AI-driven text mining of the female reproductive system: enabling multiscale biomedical modeling and personalized medicine.

We used AI methodologies to uncover hidden topics from the collected articles on the female reproductive system. Figure  1 illustrates the pipeline of the unsupervised topic discovery procedure. Overall, we employed BioBERT to transform the textual information in the article abstracts into contextualized vectors with biomedical semantics and applied BERTopic (Fig.  1 ) to extract key topics related to the female reproductive system, without relying on supervision from human experts or predefined topic sets. Specifically, the pipeline consists of three stages: (i) biomedical contextualization, (ii) unsupervised topic modeling, and (iii) topic map visualization. Fig. 1 Overview of the BERTopic pipeline for unsupervised topic discovery on abstracts of peer-reviewed articles about the female reproductive system. The process includes biomedical contextualization, topic modeling, and topic visualization Overview of the BERTopic pipeline for unsupervised topic discovery on abstracts of peer-reviewed articles about the female reproductive system. The process includes biomedical contextualization, topic modeling, and topic visualization We conducted a comprehensive and systematic collection of research articles focusing on bioengineered models of the female reproductive system, encompassing studies published between 2000 and 2024. The search was carried out across multiple international databases, including PubMed, Web of Science, and Scopus, using a predefined set of keywords such as “endometrium-on-a-chip,” “placenta organoid,” “ovarian bioprinting,” “cervix chip,” and “fallopian tube model.” These keywords were carefully selected to encompass a broad range of experimental platforms and organ-specific applications, ensuring that relevant studies across various engineering approaches were captured. To maintain rigor and reproducibility, only peer-reviewed journal articles written in English and providing sufficient methodological detail were included in the dataset. Review articles, editorials, conference abstracts, and studies lacking experimental reproducibility were excluded to minimize bias and ensure the inclusion of high-quality evidence. After collection, the studies were organized into five anatomical domains: endometrium, placenta, ovary, cervix, and fallopian tubes, which allowed us to examine research trends within each reproductive organ. Within each domain, the literature was further classified based on the technological platform employed, including chip-based models (2D and 3D microfluidic systems), bioprinting approaches, conventional 2D cell culture, advanced 3D cell culture systems, and organoid-based models. This multi-tiered classification strategy was essential for capturing the breadth of engineering strategies and for identifying overlaps where similar methodologies have been applied across different reproductive tissues. Importantly, our dataset includes multiple instances of each category, for example, repeated applications of chip (2D/3D) and bioprinting platforms, alongside a particularly extensive use of organoid systems—as well as numerous studies employing both 2D and 3D cell culture methodologies. By recording detailed metadata for each publication, such as year of publication, authorship, journal source, methodological approach, and specific platform characteristics, we ensured that the dataset could be systematically compared and cross-referenced across organs and technologies. Through this structured process, we established a robust dataset that not only reflects the diversity of engineering strategies applied to reproductive biology but also provides a clear framework by which the literature can be categorized according to both organ system and platform type. In this way, the dataset enables systematic examination of how different technological approaches have been utilized to model distinct aspects of the female reproductive system, while also highlighting areas of overlap and emerging research directions. Ultimately, this comprehensive collection and classification process serves as the methodological foundation for the analyses presented in this review. To transform plain text into contextualized vector embedding with biomedical semantics, we employed the standard WordPiece tokenizer from the original BERT architecture to ensure compatibility with BioBERT’s pretrained biomedical knowledge during embedding and contextualization. This process comprises three key stages: text tokenization, token embedding, and biomedical contextualized embedding. To begin with, each abstract is split into subword units using the WordPiece tokenizer. Tokenization is a crucial preprocessing step that maps unstructured text into consistent units that can be handled by transformer models. WordPiece operates at the subword level, allowing rare or compound biomedical terms (e.g., “syncytiotrophoblast”) to be decomposed into smaller, more frequent subunits (e.g., “syncytio”, “tropho”, “blast”), thus mitigating the out-of- vocabulary problem common in specialized domains. Following tokenization, special tokens are inserted to structure the input. A classification token, [CLS], is prepended to the sequence to represent the entire abstract, while [SEP] tokens are used to separate distinct sentences or segments. This [CLS] token is also used as a representative of the whole abstract of a paper. Each token is then represented by the sum of three embeddings: a token embedding (from the vocabulary), a segment embedding (indicating sentence membership), and a positional embedding (encoding token order). In particular, token embeddings are derived from the BioBERT vocabulary, which is specialized for biomedical terminology. Each token is linked to a vocabulary index, which retrieves its vector representation from a pre-trained embedding matrix. This matrix is learned during the model’s pretraining phase using large biomedical corpora including PubMed and PMC. It can capture semantic relationships between tokens. This process serves as a bridge between the raw biomedical language and the numerical input required by machine learning models. Once embedded, the token sequence is passed through BioBERT, a domain adapted version of BERT (Bidirectional Encoder Representations from Transformers). BioBERT retains the architecture of BERT-base, consisting of 12 stacked transformer encoder layers with a hidden size of 768, but it is further pre-trained on biomedical corpora. Each transformer layer consists of two main components: a multi-head self attention mechanism and a position-wise feedforward network (FFN). These components are wrapped in residual connections and layer normalization, enabling deep stacking while mitigating vanishing gradient issues. BERT generates contextual embeddings by jointly attending to the left and right contexts of each token. It is trained with two pretraining objectives: masked language modeling (MLM), where random tokens are masked and predicted, and next sentence prediction (NSP), which models inter-sentence coherence. This bidirectional training allows the model to capture long-range dependencies and sentence-level meaning—capabilities that are particularly useful for biomedical abstracts with complex clause structures. In the self-attention mechanism, each token is projected into a query Q, a key K, and a value V, and attention is computed as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \begin{aligned} Attention\left( X \right) \\ = & softmax\left( {\frac{{QK^{T} }}{{\sqrt {d_{k} } }}V} \right) \\ = & softmax\left( {\frac{{(XW^{Q} )(XW^{K} )T}}{{\sqrt {d_{k} } }}XW^{V} } \right) \\ \end{aligned} $$\end{document} This operation allows each token to dynamically weight the importance of all other tokens based on semantic similarity. For example, in the phrase "implantation occurs during the window of endometrial receptivity", the word "receptivity" can attend to both "implantation" and "window" to refine its meaning in context. BioBERT applies this process across 12 layers, with each layer refining the contextual embedding. The final output consists of contextualized vectors for each token (dimension 768), with the [CLS] token vector serving as a sentence-level embedding summarizing the abstract. To uncover latent research themes in biomedical abstracts, we employed BERTopic, which enables unsupervised, non-parametric topic discovery based on semantic similarity. This design is particularly well-suited for biomedical corpora, which often contain domain-specific vocabulary, hierarchical concepts, and syntactically dense abstracts. The pipeline begins with high-dimensional BioBERT embeddings, which are projected into a lower-dimensional space using UMAP. Clustering is then performed using HDBSCAN, followed by topic labeling via class-based TF–IDF (c-TF–IDF). BioBERT embeddings produce 768-dimensional vector representations for each abstract, capturing rich contextual information. However, directly clustering or visualizing data in such high-dimensional spaces is often ineffective due to the “curse of dimensionality” a phenomenon in which distances between points become less meaningful as dimensionality increases. This not only degrades the reliability of similarity metrics but also makes clustering and density estimation unstable. To address this, we apply UMAP (Uniform Manifold Approximation and Projection), a nonlinear dimensionality reduction technique that preserves local neighborhood structures in the data. UMAP works by constructing a weighted k-nearest neighbor graph in the high-dimensional space, modeling it as a fuzzy simplicial set, a mathematical object that captures neighborhood connectivity with probabilistic weights. It then finds a low-dimensional embedding that minimizes the cross-entropy between the high-dimensional and low-dimensional graphs. Intuitively, UMAP tries to preserve which points are “close” to one another without distorting the overall manifold structure. In biomedical corpora, where semantically similar abstracts may differ by only a few technical terms (e.g., “endometrial receptivity” vs. “embryo implantation”), preserving fine-grained local relationships is crucial. To capture such subtle distinctions, we set the number of neighbors in UMAP to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{neighbors}$$\end{document} = 10. This parameter controls the trade-off between local and global structure: smaller values emphasize fine-grained neighborhood preservation, which is particularly important in biomedical text where topic boundaries are often nuanced and lexically sparse. We reduced the embedding dimensionality to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{component}$$\end{document} = 5, based on empirical findings that lower dimensions (e.g., 2 or 3) tend to oversimplify the latent structure and lead to overly coarse topic groupings. Although this setting complicates direct visualization, our primary objective was to enhance topic separability for downstream clustering. In this reduced 5-dimensional space, documents form more compact and semantically coherent clusters, better reflecting the thematic organization of the corpus. It is important to note that this representation space is distinct from the 2D projection used for topic map visualization. While the 5D embeddings serve as input to the clustering algorithm, a separate 2D projection is later used solely for illustrative purposes. This ensures that clustering quality is not compromised by constraints imposed by visualization, while still allowing interpretable topic landscape mapping. After projection, the reduced embeddings are clustered using Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). Unlike centroid-based algorithms such as k-means, which assume that clusters are convex and of similar size, HDBSCAN identifies clusters as contiguous regions of high density in the data manifold. Sparse regions are labeled as noise, enabling the algorithm to filter out outlier points that do not fit any dominant group. In essence, HDBSCAN constructs a hierarchy of clusters based on the mutual reachability distance between points and then condenses this hierarchy into a flat clustering by selecting the most persistent density peaks. This approach is non-parametric in nature: it does not require the number of clusters to be specified in advance and adapts naturally to the shape, size, and density variation of clusters in real-world data. HDBSCAN exposes two main parameters: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{m}\mathrm{i}\mathrm{n}\_cluster\_size$$\end{document} , which determines the smallest size of a cluster, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{m}\mathrm{i}\mathrm{n}\_\mathrm{s}\mathrm{a}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{e}\mathrm{s}$$\end{document} which controls the sensitivity to local density variations. In our setting, these parameters were selected to balance granularity and noise tolerance, preserving meaningful groupings while avoiding fragmentation. This flexibility is particularly valuable in biomedical literature, where abstract content ranges from broad conceptual reviews to narrowly focused experimental findings. For example, documents on “placental biology” may vary widely in methodology and terminology but still fall within a coherent topical space. HDBSCAN can accommodate such internal variation without forcing sharp boundaries. At the same time, it identifies scattered or idiosyncratic abstracts such as highly technical methods papers or interdisciplinary edge cases as noise, preventing them from distorting the semantic structure of the main topics. Overall, HDBSCAN’s ability to discover irregular, non-spherical, and variably sized clusters without explicit assumptions makes it especially suitable for unsupervised topic modeling in complex, heterogeneous biomedical corpora. Once clusters are established, we extract interpretable topic descriptors using class-based term frequency-inverse document frequency (c-TF–IDF). This technique adapts the traditional TF–IDF formulation to a cluster-based setting, where each cluster is treated as a single “meta-document” formed by concatenating all abstracts within the cluster. In standard TF–IDF, the importance of a term in a document is calculated based on how frequently it appears in that document relative to how common it is across the entire corpus. However, this approach operates on individual documents and is thus not well suited for capturing inter-cluster salience. In contrast, c-TF–IDF generalizes this idea to the topic level, identifying terms that are discriminative across clusters—that is, terms that appear frequently within one cluster but are rare in others. Formally, the weight of term x in cluster c is defined as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${W}_{x,c}=\Vert {tf}_{x,c}\Vert \times log(1+\frac{A}{{f}_{x}})$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${tf}_{x,c}=\frac{{f}_{x,c}}{\left|c\right|}$$\end{document} is the normalized term frequency of x in cluster \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c$$\end{document} , with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{x,c}$$\end{document} the raw count of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x$$\end{document} , and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|c\right|$$\end{document} the total number of tokens in the cluster document. The global frequency of term x across all clusters is: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ f_{x} = \sum\limits_{{c\prime }} {f_{{x,c^{\prime } }} } $$\end{document} and the average number of tokens across clusters is: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ A = \frac{1}{N}\sum\limits_{{c\prime }}^{N} {\left| {c^{\prime } } \right|} $$\end{document} This formulation ensures that terms selected for topic