Results

In our first attempt to work with ASOs in the lab, we designed 18 ASOs of varying target regions in the GFP mRNA to investigate efficiency in mammalian cells expressing GFP (HEK293-GFP cell line). This helped us understand ASO activity and refine our design strategy for future experiments. The ASO sequences are available here. All ASO’s were ordered from IDT - one of iGEM’s official sponsors. These ASOs did not contain any chemical modifications (Figure 1).

No significant reduction in GFP expression was detected for any of the 18 unmodified ASOs tested (Figure 2). Likewise, the scrambled controls and the designated positive control failed to elicit significant silencing activity. These findings suggest that the lack of knockdown may reflect limitations in ASO uptake, intracellular stability, or effective concentration under the experimental conditions.

Conclusion and Next Steps

The lack of detectable GFP knockdown may be attributed to one or more of the following factors:

1. Inefficient ASO transfection.
2. Suboptimal ASO concentration insufficient to elicit measurable gene silencing.
3. Rapid ASO degradation due to the absence of chemical modifications.

To investigate these possibilities, our next iteration focused on:

Testing transfection efficiency directly using a fluorescently labeled oligonucleotide.

We transfected GFP-expressing HEK293 cells with chemically unmodified and 5’-ROX-labeled ASOs using Oligofectamine. Flow cytometry confirmed efficient cellular uptake of the ASOs, as shown by the ROX signal (Figure 3). Despite successful delivery, GFP expression remained unchanged, indicating that the ASOs reached the cytoplasm but did not induce target degradation. These findings suggest that unmodified ASOs are insufficiently stable or active within cells, and that chemical modifications (e.g., 2′-MOE, phosphorothioate) will be required to achieve effective silencing.

To validate antisense oligonucleotide activity in our system, we tested a chemically modified ASO targeting the long non-coding RNA MALAT1 (sequence: GGCATATGCAGATAATGTTC; 2′-MOE modifications on five bases at both ends; full phosphorothioate backbone; ordered from IDT). This ASO was supplied as a validated industry control by IDT [2]. We performed this experiment since the lack of chemical modifications in our initial ASOs might explain their limited activity.

Using the same Oligofectamine transfection conditions applied in our GFP experiments, we transfected HEK293 and A549 cells with the chemically modified MALAT1-targeting ASO. After incubation, total RNA was extracted, reverse-transcribed to cDNA, and MALAT1 expression was quantified by RT–qPCR relative to a housekeeping gene (β-actin/ACTB) (Figure 4).

The MALAT1 ASO produced a significant reduction in MALAT1 levels in both cell lines (Figure 5), confirming its activity and validating our transfection.

Conclusions

The antisense oligonucleotide effectively suppressed MALAT1 expression in both cell lines, confirming functional delivery and the importance of chemical modifications for ASO activity. In HEK293 cells, knockdown was dose-dependent, with the strongest reduction observed at 250–750 nM. In A549 cells, significant suppression occurred at 5–250 nM, but the effect declined slightly at 750 nM, suggesting possible saturation or compensatory mechanisms at higher concentrations. Control treatments (Lipofectamine only, untreated) showed no reduction, verifying the specificity of the response. These results established A549 as the preferred model for subsequent experiments.

After confirming that our protocols were properly calibrated with the reference MALAT1 ASO, we applied our computational design pipeline (see our model section) to generate three novel ASOs targeting MALAT1 mRNA (Table 1). These candidate ASOs were ordered from IDT (2′-MOE modifications on five bases at both ends; full phosphorothioate backbone) and subsequently tested in A549 cells under the same transfection conditions, using the same protocol as previously described in Figure 4. This setup allowed us to directly evaluate their knockdown efficiency relative to the validated control.

