An ER stress-responsive RNA rheostat for Programmable Gene and mRNA therapies

Summary

Therapeutic protein overexpression can overwhelm endoplasmic reticulum (ER) folding capacity, trigger unfolded protein response (UPR) signaling, and compromise the safety of gene and mRNA therapies. Here, we engineer stress-responsive RNA rheostats that couple transgene expression to endogenous ER stress sensing. Short RNA elements derived from X-box-binding protein 1 (XBP1) mRNA undergo IRE1α-dependent splicing under ER stress, inducing a frameshift that attenuates downstream protein expression. XBP1 switches function across DNA and mRNA delivery platforms and regulate expression of fluorescent reporters, coagulation factor VIII, and Leronlimab, a therapeutic anti-CCR5 monoclonal antibody. Switch activation reduces ER stress markers while preserving expression under homeostatic conditions. We further demonstrate regulation of Leronlimab expression in vivo using recombinant adeno-associated virus vectors. Together, these findings establish programmable RNA feedback control as a strategy for linking cellular proteostasis to therapeutic protein expression and improving the safety of gene and mRNA therapies.

eTOC blurb

Proteostasis pathways offer opportunities for engineering cellular control systems. Ogurlu et al. repurpose IRE1α-dependent XBP1 splicing to create RNA rheostats that link ER stress sensing with transgene expression, establishing a feedback mechanism that limits protein overexpression and improves control of gene and mRNA therapies.

Graphical Abstract

Introduction

Gene and mRNA therapies to restore or augment therapeutic protein levels have emerged as promising treatment options to combat genetic diseases. Recombinant adeno-associated viruses (rAAV) and lipid nanoparticles (LNP) are the leading delivery platforms for introducing therapeutic protein coding DNA or mRNA to cells, respectively.^1–3 Various classes of proteins such enzymes, receptors/transporters, mitochondrial/nuclear factors, and secreted blood factors are expressed by these platforms in a wide range of cell and tissue types.^2,4 While promising, such therapies can also pose complexities such as inflammatory responses and liver toxicities that present clinical challenges. Further, high AAV vector or LNP-mRNA doses required to achieve therapeutic efficacy compounded with high liver uptake and transgene expression levels can cause unintended side effects.^5–8 Specifically, hepatotoxicity has been reported to be the most common side effect in AAV clinical trials and a safety concern in LNP-mRNA preclinical studies in recent years.^4,6

Supraphysiological protein expression from DNA/mRNA can result in overload in the endoplasmic reticulum (ER) and trigger ER stress, which has been implicated in hepatotoxicity caused by nucleic acid products.^9–11 The ER is a eukaryotic organelle consisting of a network of branching tubules and flattened sacs that have an essential function in synthesis, processing, and transport of more than one third of the proteins produced in the cell. In particular, most secretory proteins are translated in the ER. When protein production and folding demands overwhelm ER capacity, misfolded proteins accumulate in the lumen and induce ER stress. In order to relieve ER stress and re-establish homeostasis, three principal signaling branches of the unfolded protein response (UPR) pathway initiate a series of adaptive transcriptional and translational events through ER transmembrane proteins inositol-requiring enzyme 1α (IRE1α), protein kinase R-like endoplasmic reticulum kinase (PERK), and activating transcription factor 6 (ATF6). However, if ER stress is unresolved, an alternate ‘terminal’ UPR program promotes apoptosis.^12–14

Under ER stress conditions, IRE1α oligomerization and autophosphorylation stimulates a conformational change that activates its cytoplasmic RNase domain. In its active state, IRE1α’s RNase removes a 26-nucleotide intron from X-box-binding protein 1 (XBP1) mRNA, leading to a shift in the protein reading frame. The shorter XBP1 protein expressed from the new reading frame is a potent transcription factor that modulates transcription of UPR target genes.^12–16 Specifically, XBP1 mRNA splicing on ER membrane takes place with the following sequence of events – ribosomes begin translating the unspliced XBP1 mRNA (XBP1u), an evolutionarily conserved peptide module at the carboxyl terminus (ribosomal arrest peptide) pauses XBP1u translation, signal recognition particle (SRP) recognizes hydrophobic region 2 (HR2) on XBP1u nascent peptide and recruits the mRNA-ribosome-nascent chain complex to the ER membrane, where IRE1α cleaves two specific sites on the mRNA, RtcB ligase and Archease join remaining 5′ and 3′ fragments after intron removal, and the spliced XBP1 mRNA (XBP1s) is produced (Figure 1A).^17–20

Figure 1. XBP1 mRNA splicing based RNA structural switch working principle.

(A) Under ER stress conditions and when UPR is triggered, signal recognition particle (not shown) recognizes hydrophobic region 2 (HR2) on XBP1u nascent chain and drags the mRNA-ribosome-nascent chain complex to ER membrane. Oligomerized IRE1α cleaves two specific sites on the mRNA which results in removal of a 26-nt intron. (B) Our ER stress sensing switch leverages this splicing mechanism by incorporating the protein coding sequence of XBP1u and its functional fragments (XBP1F) at the 5′end of a gene expression modality. When overexpression of a therapeutic protein encoded in a gene expression modality induces ER stress and stimulates UPR, IRE1α splices out the XBP1u intron, and RtcB ligase/Archease seal the 5′ and 3′ fragments. These events lead to a frameshift in the protein coding frame and termination of protein expression.

Design

Previous studies leveraging the IRE1α-XBP1 pathway generated XBP1 splicing-based fluorescence and luminescence reporters to monitor splicing and enable mechanistic studies.^21,22 Building on this concept, we developed a feedback strategy to control protein expression from nucleic acids encoding ER stress-inducing proteins. Specifically, we designed ER stress-responsive RNA structural switches by incorporating stretches of XBP1 mRNA upstream of reporter and therapeutic transcripts. We hypothesized that overexpression of an ER stress-inducing protein would subject the switch-containing mRNA to UPR regulation through IRE1α-mediated splicing. The frameshift mutation arising from this splicing event is expected to reduce protein expression, thereby mitigating ER stress and limiting downstream cellular consequences (Figure 1B). We explored multiple XBP1-derived configurations to achieve efficient stress-responsive regulation while maintaining productive protein expression.

Results

Synthetic XBP1u-EGFP mRNA undergoes unconventional cytosolic mRNA splicing under chemically triggered ER stress.

We engineered a reporter construct by inserting the XBP1u coding sequence without a stop codon in-frame and placed upstream of EGFP lacking an AUG start codon. In this fusion construct transcribed from the CBA promoter, ER stress-driven splicing causes a frameshift in XBP1u-EGFP ORF and incorporation of two stop codons before the EGFP reading frame (Figure 2A). In order to evaluate reporter splicing under ER stress, we added thapsigargin to reporter plasmid transfected HEK293 cells. Treatment with thapsigargin, a small molecule inhibitor of sarcoplasmic/endoplasmic reticulum Ca²⁺-ATPase (SERCA), elicits ER stress by exhausting ER calcium levels and disturbing protein folding.²³ The ER chaperone binding immunoglobulin protein, also known as glucose-regulated protein 78, (BiP/GRP78) is an ER stress marker that plays a central role in IRE1α activation in the presence of misfolded proteins.^12,13,23,24 Incubation with thapsigargin upregulated BiP/GRP78 mRNA expression 3-fold at 1 hour and 4-fold at 3 hours relative to the DMSO control (Figure 2B). Correspondingly, full-length XBP1u-EGFP expression was downregulated 2-fold at 1 hour and 3-fold at 3 hours, confirming excision of XBP1u-EGFP intron in response to ER stress (Figure 2D). Furthermore, formation of XBP1s-EGFP after 3 hours of thapsigargin incubation was validated by PCR for both plasmid and IVT mRNA transfected cells which confirms that the splicing event occurred in the cytosol, post-nuclear export (Figure 2C). Collectively, these data suggest that appending XBP1u to EGFP mRNA imparts the ability to be spliced by IRE1α.

Figure 2. Chemical ER stress induction reduces XBP1u-EGFP mRNA abundance and EGFP-degron (EGFPd2) protein expression in HEK293 cells.

(A) XBP1u sequence was incorporated at the 5′end of EGFP transcript. During UPR, the 26-nucleotide intron in XBP1u is removed by IRE1α, causing a frameshift and introducing two stop codons before EGFP. (B, D) HEK293 cells were transfected with a plasmid encoding for XBP1u-EGFP and supplied with 1μM thapsigargin or DMSO after 48 hours. RNA from replicate samples was reverse transcribed and cDNA was used to perform RT-qPCR for BiP/GRP78 and full-length XBP1u-EGFP. Data is normalized to RPL19 and reported relative to cells only control. (left: 1 hour, right: 3 hours) (n = 3) (C) RT-PCR with primers spanning the XBP1u intron was conducted for 3-hour samples (n = 4). (top: XBP1u-EGFP plasmid, bottom: XBP1u-EGFP IVT mRNA) (E) XBP1F1 and XBP1F2, fragments of XBP1u mRNA, were cloned upstream of a sequence encoding a short half-life (~2 hours) EGFP (EGFPd2). Porcine teschovirus-1 2A (P2A) self-cleaving peptide was included between XBP1F and EGFPd2. White box is the 26-nt XBP1u intron. (F) 48 hours after transfection, 1μM thapsigargin was added to cells transfected with EGFPd2 or XBP1F1-EGFPd2. Mean fluorescence intensity (MFI) was quantified by flow cytometry every 4 hours (0-12 hours) and percent change was calculated relative to MFI at Time 0 (n = 5). (G) Media was changed 12 hours after incubation with thapsigargin and flow cytometry experiment was repeated every 4 hours (16-28 hours) post-washout (n = 5). (H) Flow quantification of change in MFI with thapsigargin supplementation was repeated for the longer fragment of XBP1u (XBP1F2) (n = 5). In F, G, and H, the EGFPd2 reporter lacking the XBP1F element serves as a control for ER stress-induced global translational attenuation. Unpaired two-tailed t-test was applied to B, D and two-way ANOVA with Sidak’s multiple comparison test was applied to F, G, and H. Where indicated *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001. See also Figures S1 and S2.

