Spatiotemporal targeting of messenger RNA lipid nanoparticles to the endometrium for the treatment of reproductive disorders.

The endometrium, the inner lining of the uterus, has important barrier and reproductive functions 1 . Endometrial dysfunction is characteristic of a myriad of women’s health conditions, including uterine adhesions, Asherman’s syndrome, endometriosis, infections and cancers 2 – 5 . Endometrial dysfunction can lead to poor receptivity and embryo implantation failure. Thinning in the endometrial lining, spontaneous or caused by a gynaecologic condition, is associated with implantation failures and poor pregnancy outcomes 6 – 8 . Women with a thin endometrium (TE) are less likely to become pregnant, even with assisted reproductive technology (ART) 9 – 13 . In addition, many women experience recurrent implantation failure (RIF) in multiple ART cycles even when the endometrial thickness is normal and embryos quality is good 14 , 15 . There are limited evidence-based treatment options for poor endometrial receptivity 16 . Cell therapies can improve endometrial function 17 , 18 , but a pharmacologic approach would be more accessible to patients. A plethora of studies have demonstrated the potential of intrauterine administration of growth hormones or cytokines in enhancing uterine blood flow, endometrial thickness and receptivity in women with TE and/or RIF 19 – 22 . However, challenges such as insufficient endometrial targeting, systemic exposure and the inability to sustain adequate therapeutic protein levels within the uterus remain significant obstacles 23 . Messenger RNA can induce intracellular protein production and sustain its expression for several hours to a few days and allow for multiple dosing 24 , 25 . Lipid nanoparticles (LNPs) are efficient tools for the intracellular introduction of mRNA, but an endometrium-targeted mRNA delivery system has not been described. Several studies showed promise in delivering mRNA-LNPs to the placenta by systemic administration; however, substantial off-target expression in the maternal liver and spleen has been observed 26 , 27 . Locally administered mRNA-LNPs can still undergo systemic distribution and induce protein expression in the liver and spleen 28 . In particular, the rich blood supply, extensive lymphatic network and larger size of tight junctions in the uterus, compared with those in the gastrointestinal or nasal mucosa 29 , may facilitate the systemic absorption of intrauterine therapeutics. Here we aimed to localize mRNA delivery to the endometrium by intrauterine infusion, a minimally invasive method routinely used to administer embryos or therapeutics in ART 22 . The endometrium undergoes significant structural and functional changes during the menstrual cycle in humans 30 . The overexpression of cell adhesion molecules, including αvβ3, α4β1 and α1β1 integrins, on the luminal epithelium during the window of implantation (WOI) primes the uterus for blastocyst attachment 30 – 33 . In this paper, we functionalized the LNPs with arginylglycylaspartic acid (RGD), a peptide motif with binding affinity to integrins, to improve mRNA delivery to the endometrium during the WOI. Surprisingly, we discovered that attaching an RGD peptide directly to the lipid component of the LNP conferred a bifunctional advantage: it not only induced higher encoded protein expression in the endometrium but also reduced off-target protein expression in the liver and spleen. As a model therapeutic cytokine, we focused on granulocyte-macrophage colony-stimulating factor (GM-CSF) because of its effects on macrophage proliferation and stromal cell migration, both of which have been shown to improve endometrial thickness and receptivity 19 , 34 . A single intrauterine infusion of RGD-modified LNP containing GM-CSF mRNA administered during the WOI provided sustained, localized protein expression for up to 24 h with 60-fold reduced systemic protein exposure compared with intrauterine infusion of recombinant GM-CSF. In a murine model of endometrial injury, intrauterine infusion of targeted GM-CSF mRNA-LNP restored embryo implantation, outperforming the recombinant protein. The endometrium-targeted mRNA delivery strategy described herein could expand the treatment options for women’s health conditions, including endometrial cancer and endometriosis.

We infused GM-CSF mRNA-loaded LNP B on p.c. day 2 in mice with endometrial injury to allow for sufficient endometrial repair before embryo implantation would occur around p.c. day 5 (Fig. 5a ). As described above, we confirmed that administration of LNP B on p.c. day 2 resulted in reduced off-target protein expression (Supplementary Fig. 1 ), and avoided fetal resorption that occurred when administering fluids to the uterus as late as p.c. day 5 (Supplementary Fig. 12 ). Ethanol-induced injury with PBS treatment resulted in a 67% reduction in embryo implantation sites (Fig. 5b–e ) and a 110 μm reduction in endometrial thickness (Fig. 5f,g ) compared with the non-injured, untreated control uterus. Interestingly, GM-CSF mRNA-loaded LNP B restored implantation to levels comparable to the non-injured, untreated control horn (Fig. 5b–e ), with no signs of acute toxicity in the uterus (Supplementary Fig. 13 ) or other tissues (Supplementary Fig. 14 ). Mice treated with recombinant GM-CSF had 67% lower implantation rates in the injured horn relative to non-injured, untreated control uterus (Fig. 5b–e ). While systemic treatment with recombinant GM-CSF was shown to be effective in a previous mouse study 34 , GM-CSF was administered over several days with a total dose 20-fold higher than this study. Fig. 5 Therapeutic utility of GM-CSF mRNA-loaded LNP B in an endometrial injury model. a , Induction of endometrium injury and treatment schedule. b , Representative images showing embryo implantation sites on p.c. days 7–9. The injured uterine horn was infused with 20 μl PBS solution containing nothing, 4 μg GM-CSF mRNA in LNP B or 3.6 μg recombinant mouse GM-CSF. c , Numbers of embryo implantation sites in each injured and non-injured horn of treated and untreated groups. Data represent the mean ± s.e.m. ( n  = 6 mice for LNP B and GM-CSF protein groups, n  = 9 for the PBS group). Two-tailed, paired t -tests. d , Compiled data showing the total number of embryo implantation sites, with the numbers from non-injured uterine horns from all three groups pooled into one group (grey bar). Data represent the mean ± s.e.m. ( n  = 6 uterine horns for LNP B and GM-CSF protein groups, n  = 9 for the PBS group and n  = 21 for the non-injured group). One-way ANOVA followed by Dunnett’s test. e , The percentage of uteri positive to at least one implantation as calculated from d . Data represent the mean ± s.e.m. ( n  = 6 uterine horns for LNP B and GM-CSF protein groups, n  = 9 for the PBS group and n  = 21 for the non-injured group). Two-tailed, Fisher’s exact tests against