Table 1: Sequences of MALAT1-Targeting ASOs, Positive Control [2], and Scrambled Control

Transfection of A549 cells with the chemically modified reference MALAT1 ASO resulted in a strong and consistent reduction of MALAT1 expression compared to untreated, Lipofectamine-only, and scrambled controls. Notably, ASO2 and ASO3, computationally designed using our modeling pipeline, achieved even greater knockdown efficiency than the commercial reference ASO (Figure 6).

Conclusion

This experiment demonstrates that our computational system reliably predictsASO that could mediate highly efficient knockdown. The validated commercial MALAT1 ASO confirmed assay robustness, while all of our computationally designed ASOs demonstrated effective silencing, with two of them (ASO2 and ASO3) achieving even stronger silencing of MALAT1 than the commercial reference [2]. These results establish proof-of-concept that computational design can yield highly effective ASOs and support further exploration of this approach.

Related sequences and registry information are available on our Parts page.

Following our initial screening of MALAT1-targeting ASOs, we aimed to enhance their silencing efficiency by applying our improved computational prediction model. The new model included many new features pre-selection, such as codon bias, RNase-related, and off-target features, with a different subset ultimately selected. Feature selection aimed to maximize NDCG performance on unseen cell lines, and the ranking model placed heavier emphasis on top performers, reducing the NDCG evaluation from all samples to the top 200 per group.

To evaluate whether the new model produced more effective candidates, we designed a second generation of ASOs (V2) targeting the same gene as our first iteration (V1) and compared their performance to both the standard reference MALAT1 ASO and a scrambled (SCR) control (Table 2). A549 cells were transfected with each ASO, and MALAT1 expression was quantified by RT–qPCR 24 hours post-transfection (Figure 7).

Table 2: Sequences of MALAT1-Targeting ASOs, Positive Control [2], and Scrambled Control

Conclusion

The second-generation ASOs (V2) designed with our updated model did not outperform the first-generation candidates (V1) or the reference MALAT1 ASO. Knockdown levels were comparable at best, and in some cases lower, indicating that the current model update did not translate into improved potency under our assay conditions. The second model performed worse on ASO inhibition, likely due to several factors. It may have overfit to melanoma and lymphoma cell lines after liver lines were removed—an intended step to improve biological relevance, since melanoma is closer to NSCLC—but this reduced diversity likely amplified dataset-specific biases. Although extensive feature selection was applied, it may not have been optimal. The model might have placed greater emphasis on newer RNase-related features, which may not generalize to A549, and residual experimental noise could have further obscured real signal.

Related sequences and registry information are available on our Parts page.

After confirming that our protocols were properly calibrated with the reference MALAT1 ASO, and further validating our computational model with MALAT1-targeting ASOs, we applied the same design pipeline to generate three novel ASOs targeting GFP mRNA (Table 3).

Table 3: Sequences of GFP-Targeting ASOs, Scrambled Control and ref ASO [1].

These candidate ASOs were ordered from IDT (2′-MOE modifications on five bases at both ends; full phosphorothioate backbone) and subsequently tested in A549-GFP cells under the same transfection conditions, allowing us to directly evaluate their knockdown efficiency relative to the validated control. To validate GFP reduction at both the mRNA and protein levels, we employed RT–qPCR and flow cytometry, as shown in Figure 8.

Flow cytometry and RT–qPCR analyses consistently demonstrated that GFP-targeting ASOs effectively reduced GFP expression in A549-GFP cells (Figure 9). Flow cytometry confirmed a decrease in GFP protein levels following transfection with the reference GFP ASO as well as our computationally designed candidates, with ASO2 and ASO3 producing the most pronounced knockdown compared to scrambled and Lipofectamine-only controls. Complementary RT–qPCR analysis of GFP mRNA levels at 24 h and 48 h post-transfection further validated these findings, showing significant reductions in GFP transcripts for all GFP-targeting ASOs and the reference ASO, relative to controls . Notably, the strongest knockdown was consistently observed with ASO2, ASO3, and the reference ASO, highlighting both the reliability of our system and the capacity of computationally designed ASOs to achieve potent silencing at both the mRNA and protein levels. Together, these results provide robust evidence for the sequence-specific activity of our GFP ASOs and establish a solid foundation for further optimization.