Reporter protein expression is responsive to ER stress when XBP1 fragments are included at the 5′ of chimeric mRNA.

We truncated the protein coding region of the XBP1u mRNA into its functional components to identify the shortest mRNA stretch that undergoes efficient splicing in our experimental setting. We identified XBP1 mRNA functional components on previously published papers that study the essential elements required for XBP1 mRNA splicing by IRE1α.^15,17–19 Notably, the double-stem loop structure is indispensable for recognition by IRE1α and is preserved in our shortest fragment, XBP1F1 (Figure S1A, B). XBP1F1 covers a 102-nucleotide region spanning 40-nucleotides upstream and 36-nucleotides downstream of the XBP1u intron. We extended XBP1F1 to encompass the HR2 domain to generate XBP1F2. After XBP1F is a P2A ribosomal skipping element followed by an EGFP reporter with a C-terminally fused mouse ornithine decarboxylase degradation signal peptide (EGFPd2) which shortens protein half-life to ~2 hours (Figure 2E). When in-frame, P2A promotes ribosomal skipping, and thus detaching EGFPd2 from XBP1F peptide. Sequences encoding for P2A and EGFPd2 are separated by an ‘AAC’ Asn codon to accommodate integration of a ‘UAA’ stop codon when splicing leads to a +2 frameshift, preventing reporter protein expression. Notably, previous XBP1 splicing-based reporter constructs function as readouts of IRE1α-dependent XBP1 mRNA splicing, whereas XBP1F splicing introduces a premature stop codon that attenuates translation of the downstream gene. We confirmed XBP1F1-EGFPd2 and XBP1F2-EGFPd2 splicing under chemically triggered ER stress in HEK293 cells via gel electrophoresis (Figure S2A). Then, we studied the effects of XBP1F1 and XBP1F2 splicing domains on EGFPd2 protein expression in HEK293 cells by measuring the change in mean fluorescence intensity (MFI), via flow cytometry and subsequent gating (Figure S2B), over time after ER stress induction with thapsigargin. We observed a significant decline in percent change in MFI at 8 and 12 hours with XBP1F1-EGFPd2 compared to a construct lacking an XBP1F segment (Figure 2F). Removing thapsigargin reversed the decline and percent change in MFI was equalized for the constructs at later timepoints, corroborating that the decrease in protein expression efficiency under ER stress was transient (Figure 2G). Decrease in percent change in MFI was steeper for XBP1F2-EGFPd2, dropping by 11.89% and 31.17% relative to EGFPd2 alone after 8 and 12 hours of thapsigargin incubation, respectively (Figure 2H). Improved responsiveness to ER stress with XBP1F2 can be attributed to the HR2 domain, which is critical for efficient splicing as it drives ER membrane localization of the XBP1 mRNA.¹⁸ In another EGFPd2 protein expression assay, we measured MFI after incubating EGFPd2 and XBP1F2-EGFPd2 transfected cells with 4μ8C, an IRE1α RNase inhibitor, and IXA4, an IRE1α activator, to corroborate IRE1α dependency of our proposed mechanism.^25,26 While the change in MFI was non-significant across conditions for the cells expressing EGFPd2 without a rheostat element, MFI was higher with 4μ8C incubation and lower with IXA4 incubation compared to DMSO control in XBP1F2-EGFPd2 transfected cells. Co-incubation with 4μ8C and IXA4 resulted in a modest but statistically significant partial expression rescue relative to activator-incubated cells (Figure S2C, D). Although cytosolic EGFP and its derivative EGFPd2 do not stimulate ER stress in their native forms when overexpressed, these studies illustrate that XBP1 mRNA fragments can help regulate protein expression efficiency in an IRE1α-dependent manner under induced ER stress conditions.²⁷

XBP1F splicing can alleviate ER stress response induced by B domain deleted coagulation factor VIII (BDD-FVIII) overexpression.

Factor VIII (FVIII) is an essential blood coagulation glycoprotein whose deficiency results in hemophilia A, an X-linked bleeding disorder. Overexpression of FVIII and its derivatives has been linked to unfolded protein response (UPR) initiation in preclinical studies and hepatotoxicity in clinical studies with AAV-mediated delivery.^9,28–33 The FVIII B domain is expendable and has been excluded from AAV vector cassette design, since full-length FVIII cDNA is above AAV packaging limit (~4.7kb).^28,33 The BDD-FVIII structure is composed of a heavy-chain and a light chain (PDB: 3CDZ) (Figure 3A).^34,35 In our experimental design, the BDD-FVIII sequence encodes a single polypeptide that includes both the heavy and light chain domains of the protein. We integrated XBP1F into BDD-FVIII mRNA to examine off-switch design elements in cell culture models for efficient cytosolic splicing and regulation of ER stress markers. To further probe elements of the off-switch design, we constructed XBP1F3 by elongating XBP1F2 to incorporate the remainder of XBP1u coding sequence, XBP1F4 by swapping HR2 domain in XBP1F3 with N-terminal signal peptide of ER-luminal protein calreticulin, and XBP1F5 by duplicating the XBP1F1 segment in XBP1F3 (Figure 3B, Figure S1A, B).¹⁸ Specifically, doubling the XBP1F1 stretch in XBP1F3 introduces a second XBP1u intron to the switch. We then calculated the percentages of accurately spliced, unspliced, and other transcripts from the RNA sequencing data. We first confirmed XBP1F2-BDD FVIII splicing visually before evaluating all constructs (Figure S3A). Splicing efficiency was ~60% for XBP1F1 and XBP1F2, whereas it was ~50% for XBP1F3 and XBP1F4. Meanwhile, the percentage of unspliced XBP1F transcripts was ~10% higher for XBP1F3 and XBP1F4 than for XBP1F1 and XBP1F2, which accounts for the decrease in splicing efficiency for the former. Since all transcripts containing 20-bp regions on either side of the XBP1u intron are in the dataset, we labeled sequences outside of the top two most abundant ‘XBP1F-spliced’ and ‘XBP1F-unspliced’ as ‘other’. These sequences mainly include sequencing-related SNP errors rather than splicing defects, and therefore are omitted from the analysis (Figure 3C). XBP1F5 was eliminated from subsequent experiments due to the formation of cryptic RNA products (Figure S3B). XBP1F1 and XBP1F2 splicing efficiency was similar (ranging between 50-60%) between Huh7 cells and HEK293 cells. In contrast, XBP1F3 and XBP1F4 splicing efficiency went down to ~20% compared to ~50% in case of HEK293s, implying the inhibitory effect of XBP1u sequence after the HR2 domain on either splicing or BDD-FVIII translation across cell lines (Figure 3D). Thus, XBP1F1 and XBP1F2 are the more promising candidates for the final design.

Figure 3. XBP1F splicing lowers ER stress responses triggered by B domain deleted coagulation factor VIII (BDD-FVIII) overexpression.

(A) BDD-FVIII protein structure (purple: heavy chain, pink: light chain, PDB: 3CDZ) was visualized on ChimeraX.⁴⁸ (B) XBP1F variants were cloned upstream of BDD FVIII. (AP: arrest peptide, CP: calreticulin peptide). 48 hours after plasmid transfection, RNA was collected and converted to cDNA. cDNA was PCR amplified with primers spanning the intron region, and amplicons were subjected to Illumina sequencing for (C) HEK293 (n = 3) and (D) Huh7 (n = 4) cells to calculate percentage of spliced transcripts. RT-qPCR was performed to quantify BiP/GRP78 mRNA expression in (E) HEK293 (n = 5, Thapsigargin n = 3, PEI n = 4) and (F) Huh7 cells (n = 4). As a positive control, non-transfected cells were treated with 1 μM thapsigargin for 6 hours prior to collection. Data is represented relative to RPL19 housekeeping gene expression. BDD-FVIII transcript abundance was calculated from RT-qPCR results in (G) HEK293 (n = 5) and (H) Huh7 cells (n = 5). Data is normalized to RPL19 and reported relative to cells only control. (I) cDNA from non-transfected and transfected HEK293 cells were PCR amplified with primers spanning the endogenous XBP1u intron. Amplicons underwent Illumina sequencing to determine percentage of spliced transcripts (n = 3). One-way ANOVA with Dunnett’s post-test was applied to results for statistical analysis. Where indicated ns = non-significant, * p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001. Data are represented as mean ± SEM in C, D, and I. See also Figures S1 and S3.