labeling are both representative—frequent within their cluster—and distinctive—rare across other clusters. Such discriminative power is particularly valuable in biomedical corpora, where general-purpose terms like “cell” or “expression” occur widely, while specialized terms such as “syncytiotrophoblast” or “zona pellucida” are closely tied to specific biological processes. For example, one cluster was characterized by terms such as trophoblast , EMSFs , endometrial , and placenta , illustrating a semantically coherent theme around implantation and endometrial receptivity. Even seemingly generic words like long or long term gained discriminative value within this context, due to their frequent use in longitudinal reproductive studies. By identifying these high-weight terms, c-TF–IDF enables concise and interpretable labeling of otherwise opaque clusters, facilitating expert validation, comparative analysis, and downstream visualizations. This final step of the topic modeling pipeline transforms abstract embeddings and unsupervised groupings into human-readable scientific themes. Following topic extraction using HDBSCAN, which yielded 15 fine-grained clusters, we conducted a manual postprocessing step to improve thematic coherence and interpretability. This involved grouping semantically related clusters into broader topic categories. The aggregation process was informed by a combination of cluster-level keywords extracted using class-based TF–IDF and domain expertise in reproductive biology. To facilitate expert interpretation, each cluster was labeled using representative keywords, which were then summarized into human-readable topic names with the assistance of the GPT-4o model. These initial cluster names and their associated keywords served as a guide for domain experts to manually merge similar clusters into six overarching major themes. This manual merging process enabled us to move from an unsupervised clustering output to a semantically meaningful topic structure that reflects key biological themes. Figures  2 , 3 , and 4 summarize the mapping from original clusters to their aggregated topics. Fig. 2 Embedding pipeline from raw text to vectorized input. Tokenization breaks text into subwords, which are then encoded using token, segment, and positional embeddings before being passed into the BioBERT encoder Fig. 3 Architecture of BioBERT. Input embeddings are processed through 12 transformer layers consisting of multi-head self-attention and feedforward networks, producing contextualized token and sentence representations Fig. 4 Workflow of the AI-driven topic modeling pipeline. A total of 347 peer-reviewed articles were collected and categorized across five anatomical domains of the female reproductive system. Abstracts were embedded using BioBERT, followed by unsupervised topic modeling with BERTopic. Clusters were refined through HDBSCAN-based filtering and expert-guided merging, resulting in six major thematic categories used for downstream analysis Embedding pipeline from raw text to vectorized input. Tokenization breaks text into subwords, which are then encoded using token, segment, and positional embeddings before being passed into the BioBERT encoder Architecture of BioBERT. Input embeddings are processed through 12 transformer layers consisting of multi-head self-attention and feedforward networks, producing contextualized token and sentence representations Workflow of the AI-driven topic modeling pipeline. A total of 347 peer-reviewed articles were collected and categorized across five anatomical domains of the female reproductive system. Abstracts were embedded using BioBERT, followed by unsupervised topic modeling with BERTopic. Clusters were refined through HDBSCAN-based filtering and expert-guided merging, resulting in six major thematic categories used for downstream analysis To visualize the distribution of abstracts in semantic space, we plotted a 2D scatter map using UMAP-reduced embeddings. Each abstract was first embedded using BioBERT, followed by dimensionality reduction via UMAP to preserve local and global structures. The resulting 2D coordinates were plotted using Seaborn’s scatterplot functionality in Python. To enhance visual distinction between overlapping clusters, we applied Gaussian jittering to each point, adding normally distributed noise (σ = 3) to the UMAP coordinates. Points are colored by the final merged topic group, and the legend shows the manually assigned topic names based on expert review of cluster keywords (Table  1 ). Table 1 AI-derived major topics, subtopics, and associated keywords in female reproductive system research Major topic Subtopics (Refined) Representative keywords Dominant platforms Key biological focus Implantation & endometrial receptivity Endometrial profiling, trophoblast–stromal interactions within the implantation environment, endometrial repair and regeneration Endometrial, trophoblast, stromal, menstrual flow, trophectoderm, long term, endometrial gland, escs, eeo, hpmcs, endometriotic, ahr, rat endometrial Organoid, organ-on-chip, 3D bioprinted hydrogel Hormone-responsive remodeling; epithelial–stromal crosstalk; implantation microenvironment Placental interface & maternal–fetal barrier Placental barrier, placental invasion, infectious placenta Placental, placenta, trophoblast, barrier, placental barrier, fetal, glucose, pericytes, invasion, preeclampsia, epithelial, infections, chlamydia, zikv, evt, evts Organ-on-chip, microfluidics Maternal–fetal exchange; barrier integrity; immune interaction Gamete interaction & fertilization Fimbriae epithelium, oviduct EVs Sperm, chip, microfluidic, selection, bioactivity, viscosity, bidirectional endocrine, oviduct, evs, oviduct evs, fimbriae, fte, curvature, epithelial Microfluidic chip, organ-on-chip Gamete transport; fertilization niche; endocrine-regulated interaction Follicle & ovary microenvironment Follicle development, endocrine signaling Follicles, follicle, oocyte, oocytes, primordial, immature, preantral, fertility, granulosa, ovulation, endocrine loops 3D culture, organoid Follicle development; hormone secretion; ovarian niche ECM and tissue engineering platforms Fibrin matrix, hydrogel maturation Hydrogels, hydrogel, ECM, scaffold, bioinks, fibrin, fibrin matrices, collagen, synthetic, maturation, ECM components, mscs 3D bioprinting, scaffold, hydrogel ECM remodeling; tissue repair; angiogenesis Tumor models & cancer spheroids Cancer spheroids, drug delivery tumor Cancer, tumor, tumors, spheroid, spheroids, ascites, hgsoc, eoc, hgsc, drug, sensitivity, pdo, ezh2, tissueoriginated 3D spheroid, bioprinting, microfluidics Tumor–stroma interaction; drug penetration; resistance AI-derived major topics, subtopics, and associated keywords in female reproductive system research To systematically analyze research trends in the female reproductive system, a total of 347 peer-reviewed articles were identified through comprehensive database searches and categorized across five anatomical domains (Fig.  4 ). Abstracts from these studies were embedded using BioBERT to capture contextual semantic relationships and subsequently analyzed using BERTopic for unsupervised topic clustering. Of the initial corpus, 299 documents were assigned to valid clusters, while 48 were identified as noise by HDBSCAN and excluded from further analysis. Initially, 15 fine-grained subtopics were generated and subsequently refined through expert-guided merging into six major thematic categories. An additional curation step removed topics not directly aligned with reproductive bioengineering (e.g., sperm selection). The final dataset of 280 articles was used for downstream topic map visualization and trend analysis. The figure (Fig.  5 ) provides an interpretable overview of how biomedical abstracts cluster into high-level themes across the corpus. Fig. 5 Topic map of the discovered clusters. Each point represents an article abstract embedded with BioBERT and projected to a 2D space using UMAP Topic map of the discovered clusters. Each point represents an article abstract embedded with BioBERT and projected to a 2D space using UMAP Fifteen clusters were initially generated using HDBSCAN and subsequently aggregated into six high-level thematic categories through manual review by domain experts, guided by class-based TF–IDF keywords and GPT-4o assisted topic naming.

In this study, we implemented an AI-driven, unsupervised topic discovery framework to systematically synthesize and reinterpret two decades of research on bioengineered models of the female reproductive system. Unlike conventional narrative reviews, which rely on manual article selection and predefined thematic categories, our approach enables data-driven comparison across thousands of publications by extracting latent semantic structures directly from biomedical text. To improve clarity and methodological transparency, the analytical workflow was explicitly structured into sequential steps, including corpus preprocessing, contextual embedding generation using BioBERT, and topic extraction using BERTopic with density-based clustering. This stepwise framework ensures reproducibility and facilitates interpretation of how thematic structures are derived from the underlying text data. By integrating BioBERT-based contextual embeddings with density-based clustering, we reconstructed the research landscape of the endometrium, placenta, ovary, fallopian tube, and cervix in a reproducible and scalable manner. Advances in AI technologies are transforming the research framework of the female reproductive system and are emerging as powerful tools in this field. Using this AI-driven framework, research trends and highly emphasized keywords within each reproductive organ category can be rapidly and systematically identified. By organizing large-scale literature into semantically coherent topic clusters, the framework allows efficient comparison of dominant research themes and frequently discussed biological and engineering concepts across organ domains. Beyond literature summarization, the cluster-based topic analysis provides a quantitative basis for gap analysis across reproductive organs and technological platforms. High-density topic clusters were predominantly associated with endometrial and placental models, reflecting sustained methodological development in organoids, barrier systems, and vascularized microphysiological platforms. In contrast, comparatively sparse or fragmented topic representations were observed for biologically critical yet technically challenging areas such as fallopian tube modeling and ovarian aging. These gaps are consistent with prior reports highlighting persistent limitations in cell sourcing, long-term culture stability, and hormone-responsive modeling for these organs. Importantly, the ability of AI-based topic modeling to reveal such imbalances enables underexplored yet clinically relevant domains to be identified in an unbiased manner, thereby informing future research prioritization. Importantly, this framework also enables objective identification of evolving research trends. Analysis of topic density and publication distribution over time reveals a clear shift in recent years toward physiologically integrated and functionally dynamic systems. Emerging themes increasingly emphasize endocrine regulation, vascular remodeling, and cyclic hormonal responsiveness, reflecting a broader transition from static single-organ models toward systems-level representations of female reproductive biology. These trends align with the growing adoption of multi-organ microphysiological systems and body-on-a-chip platforms designed to capture hormonal feedback loops and inter-organ communication. In addition to identifying research gaps and temporal trends, the semantic proximity between topic clusters provides insight into multi-organ integration potential. Topics related to ovarian steroidogenesis, endometrial receptivity, and placental vascularization frequently co-localized in the embedding space, suggesting biologically coherent axes that are particularly well-suited for multi-organ modeling. Such findings support the rational design of ovary–endometrium–placenta platforms to study cyclic regulation, implantation failure, and pregnancy-associated disorders, while also highlighting opportunities for integrating cervix–immune–infection modules or endocrine–metabolic interfaces involving the liver and hypothalamic–pituitary axis [ 128 ]. For instance, endometrial receptivity clusters align with patient-derived endometrium-on-a-chip platforms enabling personalized assessment of endometrial receptivity-associated markers [ 30 ], trophoblast invasion clusters correspond to placental barrier-on-chip systems modeling maternal–fetal drug transport [ 129 ], and ovarian folliculogenesis clusters point to organoid-based platforms that capture tumor heterogeneity and drug responses across ovarian cancer subtypes [ 130 ]. In this context, AI-derived topic maps function as conceptual blueprints for multi-organ system architecture rather than as passive descriptive tools. From a translational perspective, the topic-guided framework presented here can be leveraged in multiple ways. The proposed framework can serve as a decision-support tool for experimental design, guiding platform developers toward biologically prioritized readouts and underrepresented organ combinations. Second, it enables benchmarking of engineering modalities including organoids, organ-on-a-chip, and 3D bioprinting by revealing how different technologies converge on shared biological themes [ 11 ]. Third, it provides a foundation for hypothesis generation by identifying emerging topic intersections that have yet to be systematically explored. Collectively, these applications position AI-assisted topic modeling as a bridge between large-scale literature synthesis and experimental strategy development in reproductive bioengineering. Nevertheless, several limitations of the current framework should be acknowledged. Topic modeling was primarily conducted using article abstracts, which may not fully capture experimental nuance, methodological constraints, or negative findings that are often described only in full texts. Furthermore, while embedding-based clustering effectively captures semantic relationships, expert-driven interpretation remains essential to assign topic labels that are biologically and clinically meaningful within the context of female reproductive system research. Future work incorporating full-text embeddings, citation network analysis, and longitudinal topic modeling will further enhance the resolution, robustness, and interpretability of AI-assisted synthesis [ 131 ]. Building upon these findings, future extensions of the present framework could move beyond topic-level analysis toward more integrative and predictive modeling approaches. For instance, incorporation of knowledge graph representations may enable explicit mapping of relationships among reproductive organs, biological processes, and engineering platforms. In parallel, graph-based learning methods, including graph neural networks, could facilitate the identification of higher-order interactions that are not readily captured by embedding-based clustering alone. Furthermore, integrating literature-derived semantic structures with experimental and clinical datasets may enhance the translational utility of the framework. Such multimodal integration could support the alignment of AI-inferred research trends with measurable biological outcomes, thereby improving the interpretability and applicability of the franework. Importantly, the adoption of expert-in-the-loop strategies will remain critical to ensure that AI-driven outputs are grounded in biological relevance and clinical context. Rather than replacing domain expertise, these approaches are expected to augment hypothesis generation and guide rational experimental design in reproductive bioengineering.

