Background

Modular, programmable RNA sensing using ADAR editing in living cells

The ability to sense RNA inside living cells has the potential to allow diagnostics and therapeutics precise control over cellular behavior. Traditional RNA-based sensing mechanisms have mostly relied on microRNAs or engineered ribozymes and guide RNAs, but these approaches often suffer from limited programmability or interference with host pathways (Kaseniit et al., 2022). To overcome these limitations, Kaseniit et al. (2023), Jiang et al. (2023), and Qian et al. (2022) developed a modular and programmable system called RADAR using the natural activity of adenosine deaminases acting on RNA (ADARs) for RNA sensing. ADARs convert adenosine to inosine within double-stranded RNA, and RADAR uses this process to remove stop codons that block translation of a downstream protein, thereby linking the presence of a specific RNA sequence directly to a measurable and potentially therapeutic output. RADAR sensors have been validated in human and mouse transcriptomes and shown to differentiate between disease-relevant mutations and wild-type transcripts.

RADAR converts presence of target RNA seq into a functional protein output

Hepatocellular Carcinoma development from chronic Hepatitis B Virus conditions

Patients chronically infected with hepatitis B virus (HBV) are at high risk of developing chronic liver disease (CLD), which progresses from hepatitis to fibrosis, cirrhosis, and ultimately hepatocellular carcinoma (HCC) (Shen et al., 2023). A major contributor to this process is the hepatitis B x antigen (HBx), which not only promotes HBV gene expression and replication from covalently closed circular (ccc) DNA, but also induces oxidative stress, mitochondrial damage, and persistent activation of transcription factors, maintaining cycles of inflammation, hepatocellular destruction, and regeneration (Feitelson et al., 2022). Importantly for us, HBx often persists in the liver through integration of HBV DNA into the host genome, allowing continuous HBx expression even in patients with little or no viral replication. This integration causes cancer by disrupting normal host genes and chromosomal stability, leading to uncontrolled cell growth and survival driving tumor development (Levrero & Zucman-Rossi, 2016). In this context, multiple host biomarkers are closely linked with HBV-related HCC, including glypican-3 (GPC3), a cell-surface proteoglycan that enhances Wnt/β-catenin signaling and serves as a diagnostic marker distinguishing HCC from benign liver disease, and aldo-keto reductase family 1 member B10 (AKR1B10), an enzyme upregulated by oxidative stress that removes byproducts of lipid damage, allowing for tumor cell survival and metabolic adaptation (Zhou et al., 2017; Wang et al., 2009).

Goals and Objectives

Our primary objective is to develop and validate the synthetic RNA sensing system that will only selectively detect HBV-driven hepatocellular carcinoma cells based on transcriptomic signatures. Rather than relying on traditional surface biomarkers, our objective is to build a programmable intracellular sensor capable of responding to combinations of RNA transcripts present in cancerous liver cells, but absent or minimally expressed in healthy tissue. Our goals specifically include engineering this system to drive expression of an therapeutic output protein only in target cells, optimizing its specificity and reliability in liver cancer-relevant models such as that of HBV transformed HCC. We test the system using artificially induced trigger environments in HEK293T cells and plan to test the system in vitro in engineered liver cell lines that naturally express our desired triggers and more closely resemble the liver cancerous tumor environment and finally in vivo. Ultimately, we aim to create a proof-of-concept for a modular therapeutic platform that can be adapted to other cancers lacking one specific biomarker, but possessing multiple transcriptional profiles instead, with our system even demonstrating the capability of sensing the edge case of a viral indicator.

