Segmented poly(A) tails with microRNA target sites confer tissue-specific regulation for mRNA therapeutics

Abstract

Targeted delivery and controlled expression of mRNA encapsulated by lipid nanoparticles (mRNA-LNPs) are critical for the development of safe and effective mRNA medicines. However, efficient post-delivery regulation of mRNA-LNP expression remains challenging, and conventional approaches largely rely on modifying the 3′ UTR at the expense of other regulatory elements. In this study, we engineered segmented poly(A) tail variants that function as gene-specific regulatory elements for synthetic mRNAs, providing an alternative regulatory module that preserves UTR integrity. Insertion of miR-122 or miR-142 target sites (MTSs) at various positions within the poly(A) tail suppressed luciferase expression in non-target cells in vitro and in vivo in a position-dependent manner. Furthermore, by incorporating triple-MTS sequences for miR-142, miR-126, and miR-148a in all possible combinations at the 5′ end of the poly(A) tail, we identified specific arrangements that simultaneously reduced luciferase activity in three non-target hepatic cell types while preserving robust expression in hepatocytes. MTS insertion order proved critical for optimal silencing, highlighting a design parameter not observed in dual-MTS constructs. These findings establish the poly(A) tail as a programmable platform for cell-type-selective regulation, complementing tissue-tropic LNPs and expanding the toolbox for mRNA therapeutic design.

Graphical abstract

Zhang and colleagues engineered the poly(A) tail as a programmable regulatory element, showing that embedding cell-type-specific microRNA target sites directly within it confers robust, position-dependent silencing in off-target tissues while preserving activity in target cells. This strategy offers a new modular tool to enhance mRNA therapeutic safety and precision.

Introduction

Messenger RNA (mRNA)-based technology demonstrated the huge potential as a platform for vaccines, protein replacement, and gene editing. The success of SARS-CoV-2 mRNA vaccines by Moderna and Pfizer/BioNTech highlights the utility of antigen-encoding mRNA encapsulated by lipid nanoparticles (LNPs) to elicit robust immune responses.^1,2 Beyond vaccines, mRNA-LNP technology has also been actively explored for treating rare genetic disorders and other diseases by delivering mRNAs encoding therapeutic proteins or gene-editing machinery.^3,4

However, precise spatiotemporal control of mRNA expression remains a critical unresolved problem.⁵ Off-target expression can contribute to immune reactogenicity in vaccines (e.g., leaky expression of SARS-CoV-2 Spike protein in liver after intramuscular administration of mRNA vaccine) or toxicity in gene therapies (e.g., ectopic expression of gene editing tool in off-target tissues or cell types within the correct tissue).⁶ Although advances in lipid discovery and LNP formulation facilitated the successful development of mRNA vaccines, achieving extrahepatic delivery or cell-type-specific control within hepatic tissues has been challenging due to an incomplete understanding of LNP biodistribution and cellular uptake mechanisms for different organs and tissues.⁷ (4S)-KEL12 is a ketal-ester ionizable lipid we previously developed that exhibits reduced hepatotoxicity, enhanced spleen tropism, and potent mRNA delivery capacity in preclinical models. This LNP platform is used to deliver an HPV16/18 therapeutic mRNA vaccine, RG002, that received IND clearance from both the US FDA (October 2023) and China CDE and is currently in Phase 1/2 clinical trials for HPV16/18-associated CIN2/3 (NCT06273553; CTR20251020). To our knowledge, RG002 is the first mRNA-LNP therapeutic vaccine to receive IND approval for this indication globally, underscoring the clinical relevance of the KEL12-based LNP delivery platform used in this study. Thus, while LNP-based hepatic delivery has been achieved, strategies to restrict expression to specific hepatic cell types (e.g., hepatocytes versus non-parenchymal cells) remain a challenge that our poly(A) MTS platform addresses by engineering the mRNA cargo itself.

To address this limitation, a promising complementary strategy is to engineer the mRNA cargos to enhance tissue specificity.⁸ Synthetic mRNAs typically consist of an open reading frame (ORF) encoding the protein of interest and several essential regulatory elements, including 5′ cap, 5′ untranslated region (UTR), 3′ UTR, and a poly(A) tail for optimized expression.⁹ Particularly, the 3′ UTR naturally contains microRNA (miRNA) target sites (MTSs) and other regulatory motifs that control mRNA stability and translation.¹⁰ miRNAs are a class of small non-coding RNAs showing tissue- and cell type-specific patterns.^11,12 After being incorporated into the RNA-induced silencing complex (RISC), they can mediate mRNA degradation or translational repression by binding complementary sequences in target mRNAs.¹³

The miRNA-based mechanism has been utilized for tissue-selective expression of therapeutic mRNAs.¹⁴ By incorporating MTS into the 3′ UTR, several groups have successfully reduced off-target expression in mice. For instance, miR-122 is abundant in hepatocytes but absent in hematopoietic cells,¹⁵ while miR-142 shows the opposite pattern.¹⁶ miR-122 or miR-142 MTS inserted in the 3′ UTR showed significant silencing effects on mRNA expression in liver and spleen, respectively.¹⁷ Others have placed MTS in the 5′ UTR or ORF to achieve similar effects.^14,18 Circular RNAs with miRNA-responsive internal ribosome entry sites (IRES) further demonstrated the potential of miRNA-mediated regulation.¹⁹ Collectively, these studies underscored the utility of endogenous miRNA networks for mRNA therapeutics.