Misfolded FVIII and its derivatives bind to BiP/GRP78 in the ER lumen and initiate UPR signaling pathways, thus BDD-FVIII overexpression upregulates the ER chaperone expression.^{9,28,29,36,37} Indeed, in our experiment, BDD-FVIII overexpression elevated BiP/GRP78 mRNA levels 20-fold in HEK293 cells and 4-fold in Huh7 cells relative to the transfection reagent-only controls. In both cell lines, 5′ XBP1F sequences curtail BiP/GRP78 mRNA expression upregulation with XBP1F3 and XBP1F4 having a greater impact than XBP1F1 and XBP1F2 (Figure 3E, F). Variability in BDD-FVIII mRNA expression levels is negligible for all constructs in both cell lines which excludes the likelihood of nonsense-mediated decay (NMD) as a plausible alternative mechanism for ER switch function (Figure 3G, H). Persistent IRE1α activity during ER stress can lead to activation and phosphorylation of c-Jun N-terminal kinase (JNK), which in turn associates with apoptosis regulating proteins.^12,14,27,38 We observed that BDD-FVIII overexpression in HEK293 cells increases phosphorylated JNK protein levels and is mitigated by XBP1F3 and XBP1F4 sequences. (Figure S3C). Approximately 20% of endogenous XBP1 mRNA transcripts exist in the spliced form in HEK293 cells. When BDD-FVIII is overexpressed, this number is tripled and curbed by incorporating 5′ XBP1F elements (Figure 3I). Overall, our results suggest that BDD-FVIII overexpression in cell culture models activates ER transmembrane protein sensor IRE1α, which cleaves XBP1F-BDD FVIII mRNA and attenuates UPR induction, constituting a self-regulated feedback mechanism.

XBP1F regulatory switches can mitigate ER stress response induced by lipid-complexed mRNAs encoding monoclonal antibodies.

First, to assess whether XBP1F regulatory switches can be expanded further to broader applications, we chose Leronlimab, an IgG4 monoclonal antibody to CCR5 that is currently being investigated in clinical trials as anti-HIV therapy, as a model.³⁹ Similar to BDD-FVIII (and more broadly, antibodies in general), the Leronlimab structure predicted by AlphaFold 3⁴⁰ shows heavy chains and light chains each requiring folding and oligomerization within the ER (Figure 4A). XBP1F constructs carrying the Leronlimab sequence split by a ribosomal skipping element between the heavy and the light chains were generated. We anticipated that Leronlimab overexpression and secretion would require accumulation in the ER, thereby triggering UPR and the stress response.

Figure 4. XBP1F switch incorporation can alleviate ER stress response induced by lipid-complexed mRNAs encoding monoclonal antibodies.

(A) Rhesus-optimized Leronlimab protein structure was generated by AlphaFold3 (purple: heavy chain, pink: light chain) and visualized on ChimeraX.⁴⁸ (B) Leronlimab and XBP1F-Leronlimab IVT mRNAs were complexed with a lipid for intracellular delivery. Two ribosomal skipping elements, P2A and T2A (Thosea asigna virus), were included between the Leronlimab heavy and light chains. (C) HEK293 cells were transfected with the RNA-lipid complexes. 24 hours after transfection, RNA was extracted and converted to cDNA. BiP/GRP78 mRNA abundance was analyzed from RT-qPCR data. As a positive control, non-transfected cells were treated with 1 μM thapsigargin for 6 hours prior to collection. Data is represented relative to RPL19 housekeeping gene expression (n = 4, Leronlimab n = 5, XBP1F1 n = 5, XBP1F2 n = 5). (D) Cellular Leronlimab mRNA levels were quantified by RT-qPCR. Data is normalized to RPL19 and reported relative to cells only control (n = 5, XBP1F3 n = 4, XBP1F4 n = 4). (E) Various quantities of Leronlimab and XBP1F1-Leronlimab mRNAs were transfected into HEK293 cells. BiP/GRP78 expression was calculated from RT-qPCR data from RNA collected 24 hours after transfection. Data is represented relative to RPL19 housekeeping gene expression (n = 4). (F) Two sets of mutations known to eliminate XBP1 mRNA splicing were introduced to the XBP1F1-Leronlimab mRNA construct. (G) Mut A-Leronlimab and Mut B-Leronlimab IVT mRNAs were transfected into HEK293 cells along with Leronlimab and XBP1F1-Leronlimab IVT mRNAs. RT-qPCR was performed to measure BiP/GRP78 mRNA abundance in cells 24 hours later. Data is represented relative to RPL19 housekeeping gene expression (n = 5). (H) XBP1F1-Leronlimab and Leronlimab IVT mRNA constructs were transfected into HEK293 cells and media was collected 48 hours after transfection. Secreted Leronlimab was concentrated by Protein A purification and detected on a western blot with an anti-kappa light chain antibody (n = 3). A PEI only control was included as a negative control (n = 1). One-way ANOVA with Dunnett’s post-test was applied to C, D, G and two-way ANOVA with Sidak’s multiple comparison test was applied to E. Where indicated *p≤0.05, **p≤0.01, ****p≤0.0001. See also Figures S1 and S4.

Secondly, as proof-of-concept for the applicability of XBP1F regulatory switches to an mRNA platform, we generated in vitro transcribed Leronlimab and XBP1F-Leronlimab mRNAs (Figure S4A). Transfection of lipid complexed mRNAs into HEK293 cells in vitro (Figure 4B) resulted in ~13-fold increase in BiP/GRP78 mRNA levels compared to the transfection reagent only control. Introducing XBP1F switches into Leronlimab IVT mRNA restored BiP/GRP78 expression to near homeostasis levels. Rapid initiation of Leronlimab protein expression and lack of cellular XBP1F-Leronlimab transcription could be the key drivers of peak BiP/GRP78 mRNA upregulation with Leronlimab IVT mRNA expression and the steep decline with XBP1F incorporation, respectively (Figure 4C). Although XBP1F3-Leronlimab IVT mRNA abundance is slightly below the other constructs, net IVT mRNA quantities in HEK293 cells are generally unaffected (Figure 4D). BiP/GRP78 mRNA expression levels appear closely correlated with increasing amounts of transfected Leronlimab and XBP1F1-Leronlimab IVT mRNAs. Both BiP/GRP78 mRNA abundance and mRNA expression growth rate are larger for Leronlimab than for XBP1F1-Leronlimab at every quantity transfected. Therefore, including the XBP1F1 switch upstream of the Leronlimab IVT mRNA exponentially downregulates BiP/GRP78 mRNA expression (Figure 4E).

We then introduced two sets of point mutations outside and within the XBP1 intron in XBP1F1, named Mut A and Mut B, respectively to abolish splicing as a negative control (Figure 4F).¹⁵ For the purpose of uncoupling the contributions of XBP1F splicing and translational inhibition by XBP1F sequences on BiP/GRP78 mRNA expression, we transfected Leronlimab, XBP1F1-Leronlimab, Mut A-Leronlimab, and Mut B-Leronlimab IVT mRNAs into HEK293 cells. BiP/GRP78 mRNA expression in XBP1F1-Leronlimab IVT mRNA transfected cells was 3-fold lower than Leronlimab, Mut A-Leronlimab, and Mut B-Leronlimab IVT mRNA transfected cells, indicating that XBP1F splicing and not translational inhibition by XBP1F addition was responsible for ER stress mitigation (Figure 4G).

We quantified secreted Leronlimab in HEK293 cell media 48 hours after transfection with Leronlimab and XBP1F1-Leronlimab IVT mRNA constructs by probing Protein A-purified IgG with an anti-κ light chain antibody. XBP1F1-Leronlimab yielded substantially less protein than unfused Leronlimab (Figure 4H, S4C). Moreover, we calculated the percentage of spliced XBP1F1-Leronlimab at 12, 24, and 36 hours after IVT mRNA transfection from RNA sequencing data, enabling direct comparison of protein output with splicing dynamics (Figure S4B). Taken together, these results show that despite modest splicing fractions (approximately 2%, 5%, and 8%, respectively), XBP1F1 is able to regulate protein expression in the context of IVT mRNA overexpression.

XBP1F switches modulate Leronlimab protein expression through feedback control.

