic diseases, and even transformation of cells into living therapeutics and sensors.¹ Realizing this promise requires the development of genetic circuits capable of finely tuning the relative expression stoichiometries of multiple proteins to produce functional multimeric protein assemblies, multi-component signaling systems, or multi-enzyme biosynthetic pathways.^2–6 Current approaches to doing so at the DNA level (e.g., varying promoter strength^7,8 or titrating copy numbers of each gene^2,9) yield lengthy DNA constructs that often must be packaged into multiple delivery vectors and lead to undesirable cell-to-cell variability due to both stochastic gene delivery and integration across the population. Additionally, attainable protein expression stoichiometries are limited by the transcriptional strengths of a relatively small set of curated promoters that often demonstrate cell-to-cell variability.^7,8,10

Post-transcriptionally, sequence motifs such as internal ribosome entry sites (IRES) or 2A self-cleaving peptides may be used to encode multiple open reading frames (ORF) into a single transcript and tune protein stoichiometries using relative translation levels.^11,12 Such post-transcriptional mechanisms are also useful for mRNA-based, multi-gene expression systems, which are of relevance to mRNA vaccine and therapeutic development.¹³ Although powerful tools, these genetic parts have their disadvantages. For instance, IRES sequences have a considerable genetic footprint (~200–600 bps),¹⁴ leading to lengthier genetic constructs which may reduce viral packaging efficiency. Likewise, 2A self-cleaving peptides leave peptide scars and can yield undesired fusion proteins, both of which can be detrimental to protein function.¹⁵ In addition, when employed in bicistronic vectors, these genetic parts can strongly attenuate the expression of the second ORF relative to the expression of the first ORF.^12,16 Here, we present an alternative approach that enables compact and robustly tunable polycistronic expression in mammalian cells using plasmid- and mRNA-based expression systems. We call this approach “Stoichiometric Expression of mRNA Polycistrons by Eukaryotic Ribosomes” (SEMPER).

SEMPER uses the canonical cap-dependent ribosome recruitment and translation mechanism in mammalian systems, which begins when the 43S preinitiation complex (PIC) of the ribosome is loaded onto the 5’ end of mRNA.^17,18 This complex then scans the 5’ untranslated region (5’ UTR) until it encounters a translation initiation site (TIS), consisting of the start codon (AUG) and ~3–10 neighboring nucleotides.¹⁹ The TIS sets the translational reading frame and initiates translation by engaging with the 60S ribosomal subunit. The full ribosome then translates the mRNA into protein until it encounters a stop codon, where it terminates translation and disengages the transcript. With some frequency, the 43S PIC may scan through the first TIS and initiate translation from a downstream ORF starting at another TIS, a phenomenon called leaky ribosomal scanning (LRS).²⁰ As determined by its sequence, a strong TIS (e.g. the Kozak consensus sequence) will reliably initiate translation while weaker ones will more frequently allow the 43S PIC to scan past.²¹

By employing a short ORF (uORF) upstream of a gene of interest (GOI), mammalian cells naturally use LRS and alternate TISs to divert a portion of the ribosome flux away from a GOI, effectively downregulating its translation.²² Recently, Ferreira and colleagues demonstrated that it is possible to use synthetic uORFs to regulate the expression of a downstream recombinant GOI.²³ They also empirically determined the strength of various translation initiation sequences and showed that it is possible to divert varying amounts of ribosomal flux away from the GOI by varying the uORF TIS strength. uORFs have also been used in synthetic systems to tune an endoribonuclease-based feedforward controller to manage resource limitation.²⁴ Others have also utilized translation initiation strength variability to tune the expression of a GOI in the context of modular genetic design.²⁵ The SEMPER approach replaces non-protein-coding uORFs with longer, functional coding sequences. In this study, we show that by chaining together multiple ORFs while varying the translation initiation strength of each ORF, this approach achieves polycistronic expression of multiple proteins with tunable translation rates. We further demonstrate that this framework is functional in in vitro transcribed mRNA, paving the way for advances in mRNA-based protein therapeutics of higher complexity.

RESULTS

Tunable, plasmid-based SEMPER framework for expressing two ORFs.

To test tunable translation levels of two recombinant proteins from single transcripts, we encoded fluorescent proteins (FPs) with minimal spectral overlap into the first two ORFs of our SEMPER plasmid vector (Figure 1A). We included 11 bps of distance between these ORFs to ensure that read-through of the stop codon of the first ORF would not result in a fusion containing both proteins. We used the TIS sequence NNNAUGG to initiate translation of our ORFs, where NNN represents one of the following sequences: ACC, CCC, TTT. These sequences were described as having strong, medium, and weak translation initiation efficiency respectively by Ferreira et al.²³ We engineered all our FPs to contain a valine residue (GTG or GTT) following the N-terminal methionine to ensure that changes in translation initiation were due to the trinucleotide preceding the start codon and not the dinucleotide succeeding it. TIS sequences will be referred to by their variable NNN sequence. We used one other sequence (TTTCCAT), referred to as ***, to scrub the TIS entirely and prevent the ORF from being translated. For the first ORF, we generated a methionine-less monomeric Superfolder GFP (msfGFP[r5M]) in which all methionines—except the N-terminal one—were mutated to other amino acids.^26,27 In addition, out-of-frame AUGs were removed from the msfGFP[r5M] coding sequence using synonymous mutations. These mutations effectively removed all internal TISs within ORF 1 that could reduce ribosomal flux to downstream ORFs (Figure 1B). In the second ORF, we encoded mEBFP2 with all its natural methionines. Finally, downstream of the SEMPER ORFs, we included an IRES followed by mCherry (IRES-mCherry) for normalization. As ribosomal binding to the IRES and subsequent translation of mCherry are conducted independently of ORF 1 and ORF 2 translation,^28,29 the mCherry allowed us to normalize single-cell fluorescence measurements for msfGFP[r5M] and mEBFP2, a strategy common to LRS-focused studies.^23,30 Because the analyte ORFs and IRES-mCherry were encoded on the same transcript, this normalization scheme accounted for variations in transfection efficiency, transcription, and mRNA decay. mCherry fluorescence also served as a proxy for transcript abundance in each cell (Figure S1 and Table S1A-1D).

Figure 1: 2-ORF SEMPER constructs demonstrate tunable, bicistronic expression.