the non-injured group. f , g , Endometrial thickness as measured in sections with haematoxylin-and-eosin staining. Representative images of the treated uterus at p.c. days 7–9 ( f ). Average endometrial thickness ( g ). Measurements were taken between implantation sites where a contiguous and visible luminal space was present. Scale bars, 200 μm. Data represent the mean ± s.e.m. ( n  = 4 uterine horns for LNP B, n  = 5 for GM-CSF protein, n  = 7 for PBS and n  = 9 for non-injured groups). One-way ANOVA followed by Dunnett’s test. Illustration in a created with BioRender.com . Source data a , Induction of endometrium injury and treatment schedule. b , Representative images showing embryo implantation sites on p.c. days 7–9. The injured uterine horn was infused with 20 μl PBS solution containing nothing, 4 μg GM-CSF mRNA in LNP B or 3.6 μg recombinant mouse GM-CSF. c , Numbers of embryo implantation sites in each injured and non-injured horn of treated and untreated groups. Data represent the mean ± s.e.m. ( n  = 6 mice for LNP B and GM-CSF protein groups, n  = 9 for the PBS group). Two-tailed, paired t -tests. d , Compiled data showing the total number of embryo implantation sites, with the numbers from non-injured uterine horns from all three groups pooled into one group (grey bar). Data represent the mean ± s.e.m. ( n  = 6 uterine horns for LNP B and GM-CSF protein groups, n  = 9 for the PBS group and n  = 21 for the non-injured group). One-way ANOVA followed by Dunnett’s test. e , The percentage of uteri positive to at least one implantation as calculated from d . Data represent the mean ± s.e.m. ( n  = 6 uterine horns for LNP B and GM-CSF protein groups, n  = 9 for the PBS group and n  = 21 for the non-injured group). Two-tailed, Fisher’s exact tests against the non-injured group. f , g , Endometrial thickness as measured in sections with haematoxylin-and-eosin staining. Representative images of the treated uterus at p.c. days 7–9 ( f ). Average endometrial thickness ( g ). Measurements were taken between implantation sites where a contiguous and visible luminal space was present. Scale bars, 200 μm. Data represent the mean ± s.e.m. ( n  = 4 uterine horns for LNP B, n  = 5 for GM-CSF protein, n  = 7 for PBS and n  = 9 for non-injured groups). One-way ANOVA followed by Dunnett’s test. Illustration in a created with BioRender.com . Source data We further evaluated the effects of the LNP components on implantation rates. The empty LNP B failed to increase the number of implantations in the injured horn, which remained 66% lower than its control horn (Supplementary Fig. 15 ). Notably, the non-injured, untreated horn of the empty LNP B group had a similar average number of implantations as the non-injured, untreated horn of the PBS group, indicating the low influence of LNP components on embryo implantation in the neighbouring healthy uterine horn (Supplementary Fig. 15a ). Furthermore, we demonstrated that protein expression could be modulated through dose adjustment or RNA modifications for future considerations (Supplementary Fig. 16 ).

CleanCap AG fLuc, EGFP and m1Ψ-modified mouse GM-CSF ( CSF2 ; Supplementary Table 2 ) mRNA prepared by in vitro transcription were purchased from Trilink Biotechnologies. DSPC, DSPE-PEG2000 (C18-PEG2000), DMG-PEG2000 (C14-PEG2000), DSPE-PEG350 (C18-PEG350), DMPE-PEG350 (C14-PEG350) and DSPE-RGD (C18-RGD) were purchased from Avanti Research. D-Lin-MC3-DMA and vitronectin protein, mouse (HEK293, His) were purchased from MedChemExpress. Cholesterol was obtained from Sigma. 1,1′-Dioctadecyl-3,3,3,3′-tetramethylindodicarbocyanine (DiD) was purchased from Invitrogen. DSPE-PEG2000-RGD (C18-PEG2000-RGD) was obtained from BOC Sciences. Mouse recombinant GM-CSF was obtained from Miltenyi Biotec. Luciferin assay system and reporter lysis buffer were purchased from Promega. FluoSpheres carboxylate-modified 0.1 μm polystyrene beads with a 580/605 fluorophore, the Pierce BCA protein assay kit and the Quant-it RiboGreen RNA assay kit were obtained from Thermo Fisher Scientific. Cy5 Label IT nucleic acid labelling kit was purchased from Mirus. A LipidLaunch LNP Apparent pKa assay kit was provided by Cayman. OCT compound was purchased from SciGen. Rabbit GFP polyclonal primary antibody (Thermo, A11122) and goat anti-rabbit IgG conjugated to Alexa Fluor 488 (Thermo, A11034) were obtained from Thermo. Alexa Fluor 647 anti-pan cytokeratin antibody (Abcam, C-11), rabbit Alexa Fluor 647 anti-vimentin antibody (Abcam, EPR3776), goat FITC anti-GFP antibody (Abcam, Ab6662) and D-Luciferin sodium salt were purchased from Abcam. DAPI Fluoromount-G was obtained from SouthernBiotech. LNPs were formulated by a rapid mixing method. BLOC LNPs were identified by studying the impact of various parameters of RGD conjugation onto LNPs, including ligand density, ligand spacer length and the type of PEG-lipid stabilizer used (Supplementary Table 1 ). We used a broad affinity, linear RGD sequence to promote interactions with the range of RGD-binding integrins expressed on the endometrium 61 . All lipids were purchased in powders and dissolved in 99% ethanol to make stock solutions. The concentration of mRNA stock solutions was determined using a NanoDrop ND-1000 spectrophotometer. For the preparation of fLuc or EGFP mRNA-loaded LNPs, 200 µl of an ethanolic solution containing D-Lin-MC3-DMA, DSPC, cholesterol, PEG-lipid and RGD-conjugated lipid was prepared and mixed with 25 µl of 1 mg ml −1 mRNA dissolved in 1 mM sodium citrate pH 6.4. C18-RGD was added at 0–30 mol% of total lipids. The mixture was then added dropwise to 600 µl of 50 mM sodium citrate buffer (pH 3) and let to equilibrate for 5 min at room temperature. The resulting mixture was then diluted in 30 ml PBS and spun using 30 kDa Amicon centrifugal filters (Millipore) for buffer exchange and removal of residual ethanol. The final formulation volume was brought to 250 µl and used for in vivo experiments or physicochemical characterization. Particle size and the PDI were measured using dynamic light scattering equipped with a diode laser ( λ  = 532 nm) with a scattering angle of 173° (Zetasizer Nano-ZS, Malvern Instruments) after 10× dilution in 1× PBS. The ζ -potential was determined by laser-Doppler electrophoresis using the same Zetasizer Nano-ZS at 50× dilution in 1× PBS. For fluorescence labelling in multiple-particle tracking studies, DiD was added to the lipid phase at 0.1 mol% of total lipids. For quantifying mRNA biodistribution, mRNA was labelled with Cy5 using the Mirus labelling kit. Encapsulation ratios were determined using the RiboGreen assay. The degree of LNP protonation at different pH values was determined using the 6-( p -toluidino)-2-naphthalene sulfonic acid sodium salt (TNS) method. The pKa was calculated by nonlinear regression using a four-parameter model. For making GM-CSF mRNA-loaded LNPs, 340 µl of an ethanolic solution containing lipids was mixed