Extracellular vesicle-derived miRNA-mediated cell–cell communication inference for single-cell transcriptomic data with miRTalk

Shao, Xin; Yu, Lingqi; Li, Chengyu; Qian, Jingyang; Yang, Xinyu; Yang, Haihong; Liao, Jie; Fan, Xueru; Xu, Xiao; Fan, Xiaohui

doi:10.1186/s13059-025-03566-x

Software
Open access
Published: 14 April 2025

Extracellular vesicle-derived miRNA-mediated cell–cell communication inference for single-cell transcriptomic data with miRTalk

Xin Shao^1,2,3,4^na1,
Lingqi Yu^1,2,3,6^na1,
Chengyu Li^2,3,
Jingyang Qian^2,3,
Xinyu Yang^5,6,
Haihong Yang⁷,
Jie Liao^2,3,
Xueru Fan^2,3,
Xiao Xu^8,9 &
…
Xiaohui Fan ORCID: orcid.org/0000-0002-6336-3007^1,2,3,4

Genome Biology volume 26, Article number: 95 (2025) Cite this article

1411 Accesses
1 Altmetric
Metrics details

Abstract

MicroRNAs are released from cells in extracellular vesicles (EVs), representing an essential mode of cell–cell communication (CCC) via a regulatory effect on gene expression. Single-cell RNA-sequencing technologies have ushered in an era of elucidating CCC at single-cell resolution. Herein, we present miRTalk, a pioneering approach for inferring CCC mediated by EV-derived miRNA-target interactions (MiTIs). The benchmarking against simulated and real-world datasets demonstrates the superior performance of miRTalk, and the application to four disease scenarios reveals the in-depth MiTI-mediated CCC mechanisms. Collectively, miRTalk can infer EV-derived MiTI-mediated CCC with scRNA-seq data, providing new insights into the intercellular dynamics of biological processes.

Background

MicroRNAs (miRNAs) comprise a class of small RNA molecules that predominantly orchestrate numerous cellular processes through post-transcriptional gene silencing or, although less often, activation in eukaryotes [1]. Over the past decade, miRNAs have been discovered within extracellular environments and have demonstrated their capacity to facilitate functional cell–cell communication (CCC) [2]. MiRNAs produced and released from one cell via extracellular vesicles (EVs) such as exosomes and microvesicles are captured by distant cells, subsequently leading to alterations in gene expression, cellular behavior, and function [2]. To date, EV-derived miRNAs have achieved widespread recognition as a distinctive mode of CCC, with mounting evidence demonstrating the pivotal role of EV-derived miRNA-mediated CCC in the emergence and progression of diseases [3, 4]. For instance, Wang et al. discovered that the depletion of miR- 9- 3p in adipose tissue-derived EVs significantly mitigates their deleterious effects on cognitive function in high-fat diet mice or patients with diabetes, thus unveiling the crucial role of the adipose tissue-neuron communication mediated by EV-derived miR- 9- 3p in promoting synaptic loss and cognitive impairment [5]. He et al. reported that EV-derived miR- 223 from neutrophils functioned to impede hepatic inflammatory and fibrogenic gene expression, revealing the neutrophil-to-hepatocyte communication via miR- 223-enriched EV transfer in the alleviation of nonalcoholic steatohepatitis [6]. Nonetheless, due to the limitations of related technologies and the pervasive cellular heterogeneity within a given tissue, the large-scale systematic investigation of EV-derived miRNA-mediated CCC among heterogeneous cell populations becomes substantially difficult, severely constraining the advancement of this field.