A) Architecture of an mRNA transcript produced by transfected 2-ORF SEMPER plasmid DNA. Cap-dependent ribosomes translate ORF 1, msfGFP[r5M], or ORF 2, mEBFP2, with frequencies dependent on the trinucleotide (NNN) upstream of each ORF. The relative strengths of the trinucleotides and TISs they represent are depicted. *** (a.k.a. TTTCCAT) does not contain a start codon, preventing translation of the ORF. IRES-mCherry is included to normalize fluorescence measurements. B) mEBFP2 distribution plots for constructs containing different versions of monomeric Superfolder GFP encoded in ORF 1. Removal of internal methionines from the msfGFP leads to much stronger expression of the downstream mEBFP2. Both ORFs used the strong ACC TIS (N=1, representative of four biological replicates). C) Flow cytometry plots of mCherry positive HEK293T cells transfected with 2-ORF SEMPER constructs. The legend contains the ORF 1 TIS and the ORF 2 TIS separated by a slash. (Left) Increasing the strength of the first TIS increases msfGFP[r5M] fluorescence relative to that of mEBFP2 (N=1, representative of four biological replicates). (Right) Binning cells into three mCherry fluorescence ranges produces unique clusters of cells (N=1, representative of four biological replicates). D) Min-max normalized fluorescence values for each GOI relative to ACC/*** and ***/ACC. Fluorescence measurements were first normalized by mCherry. Cluster means were then calculated followed by min-max normalization. Error bars indicate mean ± SEM (N=4 biological replicates). Welch’s ANOVA tests were conducted for blue and green bars independently (p < 0.0001 for all colors across all bins) followed by Dunnett’s T3 multiple comparisons tests. All pairwise comparisons are provided in Table S2. E) Violin plots of log10(msfGFP[r5M]/mEBFP2) values for all mCherry positive cells for four combined biological replicates of TTT/ACC, CCC/ACC, and ACC/ACC plasmids transfected into HEK293T and CHO-K1 cell lines. Each distribution is produced from the measurements of at least 15,000 cells. The median and quartiles of the distribution are represented by the solid and dotted lines respectively. Statistical analysis was conducted using one-way Welch’s ANOVA tests (p < 0.0001 (Left) and p < 0.0001 (Right)) followed by Games-Howell’s multiple comparisons test to yield pairwise comparisons.

After cloning various combinations of TISs in front of our two ORFs, we transfected these “2-ORF SEMPER” constructs into human HEK293T cells—a widely used research model for mammalian cell biology. Three single-color control plasmids with msfGFP[r5M], mEBFP2, and mCherry were also transfected. We screened the transfected cells using flow cytometry, utilizing the single-color control plasmids for compensation and correction of fluorescence spillover emissions (Figure S2-3). As hypothesized, our cell lines produced both msfGFP[r5M] and mEBFP2 from single transcripts (Figure 1C). Furthermore, the tested TIS combinations (TIS for ORF 1 / TIS for ORF 2) yielded unique relationships between the fluorescence of the first ORF and that of the second ORF, with the relative expression of the former vs the latter following the strength of the first TIS. To analyze the 2-ORF SEMPER performance as a function of transcript abundance or “copy number”, we binned cells into three categories (low copy, medium copy, and high copy) based on their mCherry fluorescence. This yielded distinct clusters. We then calculated the means of msfGFP[r5M]/mCherry, mEBFP2/mCherry, and msfGFP[r5M]/mEBFP2 for the cells in each bin for each tested TIS combination. We conducted min-max normalization for each ORF separately using two constructs from our screen (***/ACC and ACC/***). We reasoned that ***/ACC would produce the maximum amount of mEBFP2 and the minimum amount of msfGFP[r5M], as all 43S PIC flux should bypass the first ORF and initiate translation at the second. Conversely, we expected ACC/*** to produce the maximum amount of msfGFP[r5M] and the minimum amount of mEBFP2. Conducted for each mCherry bin independently, this analysis enabled us to compare translation levels of a particular FP across various TIS combinations relative to its hypothesized maximum and minimum (Figure 1D). Pairwise statistical comparisons for these TIS combinations are provided in Table S2.

As we increased the TIS strength in front of msfGFP[r5M], we observed increases in msfGFP[r5M] translation levels and decreases in mEBFP2 translation levels for all mCherry bins. Across mCherry bins, we found that the rank order of relative translation levels for mEBFP2 for our different TIS combinations was conserved. Additional TIS combinations were also tested in HEK293T cells, showing that they can provide additional degrees of tuning in expression ratios (Figure S4).

To confirm generalizability across species and cell types, we also demonstrated that the 2-ORF SEMPER constructs yielded TIS combination-dependent relative translation levels of our ORFs in CHO-K1 cells, a widely used Chinese hamster ovary cell line for the production of biologics (Figure S5, Table S3).³¹ Upon comparing the distributions of log10(msfGFP[r5M]/mEBFP2) values between cell types (Figure 1E), we determined that the three TIS combinations tested maintained the same rank order in both HEK293T and CHO-K1 cell lines.

Scaling plasmid-based SEMPER constructs to three ORFs.

Enzymatic pathways and macromolecular assemblies are often composed of more than two proteins or peptide subunits.^2,6,32 To determine if the SEMPER system could be scaled beyond two ORFs to meet the needs of increasingly complex pathways and assemblies, we screened a variety of “3-ORF SEMPER” constructs. Following a similar strategy to the 2-ORF system, we first cloned and tested a methionine-less monomeric Superfolder BFP (msfBFP[r5M]).

We then encoded msfBFP[r5M], msfGFP[r5M], and emiRFP670—a far-red fluorescent protein—into a variety of 3-ORF SEMPER plasmids, maintaining IRES-mCherry for normalization (Figure 2A). To conduct min-max normalization, we cloned three plasmids where the starting TISs were removed from two of the ORFs using the *** sequence, while maintaining a strong TIS on the third: i) ***/***/ACC, ii) ***/ACC/***, iii) ACC/***/*** (Figure S6). The emiRFP670 coding sequence still encoded alternative translation initiation sites within it, yielding some observable far-red fluorescence in constructs ii) and iii). Additionally, we constructed and transiently transfected six other 3-ORF SEMPER plasmids with unique TIS combinations into HEK293T cells and measured their output using flow cytometry. Four single-color controls (msfBFP[r5M], msfGFP[r5M], emiRFP670, and mCherry) were used for fluorescence compensation (Figure S7-8). Following an mCherry binning strategy similar to that employed for 2-ORF SEMPER constructs, we analyzed the mCherry-normalized translation levels of our three ORFs relative to their own theoretical maxima and minima from i), ii), iii). With the tested TIS combinations, we achieved a wide range of relative translation levels for each ORF, demonstrating tunable co-expression of three genes from a single transcriptional unit (Figure 2B, Left).

Figure 2: Scaling tunable polycistronic expression beyond two ORFs.

A) Architecture of an mRNA transcript produced by transfected 3-ORF SEMPER plasmid DNA. B) (Left) Min-max normalized relative translation levels for each fluorescent protein. Cells were first clustered into three mCherry fluorescence bins (low, medium, high). Error bars indicate mean ± SEM (N=4 biological replicates). Welch’s ANOVA tests were conducted for blue, green, and red bars independently (p < 0.0001 for all colors across all bins). Post-hoc Dunnett’s T3 multiple comparisons tests are provided in Table S4. (Right) Comparisons of msfBFP[r5M] relative translation levels between the max control and those TIS combinations with multiple ACC driven ORFs. Error bars indicate mean ± SEM (N=4 biological replicates). Post-hoc Dunnet’s T3 comparisons tests demonstrate a significant upregulation in relative translation of msfBFP[r5M] for depicted constructs compared to that observed for ACC/***/***.