with 85 µl of 0.6 mg ml −1 GM-CSF mRNA dissolved in 1 mM sodium citrate (pH 6.4). The mixture was then added dropwise to 1 ml 50 mM sodium citrate buffer (pH 3) and further diluted in 50 ml PBS. The mixture was concentrated to a final volume of 250 µl and used for efficacy, toxicity or pharmacokinetic studies. All experimental procedures were approved by the Johns Hopkins University Animal Care and Use Committee. Johns Hopkins Program of Animal Care and Use is accredited by AAALAC International. Animal care and procedures follow the Guide for the Care and Use of Laboratory Animals 8th Edition 62 . CD-1 mice were purchased from Charles River Laboratories at 6–8 weeks and were housed in a 12 h light/12 h dark cycle room. Time-mated female mice were purchased from Charles River Laboratories and delivered on p.c. day 4 and allowed to acclimate until dosing on p.c. day 5. In the efficacy studies, male CD-1 (8–12 weeks) mice were mated with female mice with endometrial injury at a 1:1 ratio in individual cages. Male CD-1 mice were individually housed and acclimated for at least a week before mating with female mice. Successful pregnancy was confirmed by the observation of a copulatory, vaginal plug (Supplementary Fig. 17a ). The day a vaginal plug was observed was assigned as p.c. day 1. Estrous cycle stage was determined by visual examination of the external vaginal opening (Supplementary Fig. 17b ). To characterize the mobility of mRNA-LNPs on endometrial mucosal surface after infusion in the uterine cavity, particle diffusion was visualized on the surface of excised mouse endometrial tissue. Around 2 mm of uterus fragment was excised on p.c. day 5, flattened and placed on a slide with a cavity of sufficient depth to allow a coverslip to contact the endometrial surface without compressing the tissue. The surface of the endometrium was unwashed, leaving the mucous membrane intact. DiD-labelled mRNA-LNP (0.5 µl, 5 ng mRNA) was carefully added to the surface of the endometrium to prevent any pooling or dilution. The sample was then covered with a glass coverslip and sealed with a mountant. Multiple-particle tracking analysis was carried out using a ZEISS Axiovert epifluorescence microscope equipped with a 100× oil immersion objective. Trajectories were analysed in each measurement, and the coordinates of LNP centroids were transformed into time-averaged mean square displacements () using Matlab software (v. 23.2). The experiment was performed in triplicate, and 8–12 videos were acquired in each replicate, taking care to avoid any areas with fluid convection. An average of 391 ± 144, 398 ± 149 and 335 ± 176 particles of PS-COOH, LNP A and LNP B were analysed in each measurement, respectively. Distribution histograms of individual particle MSD at a timescale = 1 s were fitted into a Gaussian model using a nonlinear regression of 189 degrees of freedom to calculate the mean MSD at t  = 1 s. Mice were induced in a chamber and maintained under isoflurane anaesthesia using a nose cone system. The abdomen was swabbed with 10% povidone-iodine solution, and a midline laparotomy was performed. Using surgical microscopes, the right uterine horn was isolated, and 20 µl solution containing the mRNA-LNP (2 µg fLuc mRNA or EGFP mRNA) was infused into the lumen of the right uterine horn using a 30 G needle injected at the mid-distal position. The successful injection was confirmed by the observation of a bulge in the uterus because of a fluid-filled lumen. All injections were carried out by an investigator who remained blinded to the treatment. Closure was performed using 5-0 polyglycolic acid sutures in a simple continuous pattern for the peritoneum, followed by staple application for skin closure. Short-acting buprenorphine analgesia (0.1 mg kg −1 ) was injected subcutaneously near the incision site, while the mouse was still under anaesthesia. Mice were killed 4 h post-injection for the evaluation of luciferase expression in organs, or 24 h for microscopic observation of EGFP tissue distribution. For intravenous injections, 200 µl solution containing 2 µg fLuc mRNA or Cy5-labelled fLuc mRNA loaded in LNPs was administered by the tail vein in unmated female CD-1 mice. After the mice were killed 4 h post-treatment, the whole uterus, cervix, liver and spleen were collected and placed in screw-cap tubes each containing 0.5 ml 1× reporter lysis buffer. The tissues were then homogenized after the addition of stainless-steel beads and shaken at 6 m s −1 for 40 s using an MP Biomedical FastPrep-24 5G Homogenizer. The samples were subjected to a single freeze–thaw cycle at −80 °C. The tubes were spun at 8,000 ×  g for 5 min and 4 °C. A 20 µl of the supernatant was transferred into a 96-well, clear-bottom white microplate. Luminescence was measured after auto-injection of a 100 µl luciferin substrate into each well using a Tecan microplate reader. Data were presented as photons per second exposure. To compare the delivery efficiency between unmated, p.c. day 2 and p.c. day 5 mice, luciferase expression was normalized to total protein content in each tissue homogenate, which was measured using the BCA assay. For tracking the biodistribution of Cy5-labelled mRNA, mice were killed 1 h post-injection of LNP A or B, and organs were collected and imaged using a PerkinElmer Lumina In Vivo Animal Imaging System (IVIS) at excitation and emission wavelength 620 nm and 670 nm, respectively. Luminescence measurements using IVIS were performed as previously described 63 . In brief, images were acquired at 10 s exposure 10 min after intraperitoneal injection with 200 µl of 15 mg ml −1 luciferin. Tissue fixation was carried out as described before 24 . In brief, the extracted organs were immersed in 4% paraformaldehyde in PBS at 4 °C overnight. The next day, organs were incubated in 10% sucrose in PBS for 4 h, 15% sucrose in PBS for 4 h and 20% sucrose in PBS overnight at 4 °C. The organs were then embedded in OCT compound and frozen gradually at −80 °C. Serial cross-sections of the uterus were then prepared at 14 μm thickness using a cryostat (Leica Biosystems) and placed onto positively charged slides. For immunostaining, the sections were thoroughly washed and treated with a blocking buffer (2% BSA in PBS containing 0.1% w/v Tween 20) for 1 h at room temperature. EGFP was stained with a rabbit GFP primary antibody (Thermo, A11122) at 300× dilution in blocking buffer at 4 °C overnight. The next day, the primary antibody was washed off, and the sections were incubated with goat anti-rabbit IgG conjugated to Alexa Fluor 488 (Thermo, A11034) at 500× dilution for 1 h. After the last washing step, the slides were stained with DAPI and observed using an Axiovert epifluorescence microscope or LSM 710 confocal laser scanning microscope (Zeiss). For co-staining with an epithelial cell marker, the sections were labelled with a mouse Alexa Fluor 647 anti-pan cytokeratin antibody (Abcam, C-11) mixed at 300× dilution with the GFP primary antibody (Thermo, A11122). For co-staining with a stromal cell marker, rabbit Alexa Fluor 647 anti-vimentin antibody (Abcam, EPR3776) and goat FITC anti-GFP antibody (Abcam, Ab6662) were both diluted at 300× and incubated at 4 °C overnight before microscopic evaluation as described above. Unmated or mated (p.c. day 5) CD-1 mice were infused 20 µl containing 4 µg GM-CSF mRNA-loaded LNPs or 3.6 µg recombinant mouse GM-CSF in the lumen of the right uterine horn, a dose matching previous intrauterine protein infusion studies in rodents 64 . At definite time points, mice were killed, and the organs were collected, weighted and homogenized in 0.5 ml 1× reporter lysis buffer. The blood was collected in heparinized tubes from the inferior vena cava of mice under isoflurane anaesthesia just before killing. The plasma was collected after centrifugation at 2,000 ×  g for 5 min at 4 °C and stored at −80 °C until analysed. The plasma and the tissue homogenate supernatants were adequately diluted and assayed for the content of GM-CSF using a mouse GM-CSF enzyme-linked immunosorbent assay (ELISA) kit according to the manufacturer’s protocol. To evaluate receptor-bound and tissue-retained GM-CSF protein fraction in the pharmacokinetic study, the uterine lumen was flushed with 1 ml PBS after tissue collection to remove unbound protein at each time point. An endometrial injury model was induced by infusing ethanol into the uterus as previously described 34 . In brief, the uterine horns were exposed and isolated as described above. Twenty-five microlitres of 95% ethanol was infused into the right horn, while the left horn was ligated near the cervix using a surgical clamp to prevent ethanol leakage. The non-injured left uterine horn was untreated, without buffer administration, and used as an internal control for successful implantation after mating. Extended-release buprenorphine (3.25 mg kg −1 ) was injected at the site of surgery. Mice were then housed for 7 days before mating with male mice (1:1 ratio). On p.c. day 2, mice were randomized and the second infusion (20 µl containing PBS, recombinant GM-CSF, GM-CSF mRNA-loaded LNP B or empty LNP B) was administered to the right injured horn, while the left non-injured uterine horn was untreated, without buffer administration. Mice were injected with another dose of buprenorphine (3.25 mg kg −1 ) at the incision site. On p.c. days 7–9, mice were killed, and uteruses were extracted for counting embryo implantation sites and histopathology. Uterus segments that lie between embryo attachments were excised and fixed in 4% paraformaldehyde at 4 °C overnight. Tissues were then sent to the JHMI Reference Histology core for paraffin embedding, cross-sectioning at 6 μm thickness and staining with haematoxylin and eosin. Sections were imaged and analysed using a Nikon light microscope. The endometrium thickness was calculated as the average of four measurements capturing the endometrial layer length extending from the luminal layer to the myometrial layer. Evaluable uterine sections obtained from inter-attachment sites showing a contiguous and visible luminal space at p.c. days 7–9 were reported. The statistical significance between the two groups was analysed using an unpaired, two-tailed Student’s t -test. In the fertility experiment comparing embryo counts between the left and right uterine horns in the same mouse, two-tailed, paired t -tests were used. Multiple comparisons among three or more groups were performed using one- or two-way analysis of variance (ANOVA) followed by Tukey’s post hoc test unless indicated otherwise. Comparison against untreated controls was performed using Dunnett’s test. The statistical significance in the % of animals showing at least 1 implantation site was calculated using a 2 × 2 contingency table (two-tailed, Fisher’s exact test) against the non-injured controls. A statistically significant difference was set at P  < 0.05. The P value below 0.0001 is reported as P  < 0.0001. No data were excluded in the analyses. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Untargeted LNPs containing 0% RGD-lipid (LNPs A, F and I) showed high luciferase expression in the liver after intravenous injections (Fig. 3h ). Increasing the amount of C18-RGD from 0 mol% to 5 mol% reduced the luciferase expression by 100-fold (LNP A versus B), 10-fold (LNP F versus H) and 4-fold (LNP I versus K) (Fig. 3h ). Intriguingly, LNP D (1.5 mol% C18-PEG2000-RGD) exhibited 2.4- and 14-fold higher liver expression compared with LNPs C and G, respectively, both of which contain 1.5 mol% C18-RGD and 1.5 mol% C18-PEG lipid (Fig. 3h ). This further suggests that attaching RGD directly to the C18 lipid could reduce luciferase expression in the liver. While the intravenous injection of LNPs A–L showed a less predictable effect on luciferase expression in the spleen, LNP B reduced expression in the spleen by >2-fold compared with LNP A (Supplementary Fig. 7a ). Intravenous injection of LNPs A–L resulted in negligible luciferase expression in the uterus (Supplementary Fig. 7b ). We then normalized luciferase expression in the liver for each LNP to that of LNP B, which was the lowest (Fig. 3i ). We also confirmed that intravenous or intrauterine administration of LNP B induces dramatically reduced luciferase expression in the mouse lungs, heart and kidneys (Supplementary Figs. 8 and 9 ). The reduced luciferase expression in the liver, spleen and other off-target tissues observed by LNP B could be due to limited mRNA biodistribution or poor cellular/intracellular trafficking. Using Cy5-labelled mRNA loaded in LNPs, we observed both LNPs A and B accumulated similarly in the liver and spleen after intravenous administration (Supplementary Fig. 10a–f ). This suggests that LNP B can access the liver and spleen but shows reduced functional mRNA delivery in these organs. This is consistent with a previous study demonstrating that mRNA-LNPs targeted to other organs can still accumulate in the liver while functional protein expression was enhanced in the target organ 43 . In addition, both LNPs A and B have pKa values of 6.2–6.4 (Supplementary Fig. 10g,h ), which is in the range required to deliver nucleic acids to hepatocytes efficiently 43 . This suggests that pKa alone may not be sufficient to assess the tissue tropism of ligand-conjugated ionizable LNPs. While a full mechanistic analysis is required, it could be argued that the direct addition of RGD to surface of LNP B interfered with its intracellular trafficking in the liver or spleen, resulting in reduced luciferase expression. Taken together, we show that engineering LNP surfaces with C18-RGD and an appropriate PEG-lipid stabilizer component (termed bifunctional ligands via organized conjugation, or BLOC) can maximize local targeting efficiency and reduce off-target systemic expression (Supplementary Fig. 11 ).

