Engineering


Introduction

We set out to create a novel therapeutic for treating glioma and have decided on an mRNA-based approach to selectively kill tumor cells by expressing a toxic protein. This requires some kind of mechanism to differentiate between healthy cells and tumor cells, and to limit the expression of the toxic payload accordingly to the cell type. To achieve this, we utilized the DART VADAR technology to create mRNA sensors that couple the detection of specific mRNA molecules to the conditional translation of a payload gene.1

After deciding on the mode of action, we had to find suitable targets for the sensors, i.e. glioma-specific mRNA transcripts that could be detected by our therapeutic mRNA. Since there are different types of glioma with different expression profiles, we decided on two different mRNA molecules as targets after a thorough literature research: the wild-type transcript of the epidermal growth factor receptor (EGFR), which is commonly overexpressed in glioblastoma, and the mutant IDH1(p.R132H) transcript of the isocitrate dehydrogenase 1 gene. This mutation is observed in about 90% of astrocytoma and oligodendroglioma cases.2,3 We obtained consensus mRNA sequences of both the human EGFR and the IDH1 transcripts from the NCBI RefSeq database (accession numbers NM_005228.5 and NM_005896.4), and modified the IDH1 transcript accordingly to carry the p.R132H mutation (CGT>CAT at codon 132).2


Engineering the pASTERISK expression plasmids

We envisioned our therapeutics to be transfectable mRNA packaged inside lipid nanoparticles or liposomes. Hence, we had to find a way to produce our mRNA therapeutics in a way that would be applicable in a real pharmaceutical production setting. We discussed the bulk expression of mRNA in E. coli with subsequent purification due to the low cost of this approach, but dismissed it because of the risks of introducing too many impurities. Therefore, we opted for in vitro transcription as a means of mRNA production instead.

Our first task was to create a suitable plasmid containing the open reading frame (ORF) of our therapeutic construct under the control of a T7 promoter, so it could be linearized and transcribed in vitro using T7 polymerase. We used the published DART VADAR plasmid as a starting point for designing our own plasmid, that we termed pASTERISK. For the backbone, we chose the commonly used pSB1C3 plasmid. We now had to design the ORF that would encode our therapeutic construct under the control of a T7 promoter, minus the sensor sequence, because these should be cloned into the plasmid afterwards for each target. Figure 1 shows the final design of this construct. In fact, we designed to versions of this construct, called ASTERISK-HBA1 and ASTERISK-HBB, which we provide as BioBrick parts to the parts registry. The difference between these lies in the 5'- and 3'-untranslated regions (UTRs). ASTERISK-HBA1 contains UTRs from the human alpha globin 1 gene (HBA1), while ASTERISK-HBB contains UTRs from the human beta globin gene (HBB). We decided to try out two different UTR combinations, with the goal of improving mRNA stability. These UTRs were chosen due to their use in mRNA vaccines and stabilizing effects on mRNA.4


ASTERISK-HBA1 / ASTERISK-HBB + T7 promoter and terminator
Figure 1: Circuit diagram of the ASTERISK-HBA1 and ASTERISK-HBB parts. Both of these constructs were cloned into pSB1C3 plasmids.


We designed the ASTERISK-HBA1 and ASTERISK-HBB parts by using the published DART VADAR ORF as a starting point, and introduced many changes and modifications. The ORF starts with either an HBA1 or HBB 5'UTR, followed by a Kozak sequence and an mCherry transfection marker that we substituted for the original TagBFP marker. A so-called 2A peptide (a 'self-cleaving peptide') separates the next segment of the ORF from the transfection marker by interrupting the peptide linkage during ribosomal translation. What follows is the sensor cloning site, in which a 123 bp sensor sequence can be cloned into by HindIII digestion followed by Gibson assembly. Another 2A peptide separates the cloning site from the payload sequence. We introduced additional SphI and SalI restriction sites flanking the payload sequence, so that it can be easily changed and modified. We used the original mNeonGreen marker as a surrogate payload for our experiments. Another 2A peptide isolates the payload from the final coding sequence, MCP-ADAR2dd. This fusion protein, consisting of the MS2 bacteriophage major coat protein (MCP) and the deaminase domain of ADAR2 (ADAR2dd), is responsible for driving the positive sensor activation feedback loop that characterizes the DART VADAR circuit. Finally, we replaced the original 3'UTR with an HBA1 or HBB 3'UTR and removed the PolyA-signal, because we wanted to add the PolyA-tail in a subsequent step after IVT. Throughout the entire original ORF, there were several illegal restriction sites that we silently mutated for RFC10 BioBrick conformity.