Fortunately, increasing progress in high-throughput single-cell RNA-sequencing (scRNA-seq) technologies has facilitated the classification of thousands of cells in a single assay based on transcriptome profiling of coding and non-coding genes [7, 8], enabling the elucidation of EV-derived miRNA-mediated CCC mechanisms underlying physiological and pathological processes at single-cell resolution. As an illustration, Ramanujam et al. identified the cardiac macrophage-derived miR- 21 as vital in the transition from quiescent fibroblasts to myofibroblasts through single-cell analysis of pressure-overloaded hearts from mice featuring macrophage-specific ablation of miR- 21, thereby revealing the involvement of miR- 21 in guiding macrophage-fibroblast communication regarding myocardial homeostasis and malady-associated remodeling [9]. Zhang et al. performed scRNA-seq analyses of human intrahepatic cholangiocarcinoma (CHOL) and neighboring samples and found that cancer cell (CC)-derived exosomal miR- 9- 5p engendered elevated expression of IL- 6 in vascular cancer-associated fibroblasts (vCAFs) to expedite tumor progression, underscoring the significance of miR- 9- 5p-dependent CCC between CCs and vCAFs within the tumor microenvironment (TME) [10]. Nevertheless, the absence of a computational approach for inferring EV-derived miRNA-mediated CCC based on scRNA-seq data presents a great challenge.

In this study, we developed a distinctive computational method, miRTalk (https://github.com/multitalk/miRTalk), to infer EV-derived miRNA-mediated CCC based on scRNA-seq data. By integrating experimentally verified data of EV-derived miRNAs and their corresponding miRNA-target interactions (MiTIs) as well as associated pathways from diverse repositories, we have constructed an integrated database called miRTalkDB that encompasses EV-derived MiTIs for humans, mice, and rats and functional annotations for EV-derived miRNAs. With miRTalkDB as the underlying reference, our model aims to predict the likelihood of EV-derived miRNA-mediated CCC based on the potential of producing EVs and the expression of miRNAs in senders as well as the activation of miRNA processing machinery and the expression of target genes in receivers. The performance of miRTalk was validated through simulated and real-world benchmarking datasets. Furthermore, in exploring scRNA-seq and spatial transcriptomics (ST) data from four disease scenarios—human glioblastoma, mouse kidney fibrosis, rat fatty liver transplantation, and mouse skeletal muscle injury—miRTalk elucidated the intricate, yet hitherto unseen, MiTI-mediated CCC. In summary, as the first of its kind, miRTalk paves the way for the computational assessment of EV-derived cellular crosstalk from scRNA-seq or ST data, offering insights into the understanding of intercellular dynamics from the aspect of EV-derived miRNA-mediated communication.

Results

Overview of the miRTalk method

First, we constructed an integrated EV-derived MiTI database named miRTalkDB as the fundamental reference. As depicted in Fig. 1a, we collected and manually verified the EV-derived miRNAs of human, mouse, and rat miRNAs from existing databases—ExoCarta [11], Vesiclepedia [12], EVmiRNA [13]—and primary literature, thereby ensuring the inclusion of miRNAs present within these established repositories that record miRNAs found in EVs or literature supporting the existence of miRNAs in EVs as the EV-derived miRNAs. By incorporating experimentally verified information on MiTIs derived from miRTarBase [14] and TarBase [15] as well as associated pathways of EV-derived miRNAs obtained from miRPathDB [16], the miRTalkDB included 2567, 935, and 179 EV-derived miRNAs, 1,595,450; 484,123; and 1626 MiTIs of humans, mice, and rats respectively, and a total of 8351 functional annotations of EV-derived miRNAs.

