Enzymatic cross-linking collagen-based bioink for 3D bioprinting of endometrial model with endometrial epithelial organoids.

Huimin Shi: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Conceptualization. Xin Xu: Writing – original draft, Methodology, Investigation, Formal analysis. Qingru Huang: Resources, Investigation, Formal analysis. Mengying Liu: Methodology, Formal analysis. Xuechai Bai: Resources, Investigation. Yana Ma: Methodology, Formal analysis. Yifan Wang: Resources, Methodology. Xiaona Lin: Writing – review & editing, Supervision, Resources, Methodology, Funding acquisition, Formal analysis, Conceptualization. Jun Yin: Writing – review & editing, Supervision, Resources, Methodology, Funding acquisition, Formal analysis, Conceptualization.

This study was approved by the Ethics Committee of Sir Run Run Shaw Hospital, Zhejiang University School of Medicine (Approval No. SRRSH-2025-0026) and the Laboratory Animal Welfare and Ethics Review Committee, Zhejiang University (Approval No. 31226). Endometrial tissue biopsies were collected from donors at Sir Run Run Shaw Hospital following a standardized protocol. Participants provided written informed consent for biological sample collection and voluntarily contributed endometrial tissues for scientific research purposes.

As illustrated in Fig. 2 a, collagen sponges were converted into suspensions, with acid collagen subjected to stirring at 37 °C for 4, 8, or 12 h. Circular dichroism (CD) spectra ( Fig. 2 b) revealed that all treated groups, along with the untreated control, exhibited a characteristic positive peak at approximately 221 nm and a negative peak around 198 nm [40] , [50] , while sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) analysis ( Fig. 2 e) confirmed the presence of distinct bands near 250 kDa (β-chain) and 130 kDa (α1 and α2 chains) [40] . These findings suggest that stirring at 37 °C for up to 12 h did not result in detectable alterations in the triple-helix-associated structural features or chain integrity of collagen under the tested conditions. However, dynamic light scattering (DLS) measurements ( Fig. 2 c and d) showed a progressive decrease in apparent particle size with increasing stirring duration under dilute measurement conditions. Rheological characterization ( Fig. 3 a) revealed that all three treated groups exhibited increased storage modulus ( G′ ) and loss modulus ( G″ ), accompanied by decreased loss tangent (tan δ  =  G" / G′ ), as temperature decreased from 27 °C to 4 °C. At equivalent temperatures, longer stirring durations resulted in lower G′ and G″ values and higher tan δ values. Based on these observations, and considering the balance between enzymatic crosslinking rate and hydrogel strength, the 8-h stirring condition was selected for subsequent bioink preparation and printing experiments. Fig. 2 Preparation and characterization of materials. ( a ) Overview of the key steps in preparation of collagen suspension. ( b ) CD spectra of the suspension. ( c ) Particle distributions with different sizes of the suspension. ( d ) Mean sizes of the suspension ( n  = 3). ( e ) SDS-PAGE of collagen suspension. ( f ) In vitro degradation of collagen-based hydrogels ( n  = 5). ( g ) In vivo degradation of C5H2T1 hydrogel ( n  = 2 animals per time point). ( h ) Mechanism of TG-catalyzed collagen crosslinking. ( i ) Stress-strain curves at 37 °C. ( j ) Compressive moduli of hydrogels ( n  = 5) at 37 °C. ( k ) Equilibrium swelling properties of hydrogels ( n  = 8). Data are presented as mean ± SD. Statistical significance was assessed using one-way ANOVA followed by Tukey's multiple comparisons test. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. Fig. 3 Bioink printability. ( a ) Effect of temperature on the moduli (ⅰ) and tan δ (ⅱ) of collagen suspension. ( b ) Time-dependent changes in G′ of bioinks with different TG concentrations. ( c ) Curing states of bioinks. ( d ) Schematic illustration of the designed printing process. ( e ) Experimental data and fitted curve of G′ (i) and tan δ (ii), and the printable time of the C5H2T1 bioink (iii). The R 2 values of the G ′– t fitting curves were consistently greater than 0.9, indicating good fitting accuracy. The tan δ – t fitting curve also showed good agreement with the experimental data after 0.4 h, with an R 2 value of 0.9468. ( f ) Representative structures printed using the C5H2T1 bioink. Preparation and characterization of materials. ( a ) Overview of the key steps in preparation of collagen suspension. ( b ) CD spectra of the suspension. ( c ) Particle distributions with different sizes of the suspension. ( d ) Mean sizes of the suspension ( n  = 3). ( e ) SDS-PAGE of collagen suspension. ( f ) In vitro degradation of collagen-based hydrogels ( n  = 5). ( g ) In vivo degradation of C5H2T1 hydrogel ( n  = 2 animals per time point). ( h ) Mechanism of TG-catalyzed collagen crosslinking. ( i ) Stress-strain curves at 37 °C. ( j ) Compressive moduli of hydrogels ( n  = 5) at 37 °C. ( k ) Equilibrium swelling properties of hydrogels ( n  = 8). Data are presented as mean ± SD. Statistical significance was assessed using one-way ANOVA followed by Tukey's multiple comparisons test. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. Bioink printability. ( a ) Effect of temperature on the moduli (ⅰ) and tan δ (ⅱ) of collagen suspension. ( b ) Time-dependent changes in G′ of bioinks with different TG concentrations. ( c ) Curing states of bioinks. ( d ) Schematic illustration of the designed printing process. ( e ) Experimental data and fitted curve of G′ (i) and tan δ (ii), and the printable time of the C5H2T1 bioink (iii). The R 2 values of the G ′– t fitting curves were consistently greater than 0.9, indicating good fitting accuracy. The tan δ – t fitting curve also showed good agreement with the experimental data after 0.4 h, with an R 2 value of 0.9468. ( f ) Representative structures printed using the C5H2T1 bioink. Fig. 2 h illustrates the mechanism of TG-catalyzed collagen crosslinking, which proceeds through an acyl-transfer reaction between the γ-carboxamide groups of glutamine residues and the ε-amino groups of lysine residues in collagen. This reaction generates ε-(γ-glutamyl)lysine isopeptide bonds that covalently crosslink collagen molecules, with each crosslinking event releasing one molecule of ammonia [49] . As shown in Fig. 2 i and j, at 37 °C, the C5H0T1 group exhibited the highest compressive modulus (3.55 ± 0.44 kPa), followed by the C5H2T2 (2.76 ± 0.21 kPa) and C5H2T1 (1.05 ± 0.21 kPa) groups. No significant difference was observed between the C5H2T1 group and Matrigel (0.72 ± 0.16 kPa). As illustrated in Fig. 2 k, Matrigel exhibited the highest swelling ratio, reaching approximately 70. The swelling ratio of the C5H2T1 group (16.74 ± 1.70) was significantly higher than that of C5H0T1 group (11.90 ± 0.49). Notably, the swelling ratio of the C5H2T2 group (13.45 ± 3.05) was slightly lower than that of C5H2T1 group. A FITC-dextran diffusion assay was conducted to qualitatively assess molecular transport within the hydrogels. As shown in Supplementary Fig. S1 , FITC-dextran gradually penetrated from the hydrogel surface toward the interior over time. At the same time point, the penetration depth followed the order C5H2T1 > C5H2T2 > C5H0T1. All three groups showed a gradual decrease in remaining mass over 24 h ( Fig. 2 f). The C5H2T2 group generally retained a relatively higher percentage of its initial mass during the intermediate and later stages, whereas the C5H2T1 group showed lower percentages, with the C5H0T1 group displaying intermediate values. By the final time point, the remaining mass percentages of all three groups had further decreased and became more comparable. As shown in Fig. 2 g, the C5H2T1 hydrogel showed a progressive decrease in residual mass during in vivo degradation, with approximately 80%, 45%, and 20% of the initial mass remaining on days 3, 8, and 13, respectively. The collagen suspension, with a pH range of 7.2–7.4, exhibited temperature-sensitive properties ( Fig. 3 a), which displays inadequate rheological properties at 22 °C [43] , [44] , [45] . As depicted in Fig. 3 b and c, collagen-based bioinks containing 0 wt%, 2 wt%, or 4 wt% TG formed gel-like states after incubation at 22 °C for 2 h. Upon transfer to 37 °C for 0.5 h, the TG-free formulation reverted to a liquid-like state, indicating the formation of a weak and thermoreversible network rather than a stable crosslinked hydrogel. In contrast, TG-crosslinked collagen bioinks maintained their hydrogel state and exhibited favorable processability, enabling successful filament extrusion ( Supplementary Fig. S3 ). Complete bioprinting process composes of five sequential steps ( Fig. 3 d). First, the bioink was prepared in a liquid state under 37 °C. Second, the bioink was loaded into the extrusion syringe. Third, the bioink underwent pre-crosslinking at 22 °C for a specific time ( t p ) to modulate the rheology. Fourth, the bioink was extruded to fabricate complex structures using a 3D printer at 22 °C within the time period ( t d ). Finally, the printed constructs were maintained at 22 °C for 1 h to allow further crosslinking and stabilization of the hydrogel structure. Consistent with prior experimental findings [51] , [52] , the bioink can be extruded into filaments when G′ exceeds approximately 30 Pa within the linear viscoelastic region, a criterion used to determine the pre-crosslinking time ( t p ). Gao et al. [53] demonstrated that the loss tangent (tan δ ) of bioinks closely correlates with the uniformity (smoothness) and structural integrity of extruded microfilaments. Here, the printable process was defined to terminate when tan δ  = 0.3, ensuring optimal print quality and structural fidelity. By fitting the time-dependent tan δ curve of the target bioink, the printing duration ( t d ) can be estimated, providing a quantitative basis for guiding process optimization and improving printing consistency. To analytically determine t p and t d for bioprinting, a quantitative model was established to describe the time-dependent rheological evolution of the collagen-based bioink. The rheological evolution at room temperature was hypothetically described using a two-step consecutive first-order reaction scheme, and the corresponding analytical solution was used to formulate the model [54] , [55] , [56] (Eq. (1) ). This kinetic scheme served as a hypothetical phenomenological mathematical representation of the observed rheological evolution. (1) B → k 1 Reaction 1 D → k 2 Reaction 2 P where B represents the initial bioink, D represents the product formed after the first reaction with an apparent reaction rate constant k 1 , and P represents the final product formed after the second reaction with an apparent reaction rate constant k 2 . The governing equations of the two consecutive reactions (Eq. (1) ) can be described as (2) { − d c B ( t ) d t = k 1 c B ( t ) d c D ( t ) d t = k 1 c B ( t ) − k 2 c D ( t ) d c P ( t ) d t = k 2 c D ( t ) c B ( t ) + c D ( t ) + c P ( t ) = a with the initial conditions of c B (0) =  a , c D (0) = 0, c P (0) = 0, where c B  ( t ), c D  ( t ) and c P  ( t ) are the concentrations of reactant B , intermediate product D and final product P , respectively, and the constant a represents the initial concentration of reactant B . The solution of Eq. (2) leads to (3) { c B ( t ) = a exp ( − k 1 t ) c D ( t ) = k 1 a k 2 − k 1 [ exp ( − k 1 t ) − exp ( − k 2 t ) ] c P ( t ) = a [ 1 − k 2 k 2 − k 1 exp ( − k 1 t ) + k 1 k 2 − k 1 exp ( − k 2 t ) ] Assuming the dynamic storage modulus G' t increases linearly with c P ( t ) [54] , G' t can be represented by: (4) G t ′ = G 0 ′ + m c P ( t ) = G 0 ′ + m ( 1 − k 2 k 2 − k 1 exp ( − k 1 t ) + k 1 k 2 − k 1 exp ( − k 2 t ) ] where m is the linear constant. G′ 0 and m can be obtained by fitting the experimental data pertaining to the time-dependent storage modulus ( Fig. 3 e-ⅰ). Additionally, given that tan δ is intrinsically linked to the uniformity (smoothness) and structural integrity of the extruded microfilaments [53] , the rheological profile of tan δ can be fitted by (5) tan δ ( t ) = μ exp ( − λ t ) + φ where μ and λ are fitting constants. Measured data points and fitted curves are shown in Fig. 3 e-ⅱ, and values of parameters are listed in Supplementary Table S3 and Table S4 . Fitted curves show good agreement with the measured experimental data points, with all goodness-of-fit ( R 2 ) values exceeding 0.9. To ensure a stable bioprinting process, two critical parameters should be determined for each bioink: t p and t d . For cell-laden