While the poly(A) tail is traditionally viewed as a passive stabilizer, recent work suggested that engineered poly(A) variants could unlock new regulatory dimensions. In eukaryotes, the poly(A) tail protects mRNA from exonuclease degradation and recruits poly(A)-binding protein (PABP) to enhance translation initiation.²⁰ Although natural poly(A) tails lack gene-specific functions, their length and sequence can be modulated to facilitate mRNA manufacturing and protein yield. For example, a 120 nt poly(A) tail maximizes protein expression compared with those of shorter lengths,²¹ while a segmented A30-70 poly(A) tail variant developed by BioNTech improves template DNA stability in bacteria for IVT mRNA production.²² Cytosine-rich spacers or branched poly(A) structures have been shown to boost mRNA stability and expression.^23,24 However, these poly(A) engineering efforts primarily focus on preserving or enhancing its canonical functions for mRNA stabilization and translational regulation. Therefore, tissue-specific expression regulation through engineered poly(A) tail represents an excellent opportunity for expanding the regulatory capacity of mRNA medicines.

In this study, we modified the poly(A) tail into a novel miRNA-responsive regulatory element. We designed segmented poly(A) tail variants with embedded miR-122 or miR-142 target sites. These poly(A) variants potently suppressed mRNA expression in non-target tissues with efficacies impacted by their sites of insertion, while maintaining robust activity in desired organs in vivo. Furthermore, by sequentially incorporating multiple distinct MTS at the 5′ end of poly(A), we achieved simultaneous silencing effects in multiple hepatic cell types while allowing robust expression in cultured primary hepatocytes. This modular and straightforward platform holds promise for tissue-specific expression for safe mRNA medicines.

Results

Insertion of MTSs in poly(A) tail suppresses mRNA expression in cultured cells

To systematically evaluate the silencing effects of MTSs at different 3′ terminal positions, we engineered firefly luciferase (fLuc) mRNAs containing either miR-122 or miR-142 MTS downstream of the α-globin 3′ UTR. These MTS were inserted at five distinct positions (0A, 14A, 19A, 30A, and 60A) within a segmented RG2 poly(A) variant,²⁵ generating constructs designated as 122MTS-0A, 122MTS-14A, 122MTS-19A, 122MTS-30A, and 122MTS-60A, and their 142MTS counterparts (142MTS-0A, 142MTS-14A, 142MTS-19A, 142MTS-30A, and 142MTS-60A) (Figure 1A). Control constructs included one that lacks any MTS sequences (hereafter referred to as noMTS) and two additional ones with either miR-122 or miR-142 MTS in the α-globin 3′ UTR (hereafter referred to as 122MTS-3UTR and 142MTS-3UTR, respectively) as previously reported.¹⁷

Figure 1.

Single miR-122 and miR-142 target site insertions in the poly(A) tail of synthetic mRNAs specifically downregulate luciferase activity in non-target cell types, respectively

(A) Schematic diagram of miRNA target site (MTS) insertions for miR-122 and miR-142 at different positions of 3′ UTR and poly(A) downstream of the luciferase ORF. Only the 3′ UTR and the poly(A) regions were depicted. No insertion control (noMTS) and insertions in 3′ UTR (MTS-3UTR) or at five distinct positions of poly(A) were indicated as noMTS, MTS-3UTR, -0A, −14A, −19A, −30A, and −60A, respectively. (B) miR-122 target site insertions at distinct positions of the poly(A) tail effectively downregulate luciferase activity in primary mouse hepatocytes (PMHs) compared with miR-142 target site insertions at the same locations or noMTS control. Bar graph shows the means of relative unit of luminescence (RLU) with SEM. (C) miR-142 target site insertions at distinct positions of the poly(A) tail effectively downregulate luciferase activity in RAW264.7 macrophages compared with miR-122 target site insertions at the same locations or noMTS control. Bar graph shows the mean of RLU with SEM. ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001 (one-way ANOVA).

We determined the impact of a single miR-122 or miR-142 target site inserted in the poly(A) tail on luciferase activity using primary mouse hepatocytes (PMHs; high miR-122 and low miR-142) and RAW264.7 macrophages (low miR-122 and high miR-142), based on their reciprocal expression patterns for the two miRNAs.^15,16 In PMH cells, all fLuc-142MTS constructs exhibited luciferase activities comparable with the fLuc-noMTS control, reflecting the low endogenous miR-142 levels in these cells (Figure 1B and data not shown). In contrast, all fLuc-122MTS constructs showed significant reductions in luciferase activity, consistent with miR-122-mediated silencing (Figure 1B). In addition, proximal insertions (0A, 14A, and 19A) mediated strong silencing comparable with 122MTS-3UTR (p > 0.05, Student’s t test), whereas distal insertions (30A and 60A) exhibited weaker suppression (Figure 1B).