We aim to successfully develop a proof of concept for an logic-gated RNA-sensing system for the generation of an effective therapeutic response targeted to a specific cancerous area. The platform can be adapted to output diverse molecules, including apoptotic proteins, cytokines, immune checkpoint inhibitors etc. that recruit immune cells to the tumor microenvironment (TME). With this in mind, we aim to demonstrate that our sensors for HBV-induced HCC are reliable and marked by GFP production, which serves as a placeholder for our upcoming experiments including producing IL-2, a cytokine known to enhance immune responses (Jiang et al., 2016). To ensure specificity and functionality even within physiologically relevant transcriptional environments, we test this model in HEK293T cells and are in the process of testing within the Hep3B cell line derived from HBV-induced HCC tumors as well. Future progress includes testing the system in other cancers induced by viral integration or possessing another biomarker specific to a certain stage of progression.

Results

We first evaluated the performance of the viral HBx sensor. In HEK293T cells co-transfected with synthetic triggers, the Broad HBx sensor exhibited a 23-fold increase in GFP fluorescence relative to the scramble controls (p = 0.00262 following t-test). A second construct, the Hep 3B-specific HBx sensor, demonstrated a 13-fold increase in GFP signal compared to scramble (p = 0.00297 following t-test). Importantly, the Hep 3B-specific sensor maintained detection capacity even when challenged with a synthetic trigger containing up to 5% sequence mismatch, reflecting resistance to naturally occurring viral sequence variation. Comparison of matched and partially matched triggers confirmed significant differences in GFP output (p = 0.0233), supporting the simultaneous specificity and universality of the Broad design (Figure 1, 2, and 3).

Figure 1. Validation of HBx sensor performance in HEK293T cells.
HBx_Broad sensor shows GFP fluorescence output in response to matched versus scrambled synthetic triggers. HBx_Hep3B-specific sensor shows GFP fluorescence output in response to matched (HBx_Broad trigger with 5% mismatch) versus scrambled synthetic triggers. White signals indicate GFP-positive cells; red signals represent sensor presence. Both sensors showed strong responses to matched triggers and negligible responses to scramble controls, confirming specificity of detection.

Figure 2. Single replicate quantitative evaluation of HBx sensor performance. Flow cytometry analysis of GFP fluorescence in HEK293T cells transfected with HBx_Broad and HBx_Hep3B sensors in response to matched and mismatched synthetic triggers. Data was collected for 100,000 cell events and were gated on sensor-high cells. (Top) Histogram overlays show a clear rightward shift in GFP intensity for matched triggers (blue) compared to scramble controls (red), with the Hep3B sensor retaining activity even against 5% mismatched sequences. (Bottom) Tabulated mean fluorescence intensity (MFI) values demonstrate 13 fold and 23 fold increases in GFP output relative to scramble controls.

Figure 3. HBx sensor discrimination between matched and partially matched triggers. Dot plot showing GFP fluorescence output in HEK293T cells transfected with HBx sensors in the presence of complete matched or partial matched synthetic triggers. Complete match triggers produced strong GFP activation, while partial match triggers (containing 5% mismatched sequence) still elicited a measurable, but reduced signal (p < 0.01). As expected, complete match triggers also produced a statistically significant increase in fluorescent intensity compared to partial match triggers (p=0.0233). Results confirm both the sensitivity and ability to tolerate limited sequence variation in viral sequences.

Next, we tested sensors targeting hepatocellular carcinoma (HCC)-specific transcripts, testing two targets strongly expressed in HCC, GPC3 surface protein and AKR1B10 metabolic enzyme. The GPC3 sensor showed clear activation in response to its respective trigger while remaining silent under scramble conditions (Figure 4). Similarly, the AKR1B10 sensor displayed strong discrimination between scramble and matched triggers, with event counts once again confirming specificity in functionality (Figure 4). Mean calculations of our flow cytometry values revealed that our AKR1B10 sensor exhibited a 20-fold increase in GFP fluorescence in the top 1% of sensor expressing cells (p = 0.08 following t-test) whereas GPC3 sensor exhibited a 15-fold increase in GFP fluorescence (p = 0.00262 following t-test) (Figure 5, 6, 7).