bioprinting, minimizing the pre-crosslinking time is crucial to mitigate potential cellular damage, while maintaining a sufficiently long printable duration is necessary for fabricating large-volume 3D structures. To illustrate this optimization process, an analytical model was developed to predict the rheological properties of 1 wt%-TG-crosslinked collagen bioink (C5H2T1), which guided the determination of optimal t p and t d for printing experiments. Our analysis revealed a linear correlation between the reaction rate constants ( k 1 , k 2 ) and TG concentration, as demonstrated in Supplementary Table S3 and Supplementary Fig. S2a–b . Utilizing the theoretical time-dependent equation of G' (Eq. (4) and Supplementary Eq. S(1) ), we generated the theoretically fitted curve of C5H2T1 bioink, presented in Fig. 3 e-ⅲ. Furthermore, we analyzed parameters μ and λ in Eq. (5) , establishing their linear relationship with TG concentration based on Supplementary Table S4 , with the corresponding curves displayed in Fig. S2c–d . Through this comprehensive analysis, the time-dependent curve of tan δ for C5H2T1 bioink (Supplementary Eq. S(2) ) was obtained and plotted with experimentally measured data ( Fig. 3 e-ⅲ). The determination of critical bioprinting parameters was achieved through systematic analysis of the rheological properties. Specifically, t p was identified as the intersection point between the threshold storage modulus ( G'  = 30 Pa, represented by the blue line) and the fitted G′ - t curve ( Fig. 3 e-ⅲ). Similarly, the endpoint of printing ( t p   +   t d ) was determined by the intersection between the threshold loss tangent (tan δ  = 0.3, indicated by the green line) and the fitted tan δ - t curve. For the C5H2T1 bioink, the theoretical model predicted a pre-crosslinking time ( t p ) of 25.7 min and an endpoint of printing ( t p  +  t d ) of 78.2 min, which demonstrated excellent consistency with experimental observations ( t p  ≤ 30 min, t d  ≥ 40 min) as shown in Supplementary Fig. S3 . This agreement between theoretical predictions and experimental results validated the robustness of our analytical model. Based on comprehensive theoretical analysis and experimental validation, we have demonstrated that the C5H2T1 bioink formulation represents a selected formulation for subsequent bioprinting experiments. The recommended processing parameters include a pre-crosslinking time of approximately 30 min and a printable duration of at least 40 min. Under these conditions, the bioink enabled reliable filament extrusion and reproducible fabrication of self-supporting 3D structures with defined geometries. These optimized parameters provide a practical processing window for C5H2T1 bioink, supporting its application in extrusion-based cell-laden bioprinting of tissue engineering constructs. Using the optimized C5H2T1 bioink, we fabricated large-area 2D structures with complex curvilinear patterns ( Fig. 3 f). After printing, the structures were maintained at room temperature for approximately 1 h topromote further crosslinking. The intricate structures ( Fig. 3 f–i, ii, iii) surpass the printing capabilities using conventional collagen-based bioinks [29] , [31] , [57] , [58] , which often fail to maintain structural integrity due to insufficient mechanical strength and uncontrolled spreading. In contrast, our TG-crosslinked collagen bioink achieved improved structural stability under the optimized printing conditions. In addition, large-volume 3D structures were successfully fabricated ( Fig. 3f–iv, v , and vi), indicating improved self-supporting capability compared with the limitations commonly reported for conventional collagen bioinks [29] , [39] . These results suggest that TG-mediated pre-crosslinking can improve the printability and structural stability of collagen-based bioinks for the fabrication of complex 3D constructs. We evaluated the biological performance of EEOs cultured in Matrigel and collagen-based hydrogels (C5H2T1, C5H0T1), respectively ( Fig. 4 a and Supplementary Fig. S5a ), including EEO growth dynamics ( Fig. 4 b–f), cellular metabolic activity ( Fig. 4 g), characteristic protein expression ( Fig. 4 h–j), and hormonal response ( Supplementary Fig. S5d–g ). Cell viability was monitored by Calcein AM/propidium iodide (PI) staining on days 1, 4, and 7 ( Fig. 4 e), and good cell viability was maintained in all three hydrogel systems throughout the culture period. EEO diameters increased progressively with culture time in all three hydrogel systems ( Fig. 4 d), with the C5H0T1 group showing a broader size distribution by day 7, while EEOs in the C5H2T1 group tended to exhibit larger diameters than those in Matrigel. EEO reformation was comparable among the three groups on day 1 ( Fig. 4 f). On day 4, C5H2T1 exhibited higher EEO reformation than C5H0T1, whereas no significant difference was observed between C5H2T1 and Matrigel. By day 7, Matrigel showed higher EEO reformation than both collagen-based hydrogel groups, while no significant difference was observed between C5H0T1 and C5H2T1. Cellular metabolic activity showed no significant differences among the three groups on days 1 and 4 but increased significantly in the collagen-based hydrogel groups by day 7, with C5H2T1 exhibiting the highest value ( Fig. 4 g). Fig. 4 3D monoculture of EEOs in collagen-based hydrogels and Matrigel. ( a ) Preparation of EEOs-laden bioink and hydrogels. ( b ) Representative bright-field images captured by light microscopy. ( c ) 3D images acquired by confocal microscopy. Blue Hoechst fluorescence indicates cell nuclei. ( d ) Diameter distribution of EEOs. ( e ) Representative live/dead staining images. Green and red fluorescence represent live and dead cells, respectively. ( f ) Quantification of EEO reformation ( n  = 3 biologically independent donor-derived samples). ( g ) CCK-8 assay of EEO metabolic activity in different hydrogels ( n  = 3 biologically independent donor-derived samples). ( h ) Fluorescence staining images of characteristic proteins. Green fluorescence indicates the corresponding marker protein (FOXA2, E-cad, Ki67, or PR) and blue fluorescence represents cell nuclei. ( i ) Quantification of Ki67-positive cells in EEOs ( n  = 3 biologically independent donor-derived samples). ( j ) Quantification of PR-positive cells in EEOs ( n  = 3 biologically independent donor-derived samples). Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Tukey's multiple comparisons test for ( f , g ), and one-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Tukey's multiple comparisons test for ( i , j ). ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. 3D monoculture of EEOs in collagen-based hydrogels and Matrigel. ( a ) Preparation of EEOs-laden bioink and hydrogels. ( b ) Representative bright-field images captured by light microscopy. ( c ) 3D images acquired by confocal microscopy. Blue Hoechst fluorescence indicates cell nuclei. ( d ) Diameter distribution of EEOs. ( e ) Representative live/dead staining images. Green and red fluorescence represent live and dead cells, respectively. ( f ) Quantification of EEO reformation ( n  = 3 biologically independent donor-derived samples). ( g ) CCK-8 assay of EEO metabolic activity in different hydrogels ( n  = 3 biologically independent donor-derived samples). ( h ) Fluorescence staining images of characteristic proteins. Green fluorescence indicates the corresponding marker protein (FOXA2, E-cad, Ki67, or PR) and blue fluorescence represents cell nuclei. ( i ) Quantification of Ki67-positive cells in EEOs ( n  = 3 biologically independent donor-derived samples). ( j ) Quantification of PR-positive cells in EEOs ( n  = 3 biologically independent donor-derived samples). Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Tukey's multiple comparisons test for ( f , g ), and one-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Tukey's multiple comparisons test for ( i , j ). ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. Then, the expression of key proteins and proliferative capacity of EEOs in three hydrogel systems were evaluated. Forkhead box protein A2 (FOXA2), a transcription factor essential for endometrial gland differentiation and fertility in mice [59] , was expressed in EEOs across all matrices ( Fig. 4 h). E-cadherin (E-cad), important for structural, morphological, and functional integrity of epithelial tissue, showed well-organized staining patterns in EEOs cultured in all hydrogels, maintaining columnar morphology consistent with native endometrial architecture [7] ( Fig. 4 h). Ki67, a nuclear protein and marker of cellular proliferation, was significantly elevated in EEOs cultured in C5H2T1 hydrogel (72.80%) compared to Matrigel (54.14%) and C5H0T1 (66.15%) groups, representing a 1.34-fold increase (approximately 34% enhancement) over Matrigel ( Fig. 4 h and i). This increased proliferation indicates that C5H2T1 supported EEO expansion in 3D culture. Analysis of progesterone receptor (PR) expression, which is involved in mediating progesterone-associated reproductive functions [60] , revealed higher proportions of PR-positive epithelial cells in collagen-based hydrogels (C5H0T1: 57.39%; C5H2T1: 57.60%) than in Matrigel (41.74%) ( Fig. 4 h–j). C5H2T1 exhibited 1.37-fold higher PR expression than Matrigel, indicating increased PR expression in EEOs cultured in this hydrogel. Furthermore, EEOs in C5H2T1 displayed hormone-responsive changes consistent with the applied menstrual cycle-mimicking conditions ( Supplementary Fig. S5d–g ), supporting the maintenance of EEO functionality. Biological performance of ESCs cultured in Matrigel or collagen-based hydrogels was also evaluated. ESCs grown in the C5H2T1 hydrogel exhibited spindle-shaped cytoskeletal structures, while those in Matrigel displayed a rounded morphology after 7 days ( Supplementary Fig. S6 ). Co-culture of EEOs and ESCs exhibited enhanced secretory-phase-associated responses compared with EEO or ESC monocultures under their respective induction conditions ( Supplementary Fig. S7 ), suggesting that this integrated system can better represent the coordinated functional characteristics ( Supplementary Fig. S11 ) of the endometrium. Thus, we established a 3D co-culture system of EEOs and ESCs using C5H2T1 bioink. After 5 days of co-culture, EEOs exhibited spherical growth morphology, while a portion of ESCs localized to the surface of EEOs ( Fig. 5 c and d), indicating direct cell–cell contact and suggesting potential intercellular interactions between the two cell types ( Supplementary Fig. S11 ), while both cell types maintained favorable growth characteristics. To assess hormone responsiveness, we subjected the co-culture system to hormonal treatment and analyzed gene expression profiles ( Fig. 5 e) and secreted factors ( Supplementary Fig. S8 ). The embryo implantation “window of receptivity”, occurring during the secretory phase is characterized by endometrial receptivity induced by ovarian steroid hormones. During blastocyst positioning and adhesion, EECs and ESCs produce distinct molecular mediators that regulate uterine status through autocrine and paracrine mechanisms, including cytokines and growth factors essential for implantation. Insulin-like growth factor binding protein 1 (IGFBP1) is a well-established marker of endometrial stromal cell decidualization and regulates IGF-mediated cellular responses, including cell growth and differentiation. In our system, IGFBP1 expression in the secretory-phase group was increased by 42.22-fold compared with the control group ( Fig. 5 b–e-i). Consistently, the level of IGFBP1 secreted into the culture supernatant was approximately 2.5-fold higher in the secretory-phase group than in the control group ( Supplementary Fig. S8a ), supporting the successful induction of decidualization. Prolactin (PRL) is another classical decidualization-associated marker produced by decidualized endometrial stromal cells. In the secretory-phase group, PRL expression was upregulated by 62.67-fold compared with the control group ( Fig. 5 e–ii). In agreement with this result, the level of PRL secreted into the culture supernatant was also approximately 3-fold higher in the secretory-phase group than in the control group ( Supplementary Fig. S8b ). These findings further confirm that the system exhibited a decidualized phenotype under secretory-phase hormonal stimulation [61] . HSD17β2 expression ( Fig. 5 e–iii) progressively increased throughout the menstrual cycle, with the proliferative phase showing an 11.12-fold increase and the secretory phase reaching 22.32-fold elevation over controls, indicating regulation of estrogen activity. Progesterone receptor gene (PGR) expression ( Fig. 5 e-ⅳ) was significantly upregulated in the proliferative phase (2.13-fold increase versus control), indicating estrogen-associated induction of PGR expression. Leukemia inhibitory factor (LIF), a multifunctional glycoprotein, emerges as a potential endometrial receptivity biomarker. LIF regulates trophoblast invasiveness, promotes immune tolerance, facilitates embryo implantation, and supports stromal cell decidualization. The secretory phase group showed 1.24-fold higher LIF expression than controls ( Fig. 5 e–v), indicating enhanced endometrial receptivity and optimal preparation for embryo implantation. Fig. 5 EEO/ESC co-culture in C5H2T1 hydrogel. ( a ) Schematic illustration of native endometrial changes during the menstrual cycle. ( b ) In vitro hormonal treatment strategy used to mimic the menstrual cycle. ( c ) Representative bright-field images showing the growth status of EEO/ESC co-cultures. ( d ) Representative confocal fluorescence images showing the growth status of EEO/ESC co-cultures. PanCK indicates epithelial cells, vimentin indicates ESCs, and Hoechst indicates cell nuclei. ( e ) Gene expression profiles after hormonal treatment ( n  = 3 biologically independent donor-derived samples). ( f–i) Representative confocal fluorescence images showing the expression of Ki67, ER, MUC1, and PR after hormonal treatment. ( j–m) Quantification of Ki67, ER, MUC1, and PR expression after hormonal treatment ( n  = 5 biologically independent donor-derived samples). Data are presented as mean ± SD. Statistical significance was assessed using one-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Dunnett's multiple comparisons test. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. EEO/ESC co-culture in C5H2T1 hydrogel. ( a ) Schematic illustration of native endometrial changes during the menstrual cycle. ( b ) In vitro hormonal treatment strategy used to mimic the menstrual cycle. ( c ) Representative bright-field images showing the growth status of EEO/ESC co-cultures. ( d ) Representative confocal fluorescence images showing the growth status of EEO/ESC co-cultures. PanCK indicates epithelial cells, vimentin indicates ESCs, and Hoechst indicates cell nuclei. ( e ) Gene expression profiles after hormonal treatment ( n  = 3 biologically independent donor-derived samples). ( f–i) Representative confocal fluorescence images showing the expression of Ki67, ER, MUC1, and PR after hormonal treatment. ( j–m) Quantification of Ki67, ER, MUC1, and PR expression after hormonal treatment ( n  = 5 biologically independent donor-derived samples). Data are presented as mean ± SD. Statistical significance was assessed using one-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Dunnett's multiple comparisons test. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. To further characterize cellular dynamics, we performed immunofluorescence (IF) staining for Ki67, MUC1, PR, and ER proteins at distinct time points. Ki67 expression in EEOs ( Fig. 5 f) exhibited a biphasic pattern, initially increasing and subsequently decreasing from the proliferative phase to the secretory phase. Quantitative analysis ( Fig. 5 j) revealed that Ki67-positive cells peaked at 80.35% in the proliferative phase, significantly higher than control (55.17%), but declined to 9.26% in the secretory phase, indicating maximal proliferative activity during the proliferative phase and its attenuation upon decidualization. ER expression ( Fig. 5 g) remained comparable between the control (61.96%) and proliferative phase (69.43%) groups, but markedly decreased in the secretory phase (10.63%) ( Fig. 5 k). This downregulation following progesterone stimulation facilitates the opening of the embryo implantation “window period”. MUC1 fluorescence intensity ( Fig. 5 h) progressively increased from proliferative to secretory phases, with mean intensities of 28.33 (control), 39.65 (proliferative), and 47.23 (secretory) ( Fig. 5 l), mirroring physiological transmembrane protein upregulation during the human menstrual cycle's secretory phase. PR ( Fig. 5 i) showed a distinct pattern, increasing in the proliferative phase but decreasing in the secretory phase. Statistical analysis ( Fig. 5 m) revealed PR expression levels of 22.49% (control), 67.53% (proliferative), and 49.92% (secretory), consistent with gene expression data ( Fig. 5 e–iv) and confirming estrogen-mediated upregulation during the proliferative phase. ESCs primarily reside within the endometrial lamina propria, whereas EECs are primarily localized in the superficial endometrial layer, with a subset extending into deeper regions to establish contact with stromal cells [2] . As illustrated in Fig. 6 a, we implemented a dual-nozzle extrusion-based bioprinting strategy to construct a bilayered cellular architecture, comprising EEOs in the upper layer and ESCs/EEOs in the lower layer, thereby recapitulating the native endometrial structure. As shown in Fig. 6 b and Supplementary Fig. S9 , live and dead EEOs or ESCs within constructs printed using different needle sizes were stained with green-fluorescent Calcein AM and red-fluorescent PI, respectively. EEOs or ESCs printed with the 300 μm needle exhibited stronger red fluorescence, indicating increased printing-induced cell death. In contrast, the 500 μm and 700 μm groups showed abundant green fluorescence with minimal red fluorescence, indicating lower cell death rates. However, the 700 μm needle produced microfilaments with lower uniformity and structural fidelity than the 500 μm needle. Through systematic optimization, the 500 μm-diameter nozzle was used as the optimal configuration for both EEOs-laden and ESCs-laden bioprinting, achieving an optimal balance between printing precision and cell viability ( Fig. 6 b and c and Supplementary Fig. S9 ). The 3D constructs containing EEOs or ESCs printed using a 500 μm needle were stained for confocal imaging. EEO nuclei were stained with DAPI (blue), while ESC nuclei and cytoskeleton were labeled with DAPI (blue) and F-actin (green), respectively ( Fig. 6 c). Fig. 6 Cell-laden bioprinting of C5H2T1 bioink. ( a ) Schematic illustration of the cell-laden extrusion bioprinting process for constructing the endometrial model. ( b ) Growth of EEOs or ESCs in the printed bioink after extrusion using a needle with an inner diameter of 500 μm. ( c ) 3D fluorescence images of EEOs and ESCs acquired by confocal microscopy. ( d ) Gene expression profiles of EEOs/ESCs-laden scaffolds with different macroporosities ( n  = 3 biologically independent donor-derived samples). ( e ) Design of EEO/ESCs-laden scaffolds with different macroporosities. f ) Cell-laden collagen-printed model. ( g ) Top and side views of the printed structure. ( h ) Cell-embedded Matrigel-cast model. ( i ) Top and side views of the cast structure. Data are presented as mean ± SD. Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Dunnett's multiple comparisons test. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. Cell-laden bioprinting of C5H2T1 bioink. ( a ) Schematic illustration of the cell-laden extrusion bioprinting process for constructing the endometrial model. ( b ) Growth of EEOs or ESCs in the printed bioink after extrusion using a needle with an inner diameter of 500 μm. ( c ) 3D fluorescence images of EEOs and ESCs acquired by confocal microscopy. ( d ) Gene expression profiles of EEOs/ESCs-laden scaffolds with different macroporosities ( n  = 3 biologically independent donor-derived samples). ( e ) Design of EEO/ESCs-laden scaffolds with different macroporosities. f ) Cell-laden collagen-printed model. ( g ) Top and side views of the printed structure. ( h ) Cell-embedded Matrigel-cast model. ( i ) Top and side views of the cast structure. Data are presented as mean ± SD. Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser–Greenhouse correction followed by Dunnett's multiple comparisons test. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. To elucidate the influence of EEOs/ESCs-scaffold architecture on cellular functionality, we investigated three distinct printing grid porosities (21.5%, 57.7%, 75.8%) ( Fig. 6 d and e). This comprehensive analysis encompassed the evaluation of stemness-related gene expression (SOX9, SSEA1, ALDH1A1, PROM1) and PGR expression, with a non-printed cast mold group serving as the control ( Fig. 6 e). Notably, scaffolds with 21.5% porosity exhibited significantly reduced expression levels of stemness-related genes, ALDH1A1 and PROM1, which we attributed to shear-induced cellular damage during the printing process. In contrast, scaffolds with higher porosities (57.7% and 75.8%) demonstrated enhanced gene expression levels compared to the cast-molded control group. We designed a collagen-printed in vitro endometrial co-culture model, as shown in Fig. 6 f, with EEOs in the upper layer and EEOs/ESCs in the lower layer. Since Matrigel is difficult to print, a Matrigel-cast model ( Fig. 6 h) was used as the control. Before printing, EEOs were labeled with DiO (green), and ESCs were labeled with DiI (red). In the printed 3D constructs, the two cell types were distributed according to the designed architecture ( Fig. 6 g). In contrast, cells in the Matrigel casting control were randomly distributed without clear spatial orientation ( Fig. 6 i). For the endometrial model designed above, we implemented a hormonal treatment protocol consistent with that used in the EEOs/ESCs co-culture system. As depicted in Fig. 7 a (where the cell-laden collagen-printed group is denoted as “collagen” and the cell-embedded Matrigel-cast group as “Matrigel”), IGFBP1 exhibited significantly lower expression in Matrigel group compared to the collagen group under identical hormonal treatment conditions ( Fig. 7 a-ⅰ). This observation suggests that the 3D bioprinted collagen scaffold may provide a more favorable environment for decidualization-associated gene expression. SOX9, a marker associated with cellular stemness, reflects the potential for self-renewal and differentiation. As shown in Fig. 7 a-ⅱ, cells within the collagen group exhibited higher SOX9 expression than those in the Matrigel group, suggesting better maintenance of stem-like characteristics. In the control group, PGR expression was notably higher in the collagen-printed structures than in the Matrigel-cast structures ( Fig. 7 a-ⅲ), suggesting that the collagen-printed scaffolds may support higher PGR expression under these conditions. However, no significant differences in PGR expression were observed between the collagen-printed and Matrigel-cast structures in either the proliferative or secretory phase groups. PAEP, which encodes essentials for creating a uterine environment conducive to pregnancy, also exhibited significantly higher gene expression in the collagen group compared to the Matrigel group ( Fig. 7 a-ⅳ). The expression of the HSD17β2, a gene involved in the regulation of estrogen and progesterone metabolism, was also assessed ( Fig. 7 a-ⅴ). Higher HSD17β2 expression may reflect altered steroid hormone metabolic activity. Fig. 7 Comparison of the collagen-printed model using C5H2T1 bioink and the Matrigel-cast model. ( a ) Gene expression profiles after hormonal treatment ( n  = 3 biologically independent donor-derived samples). Data are presented as mean ± SD. Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser–Greenhouse correction, followed by Dunnett's multiple comparisons test for comparisons with control within each group and Šídák's multiple comparisons test for comparisons between Matrigel and collagen under the same treatment condition. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. ( b ) UMAP plot showing the distribution of cell subpopulations. ( c ) Bubble plot showing gene expression patterns across cell subpopulations. Bubble size represents the proportion of cells expressing a given gene within each cluster, while the color gradient from blue to red indicates the average expression level from low to high. ( d ) RNA velocity analysis. ( e ) UMAP plot comparing the collagen-printed and Matrigel-cast groups. ( f ) UMAP visualization of DPP4 and KLF5 expression. ( g ) GO enrichment analysis of significantly upregulated genes in SOX9 + _Epi from the collagen-printed group compared with the Matrigel-cast group. ( h ) UMAP visualization of MSLN and WNT7A expression. i ) GO enrichment analysis of significantly downregulated genes in SOX9 + _Epi from the collagen-printed group compared with the Matrigel-cast group. Comparison of the collagen-printed model using C5H2T1 bioink and the Matrigel-cast model. ( a ) Gene expression profiles after hormonal treatment ( n  = 3 biologically independent donor-derived samples). Data are presented as mean ± SD. Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser–Greenhouse correction, followed by Dunnett's multiple comparisons test for comparisons with control within each group and Šídák's multiple comparisons test for comparisons between Matrigel and collagen under the same treatment condition. ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001. ( b ) UMAP plot showing the distribution of cell subpopulations. ( c ) Bubble plot showing gene expression patterns across cell subpopulations. Bubble size represents the proportion of cells expressing a given gene within each cluster, while the color gradient from blue to red indicates the average expression level from low to high. ( d ) RNA velocity analysis. ( e ) UMAP plot comparing the collagen-printed and Matrigel-cast groups. ( f ) UMAP visualization of DPP4 and KLF5 expression. ( g ) GO enrichment analysis of significantly upregulated genes in SOX9 + _Epi from the collagen-printed group compared with the Matrigel-cast group. ( h ) UMAP visualization of MSLN and WNT7A expression. i ) GO enrichment analysis of significantly downregulated genes in SOX9 + _Epi from the collagen-printed group compared with the Matrigel-cast group. In this study, we leveraged 10× Genomics single-cell transcriptome sequencing technology to systematically evaluate the influence of collagen-printed structures and Matrigel-cast structures on the transcriptional profiles of individual cells or specific cell clusters. Utilizing the uniform manifold approximation and projection (UMAP) dimensionality reduction algorithm, we projected high-dimensional transcriptomic data into a two-dimensional space, revealing 17 distinct cell clusters with significant transcriptional heterogeneity ( Supplementary Fig. S10 ). The observed variations in gene expression patterns across these clusters underscored the high quality of the samples and the effectiveness of clustering, providing a solid foundation for subsequent cell type annotation. Initial classification based on classical epithelial cell markers (EPCAM) and stromal cell markers (MME, COL6A2) segregated the cells into two broad categories: epithelial cells and stromal cells. Further refinement of epithelial cells was achieved using stem cell markers (SOX9) [62] , glandular epithelial markers (FOXA2, MUC16), and luminal epithelial markers (WNT7A) [63] , resulting in the identification of SOX9 + epithelial cells (SOX9 + _Epi), glandular epithelial cells (GE), and luminal epithelial cells (LE). Similarly, stromal cells were subdivided into Str1 and Str2 subpopulations according to their cell-cycle-associated transcriptional profiles, with MKI67 and TOP2A [64] enriched in Str2. Str2 exhibited stronger proliferative signatures, whereas Str1 showed relatively lower cell-cycle activity. The defined subpopulations exhibited distinct transcriptional characteristics, supporting the reliability of our classification approach. The five final annotated cell subpopulations formed distinct clusters in the UMAP embedding ( Fig. 7 b). Comparative UMAP visualization of the collagen and Matrigel groups ( Fig. 7 e) showed that the major annotated cell populations were represented in both conditions with broadly similar transcriptional distributions. RNA velocity analysis was further used to explore potential transcriptional state-transition tendencies among epithelial subpopulations ( Fig. 7 d). The velocity streamlines suggested potential state-transition tendencies among SOX9 + _Epi cells, GE, and LE populations. These patterns were interpreted as exploratory transcriptional dynamics rather than evidence of definitive lineage relationships. Comparative analysis between the collagen group and the Matrigel group revealed specific expression patterns of key genes across cell populations. For instance, DPP4 was highly expressed in GE, KLF5 showed elevated expression in stromal cells, and MSLN and WNT7A were preferentially expressed in LE ( Fig. 7 f and h). Further analysis through Gene Ontology (GO) enrichment of SOX9 + _Epi populations revealed distinct functional differences between the collagen-printed and Matrigel-cast groups. In the collagen group, significantly upregulated genes were primarily enriched in biological processes such as “Response to chemokine”, “Positive regulation of ATP-dependent activity”, “Positive regulation of cell cycle”, “Cellular homeostasis”, and “Cell division” ( Fig. 7 g). The enrichment in chemokine response suggests potential involvement of SOX9 + _Epi cells in chemokine-responsive processes associated with cell migration and tissue repair in the collagen group [65] . The upregulation of ATP-dependent processes and cell cycle-related genes further further suggests an active proliferative state of SOX9 + _Epi cells in the collagen group, consistent with our earlier experimental observations ( Fig. 7 a). This observation was further supported by IF staining ( Fig. 4 h and i), which showed higher levels of the proliferation marker Ki67 in EEOs cultured in the C5H2T1 hydrogel than in those cultured in Matrigel, further supporting the increased proliferative activity observed in the collagen group. Conversely, genes significantly downregulated in the collagen group, relative to the Matrigel group, were predominantly enriched in processes such as “Aging”, “Negative regulation of defense response,” and “Inflammatory response” ( Fig. 7 i). This downregulation suggests that collagen-based system may be associated with reduced inflammatory- and defense response-related transcriptional activity, potentially providing a more favorable microenvironment for cellular function and tissue regeneration than Matrigel. In GE, the collagen group showed greater enrichment of biological processes related to hormone secretion and angiogenesis than the Matrigel group ( Supplementary Fig. S12a–b ), suggesting potential differences in physiological function. For LE, the collagen-printed construct was associated with reduced senescence- and cytotoxicity-related processes and altered energy metabolism ( Supplementary Fig. S12c–d ), potentially providing a cellular microenvironment favorable for long-term cell culture and tissue engineering applications. Furthermore, stromal cells in the collagen group exhibited greater enrichment of proliferation- and decidualization-related processes [66] , [67] ( Supplementary Fig. S12e–h ), suggesting differences in their functional states compared with the Matrigel group. To provide additional context for interpreting the differences between the collagen-printed and Matrigel-cast models, we also analyzed a collagen-cast EEO/ESC bilayer model, enabling a three-group comparison among the Matrigel-cast, collagen-cast, and collagen-printed conditions ( Supplementary Fig. S13 ). The collagen-cast group generally exhibited an intermediate transcriptional profile across several epithelial- and stromal-associated markers. These findings suggest that matrix composition may account for part of the observed differences, while additional changes were associated with the collagen-printed configuration.

The in vitro reconstruction of the human endometrium remains challenging because of its diverse cellular composition, complex tissue architecture [2] , and dynamic hormone-responsive behavior [1] . Matrigel has been widely used as the gold-standard matrix for endometrial cell culture. However, its undefined composition, tumor-derived origin, and inadequate mechanical properties for fabricating complex 3D structures limit its translational potential and future clinical applications [11] . Collagen, a major component of the native endometrial extracellular matrix [10] , possesses several biologically favorable properties, including cell-adhesive motifs such as RGD sequences [20] , a characteristic triple-helical structure that contributes to matrix organization and structural integrity [15] , and MMP-mediated degradability [23] . These features make collagen a promising bioactive matrix for constructing physiologically relevant in vitro endometrial models. Nevertheless, previous attempts to apply collagen-based matrices to endometrial model construction have shown several limitations. For example, Muruganandan et al. [24] established a 3D culture system for endometrial tissue slices within collagen gels using a dual-dish culture method. Although this model successfully exhibited functional responses to estrogen and progesterone stimulation, the limited availability of human endometrial tissue slices, together with ethical constraints, restricts its scalability and broader application. In the model reported by Bentin-Ley et al. [25] , [26] , ESCs were cultured within collagen gels to investigate paracrine interactions between ESCs and EECs; however, direct contact between ESCs and EECs was not recapitulated, and the functional characteristics of the model were not extensively explored. In addition, although the models developed by Shibata et al. [27] or Tian et al. [7] supported the long-term functional culture of EEOs, the matrices used in these studies were collagen-Matrigel mixtures, and hydrogel contraction occurred over time. In contrast, this study used EEOs instead of tissue slices and established a collagen-based hydrogel system that supported the long-term co-culture of EEOs and ESCs in a native tissue-like spatial organization. EEOs and ESCs could be maintained within the contraction-resistant collagen-based hydrogel for an extended period (≥11 days). In addition, EEOs cultured in the collagen-based hydrogel showed a reformation rate comparable to that in Matrigel at day 4, together with enhanced cellular metabolic activity and protein expression levels at day 7 ( Fig. 4 ). ESCs also maintained a normal spindle-like spreading morphology within the collagen matrix ( Supplementary Fig. S6 ). These findings suggest that the collagen-based hydrogel developed in this study provides a supportive microenvironment for EEOs and ESCs and may serve as a promising alternative to Matrigel for constructing in vitro endometrial models. The endometrium consists of the functional layer and the basal layer, with the functional layer containing abundant glandular tubular structures [2] and exhibiting a highly complex tissue architecture [68] . Therefore, 3D bioprinting provides a promising strategy to reproduce such structural complexity in vitro. However, current collagen-based bioinks still face important limitations, particularly the difficulty in balancing biological compatibility with printability [20] , [21] , [29] , [31] , [32] , [39] . In this study, we sought to address this challenge by optimizing both the crosslinking strategy and the composition of the collagen-based bioink. Specifically, TG and HA were incorporated as auxiliary components. TG was used to promote enzymatic crosslinking and improve structural stability [69] , whereas HA was introduced because of its relevance to the dynamic remodeling of the endometrial extracellular matrix [48] . Since the physical properties of hydrogels define the 3D microenvironment for embedded cells, parameters such as mechanical properties, swelling behavior, and molecular transport capacity can directly influence cell growth and function. Therefore, using Matrigel as a reference, we systematically evaluated the physicochemical properties of hydrogels with different formulations. Compression testing ( Fig. 2 i and j) showed that, at 37 °C, the mechanical properties of the HA-containing hydrogel (C5H2T1) were not significantly different from those of Matrigel but were lower than those of the HA-free formulation (C5H0T1). Given the close relationship between hydrogel swelling and molecular exchange, we further assessed the swelling behavior of the hydrogels ( Fig. 2 k) together with FITC-dextran penetration under hydrated conditions ( Supplementary Fig. S1 ). The C5H2T1 hydrogel showed a higher swelling ratio and faster FITC-dextran penetration than C5H0T1, suggesting that C5H2T1 formed a more hydrated and permeable network that may provide a more favorable microenvironment for long-term cell growth within the hydrogel. This finding suggests that HA may be one of the factors contributing to the reduced compressive modulus, enhanced swelling capacity, and greater penetration observed in the C5H2T1 hydrogel. However, the underlying mechanisms and the specific contribution of HA to these changes remain to be clarified and will require more systematic investigation in future studies. TG