In RAW264.7 macrophages (high miR-142 and low miR-122), all fLuc-122MTS constructs displayed luciferase activities comparable to the noMTS control, consistent with the low endogenous miR-122 levels in these cells (Figure 1C). Conversely, all fLuc-142MTS constructs showed significant reductions in luciferase activity with a position effect, mirroring the pattern observed for fLuc-122MTS in PMH (Figure 1B). Two proximal insertions (142MTS-0A and −14A) mediated approximately 90% suppression in luciferase activity, comparable with 142MTS-3UTR (Figure 1C). In comparison, the three more distal insertions (19A, 30A, and 60A) exhibited progressively weaker silencing. Time course RT-qPCR confirmed rapid degradation of fLuc-142MTS-0A mRNA in macrophages (Figure S1). To determine whether the presence of MTS-containing mRNAs affects endogenous miRNA levels, we measured miR-142 abundance in RAW264.7 macrophages transfected with 142MTS-0A, 142MTS-19A, or 142MTS-60A mRNAs (Figure S2). miR-142 levels never fell below the noMTS control; instead, they transiently increased before returning to levels slightly above control, implying a mild miRNA sponge effect without triggering miRNA degradation. Together, these results demonstrate that MTS embedded in segmented poly(A) tails enable cell type-specific mRNA down-regulation, with silencing efficiency affected by both miRNA identity and insertion position.

To further confirm the specificity of miR-122 MTS, we evaluated 122MTS constructs across all four cell types used in this study. As expected, 122MTS significantly reduced luciferase expression only in PMH cells, with no inhibitory effect in RAW264.7 macrophages, liver sinusoidal endothelial cells (LSECs), or hepatic stellate cells (HSCs) (Figures 1, S3A, and S3E). This functional specificity is consistent with the established liver-specific expression pattern of miR-122 (15).

To determine whether the poly(A)-MTS function depends on the specific 3′ UTR context, we constructed a new reporter using a structurally distinct chimeric 3′ UTR (PNLIP-RPS3A) and inserted miR-142 MTS at five poly(A) positions (0A, 14A, 19A, 30A, and 60A). In RAW264.7 macrophages, the silencing pattern was identical to that observed with the α-globin 3′ UTR. Proximal insertions (0A, 14A, and 19A) achieved strong silencing (∼90% reduction), while distal insertions (30A and 60A) showed progressively weaker effects. In non-target 293T cells, none of the 142MTS constructs significantly affected luciferase expression (Figure S4). These results confirm that position-dependent silencing conferred by poly(A)-embedded MTS is independent of the upstream 3′ UTR sequence.

Combinatorial miRNA targeting inhibits expression in multiple hepatic cell types while preserving expression in hepatocytes

To minimize off-target mRNA expression in multiple cell types simultaneously, we engineered mRNAs with tandem MTSs at the 0A position of the poly(A) tail, where single MTS insertions demonstrated strong, consistent silencing. We designed two dual-MTS constructs containing both miR-122 and miR-142 target sites at the 0A position in different orders (122/142MTS-0A and 142/122MTS-0A; Figure 2A). In both PMH cells and RAW264.7 macrophages, these constructs reduced luciferase activity to near-background levels, while the noMTS control maintained robust expression (Figures 2B and 2C). These results demonstrate that the 3′ UTR-poly(A) junction can accommodate multiple MTS for potent, simultaneous silencing in different cell types, regardless of insertion order.

Figure 2.

Dual-miRNA target sites for miR-122 and miR-142 at the 5′ end of poly(A) tail simultaneously downregulate mRNA expression in non-target cells

(A) Schematic diagram of dual miRNA target site (dual-MTS) insertions for miR-122 (blue) and miR-142 (orange) at 0A of position poly(A). Only the 3′ UTR and the poly(A) regions were depicted. No insertion control (noMTS) and the two insertions at 0A of poly(A) with opposite orders were indicated as noMTS, 122/142MTS-0A and 142/122MTS-0A. (B–C) Sequential insertions of miR-122 and miR-142 dual-MTS at 0A position of the poly(A) tail (122/142MTS-0A and 142/122MTS-0A) effectively downregulate luciferase activity in (B) primary mouse hepatocytes (PMHs) and (C) RAW264.7 macrophages compared with noMTS control regardless of the order of insertion. Bar graph shows the means of relative unit of luminescence (RLU) with SEM (∗∗p < 0.01; ∗∗∗p < 0.001, one-way ANOVA).

To achieve silencing across multiple non-parenchymal liver cell populations while maintaining mRNA expression in hepatocytes, we decided to engineer a triple MTS system targeting macrophages (miR-142), HSCs (miR-148a), and LSECs (miR-126) for mRNA silencing.

To provide background characterization for the miRNAs used in the triple-MTS system, we first evaluated single miR-126 and miR-148a MTS at the 0A and 14A positions across all relevant cell types. 126MTS-0A and 126MTS-14A specifically silenced luciferase expression in LSEC (∼70% reduction) without affecting HSC, RAW264.7 macrophages or PMH (Figure S3). Similarly, 148aMTS-0A and 148aMTS-14A silenced expression in HSC (∼65% reduction) without affecting RAW264.7 macrophages or LSEC, except for 148aMTS-14A affecting expression mildly in PMH (Figure S3). These data confirm the target specificity of each individual MTS and provide the functional basis for their inclusion in the triple-MTS combinatorial design.

We constructed all six possible MTS permutations at 0A position of the poly(A) tail (Figure 3A). In PMH cells, which express low levels of all three miRNAs, five out of six triple-MTS maintained luciferase activity comparable with the noMTS control as expected (p > 0.05, Student’s t test; Figure 3B). However, the 148a/142/126MTS-0A variant showed significantly reduced activity (59.0% suppression, p < 0.001, Student’s t test), indicating that MTS order critically influences mRNA expression (Figure 3B).

Figure 3.