Figure 4. Validation of HCC-specific mRNA sensor performance in HEK293T cells.
The GPC3 sensor showed GFP fluorescence output in response to matched versus scrambled synthetic triggers. The AKR1B10 sensor showed GFP fluorescence output in response to matched versus scrambled synthetic triggers. White signals indicate GFP-positive cells; red signals represent sensor presence. Both sensors showed strong responses to matched triggers and negligible responses to scramble controls, confirming specificity of detection.

Figure 5. Single replicate quantitative evaluation of AKR1B10 sensor performance. Flow cytometry analysis of GFP fluorescence in HEK293T cells transfected with the AKR1B10 sensor in response to matched and mismatched synthetic triggers. (Top) Histogram overlays show a clear rightward shift in GFP intensity for matched triggers (dark blue) compared to scramble controls (light blue). (Bottom) Tabulated mean fluorescence intensity (MFI) values demonstrate 20-fold increases in GFP output relative to scramble controls for this single replicate evaluation.

Figure 6. Single replicate quantitative evaluation of GPC3 sensor performance. Flow cytometry analysis of GFP fluorescence in HEK293T cells transfected with the GPC3 sensor in response to matched and mismatched synthetic triggers. (Top) Histogram overlays show a clear rightward shift in GFP intensity for matched triggers (bright orange) compared to scramble controls (light orange). (Bottom) Tabulated mean fluorescence intensity (MFI) values demonstrate 16-fold increases in GFP output relative to scramble controls, for the single replicate quantitative evaluation.

Figure 7. GPC3 and AKR1B10 sensor discrimination between matched and mismatched triggers. Dot plot showing GFP fluorescence output in HEK293T cells transfected with GPC3 and AKR1B10 sensors in the presence of complete matched or mismatched synthetic triggers. Complete match triggers produced strong GFP activation, while mismatch triggers elicited a negligible signal. GPC3 exhibited strong output signalling (p = 0.00262 following t-test)), while AKR1B10 exhibited a nonsignificant output signal (p = 0.08 following t-test statistical analysis). These findings demonstrate that the GPC3 sensor produces a sensitive and measurable output, in contrast to the weaker response observed with the AKR1B10 sensor.

We next tested whether combining the viral and tumor sensors in a Boolean logic framework could enable selective signal production, ultimately translating to a cell-targeted therapeutic response. The two halves of the full GFP protein were first co-transfected to assess protein trans-splicing efficiency. A strong GFP output resulted when both halves expressed under a constitutive promoter were co-transfected in HEK293T cells compared to negligible output produced in the cases of each transfected alone (Figure 9). When the HBx and GPC3 sensors were integrated into an AND-gated circuit controlling split GFP, fluorescence was observed only in the presence of both triggers (Figure 10). Neither input alone was sufficient to produce the signal. HEK293T cells transfected with both sensors and their triggers exhibited a 1.6-fold change compared to a partial match containing only the GPC3 trigger and 1.7-fold change compared to the partial match containing the HBx trigger. Relative to the trial containing a complete mismatch among both sensors, there was a 1.6-fold change compared to the complete mismatch containing no triggers cognate to the sensors (Figure 11). These results demonstrate that the dual-sensor framework operates with strict AND logic, confirming compatibility of RADAR-based components in multi-input intracellular circuits for cell-targeted therapeutics.

Figure 9. Split GFP reconstitution validates sensor output design. HEK293T cells were transfected with either spGFP1-10 alone, spGFP11 alone, or both simultaneously. No fluorescence was observed with either half alone, while measurable GFP signal was detected only when both fragments were co-expressed. This confirms proper reassembly of the split GFP protein for the AND-gated sensor to be tested.

Figure 10. Microscopy validation of AND-gated GFP proof of concept. HEK293T cells were transfected with GPC3 + HBx AND-gated sensors and co-transfected with either both GPC3 and HBx_Broad triggers (dual match), just one of them, or neither of them (scramble). The dual match showed GFP output compared to negligible output in the remaining three samples. These results allow us to qualitatively validate the AND-gate sensor-output actuation design.