concentration was also closely associated with the crosslinking density and network structure of the hydrogels, with a higher TG concentration leading to a more densely crosslinked network. Accordingly, compared with C5H2T1, the C5H2T2 hydrogel exhibited a higher compressive modulus, a lower swelling ratio, and slower FITC-dextran penetration. Taken together with the biological results from EEO and ESC monoculture experiments in C5H2T1, C5H0T1, and Matrigel, these physicochemical findings supported the selection of C5H2T1 as the optimized bioink formulation for subsequent cell-laden bioprinting and endometrial model construction. Whether collagen can directly self-assemble into a hydrogel [36] through changes in temperature, pH, and ionic strength is closely related to its processing method, purity, terminal telopeptide structure, concentration, α1/α2 chain composition, and the formulation environment [70] . In addition, high-concentration collagen hydrogels formed solely through self-assembly, often possess a highly dense 3D network [38] , which may not be suitable for the long-term 3D culture of embedded cells. The collagen used in this study exhibited reversible thermoresponsive behavior ( Fig. 3 a–c), with gelation occurring at lower temperature and liquefaction upon heating. Based on this property, we designed the printing process shown in Fig. 3 d. Briefly, the TG-containing collagen bioink was first prepared at 37 °C to ensure compositional homogeneity and maintain a flowable liquid state, thereby facilitating uniform cell mixing. TG-mediated crosslinking may already begin during the short mixing and loading period. However, because this handling interval was brief and strictly controlled across all experiments, its contribution to the overall crosslinking process was considered minor and was therefore neglected as an approximation in the subsequent analysis. Then, the bioink was pre-crosslinked at room temperature for a defined period. During this stage, the time-dependent rheological evolution of the bioink may arise from the combined contributions of physical interactions, such as hydrogen bonding, van der Waals forces, and chain entanglement, and TG-catalyzed chemical crosslinking, allowing the bioink to progressively reach an extrudable and filament-forming state ( Supplementary Fig. S3 ). The bioink was then printed within the printable time window ( Fig. 3 e). To quantitatively describe the time-dependent evolution of the storage modulus during this process, a hypothetical two-step consecutive first-order kinetic model (Eqs. (1) , (2) , (3) )) was established based on the observed rheological behavior. This model enabled relatively accurate prediction of the pre-crosslinking time and printable duration ( Fig. 3 e and Supplementary Fig. S3c ), thereby providing quantitative guidance for the printing of the collagen-based bioink. However, the actual mechanistic relationship between the two steps represented in the kinetic model and the respective contributions of physical interactions and TG-mediated chemical crosslinking remains unclear and warrants further investigation. After printing, the constructs were further crosslinked at 22 °C for approximately 1 h to promote further network formation and stabilize the constructs prior to subsequent culture at 37 °C. The resulting 3D cell-laden constructs supported favorable cell growth ( Fig. 6 b–d, Supplementary Fig. S9 ), showed no obvious contraction, and maintained structural stability during long-term culture. These results indicate that the kinetic model-guided printing strategy developed in this study effectively balanced the mechanical requirements of extrusion-based bioprinting with the need to maintain a cytocompatible 3D microenvironment within a collagen-dominant hydrogel matrix. In this study, our primary objective was to develop a collagen-based 3D-bioprinted in vitro endometrial model [71] and to evaluate its suitability for potential future studies relative to the commonly used Matrigel-based casting models reported in previous literature. Accordingly, the biological evaluation focused primarily on comparing the collagen-printed and Matrigel-cast models as two complete model systems. This comparison was intended to assess their overall biological performance rather than to isolate the independent effects of matrix composition or fabrication method. To provide additional context, we also performed a preliminary comparison involving the collagen-cast EEO/ESC bilayer model, together with the Matrigel-cast and collagen-printed models, and the corresponding analyses are presented in the Supplementary Information ( Supplementary Fig. S13 ). These additional comparisons provide some information regarding differences under matched casting conditions and within the same collagen-based formulation. And more detailed analyses will be conducted in future studies if necessary. Several limitations of the present study should also be acknowledged. In particular, the collagen-printed model was not directly compared with native endometrial tissue derived from the same donors. Therefore, it remains to be further determined whether the cellular states, tissue organization, and hormone-responsive behavior of the model recapitulate those of native endometrium. Future studies should incorporate native endometrial tissue or endometrium-derived extracellular matrix as references to further evaluate the physiological relevance of the printed model. Despite these limitations, the collagen-based bioink and bioprinting strategy developed in this study show promising potential for future applications. This platform may be useful for in vitro endometrial modeling, drug screening, patient-derived disease modeling, and studies of cell–cell and cell–matrix interactions. Future work will focus on incorporating additional endometrial cell types, fabricating more complex 3D structures, and investigating interactions with embryos or trophoblast spheroids. This model may also be extended to endometrial pathologies, such as endometritis, endometrial polyps, and adenomyosis, supporting future applications in personalized disease modeling, therapeutic screening, and endometrial tissue engineering.

As illustrated in Fig. 2 a, type I collagen sponges derived from porcine skin tissue (100-300 kDa, Chengdu Kele, China) were sterilized by immersion in 75% (v/v) ethanol (Aladdin, China) for 30 min, followed by air-drying for 24 h on super clean bench. The sterilized collagen solid was then mixed with 0.5% (v/v) acetic acid (HAc, Aladdin, China) and stored at 4 °C for 3 days to facilitate dissolution. Subsequently, the mixture was stirred at 37 °C for 4, 8, or 12 h to achieve homogeneous dispersion and complete dissolution of the collagen solid. The pH of the resulting mixture was carefully adjusted to 7.2–7.4 using 1 M NaOH (Sinopharm, China) to form a concentrated stock collagen suspension suitable for cell culture. This stock suspension, with a concentration of 10 wt%, was stored at 4 °C. Notably, the stock collagen suspension exhibited gelation over time at room temperature; thus, during subsequent bioink preparation, the stock collagen gel was maintained at 37 °C to preserve its suspension state for further processing. The suspension remained flowable at 37 °C but underwent macroscopic gelation over time at room temperature, exhibiting temperature-dependent behavior and rheological trends similar to those shown in Fig. 3 a and c. During the preparation of sterile acidic collagen suspensions from collagen solids, stirring and dispersion at 37 °C constituted a critical step. The duration of stirring affected the particle size of the suspension, which in turn may influenced the enzymatic crosslinking time and the mechanical strength of the resulting hydrogel. Furthermore, prolonged heating and stirring under acidic conditions may affect thermal stability of collagen [72] . To evaluate these effects, collagen was dissolved at 37 °C for varying durations (4, 8, or 12 h), and the pH of the resulting suspensions was adjusted to 7.2–7.4. These treated collagen suspensions were then compared with a collagen solution obtained from the same manufacturer (∼1.5 wt%; Chengdu Kele, China) that had not undergone either the 75% (v/v) ethanol treatment or the subsequent 37 °C heating and stirring process. For comparison, the pH of this solution was adjusted to 7.2–7.4 using NaOH. The collagen solids used in this study were produced by the manufacturer through lyophilization of this collagen solution. The molecular properties of each group under diluted conditions were characterized using CD spectroscopy and SDS-PAGE, while particle size distributions were analyzed by DLS. CD spectroscopy measurements were conducted using a J-1500-150STCD Spectrometer (JASCO, Japan) to assess structural features associated with the collagen triple helix. The study included untreated collagen in solution state (0.02 wt%) and collagen samples that were stirred and dissolved in 0.5% (v/v) HAc for 4, 8, or 12 h, followed by neutralization. The sample solutions were loaded into quartz cells with a 0.1 cm path length. CD spectra were acquired over a wavelength range of 190 to 260 nm at a scan of 1 nm/s. To ensure accurate analysis, the spectrum of sodium acetate (NaAc) buffer (pH 7.2–7.4) was subtracted from all sample spectra to eliminate background interference. This approach enabled the characterization of collagen secondary structural features under different processing conditions. SDS-PAGE was carried out using a Mini-PROTEAN Tetra electrophoresis tank (Bio-Rad, USA). Samples of untreated collagen (0.25 wt%) and collagen dissolved in 0.5% (v/v) HAc under stirring for 4, 8, or 12 h and subsequently neutralized were centrifuged at 12,000 × g for 10 min to remove insoluble components. The supernatant of each sample was mixed with 5 × sample loading buffer (Beyotime, China) at a sample-to-buffer volume ratio of 4:1. The mixture was then heated at 100 °C for 10 min to denature the proteins. Subsequently, 5 μL of each denatured sample was loaded onto a pre-cast gel (Beyotime, China) for electrophoretic separation at 140 V for 1 h. After electrophoresis, the gel was rinsed three times with deionized water to remove residual buffer. Protein bands were visualized by staining the gel with a quick staining solution (Beyotime, China) for 30 min, followed by destaining in deionized water for 16 h to enhance band clarity. The molecular weights of the protein bands were estimated by comparison with a protein molecular weight marker (Beyotime, China), allowing evaluation of collagen band patterns and potential degradation under different processing conditions. To investigate the influence of pretreatment conditions on the particle size distribution of collagen suspensions, we prepared four distinct material groups: untreated collagen (0.05 wt%) and collagen samples subjected to stirring and dissolution in 0.5% (v/v) HAc for 4, 8, or 12 h, followed by neutralization. The particle size distributions of these samples were analyzed over a measurement range of 10 nm to 10 μm using a zeta potential analyzer (Malvern Zetasizer Nano ZS90, UK) at room temperature. For each group, we recorded the particle size distribution by quantity and calculated the mean particle size to facilitate comparative analysis. To further characterize the thermosensitive behavior of collagen suspensions under different pretreatment conditions, we performed rheological measurements using a rotational rheometer (MCR102, Anton Paar, Austria) fitted with a 25 mm parallel plate geometry. The measurement gap was maintained at 1 mm throughout the experiments, with an oscillatory frequency of 2 Hz. The temperature-dependent viscoelastic properties of 5 wt% collagen suspensions (pH 7.2–7.4) were evaluated by recording G′ and G″ during a controlled temperature ramp from 37 °C to 4 °C at a cooling rate of −1 °C/min. HA (10 kDa, Rhawn, China) was dissolved in PBS (Solarbio, China) at room temperature and subsequently sterilized by filtration through a 0.22-μm membrane filter (Merck Millipore, Germany). Similarly, a commercial TG preparation (100 U/g, lot No. 240315895; Shandong Xitang Biotechnology, China) was dissolved in PBS at 37 °C and sterilized using the same filtration procedure. Based on the target concentrations of collagen, HA, and TG in the final bioink formulation, required volumes of each component were combined and thoroughly mixed to yield a homogeneous solution at 37 °C. For cell-laden bioink preparation, cell pellets were gently resuspended in the pre-mixed bioink solution (containing collagen, HA, and TG) and pipetted