In this study, we found that functionalizing mRNA-LNPs with RGD directly attached to C18 induced high endometrial targeting and reduced off-targeted systemic protein expression when administered during the WOI. Previously described conjugation approaches have involved attaching the RGD peptides on the terminus of a polymer spacer, typically PEG, to facilitate interactions with integrins after systemic administration 45 , 46 . The PEG spacer-free ligand conjugation approach described herein could have improved endometrial targeting efficiency for two reasons. First, PEG-lipid above 1.5–3 mol% can negatively affect cellular uptake and mRNA entrapment 40 , 47 . Therefore, functionalizing LNP B with a high ligand density up to 5 mol% could have promoted multivalent interactions with integrins, without using excessive amount of PEG spacers. Second, the high flexibility of PEG chains can lead to ligand entanglement and shrouding, thereby reducing the number of ligand–receptor binding events 48 . Notably, RGD binding sites can be reached by peptides that extend only by 1–3 nm from the surface of particles 49 . While protein corona formation on the LNP can interfere with receptor interactions, the protein concentration in the uterine tubal fluids is approximately 5–10% of that in serum 50 . Therefore, the impact of protein corona formation on LNPs administered in the uterine cavity on cellular uptake and receptor interactions could be reduced. We unexpectedly discovered that attaching RGD directly to C18 in mRNA-LNPs reduced encoded protein expression in the liver, spleen and other off-target organs. This effect was observed irrespective of the delivery window (during, just before or outside the WOI), the formulation (LNPs containing various PEG-lipid stabilizers) and the route of administration (intrauterine or intravenous administrations). High-density RGD functionalization altered LNP surface properties, potentially changing protein corona formation or intracellular trafficking, leading to reduced protein expression in off-target tissues. We herein present a new concept for functionalizing mRNA-LNPs with BLOC as a strategy for maximizing local targeting efficiency and reducing off-target systemic protein expression. Here we demonstrated that a single intrauterine infusion of GM-CSF mRNA-loaded LNP B restored fertility rates, outperforming recombinant GM-CSF infusion. While CSF cytokines showed promise in enhancing implantation rates in women with RIF and/or TE, the timing of administration might be critical. Several studies have shown positive results in RIF by administering CSF cytokines 3 days before 51 , on the same day 52 or only 1–2 min before embryo transfer in an ART cycle 53 . In women with TE, CSF cytokines are typically administered during ovarian stimulation to allow sufficient increase in endometrial thickness before embryo transfer. RIF may be a more straightforward indication to demonstrate the full potential for integrin targeting in improving mRNA delivery to the endometrium during WOI. Moreover, since TE is a chronic condition, it is conceivable that women could undergo monthly treatments, delivering LNP B at the peak of the WOI to induce a persistent increase in endometrial thickness before embryo transfer. That being said, there are two biological limitations of this study. First, the ethanol injury model may not accurately reflect endometrial injury in humans, as most rodents do not menstruate and endometrial repair mechanisms may differ. Second, it has been suggested that women with unexplained infertility may exhibit reduced integrin expression during the WOI 54 . However, the reduction was seen in specific subunits, and not all integrin receptors. Further, a more recent study found no difference in αvβ3 expression between fertile and infertile women 55 . The impact of reduced integrin expression in infertility on LNP delivery efficiency remains to be carefully characterized. From a safety viewpoint, CSF cytokines can lead to tumour progression, migration and other immune-related adverse effects 56 , 57 . LNP B could reduce cytokine dose-limiting toxicity by reducing systemic protein exposure in the blood or other tissues. While the potential for inflammation caused by ionizable cationic lipids in the uterus remains to be studied, previous reports showed that intrauterine inflammation induced by endometrial scratching did not impair embryo implantation in humans 58 – 60 . LNP B could be a valuable tool to introduce a range of therapeutic proteins into the endometrium and modulate their expression kinetics for treatment of endometrial pathologies with low safety concerns.

As an initial targeting strategy, we used an RGD attached to distearoyl phosphatidylethanolamine (C18-RGD) for its potential role in enhancing binding avidity to cell surface-expressed integrins 35 . To minimize steric hindrance that could impede RGD-integrin binding, the polyethylene glycol (PEG)-lipid formulation stabilizer used had a relatively low PEG molecular weight (C18-PEG350). Untargeted LNP (LNP A) and targeted LNP (LNP B) showed comparable Z -average size, polydispersity indices (PDI) and encapsulation ratios (Fig. 1a,b and Supplementary Table 1 ). LNP B showed significantly lower ζ -potential than LNP A (Fig. 1b ), presumably owing to the net negative charge of C18-RGD. Fig. 1 Physicochemical characterization and in vivo delivery efficiency of RGD-modified LNPs. a , The composition of untargeted LNP (LNP A) and RGD-modified LNP (LNP B). LNP A was prepared by mixing ionizable lipid, phospholipid, cholesterol and C18-PEG350 at a molar ratio of 50:10:38.5:1.5. LNP B was prepared by adding C18-RGD to LNP A at 5 mol% of total lipids. b , Physicochemical characteristics of LNPs A and B. Size is reported as the Z -averaged hydrodynamic diameter. Data represent the mean ± s.e.m. of three independent formulations. Unpaired, two-tailed Student’s t -test. c – e , The diffusion of nanoparticles on mouse endometrium excised on p.c. day 5. LNPs A and B were labelled with 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindodicarbocyanine, 4-chlorobenzenesulfonate (DiD) and PS-COOH with a 580/605 fluorophore. Time-resolved changes in MSD of PS-COOH, LNP A and LNP B ( c ). Data represent the mean ± s.e.m. of multiple video measurements obtained from various locations on the same mouse endometrial surface (PS-COOH, n  = 8; LNP A, n  = 10; LNP B, n  = 12). Distribution of individual particle MSD at a timescale = 1 s ( d ). Data represent the mean ± s.e.m. of 3 mice, each replicate representing the average of 8–12 video measurements. The highlighted area represents the error bands. The % particle distribution histograms were fitted into a Gaussian model using a nonlinear regression of 189 degrees of freedom to calculate the mean log 10 (MSD t  = 1 s) ± s.e.m. The dotted vertical red line indicates the highly mobile fraction of nanoparticles showing log 10 (MSD t  = 1 s) > −1. Quantification of highly mobile nanoparticle fraction showing log 10 (MSD t  = 1 s) > −1 ( e ). Data represent the mean ± s.e.m. ( n  = 3 mice). One-way ANOVA followed by Tukey’s post hoc test. f , Luciferase expression normalized to mg protein in tissue homogenates 4 h post intrauterine infusion of 2 µg fLuc mRNA loaded into LNPs in CD-1 mice at p.c. day 