Notably, two combinations, ACC/TTT/ACC and ACC/ACC/ACC, yielded relative translation levels for msfBFP[r5M] greater than 1.0 in HEK293T cells across all mCherry bins (Figure 2B, Right). We also observed relative translation values for msfBFP[r5M] greater than 1.0 in CHO-K1 cells for TIS combinations in both medium and high bins (Figure S9, Table S5). Consistent with our observations, Wu et al. previously reported that the presence of downstream ORFs can increase the translation of upstream ones.³³ We hypothesize that constructs containing multiple ACC TISs have a higher prevalence of translating ribosomes, allowing these ribosomes and subunits—with their documented helicase activity—to maintain the mRNA secondary structure in an open conformation that is more amenable for ribosomal scanning, translation initiation, and subsequent protein expression.³⁴ However, more work is required to uncover the exact mechanisms that elicit this observed phenotype. For multiple TIS combinations, we found that the second and third ORFs achieved relative translation levels greater than 50% while still producing substantial levels of upstream ORF products. These results suggest there may be sufficient ribosomal flux to effectively scale the SEMPER paradigm to 4+ ORFs.

Based on the SEMPER mechanism, the relative expression of ORFs is expected to be tunable regardless of their order in the construct, although the exact expression levels could be affected by any resulting differences in RNA structure and stability. To examine this possibility, we interchanged the first and second ORFs (both of which lack methionines) in our 3-ORF constructs and found that the expected rank of relative expression was preserved (Figure S10). However, the exact expression levels varied with different ORF ordering, suggesting that some iterative tuning is required to achieve specific target values.

Producing gas vesicle ultrasound reporters using 2-ORF SEMPER.

Next, we set out to demonstrate that 2-ORF SEMPER constructs could be applied to encoding a multimeric protein complex by expressing gas vesicles (GV) using mammalian acoustic reporter genes (mARGs).³⁵ Originally evolved in prokaryotes, GVs were recently introduced as genetically-encodable reporters for ultrasound imaging, enabling the noninvasive imaging of dynamic cellular processes in living organisms.^2,36 A single GV is made up of many GvpA structural units that are assembled together in a helical pattern through the cooperative activity of six heterologous assembly factors and minor constituents, referred to collectively as GvpNJKFGW.^37,38 Using a two-vector system, our group has previously expressed GVs in mammalian cells by co-transfecting one plasmid encoding the structural unit upstream of IRES-mCherry (pgvpA-IRES-mCherry) along with another plasmid encoding the assembly factors and a terminal Emerald GFP (EmGFP), all linked together by P2A elements (pgvpNJKFGW-EmGFP). Additionally, we have found that co-transfecting the pgvpA-IRES-mCherry in excess of pgvpNJKFGW-EmGFP improves acoustic contrast.³⁵ Likewise, Anabaena flosaquae—the organism from which mARGs used in this study are derived—contains more copies of gvpA relative to the other gvps in its GV gene cluster.³⁹

Using the 2-ORF SEMPER strategy, we cloned single-vector systems for producing GVs in mammalian cells. We refer to these constructs as SEMPER mARGs. As gvpA does not contain any internal methionines, we encoded it directly into the first ORF. We tested our panel of TIS sequences in front of gvpA while maintaining the strong ACC TIS in front of the gvpNJKFGW-EmGFP ORF (Figure 3A). We compared these SEMPER constructs to our published two-vector expression system, mixing the gvpA-IRES-mCherry plasmid in 4-fold molar excess of the gvpNJKFGW-EmGFP plasmid. We transfected these plasmids into HEK293T cells, maintaining the same total mass of plasmid for each transient transfection. As the strength of the TIS in front of gvpA increased, we found that the translation level of gvpNJKFGW-EmGFP, as measured by Emerald/mCherry, decreased (Figure 3B), as expected from our 2-ORF SEMPER FP experiments. As measured by BURST ultrasound imaging,⁴⁰ the ACC/ACC combination—predicted to yield the highest ratio of GvpA to GvpNJKFGW—produced the strongest acoustic contrast compared to the other SEMPER mARG plasmids (Figure 3C-D).

Figure 3: Utilizing 2-ORF SEMPER constructs to express gas vesicles in mammalian cells.

A) Architecture of an mRNA transcript transcribed from transfected SEMPER mARG plasmid DNA. The first ORF, gvpA, encodes the main structural protein, while the second ORF, gvpNJKFGW-EmGFP, encodes all necessary accessory proteins strung together with P2A self-cleaving peptides. B) Flow cytometry distributions of Emerald GFP normalized by mCherry values for each TIS combination tested in addition to an IRES-mCherry control. The thick horizontal lines depict the means of the distributions. Individual cells are represented as individual points. A minimum of 4,000 cells are depicted per condition. Error bars indicate mean ± SEM. p-value <0.0001 (Kruskal-Wallis test). p-values of multiple comparisons are found in Table S6. C) BURST images of acoustic contrast due to gas vesicle expression within HEK293T cells. Depicted are HEK293T cells loaded into agarose phantoms three days after transfection of SEMPER mARG plasmids or the leading two-plasmid system. The color bar represents the magnitude of BURST signal measured in linear arbitrary units. The floor and ceiling of the images are set to 0 and 5000 respectively (N=1, representative of four biological replicates). D) BURST signal quantification of gas vesicle acoustic contrast for HEK293T samples transfected with SEMPER mARG plasmids or the leading two plasmid system. Error bars indicate mean ± SEM (N=4 biological replicates). Statistical analysis was conducted using a one-way ANOVA (p < 0.0001) followed by Fisher's LSD post-hoc test with a single pooled variance to compare each treatment group with every other. Pairwise p-values are reported in Table S7. E) Annexin V staining assays to quantify the number of apoptotic cells following expression of mARG vectors. HEK293T cells transfected with pgvpA-IRES-mCherry with the start codon removed in front of gvpA were used to establish a baseline (Not Treated). A subset of these cells was treated with Raptinal to induce apoptosis (Raptinal Treated). Other cell populations were transfected solely with a fully functional pgvpA-IRES-mCherry (gvpA-IRES-mCherry), the two-vector system (Two-vector), or the ACC/ACC SEMPER mARG plasmid (SEMPER mARG). Error bars indicate mean ± SEM (N=4 biological replicates). Statistical analysis was conducted using a one-way ANOVA (p < 0.0001) followed by Fisher's LSD post-hoc test with a single pooled variance to compare each treatment group to the not treated control. Pairwise p-values are indicated above the bars.

SEMPER gas vesicle expression system reduces cell toxicity.

A critical issue with multimeric protein assemblies is the potential for cellular burden or toxicity due to imperfect stoichiometry or the absence of an essential assembly component or chaperone. In our experiments with mARGs, we observed that samples transfected with the two-vector system contained a larger fraction of cells that were positive for pgvpA-IRES-mCherry but negative for pgvpNJKFGW-EmGFP (39.03% ±1.42%) compared to cells transfected with SEMPER mARG plasmids (13.40%±0.85%) (Figure S11). This is expected due to the inherent stochasticity of transient co-transfection. As GvpA subunits have been speculated to nonspecifically aggregate when expressed without GV assembly factors,^41,42 we hypothesized that cells receiving a sub-optimal ratio of pgvpA-IRES-mCherry : pgvpNJKFGW-EmGFP or only pgvpA-IRES-mCherry may have higher incidences of apoptosis due to the formation of cytotoxic GvpA aggregates in the cytoplasm.