The human endometrium is a highly dynamic, hormone-responsive mucosal tissue that undergoes cyclic remodeling across the menstrual cycle to prepare the uterus for embryo implantation. Fluctuating ovarian hormones orchestrate these periodic transformations, which, if implantation does not occur, culminate in tissue breakdown and rapid repair during menstruation [ 1 ]. During the proliferative phase, rising estrogen from developing ovarian follicles stimulates regeneration of the functional layer, with stromal cells and glands proliferating rapidly and the endometrium thickening to approximately 8–12 mm by the time of ovulation [ 2 ]. Following ovulation, progesterone secreted by the corpus luteum drives the secretory transformation of the endometrium: glands become highly coiled and glycogen-rich, spiral arteries elongate, and stromal cells differentiate into a receptive state. In the absence of fertilization, withdrawal of ovarian steroids triggers spiral artery constriction, ischemia, and eventual shedding of the functional layer as menstruation. Over the past decade, advances in high throughput molecular and spatial profiling encompassing transcriptomics, proteomics, epigenomics, and single cell analyses have greatly expanded our understanding of endometrial cellular heterogeneity, signaling pathways, and functional states across these phases [ 3 ]. Such comprehensive endometrial profiling reveals distinct epithelial, stromal, immune, and vascular populations and their dynamic interactions, providing critical insights into the determinants of endometrial receptivity and implantation success [ 24 ]. Successful embryo implantation depends on highly orchestrated interactions between the maternal endometrium and the invading trophoblast. Specifically, trophoblast–stromal interactions within the implantation environment play a pivotal role in establishing maternal–fetal tolerance, modulating immune responses, and remodeling the decidual extracellular matrix to allow for embryo invasion and placentation. Dysregulation of these processes is implicated in recurrent implantation failure and early pregnancy loss [ 25 ]. In parallel, the endometrium exhibits remarkable capacity for endometrial repair and regeneration after each menstrual shedding. Endometrial stem/progenitor cell populations, angiogenic signaling, and tissue-resident immune cells cooperate to rapidly restore tissue integrity while preserving reproductive function. Disruption of these regenerative processes contributes to pathologies including Asherman’s syndrome, thin endometrium, endometrial atrophy, and infertility. Understanding the mechanisms of endometrial repair and regeneration has significant translational implications for reproductive medicine, regenerative therapies, and tissue engineering approaches such as organ-on-a-chip models of implantation [ 5 ]. Recent studies have demonstrated that endometrial organoids possess regenerative potential by recapitulating native tissue architecture and function, and can be directly applied as a therapeutic strategy. Notably, transplantation of endometrial organoids into a murine model of Asherman’s syndrome significantly restored endometrial structure and function, reducing fibrotic lesions, enhancing angiogenesis, and improving implantation outcomes. Mechanistically, these effects were mediated in part by the transfer of functional mitochondria from organoids to damaged endometrial cells, leading to metabolic restoration and reversal of fibrosis-associated dysfunction. These findings highlight the importance of integrating cellular, metabolic, and microenvironmental cues in developing next-generation regenerative platforms for endometrial repair [ 26 ]. Impaired endometrial development and function constitute a major determinant of implantation failure and infertility. Under normal physiological conditions, ovarian hormones orchestrate epithelial proliferation, stromal differentiation, and vascular remodeling to establish a receptive endometrial lining. Disruption of these coordinated processes impairs tissue architecture, attenuates hormone responsiveness, and diminishes the capacity for embryo implantation. A thin endometrium exemplifies this impairment. During the mid-luteal phase, an endometrial thickness of ≤ 7 mm reflects inadequate proliferative expansion or incomplete progesterone-driven maturation [ 7 ]. Such structural insufficiency reduces glandular complexity and vascular support, thereby compromising the histological and molecular milieu required for embryo attachment and implantation. Endometriosis provides a further illustration of dysregulated endometrial function. Beyond ectopic lesions, chronic inflammation and altered immune signaling in the eutopic endometrium induce progesterone resistance and hinder stromal decidualization [ 8 ]. As a result, the endometrium fails to achieve full receptivity, reducing implantation competence even when embryo quality is optimal. Luteal phase deficiency also undermines secretory transformation through endocrine insufficiency. Inadequate progesterone secretion prevents complete glandular differentiation, limits stromal decidualization, and impairs vascular remodeling. This incomplete maturation is frequently associated with recurrent implantation failure and early pregnancy loss [ 18 ]. Collectively, these conditions underscore that endometrial receptivity cannot be inferred from thickness alone. Rather, it emerges from the integrated regulation of tissue architecture, endocrine responsiveness, immune balance, and intercellular communication. Elucidating how these elements converge to establish a receptive state provides a foundation for more precise diagnostic criteria, targeted therapeutic strategies, and advanced in vitro platforms that more faithfully recapitulate endometrial physiology and dysfunction, thereby facilitating improved implantation outcomes and reproductive success. Topics: endometrial profiling, trophoblast–stromal interactions within the implantation environment, endometrial repair and regeneration A topic map-based analysis highlights three interconnected research domains: endometrial profiling, trophoblast–stromal interactions within the implantation environment, and endometrial repair and regeneration. Endometrial profiling defines the molecular and structural attributes that enable the uterus to achieve a receptive state, particularly through glandular activity and stromal cell responsiveness. The implantation environment emphasizes the interactions between trophoblasts and maternal stromal or immune cells, which regulate embryo attachment, invasion, and the transition toward placental development. When endometrial receptivity is impaired by abnormal endometrial conditions, including endometriosis, intrauterine adhesions, and thin endometrium, repair and regeneration emerge as essential processes to restore functionality and re-establish an implantation-competent environment. Collectively, these domains represent a spectrum of processes that illustrate the central role of endometrial receptivity in determining implantation outcomes. The endometrial profile has shifted from viewing the endometrium as a uniform mucosal lining to recognizing it as a dynamic, multicellular ecosystem shaped by hormonal signaling and spatially organized cell states. High-resolution single-cell and spatial transcriptomic analyses have revealed that the endometrium is composed of diverse epithelial, stromal, endothelial, and immune subsets whose distributions vary across the menstrual cycle. Progenitor-like SOX9 + LGR5 + epithelial populations, perivascular stromal niches, and immune subclusters orchestrate cyclical remodeling, underscoring that the endometrial profile is inherently heterogeneous and temporally dynamic (Fig.  6 a) [ 27 ]. Complementing these molecular and spatial insights, organoid technologies have provided functional validation of epithelial diversity. Endometrial gland organoids derived from term placental tissue retain epithelial identity, express canonical markers such as E-cadherin and Cytokeratin 7, and respond to estrogen and progesterone by upregulating secretory genes including PAEP and osteopontin (SPP1) (Fig.  6 b) [ 28 ]. These models demonstrate that the hormonal responsiveness observed in vivo is preserved ex vivo. Importantly, they enable the generation of patient-specific organoids whose functional characteristics can be correlated with clinical pregnancy. To further capture multicellular interactions, engineered biomaterial platforms have reconstructed microenvironments that include stromal, endothelial, epithelial, and trophoblast compartments (Fig.  6 c) [ 29 ]. A gelatin-based hydrogel system supported stromal decidualization and angiogenesis, while also enabling quantitative assessment of trophoblast invasion. Fig. 6 Topic-based review of bioengineered platforms for studying endometrial profiling, trophoblast–stromal interactions within the implantation environment, endometrial repair and regeneration. a Single-nucleus RNA-seq of proliferative-phase endometrium identifying major uterine cell populations and validating cell-type classification. Reproduced with permission from Ref [ 27 ] © 2021 Nat Genet. b Immunofluorescence (IF) staining of endometrial gland organoids showing expression of E-cadherin (E-CAD), cytokeratin 7 (KRT7), and laminin (LAMA4). Scale bars, 50 μm Reproduced with permission from Ref [ 28 ] © 2020 Placenta. c GelMA hydrogel system supporting trophoblast spheroid invasion for implantation modeling. Reproduced with permission from Ref [ 29 ] © 2019 Interface Focus. Scale bars, 250 µm. d IF staining of endometrial and decidual organoids showing epithelial identity and proliferative activity. Scale bar, 50 μm. Reproduced with permission from Ref [ 13 ] © 2017 Nat Cell Biol. e Time-course IF analysis of WNT/CTNNB1 signaling during differentiation. Scale bars, 20 μm Reproduced with permission from Ref [ 31 ] © 2018 Stem Cell Reports. f Hormone-responsive endometrial organoids expressing ER, AR, and PR consistent with native tissue. Scale bars: 100 μm (bottom panels) and 20 μm (top panels). Reproduced with permission from Ref [ 32 ] © 2019 J Vis Exp. g Schematic of full-thickness endometrial defect model with layered cell sheet transplantation. Reproduced with permission from Ref [ 33 ] © 2018 Fertil Steril. h 3D bioprinted hiMSC-laden hydrogel scaffolds for endometrial regeneration. Reproduced with permission from Ref [ 35 ] © 2020 Acta Biomater. i 3D bioprinted bilayer endometrial construct restoring tissue architecture and function. Reproduced with permission from Ref [ 36 ] © 2023 Acta Biomater Topic-based review of bioengineered platforms for studying endometrial profiling, trophoblast–stromal interactions within the implantation environment, endometrial repair and regeneration. a Single-nucleus RNA-seq of proliferative-phase endometrium identifying major uterine cell populations and validating cell-type classification. Reproduced with permission from Ref [ 27 ] © 2021 Nat Genet. b Immunofluorescence (IF) staining of endometrial gland organoids showing expression of E-cadherin (E-CAD), cytokeratin 7 (KRT7), and laminin (LAMA4). Scale bars, 50 μm Reproduced with permission from Ref [ 28 ] © 2020 Placenta. c GelMA hydrogel system supporting trophoblast spheroid invasion for implantation modeling. Reproduced with permission from Ref [ 29 ] © 2019 Interface Focus. Scale bars, 250 µm. d IF staining of endometrial and decidual organoids showing epithelial identity and proliferative activity. Scale bar, 50 μm. Reproduced with permission from Ref [ 13 ] © 2017 Nat Cell Biol. e Time-course IF analysis of WNT/CTNNB1 signaling during differentiation. Scale bars, 20 μm Reproduced with permission from Ref [ 31 ] © 2018 Stem Cell Reports. f Hormone-responsive endometrial organoids expressing ER, AR, and PR consistent with native tissue. Scale bars: 100 μm (bottom panels) and 20 μm (top panels). Reproduced with permission from Ref [ 32 ] © 2019 J Vis Exp. g Schematic of full-thickness endometrial defect model with layered cell sheet transplantation. Reproduced with permission from Ref [ 33 ] © 2018 Fertil Steril. h 3D bioprinted hiMSC-laden hydrogel scaffolds for endometrial regeneration. Reproduced with permission from Ref [ 35 ] © 2020 Acta Biomater. i 3D bioprinted bilayer endometrial construct restoring tissue architecture and function. Reproduced with permission from Ref [ 36 ] © 2023 Acta Biomater More recently, microfluidic organ-on-a-chip platforms have further advanced this paradigm by integrating cellular heterogeneity, hormonal responsiveness, and multicellular interactions into a dynamic microenvironment. A patient-derived vascularized endometrium-on-a-chip model reconstructs epithelial, stromal, and endothelial compartments, enabling real-time analysis of angiogenesis and tissue-level responses. Importantly, this system allows quantitative and personalized evaluation of endometrial receptivity by linking molecular signatures with functional outcomes [ 30 ]. Together, these approaches redefine the endometrial profile as a composite of cellular heterogeneity, hormonal responsiveness, and multicellular interaction. By linking in vivo reference maps, patient-derived organoids, and bioengineered models, current research frames the endometrium not as a static tissue, but as a dynamic and clinically relevant system that underlies reproductive success and disease. The order in which these three studies are considered, namely functional hydrogel modeling, patient-specific organoids, and single-cell atlases reflects not only chronological progression but also the logical expansion of endometrial profiling. The hydrogel system established the environmental and functional context, whereas the organoid platform extended profiling to capture patient-specific and clinical variability, and the atlas integrated these dimensions into a universal molecular framework. Taken together, they illustrate that endometrial profiling cannot be reduced to a single methodology. Rather, it is an integrative field that requires linking functional responses, individual variability, and molecular resolution to fully define the receptive endometrium. This complementarity is precisely why these three studies are regarded as the most relevant contributions to the endometrial profiling cluster: each compensates for the limitations of the others, and together they articulate the multidimensionality that makes endometrial profiling indispensable for both basic and translational reproductive research. The endometrial profile provides the framework for an implantation-competent environment, with two elements being particularly critical: the secretory function of the glands and the progesterone responsiveness of stromal cells. First, a significant advancement was the development of long-term, hormone-responsive endometrial organoids (Fig.  6 d) [ 13 ]. These cultures recapitulated glandular morphology, retained genomic stability, and responded to estrogen and progesterone with secretory differentiation. When exposed to pregnancy-associated factors such as hCG and hPL, the organoids produced glycodelin (PAEP) and SPP1, positioning them as the first robust in vitro model to capture the glandular contribution to implantation. Subsequent advances extended this concept by generating multicellular organoids that combine epithelial and stromal populations into self-organizing structures. Within these assemblies, stromal cells produced extracellular matrix components, while epithelial cells polarized and exhibited hormone-dependent secretory activity. Importantly, paracrine reciprocity emerged, whereby stromal signals enhanced epithelial organization, and epithelial cues, in turn, influenced stromal decidualization, mirroring the cooperative architecture of the in vivo endometrium. Parallel work has underscored the indispensability of stromal responsiveness. Beyond glandular function, stromal progesterone responsiveness is equally critical for establishing receptivity (Fig.  6 e) [ 31 ]. iPSC-derived endometrial stromal fibroblasts generated through WNT/CTNNB1-dependent developmental trajectories acquired progesterone receptor expression and undergo decidualization upon hormonal stimulation, faithfully mimicking the transformation required for implantation. Such models not only trace the embryological origins of stromal lineages but also emphasize the centrality of progesterone-sensitive stroma in creating an implantation-competent environment. To capture the dynamic interplay between these compartments, multicellular organoid systems integrate epithelial and stromal populations into self-organizing structures (Fig.  6 f) [ 32 ]. These models retain hormone receptor expression, promote epithelial polarity, and demonstrate paracrine reciprocity, whereby stromal cells influence epithelial secretory activity and organization. This layered architecture mirrors the cooperative nature of the endometrium in vivo, in which epithelial histotrophic support and stromal decidualization jointly sustain embryo implantation. Taken together, these studies illustrate that implantation competence is not a property of glands or stroma in isolation but arises from their coordinated interaction. Organoid models highlight the