Next, we designed a custom framework to infer MiTI-mediated CCC based on scRNA-seq data (Fig. 1b). We reasoned that CCC mediated by negatively regulated MiTIs occurs when the module score of EV biogenesis and secretion (EVBS) signatures and the expression of a given miRNA are heightened in sender cells, concomitant with the low expression of the corresponding target genes and the significantly high module score of RNA-induced silencing complex (RISC)-related gene signature [17] in receiver cells representing the evidence showing the vesicles have been delivered into the receiver cells (Additional file 1:Fig. S1), as the module scoring strategy has been validated based on the experimental datasets with expected behavior [18,19,20,21,22] (Additional file 1:Fig. S2). In practice, to remove the stably expressed genes representing that they might not be regulated by miRNAs before the subsequent inference, we included widely expressed miRNAs and highly variable target genes, by referring to the miRNAs and their corresponding target genes in miRTalkDB. Then, we combined the product value of the expression of a given miRNA and the module score of EVBS signature with the product values of the ranked value of a given target gene based on the descending order of expression and the module score of RISC-related gene signatures to calculate the communication score for a given MiTI between pairwise cell types. By incorporating the permutation test, the significantly enriched MiTIs were obtained. Considering the happened activation of target genes by miRNAs, we used the ranked value of a given target gene based on the ascending order of expression and RNA-induced transcriptional activation complex (RITAC)-related gene signature [23, 24] to replace the RISC-related gene signature for the inference of the positively regulated MiTIs. This approach ensures that inferred target genes in receiver cells demonstrate the inhibitory or activated influence of EV-derived miRNA originating from sender cells. Since certain mature miRNAs are encoded by multiple miRNA genes, we selected the maximum score to evaluate EV-derived miRNA-mediated CCC, MiTIs, and specificity analysis (Fig. 1c,d). Through the integration of chord, circle, bubble, and heatmap plots, miRTalk enables the visualization of CCC mediated by EV-derived miRNA-target interactions from multiple perspectives, wherein the miRNA score mainly reflects the miRNA gene abundance in senders while the MiTI score basically represents the possibility of CCC mediated by the MiTI between senders and receivers. Given the widely existing negative regulation between miRNAs and their targets, we mainly focused on the inference and analysis of negatively regulated MiTIs.

Performance evaluation on simulated benchmarking datasets

First, we evaluate the performance of inference of MiTIs with negative and positive regulation by considering the different combinations of the score of EVBS, the expression of miRNA, the score of RISC, and the expression of the target gene. The true prediction is the high score of EVBS, the high expression of miRNA, the high score of RISC, and the low expression of target gene and other combinations are regarded as false predictions for the inference of MiTIs with negative regulation, while the true prediction is the high score of EVBS, the high expression of miRNA, the high score of RISC, and the high expression of target gene and other combinations are regarded as false predictions for the inference of MiTIs with positive regulation (Fig. 2a,b). In practice, the performance of miRTalk was evaluated with simulated benchmark scRNA-seq datasets generated by the ESCO R package [25]. By modulating key parameters such as gene number (1000, 800, 600, 400, and 200), ligands/receptors coverage (50%, 40%, 30%, 20%, and 10%), cell number (1000, 800, 600, 400, and 200), and sender/receiver ratio (8:2, 6:4, 5:5, 4:6, and 2:8), we constructed 20 different simulated datasets under four discrete contexts to systematically evaluate the performance of miRTalk. Additionally, we added the Pearson correlation and random approaches, namely miRTalk@P and miRTalk@R, respectively, as additional baseline methods and included them in the benchmarking of the simulated data. Consequently, miRTalk can achieve a high area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC) on the simulated data, outperforming miRTalk@P and miRTalk@R, indicating the reliability of miRTalk on the inference of CCC mediated by MiTIs with negative or positive regulation (Fig. 2c).

Next, we compared the performance of our method and other popular inference methods. Notably, we used NicheNet [26] as the baseline evaluation of the negatively regulated paracrine communication for it also enables the inference of ligand-target interactions (LTIs) by weighing potential ligands with their perturbed targets. In practice, we reasoned that gene–gene pairs distinguished by high expression in sender cells and low expression in receiver cells were true predictions, while the simultaneous high expression and simultaneous low expression of genes as well as interactions with low expression in sender cells and high expression in receiver cells were false ones, for the evaluation of miRTalk, miRTalk@P, miRTalk@, and NicheNet (Fig. 3a). Consequently, miRTalk significantly outperforms other methods in identifying the MiTIs with negative regulation with median AUROC and AUPRC reaching 0.97 and 0.99, respectively. In addition, we evaluated the specificity of MiTIs inferred by miRTalk regarding competitive situations. Among the true MiTIs with the high score of EVBS, the high expression of miRNA, the high score of RISC, and the low expression of the target gene, we simulated the true specificity by considering the expression of miRNAs that negatively regulate the same target and compared the Pearson correlation between the true specificity and inferred specificity by miRTalk for each target in receivers based on the true MiTIs (Additional file 1:Fig. S3a). Consequently, despite the low Pearson coefficient of specificity for some targets, most predictions by miRTalk showed a high correlation with the ground truth on the simulated data (Additional file 1:Fig. S3b).