5. Treatment was infused into the mid-distal right uterine horn. Data represent the mean ± s.e.m. ( n  = 5 animals for LNP groups and n  = 3 for the untreated group). Two-way ANOVA followed by Tukey’s post hoc test. g , Luciferase expression normalized to mg protein 4 h post intrauterine infusion of 2 µg fLuc mRNA loaded into LNPs in unmated CD-1 mice (that is, outside WOI). Mice in diestrus are represented by open circles or squares, whereas mice in non-diestrus (estrus, proestrus or metestrus) are represented by closed circles or squares. Data represent the mean ± s.e.m. ( n  = 6 mice). Two-tailed, unpaired t -tests. h , Uterus-to-liver luciferase expression ratios calculated from f and g . Data represent the mean ± s.e.m. ( n  = 5 mice for the WOI group, n  = 6 for the unmated group). Two-way ANOVA followed by Fisher’s least significant difference (LSD) post hoc test. i , Uterus-to-spleen luciferase expression ratios calculated from f and g . Data represent the mean ± s.e.m. ( n  = 5 mice for the WOI group, n  = 6 for the unmated group). Two-way ANOVA followed by Fisher’s LSD post hoc test. Mice in diestrus are represented by open circles, whereas mice in non-diestrus (estrus, proestrus or metestrus) are represented by closed circles. Illustrations in a , f and g created with BioRender.com . Source data a , The composition of untargeted LNP (LNP A) and RGD-modified LNP (LNP B). LNP A was prepared by mixing ionizable lipid, phospholipid, cholesterol and C18-PEG350 at a molar ratio of 50:10:38.5:1.5. LNP B was prepared by adding C18-RGD to LNP A at 5 mol% of total lipids. b , Physicochemical characteristics of LNPs A and B. Size is reported as the Z -averaged hydrodynamic diameter. Data represent the mean ± s.e.m. of three independent formulations. Unpaired, two-tailed Student’s t -test. c – e , The diffusion of nanoparticles on mouse endometrium excised on p.c. day 5. LNPs A and B were labelled with 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindodicarbocyanine, 4-chlorobenzenesulfonate (DiD) and PS-COOH with a 580/605 fluorophore. Time-resolved changes in MSD of PS-COOH, LNP A and LNP B ( c ). Data represent the mean ± s.e.m. of multiple video measurements obtained from various locations on the same mouse endometrial surface (PS-COOH, n  = 8; LNP A, n  = 10; LNP B, n  = 12). Distribution of individual particle MSD at a timescale = 1 s ( d ). Data represent the mean ± s.e.m. of 3 mice, each replicate representing the average of 8–12 video measurements. The highlighted area represents the error bands. The % particle distribution histograms were fitted into a Gaussian model using a nonlinear regression of 189 degrees of freedom to calculate the mean log 10 (MSD t  = 1 s) ± s.e.m. The dotted vertical red line indicates the highly mobile fraction of nanoparticles showing log 10 (MSD t  = 1 s) > −1. Quantification of highly mobile nanoparticle fraction showing log 10 (MSD t  = 1 s) > −1 ( e ). Data represent the mean ± s.e.m. ( n  = 3 mice). One-way ANOVA followed by Tukey’s post hoc test. f , Luciferase expression normalized to mg protein in tissue homogenates 4 h post intrauterine infusion of 2 µg fLuc mRNA loaded into LNPs in CD-1 mice at p.c. day 5. Treatment was infused into the mid-distal right uterine horn. Data represent the mean ± s.e.m. ( n  = 5 animals for LNP groups and n  = 3 for the untreated group). Two-way ANOVA followed by Tukey’s post hoc test. g , Luciferase expression normalized to mg protein 4 h post intrauterine infusion of 2 µg fLuc mRNA loaded into LNPs in unmated CD-1 mice (that is, outside WOI). Mice in diestrus are represented by open circles or squares, whereas mice in non-diestrus (estrus, proestrus or metestrus) are represented by closed circles or squares. Data represent the mean ± s.e.m. ( n  = 6 mice). Two-tailed, unpaired t -tests. h , Uterus-to-liver luciferase expression ratios calculated from f and g . Data represent the mean ± s.e.m. ( n  = 5 mice for the WOI group, n  = 6 for the unmated group). Two-way ANOVA followed by Fisher’s least significant difference (LSD) post hoc test. i , Uterus-to-spleen luciferase expression ratios calculated from f and g . Data represent the mean ± s.e.m. ( n  = 5 mice for the WOI group, n  = 6 for the unmated group). Two-way ANOVA followed by Fisher’s LSD post hoc test. Mice in diestrus are represented by open circles, whereas mice in non-diestrus (estrus, proestrus or metestrus) are represented by closed circles. Illustrations in a , f and g created with BioRender.com . Source data We then sought to characterize the mobility of LNPs A and B on the mucosal surface of intact mouse endometrium during the WOI. While integrin overexpression peaks during the WOI (days 20–24) of the human menstrual cycle 32 , in mice, WOI occurs on post-coitus (p.c.) days 4–5 (ref. 36 ). We used fluorescently labelled carboxylate functionalized polystyrene nanoparticle (PS-COOH) as a negative control for poor mucus penetration 37 . PS-COOH showed poor diffusion on the endometrial surface as indicated by low mean squared displacement (MSD) (Fig. 1c ). LNPs A and B showed 40- and 80-fold increase in MSD at t  = 1 s compared with PS-COOH, respectively ( P  < 0.001, P  < 0.001; two-tailed, unpaired Student’s t -test; Fig. 1c ). Histograms of individual particle MSD values at a timescale of 1 s showed that both LNPs A and B exhibited ~2-fold higher mean log 10 (MSD t  = 1 s) than PS-COOH (Fig. 1d ). Approximately, 64% of LNP A and 48% of LNP B particles were classified as ‘highly mobile’ (mean log 10 (MSD t  = 1 s) > −1), while only 11.6% of PS-COOH particles were highly mobile (Fig. 1e ). These results indicate the high mobility of both LNPs A and B on mouse endometrium. LNP B showed slightly reduced mobility, which could be due to increased binding of RGD-functionalized LNP B on the endometrial surface. When infused during the WOI, LNP B loaded with firefly luciferase (fLuc) mRNA provided a 3.9-fold increase in luciferase expression in the uterus compared with LNP A loaded with the same mRNA (Fig. 1f ). Surprisingly, luciferase expression in the liver and spleen was 450- and 65-fold reduced, respectively, with LNP B relative to LNP A (Fig. 1f ). When infused in mice outside the WOI (that is, unmated), there was no significant difference in luciferase expression in the uterus, but a non-significant trend for reduced expression in the liver and spleen for LNP B compared with LNP A was seen (Fig. 1g ). No apparent impact of the estrous cycle (diestrus versus non-diestrus) on the delivery efficiency of LNPs A or B was observed in unmated mice (Fig. 1g ). However, when dosed during the WOI, LNP B showed a significant increase in the ratios of luciferase expression in the uterus relative to the liver (uterus/liver, 6,300-fold) and spleen (uterus/spleen, 1,100-fold) compared with LNP A (Fig. 1h,i ). In addition, only LNP B showed a significant increase in uterus/liver and uterus/spleen luciferase expression in mice dosed during the WOI compared with dosing in unmated mice (Fig. 1h,i ). Further, when dosed at p.c. day 2 (that is, before the peak of integrin expression), luciferase expression in the uterus was similar between LNP A and LNP B, although LNP B showed reduced luciferase expression in the liver and spleen compared with LNP A (Supplementary Fig. 1 ).