To test this hypothesis, we performed Annexin V-based apoptosis assays on cells transfected with either pgvpA-IRES-mCherry alone, the two-vector mARG expression system at optimal transfection ratio, or the ACC/ACC SEMPER mARG (Figure 3E). In each condition, molar amounts of gvpA gene were equalized in each transfection mixture. In addition, pgvpA-IRES-mCherry plasmid without a start codon in front of gvpA was transfected into HEK293T cells to establish a negative control. A subset of these samples was then treated with Raptinal to induce apoptosis, establishing a positive control. Notably, expressing GvpA without its assembly factors led to high levels of apoptosis. While co-transfecting pgvpNJKFGW-EmGFP reduced some of the observed toxicity, the ACC/ACC SEMPER mARG transfected cells were healthier, with apoptosis levels indistinguishable from untreated negative controls. Taken together, these results suggest that the SEMPER mARG plasmid reduces cell toxicity by ensuring that assembly factors are consistently co-expressed with structural proteins in an appropriate ratio.

SEMPER enables recombinant monoclonal antibody expression from a single construct.

As a further demonstration of the generalizability and biotechnological utility of SEMPER, we used the system to express secreted recombinant monoclonal antibodies in HEK293T cells. As a model construct, we used the b12 antibody, which was discovered as part of early efforts to identify broadly neutralizing antibodies against HIV-1.^43,44 Contrary to the traditional approaches, the SEMPER-b12 plasmid was designed to express both the heavy and the light chains of the b12 IgG from a single mRNA transcript with tunable heavy chain (HC) to light chain (LC) stoichiometry (Figure 4A). IgG production was quantified from a sample of media supernatant via an enzyme-linked immunosorbent assay (ELISA).

Figure 4: Recombinant expression of monoclonal antibodies using SEMPER.

A) Architecture of an mRNA transcript transcribed from transfected SEMPER-b12 plasmid DNA. The first ORF, LC, encodes the b12 IgG light chain, while the second ORF, HC, encodes the b12 IgG heavy chain. Both polypeptides contain a signal peptide for secretion. The TIS of the first ORF, NNN, was varied while the TIS of the second ORF, ACC, remained constant in different constructs. B) Total human IgG ELISA results following expression and secretion of SEMPER-b12 vectors. The total amount of transfected DNA was held constant across all conditions. HEK293T cells transfected with either the light chain (ACC/***) or heavy chain (***/ACC) produced no detectable IgG—a similar result to the no plasmid transfection condition. SEMPER-b12 constructs with varying expression stoichiometry produced intermediate amounts of IgG with CCC/ACC producing the highest yield. Error bars indicate mean ± SEM (N=6 biological replicates). Statistical analysis was conducted using a one-way ANOVA (p < 0.0001) followed by Fisher's LSD post-hoc test with a single pooled variance to compare each treatment group with every other. Pairwise p-values are reported in Table S8.

The SEMPER system was able to produce high levels of functional IgG due to the ability to tune the ratio of HC to LC expression, reaching a maximum using the CCC/ACC SEMPER-b12 construct (Figure 4B). This production level was similar to that achieved by co-transfecting two separate plasmids (Figure S12). This finding demonstrates the ability of SEMPER to streamline the production of a valuable biologic product by simplifying the genetic engineering required for complex protein expression, while highlighting the importance of being able to tune component protein stoichiometry.

Demonstrating SEMPER 2-ORF efficacy from transfected synthetic mRNA.

With accelerating advancements in mRNA vaccine technology and growing development efforts in mRNA-based therapeutics,^13,45 we sought to demonstrate the efficacy of the SEMPER 2-ORF system to tune relative protein expression stoichiometries using mRNA-based expression systems (Figure 5A). Our in vitro transcribed (IVT) mRNA constructs do not include IRES-mCherry, as we observed reduced expression of our GOIs in IVT mRNA constructs containing this element (Figure S13).

Figure 5: 2-ORF SEMPER using in vitro transcribed mRNA.

A) Schematic of 2-ORF SEMPER mRNA produced from in vitro transcription. The first and second ORFs encode msfGFP[r5M] and mEBFP2 respectively. B) Flow cytometry data from transient transfection of IVT mRNA encoding various TIS combinations into HEK293T cells (N=1, representative of four biological replicates). IVT mRNA constructs were made with (Left) standard uridine triphosphate or (Right) N¹-methylpseudouridine-5’-triphosphate. C) Violin plots of log10(msfGFP[r5M]/mEBFP2) values for msfGFP[r5M] and mEBFP2 double-positive cells transfected with TTT/ACC, CCC/ACC, or ACC/ACC IVT mRNA produced with different uridine nucleotides. Each distribution represents measurements of at least 25,000 cells from four combined biological replicates. The median and quartiles of the distribution are represented by the solid and dotted lines respectively. Statistical analysis was conducted using a one-way Welch’s ANOVA (p < 0.0001 (Left) and p < 0.0001 (Right)) followed by Games-Howell’s multiple comparisons test to compare TIS combinations.

Through IVT and subsequent purification steps, we produced concentrated mRNA samples with the same TIS combinations and ORFs as tested for the plasmid-based 2-ORF SEMPER constructs. As nucleotide chemistry has been shown to impact translation efficiency,⁴⁶ we hypothesized that changing the nucleotide chemistry of our IVT mRNA may lead to changes in expression of our two ORFs. Therefore, we created a set of mRNA samples using standard uridine triphosphate (UTP) nucleotides in the IVT reaction, while another set was made using N¹-methylpseudouridine-5'-triphosphate (m1Ѱ), commonly used to manufacture mRNA for clinical use.⁴⁷ We then transiently transfected these mRNA samples into HEK293T cells and screened them using flow cytometry. Additionally, we transfected and collected flow cytometry data for two plasmid-based single-color controls for compensation.

Flow cytometry showed that some TIS combinations yielded unique relationships between msfGFP[r5M] and mEBFP2 fluorescence (Figure 5B-C). When using standard UTPs, the relationship for TTT/ACC was substantially different than those of CCC/ACC and ACC/ACC, demonstrating that multiple product stoichiometries are accessible from IVT mRNA SEMPER constructs. However, the specific expression ratios for CCC/ACC and ACC/ACC were not as distinct as those achieved using plasmid constructs, suggesting that further IVT-specific tuning is required, potentially in combination with additional mRNA engineering to improve expression efficiency (Figure S13). In mRNA constructs produced with m1Ѱ, the mEBFP2 vs msfGFP[r5M] relationships for TTT/ACC, CCC/ACC, and ACC/ACC were less distinct, confirming that nucleotide chemistry impacts translation initiation.

Basic performance of the SEMPER system is captured by a simple Monte Carlo model.