Conclusion

GFP-targeting ASOs designed with our computational pipeline achieved efficient and sequence-specific knockdown of GFP in A549 cells, as confirmed at both the mRNA and protein levels. ASO2 and ASO3 showed the strongest activity, performing comparably or better than the reference ASO, thereby validating the effectiveness of our design approach.

Related sequences and registry information are available on our Parts page.

Another aspect we tested was whether ASOs could work as a new gene-silencing tool in yeast. To do this, we engineered yeast to express GFP so we could try our GFP ASOs in this organism.

While GFP expression in yeast is common, our goal was not simply to generate a fluorescent strain - it was to create a cross-species comparative platform for studying antisense oligonucleotides (ASOs). Specifically, we aimed to express the exact same human-codon-optimized GFP sequence used in our mammalian (HEK293/A549) model inside yeast. This cross-organism setup allows us to test whether ASOs designed against a human transcript retain silencing activity in a phylogenetically distant eukaryotic system.

Rationale

Yeast offers a simple, genetically tractable, and well-characterized eukaryotic environment that enables rapid iteration and mechanistic exploration. However, introducing a human-optimized sequence into yeast poses challenges due to differences in codon usage, mRNA processing, and translation efficiency. Achieving stable and high-level expression of this sequence was therefore crucial before any ASO testing could be performed.

To enable this we engineered a GFP-expressing yeast strain. Using the iGEM Engineering Cycle (Design → Build → Test → Learn), we developed and iteratively optimized the model for stable and robust humanized GFP expression in yeast. (Figure 10)

First Iteration: Low GFP Expression

Our initial construct used a human-optimized GFP gene under a constitutive promoter. Although fluorescence was visible under the microscope, quantitative analysis showed a very low signal, similar to the negative control. We concluded that poor codon compatibility and weak promoter strength likely limited translation efficiency.

Second Iteration: Optimized System

Based on what we learned, we redesigned the construct with key improvements (see our Engineering page), including a stronger promoter, partial codon optimization for yeast, and a degron tag for future protein degradation control. These changes led to a substantial increase in GFP fluorescence, confirmed by plate reader analysis (Figure 11).

This optimized system is now ready to test the efficiency of our ASO candidates (Figure 12). By systematically identifying and addressing expression bottlenecks, we established a reliable model for quantifying ASO-induced mRNA knockdown in yeast.

Related sequences and registry information are available on our Parts page.

To determine the optimal method for delivering antisense oligonucleotides (ASOs) into yeast, we compared three transformation approaches: electroporation, heat shock, and Zymolyase + Lipofectamine. We used a fluorescently labeled oligo (ROX-ASO) to track uptake by confocal microscopy (Figure 13).

Among the three methods tested, electroporation was the only one that consistently resulted in detectable intracellular fluorescence, indicating successful delivery of the fluorescent oligo into S. cerevisiae cells. The other treatments caused substantial cell death, leaving very few intact cells for imaging. The absence of signal in the wash controls suggests that the fluorescence observed in electroporated samples originated from internalized oligo rather than surface-bound molecules.

Next Step

Our next step is to try ASO against GFP in yeast and measure its RNA levels. To establish a quantitative baseline before testing ASO activity, we performed qPCR on our engineered humanized-GFP-expressing yeast strain. As shown in Figure 14, GFP mRNA levels were readily detectable and significantly higher than in wild-type cells, confirming successful expression and normalization across replicates. This calibration step ensures that any future decrease in GFP levels can be confidently attributed to ASO-induced silencing rather than variability in gene expression or sample handling.

Related sequences and registry information are available on our Parts page.