Figure 11. Single replicate quantitative evaluation of AND-Gated sensor performance.
Flow cytometry analysis of GFP fluorescence in HEK293T cells transfected with the GPC3-HBx AND-gated sensors in response to matched and mismatched synthetic triggers. Complete match triggers relative to both sensors (Specimen 4) produced the highest GFP output with a mean fluorescence of ~413.6 compared to all other samples, including complete mismatches and partial matches. Both partial mismatched and complete mismatched specimens showed lower signal relative to the dual match, confirming high specificity. These results demonstrate accurate distinction between matched and mismatched inputs, with a 1.6-fold activation observed only under complete match conditions. However, since cells from three technical replicates in a 48-well format were harvested and combined into one sample for sufficient cell counts to run flow cytometry, more biological and technical replicates will be conducted in the future to allow statistical tests to be done.

To move beyond proof-of-concept, we designed a split IL-2 therapeutic output to replace the GFP split outputs. Similar to the GFP prototype, upon activation of both sensors, it is expected that the IL-2 fragments will reconstitute to produce the functional protein, establishing that the system can be readily adapted from sensing to therapeutic actuation. This would validate the modularity of the LiRA platform for any case of cancer so long as two transcriptional inputs can be combined to produce the therapeutic protein of choice. We hope to validate this system in the future.

Finally, we are currently evaluating the tumor-specific GPC3 sensor in the Hep3B cell line, a hepatocellular carcinoma line that integrates HBV sequences and exhibits high GPC3 expression. We aim for the GPC3 sensor to display a measurable activation in the Hep3B cell, consistent with predictions from our dry lab algorithm, which evaluated log-fold change (HCC tissue relative to healthy counterparts) and TPM values for the Hep3B cell line to rank GPC3 as the top transcript in this cell line. This cross-validation between computational and experimental findings will strengthen our confidence in both the design pipeline and biological performance in patient cases of HCC.

Collectively, these results demonstrate that LiRA can:

• Identified viral/endogenous mRNA sequences and validated sensor function for those sequences using high copy synthetic trigger mRNAs provided by transfection
• Integrate signals through an AND-gated design to deliver a cell-specific response
• Actuate a diagnostic (ex. Split GFP) as an actuator response

The system functioned as predicted in both engineered HEK293T cells, supporting the feasibility of LiRA as a programmable platform for selective recognition and treatment on a cancer case lacking specific biomarkers given any two transcriptomic signatures.

Discussion

The successful results in the HEK293T demonstrate that two independent transcripts can be combined to drive selective activation of a therapeutic output, offering an alternative to conventional biomarker based therapies, which often suffer from poor specificity (Chakraborty & Rahman, 2012). By requiring the simultaneous presence of both, a viral integration product (HBx) and an HCC-associated transcript (GPC3), our LiRA system reduces the risk of off-target activation in healthy tissue.

One of the important findings from this work is the ability of our viral HBx sensor to maintain detection despite up to 5% sequence mismatch. Given the high mutation rate of HBV, this ensures the viralescape variants are unlikely to undermine the HBx sensor we designed using a highly conserved 3'UTR HBx sequence. In parallel, the GPC3 and AKR1B10 sensors were validated as reliable tumor-specific inputs, with GPC3 further confirmed as a high-confidence target in Hep3B cells akin to diseased cells. Validation of an AND gate using these sensors led to a more translational step. Our transition from a diagnostic GFP output to a therapeutic split IL-2 protein represents an application of this model. Il-2 is a clinically established cytokine with known immunostimulatory effects and its modular assembly through the split protein design gives us a framework for adapting LiRA to diverse therapeutic efforts. This adaptability positions LiRA not only as a way to diagnose cancers lacking specific biomarkers, but also as a system capable of real-time detection and treatment within the diseased cells themselves.

In summary, we developed a programmable intracellular RNA logic-gated therapeutic platform capable of both sensing and actuation. By combining detection of multiple RNA transcripts to therapeutic protein production, LiRA offers an alternative method to selectively target cancers that lack a single defining biomarker. Our results show proof of concept in engineered HEK293T cells, supporting the logic of LiRA as a modular system for precise cancer therapy.