To provide insights on the essential SEMPER mechanism and facilitate future construct design, we developed a simple Monte Carlo simulation that captures the probabilistic nature of tunable, multi-ORF expression based on the LRS model. Briefly, for each Monte Carlo mRNA simulation, the k^th scanning ribosome can be consumed by one of three cases (i, ii, iii) as it traverses from the 5’ end to the 3’ end of the simulated transcript (Figure 6A). Implemented using sequential Boolean statements, the k^th scanning ribosome will first encounter i) a TIS in front of the first ORF. A Bernoulli random variable is then sampled with success probability p_TIS1 dictated by the strength of the TIS. This is equivalent to flipping a biased coin. If the instance of the random variable denotes successful initiation, the ribosome translates the first ORF, producing a single unit of Protein 1 and the simulation moves on to load the next k+1 scanning ribosome. However, if the instance of the random variable denotes non-initiation, the k^th scanning ribosome moves on to ii) the TIS sequence in front of the second ORF. Similarly, by sampling a Bernoulli random variable with success probability p_TIS2, the scanning ribosome will either translate a single unit of Protein 2, advancing the simulation to the k+1 iteration, or it will fail to initiate translation. In the absence of initiation, the k^th scanning ribosome is considered to iii) produce no translation product and the simulation moves on to load the next k+1 scanning ribosome. The aggregate totals of Protein 1 and Protein 2 from each simulated mRNA were then stored. The model does not implement IRES-mCherry expression as mCherry was only used for experimental normalization.

Figure 6: Simulating SEMPER-based gene expression.

A) Depiction of an individual Monte Carlo simulation unit referred to as a “Monte Carlo mRNA”. In each mRNA simulation, scanning ribosomes are sequentially loaded onto the mRNA. Each ribosome traverses the mRNA until it is consumed by one of three Boolean stopping conditions. i) The ribosome may successfully initiate translation of ORF 1 at the first TIS to produce a single unit of Protein 1. If successful, the simulation loads the next ribosome. If unsuccessful, the ribosome continues to the TIS in front of ORF 2 and ii) the ribosome may successfully initiate translation of ORF 2 at the second TIS to produce a single unit of Protein 2. If successful, the simulation loads the next ribosome. If unsuccessful, the ribosome is discarded iii) without producing any translation product. The probability the ribosome will translate a given ORF is modeled by a Bernoulli random variable with success probability p_TIS. p_TIS values are based on the identity of the TIS in front of the ORF. This process is repeated for multiple scanning ribosomes. B) Simulated flow cytometry plots for TIS combinations depicted in Figure 1C. For each TIS combination, hundreds of thousands of Monte Carlo mRNAs were simulated. The results of these simulations were aggregated into 10,000 smaller groups or “cells” and then plotted. C) Modeling expression changes in ORF 2 due to the presence of an internal methionine (iMet) in ORF 1. (Left) Simulated results of the expression distributions for cells transfected with plasmids utilizing TTT and ACC for ORF 1 and ORF 2 respectively. The left-shifted, light-blue distribution is caused by the consumption of ribosomes that initiate translation at the iMet, reducing ribosomal flux to ORF 2. 10,000 cells were simulated for each distribution. (Right) TTT/ACC constructs were tested experimentally, one with msfGFP[r5M] encoded in ORF 1 and one where an iMet was added back into a canonical position of msfGFP[r5M]. The presence of the iMet led to a reduction in ORF 2 translation, represented by mEBFP2 expression (N=1, representative of four biological replicates).

p_TIS success probabilities for ***, TTT, CCC, and ACC were chosen as 0.0001, 0.1, 0.5, and 0.9 respectively. p_TIS for *** was chosen to be nonzero to represent minute translation of functional protein initiated at in-frame, non-canonical start sites.⁴⁸ For each TIS combination depicted in Figure 1, we conducted nearly 500,000 Monte Carlo mRNA simulations. An average of 100 scanning ribosomes were simulated for each Monte Carlo mRNA simulation. Using standard resampling techniques, the simulated expression results of 100 mRNAs, on average, were then aggregated together into approximately 10,000 “single cells” to model transient transfection and then transcription of plasmid DNA. The results of our Monte Carlo simulations depicted in Figure 6B capture the major observed expression differences seen in Figure 1C caused by changing the TIS in front of the first ORF.

To extend our model’s ability to capture practical design rules, we simulated results for a TTT/ACC construct containing one internal methionine (iMet) within ORF 1. As seen in Figure 1B, internal methionines within ORF 1 reduced the flux of scanning ribosomes to ORF 2, thus shifting the expression of Protein 2 (mEBFP2) to lower values. To implement this in our simulation, we added an additional TIS (with p_TIS=0.5) between ORF 1 and ORF 2 in our model to represent an iMet within ORF 1. As expected, the presence of an iMet shifted the expression distribution of Protein 2 to lower values (Figure 6C, Left). For experimental validation, we added one canonical methionine back into our msfGFP[r5M] CDS, producing msfGFP[r4M]. The TIS created by the inclusion of this iMet was predicted to have a TIS strength similar to that of CCC based on quantitative results and modeling conducted by Noderer et al.³⁰ In HEK293T cells, we then compared two TTT/ACC plasmids, both of which contained mEBFP2 in ORF 2, while ORF 1 was varied between msfGFP[r5M] and msfGFP[r4M]. Consistent with modeling, the construct encoding msfGFP[r4M] yielded less mEBFP2 expression compared to the one without iMet (Figure 6C, Right), although differences in the magnitude of this effect suggest that future optimization of p_TIS values and other parameters in the model are needed to improve its quantitative prediction accuracy. For more detailed descriptions of model architecture, design choices, parameters, and assumptions, please see Methods.

DISCUSSION

Our results demonstrate that SEMPER enables tunable expression of multiple recombinant ORFs from single transcripts in mammalian cells by using leaky ribosomal scanning and TISs of varying strength. The SEMPER paradigm is applicable across cells from multiple species and can yield a range of user-tunable expression stoichiometries for at least three proteins on a single compact mRNA. Our results with SEMPER mARGs in GV production exemplify its application in encoding complex multi-gene constructs, allowing for precise modulation of expression ratios to optimize output while reducing cellular toxicity. Such tunability could have similar value in other scenarios demanding specific subunit/chaperone or toxin/antitoxin ratios, where co-transfection would result in a wider distribution of ratios. Additionally, the application of SEMPER to secreted mAb expression highlights the method’s versatility and potential utility in therapeutic production. Our results also show that the SEMPER framework can be applied directly to mRNA-based genetic circuits by demonstrating polycistronic expression with 2-ORF SEMPER mRNA constructs using IVT. Furthermore, we show that nucleotide chemistry can alter the translation levels of encoded ORFs in IVT mRNA, opening avenues to further tune protein stoichiometries with synthetic mRNA. The basic performance of SEMPER is captured by a relatively simple Monte Carlo model, which will help elucidate core design rules for future users.

While SEMPER has demonstrated tunability and rank-order relationships between TIS sets across multiple cell types, gene delivery methods, and protein outputs, the precise expression levels are context-dependent. Differences in these levels may arise from complex gene expression dynamics influenced by factors such as ORF length, RNA secondary structure, and RNA processing.⁴⁹ Nevertheless, the consistent rank order of expression ratio maintained across the set of TIS combinations tested in this work provides a convenient starting point for construct design, necessitating only limited empirical screening (which is required in most other synthetic biology approaches). The ability of SEMPER to provide tunability of polycistronic expression within a defined range is supported by our experiments with diverse ORFs, including fluorescent proteins, prokaryotic GVs, and the heavy and light chains of recombinant IgG.