up and down to ensure uniform cell distribution prior to subsequent processing. The resulting cell-laden bioink was then pre-crosslinked at 22 °C for a defined period ( t p ) before extrusion bioprinting ( Fig. 3 d). In this study, we adopted a standardized nomenclature for bioink formulations, where CxHyTz denotes the weight percentages (wt%) of collagen (x), HA (y), and TG (z), respectively ( Table 1 ). For instance, the designation C5H2T1 corresponds to a bioink composition containing 5 wt% collagen, 2 wt% HA, and 1 wt% TG. Table 1 The composition of bioink groups. Group Collagen (wt%) HA (wt%) TG (wt%) C5H0T0 5 0 0 C5H0T1 5 0 1 C5H2T0 5 2 0 C5H2T1 5 2 1 C5H2T2 5 2 2 The composition of bioink groups. The mechanical characterization of crosslinked bioinks was performed using a dynamic mechanical analysis instrument (ElectroForce, TA Instruments, USA) at 22 °C and 37 °C. For the experimental groups, bioink formulations (C5H0T1, C5H2T1, and C5H2T2) were crosslinked at 22 °C for 1 h to form disc-shaped hydrogels with standardized dimensions (8 mm in diameter × 4 mm in thickness). As a control, The Matrigel group, consisting of 70% (v/v) Matrigel and 30% (v/v) PBS, was crosslinked at 37 °C for 30 min. Following crosslinking, all samples were immersed in PBS at 37 °C for 24 h. Mechanical testing was conducted by applying uniaxial compression at a constant displacement rate of 1 mm/min. The compressive modulus was determined from the slope of the stress-strain curve with the strain range of 0–10%. The swelling behavior of four hydrogels (C5H0T1, C5H2T1, C5H2T2 and Matrigel) was systematically evaluated at 37 °C. First, the samples were immersed in PBS at 37 °C for 24 h to reach swelling equilibrium. Then, the samples were removed from PBS, and excess surface moisture was gently blotted away. The mass of each swollen sample was recorded as the swollen weight ( W s ). Subsequently, the samples were lyophilized, and the resulting mass was recorded as the dry weight ( W d ). The equilibrium swelling ratio ( S r ) was calculated as follows: (6) S r = W s − W d W d To qualitatively investigate molecular transport within different hydrogels (C5H0T1, C5H2T1, and C5H2T2) under hydrated conditions, a FITC-dextran penetration assay was performed using FITC-dextran (70 kDa; Maokangbio, China) as a fluorescent tracer. First, the bioink formulations were crosslinked at 22 °C for 1 h to form disc-shaped hydrogels with standardized dimensions (10 mm in diameter × 1 mm in thickness). Second, all hydrogel samples were immersed in PBS at 37 °C for 24 h. Third, PBS was removed and a 0.025 wt% FITC-dextran solution was applied exclusively to the top surface of each hydrogel, while the lateral and bottom surfaces were not exposed to the FITC-dextran solution, thereby establishing a top-down penetration condition. The samples were maintained at 37 °C during the assay. Three-dimensional fluorescence images were acquired at 0.5, 1, 2, 12, and 24 h to visualize the penetration of FITC-dextran into the hydrogels. To comprehensively evaluate the degradation profiles of the hydrogels (C5H0T1, C5H2T1, and C5H2T2), in vitro enzymatic degradation assay was conducted. Disc-shaped hydrogel samples (8 mm in diameter × 2 mm in thickness) were incubated in 3 mL of sterile PBS containing 2.5 U/mL type I collagenase (Sigma-Aldrich, USA) at 37 °C under static conditions [22] , [73] . The collagenase solution was not replaced during the degradation experiment. At predetermined time points (0, 2, 4, 8, 12, and 24 h), the samples were carefully retrieved, gently blotted to remove excess surface moisture, and subsequently lyophilized. The dry weight of the sample collected at 0 h was recorded as the initial dry weight ( W o ), while the dry weights of the samples collected at subsequent time points were recorded as the remaining dry weight ( W r ). The residual mass ratio ( R s ) was calculated as follows: (7) R s = W r W o × 100 % Because C5H2T1 exhibited relatively favorable overall performance in the in vitro evaluations, it was selected as the optimized formulation for subsequent assessment. Accordingly, the in vivo degradation assay was performed only for C5H2T1 to preliminarily evaluate the degradation behavior of the selected formulation under physiological conditions. For the in vivo degradation assessment, hydrogel samples were subcutaneously implanted in Sprague-Dawley rats following standard surgical procedures. Female Sprague-Dawley rats, aged 8 weeks and weighing 200-230 g, were used in this study. Prior to implantation, disc-shaped hydrogel samples (8 mm in diameter × 2 mm in thickness) were prepared, and sterilized by UV irradiation. Before surgery, the rats were anesthetized with 1.25% tribromoethanol (25 mL/kg; Meilunbio, China). The dorsal hair was shaved, and the surgical area was disinfected with iodophor (Aladdin, China). Four small skin incisions were made bilaterally along the paravertebral region, and subcutaneous pockets were carefully created by blunt dissection. Four hydrogel discs were implanted into each rat, with two samples placed on each side of the paravertebral region. The incisions were then closed with sutures, and the animals were monitored until recovery from anesthesia. Postoperative care was provided in accordance with institutional animal care guidelines. At predetermined time points (days 0, 3, 8, 10, and 13 after implantation), hydrogel samples were collected from two rats per time point, with four implants retrieved from each animal. The four implants within each rat were considered subsamples rather than independent biological replicates. At each retrieval time point, the rats were euthanized by carbon dioxide (CO 2 ) asphyxiation, and the implanted hydrogels were carefully excised together with the surrounding tissue. The explanted samples were gently separated from adherent tissue, rinsed with PBS to remove residual blood and loosely attached tissue. Lyophilized samples collected on day 0 were weighed to determine the initial dry weight ( W o ). Samples collected at the subsequent time points were also lyophilized and weighed to determine the remaining dry mass ( W r ). The residual mass ratio was also calculated using Eq. (7) . Enzymatic crosslinking was found to markedly affect the rheological properties of the collagen-based bioinks, thereby influencing their printability. To systematically investigate this effect, a bioink formulation containing 5 wt% collagen and 2 wt% HA was selected based on considerations of biological relevance and material performance. TG concentrations of 0 wt%, 2 wt%, 4 wt%, 6 wt%, and 8 wt% were used to evaluate changes in the rheological properties of the bioink at different TG concentrations. Rheological characterization was performed using a rotational rheometer equipped with a 25 mm parallel-plate geometry, with a constant measurement gap of 1 mm. The crosslinking kinetics were monitored by time-dependent oscillatory measurements under controlled conditions (0.1% strain and 1 Hz frequency) over a 2 h period. The evolution of the rheological moduli was recorded as a function of crosslinking time. Endometrial tissue biopsies were collected from donors following a standardized protocol. All participants provided written informed consent for biological sample collection and voluntarily donated endometrial tissues for scientific research purposes. Donors had normal ovarian reserve and physiological hormone levels. Individuals who had received hormone therapy within the previous two months or had uterine pathologies (including, but not limited to, endometriosis and adenomyosis) or polycystic ovary syndrome were excluded. Endometrial biopsies were obtained from proliferative-phase endometrium. Following collection, tissue specimens were immediately transferred into sterile PBS supplemented with 1 wt% penicillin-streptomycin (P/S, Solarbio, China) and maintained at 4 °C. Subsequent cell isolation and extraction procedures were completed within 2–4 h after collection. Primary EECs and ESCs were isolated from independent human donors, with the number of donors varying depending on the specific experiments. Cell isolation and extraction procedures were performed according to an established protocol with modifications [74] . Briefly, endometrial tissue biopsies were washed three times with PBS supplemented with 1 wt% P/S, and then minced into small fragments in sterile centrifuge tubes, followed by the addition of type I collagenase solution (1 mg/mL). The tubes were placed on an orbital shaker (120 rpm) and incubated at 37 °C for 2 h to facilitate enzymatic digestion. The digestion process was terminated by adding fetal bovine serum (FBS, Gibco, USA). Subsequently, the cell suspension was filtered through a 100 μm cell strainer (Corning, USA) to remove undigested tissue fragments, and the filter was rinsed thoroughly with DMEM/F12 medium (Gibco, USA) to collect the filtrate. The collected filtrate was further passed through a 40 μm cell strainer (Corning, USA), where EECs were retained on the filter membrane and ESCs were collected in the flow-through fraction. The EECs retained on the filter membrane were backwashed, and centrifuged at 600 × g for 5 min. After centrifugation, the supernatants were carefully aspirated to obtain purified EEC pellets. The EEC pellets were resuspended in 70% (v/v) Matrigel (Corning, USA) which was pre-diluted with 30% (v/v) DMEM/F12 medium on ice, to achieve a final cell density of 8 × 10 5  cells/mL. After gently mixing by pipetting, 35 μL of the cell-Matrigel mixture was added to each well of a 48-well plate (Corning, USA). The plate was then incubated at 37 °C for 30 min to allow gelation. Subsequently, 250 μL of Expansion Medium (ExM) was added to each well. The ExM consisted of 93 wt% DMEM/F12 medium, 1 wt% P/S, 1 wt% N-2 supplement (Life, USA), 2 wt% B-27 supplement (Life, USA), 1 mM nicotinamide (MCE, USA), 0.5 μM A-83-01 (MCE, USA), 50 ng/mL epidermal growth factor (EGF, PeproTech, USA), 50 ng/mL fibroblast growth factor 10 (FGF-10, Biolegend, USA), 100 ng/mL noggin (PeproTech, USA), 9 μM Y-27632, 200 ng/mL R-Spondin-1 (PeproTech, USA), 1 wt% insulin-transferrin-selenium (ITS, Gibco, USA) and 1.25 mM N-Acetyl-L-cysteine (N-Acetylcysteine) (NAC, Meilunbio, China). The 48-well plate was subsequently transferred to a humidified incubator maintained at 37 °C with 5% CO 2 . Over time, the EECs self-organized into EEOs. After 5–7 days of culture, when EEOs reached a high density, they were passaged using TrypLE (Gibco, USA) at a split ratio of 1:2 to 1:3. The ESCs collected from the flow-through fraction were centrifuged at 800 rpm for 5 min to obtain purified ESC pellets. The ESC pellets were resuspended in DMEM/F12 complete medium (DCM), consisting of 89 wt% DMEM/F12 medium, 10 wt% charcoal stripped fetal bovine serum (CS-FBS, XP Biomed, Shanghai, China) and 1 wt% P/S. The resuspended cells were subsequently seeded into culture dishes for 2D expansion. CS-FBS was used to provide essential nutrients for cell growth while minimizing potential interference from endogenous hormones. The cells were maintained in a humidified incubator at 37 °C with 5% CO 2 . When the ESCs reached approximately 90% confluence, they were dissociated using 0.25% trypsin-EDTA solution (Gibco, USA) and passaged at a split ratio of 1:2 to 1:3. The procedure for using collagen-based hydrogel as the matrix material for EEO culture was similar to that used for Matrigel culture. Briefly, when the EEOs reached a high density in Matrigel after 5–7 days of culture, the surrounding Matrigel was washed three times with PBS, and the EEOs were gently dissociated into fragments by pipetting. After centrifugation, the resulting EEO fragments were thoroughly mixed with collagen-based bioink by gentle pipetting. A 35 μL aliquot of the EEO-laden bioink was added to each well of a 48-well plate, followed by crosslinking at room temperature for 1 h. After curing, PBS was added to each well and incubated at 37 °C for approximately 10 min to remove unbound TG from the hydrogel surface. Finally, ExM was added for cell culture. In this study, the initial density of EECs embedded in the hydrogel was 8 × 10 5  cells/mL, and subsequent experiments were performed using EEOs from passages 3–10. To quantify organoid diameter, organoid reformation, and cell metabolic activity, EEO fragments were encapsulated in collagen-based hydrogels or Matrigel and cultured for 7 days. Organoids exhibiting a clear lumen and a diameter greater than 100 μm were included for diameter measurement and reformation analysis. Organoid reformation was normalized by setting the number