We transfected XBP1F-Leronlimab plasmids in vitro with the objective of choosing an optimal switch for in vivo analysis. We first measured the percentages of accurately spliced, unspliced and other XBP1F-Leronlimab transcripts by inputting RNA sequencing files into a custom pipeline. In HEK293 cells, XBP1F splicing efficiencies were 50-75% below the XBP1F-BDD FVIII values. The reduced splicing efficiency observed in this context may reflect differences in the expression context or the nature of ER stress induced by Leronlimab overexpression. (Figure 5A). In Huh7 cells, splicing efficiency was ~45% for XBP1F1 and XBP1F2, while ~14% for XBP1F3 and XBP1F4. The difference in splicing efficiencies between XBP1F-Leronlimab and XBP1F-BDD FVIII mRNAs was smaller in Huh7 cells, denoting variability in XBP1F switch functional parameters depending on the cell type and the therapeutic mRNA (Figure 5B). Consistent with these observations, ER stress response to Leronlimab overexpression was more subdued since BiP/GRP78 mRNA expression in HEK293 cells was upregulated ~9x compared to ~20x for BDD-FVIII overexpression relative to the transfection reagent only controls. Inserting 5′ XBP1F switches preceding Leronlimab mRNA reduced BiP/GRP78 upregulation to varying degrees (Figure 5C). Even though spliced XBP1F1 and XBP1F2 products were more abundant than spliced XBP1F3 and XBP1F4 products, the change in BiP/GRP78 mRNA expression was smaller for the former which points to a non-linear correlation between the two events. Moreover, the decline in BiP/GRP78 mRNA levels were less significant for XBP1F1 and XBP1F2 in Huh7 cells despite more efficient splicing, pointing to cell type effects in switch regulation (Figure 5D). Leronlimab mRNA abundance was comparable across XBP1F switches in HEK293 cells (Figure S5B). However, Leronlimab mRNA quantities were diminished in Huh7 cells either due to mRNA sequence related transcriptional deceleration or impaired mRNA stability dynamics (Figure S5C). In a complementary assay, inhibiting IRE1α’s RNase activity with 4μ8C significantly decreased spliced XBP1F1-Leronlimab population in HEK293 cells, supporting an IRE1α-dependent process (Figure S5A).

Figure 5. XBP1F switches regulate Leronlimab secretion and stabilize ER stress response caused by overexpression of the monoclonal antibody.

Leronlimab and XBP1F-Leronlimab encoding plasmids were transfected into HEK293 and Huh7 cells. 48 hours after plasmid transfection, RNA was extracted and reverse transcribed. cDNA was PCR amplified with primers spanning the intron region, and amplicons were subjected to Illumina sequencing for (A) HEK293 (n = 3) and (B) Huh7 (n = 4) cells to calculate percentage of spliced transcripts. RT-qPCR was performed to measure BiP/GRP78 mRNA expression in (C) HEK293 (n = 5, Leronlimab n = 4, Thapsigargin n = 3, PEI n = 4) and (D) Huh7 cells (n = 4, Leronlimab n = 3, XBP1F1 n = 3). As a positive control, non-transfected cells were treated with 1 μM thapsigargin for 6 hours prior to collection. Data is represented relative to RPL19 housekeeping gene expression. (E) Huh7 cells were transfected with Leronlimab, XBP1F1-Leronlimab, and XBP1F2-Leronlimab. Secreted Leronlimab protein was probed with an Anti-Fc antibody at different timepoints. Total protein expression was assessed by Ponceau S staining. (F) HEK293 cells were transfected with Leronlimab and XBP1F1-Leronlimab. Fresh media with 1 μM thapsigargin or DMSO was added to the wells 48 hours after transfection. Media was changed 6 hours after incubation with the ER stress inducer. In order to measure secreted Leronlimab, media was collected at the time of thapsigargin addition, at the time of media change, and 18 hours after the media change. For analyzing XBP1F1 splicing, cells were harvested at the same timepoints with an additional 3-hour thapsigargin incubation timepoint. (G) RNA was extracted and reverse transcribed. RT-qPCR was performed to quantify unspliced and total XBP1F1-Leronlimab mRNA abundances relative to RPL19. The ratio of unspliced to total transcripts was converted to a percentage and subtracted from 100% to calculate percent spliced XBP1F1-Leronlimab (n = 4). (H) BiP/GRP78 mRNA abundance at 3, 6, and 24-hour timepoints was quantified by RT-qPCR. Data is represented relative to RPL19 housekeeping gene expression (n = 4). (I) Leronlimab protein in media was assessed with an ELISA assay and plotted as a percent of secreted Leronlimab at the time of thapsigargin addition (n = 4). One-way ANOVA with Dunnett’s post-test was applied to C and D. Two-way ANOVA with Sidak’s multiple comparison test was applied to G, H, and I. Data are represented as mean ± SEM in A and B. Where indicated ns = non-significant, *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001. See also Figures S1 and S5.

We also confirmed Leronlimab protein secretion from Huh7 cells. The band corresponding to the IgG Fc fragment was at the correct size and no larger bands were detected, verifying successful P2A ribosomal skipping. Less Leronlimab was secreted from the XBP1F2-Leronlimab construct consistent with its higher XBP1F2 splicing rate and potentially sequence-dependent effects on protein expression. The band at ~50 kDa on the Ponceau S stained total protein gel is albumin from the fetal bovine serum in media (Figure 5E). Reduced secretion from XBP1F2-Leronlimab under continuous culture conditions was verified by probing the media with an anti-κ light chain antibody. At 48 hours post-transfection, comparable levels of secreted light chain were detected in the media from Leronlimab and XBP1F1-Leronlimab transfected cells. However, lower protein accumulation was observed under XBP1F1 control after the media change (Figure S5D). In order to further assess the feedback mechanism of our switch design, we performed a timepoint assay with thapsigargin incubation in HEK293 cells transfected with Leronlimab and XBP1F1-Leronlimab. 48 hours after transfection, we replaced media with thapsigargin- or DMSO-containing fresh media. We replenished the media 6 hours after incubation with the agents. We collected media at the time of thapsigargin addition, after thapsigargin incubation, and 18 hours after the thapsigargin washout. We harvested cells for RNA extraction at the same timepoints, with an additional harvest 3 hours into thapsigargin incubation (Figure 5F). XBP1F1-Leronlimab splicing increased significantly compared to the DMSO control at 3 hours and remained elevated above DMSO at the 6 and 24 hour timepoints, although it declined modestly over time, while splicing remained constant in the DMSO control (Figure 5G). BiP/GRP78 mRNA increased during and following thapsigargin incubation and after washout, while the levels were relatively constant for the DMSO control (Figure 5H). Thapsigargin incubation resulted in a near-complete loss of detectable secreted Leronlimab in the XBP1F1-Leronlimab transfected cells during thapsigargin incubation and after washout. However, XBP1F1-Leronlimab secretion resumed in the DMSO control and reached more than 50% of the baseline expression in some wells (Figure 5I). Because thapsigargin induces global ER stress and broadly perturbs protein folding and secretion, this condition does not isolate switch-specific effects. In the Leronlimab transfected cells, thapsigargin incubation modestly upregulated Leronlimab secretion after thapsigargin incubation compared to the DMSO control, yet secretion was higher for the DMSO control after the washout (Figure S5E). Taken together, this experiment demonstrates that XBP1F switches respond dynamically to ER stress conditions and are associated with reduced protein output under pharmacological stress. However, given the global effects of thapsigargin on ER function, the observed reduction in secretion likely reflects both switch-mediated regulation and broader stress-induced suppression of the secretory pathway.

XBP1F1 regulatory switch retains activity in vivo.

We packaged Leronlimab and XBP1F1-Leronlimab cassettes flanked by inverted terminal repeats (ITRs) into AAV serotype 8 capsids to target the liver (Figure 6A). XBP1F1 rather than other XBP1F derivatives was chosen for in vivo studies due to its reproducible splicing and regulatory activity in therapeutic construct contexts. The mouse ortholog of XBP1 exhibits differences from the human gene; therefore, before delivering the vectors to mice, we transfected C2C12 cells, a mouse myoblast cell line, with XBP1F1-Leronlimab to verify splicing (Figure S5F).

Figure 6. Leronlimab encoding AAV vectors equipped with XBP1F regulatory switches are functional in vivo.

(A) AAV8 vectors packaging CBA-Leronlimab and CBA-XBP1F1-Leronlimab were administered to C57/BL6 mice intravenously. Serum was collected on days 0, 7, and 28. A PBS injected group was included as control. (B) Liver tissue was harvested on day 28, genomic DNA was extracted from liver samples, and vector genomes were quantified by qPCR. CBA-XBP1F1-Leronlimab transgene cassette was isolated from the ITR plasmid and serially diluted to employ as a qPCR standard (n = 5). (C) Serum Leronlimab protein concentration was measured with an ELISA assay (n = 5, repeated serum collection). (D) RNA was isolated from the same liver samples and converted to cDNA. BiP/GRP78 mRNA abundance was determined by RT-qPCR. Data is represented relative to RPL19 housekeeping gene expression (n = 5). (E) cDNA was PCR amplified with primers spanning the XBP1F1 intron. The amplicons underwent Illumina sequencing to calculate the percentage of spliced XBP1F1-Leronlimab mRNA with a custom pipeline (n = 3). Unpaired two-tailed t-test with Welch’s correction was applied to Leronlimab and XBP1F1-Leronlimab in B, two-way ANOVA with Sidak’s multiple comparison test was applied to C, and one-way ANOVA with Dunnett’s post-test was applied to D. Data are represented as mean ± SEM in E. Where indicated ns = non-significant, ***p≤0.001, ****p≤0.0001. See also Figure S1, S5, and S6.