Limitations and Future Directions

That said, some limitations remain. First, sensors require relatively high transcript expression to achieve activation and measurable output, which may limit applicability in tumors with lower biomarker abundance. Although our computational projects have developed a screening method capable of highlighting sequences highest in transcript prevalence for a given cell line, additional testing across a broader panel of HCC cell lines will be necessary to establish generalizability. Second, while IL-2 served as an effective demonstration of a therapeutic application using our modular system, future work should expand to alternative outputs, including pro-apoptotic proteins or immunomodulatory effectors, to diversify therapeutic potential and cover a broader range of cancers the system can combat. Finally, delivery remains a discussion for the system. Given this prototype, we envision the system could be administered intravenously to target the liver, where HBV-associated HCC arises. However, adapting this approach to other cancers would require new strategies to ensure delivery to the diseased tissues. Regardless of the specificity of delivery, the system's precision in actuation eases the worries of off-target harm.

Despite these questions, LiRA demonstrates the promise of RNA logic-gated systems for cancer therapy, enabling both precise detection and therapeutic response in a single system. With further optimization, this approach could be extended beyond HBV-associated HCC to other cancers defined by multiple RNA signatures.

References

1. Chakraborty, S., & Rahman, T. (2012). The Difficulties in Cancer Treatment. Ecancermedicalscience, 6(16). https://doi.org/10.3332/ecancer.2012.ed16
2. Feitelson, M. A., Arzumanyan, A., Spector, I., & Arvin Medhat. (2022). Hepatitis B x (HBx) as a Component of a Functional Cure for Chronic Hepatitis B. Biomedicines, 10(9), 2210–2210. https://doi.org/10.3390/biomedicines10092210
3. Jiang, K., Koob, J., Chen, X.D. et al. (2023). Programmable eukaryotic protein synthesis with RNA sensors by harnessing ADAR. Nat Biotechnol, 41, 698–707. https://doi.org/10.1038/s41587-022-01534-5
4. Jiang, T., Zhou, C., & Ren, S. (2016). Role of IL-2 in cancer immunotherapy. OncoImmunology, 5(6), e1163462. https://doi.org/10.1080/2162402x.2016.1163462
5. Kaseniit, K. E., Katz, N., Kolber, N. S., Call, C. C., Wengier, D. L., Cody, W. B., Sattely, E. S., & Gao, X. J. (2022). Modular, programmable RNA sensing using ADAR editing in living cells. Nature Biotechnology. https://doi.org/10.1038/s41587-022-01493-x
6. Levrero, M., & Zucman-Rossi, J. (2016). Mechanisms of HBV-induced hepatocellular carcinoma. Journal of Hepatology, 64(1), S84–S101. https://doi.org/10.1016/j.jhep.2016.02.021
7. Qian, Y., Li, J., Zhao, S. et al. (2022). Programmable RNA sensing for cell monitoring and manipulation. Nature, 610, 713–721. https://doi.org/10.1038/s41586-022-05280-1
8. Shen, C., Jiang, X., Li, M., & Luo, Y. (2023). Hepatitis Virus and Hepatocellular Carcinoma: Recent Advances. Cancers, 15(2), 533. https://doi.org/10.3390/cancers15020533
9. Wang, C., Yan, R., Luo, D., Watabe, K., Liao, D.-F., & Cao, D. (2009). Aldo-keto Reductase Family 1 Member B10 Promotes Cell Survival by Regulating Lipid Synthesis and Eliminating Carbonyls. Journal of Biological Chemistry, 284(39), 26742–26748. https://doi.org/10.1074/jbc.m109.022897
10. Zhou, F., Shang, W., Yu, X., & Tian, J. (2017). Glypican-3: A promising biomarker for hepatocellular carcinoma diagnosis and treatment. Medicinal Research Reviews, 38(2), 741–767. https://doi.org/10.1002/med.21455