Considering the negative regulation of miRNAs on their target genes, we also measured the anticorrelation between miRNAs in sender cells and target genes in receiver cells based on the simulated datasets (Fig. 3b). For the calculation of Pearson coefficient, the autocrine MiTI pairs were included as the baseline evaluation. As a result, most inferred MiTIs by miRTalk were significantly negatively correlated, for example, the inferred miRNA346-target180 interaction in the simulated dataset with Pearson coefficient and P value equal to 0.55 and 1.54e − 78. Despite a small portion of positively regulated MiTIs identified in each simulated dataset, most inferred MiTIs were significantly negatively correlated and the ratios of significant MiTIs with negative regulation were mostly more than 75% across the twenty simulated datasets, while CSmiR [27], a computational method to identify MiTIs at single-cell resolution, identified more than 50% significant negative MiTIs among most simulated datasets (Fig. 3c). Notably, miRTalk and CSmiR had a similar trend in the ratios of significant MiTIs with negative regulation on the simulated datasets with different cell number, ratio of senders and receivers, gene number, and miRNA/target coverage, with superior performance observed for our method in the inference of significant MiTIs with negative correlation.

Prediction of miRTalk consistent with the literature evidence on real datasets

To further validate that our proposed method enables the identification of biologically meaningful miRNAs and MiTIs with negative regulation, we performed an analysis by comparing the miRNA and MiTI score inferred by miRTalk from the RNA-seq data of cells and real levels of miRNAs inside their secreted EVs measured by miRNA-seq [28] (Fig. 4a). For each miRNA, the max MiTI score was used to perform the linear regression analysis. Consequently, the mature miRNAs with high levels in exosomes and microvesicles secreted by the human colorectal cancer cells exhibited significantly higher miRNA and MiTI scores in cells than that with low levels in EVs (Fig. 4b). In addition, a significant positive correlation between miRNA/MiTI scores and real miRNA levels in EVs was observed. Similar findings were also observed in another independent dataset (Additional file 1:Fig. S4).

Given the well-established role of oncogenic (onco) miRNAs and tumor suppressor (TS)-miRNAs within the TME, we reasoned that in the absence of external intervention, the score of onco-miRNAs in tumor samples will supersede that in normal samples, while the score of TS-miRNAs will decrease in tumor samples. Accordingly, we collected scRNA-seq datasets of 8 neoplastic specimens, derived from the bladder (BLCA) [29], cholangiocarcinoma (CHOL) [10], and ovarian (OV) cancers [30] for validation (Additional file 2). Specifically, miR- 24 - 3p is known to propagate tumorigenesis [31], while miR- 29a- 3p has been implicated in the suppression of malignant neoplasms in BLCA [32], as evidenced by the consistent trend of change in the cancer genome atlas (TCGA) BLCA cohort (Fig. 4c). Remarkably, our method significantly enriched miR- 24 - 3p and miR- 29a in BLCA cancer cells (CCs) in the scRNA-seq dataset (Fig. 4d, Additional file 3), with increased and decreased miRNA scores for onco-miR- 24 - 3p and TS-miR- 29a in tumor samples relative to normal samples in the bulk dataset (Fig. 4e), mirroring findings described within the literature. For CHOL, the exosomal miR- 24 - 3p, a well-established onco-miRNA in CCs, is known to target and adversely affect the proliferative and differentiation capacities of T cells [33, 34], consequently fostering tumor propagation and dissemination, as evidenced by the differential expression between normal and CHOL tumor samples (Fig. 4f). In contrast, macrophages secret exosomal miR- 142 - 3p which selectively targets and effectively suppresses the proliferation of CCs within the TME [35, 36]. Concordantly, the miR- 24 - 3p-mediated communication from CCs to T cells and the miR- 142 - 3p-mediated communication from MPs to CCs were significantly enriched based on the scRNA-seq dataset (Fig. 4g, Additional file 3), with consistently increased and decreased MiTI scores of onco-miR- 24 - 3p and TS-miR- 29a in tumor samples relative to normal samples in the bulk dataset (Fig. 4h).