We injected 1.5e13 vg/kg of AAV8 vectors intravenously into 8-10 weeks old C57/BL6 mice via the tail vein and collected serum on days 0, 7, and 28. On day 28, we harvested liver tissue and quantified vector copy numbers, which were comparable in all animals dosed with both AAV constructs (Figure 6B). On day 7, serum Leronlimab protein levels in AAV8-CBA-Leronlimab injected mice were ~5x higher than in AAV8-CBA-XBP1F1-Leronlimab injected mice. At the terminal endpoint on day 28, Leronlimab levels appear equivalent. However, it is important to note the high variability in Leronlimab expression levels and in particular, nearly undetectable expression in two mice with the conventional AAV8-CBA-Leronlimab injected cohort. In contrast, XBP1F1 regulated expression results in a gradual increase in Leronlimab secretion into the serum with minimal variability and potentially increasing the likelihood of sustained long-term expression (Figure 6C). This variability was also reflected at the transcript level, where liver Leronlimab mRNA abundance showed substantial heterogeneity in the AAV8-CBA-Leronlimab cohort, whereas XBP1F1 incorporation resulted in more uniform expression (Figure S6A). BiP/GRP78 levels for both AAV-treated cohorts were comparable to the PBS-injected cohort on day 28 (Figure 6D). Similarly, hepatic CHOP mRNA levels were not significantly different between PBS, Leronlimab, and XBP1F1-Leronlimab treated mice at day 28 (Figure S6B). Approximately 6% of XBP1F1-Leronlimab transcripts were spliced as calculated from the RNA sequencing data, supporting switch engagement despite modest splicing efficiency (Figure 6E). Together, these results support engagement of the XBP1F1 switch in vivo, resulting in attenuated early expression and more uniform serum Leronlimab levels without detectable increases in ER stress markers.

Discussion

In this study, we add to the emerging toolbox of RNA switches for nucleic acid therapies by engineering ER stress responsive RNAs. Our “rheostat” design harnesses conditional splicing of XBP1 mRNA by IRE1α during UPR to control protein expression from different therapeutic transcripts that can often trigger ER stress and deleterious side effects. Importantly, previous research on misfolded membrane proteins, such as ΔF508 cystic fibrosis transmembrane conductance regulator (CFTR), has indicated that UPR is induced only when the accumulation of misfolded proteins exceeds the capacity of ER quality control mechanisms.⁴¹ In line with this, our XBP1F cassette remains unspliced under homeostasis conditions and is activated only when ER stress triggers IRE1α-dependent splicing, which imposes a built-in expression threshold.

We first validated XBP1u-EGFP reporter mRNA splicing by either plasmid or IVT mRNA delivery in vitro. Then, we generated XBP1F switches consisting of XBP1 mRNA functional elements. The shortest spliceable switches, XBP1F1 and XBP1F2, were able to decrease the rate of EGFPd2 expression, validating the responsiveness of our switch to simulated ER stress. Encouraged by the reporter mRNA experiments, we explored model therapeutic proteins to substantiate practical applicability of the XBP1F switches. Plasmid overexpression of BDD-FVIII and Leronlimab, both of which are secreted and heavy/light chain composed, was adequate to trigger UPR and ER stress as evidenced by upregulated BiP/GRP78 mRNA expression, phospho-SAPK/JNK protein expression, and endogenous XBP1 mRNA splicing in vitro. Incorporation of XBP1F switches of different configurations resulted in cytosolic splicing at varying efficiencies for BDD-FVIII and Leronlimab constructs, although all tested constructs mitigated ER stress marker levels in an NMD-independent manner. Perhaps other factors such as translation inhibition by the different RNA structural elements may influence cell type specific outcomes.^12,13,42

We then assessed the applicability of XBP1F switches to lipid complexed mRNA format using Leronlimab as a model. Interestingly, BiP/GRP78 mRNA expression was higher for Leronlimab delivered by IVT mRNA than by plasmid. Faster protein expression kinetics of IVT mRNA resulting in rapid onset of UPR is a plausible explanation for this phenomenon. We postulate this could be a significant contributing factor that attenuates expression efficiency and triggers inflammatory responses. While IVT mRNA is translated immediately after cytosolic delivery, plasmid DNA requires nuclear delivery, transcription and nucleo-cytoplasmic shuttling for translation initiation.⁴³ Additionally, the decrease in BiP/GRP78 mRNA expression with XBP1F insertion was larger for IVT mRNA than for plasmid delivery of Leronlimab. Continuous Leronlimab mRNA transcription from the plasmid leads to more sustained protein expression which could explain relatively elevated BiP/GRP78 mRNA abundance even with XBP1F switches.⁴³ Linearly increasing the amount of transfected Leronlimab and XBP1F1-Leronlimab IVT mRNAs in vitro resulted in an exponential increase in BiP/GRP78 mRNA expression. The rate of increase was higher for Leronlimab IVT mRNA at each interval, emphasizing the potential benefits of XBP1F switches especially when high doses are essential for therapeutic benefit. Mechanistically, we were able to demonstrate that stem loop mutations (Mut A and B) abolished splicing, and were consequently unable to downregulate BiP/GRP78 mRNA expression. Along with bidirectional expression dynamics with 4μ8C and IXA4 incubation, the mutation experiment indicates the regulation is driven by IRE1α-dependent splicing and rather than being explained solely by non-specific translational inhibition from RNA structure. The genetic and chemical controls in the study support the intended mechanism while transcriptomic and proteomic profiling will be valuable for future work.

Previous attempts to minimize deleterious effects of ER stress on therapy efficacy have relied on nucleic acid design optimization, protein engineering, and use of antioxidants or immune modulators.^33,44,45 While these methods may vary for different constructs or be universally applicable in the clinic, our rheostat approach can potentially be utilized as an independent platform and/or in a synergistic manner. In addition to therapeutic modalities focused on liver expression, another exciting prospect for XBP1F switches is the potential to exploit pre-existing ER stress in disease states (e.g., ocular and neurodegenerative diseases including Alzheimer’s and Parkinson’s).⁴⁶ For example, controlled expression of different monoclonal antibodies with our constructs in diseased cells and tissues with ongoing ER stress should be feasible.⁴⁷

Limitations of the study

We acknowledge that several layers of refinement, optimization, and testing are warranted prior to further clinical development. For instance, differences in ER stress involving XBP1 cytosolic splicing between different tissue types (e.g., brain versus liver) and cell types (e.g., neurons vs hepatocytes) require further investigation.¹² Another potential limitation is that ER stress can occur through XBP1/IREα independent pathways and the current platform may not be responsive to other forms of cellular stress.^12–14 In addition, pharmacological modulation of IRE1α activity using small molecules is expected to produce partial and variable changes in RNase activity, rather than complete inhibition or activation. As a result, the dynamic range of reporter-based measurements in these experiments may be limited, and population-averaged readouts may underestimate the extent of regulation occurring at the level of individual transcripts or cells. Furthermore, thapsigargin induces ER stress through global perturbation of ER calcium homeostasis and can affect multiple aspects of protein folding and secretion, which may complicate interpretation of switch-specific effects in stress challenge experiments. Notably, the relationship between XBP1F splicing efficiency and protein output was not directly proportional, as relatively modest increases in splicing were associated with larger reductions in secreted Leronlimab in certain contexts. This indicates that splicing measurements alone do not provide a simple quantitative predictor of protein output. As a result, the current implementation of the XBP1F switch may not support precise, linear tuning of protein expression based solely on measured splicing levels, and the magnitude of the effect may vary depending on context.

Resource Availability

Lead Contact

Further information and requests for resources, reagents, and code must be directed to and will be fulfilled by the lead contact, Aravind Asokan (aravind.asokan@duke.edu).

Materials Availability

All materials stated in this paper are available upon request to the lead contact and upon completion of materials transfer agreement, if relevant.

STAR Methods

Experimental Model and Study Participant Details

Cell Culture

Cell lines are obtained and authenticated by the Duke Cell Culture Facility. The cell lines in the current study are as follows: HEK293 (female), Huh7 (male), C2C12 (female). HEK293 and Huh7 cells were cultured in Dulbecco’s modified Eagle’s medium (4.5 g/L D-glucose, 110 mg/L sodium pyruvate) supplemented with 10% fetal bovine serum (FBS) and 100 mg/mL penicillin, and mg/mL 100 streptomycin. C2C12 mouse myoblasts were cultured in Dulbecco’s modified Eagle’s medium supplemented with 20% fetal bovine serum (FBS) and 100 μg/mL penicillin, and 100 μg/mL streptomycin. All cell lines were maintained at 37°C with 5% CO₂.

Mouse Studies

All mouse procedures were performed in accordance with protocols approved by the Institutional Care and Use Committee (IACUC) at Duke University, accredited by the Association for Assessment and Accreditation of Laboratory Animal Care-International (AAALAC). C57/BL6 mice (male) (Jackson Laboratories; Jax#000664) were maintained at Duke University School of Medicine animal facility with the assistance of Duke’s Division of Laboratory Animal Resources (DLAR). Mice were kept in a temperature-controlled (18–23°C, 40–60% humidity) and enriched environment, with a 12 h light/dark cycle, and provided standard chow and water ad libitum. At 8-10 weeks of age, animals were administered intravenously via the tail vein with PBS or AAV8 at a dose of 1.5e13 vg/kg (n = 5). Blood was collected via cheek bleeding on Days 0, 7, and 28. Mice were sacrificed on day 28 and livers were harvested.

Method Details

Structure determination

BDD FVIII structure was obtained from PDB (PDB: 3CDZ).^34,35 Leronlimab amino acid sequence was submitted to AlphaFold 3 for structure modelling using the default parameters.⁴⁰ ChimeraX software version 1.9⁴⁸ was used to visualize BDD FVIII and Leronlimab protein structures. For both structures purple color indicates the heavy chain, and the pink color indicates the light chain. RNAfold⁴⁹ was used to visualize minimum free energy XBP1F secondary RNA structures using the default parameters.