The translation-based stoichiometry control provided by SEMPER complements transcription-based approaches that control relative expression using synthetic promoters and programmable transcriptional systems^7,8. SEMPER’s translation-focused approach allows tuning to be more compact and encodable at both DNA and mRNA levels. Combining the two classes of methods would enable the construction of gene circuits with greater complexity if needed for synthetic biology applications. SEMPER can also be combined with advanced circuits designed to overcome resource limitation and DNA copy number variability.^24,50

Although the genetic constructs described in this work should be immediately useful in a variety of contexts, future work is needed to demonstrate the SEMPER framework’s utility upon integration into the genome through stable transfection and viral transduction methods. Another future step is to make SEMPER compatible with methionine-containing proteins. Internal methionine codons create TISs that may undesirably consume ribosomal flux from downstream open reading frames. If mutating these methionines is not possible, an alternative approach may be to alter the nucleotide context surrounding the in-frame AUG using rationally chosen degenerate codons to effectively mask the TIS. Future work should also elucidate the maximum number of ORFs that are encodable using SEMPER at both the plasmid and mRNA levels. Additional work focused on IVT mRNA applications should examine the influence of different base chemistries, capping reagents, and untranslated regions. Lastly, future studies can further determine the mechanisms by which strong TIS sequences can enhance translation initiation at upstream and even downstream ORFs, as observed in our 3-ORF SEMPER screen with multiple ACC-containing ORFs.

We believe SEMPER represents a substantial advancement in the field of recombinant multi-protein expression and synthetic biology, enabling compact, user-friendly tuning of multiple proteins within mammalian systems. Researchers can readily adopt this framework to rapidly screen libraries relevant to expression and tuning of enzymatic pathways and circuits, assembly of multimeric protein structures, and other endeavors.⁵¹ As we have demonstrated in this study, this technology can enable simultaneous expression of a protein of interest and its folding chaperones from a single transcriptional unit at an optimal ratio, offering a strategy to tackle protein misfolding precisely. Moreover, as shown in some of the constructs used in this study, SEMPER can be used in combination with IRES and 2A elements for versatile encoding of more complex genetic constructs.

SEMPER also holds potential in the field of RNA vaccines and therapeutics. It could enhance the development of polyvalent RNA vaccines as well as the expression of mosaic virus-like particles from delivered mRNA, thereby broadening the immune response against highly variable viruses.^52,53 Further, it could improve RNA-delivered monoclonal antibody therapeutics by optimizing the ratio of heavy and light chain production in each cell, and by allowing for simultaneous production of multiple or bi-specific antibodies from a single mRNA.^54–56 This technology also paves the way for the production of cytokine cocktails through the co-expression of multiple cytokines from a single mRNA, which could offer synergistic effects to modulate immune responses. Taken together, SEMPER provides an option that can always be considered in developing polycistronic constructs for synthetic biology and medicine.

STAR★Methods

Resource availability

Lead contact

• Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, MGS (mikhail@caltech.edu).

Materials availability

• Plasmids generated in this study have been deposited to Addgene and are described in the Key Resources Table.

• This study did not generate new unique reagents.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Bacterial and virus strains
NEB Stable competent E coli	New England Biolabs	C3040H
Biological samples
Chemicals, peptides, and recombinant proteins
CleanCap AG Reagent	Trilink	N-7113
PEI-MAX	Polysciences	24765–2
Annexin V Pacific Blue stain	invitrogen	A35122
N1-Methylpseudouridine-5’-triphosphate	Trilink	N-1081
Raptinal	Med Chem Express	HY-121320
Critical commercial assays
HU IGG TOTAL COATED ELISA 96T	Life Technologies	BMS2091
HiScribe T7 High Yield RNA Synthesis Kit	New England Biolabs	NEB # E2040
Lipofectamine MessengerMAX Reagent	Invitrogen	#LMRNA008
Deposited data
Experimental models: Cell lines
HEK293T	ATCC	CLR-3216
CHO-K1	ATCC	CCL-61
Experimental models: Organisms/strains
Oligonucleotides
Primer: Forward primer to generate linear IVT templates. CGGCCGCTAATACGACTCACTATAAGGGTCAGATCGC	This paper	Referred to as p101
Primer/Ultramer: Reverse primer to generate linear IVT templates. TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTGCTAGCTCCAGGGTGTGG	This paper	Referred to as p54
Recombinant DNA
pCMV-SEMPER_ACC-msfGFP[r5M]_ACC-mEBFP2_IRES-mCherry	This paper	Addgene (#221075)
pCMV-SEMPER_GGG-msfGFP[r5M]_ACC-mEBFP2_IRES-mCherry	This paper	Addgene (#221076)
pCMV-SEMPER_CCC-msfGFP[r5M]_ACC-mEBFP2_IRES-mCherry	This paper	Addgene (#221077)
pCMV-SEMPER_TTC-msfGFP[r5M]_ACC-mEBFP2_IRES-mCherry	This paper	Addgene (#221078)
pCMV-SEMPER_TTT-msfGFP[r5M]_ACC-mEBFP2_IRES-mCherry	This paper	Addgene (#221079)
pCMV-SEMPER-mARG_ACC-gvpA_ACC-gvpNJKFGW-EmGFP_IRES-mCherry	This paper	Addgene (#221080)
pCMV-SEMPER_ACC-msfBFP[r5M]_ACC-msfGFP[r5M]_ACC-emiRFP670_IRES-mCherry	This paper	Addgene (#221081)
pT7-IVT_ACC-msfGFP[r5M]_ACC-mEBFP2	This paper	Addgene (#221082)
Software and algorithms
FlowJo 10	Becton Dickinson & Company (BD)
Matlab 2021	Mathworks
Prism 10	Graphpad
Vantage 4.6	Verasonics
Code for generating simulation results	This paper	10.5281/zenodo.11136551
Other

Data and code availability

• Data: All data reported in this paper will be shared by the lead contact upon request.

• Code: All original code has been deposited at Zenodo and is publicly available as of the date of publication. DOIs are listed in the key resources table. Original code is also available through https://github.com/shapiro-lab/semper-simulation.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental Model and Study Participant Details

Cell lines

HEK293T (female human embryonic kidney) cells (American Type Culture Collection (ATCC), CLR-3216) and CHO-K1 (female, Chinese hamster ovary) cells (American Type Culture Collection (ATCC), CCL-61) were cultured in 24-well plates at 37 °C and 5% CO₂ in a humidified incubator in 0.5 mL of DMEM (Corning, 10–013-CV) with 10% FBS (Gibco) and 1X penicillin–streptomycin.

Method Details

Vector construction

Plasmids were constructed using standard cloning techniques, including Gibson assembly and conventional restriction and ligation. All final sequences were verified using whole-plasmid sequencing through Primordium Labs. Plasmids containing two or three fluorescent protein ORFs were constructed in the following way: individual FP sequences were ordered from IDT or TWIST Biosciences as synthetic gBlocks. Modified TIS sequences and the *** sequence were introduced using overhang PCR primers.