glandular contribution to histotrophic support, iPSC-derived stromal cells emphasize the hormonal transformation of the stromal compartment, and multicellular constructs integrate both into a self-organizing ecosystem. By weaving these complementary perspectives, recent research reframes the implantation environment as a dynamic, hormone-driven network, providing new experimental platforms for dissecting receptivity and developing translational interventions in infertility. While endometrial profiling and trophoblast–stromal cell to cell interactions define the physiological basis of a receptive state, implantation success is frequently challenged by conditions in which this state is disrupted. Abnormalities such as endometriosis, intrauterine adhesions, or thin endometrium compromise the structural and molecular integrity of the endometrium, leading to impaired receptivity. In this context, endometrial repair and regeneration emerge as a critical research focus, aiming to restore a functional environment capable of supporting embryo implantation. Research on endometrial regeneration has evolved from preliminary feasibility assessments to sophisticated approaches that incorporate biomaterials and stem cell-based biofabrication techniques. A significant advancement in this field was achieved through the application of cell sheet transplantation, in which stratified epithelial and stromal cell sheets were successfully engrafted into a rat model of uterine injury. These constructs facilitated the restoration of glandular and stromal architecture and notably enabled successful fertilization and pregnancy. This study provided compelling evidence that regenerative strategies can restore not only the structural integrity but also the functional receptivity of the endometrium (Fig.  6 g) [ 33 ]. Progress in the field was further propelled by the implementation of 3D bioprinted hydrogel scaffolds incorporating mesenchymal stem cells derived from human iPSCs. Porous alginate–gelatin matrices were engineered to establish a supportive microenvironment that promoted cell viability and facilitated tissue regeneration. Upon implantation into rat models of endometrial injury, these constructs enhanced the regeneration of stromal, epithelial, and endothelial compartments, while concurrently improving receptivity-related markers, including pinopode formation, leukemia inhibitory factor (LIF), and integrin αvβ3 expression. Recent advances in nanomaterial-reinforced bioinks have further expanded the mechanical and functional capabilities of 3D bioprinted scaffolds, offering new opportunities for constructing biomimetic reproductive tissue constructs with improved structural fidelity [ 34 ]. Compared with cell-only or scaffold-only controls, the integrated cell–scaffold constructs yielded superior regenerative outcomes and partially restored implantation competence, underscoring the synergistic advantage of a biofabrication-based combinatorial strategy (Fig.  6 h) [ 35 ]. More recently, bilayered 3D bioprinted constructs have been engineered to more precisely recapitulate the native histoarchitecture of the endometrium. Utilizing alginate–hyaluronic acid hydrogels embedded with primary epithelial and stromal cells, a biomimetic structure was fabricated consisting of a compact epithelial monolayer atop a porous stromal compartment. Upon implantation into a rat model of partial full-thickness uterine excision, these constructs successfully restored the layered architecture characteristic of the native tissue, including the luminal epithelium, endometrial glands, stroma, vasculature, and smooth muscle. Notably, reproductive function was markedly improved, with pregnancy rates reaching 75% compared to 12.5% in non-printed controls, offering compelling evidence that architectural fidelity can translate into functional restoration (Fig.  6 i) [ 36 ]. Across these three domains, a consistent theme emerges: endometrial receptivity is multidimensional, integrating molecular profiles, cell–cell interactions, and regenerative capacity. Functional hydrogels provide environmental context, patient-specific organoids capture clinical variability, and single-cell atlases supply molecular resolution. None alone is sufficient together, they articulate the full scope of endometrial biology. This integrated approach reframes the endometrium as a dynamic, hormone-driven system whose integrity is essential for implantation and whose restoration offers a promising route to treating infertility. The placenta is a transient organ that develops during pregnancy and is expelled after childbirth [ 37 ]. Nevertheless, its role during pregnancy is critical, influencing both maternal and fetal health through diverse functions. One of its principal functions is to mediate the exchange of gases, nutrients, and waste products between the mother and the fetus. It simultaneously establishes a specialized environment that maintains optimal conditions for fetal growth and development. In addition, the placenta acts as a barrier formed by a multilayered interface of trophoblasts, fetal capillary endothelium, and intervening extracellular matrix [ 38 ]. This barrier performs the dual role of facilitating the efficient transfer of beneficial molecules while restricting the passage of potentially harmful xenobiotics and pathogens. Recent advances in high-resolution imaging, omics profiling, and microphysiological modeling have revealed the dynamic cellular composition, transport mechanisms, and remodeling of the placental barrier across gestation [ 39 ]. Moreover, the maternal–fetal barrier integrates immune tolerance with protective functions, ensuring maternal acceptance of the semi-allogeneic fetus. Among the processes that shape this barrier, deep trophoblast invasion represents a defining feature of human placentation (Fig.  7 a) [ 38 ]. Deep trophoblast invasion is a hallmark of human placentation, defined by the migration of extravillous trophoblasts into the maternal decidua and their penetration of spiral arteries [ 40 ]. By degrading the extracellular matrix, recruiting maternal immune cells, and remodeling uterine vessels, these cells convert spiral arteries into high-capacity, low-resistance conduits capable of sustaining the demands of pregnancy. This process is tightly regulated, as insufficient invasion and poor vascular remodeling contribute to disorders such as preeclampsia and fetal growth restriction, whereas excessive or uncontrolled invasion leads to placenta accreta spectrum, a major cause of severe maternal morbidity [ 41 ]. Elucidating the cellular and molecular mechanisms underlying this invasive behavior remains critical for improving maternal and fetal outcomes. This invasive process is unique to human pregnancy, underscoring the evolutionary specialization of placental development compared with other mammals. Fig. 7 Topic-based review of bioengineered platforms for studying trophoblast invasion and the maternal–fetal barrier. a Schematic of early human implantation and placentation [ 38 ]. b Schematic of placental terminal villi showing maternal–fetal exchange across syncytiotrophoblast and fetal endothelium. Reproduced with permission from Ref [ 44 ] © 2019 Adv Sci. c Microengineered placenta-on-a-chip platform consisting of two parallel microchannels separated by a semipermeable membrane, where trophoblasts are cultured on the apical side and villous endothelial cells on the basal side to reconstitute the placental barrier. Scale bars, 500 μm Reproduced with permission from Ref [ 45 ] © 2019 Adv Sci. d Schematic workflow for generating human trophoblast stem cell (hTSC)-derived organoids within Matrigel droplets and brightfield images showing progressive organoid growth over six days. Reproduced with permission from Ref [ 50 ] © 2023 Nat Commun. e Representative immunofluorescence images of hTSC-derived organoids showing expression of trophoblast lineage markers, including GATA3 and CK7 (pan-trophoblast), Ki67 and CDH1 (proliferative cytotrophoblasts), and CGB (syncytiotrophoblasts). Nuclei are counterstained with DAPI, and dashed lines outline multinucleated syncytiotrophoblast regions. Each marker set was analyzed in separate organoids. Scale bars, 50 μm. Reproduced with permission from Ref © 2023 Nat Commun. f TGF-β signaling regulates EVT motility in villous explants. Inhibition of TGF-β (A8301) enhances radial outgrowth and induces a migratory phenotype, as shown by F-actin reorganization. Scale bars: 1000 μm (left panels) and 50 μm (magnified and fluorescence panels). Reproduced with permission from Ref [ 50 ] © 2022 PNAS Topic-based review of bioengineered platforms for studying trophoblast invasion and the maternal–fetal barrier. a Schematic of early human implantation and placentation [ 38 ]. b Schematic of placental terminal villi showing maternal–fetal exchange across syncytiotrophoblast and fetal endothelium. Reproduced with permission from Ref [ 44 ] © 2019 Adv Sci. c Microengineered placenta-on-a-chip platform consisting of two parallel microchannels separated by a semipermeable membrane, where trophoblasts are cultured on the apical side and villous endothelial cells on the basal side to reconstitute the placental barrier. Scale bars, 500 μm Reproduced with permission from Ref [ 45 ] © 2019 Adv Sci. d Schematic workflow for generating human trophoblast stem cell (hTSC)-derived organoids within Matrigel droplets and brightfield images showing progressive organoid growth over six days. Reproduced with permission from Ref [ 50 ] © 2023 Nat Commun. e Representative immunofluorescence images of hTSC-derived organoids showing expression of trophoblast lineage markers, including GATA3 and CK7 (pan-trophoblast), Ki67 and CDH1 (proliferative cytotrophoblasts), and CGB (syncytiotrophoblasts). Nuclei are counterstained with DAPI, and dashed lines outline multinucleated syncytiotrophoblast regions. Each marker set was analyzed in separate organoids. Scale bars, 50 μm. Reproduced with permission from Ref © 2023 Nat Commun. f TGF-β signaling regulates EVT motility in villous explants. Inhibition of TGF-β (A8301) enhances radial outgrowth and induces a migratory phenotype, as shown by F-actin reorganization. Scale bars: 1000 μm (left panels) and 50 μm (magnified and fluorescence panels). Reproduced with permission from Ref [ 50 ] © 2022 PNAS Despite its protective function, the placenta does not constitute an absolute barrier. A wide range of pathogens, including viruses such as cytomegalovirus, Zika virus, and SARS-CoV-2, as well as bacterial and parasitic agents like Listeria monocytogenes and Plasmodium species, can traverse the placental interface [ 40 , 42 ]. These infections reveal the inherent vulnerability of this organ and may provoke inflammation, disrupt trophoblast activity, and impair nutrient transfer, ultimately resulting in miscarriage, preterm birth, stillbirth, growth restriction, or long-term neurodevelopmental sequelae. These outcomes underscore the need to elucidate the mechanisms by which pathogens cross the placental barrier and the ways in which host immune responses regulate this process. Understanding these interactions is essential not only for preventing vertical transmission but also for developing therapeutic strategies that safeguard both maternal and fetal health. Topics: placental barrier, placental invasion, infectious placenta Human placentation is characterized by deep invasion of extravillous trophoblasts into the maternal decidua and spiral arteries, a process that remodels uterine vessels into high-capacity, low-resistance channels essential for sustaining fetal growth. This process not only establishes the maternal–fetal circulation but also gives rise to the placental barrier, a multilayered interface formed by syncytiotrophoblasts, cytotrophoblasts, and fetal capillary endothelium [ 41 ]. The barrier facilitates the bidirectional exchange of gases, nutrients, and waste products, while restricting the passage of harmful substances and pathogens. However, despite its protective architecture, the placenta remains vulnerable to infectious agents. Viruses, bacteria, and parasites can exploit structural or immunological weaknesses, leading to infectious pathologies that disrupt trophoblast function, compromise nutrient exchange, and result in adverse pregnancy outcomes such as miscarriage, preterm birth, or congenital disorders. A hallmark of human placentation is the invasion of extravillous trophoblasts (EVTs) into the maternal decidua and spiral arteries [ 43 ]. Proliferative cytotrophoblasts differentiate into invasive EVTs that migrate through the uterine stroma and replace the endothelial lining of spiral arteries, transforming them into high-capacity, low-resistance vessels that sustain placental perfusion. This invasive behavior is regulated by a balance between proteolytic and inhibitory factors. EVTs secrete matrix metalloproteinases (MMPs), urokinase plasminogen activator (uPA), and cathepsins to degrade the decidual extracellular matrix and promote migration, while tissue inhibitors of metalloproteinases (TIMPs) and plasminogen activator inhibitors (PAIs) produced by trophoblasts and decidual cells restrain the extent of invasion. Cytokines, chemokines, and angiogenic factors further modulate this process by upregulating MMP-2 and MMP-9 expression and enhancing trophoblast motility. Dysregulation of trophoblast invasion is closely linked to pregnancy complications, with insufficient invasion contributing to preeclampsia and fetal growth restriction, and excessive invasion leading to placenta accreta spectrum and gestational trophoblastic disease. Aberrant remodeling of the maternal vasculature is a hallmark of preeclampsia, and growing evidence indicates that perivascular cells are central to this process. Under physiological conditions, the VEGF–Angiopoietin–Tie2 axis orchestrates endothelial–pericyte signaling, ensuring vessel stability, controlled permeability, and the successful adaptation of spiral arteries into high-capacity, low-resistance conduits (Fig.  7 b) [ 44 ]. This balance establishes the conditions necessary environment for trophoblast invasion and sustained maternal–fetal perfusion. However, disruption of this signaling network alters pericyte behavior, leading to excessive secretion of VEGF-A, destabilization of Tie2-mediated pathways, and a cascade of impaired angiogenesis, endothelial dysfunction, and inflammatory activation within the placental microvasculature. These disturbances hinder the establishment of adequate vascular remodeling, creating a pathological state in which blood flow to the intervillous space is restricted. The resulting hypoxic stress and maternal endothelial activation form the biological basis of the hypertensive, inflammatory, and multi-organ manifestations of preeclampsia. This conceptual framework underscores the importance of cross-talk between trophoblasts, endothelial cells, and pericytes, positioning the regulation of vascular invasion not only as a developmental necessity, but also as a determinant of maternal and fetal health outcomes. Recent advances in bioengineering have led to the development of dynamic placenta models that better replicate trophoblast–endothelium interactions. By combining 3D bioprinting with perfusion bioreactor systems, these models can reproduce angiogenic signaling, endothelial apoptosis, and trophoblast invasion under physiologically relevant shear stress conditions (Fig.  7 c) [ 45 ]. Such platforms not only overcome the limitations of static two-dimensional cultures but also provide mechanistic insights into how aberrant trophoblast–endothelium crosstalk contributes to defective vascular remodeling. Consequently, they represent an emerging approach to experimentally model preeclampsia, enabling deeper investigation into its pathogenesis and potential therapeutic interventions. The placenta is a temporal organ that develops only during pregnancy and functions as a site of nutrient and gas exchange, as well as hormone production. A representative advance in modeling the maternal–fetal interface is the development of placenta-on-a-chip platforms that recreate the structural and functional features of the placental barrier [ 46 ]. By co-culturing trophoblasts and fetal endothelial cells in opposing channels, these systems establish a polarized architecture resembling the in vivo barrier, including syncytiotrophoblast-like layers, microvillus formation, and junctional protein expression. Functional validation through glucose transport assays demonstrated selective permeability and highlighted clear improvements over conventional transwell models. Moreover, permeability assessments under stress conditions revealed how pathological or external stimuli could compromise barrier integrity. Such microphysiological models overcome the limitations of static cultures and animal studies, providing powerful tools to investigate placental barrier breakdown in disease, mechanisms of vertical transmission, and drug transfer across the maternal–fetal interface. Building on advances in placenta-on-a-chip technology, microengineered models have now progressed to recapitulate not only barrier structure but also active transport functions critical for fetal protection [ 45 ]. The system described