pASTERISK-HBA1 plasmid map
Figure 2: Plasmid map of the pASTERISK-HBA1 plasmid. The plasmid was constructed by cloning the ASTERISK-HBA1 part into a pSB1C3 backbone.


Both versions of these plasmids – pSB1C3 containing either the ASTERISK-HBA1 or ASTERISK-HBB part – were successfully cloned, giving rise to the pASTERISK-HBA1 and pASTERISK-HBB plasmids, respectively (Figure 2). We now had suitable plasmids for cloning our sensor inserts into and for producing therapeutic mRNA using IVT. The next step was to design the sensor inserts.


Engineering the EGFR sensors

In order to selectively kill glioblastoma cells that are overexpressing the EGFR gene, the fraction of activated sensors has to be high enough in EGFR-overexpressing cells to induce lethality, but low enough in healthy cells to leave them unharmed. Assuming that higher levels of EGFR expression lead to a higher fraction of activated sensor molecules, which in turn leads to a higher amount of toxic protein expression, the tumor cells should ideally cross the lethality threshold due to their EGFR overexpression, while healthy cells do not.


Good and bad EGFR sensor features
Figure 3: Schematic explanation of good and bad EGFR sensor features. A bad EGFR sensor would be either too insensitive and thus neither healthy nor cancer cells would be killed by the toxic payload, or too sensitive, which means both cell types would be killed. A good EGFR sensor responds to EGFR overexpression with lethality, but is not lethal at normal expression levels.


First, we wanted to find out how we could tweak the different properties of the sensors to change their dynamic range and efficiency, that is, their propensity to become activated on hybridization to their target mRNA. Ideally, we wanted to find ways to lower the sensor efficiency. This seems counterintuitive, but as explained, the sensor should behave in a way that only a small fraction gets activated at wild-type EGFR expression levels. If the sensor was not efficient enough to kill the tumor cells, one could increase the dosage or choose a more potent toxin as payload.

We chose three variables that would be independently varied to observe their effect on the sensor efficiency:

  • The type of central nucleotide triplet targeted by the sensor
  • The amount of matching/mismatching bases between the sensor and the trigger sequence
  • The amount of in-frame stop-codons that had to be edited by ADAR

This led to the design of four EGFR sensors, as shown in table 1. Using our DVSensor software, we first generated all possible sensors against the EGFR transcript in the 3'UTR region. We picked two sensors, EGFR_CAA5991 (targeting a CAA triplet) and EGFR_CCA7127 (targeting a CCA triplet), to compare the effect of targeting different triplets (CAA vs CCA). Next, we chose EGFR_CCA7127 as a template for designing two more sensors, EGFR_CCA7127_MM and EGFR_CCA7127_DS. In the EGFR_CCA7127_MM sensor, 20 mismatches were introduced to investigate the effect of reducing the affinity of the trigger to the sensor. The EGFR_CCA7127_DS sensor contains an additional stop-codon which is in-frame with the central stop-codon, which means an additional editing step by ADAR is required to activate the sensor.

Importantly, all four sensors that were designed in this step have the same GC-content of 52.8%, which eliminates GC-bias when comparing the sensor efficiencies, and had no other in-frame start- or stop-codons that had to be removed.