Triple miRNA target site (triple-MTS) insertions at the 5′ end of poly(A) tail allow effective expression in hepatocytes while simultaneously suppressing expression in multiple non-target hepatic cell types

(A) Schematic diagram of triple miRNA target site (triple-MTS) sequential insertions for miR-142 (orange), miR-126 (blue), and miR-148a (yellow) at 0A of position poly(A) with all possible permutations. Only the 3′ UTR and the poly(A) regions were depicted. No insertion control (noMTS) and all six possible permutations were indicated. (B–E) Bar graphs showing luciferase activity (RLU) for the negative control (mock), fLuc mRNA with no MTS control, and with triple-MTS at 0A position of the poly(A) tail in primary mouse hepatocytes (PMHs), RAW264.7, LSEC, and HSC, respectively. (B) In PMH cells, five out of six triple-MTS insertions effectively express luciferase at levels comparable with that of noMTS control with an exception showing reduction with extreme significance. (C) In RAW264.7 cells, all triple-MTS insertions show significant reductions in luciferase activity compared with the noMTS control, with five out of six showing more than 90% reductions. (D) In LSEC, all triple-MTS insertions displayed significant silencing compared with noMTS control, with one showing more than 70% reduction. (E) In HSC, five out of six displayed significant silencing compared with noMTS control, with one showing more than 70% reduction. Data were presented as mean ± SEM (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001; one-way ANOVA).

In RAW264.7 macrophages, all triple-MTS combinations significantly reduced luciferase activity (p < 0.001, Student’s t test), with five out of six configurations decreasing expression to <5% of the noMTS control (Figure 3C). As an exception, 126/142/148aMTS-0A retained ∼33% activity. Similar patterns were observed in LSECs, where all constructs mediated significant suppression, with 126/148a/142MTS-0A showing maximal reduction (76% suppression; Figure 3D). In HSCs, five configurations significantly decreased expression (69% maximal reduction for 126/148a/142MTS-0A), while 142/126/148aMTS-0A displayed insignificant reduction (Figure 3E). Notably, 126/148a/142MTS-0A emerged as the overall most potent suppressor with less than 30% of the control activity for all three cell types (Figures 3D and 3E).

Our comprehensive analysis demonstrates that most triple-MTS constructs effectively silence expression in macrophages, LSECs, and HSCs while preserving hepatocyte expression (Figure S5). Among all tested configurations, 126/148a/142MTS-0A appeared to be the optimal arrangement, achieving simultaneous and robust suppression across all three non-parenchymal cell types. Notably, unlike the dual-MTS constructs, silencing efficiency in this triple-MTS system demonstrated strong dependency on MTS order, possibly reflecting configuration-specific mRNA secondary structures that influence miRNA accessibility.

To determine whether the observed order-dependent silencing in the triple MTS system could result from the creation of unintended miRNA binding sites, we performed an miRDB analysis on all six permutations. This revealed that only two different arrangements (126/148a/142MTS-0A and 142/126/148aMTS-0A) contain the same novel predicted site for mum miR-357–3p, which may account for some unexpected repression (Figure 3). No other novel sites were identified for the remaining five permutations (Table S2). Additionally, we cannot exclude the possibility that order-specific mRNA secondary structures differentially affect RISC accessibility.

miRNA target sites confer poly(A) tails tissue-specific silencing activity in vivo

To evaluate the in vivo silencing capacity of MTSs embedded in poly(A), luciferase mRNAs containing miR-122 MTS at five distinct positions (0A, 14A, 19A, 30A, and 60A) of poly(A) were synthesized by IVT and encapsulated in (4S)-KEL12 LNPs. Six hours after intravenous (i.v.) administration, whole-body luminescence imaging revealed that the fLuc-noMTS control exhibited robust luciferase activity in the liver (Figures 4A and 4B). (4S)-KEL12-based luciferase mRNA-LNP exhibits predominantly liver-specific luciferase activity when administered intravenously in our previous study.²⁶ In contrast, the 3′ UTR-inserted 122MTS control (fLuc-122MTS-3UTR) showed 91.6% suppression compared with the noMTS control (p < 0.001, Student’s t test), consistent with published data.¹⁷ Strikingly, all MTS configurations in poly(A) achieved significant silence (p < 0.001), with efficacy strongly correlating with insertion proximity to the 3′ end (Figures 4A and 4B). The three most proximal insertions (0A, 14A, and 19A) all mediated at least 90% suppression, with two surpassing even the 3′ UTR insertion control (95.8% for 0A, 94.7% for 14A, and 91.0% for 19A, respectively). In contrast, the distal 30A and 60A insertions showed progressively weaker inhibition (71.1% and 66.6%), suggesting that silencing potency decays as MTS sites are positioned deeper into the poly(A) tail.

Figure 4.

miR-122 target sites embedded in the poly(A) tail effectively suppress luciferase activity in mouse liver

Representative images of (A) whole-body, (C) livers, and (E) spleens from animals injected with KEL12-based LNPs encapsulating fLuc mRNAs containing no MTS insertion or miR-122 MTS inserted at distinct locations of the poly(A) tail 6 hours after tail vein injection (n = 6). PBS was injected as negative control (n = 3). (B, D, and F) Bar graphs are quantifications of average radiance of images in (A), (C), and (E), respectively. Average radiance is expressed as photons/sec/cm²/sr and was measured from ROIs drawn over the whole body or individual organs. Data were shown as mean, with the error bars representing SEM (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001; one-way ANOVA).