Molecular cloning

pDONR223_XBP1_WT encoding the protein coding sequence of XBP1u was a gift from Jesse Boehm & William Hahn & David Root.⁵⁰ XBP1F1-4 synthesized with P2A were ordered from Twist Biosciences as gene fragments. Plasmid utilized as a template for plasmid transfection and AAV studies consisted of a pUC19 cassette containing an AAV2 based inverted terminal repeat (ITR) flanked EGFP reporter sequence, which was placed under the control of a CBA promoter and terminated by an SV40 polyA signal. XBP1F1-4 sequences with downstream gene of interest (GOI) were cloned into this vector by digesting it with KpnI-HF and NotI-HF restriction enzymes for 1 hour at 37°C. pCAG-GFPd2 was a kind gift from Connie Cepko and pCDNA4/Full length FVIII was a gift from Robert Peters.^51,52 Rhesus optimized Leronlimab sequence was received from Jonah Sacha and was further modified for improved secretion. Primers with overlapping regions to adjacent fragments were annealed to XBP1u, XBP1F1-4, and GOI by PCR amplification with Q5^® High-Fidelity DNA Polymerase. Digested plasmid backbone and PCR generated fragments were gel purified with Zymoclean Gel DNA Recovery Kit and assembled with NEB High-Fidelity DNA Assembly 2× Master Mix by incubating the reaction for 1 hour at 50°C. Assembled reactions were directly transformed into DH10B chemically competent E.Coli. XBP1F5 featuring vectors were constructed by digesting pTR-CBA-XBP1F3-GOI plasmids with EcoRV-HF and performing NEB HiFi assembly as described with a bridging fragment, including the second XBP1u intron, produced by Integrated DNA Technologies (IDT).

pcDNA3.3_eGFP was a gift from Derrick Rossi.⁵³ Vector backbone was comprised of a T7 promoter, a 5′ untranslated region with a strong Kozak sequence, and an alpha-globin 3′ UTR. Mut A and Mut B synthesized with P2A and homology arms to the backbone and Leronlimab were ordered from IDT as gene blocks. Leronlimab, XBP1F-Leronlimab and plasmid backbone were prepared with overlapping sequences by PCR with Q5^® High-Fidelity DNA Polymerase. Assembly components were gel extracted with Zymoclean Gel DNA Recovery Kit and cloned using NEB High-Fidelity DNA Assembly 2× Master Mix.

IVT mRNA synthesis

Linear DNA templates with T7 promoter and polyT tail for IVT reactions were synthesized by tail PCR with Q5^® High-Fidelity DNA Polymerase.⁵⁴ Desired products were isolated with Zymoclean Gel DNA Recovery Kit, followed by a second cleanup step with DNA Clean & Concentrator-25. MEGAscript^™ T7 Transcription Kit was utilized in the production of mRNAs. 250 ng PCR product, 2 μl 1:5 dilution of GTP solution, and 3 μl 40 mM 3′-O-Me-m7G(5′)ppp(5′)G RNA Cap Structure Analog were combined with other reagents in a 20 μl reaction. IVT reaction mixtures were gently mixed by pipetting, incubated at 37°C for 3 hours, and then treated with TURBO DNase provided with the MEGAscript^™ T7 Transcription Kit for 20 minutes at 37°C to remove residual linear DNA template. IVT reactions were purified with MEGAclear^™ Transcription Clean-Up Kit according to manufacturer’s instructions. 2 μg aliquots of IVT mRNA samples were mixed with RNA loading dye and incubated at 90°C for 2 minutes, after which they were run on a 1% agarose gel supplemented 1% bleach.

Plasmid and IVT mRNA transfections

150,000 cells per well were seeded in 24-well culture plates, while 750,000 cells per well were seeded in 6-well culture plates for HEK293 experiments. Huh7 cells were seeded at 50,000 cells per well in 24-well tissue culture plates and at 100,000 cells per well in 12-well tissue culture plates. C2C12 cells were seeded at 50,000 cells per well in a 24-well tissue culture plate. 800 ng or 4 μg plasmid DNA was diluted in 25 μl or 125 μl Dulbecco’s modified Eagle’s medium for 24-well or 6-well tissue culture plate transfections for HEK293 cells, respectively. 3:1 (Volume of 1 mg/mL PEI: Amount of DNA in μg) polyethylenimine (PEI) was diluted in the same volume of Dulbecco’s modified Eagle’s medium in a second tube and combined with the plasmid solution. Mixture was briefly vortexed and incubated at room temperature (RT) for 5 minutes prior to being added to wells dropwise. For transfection reagent only wells, PEI diluted in Dulbecco’s modified Eagle’s medium was added to wells in the absence of plasmid DNA. The same volumes and incubation conditions as DNA transfections were maintained.

For Huh7 and C2C12 24-well plate transfections, first 850 ng plasmid DNA, then 2.5 ul TransIT^®-LT1 Transfection Reagent was added to 50 ul Opti-MEM^™ I Reduced-Serum Medium and mixed gently by pipetting. For 12-well plate transfections, 1400 ng plasmid DNA and 4.2 μl TransIT^®-LT1 Transfection Reagent were added to 100 ul Opti-MEM^™ I Reduced-Serum Medium and mixed gently by pipetting. Solution was incubated at RT for 20 minutes and dispensed dropwise to different areas of the well. For transfection reagent only wells, TransIT^®-LT1 Transfection Reagent diluted in Opti-MEM^™ I Reduced-Serum Medium was added to wells in the absence of plasmid DNA. The same volumes and incubation conditions as DNA transfections were maintained.

IVT mRNA was prepared in 25 or 125 μl Opti-MEM^™ I Reduced-Serum Medium (24-well or 6-well tissue culture plates). In a separate tube, 2.75 μl or 10 μl Lipofectamine^™ MessengerMAX mRNA Transfection Reagent (24-well or 6-well tissue culture plates) per μg mRNA was diluted in 25 μl or 125 μl Opti-MEM^™ I Reduced-Serum Medium (24-well or 6-well tissue culture plates) and incubated at RT for 10 minutes. Contents of the first tube was combined with the second tube, gently mixed, and incubated at RT for 5 minutes. mRNA-lipid complexes were added to cells dropwise. For transfection reagent only wells, Lipofectamine^™ MessengerMAX mRNA Transfection Reagent diluted in Opti-MEM^™ I Reduced-Serum Medium was added to wells in the absence of IVT mRNA. The same volumes and incubation conditions as IVT mRNA transfections were maintained.

RNA extractions for cell culture experiments

RNA was extracted 48 hours after plasmid transfections and 24 hours after IVT mRNA transfections unless otherwise stated. For cell culture experiments, RNA was isolated with Quick-RNA Miniprep Kit following the manufacturer’s protocol. RNA was eluted in nuclease-free water and stored in −80°C until use.

Gene expression quantitation by RT-qPCR

500 ng (HEK293,C2C12, and mouse experiments) or 400 ng (Huh7 experiments) total RNA was reverse transcribed using High-Capacity RNA-to-cDNA Kit in a 20 μl reaction mixture. cDNA was diluted 10-fold with nuclease-free water. 2 μl diluted cDNA solution was used as input into qPCR reactions with target specific primer sets. qPCR was performed using PowerUp^™ SYBR^™ Green Master Mix and QuantStudio 3 real-time PCR system (Thermo Fisher Scientific) with a three-step reaction, according to manufacturer’s directions. ΔCt or ΔΔCt method was applied for the analysis of RT-qPCR results. For ΔCt, target mRNA expression was normalized to RPL19 mRNA. For ΔΔCt, results were normalized to the quantity of mRNA of interest in cells-only control samples. For percent XBP1F splicing calculations, unspliced and total Leronlimab abundance was determined by RT-qPCR relative to RPL19. The ratio of unspliced to total transcript quantity was converted to a percentage and subtracted from 100%.

XBP1u or XBP1F-GOI splicing gel electrophoresis

cDNA from pTR-CBA-XBP1u or XBP1F-GOI plasmid and XBP1u or XBP1F-GOI IVT mRNA transfection experiments was subjected to PCR amplification with Q5^® High-Fidelity DNA Polymerase in a three-step reaction. 30 μl PCR product was separated on a 3% agarose gel. RPL19 was used as a housekeeping control. For XBP1F-EGFPd2 splicing analysis, 1μM thapsigargin or DMSO was added to wells 48 hours after transfection and RNA was collected 3 hours after incubation, reverse transcribed, and subjected to PCR and gel electrophoresis.

Flow cytometry

1μM thapsigargin, 10 μM IXA4, 50 μM 4μ8C, and/or DMSO was added to wells 48 hours after transfection with pTR-CBA-XBP1F-EGFPd2 or pTR-CBA-EGFPd2. At indicated timepoints, cells were dissociated with 100 μl 0.05% Trypsin-EDTA and were resuspended in 500 μl 1× phosphate-buffered saline supplemented with 10% fetal bovine serum. Cell suspension was passed through a 35 μm cell strainer and flow analyzed using a BD LSRFortessa^™ X-20 Cell Analyzer. Results were quantitated with FlowJo. For kinetic EGFP measurements, percent change in MFI was calculated with respect to MFI at Time 0 for each construct. For the single timepoint EGFP measurements, MFI was calculated with respect to the DMSO control.