Furthermore, we reasoned that within the TME, the onco-miRNA scores would experience notable reductions following effective therapeutic interventions. Hence, we employed miRTalk to infer the CCC mediated by EV-derived miRNAs in scRNA-seq data of OV containing the treatment-naïve and post-treatment samples, wherein the onco-miR- 24 - 3p [37] indeed saw a significant increase in expression from the normal to tumor states in OV (Fig. 4i). Consequently, our method significantly enriched onco-miR- 24 - 3p derived from OV CCs in the scRNA-seq dataset (Fig. 4j, Additional file 3), with significantly decreased miRNA scores for onco-miR- 24 - 3p in post-treatment samples relative to treatment-naïve samples in the bulk dataset (Fig. 4k), consistent with the literature evidence and our hypothesis. Additionally, we also compared the onco-miRNA scores and the levels of carbohydrate antigen 125 (CA125), a gold standard clinical tumor biomarker for monitoring patients with epithelial OV [30], and observed a significant positive correlation between them (Fig. 4l), further verifying our proposed hypnosis.

Oncogenic modulation of glioblastoma cells on the TME sensed by nonmalignant cells

We first applied miRTalk to a human glioblastoma (GBM) scRNA-seq dataset to explore the CCC mediated by EV-derived MiTIs with negative regulation within the TME, comprising malignant GBM cells, astrocytes (Astro), oligodendrocytes (Oligo), endothelial cells (Endo), oligodendrocyte precursor cells (OPC), neurons, and MPs [38] (Fig. 5a, Additional file 2). Consequently, intimate crosstalk within the GBM TME was detected, with malignant cells and MPs primarily functioning as prominent sources of miRNA secretion via EVs to other cells, rather than as receivers (Fig. 5b,c). This finding is compatible with the understanding that malignant cells modulate the TME towards tumor promotion through the secretion of various EV types [39]. Combined with the data from the cancer genome atlas (TCGA) GBM cohort, a total of 1462 MiTIs with matched miRNAs and target genes in the TCGA-GBM cohort were identified (Fig. 5d,Additional file 4). Among them, there are 130 onco-MiTIs and 171 TS-MiTIs exhibiting significant relevance with survival, wherein onco-MiTIs and TS-MiTIs showed a significant association with poor OS or DFS for the patients having high and low EV-derived MiTI scores, respectively (Additional file 1:Fig. S5a). Notably, the EV-derived MiTI scores inferred by our method for these onco-MiTIs and TS-MiTIs with significant survival relevance were higher than those MiTIs without significant survival relevance (Fig. 5e), and the EV-derived MiTI scores demonstrated its ability to separate high-risk and low-risk patients more significantly than using the expression level of the miRNA or their target gene alone (Fig. 5f). These findings suggested that miRTalk enables the identification of survival relevant MiTIs based on the EV-derived MiTI score.

Next, we performed comprehensive evaluations including the correlation between the expression of a miRNA and its target genes as well as the differential expression of miRNAs between normal and tumor samples of the TCGA-GBM cohort to identify the potential novel onco- and TS- miRNAs and associated MiTIs (Additional file 1:Fig. S5b). In detail, a total of 14 potential onco-MiTIs were identified with significantly higher expression of the miRNA in tumor samples than that in normal samples, significant association with poor OS or DFS for the patients having high MiTI scores, and significant negative correlation between the expression of the miRNA and its target gene, while 22 potential TS-MiTIs were found with significantly lower expression of the miRNA in tumor samples than that in normal samples, significant association with poor survival for the patients having low MiTI scores, and significant negative correlation between the expression of the miRNA and its target gene (Additional file 1:Fig. S5c, S6-S8). Although the inferred onco- and TS- miRNAs are known in cancers including GBM with supported literature, there are 8 potential novel onco- and TS-miRNAs identified in GBM. Concerning the identified MiTIs, only 7 MiTIs have been consistently reported in cancers except GBM. In other words, there are 29 potential novel onco- and TS- MiTIs identified in other cancers, and 36 potential novel onco- and TS- MiTIs identified in GBM (Additional file 4).