Table 1: EGFR sensors designed in the first engineering cycle iteration. The two sensors EGFR_CAA5991 and EGFR_CCA7127 were generated using our own software. EGFR_CCA7127 was used as a template for designing two more sensors, EGFR_CCA7127_DS and EGFR_CCA7127_MM, with additional stop-codons and mismatches, respectively. Depicted are the cDNA sequences of the different EGFR sensors hybridized to their respective trigger sequences (portions of the EGFR transcript).
Sensor Features Diagram
EGFR_CAA5991 Targets the CAA 5991 triplet in the 3'UTR. Sensor EGFR_CAA5991
EGFR_CCA7127 Targets the CCA 7127 triplet in the 3'UTR. Sensor EGFR_CCA7127
EGFR_CCA7127_DS Targets the CCA 7127 triplet in the 3'UTR, with an additional stop-codon opposite to a CAA triplet. Sensor EGFR_CCA7127_DS
EGFR_CCA7127_MM Targets the CCA 7127 triplet in the 3'UTR, with 20 mismatching bases introduced. Sensor EGFR_CCA7127_MM


Designing the mutant IDH1 sensor

The common p.R132H hot spot mutation in the IDH1 gene is caused by a CGT>CAT codon change. Importantly, this change leads to the introduction of a novel UCA triplet in the mRNA which can be efficiently targeted using an appropriate sensor (see Fig. 4). Therefore, we set out to construct a sensor targeting this novel triplet in the mutant IDH1 transcript. The sensor should in theory only get activated in cells carrying the mutation, thereby ensuring the necessary tumor specificity. As has been demonstrated by Qu et al., the UCG triplet, which is present in the wildtype transcript, has a neglectable efficiency for inducing an A-to-I editing event if it lies opposite to a UAG stop codon.5 The UCA triplet in the mutant transcript however, efficiently induces A-to-I editing and thereby would activate the sensor.5 Figure 5 shows our final design of the mutant IDH1 sensor.


Wildtype and mutant p.R132H IDH1 transcripts
Figure 4: Sections of the the wildtype and the mutant IDH1 transcript with the G>A point mutation. The mutation changes a UCG triplet into a UCA triplet, which can efficiently activate a sensor by inducing A-to-I editing of a UAG stop-codon.


Mutant IDH1 sensor design
Figure 5: The final design of the mutant IDH1 sensor. Depicted is the cDNA sequence of the sensor hybridized to the target region of the mutant IDH1 transcript. A total of 3 in-frame start- and stop-codons other than the central stop-codon had to be removed by changing a base, which causes the mismatches with the trigger sequence.


Putting it all together

After having successfully cloned pASTERISK-HBA1 and pASTERISK-HBB, we ordered all of our designed sensor sequences as oligonucleotides. These could now be cloned into the plasmids using a HindIII digestion at the sensor cloning site, followed by Gibson assembly to introduce the sensor insert. The plasmids obtained this way could then be linearized and used as IVT templates for producing our therapeutic mRNA.


Engineering results

In order to analyze the properties of the HBA1 and HBB UTRs in isolation, both plasmids (pASTERISK-HBA1 and pASTERISK-HBB) were used as templates for in vitro transcription using T7 polymerase with an empty sensor cloning site (i.e. without any sensor sequences). The mRNA was purified and used to transfect HEK293T cells. We expected to see both encoded fluorescence markers (mCherry and mNeonGreen) to be expressed by the cells, since there was no sensor insert that would prevent the payload mCherry sequence expression. Figure 6-8 show the results 8h, 24h and 48h after transfection.



Figure 6: Fluorescence microscopy images of transfected HEK293T cells after 8 hours.

Figure 7: Fluorescence microscopy images of transfected HEK293T cells after 24 hours.

Figure 8: Fluorescence microscopy images of transfected HEK293T cells after 48 hours.


  • 1. Gayet, R. V. et al. Autocatalytic base editing for RNA-responsive translational control. Nat Commun 14, 1339 (2023).
  • 2. Horbinski, C. What do we know about IDH1/2 mutations so far, and how do we use it? Acta Neuropathol 125, 621–636 (2013).
  • 3. Lassman, A. B. et al. Epidermal growth factor receptor (EGFR) amplification rates observed in screening patients for randomized trials in glioblastoma. J Neurooncol 144, 205–210 (2019).
  • 4. Kim, S. C. et al. Modifications of mRNA vaccine structural elements for improving mRNA stability and translation efficiency. Mol Cell Toxicol 18, 1–8 (2022).
  • 5. Qu, L. et al. Programmable RNA editing by recruiting endogenous ADAR using engineered RNAs. Nat Biotechnol 37, 1059–1069 (2019).