Moreover, imaging using dissected organs unequivocally demonstrated that 122MTS embedded in poly(A) tail mediated strong silencing in liver (Figures 4C and 4D). The proximal insertions (0A–19A) exhibited particularly potent suppression in liver luminescence (86.6%–93.9% reductions, p < 0.001, Student’s t test). 122MTS-14A and −19A achieved 93.9% and 91.1% silencing, respectively, surpassing the canonical 3′UTR MTS insertion control (86.6% silencing). This in vivo validation confirmed our findings in cultured cells and whole-body imaging experiments.

However, all fLuc-122MTS constructs exhibited measurable silencing in the spleen without statistical significance (20.2%–53.6% reduction, p > 0.05, Student’s t tests), suggesting some unexpected miR-122 activity in this organ (Figures 4E and 4F). The observed decreases of expression in spleen according to insertion proximity (62.9% for 14A vs. 25.3% for 60A) mirrored the trend in the liver, suggesting similar regulation with lower efficiency (Figures 4E and 4F). Notably, spleen-to-liver signal ratios revealed that three proximal poly(A) insertions (0A–19A) achieved over 3-fold improved tissue selectivity compared with noMTS controls (Table 1). Proximal insertions (0A–19A) also conferred tissue selectivity at least comparable with the 3′UTR control, with 14A and 19A demonstrating that genuine poly(A)-embedded MTS are highly effective (>3-fold improvement over noMTS and numerically comparable with the 3′UTR insertion). Therefore, we concluded that synthetic poly(A) tails with MTS insertions have the potential to functionally recapitulate the regulatory capacity of 3′ UTRs.

Table 1.

Relative activity of luciferase mRNA with 122MTS inserted at various locations downstream of 3′ UTR in dissected liver and spleen

fLuc-122MTS	Ratio (spleen/liver)	Normalized ratio
noMTS	3.9%	1.0
3′ UTR	13.4%	3.5
0A	31.0%	8.0
14A	16.1%	4.2
19A	16.4%	4.2
30A	5.8%	1.5
60A	5.5%	1.4

We next studied whether miR-142 target sites engineered into the poly(A) tail could similarly confer tissue-specific regulation in vivo. Following IV delivery of KEL12 LNP-encapsulated fLuc-142MTS mRNAs, these constructs maintained stable hepatic expression, as evidenced by both whole-body luminescence (Figures 5A and 5B) and dissected liver imaging (Figures 5C and 5D). Dissected organ imaging further demonstrated potent spleen-specific silencing for all tested 142MTS constructs. Five out of six 142MTS achieved 74% or more suppression in spleen (p < 0.001 vs. noMTS control, Student’s t test), with only the distal 142MTS-60A insertion showing much reduced silencing efficiency (47%) (Figures 5E and 5F). Quantitative analysis revealed significantly enhanced liver-to-spleen expression ratios for all 142MTS constructs, confirming that poly(A)-embedded MTS can simultaneously preserve target liver tissue expression while suppressing off-target spleen activity (Table 2). Quantitative analysis showed that all 142-MTS constructs significantly enhanced liver-to-spleen expression ratios compared with the noMTS control (Table 2). The 14A insertion achieved a normalized ratio (7.1) slightly higher than the canonical 3′UTR control (6.6), while other positions (19A, 30A, and 60A) gave ratios comparable with or modestly lower than the 3′UTR control. Nevertheless, all poly(A)-embedded 142MTS constructs effectively suppressed off-target spleen activity while preserving hepatic expression, with the proximal 14A position being particularly effective in this LNP context. These results suggested that we obtained novel poly(A) variants that can regulate mRNA expression in an active and tissue-specific manner.

Figure 5.

miR-142 target sites embedded in the poly(A) tail effectively suppress luciferase activity in mouse spleens

Representative images of (A) whole body, (C) livers, and spleens (E) whole-body from animals injected with KEL12-based LNPs encapsulating fLuc mRNAs containing no MTS insertion or miR-142 MTS inserted at distinct locations of the poly(A) tail 6 hours after tail vein injection (n = 6). PBS was injected as negative control (n = 3). Bar graphs in (B), (D), and (F) are quantifications of average radiance of images in (A), (C), and (E), respectively. Average radiance is expressed as photons/sec/cm²/sr and was measured from ROIs drawn over the whole body or individual organs. Data were shown as mean, with error bars representing SEM (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001; one-way ANOVA).

Table 2.

Relative activity of luciferase mRNA with 142MTS inserted at various locations downstream of 3′ UTR in dissected liver and spleen

fLuc-142MTS	Ratio (liver/spleen)	Normalized ratio
noMTS	999.0.0	1
3′ UTR	59.2	6.6
0A	24.4	2.7
14A	63.7	7.1
19A	46.3	5.1
30A	41.7	4.6
60A	22.5	2.5

In the field of mRNA therapeutic development, plasmid DNA stability during bacterial fermentation is a critical practical concern for GMP manufacturing. We compared the replication stability of plasmids carrying MTS inserted either in the 3′ UTR or at various positions within the poly(A) tail following transformation into E. coli DH5α. Insertion of MTS into the 3′UTR (for both miR-122 and miR-142) led to increased poly(A) truncations compared with the noMTS control (Figure S6). In contrast, MTS inserted into the poly(A) tail at several positions resulted in better transmission stability. Particularly at position 19A, insertions yielded 100% intact poly(A) sequences with no detectable deletions. Other poly(A) positions (0A, 14A, 30A, and 60A) showed intermediate stability (72%–88% intact), all of which were comparable with or better than 3′ UTR insertion (Figure S6). These data demonstrate that placing MTS within the poly(A) tail can enhance plasmid genetic stability, a significant advantage for large-scale GMP manufacturing of mRNA therapeutics.