of reformed EEOs in the Matrigel group on day 1 as 100%, and the values of other groups were calculated proportionally. Cellular metabolic activity was evaluated using a CCK-8 assay. Briefly, EEO-laden hydrogels were placed in 48-well plates, and 400 μL of ExM containing 10% (v/v) CCK-8 reagent (Beyotime, China) was added to each well. After incubation for 2 h, 200 μL of the supernatant was transferred to a 96-well plate, and the absorbance at 450 nm was measured using a microplate reader (ELx800, BioTek, USA). Furthermore, the hormonal response function of EEOs cultured in the C5H2T1 hydrogel was evaluated. Three experimental groups were established with different hormonal treatments. The control group consisted of EEOs cultured in ExM for 11 days. The proliferative phase group consisted of EEOs cultured in ExM for 5 days, followed by 6 days of culture in ExM supplemented with 10 nM β-estradiol (E2, Beyotime, China). The secretory phase group consisted of EEOs cultured in ExM for 5 days, followed by 2 days of culture in ExM containing 10 nM E2, and subsequently cultured for 4 days in ExM supplemented with 10 nM E2, 1 μM progesterone (P4, Sigma-Aldrich, USA), and 1 μM cyclic adenosine monophosphate (cAMP, MCE, USA). For EEO monoculture, 1 μM cAMP was used to induce secretory-phase-like responses while avoiding excessive stimulation of epithelial cells [5] , [74] . ESCs were also cultured in 3D collagen-based hydrogels or Matrigel. When ESCs cultured on petri dishes reached approximately 90% confluence, they were dissociated using 0.25% trypsin-EDTA solution and subsequently centrifuged to obtain ESC pellets. The cell pellets were thoroughly mixed with collagen-based bioink at a final density of 2.4 × 10 6  cells/mL. Then, 35 μL of ESC-laden bioink was added to each well of a 48-well plate. For collagen-based hydrogels, the samples were crosslinked at room temperature for 1 h, whereas Matrigel samples were gelled at 37 °C for 30 min. After crosslinking, PBS was added to the collagen-based hydrogels and incubated at 37 °C for approximately 10 min to remove unbound TG from the hydrogel surface. ESCs from passages 3 to 6 were used for subsequent experiments. Six-day-old intact EEOs were expanded in Matrigel and harvested as described above and subsequently harvested. In parallel, ESCs were expanded through conventional 2D culture and dissociated to generate a single-cell suspension. EEO fragments and ESCs were prepared simultaneously and encapsulation in C5H2T1 bioink at final densities of 8 × 10 5  cells/mL for EECs and 2.4 × 10 6  cells/mL for ESCs. The density ratio of EECs to ESCs was determined experimentally as previously described [7] . Subsequently, 35 μL of EEO/ESC-laden bioink was added to each well of a 48-well plate and crosslinked at room temperature for 1 h. After crosslinking, PBS was added and incubated at 37 °C for approximately 10 min to remove unbound TG from the hydrogel surface. Finally, co-cultured EEOs and ESCs were maintained in a 1:1 mixture of ExM and DCM (ExM/DCM, ExM:DCM = 1:1). The culture medium was refreshed every two days. The hormonal response function of EEOs/ESCs co-cultured in the C5H2T1 hydrogel was further investigated. Three experimental groups were established with distinct hormonal treatments, which differed from those used for EEO monoculture described above. The control group consisted of EEOs/ESCs co-cultured in ExM/DCM for 11 days. The proliferative phase group consisted of EEOs/ESCs co-cultured in ExM/DCM for 5 days, followed by 6 days of culture in ExM/DCM supplemented with 10 nM E2. The secretory phase group consisted of EEOs/ESCs co-cultured in ExM/DCM for 5 days, followed by 2 days of culture in ExM/DCM containing 10 nM E2, and subsequently cultured for 4 days in ExM/DCM supplemented with 10 nM E2, 1 μM P4, and 0.5 mM cAMP. The higher concentration of cAMP used in the co-culture condition was selected to induce ESC decidualization [75] . In this study, a 3D extrusion printing system was employed following a previously established protocol [52] . Collagen-based bioinks containing varying concentrations of collagen (4–6 wt%), HA (0–3 wt%) and TG (1–2 wt%) were rapidly mixed at 37 °C for approximately 2 min to maintain a flowable state and ensure homogeneous distribution of all components. The prepared bioinks were immediately loaded into 3 mL sterile syringes for approximately 1 min, with no intentional holding or incubation period between mixing and loading. Although TG-mediated crosslinking may begin during these operations at 37 °C, the preparation and loading durations were strictly controlled and kept consistent across all experiments. Subsequent pre-crosslinking and printing were performed at 22 °C, where the increased viscosity of the collagen-based bioink helped minimize cell sedimentation, while the retained TG activity ensured sufficient crosslinking capacity for structural stabilization. This condition enabled the controlled distribution of encapsulated cells and the formation of stable printed architectures. Prior to extrusion, the bioink underwent pre-crosslinking at 22 °C for t p , during which the storage modulus increased to a suitable range for extrusion printing. Extrusion was then initiated at 22 °C and continued until completion of the printing process ( t p   +   t d ). Following printing, the structures were subjected to post-printing crosslinking at 22 °C for 1 h to achieve stabilized hydrogel formation. All printing parameters are detailed in Table 2 . Table 2 Printing parameters. Parameter Value Syringe temperature (°C) 21–23 Extrusion pressure (MPa) 0–0.25 Receiving platform (°C) 15–20 XY plotting speed (mm/s) 3–5 Nozzle inner diameter (μm) 300, 500, 700 Printing parameters. To systematically evaluate the effects of printing parameters on cellular responses, two key variables were investigated: nozzle inner diameters (300, 500, and 700 μm) and the macroporosity of printed grid structures (21.5%, 57.7%, and 75.8%). For the nozzle diameter analysis, EEOs or ESCs were encapsulated in bioinks and printed into grids with identical macroporosity to isolate the effects of extrusion-induced shear stress. Separately, the effects of macroporosity on EEO/ESC responses were evaluated using printed grid structures with different pore sizes while maintaining a constant nozzle diameter. In this experiment, an EEO/ESC-laden bioink was used for fabrication. A cast-molded hydrogel group (0% macroporosity) was included as a non-printed control. All cell-laden constructs were co-cultured for 5 days in ExM/DCM, after which gene expression profiles were quantified by RT-qPCR. A dual-nozzle extrusion bioprinter was utilized to fabricate a bilayered cellular model, consisting of an upper layer containing EEOs and a lower layer containing EEOs/ESCs. The upper layer was printed with a thickness of 2 mm and a strut spacing of 2 mm at a density of 1.2 × 10 6  cells/mL. The lower layer was printed with a thickness of 1 mm and a strut spacing of 1 mm, containing 8 × 10 5  cells/mL for EECs and 2.4 × 10 6  cells/mL for ESCs. A cast-molded bilayered hydrogel based on Matrigel, with identical cell concentrations, cell sources, and spatial organization, was used as the non-printed control. For spatial tracking, EEOs were fluorescently labeled with DiO (green, APExBIO, USA), whereas ESCs were labeled with DiL (red, APExBIO, USA). At the designated time points, the cell-laden hydrogel scaffolds were carefully retrieved and rinsed with PBS. Subsequently, a live/dead staining solution was added, and the samples were incubated in the dark at 37 °C for 30 min. The staining solution consisted of PBS supplemented with 1 μM Calcein AM (Beyotime, China) and 2 μg/mL PI (Solarbio, China). After incubation, the scaffolds were thoroughly washed with PBS and imaged using a fluorescence microscope (Nikon Ti–U, Japan). To assess cell spreading and the 3D spatial distribution of cells within the hydrogel, fluorescent staining of the cytoskeleton and nuclei was performed using FITC-phalloidin (Solarbio, China) and DAPI (Beyotime, China), respectively. The samples were first fixed with 4% paraformaldehyde (Beyotime, China) at room temperature for 30 min and subsequently washed with PBS. Following fixation, the samples were permeabilized with an immunostaining permeabilization solution (Beyotime, China) for 15 min and rinsed three times with PBS. FITC-phalloidin diluted 1:100 in PBS was applied and incubated for 1 h in the dark, followed by DAPI staining for 20 min. Z-stack fluorescence images were then captured using a confocal microscope (Olympus BX61, Japan). Total RNA was extracted from all cells using TRIzol reagent (Life Technologies, USA). Briefly, TRIzol reagent was added to the cell-laden hydrogels, and the samples were mechanically ground at 4 °C using a cryogenic grinder (JXFSTPRP-CL-BSC, Jingxin, China). Chloroform extraction was subsequently performed, and RNA was precipitated with isopropanol. The resulting RNA pellet was washed twice with 75% (v/v) ethanol, air-dried, and resuspended in DEPC-treated water (Beyotime, China). The concentration and purity of the extracted RNA were assessed using a Nanodrop spectrophotometer (Thermo Scientific, USA). Subsequently, RNA was reverse-transcribed into cDNA using reverse transcriptase (Tiangen, China) according to the manufacturer's protocol. For RT-qPCR analysis, the synthesized cDNA was mixed with target gene-specific primers ( Table 3 ), SYBR Green reagent, and DEPC-treated water to achieve the required concentration and final volume. The reactions were then analyzed using a real-time quantitative PCR system (CFX Opus 96, Bio-Rad, USA). Table 3 Primer sequences. Prime name Sequence (F) 5′-3′ Sequence (R) 5′-3′ GAPDH GGAGCGAGATCCCTCCAAAAT GGCTGTTGTCATACTTCTCATGG SOX9 AGCGAACGCACATCAAGAC CTGTAGGCGATCTGTTGGGG LGR5 CTCCCAGGTCTGGTGTGTTG GAGGTCTAGGTAGGAGGTGAAG SSEA1 ACTGCCGAGGAGGTGGATCT TGTTGGCTCAGTTGGTGGTAGT ALDH1A1 GCTGGCGACAATGGAGTCAATG CGTGGAGAGCAGTGAGAGGAGT PROM1 CCAAGGACAAGGCGTTCACAGA CGCTGGTCAGACTGCTGCTAAG SPP1 CTCCATTGACTCGAACGACTC CAGGTCTGCGAAACTTCTTAGAT PR ACCCGCCCTATCTCAACTACC AGGACACCATAATGACAGCCT PAEP GAGATCGTTCTGCACAGATGG CGTTCGCCACCGTATAGTTGAT HSD17β2 TCTTCTCGGTGTCATGCTTCC CAAAACTCCGGCAAATACCGT OLFM4 ACCTTTCCCGTGGACAGAGT TGGACATATTCCCTCACTTTGGA LIF CCAACGTGACGGACTTCCC TACACGACTATGCGGTACAGC IGFBP1 TTTTACCTGCCAAACTGCAACA CCCATTCCAAGGGTAGACGC PRL GGAGCAAGCCCAACAGATGAA GGCTCATTCCAGGATCGCAAT Primer sequences. During the IF staining process, cell-laden hydrogel samples were washed three times with PBS and fixed with 4 wt% paraformaldehyde for 30 min. The samples were then immersed in 20 wt% sucrose solution (Aladdin, China) for 24 h, followed by incubation in 30 wt% sucrose solution for another 24 h. Subsequently, the samples were incubated in a 1:1 (w/w) mixture of 30 wt% sucrose solution and OCT compound (Sakura, Japan) for 24 h. After dehydration, the samples were cleaned, embedded in OCT compound, and frozen at −80 °C. Frozen sections with a thickness of 5 μm were then prepared using a cryostat (Leica, Germany). The sections were washed with 1 × TBST buffer, followed by blocking and permeabilization for 1 h at room temperature using PBS containing 5 wt% BSA (Solarbio, China) and 0.3 wt% Triton X-100 (Beyotime, China). After removal of the blocking solution, primary antibodies (diluted 1:200 in 5 wt% BSA) were applied and incubated overnight at 4 °C. The following day, the primary antibody solution was removed, and the sections were washed thoroughly. Secondary antibodies (diluted 1:500) and Hoechst (1:500, Abcam, UK), both diluted in 5 wt% BSA, were then applied and incubated for 1 h at room temperature in the dark. After incubation, the sections were washed again. Finally, the samples were mounted with an anti-fade mounting medium (Thermo Fisher, USA), and fluorescence images were acquired using the confocal microscope. Primary antibodies used in this study included forkhead box A2 (FOXA2, Abcam, UK), e-cadherin (E-cad, Abcam, UK), Ki67(Abcam, UK), progesterone receptor (PR, Abcam, UK), mucin 1 (MUC1, Abcam, UK), pan cytokeratin (PanCK, Abcam, UK), vimentin (Abcam, UK), and estrogen receptor (ER, Abcam, UK). Secreted factors in culture supernatants were quantified using ELISA. Commercial ELISA kits for IGFBP1 (E-EL-H0442, Elabscience, China), PRL (E-EL-H0141, Elabscience, China), SPP1 (E-EL-H1347, Elabscience, China), and PAEP (EH1142, FineTest, China) were used to measure their levels, respectively, according to the manufacturers’ protocols. Briefly, culture supernatants were collected from each group after hormonal treatment and centrifuged to remove cell debris. The clarified