integrates trophoblasts and placental endothelial cells under flow, allowing the formation of continuous junctions, microvilli development, and physiologic transepithelial resistance [ 47 ]. Importantly, the model reproduced transporter-mediated drug efflux, exemplified by BCRP-dependent exclusion of glyburide, thereby mimicking a key protective mechanism of the human placental barrier. This proof-of-principle demonstrates how such platforms can move beyond structural mimicry toward functional assays that quantify solute flux, efflux transporter activity, and barrier disruption under pharmacological or pathological conditions. As a result, they provide a versatile foundation for deeper investigation into maternal–fetal drug transfer, xenobiotic exposure, and disease-related barrier dysfunction, enabling translational insights not achievable with conventional animal or static culture models. During pregnancy, infections pose clinically significant risks to both the mother and fetus, depending on the pathogen, gestational timing, site of infection, and disease severity. Zika virus provides a well-characterized example of this vulnerability, preferentially targeting early trophoblast populations, including human trophoblast stem cells, which exhibit relatively permissive antiviral states. While receptors such as AXL and TIM-1 have been associated with viral entry, their roles remain context-dependent. Notably, viral replication is more efficient prior to differentiation into cytotrophoblasts, syncytiotrophoblasts, or extravillous trophoblasts (Fig.  7 d) [ 48 ]. The integration of nanobiosensors into organ-on-chip systems represents a promising approach for real-time monitoring of viral infection dynamics at the placental interface, offering new avenues for detecting and tracking pathogen-induced barrier disruption [ 49 ]. In trophoblast organoid models, Zika virus exposure disrupts villous architecture and impairs syncytialization, while single-cell transcriptomic analyses reveal loss of stemness, reduced cytotrophoblast proliferation, and transcriptional changes resembling pathological states such as preeclampsia. Immunofluorescent staining confirms trophoblast lineage-specific marker expression ( Fig.  7 e), with each marker set analyzed in separate organoids to avoid signal overlap and ensure accurate lineage-specific characterization. These alterations compromise the formation and maintenance of the syncytiotrophoblast layer, which is essential for maternofetal exchange and pathogen restriction. Collectively, these findings suggest that infection-induced dysfunction of trophoblast differentiation undermines barrier integrity, rather than merely enabling passive translocation across an otherwise intact interface [ 48 ]. These observations highlight the importance of regulatory signaling pathways that govern trophoblast differentiation and barrier function. TGF-β signaling provides a mechanistic bridge between trophoblast behavior, vascular remodeling, and barrier performance. Loss of WNT followed by activation of TGF-β–SMAD3 drives the transition from placental EVTs toward decidual EVTs with a secretory program, including DAO and PAPPA2, while restraining motility and promoting maturation [ 50 ]. This sequence supports arterial plugging and subsequent low-resistance remodeling, and it calibrates invasion depth through decidual cues. When this axis is mistimed or blunted, EVTs remain immature, remodeling is shallow, inflow becomes inefficient, and barrier stability degrades, which aligns with early-onset preeclampsia. Conversely, excessive TGF-β activity may over-suppress migration and skew EVTs toward secretion that perturbs local immunovascular tone. Taken together with dynamic trophoblast–endothelium models and infection studies that disrupt trophoblast differentiation, these data position TGF-β–guided EVT maturation as a central lever that links vascular change to maternal–fetal barrier integrity across both hypertensive and infectious placental pathology. The oviduct, or fallopian tube, is a critical component of the female reproductive system. This tubular structure plays a multifaceted role in female reproduction by mediating the directional transport of gametes, facilitating fertilization, and creating a physiologically supportive environment for early embryo development [ 51 ]. Specifically, after ovulation, the fimbrial end of the oviduct captures the cumulus–oocyte complex and guides it into the oviductal lumen through ciliary beating and fluid flow. Concurrently, sperm that have migrated through the female reproductive tract enter the isthmic region of the oviduct, where they undergo capacitation and are temporarily stored before being released in a time-regulated manner to reach the oocyte [ 52 ]. Fertilization typically occurs in the ampullary region, where the oocyte and capacitated sperm meet. Following fertilization, the oviductal epithelium provides biochemical and mechanical cues through secreted factors, extracellular vesicles, and dynamic fluid movement that support early zygotic development and maintain embryonic viability during transit toward the uterus [ 53 ]. Among its anatomically distinct regions, the fimbriae epithelium located at the distal end exhibits highly specialized structural and functional characteristics. This region captures the ovulated oocyte and directs it into the oviductal lumen through the coordinated action of motile cilia and secretory epithelial cells [ 54 ]. Impairments in fimbrial function, which may arise from anatomical abnormalities, inflammatory responses, or molecular defects, have been associated with reduced fertility and increased risk of ectopic pregnancy. These observations underscore the essential and non-redundant contribution of the fimbriae to reproductive success [ 55 ]. In addition to its mechanical roles, the oviductal epithelium functions as an active secretory interface that mediates biochemical communication within the reproductive tract. Epithelial cells in this tissue release extracellular vesicles (EVs), including both exosomes and microvesicles. These vesicles are enclosed by a lipid bilayer surrounding a molecular cargo composed of proteins, lipids, messenger RNAs, and small non-coding RNAs. Numerous studies have demonstrated that oviduct-derived EVs (OEVs) participate in key reproductive events. These vesicles have been shown to regulate sperm capacitation, preserve oocyte quality, enhance zygotic development, and support early embryonic competence. Experimental evidence further indicates that the addition of EVs to in vitro embryo culture systems improves blastocyst formation, promotes trophectoderm development, and upregulates genes associated with implantation success [ 56 ]. Emerging data suggest that the composition and functional properties of OEVs vary significantly along different regions of the oviduct. The fimbrial epithelium, in particular, may produce vesicles with unique molecular profiles and phase-specific functions during the peri-ovulatory window. These fimbrial-derived vesicles are hypothesized to influence early reproductive processes, including sperm–oocyte binding, zona pellucida remodeling, and zygote stabilization prior to entry into the ampullary region [ 57 ]. Despite growing interest in OEV biology, the cellular mechanisms governing EV release from the fimbrial epithelium, the precise bioactive cargo of these vesicles, and their downstream effects on gametes and embryos remain poorly defined. Topics: fimbriae epithelium, oviduct EVs Current topic mapping of fallopian tube biology and epithelial modeling identifies the fimbrial epithelium as a central hub for research on gamete interaction, fertilization, and the earliest epithelial changes leading to disease. Located at the distal end of the oviduct, the fimbria is the first anatomical site where ovulated oocytes and incoming sperm converge. Its ciliated and secretory epithelial cells capture the oocyte, modulate sperm behavior, and generate a microenvironment that supports fertilization and early embryo development. Beyond these structural and physical dynamics, vesicle-mediated signaling further refines this microenvironment, a concept that has gained increasing attention in recent years (Fig.  8 a) [ 48 ]. Fig. 8 Topic-based review of bioengineered platforms for studying fimbriae epithelium, oviduct EVs. a Cross-sections of porcine oviductal epithelial cells cultured at an air–liquid interface for 1, 3, and 6 weeks, showing progressive development of polarized, columnar morphology with cilia. Hemalum/eosin staining. Scale bar, 20 μm. Reproduced with permission from Ref [ 53 ] © 2023 Nat Commun. b Morphology of primary porcine oviduct epithelial cells (POECs) in long-term culture compared with native oviduct tissue. Hematoxylin/eosin staining. Scale bar, 20 μm. Reproduced with permission from Ref [ 64 ] © 2013 Theriogenology. c Immunohistochemical analysis of extracellular matrix and cytoskeletal markers in 2D- and 3D-cultured fallopian tube secretory epithelial cells (FTSECs) compared with primary fallopian tube. Scale bars, 100 μm. Reproduced with permission from Ref [ 65 ] © 2013 BMC Cell Biol. d Representative brightfield images of mouse fallopian tube epithelial (FTE) organoids showing progressive growth and maintenance from day 1 to 6 months. Scale bars, 100 μm. Reproduced with permission from Ref [ 66 ] © 2018 Stem Cell Res. e Phase-contrast images of fallopian tube epithelial organoids cultured with varying concentrations of Wnt and RSPO1. Scale bar, 1000 μm. Reproduced with permission from Ref [ 67 ] © 2015 Nat Commun. f Transmission electron microscopy images of bovine oviduct-derived extracellular vesicles from in vivo flushing and in vitro BOEC culture medium, showing exosome-like vesicles (blue arrows, 30–100 nm) and microvesicle-like structures (red arrows, >100 nm). Scale bars, 200 nm. Reproduced with permission from Ref [ 69 ] © 2017 Reproduction Topic-based review of bioengineered platforms for studying fimbriae epithelium, oviduct EVs. a Cross-sections of porcine oviductal epithelial cells cultured at an air–liquid interface for 1, 3, and 6 weeks, showing progressive development of polarized, columnar morphology with cilia. Hemalum/eosin staining. Scale bar, 20 μm. Reproduced with permission from Ref [ 53 ] © 2023 Nat Commun. b Morphology of primary porcine oviduct epithelial cells (POECs) in long-term culture compared with native oviduct tissue. Hematoxylin/eosin staining. Scale bar, 20 μm. Reproduced with permission from Ref [ 64 ] © 2013 Theriogenology. c Immunohistochemical analysis of extracellular matrix and cytoskeletal markers in 2D- and 3D-cultured fallopian tube secretory epithelial cells (FTSECs) compared with primary fallopian tube. Scale bars, 100 μm. Reproduced with permission from Ref [ 65 ] © 2013 BMC Cell Biol. d Representative brightfield images of mouse fallopian tube epithelial (FTE) organoids showing progressive growth and maintenance from day 1 to 6 months. Scale bars, 100 μm. Reproduced with permission from Ref [ 66 ] © 2018 Stem Cell Res. e Phase-contrast images of fallopian tube epithelial organoids cultured with varying concentrations of Wnt and RSPO1. Scale bar, 1000 μm. Reproduced with permission from Ref [ 67 ] © 2015 Nat Commun. f Transmission electron microscopy images of bovine oviduct-derived extracellular vesicles from in vivo flushing and in vitro BOEC culture medium, showing exosome-like vesicles (blue arrows, 30–100 nm) and microvesicle-like structures (red arrows, >100 nm). Scale bars, 200 nm. Reproduced with permission from Ref [ 69 ] © 2017 Reproduction To mechanistically dissect these complex biological processes, advances in oviduct epithelial modeling have enabled the development of long-term, polarized culture systems using primary porcine oviduct epithelial cells (POEC). Under optimized air–liquid interface conditions, these cells acquire distinct polarity and functional ciliation after three weeks and maintain this differentiated state for up to six weeks, preserving hormone responsiveness as indicated by increased OVGP1 and PGR expression [ 58 ]. Complementary standardized validation protocols across multiple donors have further enhanced reproducibility by systematically evaluating medium conditioning, serum supplementation, culture duration, and cryopreservation effects. Together, these approaches provide a physiologically relevant in vitro platform that closely recapitulates in vivo oviduct epithelial morphology and function (Fig.  8 b) [ 59 ]. While initially developed to study reproductive physiology, three-dimensional epithelial models have also provided critical insights into early epithelial transformation. Primary spheroids of fallopian tube secretory epithelial cells revealed more than 1,000 transcriptomic differences between 2D and 3D cultures, including reduced cell-cycle activity and a shift toward a luteal-phase-like gene expression profile. By restoring phase-specific differentiation and extracellular matrix deposition, this model recreated key features of the presumed cell of origin for high-grade serous ovarian carcinoma (Fig.  8 c) [ 60 ]. Recently, 3D organoid systems derived from mouse oviduct epithelial cells have extended these findings by enabling long-term culture under defined signaling conditions, revealing the essential roles of Notch and Wnt pathways in organoid growth and confirming that stem cells with organoid-forming potential are enriched in the distal fimbrial region. This complementary model not only supports the notion of the fimbrial epithelium as a cellular origin for epithelial ovarian cancer but also provides a powerful platform for modeling early disease stages and performing gene or drug screening (Fig.  8 d) [ 61 ]. Futhermore, long-term 3D organoid cultures from human fallopian tubes have confirmed the presence of adult stem cells in the fimbrial epithelium. Single epithelial cells generate organoids containing both ciliated and secretory cells, faithfully recapitulating the mucosal folds, epithelial polarity, and hormone responsiveness of native tissue. These organoids rely on coordinated Wnt and Notch signaling to maintain stemness and respond to estradiol and progesterone. As such, they offer a robust in vitro model of the hormonally dynamic fimbrial environment and a powerful platform to study epithelial renewal, early transformation events, and the origins of high-grade serous ovarian carcinoma (Fig.  8 e) [ 62 ]. In parallel with structural and cellular modeling advances, increasing evidence underscores the pivotal role of OEVs as molecular mediators of oviduct–embryo communication [ 63 ]. OEVs comprise a heterogeneous population of membrane-bound particles, ranging from small exosomes (30–100 nm) to larger microvesicles (> 100 nm) [ 64 ]. Proteomic analyses reveal distinct signatures between in vivo and in vitro sources, with in vivo OEVs enriched in reproductive proteins such as OVGP1, CD109, and HSP90, whereas in vitro OEVs prominently contain lactadherin. Functionally, labeled in vivo OEVs can traverse the zona pellucida and be internalized by embryos, enhancing blastocyst formation, hatching rates, and total cell numbers, particularly in cryopreserved embryos (Fig.  8 f) [ 64 ]. Collectively, these findings redefine the fimbrial epithelium not as a passive conduct but as an active regulator of early reproductive events. Integrating epithelial organoid systems, standardized polarized cultures, and vesicle-mediated molecular signaling establishes a multidimensional framework for investigating fertilization, embryo–maternal communication, and early epithelial transformation. Together, these platforms offer translational potential for improving assisted reproductive technologies. The ovary is the primary female gonad responsible for producing oocytes and steroid hormones. It is composed of an outer cortex and an inner medulla. The cortex contains connective tissue stroma and follicles at various stages of development, while the medulla is composed of loose connective tissue with abundant blood vessels and nerve fibers that support ovarian function [ 65 ]. Within this structural framework, the oocyte interacts with surrounding granulosa cells, theca cells, and stromal cells. Granulosa cells nurture the oocyte and produce estrogens. Theca cells synthesize androgenic precursors and promote vascularization. Stromal cells provide structural support and secrete extracellular matrix that shapes the follicular microenvironment [ 66 ]. Folliculogenesis begins during fetal and early postnatal life, when primordial follicles form the ovarian reserve. Each primordial follicle contains an oocyte arrested in prophase I and surrounded by flattened pre-granulosa cells. Over time, a subset of these follicles is recruited into the growing pool. The transition to a primary follicle is marked by the transformation of granulosa cells from flattened to cuboidal and the onset of oocyte growth [ 67 ]. Secondary follicles develop multiple layers of granulosa cells and acquire an organized theca cell layer, which together supports further oocyte maturation. These early stages are largely independent of circulating gonadotropins but require extensive bidirectional communication between the oocyte and somatic cells to coordinate