To further demonstrate the therapeutic utility of our platform, we performed an in vivo experiment using enhanced green fluorescent protein (EGFP) mRNA payload, as a surrogate for therapeutic protein delivery in the context of in vivo CAR-T cell generation. We inserted miR-223 target site (223MTS) at the 19A position of the poly(A) tail, a position identified as highly effective for silencing, and administered EGFP mRNA-LNPs (noMTS control or 223MTS-19A) intravenously into C57BL/6 mice. Six-hour post-injection, splenic single-cell suspensions were analyzed by flow cytometry. The results showed that in macrophages (F4/80⁺), 223MTS-19A reduced EGFP expression by >75% compared with the noMTS control (Figure S7). In dendritic cells (CD11c⁺), 223MTS-19A similarly reduced EGFP expression by >75%. In contrast, T cells (CD3⁺), which do not express miR-223, maintained robust EGFP expression comparable to the noMTS control. These data demonstrate that our poly(A)-MTS platform can achieve potent, cell type-specific silencing in vivo, with particular relevance for applications requiring preservation of mRNA expression in T cells while silencing in off-target myeloid populations.

Discussion

This study developed novel segmented poly(A) variants that can function as miRNA-responsive regulatory elements, expanding the toolbox for mRNA therapeutics. By embedding MTSs within poly(A) tail, we modified poly(A) tail from a non-gene-specific element for mRNA stabilization and translation initiation to an active regulatory platform for tissue-specific gene silencing (Figures 1, 4, and 5). We further explored a modular system using tandem MTS insertions at the 5′ end of poly(A) that simultaneously suppresses mRNA expression in multiple off-target liver cell types, such as macrophages, LSECs and HSCs, while maintaining hepatocyte expression in vitro (Figures 2 and 3). Together, this work addresses a critical need for precise expression in mRNA therapies, particularly for applications requiring stringent control in off-target cell types, such as gene editing or immunogenic protein delivery.²⁷

Our insertion strategy within poly(A) has several potential advantages over previous methods of inserting MTS in the UTRs. Our approaches avoid disrupting the primary sequence of UTR. Although optimal MTS positioning in poly(A) requires experimental validation, our in vitro and in vivo results confirm silencing efficacies matching MTS inserted in the 3′ UTR while preserving translational capacity in desired organs (Figures 4 and 5). This makes our approach an attractive alternative or even necessary for therapies where UTR integrity is critical.²⁸ Finally, our approaches are unexclusive to MTS inserted in UTRs or LNPs with organ tropism such as SORT.²⁹ Their integration could facilitate the development of more versatile organ- and cell type-selective expression vectors.

Our data suggested at least two critical factors affecting silencing efficiency for MTS inserted in poly(A). First, we observed stronger suppression with miR-122 compared with miR-142 (Figures 1B vs. 1C; Figures 4D vs. 5B), which might reflect different abundances of these miRNAs in their respective tissues or cells. Second, positional analysis demonstrated more effective silencing for proximal MTS insertions than for more distal insertions in general (Figures 1B, 1C, 4D, and 5B). Although the exact mechanisms remain unclear, we envision a possible scenario where cleavage of poly(A) by the RISC complex recruited to proximal MTS in poly(A) likely results in shorter poly(A) tail length with less protective function than if recruited to a distal MTS, potentially responsible for better silencing efficacy for proximal insertions. Future mechanistic studies will elucidate whether the position-dependent silencing observed in our poly(A) MTS system is consistent with a model in which RISC recruitment to the poly(A) tail, particularly at proximal positions, compromises mRNA stability. The rapid decay of 142MTS-0A mRNA (Figure S1) and the transient elevation of miR-142 levels (Figure S2) are compatible with multiple, non-mutually exclusive mechanisms, including RISC-mediated deadenylation and miRNA sequestration. More sensitive approaches, such as nanopore-based poly(A) tail profiling or single-molecule imaging, will be required to definitively dissect the molecular details.

In addition, we observed some unexpected decreases in luciferase expression in the spleen using KEL12-based LNPs loaded with fLuc mRNAs carrying a miR-122 target site in the 3′ UTR or poly(A) tail. In contrast, miR-122 target site inserted at the same position in α-globin 3′ UTR did not show any reduction of luciferase activity in the spleen in an earlier study.¹⁷ The observed silencing in the spleen by miR-122 MTS constructs may reflect several non-mutually exclusive possibilities. For example, low-level miR-122 could be expressed in a subset of splenic cells, and/or there could be off-target LNP delivery to rare miR-122-positive cells. Future studies using single-cell resolution approaches will be needed to distinguish between these possibilities and to better understand LNP biodistribution and miRNA expression profiles at the cellular level in the spleen.

Our in vitro studies also identified an optimal triple-MTS configuration (miR-126/142/148a) at the 5′ end of poly(A) that simultaneously suppresses mRNA expression in macrophages, LSECs and HSCs while maintaining robust hepatocyte expression (Figures 2 and 3). Future validation of this triple-MTS in vivo will offer a convenient tool for hepatocyte-selective mRNA-LNP expression for protein replacement or gene editing. Although it remains unclear about order-dependent silencing efficacy for triple MTS insertions, it is possible that MTS arrangement may influence RISC accessibility through order-specific secondary structures. We have not systematically mapped all three single MTS across all five poly(A) positions in all different cell lines. Future studies could fine-tune each MTS position independently for maximal combined silencing.