To investigate potential off-target effects of our antisense oligonucleotides (ASOs), we constructed a dual-reporter plasmid encoding two versions of GFP. The first cassette expressed wild-type GFP containing the exact ASO binding site, while the second cassette expressed a GFP variant carrying a few nucleotide changes within that site, mimicking an off-target transcript. This system enabled us to monitor how different mismatches in the ASO binding site affect knockdown efficiency, in order to better predict how ASOs may affect different potential off-target transcripts (Figure 15).

HEK293 cells were transfected with the dual-reporter plasmid and then treated with either the ASO alone or with the ASO complexed to its BROTHER strand [3]. After 24–48 hours, total RNA is isolated and analyzed by RT-qPCR using primer sets specific for each GFP variant.

This assay provided a direct readout of how our ASO treatments influenced both the intended (wild-type) transcript and the off-target mimic. By comparing the mRNA levels under the different conditions, we were able to assess the extent of off-target activity and to test whether BROTHER-assisted ASOs could help mitigate these effects.

Results will be published soon. Check out our designed part.

After confirming that our model-generated ASOs effectively reduce mRNA levels in comparison to established controls, we next applied our approach in the context of synthetic lethality. Specifically, we targeted known MTAP-deletion partners: PRMT5, RIOK1, and MAT2A in the A549 cell line, which is reported in the literature to harbor an MTAP deletion [1],[2],[3] (Figure 1). For each gene, our model generated three distinct ASO candidates, which were subsequently evaluated using a cell viability assay (Figure 2).

Treatment of A549 cells with model-generated ASOs targeting PRMT5 and RIOK1 induced significant, dose-dependent cytotoxicity compared to scrambled and lipofection-only controls (Figure 3). Among them, RIOK-targeting ASOs, particularly ASO-RIOK1, showed the most pronounced reduction in cell viability, while PRMT5 knockdown also produced a measurable effect at higher concentrations.

Video 1: These time-lapse videos show A549 cells (expressing mCherry, in red) treated with different antisense oligonucleotides (ASOs) over ~2 days.

• RIOK1-ASO3 (left panel): Cells progressively lose viability, with decreasing red signal marking cell death.
• PRMT5-ASO1 (middle panel): Similar to RIOK1 knockdown, cells gradually undergo cell death over time, confirming the synthetic lethal effect.
• Scrambled ASO control (right panel): Cells remain largely viable and proliferative, with minimal cell death observed, demonstrating specificity of the targeted ASOs.

Together, these videos illustrate that targeting RIOK1 or PRMT5 with ASOs induces synthetic lethality in A549 cells, while a scrambled ASO has no cytotoxic effect.

Conclusion

ASO-mediated targeting of PRMT5 and RIOK1 effectively reduced the viability of A549 lung cancer cells in a dose-dependent manner, as monitored by live-cell imaging. Both sets of ASOs demonstrated specific cytotoxic effects compared to scrambled controls and untreated groups, confirming that these genes represent critical vulnerabilities in A549 cells. These findings highlight the potential of ASO-based approaches to modulate oncogenic pathways and provide a strong rationale for further mechanistic studies and preclinical evaluation.

As part of our targeted therapeutic platform, we aim to enhance the specificity and efficacy of antisense oligonucleotide (ASO)-based cancer therapy by conjugating the ASO to a monoclonal antibody. This approach is designed to improve delivery of the ASO to lung cancer cells while minimizing off-target effects in healthy tissues. The antibody acts as a targeting moiety, directing the ASO to lung cells. As a first proof of concept, we selected to design and express cetuximab (also called Erbitux), a clinically validated monoclonal antibody that binds to the epidermal growth factor receptor (EGFR), which is frequently overexpressed in non-small cell lung cancer (NSCLC)[1].