growth, differentiation, and survival [ 68 ]. As development progresses, a fluid-filled antral cavity forms within the follicle, signaling the onset of the gonadotropin-dependent phase. At this stage, follicle-stimulating hormone drives granulosa cell proliferation, estrogen production, and luteinizing hormone receptor expression, while luteinizing hormone stimulates theca cells to produce androgens that serve as substrates for estrogen synthesis. Most antral follicles undergo atresia, but a small fraction reaches dominance, producing higher levels of estrogen and acquiring increased sensitivity to gonadotropins. These dominant follicles respond to the preovulatory luteinizing hormone surge, which induces completion of meiosis I, cumulus expansion, and ovulation. After ovulation, the residual follicular cells differentiate into the corpus luteum, a temporary endocrine structure that secretes progesterone essential for implantation and the maintenance of early pregnancy. Throughout these processes, ovarian function is tightly regulated by the hypothalamic–pituitary–ovarian axis, which coordinates systemic hormone levels with local intraovarian signals to maintain female reproductive capacity [ 69 ]. In addition to its role in gametogenesis, the ovary serves as a highly active endocrine organ that orchestrates female reproductive physiology and broader systemic homeostasis. It produces steroid hormones such as estrogens, progesterone and androgens, as well as peptide factors including inhibin, activin and anti-Müllerian hormone (AMH), which collectively regulate follicular development, the hypothalamic–pituitary–ovarian axis and the timing of ovulation [ 70 ]. Ovarian hormones not only modulate the endometrium and prepare the uterus for implantation but also exert significant effects on secondary sexual characteristics, bone metabolism, cardiovascular integrity and immune function [ 71 ]. Cyclical fluctuations in these secretions integrate intraovarian signaling with systemic endocrine feedback loops to ensure reproductive competence. With advancing age or pathological conditions such as polycystic ovary syndrome and premature ovarian insufficiency, alterations in the ovarian endocrine milieu disrupt both fertility and metabolic–hormonal homeostasis, underscoring the pivotal role of the ovary as an endocrine hub in female physiology [ 72 ]. Topics: follicle development, endocrine signaling Across studies that either recapitulate the ovarian niche in vitro or implant matrix-based grafts in vivo, a consistent message emerges instructive control of follicle development depends on coupling endocrine cues with an appropriate extracellular matrix (ECM) context. Early 3D culture work showed that encapsulating granulosa cell–oocyte complexes in alginate preserves follicular architecture, supports granulosa proliferation and oocyte growth, and yields oocytes competent to resume meiosis—evidence that geometric confinement and maintained cell–cell contacts are foundational to competency acquisition (Fig.  9 a) [ 73 ]. Similarly, building on this, a basement-membrane-like ECM is not merely supportive but synergistic with endocrine factors: adding activin-A to a 3D Matrigel environment increased growth and survival and even produced antral-like structures by day 5, whereas activin-A on single-component ECMs or 2D plastic failed to sustain growth beyond ~ 3–5 days, underscoring the necessity of an integrin-engaging, 3D matrix to render follicles responsive to mitogenic cues (Fig.  9 b) [ 74 ]. The principle that biomimetic microenvironments improve reproductive outcomes also appears outside the ovary proper: an oviduct-inspired microfluidic channel reduced polyspermy during porcine IVF while preserving cleavage and blastocyst formation. This implies that physiologic flow and spatial constraint can tune gamete–gamete interactions just as ECM and paracrine fields tune somatic–germ cell crosstalk (Fig.  9 c) [ 75 ]. Turning to implantation strategies, a synthetic “artificial ovary” built from PEG-VS hydrogel modified with RGD and MMP-cleavable crosslinks supported in vivo folliculogenesis to antral follicles and corpora lutea, normalized the post-ovariectomy rise in FSH for 60 days, and became vascularized showing that a degradable, cell-adhesive matrix can restore cyclical endocrine function (Fig.  9 d) [ 76 ]. In a recent study using macroporous alginate scaffolds, affinity-bound BMP-4 prolonged local morphogen presentation and enlarged developing follicles and pointing to matrix-tethered growth factors as potent levers for stage-appropriate maturation. The same constructs produced estradiol at levels characteristic of antral follicles and, after transplantation with VEGF/PDGF-BB, exhibited neovascularization aligning endocrine readouts with histologic evidence of graft integration (Fig.  9 e) [ 77 ]. Finally, an organ-derived route, decellularized ovarian scaffold recellularization, retained native ECM ultrastructure, supported estradiol production in vitro, and, upon transplantation, initiated puberty in ovariectomized mice, demonstrating that preserved ovarian architecture alone can instruct both steroidogenesis and systemic milestones (Fig.  9 f) [ 78 ]. Taken together, these lines of evidence converge on a coherent view: whether by emulating niche mechanics and transport in vitro or by implanting ECMs that present adhesive ligands, proteolytic cues, and bound morphogens in vivo, follicle development is steered most effectively when endocrine signaling is delivered within a correctly patterned microenvironment. This rationale that directly justifies focusing on follicle development and endocrine signaling as the central design variables for ovarian bioengineering. Importantly, recent single-cell transcriptomic studies have revealed that follicle development is not governed by uniform endocrine responsiveness, but rather by marked cellular heterogeneity across follicular compartments and developmental stages. High-resolution mapping of the human ovarian cortex has identified distinct populations of oocytes, granulosa cells, endothelial cells, immune cells, and stromal cells, all of which dynamically interact to regulate folliculogenesis [ 79 ]. Moreover, transcriptional profiling of individual follicles has demonstrated that extracellular matrix remodeling, theca cell signaling, and immune-related pathways vary significantly depending on follicle stage, indicating that morphology alone is insufficient to define follicular competence [ 80 ]. These findings suggest that endocrine signals such as FSH and LH are interpreted in a stage-specific and context-dependent manner, reinforcing the need to incorporate cellular heterogeneity into engineered ovarian models. Fig. 9 Topic-based review of bioengineered platforms for studying follicle development, endocrine signaling a Encapsulation of mouse granulosa–oocyte complexes (GOCs) in alginate beads. Scale bars: 100 μm (left panels) and 500 μm (right panels). Reproduced with permission from Ref. [ 73 ] © 2003 Tissue Eng. b Follicle morphology under different culture conditions, with antral formation observed in Matrigel + activin-A. Scale bars, 100 μm Reproduced with permission from Ref [ 74 ] © 2007 Reprod Sci. c Comparison of in vivo oviductal transport and a microfluidic device mimicking sperm guidance. Reproduced with permission from Ref [ 75 ] © 2005 Lab Chip. d Viability of primordial follicles in alginate scaffolds assessed by live/dead staining. Scale bars: 50 μm (top left and bottom left), 25 μm (top right), and 10 μm (bottom right). Reproduced with permission from Ref [ 76 ] © 2016 NPJ Regen Med. e Long-term folliculogenesis in PEG grafts showing follicle maturation and tissue expansion. Scale bars: 50 μm (top panels) and 100 μm (bottom panels). Reproduced with permission from Ref [ 77 ] © 2019 Biomaterials. f H&E staining of human ovarian tissue before and after decellularization, confirming ECM preservation. Reproduced with permission from Ref [ 78 ] © 2015 Biomaterials. g Microfluidic platforms enabling long-term culture and hormonal crosstalk of reproductive tissues. Reproduced with permission from Ref [ 82 ] © 2017 Nat Commun. h Microfluidic generation of epiblast-like structures from human pluripotent stem cells. Scale bars, 40 µm. Reproduced with permission from Ref [ 83 ] © 2019 Nature Topic-based review of bioengineered platforms for studying follicle development, endocrine signaling a Encapsulation of mouse granulosa–oocyte complexes (GOCs) in alginate beads. Scale bars: 100 μm (left panels) and 500 μm (right panels). Reproduced with permission from Ref. [ 73 ] © 2003 Tissue Eng. b Follicle morphology under different culture conditions, with antral formation observed in Matrigel + activin-A. Scale bars, 100 μm Reproduced with permission from Ref [ 74 ] © 2007 Reprod Sci. c Comparison of in vivo oviductal transport and a microfluidic device mimicking sperm guidance. Reproduced with permission from Ref [ 75 ] © 2005 Lab Chip. d Viability of primordial follicles in alginate scaffolds assessed by live/dead staining. Scale bars: 50 μm (top left and bottom left), 25 μm (top right), and 10 μm (bottom right). Reproduced with permission from Ref [ 76 ] © 2016 NPJ Regen Med. e Long-term folliculogenesis in PEG grafts showing follicle maturation and tissue expansion. Scale bars: 50 μm (top panels) and 100 μm (bottom panels). Reproduced with permission from Ref [ 77 ] © 2019 Biomaterials. f H&E staining of human ovarian tissue before and after decellularization, confirming ECM preservation. Reproduced with permission from Ref [ 78 ] © 2015 Biomaterials. g Microfluidic platforms enabling long-term culture and hormonal crosstalk of reproductive tissues. Reproduced with permission from Ref [ 82 ] © 2017 Nat Commun. h Microfluidic generation of epiblast-like structures from human pluripotent stem cells. Scale bars, 40 µm. Reproduced with permission from Ref [ 83 ] © 2019 Nature Endocrine signaling is effective not only because of “how much,” but because of when, where, and how signals reach their targets. Timing, compartmentalization, and transport determine whether tissues read the same hormone as a cue to grow, differentiate, or rest. This concept is further supported by recent systems-level and single-cell studies demonstrating that hormone responsiveness is tightly linked to cell-type composition, intercellular signaling networks, and extracellular matrix context within the ovary [ 81 ]. A clear example comes from a recirculating, multi-organ microfluidic system that links ovary, fallopian tube, endometrium, ectocervix, and liver. Under phased pituitary inputs, including FSH and hCG surge, the platform reproduces a 28-day cycle and reveals that connecting downstream tissues reshapes upstream hormone profiles—absolute steroid and peptide levels fall when tissues are coupled, consistent with consumption and feedback. Sustained hCG further stabilizes luteal activity and elevates progesterone, creating a pregnancy-like state. These observations place transport and tissue coupling are as determinative as hormone dose in shaping endocrine outcomes (Fig.  9 g) [ 82 ]. A complementary view at the tissue-patterning scale shows that local signaling hubs gate competence. In a microfluidic model of early human post-implantation development, amniotic ectoderm-like cells acted as a signaling center that initiated gastrulation-like events in adjacent epiblast. Asymmetric BMP4 delivery across device channels polarizes embryonic-like sacs within 36 h, illustrating how spatially resolved morphogen presentation programs fate even when factor identity is unchanged (Fig.  9 h) [ 83 ]. Together, these studies suggest a clear rule for engineering the follicle–ovary microenvironment. Treat hormones and morphogens as context-dependent instructions, not bulk additives. Phase the inputs, localize their sources, and maintain perfusion so that gradients and feedback are preserved. In practice, present FSH, LH, and activin within ECM-defined architecture and adhesive cues under controlled flow. This keeps folliculogenesis, steroidogenesis, and downstream tissue responses. Collectively, these advances indicate that next-generation ovarian bioengineering platforms must move beyond static hormone supplementation toward dynamic, multicellular, and spatially resolved systems that integrate endocrine signaling with ECM structure, vascular support, and immune regulation. Such approaches will be essential for accurately recapitulating folliculogenesis and restoring physiologic ovarian function. The female reproductive system is characterized by a dynamic and hormonally responsive extracellular matrix (ECM) that undergoes cyclic remodeling to support gamete maturation, fertilization, implantation and pregnancy. In the ovary, ECM components such as collagen types I and IV, laminin and fibronectin provide structural integrity to the follicle, mediate bidirectional signaling between oocytes and somatic cells and modulate folliculogenesis and ovulation [ 84 ]. Within the endometrium, the ECM forms a specialized basal lamina and stromal scaffold that reorganizes across the menstrual cycle to regulate epithelial–stromal interactions, immune cell trafficking and embryo receptivity. In the oviduct and cervix, region-specific ECM composition and stiffness influence sperm transport, fertilization efficiency and mucosal barrier properties [ 85 , 86 ]. These tissue- and cycle-specific ECM architectures act not merely as passive scaffolds but as active regulators of cell behavior, hormone responsiveness and vascular remodeling. Recapitulating such reproductive ECM complexity is therefore essential for building physiologically relevant in vitro models and advancing reproductive tissue engineering. The advancement of 3D culture systems from traditional 2D platforms, through the use of diverse biomaterials, has enabled the development of organotypic models that more accurately recapitulate native tissue architecture. In the human body, the extracellular matrix (ECM) plays a central role in providing structural support, as well as regulating cellular attachment, proliferation, and differentiation [ 87 , 88 ]. By mimicking ECM within in vitro culture systems through the integration of various biomaterials, researchers can now better reproduce in vivo cellular functions. Building on this advancement, biocompatible scaffolds have been adopted in regenerative medicine to facilitate tissue engineering and functional tissue reconstruction. In the female reproductive context, tissue engineering enables analysis of fertilization through reconstruction of the reproductive microenvironment [ 89 ]. Accordingly, tissue engineering with biomaterials provides controlled niches for follicle, oocyte, and embryo culture and have become key tools in reproductive research. Natural biomaterials such as collagen, gelatin, fibrin, and hyaluronic acid have transformed scaffolds from passive supports into instructive niches that actively regulate cellular behavior, particularly when recapitulated in extracellular matrix–derived systems such as decellularized ECM (Fig.  10 a) [ 89 ]. Collagen offers a fibrillar architecture with integrin ligands, and its fibrillogenesis and crosslinking can be tuned so that follicle growth and endometrial assembly respond to physiological stiffness and pore structure. Gelatin, the denatured form of collagen, retains adhesive and protease-sensitive motifs; once chemically stabilized it supports long-term culture while allowing cell-driven remodeling [ 90 ]. Fibrin behaves as a provisional matrix that cells can degrade and replace, which makes it well suited for modeling embryo–endometrium interactions and the angiogenic remodeling that accompanies implantation [ 91 ]. Hyaluronic acid contributes hydration and viscoelasticity and, through CD44 engagement and controlled functionalization, helps regulate migration and differentiation in layered constructs. Fig. 10 Topic-based review of bioengineered platforms for studying fibrin matrix and hydrogel maturation. a Decellularized extracellular matrix (dECM) is generated by removing cellular components from native tissues and provides a bioactive scaffold for applications including injectable hydrogels, 3D bioprinting, hydrogel scaffolds, and organ-on-a-chip systems. Reproduced with permission from Ref [ 87 ] © 2025 Biofabrication. b Workflow of a 3D endometrial co-culture system enabling studies of epithelial–stromal interactions and cancer invasion, with stromal cells embedded in a collagen/Matrigel matrix and either primary epithelial cells or KLE adenocarcinoma cells layered above. Reproduced with permission from Ref [ 93 ] © 2003 Cancer Lett. c Schematic overview of the experimental strategy in which human endometrial organoids were derived from biopsies, cultured with or without EndoECM supplementation, and characterized through histological, molecular, and functional analyses to assess the role of EndoECM in organoid development. Reproduced with permission from Ref [ 95 ] © 2021 J Pers Med. d Configuration of decellularized porcine uterus-derived extracellular matrices (UdECMs), showing separation into endometrium-derived (Endo-UdECM) and whole-uterus-derived (Whole-UdECM) fractions. Reproduced with permission from Ref [ 99 ] © 2023 Adv Funct