We next sought to characterize the LNP compositional elements leading to high uterus targeting and low off-target expression after intrauterine infusions. To this end, we created an initial library of 12 LNPs varying in the amount of RGD-lipid, the topology of RGD conjugation and the type of PEG-lipid stabilizer (Fig. 3a , f and Supplementary Table 1 ). LNPs A–L had Z -average sizes of 115–161 nm, PDI of 0.09–0.36, ζ -potential of –1.7 mV to –17 mV, and encapsulation ratios 51–94% (Supplementary Fig. 3 ). First, we formulated LNP C (1.5 mol% C18-RGD) as an intermediate ligand concentration between LNPs A and B. In LNPs A–C, the PEG-lipid used was 1.5 mol% C18-PEG350. Increasing the C18-RGD amount enhanced luciferase expression in the uterus and reduced the expression in the liver and spleen compared with LNP A (Fig. 3b ). There was a significant decrease in the ζ -potential by increasing the C18-RGD amounts from LNP A to C (−8.7 mV to −12.7 mV, P  = 0.0109) and C to B (−12.7 mV to −16 mV, P  = 0.0270) (Supplementary Fig. 3c ), which could be attributed to the anionic charge of C18-RGD and reduced ligand masking on the LNP surface facilitated by the relatively short PEG350. Fig. 3 Impact of RGD conjugation strategies on local and systemic mRNA-LNP delivery efficiencies. a , Schematic diagram illustrating the LNP compositions used to study the impact of RGD conjugation of LNPs administered by the intrauterine and intravenous routes. b – g , Two micrograms fLuc mRNA loaded into LNPs was administered by intrauterine infusion at p.c. day 5 in CD-1 mice, followed by evaluation of luciferase expression 4 h post-infusion. LNPs A–C formulated using varying amounts of C18-RGD and a fixed amount of C18-PEG350 (1.5 mol%) as the PEG-lipid stabilizer component ( b ). Data represent the mean ± s.e.m. ( n  = 5 mice for LNP A, n  = 4 for LNPs B and C). Ligand effects on mRNA delivery efficiency ( c ). First, the extent of luciferase expression mediated by LNP B uptake by integrin receptors was estimated by preincubation with a competitive ligand. The right uterine horn was infused with a 10 µl solution containing 0.5 mg ml −1 vitronectin 5 min before intrauterine infusion of LNP B in the same uterine horn. Second, the effect of ligand conjugation strategy was evaluated. LNPs D and E contained 1.5 mol% and 5 mol% C18-PEG2000-RGD, respectively, whereas LNP B contained 5 mol% C18-RGD. Data represent the mean ± s.e.m. ( n  = 4 mice). Data equal to zero is not shown in the graph. LNPs F–H formulated using varying amounts of C18-RGD and a fixed amount of C18-PEG2000 (1.5 mol%) ( d ). Data represent the mean ± s.e.m. ( n  = 4 mice). LNPs I–K formulated using varying amounts of C18-RGD and a fixed amount of C14-PEG2000 (1.5 mol%) ( e ). Data represent the mean ± s.e.m. ( n  = 4 mice). LNP L formulated using 5 mol% C18-RGD and 1.5 mol% C14-PEG350 ( f ). Data represent the mean ± s.e.m. ( n  = 3 mice for LNP B, n  = 4 for LNP L). Uterus targeting efficiency of LNPs A–L as measured by the percentage of luciferase expression in the uterus relative to total expression in the uterus, liver and spleen ( g ). Data represent the mean ± s.e.m. ( n  = 4–5 mice for all LNPs, except LNP B n  = 16). h , Luciferase expression in the liver 4 h following tail-vein injections of LNPs A–L containing 2 μg fLuc mRNA in unmated female CD-1 mice. Data represent the mean ± s.e.m. ( n  = 3 mice, except LNP B, n  = 6). One-way ANOVA followed by Tukey’s post hoc test. i , Normalized liver luciferase expression following tail-vein injections of LNPs A–L as calculated by fold changes relative to LNP B. Data represent the mean ± s.e.m. ( n  = 3 mice for all LNPs, except n  = 6 for LNP B). Illustrations created with BioRender.com . a , Schematic diagram illustrating the LNP compositions used to study the impact of RGD conjugation of LNPs administered by the intrauterine and intravenous routes. b – g , Two micrograms fLuc mRNA loaded into LNPs was administered by intrauterine infusion at p.c. day 5 in CD-1 mice, followed by evaluation of luciferase expression 4 h post-infusion. LNPs A–C formulated using varying amounts of C18-RGD and a fixed amount of C18-PEG350 (1.5 mol%) as the PEG-lipid stabilizer component ( b ). Data represent the mean ± s.e.m. ( n  = 5 mice for LNP A, n  = 4 for LNPs B and C). Ligand effects on mRNA delivery efficiency ( c ). First, the extent of luciferase expression mediated by LNP B uptake by integrin receptors was estimated by preincubation with a competitive ligand. The right uterine horn was infused with a 10 µl solution containing 0.5 mg ml −1 vitronectin 5 min before intrauterine infusion of LNP B in the same uterine horn. Second, the effect of ligand conjugation strategy was evaluated. LNPs D and E contained 1.5 mol% and 5 mol% C18-PEG2000-RGD, respectively, whereas LNP B contained 5 mol% C18-RGD. Data represent the mean ± s.e.m. ( n  = 4 mice). Data equal to zero is not shown in the graph. LNPs F–H formulated using varying amounts of C18-RGD and a fixed amount of C18-PEG2000 (1.5 mol%) ( d ). Data represent the mean ± s.e.m. ( n  = 4 mice). LNPs I–K formulated using varying amounts of C18-RGD and a fixed amount of C14-PEG2000 (1.5 mol%) ( e ). Data represent the mean ± s.e.m. ( n  = 4 mice). LNP L formulated using 5 mol% C18-RGD and 1.5 mol% C14-PEG350 ( f ). Data represent the mean ± s.e.m. ( n  = 3 mice for LNP B, n  = 4 for LNP L). Uterus targeting efficiency of LNPs A–L as measured by the percentage of luciferase expression in the uterus relative to total expression in the uterus, liver and spleen ( g ). Data represent the mean ± s.e.m. ( n  = 4–5 mice for all LNPs, except LNP B n  = 16). h , Luciferase expression in the liver 4 h following tail-vein injections of LNPs A–L containing 2 μg fLuc mRNA in unmated female CD-1 mice. Data represent the mean ± s.e.m. ( n  = 3 mice, except LNP B, n  = 6). One-way ANOVA followed by Tukey’s post hoc test. i , Normalized liver luciferase expression following tail-vein injections of LNPs A–L as calculated by fold changes relative to LNP B. Data represent the mean ± s.e.m. ( n  = 3 mice for all LNPs, except n  = 6 for LNP B). Illustrations created with BioRender.com . To characterize the role of integrin binding in the efficiency of mRNA delivery by RGD-modified LNP, we then pretreated the uterus with vitronectin as a competitive ligand 38 . Luciferase expression was ~10 times lower in the uterus pretreated with vitronectin followed by LNP B treatment relative to LNP B treatment with no vitronectin pretreatment, emphasizing the role of integrin-RGD interactions in mediating LNP B uptake (Fig. 3c ). We then investigated the role of RGD conjugation topology on uterus targeting. We used RGD conjugated to a C18 lipid with a 2,000 g mol −1 PEG spacer (C18-PEG2000-RGD) at 2 concentrations: 1.5 mol% (LNP D) and 5 mol% (LNP E) (Fig. 3a ). The luciferase expression in the uterus induced by LNP B remained 1.8- and 5.8-fold higher than LNPs D and E, respectively (Fig. 3c ). LNP E showed no improvement in luciferase expression