The selection of cetuximab was informed by a consultation with Prof. Itai Benhar, a leading expert in antibody engineering. Prof. Benhar confirmed that EGFR is expressed at low levels in healthy lung tissue but is significantly overexpressed in many subtypes of NSCLC, making it an appropriate target for antibody-mediated delivery. Following this recommendation, we obtained the full nucleotide sequences of the heavy and light chains of cetuximab from Prof. Itai Benhar Lab (Figure 1). These included the open reading frames (ORFs) as well as the corresponding promoter regions and 5′ untranslated regions (UTRs) .

Our antibody design workflow (Figure 2) begins with the original cetuximab heavy and light chain plasmids. We plan to generate two new optimized versions and compare them with the initial construct. In the first optimization route, the original sequences will be directly processed using the MNDL Bio platform [2] to enhance antibody expression in CHO cells. In the second route, the sequences will first be optimized using the ESO system (as implemented by the iGEM 2020 team)[3], and then further refined through MNDL Bio [2]. Each optimized sequence will be assembled into expression plasmids containing the appropriate heavy and light chain ORFs, followed by CHO expression and comparative analysis of antibody yield, binding efficiency, and internalization properties.

MNDL Bio is a biotechnology company specializing in AI-driven gene expression optimization to maximize recombinant protein yields. Their platform integrates deep learning, machine learning, and biophysical modeling to co-optimize both coding and non-coding regions - far beyond traditional codon optimization. By considering translation dynamics, vector stability, and preservation of hidden regulatory signals, MNDL delivers tailored DNA constructs for diverse hosts, including microbial, plant, and cell-free systems. This technology has achieved up to 20-fold increases in protein production, reduced production costs by 50–90%, and accelerated time to market by up to 33%, with proven success in dozens of complex protein expression projects. We used the MNDL Bio optimization tool in our work to enhance expression efficiency of the antibody.

The antibody sequences were transferred to Dr. Zohar Zafrir and Doron Armon, both researchers in Prof. Tamir Tuller’s lab, for codon optimization to ensure high expression efficiency in our mammalian expression system. They provided us with the codon optimized table (see patent IL305852A - "Codon optimization of antibody sequences”), which then was tested in the ESO system. We then continue with the construct optimization using the MNDL Bio tool.

The Evolutionary Stability Optimizer (ESO) is a computational platform developed to design DNA sequences that are both highly expressed and resistant to mutational degradation over time [3] . Unlike previous tools, ESO simultaneously detects and corrects multiple sources of instability, including simple sequence repeats (SSRs), repeat-mediated deletions (RMDs), and epigenetic hotspots such as methylation motifs, while allowing users to define custom sequences to avoid. Its optimization engine applies a two-step process: first improving codon usage and GC content according to the target host, then removing mutational and epigenetic hotspots while preserving protein sequence and locked regions. The system supports large-scale batch analysis, generates fully optimized sequences ready for synthesis, and has been shown to significantly improve predicted evolutionary stability without sacrificing expression levels. In our workflow, ESO was used after codon optimization to ensure long-term genetic stability of the cetuximab heavy and light chain sequences prior to expression in CHO cells. The ESO platform is freely available online at https://www.cs.tau.ac.il/~tamirtul/ESO/.

After optimizing the antibody sequences, we ordered the four gene fragments from Twist Bioscience with homology arms to pcDNA3.4, and cloned them by Gibson assembly (Figure 3).

Antibody Cleaning, Purification, Analysis:

To evaluate how computational sequence design influences antibody expression, the same amount of CHO cells were transfected with plasmids encoding three Erbitux variants: the original Erbitux sequence, and two model-generated constructs: MNDL Erbitux and ESO Erbitux. Untransfected cells (UT) served as a negative control. After expression, antibodies secreted into the culture medium were collected and analyzed by SDS–PAGE and Western blot. Detection was performed using an HRP-conjugated Goat Anti-Human IgG (H+L) antibody.