Mater. e Comparison between the native amniotic membrane and a 3D cell–matrix culture model designed to mimic its epithelial and mesenchymal cell organization. Reproduced with permission from Ref [ 101 ] © 2005 Am J Obstet Gynecol. f Fibrin clot within a peritoneal pocket at 7 days post-grafting, identified by the nonresorbable suture used to form the pocket. Reproduced with permission from Ref [ 103 ] © 2014 Fertil Steril Topic-based review of bioengineered platforms for studying fibrin matrix and hydrogel maturation. a Decellularized extracellular matrix (dECM) is generated by removing cellular components from native tissues and provides a bioactive scaffold for applications including injectable hydrogels, 3D bioprinting, hydrogel scaffolds, and organ-on-a-chip systems. Reproduced with permission from Ref [ 87 ] © 2025 Biofabrication. b Workflow of a 3D endometrial co-culture system enabling studies of epithelial–stromal interactions and cancer invasion, with stromal cells embedded in a collagen/Matrigel matrix and either primary epithelial cells or KLE adenocarcinoma cells layered above. Reproduced with permission from Ref [ 93 ] © 2003 Cancer Lett. c Schematic overview of the experimental strategy in which human endometrial organoids were derived from biopsies, cultured with or without EndoECM supplementation, and characterized through histological, molecular, and functional analyses to assess the role of EndoECM in organoid development. Reproduced with permission from Ref [ 95 ] © 2021 J Pers Med. d Configuration of decellularized porcine uterus-derived extracellular matrices (UdECMs), showing separation into endometrium-derived (Endo-UdECM) and whole-uterus-derived (Whole-UdECM) fractions. Reproduced with permission from Ref [ 99 ] © 2023 Adv Funct Mater. e Comparison between the native amniotic membrane and a 3D cell–matrix culture model designed to mimic its epithelial and mesenchymal cell organization. Reproduced with permission from Ref [ 101 ] © 2005 Am J Obstet Gynecol. f Fibrin clot within a peritoneal pocket at 7 days post-grafting, identified by the nonresorbable suture used to form the pocket. Reproduced with permission from Ref [ 103 ] © 2014 Fertil Steril As generic matrices mature, organ specificity becomes decisive. Uterus-derived decellularized ECM preserves endometrium-specific proteins, glycosaminoglycans, and matrix-bound factors that strengthen epithelial–stromal cross talk and hormone responsiveness in organotypic cultures [ 92 ]. Gelatin methacryloyl adds light-patternable control over stiffness, degradability, and geometry, which enables microscale compartmentalization for layered endometrium or spherical ovarian follicle niches and integrates cleanly with microfluidics and bioprinting. Topics: fibrin matrix, hydrogel maturation Hydrogel-based culture systems have become indispensable for modeling reproductive tissues, as they provide a supportive environment that closely resembles the ECM. One of the earliest approaches combined collagen I with Matrigel to generate a three-dimensional endometrial construct in which stromal cells supported the formation of polarized epithelium (Fig.  10 b) [ 93 ]. This system successfully reproduced key morphological features of the endometrium, including microvilli, cilia, and basal nuclei positioning, and enabled functional studies such as cancer cell invasion mediated by matrix metalloproteinases and their inhibitors. Such findings demonstrated that relatively simple hydrogels could sustain both the structural and physiological processes of reproductive tissue. Subsequent work moved toward the use of synthetic hydrogels that allow finer control of biochemical cues. Alginate-based matrices, modified with ECM proteins or short peptide motifs, were used to investigate ovarian follicle development [ 94 ]. These systems preserved the spherical morphology of follicles while inducing the growth and differentiation of somatic cells. Notably, the hydrogel composition not only influenced somatic behavior but also altered the meiotic competence of the enclosed oocyte, indicating that ECM cues act as dynamic regulators of folliculogenesis. The intricate biochemical composition and spatial architecture of native ECM remain challenging to replicate using synthetic platforms, thereby driving a transition toward tissue-specific, decellularized hydrogels. To address the limited molecular complexity of engineered matrices, more recent advances have focused on tissue-specific hydrogels derived from decellularized reproductive tissues [ 95 ]. Hydrogels generated from decellularized endometrium, when applied to human endometrial organoid cultures, enhanced proliferation, preserved epithelial identity, and maintained chromosomal stability over prolonged culture (Fig.  10 c). By more faithfully recapitulating the complexity of the native ECM, these bioactive materials provided a superior microenvironment for sustaining organoid growth and function compared with conventional matrices. Complementary microfluidic platforms have further demonstrated that perfusable vascularized tissue arrays can be formed in a high-throughput format, enabling scalable angiogenic phenotypic screening and advancing the design of vascularized reproductive tissue constructs [ 96 , 97 ]. Engineered microvascular network models have further advanced this field by enabling precise control over vascular architecture, providing biomimetic platforms applicable to placental barrier and endometrial angiogenesis modeling [ 98 ]. Consistent with these organoid findings, uterus-derived dECM (from endometrium or whole uterus) has restored endometrial structure and function in injury models, improving receptivity and implantation outcomes while normalizing the local immune milieu toward an implantation-permissive state (Fig.  10 d) [ 99 ]. Transcriptomic and perturbation data further suggest that these effects are mediated, at least in part, by modulation of the IGF axis and associated stromal–epithelial crosstalk, underscoring that tissue-matched matrices deliver instructive biochemical cues rather than passive support. Ex vivo responses of patient tissues also vary by dECM source, pointing to opportunities for tailoring matrix composition to specific endometrial pathologies and advancing personalized regenerative strategies across the female reproductive tract. Together, these developments highlight a clear trajectory in hydrogel maturation strategies. Initial reliance on natural ECM components demonstrated the feasibility of reproducing tissue architecture and function, engineered synthetic matrices introduced the possibility of tailoring biochemical cues to direct development, and decellularized tissue-derived hydrogels now offer a highly faithful mimic of the native microenvironment. This progression underscores the principle that the closer hydrogels approximate the native ECM, the more effectively they can support the maturation, differentiation, and long-term stability of reproductive tissues in vitro. Fibrin, a physiological polymer generated during coagulation, has emerged as a versatile scaffold in reproductive tissue engineering [ 100 ]. Its intrinsic properties a permissive three-dimensional architecture, mechanical stability, and pro-angiogenic signaling capacity—distinguish it from other natural matrices that often suffer from contraction and instability. Comparative studies have consistently underscored these advantages. In amnion construct models, for instance, collagen I scaffolds exhibited substantial volumetric contraction while fibrin matrices maintained geometric stability over more than a week [ 101 ]. Importantly, cells embedded within fibrin retained native-like viability and morphology, and supplementation with fibronectin enhanced mesenchymal migration accompanied by MMP-9 activity (Fig.  10 e). These findings indicate that fibrin not only preserves scaffold architecture but also promotes controlled, cell-driven remodeling—features particularly relevant for membrane reinforcement and repair. Beyond structural stability, fibrin provides a robust platform to recapitulate early reproductive pathophysiology. Endometrial explants cultured in fibrin matrices demonstrated stepwise progression from cellular outgrowth and invasion to the self-organization of gland-like structures, stromal integration, and eventual neovascularization, with CD31-positive sprouts detected [ 102 ]. Such data highlight fibrin’s unique capacity to support both tissue morphogenesis and angiogenesis, establishing it as a tractable in vitro system for modeling complex disorders such as endometriosis. Translational applications have further demonstrated fibrin’s utility. Artificial ovary prototypes incorporating defined fibrinogen/thrombin formulations enabled the encapsulation and autografting of preantral follicles together with ovarian stromal cells in murine models (Fig.  10 f) [ 103 ]. Grafts achieved ~ 31% follicle recovery after one week, with viable, proliferating follicles advancing to primary, secondary, or early antral stages. Notably, vascular infiltration into the fibrin constructs supported nutrient and oxygen delivery, underscoring fibrin’s capacity to integrate rapidly with host tissues while preserving follicular development an essential attribute for fertility restoration strategies. Fibrin occupies a unique niche within reproductive tissue engineering, bridging foundational studies of cell–matrix dynamics, disease modeling in vitro, and clinically oriented constructs for repair and fertility restoration. Its capacity to combine structural fidelity with bioactivity has made it a central platform at the intersection of basic discovery and translational application. Gynecological cancers are characterized by the uncontrolled growth of cells within the female reproductive system [ 104 ]. These malignancies account for approximately 17% of cancers in women worldwide, yet their risk factors, clinical manifestations, treatment responses, and prognoses vary greatly among patients. The most common gynecological cancers arise in the endometrium, ovary, and cervix (Fig.  11 a). Endometrial and ovarian cancers are most often diagnosed after menopause, while cervical cancer tends to develop at younger ages. A common treatment is the surgical removal of the uterus, cervix, ovaries, and fallopian tubes, but this procedure causes irreversible effects such as infertility and sexual dysfunction, significantly reducing quality of life. Fig. 11 Topic-based review of bioengineered platforms for studying cancer spheroids, drug delivery tumor. a Schematic overview of the female reproductive tract highlighting major gynecological cancer sites. Reproduced with permission from Ref [ 104 ] © 2020 Trends Cancer. b Representative image of stromal elements expanding within patient-derived organoid (PDO) cultures. Scale bar, 50 μm. Reproduced with permission from Ref [ 113 ] © 2017 Int J Gynecol Cancer. c Brightfield images of organoids derived from healthy endometrium (EM-O) and ectopic lesions (ECT-O) across passages, demonstrating long-term expansion capacity. A representative structure is magnified in the inset. Scale bars, 200 μm; inset, 50 μm. Reproduced with permission from Ref [ 117 ] © 2019 Nat Cell Biol d Workflow for isolating and culturing endometrial stromal cells and epithelial organoids from menstrual fluid. Reproduced with permission from Ref [ 39 ] © 2023 Front Endocrinol. e Confocal images of bioprinted constructs at day 1 showing homogeneous distribution of GFP-labeled SKOV-3 cells (green) and CMRA-labeled MeWo cells (red). Scale bar, 1 mm. Reproduced with permission from Ref [ 119 ] © 2023 Macromolecular Bioscience. f Representative images highlighting the advantage of microfluidic microwells in generating uniform spheroids from PDX-derived cells compared with Matrigel. Scale bar, 250 μm. Reproduced with permission from Ref [ 125 ] © 2020 Microsyst Nanoeng Topic-based review of bioengineered platforms for studying cancer spheroids, drug delivery tumor. a Schematic overview of the female reproductive tract highlighting major gynecological cancer sites. Reproduced with permission from Ref [ 104 ] © 2020 Trends Cancer. b Representative image of stromal elements expanding within patient-derived organoid (PDO) cultures. Scale bar, 50 μm. Reproduced with permission from Ref [ 113 ] © 2017 Int J Gynecol Cancer. c Brightfield images of organoids derived from healthy endometrium (EM-O) and ectopic lesions (ECT-O) across passages, demonstrating long-term expansion capacity. A representative structure is magnified in the inset. Scale bars, 200 μm; inset, 50 μm. Reproduced with permission from Ref [ 117 ] © 2019 Nat Cell Biol d Workflow for isolating and culturing endometrial stromal cells and epithelial organoids from menstrual fluid. Reproduced with permission from Ref [ 39 ] © 2023 Front Endocrinol. e Confocal images of bioprinted constructs at day 1 showing homogeneous distribution of GFP-labeled SKOV-3 cells (green) and CMRA-labeled MeWo cells (red). Scale bar, 1 mm. Reproduced with permission from Ref [ 119 ] © 2023 Macromolecular Bioscience. f Representative images highlighting the advantage of microfluidic microwells in generating uniform spheroids from PDX-derived cells compared with Matrigel. Scale bar, 250 μm. Reproduced with permission from Ref [ 125 ] © 2020 Microsyst Nanoeng The endometrium, which forms the inner epithelial lining of the uterus, serves as the site of origin for endometrial cancer (EC) [ 105 ]. EC is among the most prevalent gynecological cancers, with its pathogenesis associated with both genetic alterations and environmental risk factors, including obesity and hormonal dysregulation [ 106 ]. Endometrial cancer is most commonly identified after menopause, where hysterectomy generally is required as a treatment. However, a growing subset of patients is being diagnosed at younger than 40 years of age, many of whom have not had children [ 107 ]. In such cases, and particularly in advanced disease, therapeutic options that preserve fertility or ovarian function remain limited, which poses a substantial challenge for clinical management. Ovarian cancer is rarely diagnosed at an early stage, with most patients presenting after progression [ 105 , 108 ]. Although debulking surgery combined with platinum-based chemotherapy remains the standard of care and achieves high initial response rates, the majority of patients relapse within three years. A substantial proportion develop platinum resistance or refractoriness at first recurrence, and the effectiveness of non-platinum agents such as paclitaxel, docetaxel, pegylated liposomal doxorubicin, gemcitabine, and topotecan is limited [ 107 , 109 ]. These limitations highlight the need to clarify the molecular mechanisms underlying tumor progression and chemoresistance to enable the development of more effective therapeutic strategies. Cervical cancer remains a major global health challenge, with more than half a million new cases and over 300,000 deaths each year, making it the fourth most common malignancy in women worldwide [ 110 ]. High-risk subtypes of the human papillomavirus are the predominant cause, and prevention through screening and vaccination has dramatically reduced incidence and mortality in high-income countries, where rates have declined by more than half over the past three decades. In contrast, approximately 90% of cases occur in low- and middle-income countries that lack organized screening and vaccination programs, where mortality is up to 18 times higher. Treatment strategies depend on disease stage and available resources, ranging from fertility-preserving surgery in early-stage cases to chemoradiation with advanced radiotherapy techniques for locally advanced diseases. While targeted therapies such as bevacizumab and emerging immunotherapies have shown some promise in recurrent or metastatic settings, the prognosis remains poor for these patients. Topics: cancer spheroids, drug delivery tumor Compared with 2D monolayers, cancer spheroids and patient-derived organoids (PDOs) recreate cell–cell and matrix architecture with physiologic oxygen and drug gradients [ 111 ]. Recent advances have further integrated 3D spheroid platforms with biosensing capabilities, enabling real-time monitoring of cellular responses within three-dimensional constructs and expanding the utility of spheroid-based systems for drug screening and disease modeling [ 17 , 112 ]. The integration of organoid and microfluidic chip technologies into vascularized tumor-on-a-chip platforms further enhances the recapitulation of tumor pathophysiology, including perfusable microvascular networks critical for drug delivery and resistance modeling. This results in drug responses that more closely align with in vivo tumors. In gynecologic cancers, PDOs preserve the genotype and phenotype of the source tumor, achieve high establishment rates, and enable biobanking for standardized functional assays (Fig.  11 b) [ 113 ]. Immortalized cancer cell lines frequently drift genetically and narrow intratumoral heterogeneity, which limits their predictive value efforts to culture primary endometrial cancer epithelium report low success rates of about 20%. Patient -derived xenografts capture many features of the original tumor but require