RNA sequencing

Target PCR fragments were generated by PCR amplification of cDNA with Q5^® High-Fidelity DNA Polymerase in a three-step reaction. PCR reactions were electrophoresed on a 2% agarose gel, after which bands of interest were gel purified using Zymoclean Gel DNA Recovery Kit and submitted for amplicon sequencing on an Illumina^™ HiSeq^® with 250 bp paired-end reads (Genewiz). Resulting forward read (R1) fastq files were analyzed with a custom Python script designed to identify the presence of XBP1u intron sequence within the filtered data and to create an output file that lists each unique sequence and its count in the dataset. Top reads were aligned on UGENE software.⁵⁵ Percentages of spliced XBP1F transcripts, unspliced XBP1F transcripts, and other transcripts were calculated by (Number of transcripts / Number of total transcripts) * 100%.

Secreted Leronlimab Purification

Pierce^™ Protein A Agarose was utilized for purifying Leronlimab protein from cell media in the IVT mRNA experiment. Beaded agarose resin was resuspended in 1× phosphate-buffered saline according to the manufacturer’s instructions. Briefly, 2 mL media was centrifuged at 8,000g for 3 minutes to eliminate cell debris. 100 μl resin was added to the media, mixed gently, and incubated at RT for 3 hours. Supernatant was removed after centrifugation (800g for 1 minute), and beads were washed x2 with 1× phosphate-buffered saline. IgG antibodies were eluted in 250 μl 0.1 M citric acid pH 3.0 by gently pipetting and briefly spinning down the beads. Eluate was neutralized with 25 μl 1 M Tris pH 8.0.

Western blotting

RIPA lysis buffer containing Halt^™ Protease Inhibitor Cocktail was used to extract proteins from HEK293 cells. Protein concentration in lysates was determined by performing a Bradford Assay with Pierce^™ Detergent Compatible Bradford Assay Reagent. For secreted Leronlimab blots, 7.5 μl cell media or 12 μl concentrated IgG eluate was analyzed for plasmid and IVT mRNA overexpression experiments, respectively. Equal protein or volume amount of each sample was denatured at 95°C for 5 minutes in NuPAGE^™ LDS Sample Buffer and 2.5% β-mercaptoethanol, separated by SDS-PAGE on a Mini-PROTEAN^® TGX^™ Precast Gel (Bio-Rad) at 100 V for 1 hour, and transferred to a Trans-Blot Turbo Midi 0.2 μm PVDF Transfer membrane (Bio-Rad). Membranes were blocked for 1 hour with 5% milk powder in Tris-buffered saline with 0.1% Tween 20 (TBST). For secreted Leronlimab blots, membranes were stained with Ponceau S Staining Solution, imaged, and washed with TBST before blocking. After blocking, membranes were blotted overnight at 4°C with primary antibodies against Vinculin (1:1,000) or phosphorylated stress-activated protein kinase/Jun-amino-terminal kinase SAPK/JNK (Phospho-SAPK/JNK) (1:1,000). After 3x wash with TBST, membranes were incubated with stabilized horseradish peroxidase-conjugated goat anti-rabbit antibody for 1 hour at RT and 3x wash was repeated. For Leronlimab detection, HRP-conjugated secondary antibodies to the IgG Fc domain (1:10,000) or to the IgG κ light-chain within the Fab region (1:1,000) were used. After 1-hour incubation with the antibodies at RT, membranes were washed 3x with TBST. Blots were developed with SuperSignal^™ West Femto Maximum Sensitivity Substrate and imaged. Band intensities were quantified by Image Lab.

Secreted Leronlimab ELISA

Blood samples were incubated at room temperature for 10 minutes after collection and centrifuged at 10,000g for 10 minutes to separate out serum. Cell media was centrifuged at 8,000g for 3 minutes to remove cell debris. Enzyme-linked immunosorbent assay (ELISA) was used to detect free Leronlimab in plasma and in cell media. Half-area 96-well Costar Assay Plates (Corning) were coated with the anti-idiotype antibody PA-22, provided by CytoDyn, at 1.5 μg/mL in carbonate-bicarbonate buffer (Thermo Fisher) and incubated overnight. Plates were washed three times with PBS-T (PBS + 0.1% Tween-20) and blocked with blocking buffer (PBS + 0.4% Tween-20 + 10% BSA) for at least 2 h at RT. Leronlimab concentration was calculated with a standard curve using Leronlimab dilutions ranging from 13.72-10,000 ng/mL. Heat-inactivated plasma samples and cell media were also diluted with blocking buffer. After incubating for 30 min at RT, plates were washed three times with 0.5 M NaCl in PBS and incubated with anti-rhesus IgG4 [tessera]-HRP diluted 500-fold in blocking buffer for 30 min at RT. Plate was washed three times with PBS-T and developed for 2 min using 3,3′,5,5′-tetramethylbenzidine (TMB) substrate (Southern Biotech). Reaction was stopped with 1 N H2SO4, Plates were read on the Synergy HTX Milti-Mode Microplate Reader (BioTek) and data were collected using software Gen5 v3.09 at two absorbance wavelengths: 650 nm for the developing reaction and 450 nm for the developed reaction after the reaction was stopped with 1 N H2SO4. Final OD was determined by OD450 nm minus OD650 nm. The limit of detection for the assay is 22.5 ng/mL. For cell culture experiments, results are reported as percent of secreted Leronlimab 48 hours after transfection.

Recombinant AAV vector production

Recombinant AAV vectors were produced by a triple plasmid transfection method comprised of an adenovirus helper plasmid pXX680 (0.60 μg/mL), an AAV8 Rep/Cap plasmid (0.50 μg/mL), and an ITR plasmid (0.30μg/mL) in suspension HEK293 cells.⁵⁶,⁵⁷ Cells were transfected with PEI at 1 to 3 plasmid to PEI ratio and maintained in a 37°C CO₂ incubator on an orbital shaker. 6 days after transfection, cells were discarded after centrifugation at 2,000 g for 10 min and media was incubated with 12% PEG-8000 (Millipore Sigma) overnight at 4°C. PEG-media was centrifuged at 4,000 g for 1 h and the virus containing pellet was resuspended in 1× phosphate-buffered saline containing 1 mM MgCl2 and 0.001%F-68. Vectors were purified via an iodixanol gradient ultracentrifugation (30,000 rpm for 15 hours) followed by concentration with Pierce Protein Concentrator PES columns (100K MWCO, 5-20 mL, Thermo Scientific). The iodixanol gradient was prepared by layering 3 mL 17% iodixanol, 3 mL 25% iodixanol, 4 mL 40% iodixanol, and 3 mL 60% iodixanol in a clear ultracentrifuge tube (Beckman-Coulter) with the vector prep as the top layer. Vector genome titers were measured on a QIAcuity One digital PCR system (Qiagen) by performing a QIAcuity Probe PCR assay according to the manufacturer’s instructions with primers designed to specifically bind AAV2 ITRs.

RNA and genomic DNA (gDNA) extractions for the mouse study

RNA and gDNA from the same tissue piece were isolated with Quick-DNA/RNA Miniprep Plus Kit following the manufacturer’s instructions. RNA and gDNA eluted in nuclease-free water were stored in −80°C and −20°C, respectively.

Vector copy number analysis

The qPCR standard was generated by digesting CBA-XBP1F1-Leronlimab ITR plasmid to isolate the transgene. The linearized fragment was gel purified using Zymoclean Gel DNA Recovery Kit according to manufacturer’s instructions. 100 ng standard was serially diluted 1:10 to construct a standard curve. Genomic DNA was diluted to 50 ng/μl. 2 μl diluted gDNA or standard was used as input into the qPCR reaction with a transgene specific primer set. qPCR was performed using PowerUp^™ SYBR^™ Green Master Mix and QuantStudio 3 real-time PCR system (Thermo Fisher Scientific) with a three-step reaction, according to manufacturer’s directions. Vector genome copy numbers were quantified by linear regression and normalized per microgram DNA.

Quantification and Statistical Analysis

Individual experiment quantifications are reported in the Method Details section and figure legends. Statistical analysis was performed using GraphPad Prism software. Unpaired two-tailed t-test was applied for determining significance for datasets with two groups. For datasets with three groups or more, statistical analysis was carried out using one-way ANOVA with Dunnett’s post-test or Tukey’s post-test. When two or more independent variables were present, two-way ANOVA with Sidak’s multiple comparison test was conducted. The number of biological replicates (n) is indicated in the corresponding figure legends. Where indicated, ns = not significant, * = p≤0.05, ** = p≤0.01, *** = p≤0.001, **** = p≤0.0001. Box and whisker plots demarcate median, minimum, and maximum data points.

Significance.