A distinct band corresponding to the full antibody (~150 kDa) was observed in all transfected samples but not in the control, confirming successful secretion. Densitometric analysis of band intensities using the ImageJ program, and normalization to untreated cells showed that the MNDL variant exhibited the highest expression level, roughly twice that of the original Erbitux construct. The ESO variant showed expression similar to the original Erbitux; interestingly, this shows that the ESO was able to optimize the sequence for evolutionary stability without compromising the expression of Erbitux. These results indicate that computational optimization of antibody sequences can significantly modulate expression efficiency of antibodies in CHO cells (Figure 4).

Related sequences and registry information are available on our Parts page.

After successfully optimizing and producing our antibody and developing a computational and experimental framework for generating functional antisense oligonucleotides (ASOs) capable of selectively silencing essential genes in cancer cells, we identified and validated ASOs that induced measurable cytotoxic effects in cancerous cell lines, confirming their potential as therapeutic agents.Building on these two achievements, a specific antibody and a model that produces functional, cell-killing ASOs, we now aim to combine both systems into a single multifunctional therapeutic molecule.

Our next step is to chemically conjugate the optimized antibody to the ASO using NHS-ester chemistry, as was advised by Prof. Doron Shabat, an expert in organic and bioorganic chemistry. We draw inspiration from industrial approaches such as those implemented by Igor et al[4], which utilize both site-specific and random lysine conjugation strategies and diverse linker chemistries (e.g., cleavable disulfide or stable amide linkages) to balance stability and release efficiency (Figure 5).

This approach enables the covalent linkage of NHS-activated ASOs to lysine residues on the antibody through stable amide bonds, while maintaining protein integrity and biological activity under mild aqueous conditions. The resulting antibody–ASO conjugate (AOC) is expected to combine targeted delivery and gene-specific silencing, directing the ASO precisely to the cells recognized by the antibody and thereby enhancing therapeutic specificity and potency.

In our upcoming experiments, we will:

• Validate conjugation efficiency and molecular integrity using SDS-PAGE and UV/Vis analysis.
• Confirm preservation of antibody binding capability.
• Evaluate the conjugate’s ability to silence target genes and reduce cell viability in cancer cell models.

Results will be published soon.

As part of our therapeutic strategy, we sought to test whether an antibody–epitope conjugate can be internalized by cancer cells, processed through the endogenous antigen presentation machinery, and subsequently activate CD8⁺ T cells. This experiment is designed to evaluate the immunogenic arm of our platform, in which the antibody functions not only as a delivery vehicle but also as a trigger for immune recognition. By enabling the presentation of a tumor-associated epitope on MHC-I, this approach has the potential to stimulate T cells that will not only kill the targeted cancer cells but also establish long-term immune surveillance, thereby supporting the sustained elimination of cancer.

To establish a proof of concept, we selected the MART-1 (Melan-A) epitope, amino acids 26–35 (ELAGIGILTV), a well-characterized HLA-A2–restricted peptide. We collaborated with Prof. Cyrille Cohen (Bar-Ilan University), whose group has engineered T cells expressing a MART-1–specific TCR [1] . These engineered T cells provide a sensitive readout of antigen presentation on MHC-I.

For the assay, we will use A375 human melanoma cells (ATCC CRL-1619), which naturally express HLA-A2. Cells will be incubated with our designed antibody–epitope conjugates, and internalization will be assessed through receptor-mediated uptake of the antibody bound to EGFR. Upon internalization, the MART-1 peptide payload is expected to be processed by the proteasome and presented on MHC-I molecules at the cell surface.

Subsequently, MART-1–specific CD8⁺ T cells will be co-cultured with the treated A375 cells. Successful presentation of the MART-1 epitope should activate these T cells, leading to cytokine secretion (e.g., IFN-γ, TNF-α) and target-cell killing. T cell proliferation and cytokine release will be measured as indicators of activation and tumor recognition (Figure 1).

This experiment provides a controlled system to evaluate whether antibody–epitope complexes can bridge targeted delivery and adaptive immune activation, a key step toward combining precision delivery of ASOs with long-term immune-mediated tumor clearance.

Results will be published soon.