long set-up times and substantial cost, with variable take rates that hinder routine, time-sensitive decisions [ 114 ]. PDO platforms, by contrast, support faster turnaround and higher throughput, and several studies report concordance between PDO drug sensitivity and clinical responses, including in colorectal, pancreatic, and ovarian cancer cohorts. Importantly, multicellular tumor spheroids further recapitulate diffusion-limited drug penetration, hypoxia gradients, and multicellular organization, which significantly alter therapeutic responses compared with 2D systems and often lead to increased resistance to chemotherapeutic agents such as paclitaxel and cisplatin [ 115 , 116 ]. Endometrial cancer PDOs function as a patient-specific test bed rather than a descriptive culture model. The workflow starts with fresh surgical tissue, forms organoids rapidly, and yields a simple growth-inhibition endpoint on a clinically relevant clock [ 114 , 117 ]. The platform preserves tumor architecture and canonical markers, so signals that emerge hormone dependence, pathway addiction, or multi-drug tolerance reflect the biology of the source lesion rather than artifacts of long-term cell-line adaptation. As a result, PDO assays operate as the practical link between static genomic profiles and real treatment choices in gynecologic oncology. PDO-based spheroid systems justify a shift from 2D screening toward functional precision models that can standardize endpoints, support comparative drug testing, and feed directly into prospective clinical evaluation. Patient-derived endometrial organoids establish a single platform that spans healthy tissue, endometriosis, precancer with Lynch syndrome, and multiple endometrial cancer subtypes [ 117 ]. Cultures expand over extended periods, retain lineage identity, and preserve genomic and transcriptomic features that define each lesion and subtype (Fig.  11 c). Organoids reproduce architecture and marker patterns of the source tissue and form lesions after transplantation in vivo, which confirms biological fidelity. The platform enables a biobank across the benign–precancer–cancer continuum, permits dissection of tissue-specific pathways such as canonical WNT and PI3K–AKT, and preserves mismatch-repair defects, microsatellite instability, and copy-number landscapes. Standardized drug assays reveal patient-dependent responses, connect static genomic profiles with functional sensitivity data, and provide a practical chassis for precision models and drug screens within the female reproductive system . Patient-derived organoids from endometrial cancer serve as a practical test bed for individualized drug assessment. Fresh surgical tissue yields organoids within hours, and the platform delivers a growth-based readout within two weeks, which aligns with clinical decision windows. The cultures retain tumor architecture and immunohistochemical features from the source lesion and preserve pathway dependencies, including estrogen receptor dependence and STAT3 pathway activity, so the signals reflect patient-specific biology rather than artifacts from long-term cell lines. A concise workflow enzymatic dissociation, basement-membrane encapsulation in serum-free medium, and count-based endpoints establishes a standard that laboratories can reproduce without reliance on xenografts. This model links static genomic information to functional sensitivity data and defines a route toward prospective evaluation of clinical utility in gynecologic oncology. Across endometrial and ovarian tumors, effective drug delivery must account for epithelial–stromal crosstalk, stromal barriers, and patient-specific phenotypes that shape uptake and response. Patient-derived endometrial organoids reproduce clinical heterogeneity and retain genotype, hormone receptors, and pathway dependencies, which allows direct tests of endocrine agents and targeted compounds on patient tissue [ 118 ]. Menstrual-fluid protocols add a noninvasive route to paired epithelial organoids and stromal cells from the same donor, so investigators can compare epithelial drug uptake and stromal decidual or paracrine effects under matched genetics and cycle status [ 117 ]. Based on this patient-specific platform, coculture tumor constructs built by 3D bioprinting place ovarian cancer cells with cancer-associated fibroblasts inside tunable hydrogels, where fibroblast recruitment around malignant cells recreates microenvironmental barriers that alter penetration, distribution, and resistance, yet with throughput suitable for delivery screens [ 119 ]. Extending these advances, microfluidic spheroid systems add precise control over size, architecture, and medium volume, preserve epithelial markers and viability better than large-volume or Matrigel setups, and use integrated valves to load scarce biopsy material once and apply full dose series in parallel [ 120 ]. Uniform arrays of spheroids fix diffusion paths and concentration gradients, which clarifies depth-dependent effects, separates cytotoxic from cytostatic action by size change and viability, and supports quantitative comparisons across patient avatars [ 121 ]. Together these platforms define a practical foundation for tumor-focused drug delivery in the female reproductive tract, where noninvasive sampling, faithful epithelial and stromal reconstruction, and microscale control converge to guide regimen selection and formulation design. Gynecologic cancers represent a biologically diverse group of malignancies driven by distinct etiologies yet sharing common therapeutic challenges. Ovarian cancer, particularly high-grade serous ovarian carcinoma, frequently develops resistance to chemotherapy through genomic alterations such as CCNE1 amplification and dysregulation of DNA damage repair pathways [ 122 ]. In contrast, endometrial cancer is strongly influenced by endocrine signaling, where estrogen-dependent pathways regulate tumor progression through modulation of key tumor suppressors such as PTEN, as highlighted by integrative genomic analyses of endometrial carcinoma [ 123 ]. Cervical cancer is primarily driven by persistent high-risk human papillomavirus infection, which reshapes the immune microenvironment and promotes malignant transformation. Despite these differences, a unifying feature across gynecologic cancers is the presence of a complex tumor microenvironment that critically influences therapeutic response [ 124 ]. This microenvironment regulates key processes including drug penetration, stromal–epithelial interactions, immune evasion, and resistance development, thereby limiting the efficacy of conventional therapies. Consequently, advanced in vitro platforms that recapitulate these multicellular and microenvironmental dynamics, such as patient-derived organoids and microfluidic systems, have emerged as essential tools for understanding disease mechanisms and optimizing therapeutic strategies. Menstrual fluid yields paired endometrial epithelial organoids and stromal cells through a fully noninvasive workflow that suits drug delivery studies in gynecologic tumors [ 39 ]. Tissue fragments from the sample separate into epithelial clusters for organoid formation within matrix and a stromal fraction that adheres as a monolayer, which recreates the principal compartments of the endometrium (Fig.  11 d). Each culture preserves hallmark functions: organoids mount an estrogen response, stromal cells undergo decidual response under progesterone and cyclic AMP, and both retain lineage markers. Serial sampling across cycles becomes feasible, so investigators can map donor-specific hormone states and adjust carrier choice, dose, and schedule for delivery to tumor tissue that arises in the same uterine niche. Co-culture of matched epithelium and stroma enables assessment of barrier transfer, paracrine cues, and matrix constraints that govern penetration, retention, and therapeutic effect, thus providing a patient-specific platform for optimization of local and systemic delivery strategies in endometrial cancer. A coculture, 3D-printed ovarian tumor model that combines epithelial cancer cells with cancer-associated fibroblasts provides a controllable microenvironment for studies of intratumoral delivery [ 119 ]. A gelatin–alginate hydrogel supports precise geometry, high initial viability, and rapid generation of uniform constructs, which enables high-throughput assays under reproducible diffusion distances (Fig.  11 e). Within this matrix, fibroblasts move toward and encircle tumor cells to form heterotypic aggregates that mirror stromal corralling observed in vivo, a configuration known to alter penetration, sequestration, and efflux of therapeutics. Standard histology and immunolabels confirm epithelial identity and fibroblast activation, while real-time observation documents cell rearrangement that establishes nutrient and oxygen gradients across the construct. The platform therefore links architectural control with microenvironmental behaviors that shape transport and response and offers a practical test bed to compare carrier chemistry, size, and dosing schedules for delivery to ovarian tumors. The microfluidic platform for ovarian cancer offers a tumor surrogate that tests drug delivery and response with very limited tissue [ 125 ]. Microwell chambers in small volumes form uniform patient-derived spheroids, and on-chip valves switch between serial cell loading and parallel dose application so that one device evaluates multiple concentrations from a small cell input [ 126 ]. Under matched media, viability and epithelial marker expression exceed those in large-volume 3D or Matrigel systems, and the platform supports spheroid formation for models that fail on Matrigel (Fig.  11 f). Arrayed and size-controlled spheroids standardize diffusion paths and concentration gradients, which clarifies permeability, intratumoral accumulation, and layer-specific sensitivity to therapy [ 127 ]. The same device yields full dose–response curves and separates cytotoxic agents that induce cell death from cytostatic agents that arrest growth based on changes in spheroid size. The minimal cell requirement aligns with fine-needle biopsy yields and preserves phenotypic heterogeneity and relevant microenvironmental cues of ovarian cancer. This system thus serves as a preclinical tool to select and optimize drug delivery strategies in the female reproductive tract. Importantly, these platforms converge on a common principle of reconstructing patient-specific tumor microenvironments with controlled architecture and cellular composition, enabling more predictive evaluation of drug delivery and therapeutic response in gynecologic cancers.

The female reproductive system, comprising the endometrium, placenta, ovary, cervix, and fallopian tubes, plays a critical role in conception, implantation, and fetal development [ 1 ]. These tissues function as an integrated system within a tightly regulated endocrine and immune environment, primarily governed by ovarian hormones, and undergo dynamic structural and functional remodeling throughout the reproductive cycle [ 2 , 3 ]. This biological complexity extends from molecular signaling to tissue and organ level processes and presents substantial challenges for biomedical research and clinical practice [ 4 , 5 ]. The clinical importance of this system is further underscored by the high burden of reproductive disorders, with gynecological cancers alone accounting for approximately 17% of all cancers in women worldwide [ 6 ]. These facts highlight the urgent need for integrative approaches that bridge molecular mechanisms with organ-level physiology in female reproductive research. Over the past two decades, researchers have increasingly adopted bioengineered in vitro models to study and manipulate reproductive tissues under controlled conditions [ 7 , 8 ]. Conventional 2D cell cultures have long provided basic tools for exploring reproductive physiology, but lack the structural and functional complexity of living tissues. To overcome these limitations, advanced approaches have emerged, including 3D organoids [ 9 ], organ-on-a-chip [ 10 ], 3D bioprinting platforms [ 11 ], and scaffold-based constructs [ 12 ]. These next-generation models more faithfully recapitulate key aspects of reproductive physiology including cell–cell interactions, extracellular matrix remodeling, hormone responsiveness, and barrier function. Organoids derived from endometrial or ovarian tissue reproduce the cellular composition and functional characteristics of the native organ with greater precision than flat monolayers [ 13 ]. Notably, microphysiological systems (MPS), including organ-on-a-chip technologies, have emerged as advanced platforms capable of recapitulating dynamic reproductive microenvironments across multiple stages of assisted reproductive processes [ 14 – 16 ]. Within these systems, organ-on-a-chip devices enable the reproduction of dynamic environments through controlled fluid flow and hormonal regulation. Microfluidic systems, in particular, provide highly controllable and reproducible microenvironments and have been widely applied to preclinical drug delivery testing and organ-mimetic platforms [ 17 ]. 3D bioprinting strategies provide spatial control over multiple cell types within tissue-specific architectures. Collectively, these platforms have greatly expanded the experimental toolkit, but their rapid and diverse development has also resulted in a fragmented and heterogeneous body of literature that is difficult to integrate into a coherent synthesis [ 18 ]. This fragmented landscape, further complicated by inconsistent terminology and disparate research focuses, limits the ability to derive coherent insights across studies. AI-driven topic modeling addresses this challenge by systematically organizing heterogeneous literature into a structured knowledge framework, enabling the identification of hidden connections and critical research gaps that are often overlooked in traditional reviews. The necessity of AI-based integration in this field is supported by several converging factors. First, the exponential growth of biomedical publications has made manual literature synthesis increasingly impractical and prone to bias. Second, reproductive biology is inherently multidisciplinary, spanning molecular endocrinology, tissue engineering, and clinical medicine, where complex semantic relationships cannot be fully captured through conventional keyword-based approaches. Domain-specific language models such as BioBERT are designed to address this limitation by encoding contextual relationships within biomedical corpora. Finally, AI-driven topic modeling has demonstrated the ability to uncover latent thematic structures and cross-domain connections within complex datasets, enabling the identification of emerging research directions and overlooked knowledge gaps. Taken together, these considerations highlight the necessity of AI-based integration in female reproductive research, where the interplay between hormonal regulation, engineered models, and clinical translation requires a scalable and systems-level analytical approach. To systematically address this complexity, we employed an artificial intelligence (AI)-based text mining and bibliometric mapping framework. We used BioBERT, a biomedical transformer model, to encode semantic relationships among studies across the endometrium, placenta, ovary, cervix, and fallopian tube, enabling AI to capture conceptual similarities beyond keyword matching. Dimensionality reduction was performed with UMAP, followed by unsupervised clustering using HDBSCAN to identify latent thematic groups [ 19 , 20 ]. Representative keywords for each cluster were then extracted through class-based TF–IDF, which enabled interpretation and labeling of topics [ 21 , 22 ]. In parallel, curated metadata including publication years, authors, and institutional affiliations supported bibliometric analysis, allowing us to identify prolific contributors and research hubs within the field. These complementary methods provide an objective and scalable approach to distilling thousands of publications, moving beyond the limitations of traditional narrative reviews and reducing reliance on user bias [ 21 , 23 ]. Through this computational strategy, we generated a quantitative map of research activity across female reproductive system bioengineering. The topic modeling analysis systematically delineates dominant themes, areas of overlap, and emerging frontiers, while bibliometric mapping highlights key researchers and institutions driving innovation. By quantifying the distribution of studies, our approach identifies both extensively studied domains and underexplored areas. This dual perspective provides a panoramic overview of the field that links biological insights with translational priorities [ 8 , 19 ]. Importantly, by uncovering knowledge gaps and directing attention to underdeveloped areas, AI-driven analyses can inform targeted strategies for advancing personalized reproductive medicine. Integrating organ-level biological contexts with AI-based literature synthesis provides a coherent framework for clarifying the state-of-the-art in reproductive biology and guiding future studies toward clinically and therapeutically impactful directions.