Finally, all components were subcloned into pCMV-Sport-gvpA-IRES-mCherry using NEB HiFi DNA Assembly replacing the gvpA ORF. Some modifications were made to the 5’ UTR sequences. Plasmid sequences for the 2-ORF TTT/ACC construct and the 3-ORF ACC/***/*** construct as well as sequences for individual genetic parts in these plasmids are provided in Table S9 to aid in reproduction of this work. A subset of plasmids used in this work are available through Addgene (see Key Resources Table).

SEMPER mARG plasmids containing gvpA and gvpNJKFGW-EmGFP were constructed as follows: the ORF containing gvpNJKFGW-EmGFP was PCR amplified and subcloned into pCMV-Sport-gvpA-IRES-mCherry using NEB HiFi DNA Assembly between gvpA and IRES. Different gvpA TIS sequences were introduced using single-stranded oligo bridge primers with NEB HiFi Assembly.

To produce plasmids amenable for in vitro transcription of mRNA, sequences containing the 5’ UTR through the stop codon of the mEBFP2 ORF from the 2-ORF SEMPER plasmids were cloned into an IVT plasmid backbone containing a T7 promoter, a synthetic 3’ UTR sequence, and a 100 bp polyA track (Table S9) using PCR amplification and Gibson assembly methods.

Production of mRNA by in vitro transcription

Linear templates were PCR amplified from the IVT plasmids (described in Vector construction) using primer p101 and “Ultramer” p54 synthesized by Integrated DNA Technologies. Primer p101 introduced an AG dinucleotide following the T7 promoter sequence on linear templates to ensure efficient 5’ capping of IVT mRNA. Linear templates were purified using standard gel electrophoresis and DNA cleanup methods. IVT mRNA was produced using the HiScribe T7 High Yield RNA Synthesis Kit (NEB # E2040) along with 500 ng of linear template and 4 mM CleanCap AG Reagent (Trilink, N-7113). 5 mM N¹-Methylpseudouridine-5’-triphosphate (Trilink, N-1081) was substituted for the standard uridine triphosphate included in the HiScribe Kit for certain reactions. Following the IVT incubation, reactions were additionally incubated with 2 units of DNAse I to remove remaining DNA template. Purification of IVT mRNA was conducted using the Monarch RNA Cleanup Kit (NEB #T2040L). To confirm IVT mRNAs were of the correct lengths, aliquots of purified IVT mRNA were denatured at 70°C for 15 minutes and then subjected to gel electrophoresis on a 1% agarose gel in Tris acetate EDTA (TAE) stained with SYBR Safe. ssRNA ladder (NEB #N0362S) was used to confirm band sizes. mRNA concentrations for each IVT mRNA were quantified using a Qubit Fluorometer (Invitrogen). Completed IVT mRNA samples were stored at −80°C.

Cell culture and transfection

HEK293T cells (American Type Culture Collection (ATCC), CLR-3216) and CHO-K1 cells (American Type Culture Collection (ATCC), CCL-61) were cultured in 24-well plates at 37 °C and 5% CO₂ in a humidified incubator in 0.5 mL of DMEM (Corning, 10–013-CV) with 10% FBS (Gibco) and 1X penicillin–streptomycin until about 80% confluency before transfection with plasmid DNA or IVT mRNA.

For plasmid DNA transfection, transient transfection mixtures were created by mixing 500 ng of plasmid with polyethyleneimine (PEI-MAX, linear 40 kD #24765–2, Polysciences) at 4.12 μg of polyethyleneimine per microgram of DNA in 150 mM NaCl. The mixture was incubated for 12 minutes at room temperature and added dropwise to HEK293T or CHO-K1 adherent cells. Media was changed after 12–16 hours and daily thereafter. Cells were analyzed with flow-cytometry 48 hours post-transfection. Deviations from this protocol are described in the supplementary information where applicable.

For IVT mRNA transfection, transient transfection mixtures were created by mixing 500 ng of mRNA, 1.5 μL of Lipofectamine MessengerMAX Reagent (Invitrogen #LMRNA008), and 50 μL of Opti-Mem media (Gibco #31985070) according to the Lipofectamine MessengerMAX Reagent standard operating procedure. The mixture was incubated for 5 minutes at room temperature and added dropwise to HEK293T cells. Media was changed after 12–16 hours and daily thereafter. Cells were analyzed with flow-cytometry 48 hours post-transfection.

For transfection of plasmids encoding mARGs, transient transfection mixtures were created as above except that 600 ng of total DNA was mixed together as follows: 56 fmol of gvpA-IRES-mCherry plasmid with or without 14 fmol gvpNJKFGW-EmGFP plasmid or 56 fmol of a 2-ORF SEMPER mARG plasmid. Plasmid mixtures were normalized with addition of pUC19 up to 600 ng before complexing with PEI-MAX. The mixture was incubated for 12 minutes at room temperature and added dropwise to HEK293T adherent cells. Media was changed after 12–16 hours and daily thereafter. Cells were harvested 3-days post-transfection.

Expression of b12 IgG in HEK293T cells

HEK293T cells were plated at a density of 200,000 cells per well in 48-well plates using 250 μL of Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1X penicillin/streptomycin, 20 mM HEPES (pH 7.0), 1X MEM Non-Essential Amino Acids (NEAA), and 1X GlutaMAX™. The cells were cultured at 37°C in a humidified atmosphere containing 5% CO₂ for 24 hours prior to transfection.

A DNA-polyethylenimine (PEI) Max transfection mixture was prepared by dissolving 206 mg of PEI-MAX (40 kDa, Polysciences, #24765–2) in 500 mL of tissue culture (TC)-grade phosphate-buffered saline (PBS) to achieve a final concentration of 0.412 mg/mL (10.3 μM). The solution was adjusted to pH 7.0 with NaOH and filter-sterilized. Aliquots were stored at −80°C until needed. For each transfection, 350 ng of the SEMPER-b12 plasmid (7485 bp) was combined with 17.5 μL of a sterile 150 mM NaCl solution. Separately, the PEI-Max solution was diluted with 150 mM NaCl in a 1:4 ratio, and 17.5 μL of this dilution was added to the plasmid-NaCl mixture. The combined mixture was gently mixed and incubated at room temperature for 12 minutes to allow complex formation, after which it was added dropwise to each well. For the two-plasmid control, 175 ng of heavy chain (HC) and light chain (LC) plasmids were combined to total 350 ng and then complexed with the PEI-Max solution as described above.

The media in the wells were replaced after 16 hours with fresh DMEM. An additional 250 μL of the same medium was supplemented after another 24 hours, bringing the total volume in each well to 500 μL. Three days post-transfection, 5 μL of the cell culture supernatant was transferred to 95 μL of assay buffer.

IgG yields were quantified using the Human Total IgG Coated ELISA Kit (Invitrogen, #BMS2091), adhering to the manufacturer’s protocol. Absorbance was measured at 450 nm with a reference wavelength of 620 nm using the Tecan SPARK plate reader.