supernatants were stored at −80 °C until analysis. Before detection, the samples were thawed and equilibrated to room temperature. Standards and samples were added to antibody-coated 96-well plates and incubated following the kit instructions. After washing, the corresponding detection reagents and substrate solutions were added sequentially. The reaction was terminated using stop solution, and the absorbance was read at 450 nm with the microplate reader. Protein concentrations were calculated against standard curves. The cell-laden endometrial models were collected into 2 mL tubes and washed three times with PBS. The samples were subsequently lysed and centrifuged to generate single-cell suspensions. The resulting cell suspensions were then loaded onto Chromium microfluidic chips and processed using the Chromium Single Cell 3′ v3.1 chemistry on a 10× Chromium Controller (10× Genomics) for cell barcoding and library preparation. Reverse transcription of barcoded transcripts and construction of sequencing libraries were performed using the Chromium Single Cell 3′ Reagent Kit v3.1 (10× Genomics) according to the manufacturer's instructions. The libraries were sequenced on an Illumina NovaSeq 6000 platform following the manufacturer's guidelines (Illumina). Raw sequencing reads were demultiplexed and aligned to the reference genome using the 10× Genomics Cell Ranger pipeline with default parameters. Unless otherwise specified, downstream single-cell RNA sequencing analyses were performed using Cell Ranger and Seurat V4.0 [76] . Briefly, gene expression matrices were generated by counting unique molecular identifiers for each gene within each cell barcode retained after Cell Ranger processing. Additional quality control filtering was performed using Seurat, where genes detected in fewer than three cells were excluded, and cells expressing fewer than 200 genes were removed. Contaminating or non-target cells were further filtered during this step. For comparative reanalysis, feature-barcode matrices generated by Cell Ranger Count or Cell Ranger Aggr were imported into Seurat, and subsequent dimensionality reduction, clustering, and gene expression analyses were performed using the corresponding default parameters. The Seurat package was used for data normalization, dimensionality reduction, clustering, and differential expression analysis. For integrated analysis of multiple datasets, the Seurat alignment method based on canonical correlation analysis (CCA) [77] was applied. For clustering analysis, highly variable genes were identified, and principal components calculated from these genes were used to construct a shared nearest-neighbor graph. Cell clusters were subsequently identified using graph-based clustering with a resolution parameter of 0.6. Based on the filtered gene expression matrix generated by Seurat, differential expression analysis between samples was performed using the edgeR package [78] to identify sample-specific marker genes. GO enrichment analysis of the identified marker genes was conducted using the clusterProfiler R package, with gene length bias correction applied. GO terms with an adjusted P -value < 0.05 were considered significantly enriched. In addition, KEGG pathway enrichment analysis was performed using the clusterProfiler R package to evaluate the functional enrichment of marker genes. Protein-protein interaction (PPI) network analysis of the identified marker genes was performed using the STRINGdb R package. For RNA velocity analysis, spliced, unspliced, and ambiguous transcript counts were generated from Cell Ranger-aligned BAM files using velocyto (v0.17.17) based on the GRCh38 gene annotation provided with Cell Ranger 2020-A. The resulting spliced and unspliced count matrices were integrated into the multi-sample dataset and analyzed using scVelo. Preprocessing was performed using scvelo.pp.filter_and_normalize with min_shared_counts = None and n_top_genes = None, without additional gene filtering based on shared counts or highly variable gene selection. First- and second-order moments were calculated using scvelo.pp.moments with 30 nearest neighbors. The n_pcs parameter was not explicitly specified; therefore, 30 principal components were used during the internal PCA calculation according to the default setting of the software. RNA velocities were estimated using the stochastic model implemented in scvelo.tl.velocity, followed by construction of the velocity graph using scvelo.tl.velocity_graph. Velocity streamlines were visualized on the UMAP embedding using scvelo.pl.velocity_embedding_stream. No root-state assignment, or uncertainty analysis was performed, as the RNA velocity analysis was used to explore local transcriptional state-transition tendencies rather than to infer definitive lineage trajectories. For cell-free material characterization experiments, n represents independent hydrogel specimens or material samples prepared and tested under the same conditions, which were considered technical replicates. For cell-based experiments, biological replicates were defined as independent donor-derived cell preparations. Technical replicates from the same donor were averaged before statistical analysis to generate one value per biological replicate. Donor information and experimental units are summarized in Supplementary Table S5 . Unless otherwise stated, image processing, quantitative analyses, and graphical presentation were performed using ImageJ, OriginPro 2018 and GraphPad Prism (GraphPad Software, USA). Statistical analyses were performed using GraphPad Prism, and data are presented as mean ± SD. Statistical tests were selected according to the experimental design. One-way ANOVA was used for single-factor comparisons, two-way ANOVA for two independent factors, and repeated-measures ANOVA with Geisser–Greenhouse correction where applicable for matched donor-derived samples. Two-tailed paired t-tests were used for two matched groups. Appropriate post hoc tests (Tukey's, Dunnett's, or Šídák's) were applied for multiple comparisons, and multiplicity-adjusted P values were reported. Statistical significance was defined as: ns, not significant; * P  < 0.05, ** P  < 0.01, *** P  < 0.001, and **** P  < 0.0001.

The human uterus, essential for embryo implantation and pregnancy maintenance, relies on the endometrium, which undergoes cyclic remodeling [1] in response to hormonal changes. Dysfunctional endometrial regeneration, often caused by infections, surgeries, or disorders like endometriosis and Asherman's syndrome [2] , can lead to infertility and other complications [3] . Despite their clinical significance, research on these conditions has been hindered by ethical constraints and the limited translational relevance of animal models. To overcome these challenges, the development of three-dimensional (3D) human endometrial-like tissues that accurately replicate the structure and function of the native endometrium in vitro represents a transformative approach. This advancement holds promise for elucidating disease mechanisms and developing targeted therapies, addressing a critical unmet need in reproductive medicine. The development of physiologically relevant 3D endometrial models relies heavily on primary endometrial cells including endometrial epithelial cells (EECs) and endometrial stromal cells (ESCs), and organoids such as endometrial epithelial organoids (EEOs), which require highly bioactive culture matrices. While Matrigel, a tumor-derived material, has been widely used to support organoid growth [4] , [5] or modulate Ishikawa cell proliferation [6] , the undefined composition, low mechanical strength [7] , and poor processability [8] , [9] ( Fig. 1 ) limit its clinical applicability [10] , [11] and ability to replicate complex endometrial architectures. These limitations hinder the exploration of structure-function relationships in endometrial tissues. 3D bioprinting, with its capacity for precise, layer-by-layer fabrication of complex, biomimetic structures, offers advantages such as high efficiency, freeform fabrication, and personalized customization [12] . Although Zou's group demonstrated the regenerative potential of 3D bioprinted endometrial constructs for severe injuries, the development of long-term, physiologically relevant in vitro endometrial models remains challenging [13] . Thus, developing a bioink capable of balancing printability and bioactivity represents an important step toward the 3D bioprinting of in vitro endometrial models. Fig. 1 Schematic of in vitro endometrial model fabrication using printable enzymatic cross-linking collagen-based bioink comparing with unprintable Matrigel. Schematic of in vitro endometrial model fabrication using printable enzymatic cross-linking collagen-based bioink comparing with unprintable Matrigel. Collagen, a major extracellular matrix component comprising ∼30% of total body proteins [14] , plays a pivotal role in cell survival, spreading, adhesion and functional cellular communication due to its triple-helical structure [15] , [16] , [17] , [18] , [19] and arginine-glycine-aspartic acid (RGD) peptide sequences [20] . Its bioactivity [21] , [22] , combined with its degradability by matrix metalloproteinases (MMPs) [23] , makes it a promising candidate for constructing endometrial models in vitro. Previous studies have demonstrated the value of collagen-based matrices in supporting endometrial tissue slices [24] , ESCs [25] , [26] , or EEOs [7] , [27] , [28] under 3D culture conditions. Nevertheless, many of these models involve tissue slices [24] , mold-based fabrication [7] , [24] , [25] , [26] , [27] , [28] , or collagen-Matrigel composite matrices [7] , [27] , which may pose challenges for reproducibility, spatial control, or translational applicability. In addition, hydrogel contraction [26] , [27] and limited regulation of epithelial–stromal spatial organization [28] remain to be further addressed. Therefore, the development of a printable collagen-based system capable of maintaining structural stability, supporting epithelial–stromal co-culture, and enabling controlled reconstruction of endometrial architecture remains highly desirable. Developing collagen-based bioinks with both excellent printability and bioactivity remains challenging. Current cell-laden collagen bioinks, typical at concentrations below 1.5 wt% [29] , [30] , lack self-supporting properties for extrusion 3D bioprinting [29] , [31] and exhibit insufficient mechanical stability for long-term organoid [32] or ESC culture [33] . High-concentration collagen hydrogels [34] formed by self-assemble [35] , [36] present dense and short fibers [37] with low porosity and swelling ratio [38] , leading to cell death [29] , [39] . Other existing crosslinking strategies, including photocurable group modifications [40] , [41] , [42] and chemical agents like genipin [43] and tannic acid [44] , [45] , often result in cytotoxicity (photoinitiator toxicity [46] , radical toxicity or phototoxicity [47] ) with cell viability frequently below 80% [41] , [42] , or impair cell spreading and functions. These limitations highlight the urgent need for improved crosslinking mechanisms and printing strategies in collagen-based bioink development. This study presents an enzymatically crosslinked collagen-based bioink and an extrusion-based bioprinting strategy for fabricating an in vitro endometrial model ( Fig. 1 ). Collagen served as the principal bioactive matrix, while hyaluronic acid (HA) [48] and transglutaminase (TG) [49] were incorporated to optimize bioink performance. TG-catalyzed crosslinking enabled room-temperature pre-crosslinking of the collagen bioink, thereby optimizing its rheological properties for extrusion printing while maintaining high cell viability under mild and cytocompatible processing conditions. As a result, the bioink supported the co-culture of human primary endometrium-derived EEOs and ESCs, with EEOs retaining high viability and hormone-responsive function after 11 days of culture. Compared with the Matrigel-cast model, the collagen-printed model exhibited improved biofunctional performance and enhanced structural resemblance to the native endometrium. Overall, the TG-crosslinked collagen-based bioink and printed endometrial model developed in this study provide a promising platform for advancing endometrial physiology and pathology research, embryo–endometrium interaction studies, and related in vitro applications.

The authors declare no competing interests.

All the data supporting the findings of this study are available within the main text and the Supplementary Materials upon request.