To extend these findings to a more physiologically relevant setting, we next evaluated the in vivo performance of the poly(A)-MTS platform using an EGFP-encoding mRNA carrying a miR-223 target site at the 19A position (223MTS-19A; Figure S7). Following systemic administration, EGFP expression was potently silenced in splenic macrophages (F4/80⁺) and dendritic cells (CD11c⁺), both of which endogenously express miR-223, with reductions exceeding 75% compared with the noMTS control. In contrast, T cells (CD3⁺), which lack miR-223, maintained EGFP levels comparable with those in the control group, underscoring the strict miRNA-dependent specificity of the platform in vivo. Collectively, these data demonstrate that a single MTS placed within the poly(A) tail is sufficient to confer robust, cell type-restricted repression of a therapeutic mRNA payload in a complex tissue environment. This feature is particularly valuable for mRNA-based in vivo CAR-T therapy, where sustained transgene expression in T cells is often desired, while unintended expression in myeloid antigen-presenting cells may provoke off-target effects or immune tolerance. Therefore, our poly(A)-MTS strategy offers a simple yet effective means to improve the safety and precision of mRNA therapeutics.

As with all miRNA-based regulatory strategies, our poly(A)-MTS platform requires adequate endogenous miRNA expression in the off-target cells to achieve effective silencing. The magnitude of silencing achieved by our constructs is likely influenced by the abundance of the corresponding miRNAs in specific cell types, a reflection of the fundamental dependence of miRNA-mediated regulation on target miRNA abundance, rather than a limitation unique to our platform. For therapeutic applications, we recommend selecting miRNAs that are highly and specifically expressed in the off-target cell population(s) to ensure robust silence. When high-abundance miRNAs are not available for a particular cell type, combining multiple moderate-abundance miRNAs (as in our triple-MTS design) may achieve additive silencing. In cells with profoundly impaired miRNA biogenesis (e.g., DROSHA- or DICER-deficient lines), alternative non-miRNA regulatory elements would be needed. For most therapeutic applications, however, the relevant off-target tissues (liver, spleen, hematopoietic cells) express abundant tissue-specific miRNAs, making our platform broadly applicable. Importantly, this modular platform offers opportunities for potential expansion to other tissue or cell type targets through rational MTS engineering. Together, the simplicity and modularity of our approaches including repurposed poly(A) tails and multiple MTS configurations suggest broad applications for cell-type-specific expression control, both within the liver and in extrahepatic tissues. Their future combinations with cell-specific delivery systems may unlock programmable expression control across diverse tissues.

Materials and methods

Plasmids

All plasmids for in vitro transcription (IVT) used in this study were derived from a pUC57-Luc plasmid containing a T7 promoter region, 5′ α-globin UTR, fLuc ORF, 3′ α-globin UTR, and a segmented poly(A) tail variant. The sequence of the poly(A) variant is as follows: 5′- AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAGATATCAAAAAAAAAAAAAAAAAAAGAAAAAAAAAAAAAAAAAAAGAAAAAAAAAAAAAAAAA – 3’. The pUC57-Luc IVT template DNA plasmid containing the FLuc reporter gene was generated by gene synthesis service (Shanghai Dynegene Technologies Co., Ltd), using pUC57 as the vector backbone.

Sequences of modified poly(A) tail variants

Sequences of the modified poly(A) variants were listed in Table S1.

mRNA IVT and purification

Plasmid DNA was isolated from 50 mL of overnight bacterial culture using EndoFree Plasmid Maxi Kit (Qiagen, Cat. #12362), followed by BspQ1 (New England Biolabs, Cat.#R0712L) digestion to linearize DNA for IVT. IVT was performed using the High Yield T7 RNA Synthesis Kit (Hongene Biotech, Cat# ON-40) in a 100 μL reaction mixture containing 5 μg of plasmid template, ATP, GTP, CTP, 1-N-Me-PseudoUTP (final concentration is 10 mM for each) and 5 μL of Cap analog (Synthgene, Cat.# CAP3011). The mixture was incubated at 37°C for 3 h, followed by DNase I treatment to remove DNA template. Finally, synthetic mRNA was purified through LiCl precipitation. The concentration and integrity of mRNA were determined by NanoDrop Spectrophotometer and the Agilent 5200 Fragment Analyzer, respectively.

Lipid nanoparticle preparation and characterization

mRNA was diluted to 200μg/mL in acetate buffer (pH 4.0). Lipid components (50% ionizable lipid, 38.5% cholesterol, 10% DSPC, and 1.5% PEG-DMG dissolved in ethanol) were mixed with the mRNA solution at a flow rate ratio of 3:1 using a T-mixer device. The resulting LNPs were dialyzed against 2 mM acetate buffer (pH 4.0) using a 100 kDa ultrafiltration membrane to remove ethanol. Finally, the required LNP was obtained after adjusting the pH to 7.5 with Tris buffer.