Overexpression of therapeutic proteins can overwhelm cellular proteostasis networks, limiting the safety and durability of gene and mRNA therapies. Current approaches largely focus on optimizing transgenes or delivery systems rather than dynamically adapting expression to the physiological state of the cell. Here, we repurpose the endogenous IRE1α–XBP1 stress-response pathway to create RNA rheostats that sense ER stress and attenuate transgene expression through stress-induced RNA splicing. By coupling therapeutic protein output to cellular folding capacity, this strategy establishes a programmable feedback mechanism that mitigates ER stress while preserving expression under homeostatic conditions. These findings introduce a broadly applicable framework for engineering self-regulating gene expression systems for nucleic acid therapeutics.

Highlights.

• XBP1-derived RNA elements sense ER stress and regulate transgene expression

• IRE1α-mediated splicing attenuates protein output under ER stress conditions

• RNA rheostats reduce ER stress caused by therapeutic protein overexpression

• Stress-responsive control improves the safety of gene and mRNA therapies

Data Availability

Data have been deposited at the Duke Research Data Repository and are publicly available as of the date of publication. The access link/DOI is listed in the key resources table.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit monoclonal anti-Vinculin (Clone ID: E1E9V)	Cell Signaling Technology	Cat#13901; RRID: AB_2728768
Rabbit polyclonal anti-Phospho-SAPK/JNK (Thr183/Tyr185)	Cell Signaling Technology	Cat#9251; RRID: AB_331659
Goat polyclonal anti-Human IgG Fc Cross-Adsorbed Secondary Antibody, HRP	Thermo Fisher Scientific	Catalog#A18823; RRID: AB_2535600
Goat anti-Human Kappa Light Chain Secondary Antibody, HRP	Thermo Fisher Scientific	Catalog#A18853; AB_2535630
Anti-idiotype antibody PA-22	CytoDyn	N/A
Anti-rhesus IgG4 [tessera]-HRP (Clone ID: [7A8])	NIH Nonhuman Primate Reagent Resource	PR-7184; RRID: AB_2819324
Goat Anti-Rabbit Ig, Human ads-HRP	SouthernBiotech	Cat. No.: 4010-05
Bacterial and virus strains
AAV8, CBA-Leronlimab	This paper	N/A
AAV8, CBA-XBP1F1-Leronlimab	This paper	N/A
DH10B Competent Cells	Thermo Fisher Scientific	Catalog Number EC0113
Chemicals, peptides, and recombinant proteins
Thapsigargin	Tocris Bioscience	Cat. No.1138
IXA4	MedChemExpress	Cat. No.: HY-139214
4μ8C	MedChemExpress	Cat. No.: HY-19707
Ponceau S Staining Solution	Cell Signaling Technology	Cat#59803
KpnI-HF	New England Biolabs	Catalog#R3142
NotI-HF	New England Biolabs	Catalog#R3189
EcoRV-HF	New England Biolabs	Catalog#R3195
Q5® High-Fidelity DNA Polymerase	New England Biolabs	Catalog#M0491
3′-O-Me-m7G(5′)ppp(5′)G RNA Cap Structure Analog	New England Biolabs	Catalog#S1411
RNA Loading Dye, (2X)	New England Biolabs	Catalog#B0363S
Polyethylenimine (PEI MAX)	Polysciences	Product Number 24765
TransIT®-LT1 Transfection Reagent	Mirus Bio	MIR 2304
Lipofectamine™ MessengerMAX mRNA Transfection Reagent	Thermo Fischer Scientific	Catalog number LMRNA001
RIPA Buffer	Millipore Sigma	R0278
Halt™ Protease Inhibitor Cocktail (100X)	Thermo Fisher Scientific	Catalog number 78429
Pierce™ Detergent Compatible Bradford Assay Reagent	Thermo Fisher Scientific	Catalog number 23246
NuPAGE™ LDS Sample Buffer (4X)	Thermo Fisher Scientific	Catalog number NP0007
β-mercaptoethanol	Millipore Sigma	Product No. 63689
TWEEN® 20	Millipore Sigma	Product No. P7949
Opti-MEM™ I Reduced-Serum Medium	Thermo Fisher Scientific	Catalog number 31985070
Trypsin-EDTA (0.05%), phenol red	Thermo Fisher Scientific	Catalog number 25300120
DPBS, no calcium, no magnesium	Thermo Fisher Scientific	Catalog number 14190250
Critical commercial assays
Zymoclean Gel DNA Recovery Kit	Zymo Research	Cat#4001
NEBuilder® HiFi DNA Assembly Master Mix	New England Biolabs	Catalog#E2621
DNA Clean & Concentrator-25	Zymo Research	Cat#D4005
MEGAscript™ T7 Transcription Kit	Thermo Fisher Scientific	Catalog Number AM1334
MEGAclear™ Transcription Clean-Up Kit	Thermo Fisher Scientific	Catalog Number AM1908
Quick-RNA Miniprep Kit	Zymo Research	Cat#R1055
Quick-DNA/RNA Miniprep Plus Kit	Zymo Research	Cat#D7003
High-Capacity RNA-to-cDNA™ Kit	Thermo Fisher Scientific	Catalog Number 4388950
PowerUp™ SYBR™ Green Master Mix for qPCR	Thermo Fisher Scientific	Catalog Number A25742
SuperSignal™ West Femto Maximum Sensitivity Substrate	Thermo Fisher Scientific	Catalog Number 34095
Pierce™ Protein A Agarose	Thermo Fisher Scientific	Catalog Number 20334
QIAcuity Probe PCR Kit	Qiagen	Cat. No. 250101
Deposited Data & Code
BDD FVIII Structure	PDB	PDB: 3CDZ, https://www.rcsb.org/structure/3CDZ
NGS RNA sequencing data	This paper	https://research.repository.duke.edu/record/541?&ln=en
Python script code for RNA seq data	This paper	https://research.repository.duke.edu/record/541?&ln=en
Experimental models: Cell lines
HEK293	ATCC	CRL-1573
Huh7	CLS	300156
C2C12	ATCC	CRL-1772
Experimental models: Organisms/strains
C57/BL6 mice	Jackson Laboratories	000664
Oligonucleotides
See table S1 for the List of Primers Used in the Study	This paper	N/A
Recombinant DNA
pDONR223_XBP1_WT	Kim et al.50	Addgene plasmid #82168
pCAG-GFPd2	Matsuda et al.51	Addgene plasmid #14760
pCDNA4/Full length FVIII	Peters et al.52	Addgene plasmid #41036
pcDNA3.3_eGFP	Warren et al.53	Addgene plasmid #26822
pTR-CBA-XBP1u-EGFP	This paper	N/A
pcDNA3.3-XBP1u-EGFP	This paper	N/A
pTR-CBA-XBP1F1-EGFPd2	This paper	N/A
pTR-CBA-XBP1F2-EGFPd2	This paper	N/A
pTR-CBA-EGFPd2	This paper	N/A
pTR-CBA-BDD FVIII	This paper	N/A
pTR-CBA-XBP1F1-BDD FVIII	This paper	N/A
pTR-CBA-XBP1F2-BDD FVIII	This paper	N/A
pTR-CBA-XBP1F3-BDD FVIII	This paper	N/A
pTR-CBA-XBP1F4-BDD FVIII	This paper	N/A
pTR-CBA-XBP1F5-BDD FVIII	This paper	N/A
pcDNA3.3-Leronlimab	This paper	N/A
pcDNA3.3-XBP1F1-Leronlimab	This paper	N/A
pcDNA3.3-XBP1F2-Leronlimab	This paper	N/A
pcDNA3.3-XBP1F3-Leronlimab	This paper	N/A
pcDNA3.3-XBP1F4-Leronlimab	This paper	N/A
pcDNA3.3-Mut A-Leronlimab	This paper	N/A
pcDNA3.3-Mut B-Leronlimab	This paper	N/A
pTR-CBA-Leronlimab	This paper	N/A
pTR-CBA-XBP1F1-Leronlimab	This paper	N/A
pTR-CBA-XBP1F2-Leronlimab	This paper	N/A
pTR-CBA-XBP1F3-Leronlimab	This paper	N/A
pTR-CBA-XBP1F4-Leronlimab	This paper	N/A
pXX680	This paper	N/A
pLH8	This paper	N/A
Software and algorithms
GraphPad Prism software, version 10.4.1	GraphPad software	https://www.graphpad.com/
ChimeraX software, version 1.9	Meng et al.48	https://doi.org/10.1002/pro.4792; https://www.rbvi.ucsf.edu/chimerax/
Image Lab Software	Bio-Rad	https://www.bio-rad.com/en-us/product/image-lab-software
Gen5, version 3.09	Agilent Technologies	https://www.agilent.com/en/product/cell-analysis/cell-imaging-microscopy/cell-imaging-microscopy-software/biotek-gen5-software-for-imaging-microscopy-1623226
FlowJo Software, version 10.9.0	BD Life Sciences	https://www.flowjo.com/
RNAfold	Lorenz et al.49	https://doi.org/10.1186/1748-7188-6-26; http://rna.tbi.univie.ac.at/cgi-bin/RNAWebSuite/RNAfold.cgi
AlphaFold 3	Abramson et al.40	https://doi.org/10.1038/s41586-024-07487-w; https://alphafoldserver.com/
Unipro UGENE software, version 48.1	Okonechnikov et al.56	https://ugene.net/

Code Availability

This study contains a custom Python script for analyzing NGS data and has been deposited at Duke Research Data Repository. The code is publicly accessible. The access link/DOI is provided in the key resources table.

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