in the uterus relative to the LNP B group pretreated with vitronectin (Fig. 3c ) and substantially reduced mRNA encapsulation (51%; Supplementary Fig. 3d ). LNP D showed 5–10-fold higher luciferase expression in the liver compared with LNPs B and C (Fig. 3c ). This suggested that attaching RGD directly to the C18 lipid enhanced uterus targeting and reduced off-target systemic protein expression. A PEG-lipid bearing a 2,000 g mol −1 PEG is typically used as a stabilizer in LNP formulations. We formulated LNPs using 1.5 mol% C18-PEG2000 and varied the C18-RGD concentration from 0 mol% to 5 mol% (Supplementary Table 1 ). There was no enhancement in the luciferase expression in the uterus by increasing the C18-RGD amount (LNPs F–H; Fig. 3d ). There was no significant change in ζ -potential in LNPs F–H, which ranged from −6 mV to −8 mV (Supplementary Fig. 3c ). The estimated contour length of a PEG2000 in water is 12.7 nm (ref. 39 ), which can potentially block the interaction between RGD tethered on the LNP surface and integrins. Notably, 2–24-fold reduction in liver and spleen expression was still observed with the addition of C18-RGD in LNPs G and H relative to LNP F (Fig. 3d ). Further, we formulated LNPs with 1.5 mol% C14-PEG2000, an excipient used in the Food and Drug Administration (FDA)-approved LNPs such as Onpattro and Spikevax. There was a statistically insignificant increase in luciferase expression in the uterus accompanied by a 2–15-fold reduction in liver and spleen expression in LNPs J and K relative to LNP I (Fig. 3e ). There was statistically insignificant decrease in the ζ -potential by increasing the C18-RGD amounts from LNP I to J (−1.7 mV to −7 mV) and J to K (–7 mV to –13.7 mV) (Supplementary Fig. 3c ). C14-PEG2000 can undergo shedding upon dilution in biological fluids 40 , potentially increasing RGD exposure on the surface. We also show that the LNP prepared using 1.5 mol% C14-PEG350 and 5 mol% C18-RGD (LNP L) can induce high uterus expression and low liver and spleen expression similar to LNP B (Fig. 3f ). We then calculated the % luciferase expression in the uterus in LNPs A–L (Fig. 3g ). LNPs B, K and L showed the highest uterus delivery efficiency, with >97% of the total luciferase expression originating from the uterus (Fig. 3g ). We further evaluated the delivery efficiency of LNPs containing >5 mol% C18-RGD (7.5–30 mol%; LNPs M–P; Supplementary Table 1 ). While 1.5 mol%, 5 mol% and 7.5 mol% C18-RGD enhanced luciferase expression in the uterus compared with 0% modification (LNP A) (Fig. 3b ), C18-RGD densities >7.5 mol% reduced luciferase expression in the uterus in a C18-RGD concentration-dependent manner (Supplementary Fig. 4a ). Notably, luciferase expression in the liver or spleen was consistently reduced with increasing C18-RGD amounts (Supplementary Fig. 4b,c ). Five mol% C18-RGD (that is, LNP B) showed the highest uterus luciferase expression relative to the liver and spleen (Supplementary Fig. 4 ). We then calculated the theoretical RGD density on the surface of the LNP B, and the average distance between adjacent RGD peptides was estimated to be 4.8–6.8 nm (Supplementary Fig. 5 ). Such spacing is similar to the penton base protein of adenovirus 2, where the distance between immobilized RGDs was estimated to be 5.7–9 nm (ref. 41 ). Since LNP physicochemical characteristics can influence LNP uptake and biodistribution, we plotted the luciferase expression in the uterus, spleen or liver against LNP Z -average size, PDI, ζ -potential or mRNA encapsulation ratio. We did not observe any strong correlation between luciferase expression in the uterus, liver or spleen and any of the four particle characteristics (Supplementary Fig. 6a–c ). Other LNP physicochemical properties, such as stiffness and internal structure 42 , could influence endometrial targeting efficiency and warrant further studies in the future.

After intrauterine infusion of LNP B containing enhanced green fluorescent protein (EGFP) mRNA during the WOI, an EGFP signal was seen in luminal epithelium, uterine glands and stroma of the endometrium (Fig. 2a,b ). The EGFP signal penetrated the whole endometrial layer with minimal apparent expression observed in the myometrium or the perimetrium (Fig. 2b ). High co-localization between pan-cytokeratin staining and EGFP expression confirmed mRNA delivery to luminal epithelium (Fig. 2c ). Co-localization between vimentin staining and EGFP expression confirmed mRNA delivery to endometrial stromal cells (Fig. 2d ). Near the implantation site on p.c. day 6, we observed EGFP signal in endometrial epithelium (Supplementary Fig. 2a ). While EGFP expression was visualized in the epithelial and stromal endometrial cells after treatment with either LNP A or LNP B (Supplementary Fig. 2b ), a higher number of EGFP-expressing cells were observed in the liver and spleen in mice treated with LNP A relative to LNP B (Supplementary Fig. 2c,d ). Fig. 2 Intrauterine distribution of encoded protein expression. a , Untreated uterus representing the inter-attachment site at p.c. day 6 in CD-1 mice. Scale bar, 50 μm. b , Uterus treated with 2 μg EGFP mRNA loaded into LNP B and infused at p.c. day 5. After 24 h, uteruses were collected, and inter-attachment segments were observed. *, epithelium; #, uterine glands; arrows, stroma; Myo, myometrium; Peri, perimetrium; Endo, endometrium. Scale bar, 50 μm. c , Immunostaining for pan-cytokeratin. Co-localization (yellow, asterisk) between EGFP expression (green) and pan-cytokeratin (red) indicates mRNA delivery to epithelial cells. Scale bar, 10 μm. d , Immunostaining for vimentin. Co-localization (yellow, arrows) between EGFP expression (green) and vimentin (red) indicates mRNA delivery to stromal cells. Scale bar, 10 μm. Each image shown is a representative section of the right uterine horn of one mouse. EGFP expression following LNP B infusion was confirmed in three independent animals. Source data a , Untreated uterus representing the inter-attachment site at p.c. day 6 in CD-1 mice. Scale bar, 50 μm. b , Uterus treated with 2 μg EGFP mRNA loaded into LNP B and infused at p.c. day 5. After 24 h, uteruses were collected, and inter-attachment segments were observed. *, epithelium; #, uterine glands; arrows, stroma; Myo, myometrium; Peri, perimetrium; Endo, endometrium. Scale bar, 50 μm. c , Immunostaining for pan-cytokeratin. Co-localization (yellow, asterisk) between EGFP expression (green) and pan-cytokeratin (red) indicates mRNA delivery to epithelial cells. Scale bar, 10 μm. d , Immunostaining for vimentin. Co-localization (yellow, arrows) between EGFP expression (green) and vimentin (red) indicates mRNA delivery to stromal cells. Scale bar, 10 μm. Each image shown is a representative section of the right uterine horn of one mouse. EGFP expression following LNP B infusion was confirmed in three independent animals. Source data

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41565-025-02108-7.

Source Data Figs. 1, 2, 4 and 5 Source raw data. Source raw data.