LNP size and polydispersity index (PDI) were measured using a Zetasizer Pro. Instrument (Malvern). Encapsulation efficiency was determined using the Quant-iT RiboGreen Kit (Cat# R11490, Invitrogen). In brief, free mRNA in solution and total mRNA content were measured separately, and encapsulation efficiency was calculated as Encapsulation efficiency (%) = (Total mRNA content – Free mRNA content)/(Total mRNA content × 100%). These LNPs had a particle size of 70–90 nm, PDI <0.15, encapsulation efficiency >90%, and an RNA: lipid ratio of 0.05 (w/w), as detailed in Table S3.

Mammalian cell culture

Human embryonic kidney 293T (HEK293T) cells were obtained from the Stem Cell Bank, Chinese Academy of Sciences. Mouse monocyte macrophage RAW 264.7, mouse LSECs, and HSCs were purchased from Newgainbio Ltd (Wuxi, China). All four cell lines were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (FBS; Gibco). Primary mouse hepatocytes of C57BL/6 mice were purchased from Milecell Biological Science & Technology Co., Ltd. (Shanghai, China) and were cultured in Maintenance Media. All cells were cultivated at 37°C with 5% CO₂.

Animals

C57BL/6 mice (6–8 weeks old, 18–22 g) were purchased from Charles River Laboratory Animal Technology Co., Ltd (Beijing, China) and were cultivated at Ascentage Pharma (Suzhou, China) under standard and pathogen-free conditions (25°C, 50% ± 10% humidity, 12-h dark/light cycle) with free access to food and water. Animal studies were conducted in accordance with the guidelines of the Chinese Association for Laboratory Animal Sciences and approved by the IACUC of Ascentage Pharma (Approved ethical number: AS-20230621–01, AS-20240313–01, and AS-20221228–01).

Luciferase assay in vitro

Cells of HEK293T, RAW 264.7, LSEC, and HSC were seeded into 24-well plates at densities of 2 × 10⁵, 4 × 10⁵, 2 × 10⁵, and 2 × 10⁵ cells/mL, respectively (500 μL/well), and were grown overnight to the cell density of ∼80%. Next day, cells were transfected with 500 ng/well of mRNA using Lipofectamine 3000 Transfection Reagent (Thermo Fisher, Cat. #L3000015). Luciferase assay was performed according to the manufacturer’s instructions of Luciferase Assay System kit (Promega, Cat. #E1501) and luminescence signal was measured using Centro Microplate Luminometer (Bethold).

Luciferase assay in vivo

C57BL/6 mice were injected intravenously with fLuc mRNA-LNPs at the dose of 1 mg/kg, using 1×PBS as a negative control. At 6 h post-administration, mice were injected intraperitoneally with 150 mg/kg of D-luciferin monosodium salt (YEASEN, Cat. 40901ES01). Mice were subsequently anesthetized in a chamber supplied with 2.5% isoflurane, then placed on the imaging platform while being maintained on 2% isoflurane via a nose cone. The bioluminescence images were taken 10 min after the D-Luciferin injection using IVIS Lumina II (PerkinElmer). After whole-body imaging, mice were sacrificed for organ imaging. The livers and spleens of the same mice were immediately collected and imaged. After all groups were imaged, average radiance (photons/sec/cm²/sr) was quantified using IVIS Living Image software. Image analysis was performed using IVIS software (version 4.7.4, PerkinElmer). This metric represents the intensity of bioluminescent signal per unit area per unit solid angle and is the standard output for IVIS imaging systems. Theoretically, whole body and isolated organ signals are not directly comparable with each other; instead, comparisons are made within each same modality (whole body to whole body, liver to liver, and spleen to spleen). All reported fold changes and statistical analyses are based on dissected organ data, which provide the most reliable quantitative readout.

Statistics

Normality of the data was assessed using the Shapiro-Wilk test (p > 0.05 for all experimental groups), confirming that parametric assumptions were met. Comparisons among multiple groups were performed using one-way ANOVA followed by Dunnett’s post-hoc test for comparisons against the noMTS control, or Tukey’s post-hoc test for all pairwise comparisons. p values and effect sizes are reported in figure legends and supplementary tables. These analyses were performed using GraphPad Prism 10 (GraphPad Software, LLC). All in vitro experiments were performed with 3 independent biological replicates. Data are presented as mean ± SEM.

Data and code availability

All data supporting the findings of this study within the article or its supplementary materials are available upon request, including full plasmid maps and annotated sequences (in GenBank format) for all constructs, after acceptance for academic experimentation, duplication, and reuse of the data and materials.

Acknowledgments

This work was supported by the National Center of Technology Innovation for Biopharmaceuticals NCTIB2023XB02010 (to W.Z.), and CAMS Innovation Fund for Medical Sciences 2024-I2M-ZH-013 (to S.C.).

Author contributions

R.Q., R.C., and W.Z. conceived and designed the study. R.Q. and R.C. performed in vitro and in vivo experiments. R.X., L.H., Y.X., and H.B. assisted with animal experiments. J.L. and T.Z. prepared LNP formulations. N.L. and Q.L. administered LNPs via intravenous injection. K.L. provided the cationic lipid (4S)-KEL12. R.Q. and H.C. performed data analysis. H.C. and W.Z. drafted and revised the manuscript. All authors have read and edited the manuscript.

Declaration of interests

W.Z., Y.D., R.C., and R.Q. are listed as co-inventors of a patent related to this work. R.Q., R.C., H.C., R.X., L.H., Y.X., J.L., N.L., Q.L., H.B., T.Z., Y.D., and W.Z. were employees of RinuaGene, Inc., at the time of the work.