Flow cytometry and apoptosis assays

To harvest the cells, cells were dissociated using trypsin/EDTA and centrifuged at 300 g for 6 minutes at room temperature. For experiments involving mARGs, 2-ORF SEMPER plasmid transfections, and 2-ORF SEMPER mRNA transfections, cells were analyzed with MACSQuant VYB (Miltenyi Biotec) except for data depicted in Figure 6C (Right) which was gathered using a MACSQuant10 (Miltenyi Biotec). For experiments involving 3-ORF SEMPER, cells were analyzed using Cytoflex S (Beckman Coulter). Single-color controls were used to allow for compensation. In plasmid transfection experiments, cells were gated for size and doublet-discriminated before being gated and further binned by mCherry fluorescence. For IVT mRNA transfection experiments, cells were gated as above except they were gated for msfGFP[r5M] and mEBFP2 double-positivity instead of by mCherry fluorescence.

For the Annexin V apoptosis assay, we transfected cells as described above but did not perform media changes. For the positive control, cells transfected with pgvpA-IRES-mCherry (w/ start codon removed in front of gvpA) were treated with 10 μM Raptinal 18 hours before harvest. Cells were harvested two days after transfection. To do this, we collected supernatant from cells to recover the dead cells. We then trypsinized the adhered cells and added this trypsinized cell fraction to the original supernatant for maximum cell recovery. Then, we spun the cells down at 300 g for 6 minutes, carefully removed the supernatant, and resuspended the cells in 90 μL of cold annexin binding buffer with no EDTA and supplemented Ca²⁺, followed by 10 μL of Annexin V Pacific Blue stain (A35122). Following a 15-minute incubation, we added 150 μL of the same annexin binding buffer and then performed flow cytometry to obtain fluorescence values for the pacific blue stain. We gated in FlowJo by selecting the largest population in the FCS/SSC plot, keeping the lower left corner in the cell gate to include smaller apoptotic cells. We selected a 10⁴ threshold for the Pacific Blue Annexin V stain, corresponding to the right tail of the unstained control so that the Annexin V+ population of the unstained control was around 0%. Values were normalized by setting the mean value of Raptinal-treated samples to 100%.

In vitro ultrasound imaging of HEK293T cells

Cells were harvested as described above. Cells were resuspended with 1% low-melt agarose (GoldBio) in PBS at 40°C at concentrations of ~15 million cells per milliliter and then loaded into wells of pre-formed phantoms consisting of 1% molecular biology-grade agarose (Bio-Rad) in PBS. Phantoms were imaged using L22–14vX transducer (Verasonics) at 15.625 MHz while submerged in PBS on top of an acoustic absorber pad. For BURST imaging, wells were centered around the 8 mm natural focus of the transducer and a BURST pulse sequence was applied in pAM acquisition mode with the focus set to 8 mm, and the voltage was set to 2V for the first 10 frames and 15V for the remaining frames.

BURST images were produced by pixel-wise subtraction of the 11th (collapse) frame from the 54th (post-collapse) frame. Resulting differences were divided by 100. Images were quantified as follows: The sample ROIs were drawn inside the well of the agarose phantom. Average pixel value inside the ROI was calculated for each replicate.

Simulations

The RAM and CPU power of a standard laptop computer were sufficient to run the simulations depicted in Figure 6. Simulation code along with example usage and a list of dependencies are accessible through GitHub: https://github.com/shapiro-lab/semper-simulation and through the following DOI: 10.5281/zenodo.11136551.

The model assumes scanning ribosomes (also referred to below as just “ribosomes”) load at the 5’ end of an mRNA transcript and will only move across the mRNA from 5’ to 3’. The p_TIS values for ***, TTT, CCC, and ACC were estimated based on our observed results and the results of Ferreira et al. and Noderer et al.^23,30 Estimates of these relative p_TIS values can be made for other TIS sequences through analysis of the predicted and measured strengths of TISs made by Noderer et al. We assume there is inherent stochasticity with the number of scanning ribosomes that will traverse an mRNA and the number of mRNA that may exist in a cell. This stochasticity was captured by sampling these values from probabilistic distributions. Specifically, normality was used to describe these processes. The model does not consider the sequence identity of the ORFs other than the TIS sequence in front of it. The model does not account for ribosomes falling off the transcript during scanning, the length of the transcript, the length of the 5’ UTR, local structure of the mRNA, or global structure of the mRNA.

The architecture of the modeling pipeline is described below:

1. Choosing the number of mRNA to model. The number of unique mRNA simulation results needed to fill the desired number of cells to be modeled is estimated.

2. Estimating the number of ribosomes that will traverse each mRNA. Once the number of mRNA are determined, the number of ribosomes that will traverse each mRNA are sampled from a normal distribution. We sampled from a normal distribution with mean and standard deviation both equal to 100. The mean value of 100 was motivated by estimates of the number of observed translation events from single CD147 mRNAs in mammalian cells.⁵⁷

3. Simulating ribosomes traversing an mRNA. Each mRNA then undergoes a Monte Carlo simulation where R ribosomes will traverse the mRNA. To do this we iterate through a set of Boolean conditions k=1:R times, as described in Results. We say the k^th ribosome has been loaded onto the 5’ end of the mRNA when the k^th iteration of the loop begins.

4. Loading/aggregating mRNA simulation results into cells. Once all mRNAs have been simulated, the results of individual mRNA simulations are aggregated together as a proxy for transient transfection and implicit transcription of plasmid-DNA. Here, the number of individual mRNA simulation results that are aggregated together is sampled from a normal distribution with mean 100 and standard deviation of 100 to capture the stochasticity of transient transfection. This mean value is motivated by observed endogenous CD147 mRNA counts in single cells. Resampling of the mRNA simulations was used in this step as mRNA simulations were computationally expensive to calculate. Specifically, from ~500,000 individual mRNA simulations, 2,500 cells were “loaded” with aggregated results. The simulations were then randomized, and another 2,500 cells were loaded. This process was conducted four times to yield 10,000 simulated cells per condition, as plotted in Figure 6B.

The model can readily be used to consider systems with more than two ORFs by adding additional p_TIS values in the input parameters. The addition of a p_TIS can be considered as the addition of a desired ORF or an internal methionine within an ORF. The latter case was considered to produce Figure 6C, as showcased in the publicly available code. The model does not differentiate between a TIS that may exist at the canonical 5’ end of an ORF or a TIS that may be encoded in the middle of an ORF.

Quantification and Statistical Analysis

All statistical analysis was performed in GraphPad Prism 10. Pairwise statistical analysis, number of datapoints, and statistical methods used can be found in figure legends and in the referenced tables provided in the Supplementary Information. As used throughout the text, we define a single biological replicate as the transient transfection of cells with DNA or mRNA in a single culture well.

HIGHLIGHTS.

• SEMPER enables the tunable expression of multiple proteins from a single transcript

• Up to three fluorescent proteins expressed in various cell lines

• Single-transcript expression of gas vesicles and recombinant monoclonal antibodies

• Probabilistic modeling supports SEMPER's underlying mechanisms