Next, we focused on the analysis of the MiTIs separately to understand the oncogenic modulation of glioblastoma cells on the TME sensed by nonmalignant cells, considering their different biological processes and pathways. For example, the literature-supported known interaction of onco-miR- 125 - 5p and its target gene PABPC1 [40] has been verified through the significant survival relevance, correlation, and differential expression with P = 0.0045 for the OS relevance, rho = 0.19 (P = 0.0069) for the negative correlation between the expression of the miR- 125 - 5p and its target gene PABPC1, and P = 0.00078 for the over-expressed miR- 125 - 5p in tumor samples (Fig. 5g). Through the secretion of miR- 125b- 5p via EVs, malignant GBM cells engaged in extensive interactions with neighboring cells including the OPC, neuron, Oligo, and Astro, through both autocrine and paracrine (Fig. 5h, Additional file 1: Fig. S9a). Furthermore, miR- 125b- 5p showed an association with the metabolic pathways, such as glucose metabolism, metabolism of lipids, arachidonic acid metabolism, and phospholipid metabolism, and the significant negative correlation between the scaled module scores of metabolic pathway signatures and the scaled expression of the target gene PABPC1 of miR- 125b- 5p further confirmed the significant association of miR- 125b- 5p with the metabolic pathways such as the glucose metabolism and metabolism of lipids (Fig. 5i, Additional file 1: Fig. S9b). The augmented metabolic processes intertwined with tumor progression have been extensively delineated within the TME [41], implying that malignant GBM cells exert oncogenic modulation on the TME sensed by non-malignant cells through the EV-derived miR- 125b- 5p-PABPC1 interaction.

It is known that Astro constitutes approximately half of the brain’s total cell population, and their pivotal role in GBM progression has been increasingly studied [39]. Upon activation in response to tumor-derived EVs, Astro can establish an immunologically advantageous environment for tumor cells, promoting further tumor growth and invasion. Through the specificity analysis, malignant cells and MPs were the main cell types that mainly secrets miR- 125b- 5p and miR- 221 - 5p, respectively, to negatively regulate the expression of PABPC1 in astrocytes (Additional file 1:Fig. S9c). Although the gene expression of miR- 221 - 5p was significantly enriched in MPs, insignificant negative correlations were observed between the expression of miR- 221 - 5p and PABPC1 in TCGA-GBM data (Additional file 1:Fig. S9 d), further highlighting that the inhibited expression of PABPC1 in Astro was specifically regulated by the miR- 125b- 5p derived from the malignant GBM cells.

Notably, the newly identified onco-miR- 222 - 3p and its target gene TP53BP2 have been also verified with P = 0.00057 for the OS relevance, rho = 0.20 (P = 0.0048) for the negative correlation, and P = 0.00078 for the over-expressed miR- 222 - 3p in tumor samples (Fig. 5j). From the aspects of the significance of survival relevance and correlation, the identified novel onco-miR- 222 - 3p and its target gene TP53BP2 showed more significant relevance with the development and prognosis of GBM compared to the known onco-miR- 125 - 5p and its target gene PABPC1 (Additional file 1:Fig. S6). Although a significant difference was observed between patients with high and low expression of miR- 222 - 3p (P = 0.014) or TP53BP2 (P = 0.035), the GBM patients with high MiTI scores showed more significant poor overall survivals than those with low MiTI scores (P = 0.00057), indicating the superiority of the MiTI score in separating high-risk and low-risk patients (Additional file 1:Fig. S9e). Collectively, these findings suggested a more valuable role for the newly identified potential onco-MiTI between onco-miR- 222 - 3p and its target gene TP53BP2 in GBM.

Considering the happened activation machinery between the miRNAs and their targets [42], we also investigated the MiTIs with positive regulation among the GBM cell types. Consequently, we identified the significantly enriched communication between malignant cells and MPs mediated by EV-derived miR- 106b- 3p and its target gene FN1, wherein the malignant cells and vascular cells specifically highly expressed MIR106B and FN1, respectively (Fig. 5k, Additional file 1:Fig. S9f). The metabolism-related pathways for miR- 106b- 3p were observed (Fig. 5l), and the prognosis analysis demonstrated the significant relevance of this interaction with the tumor recurrence (Fig. 5m). Notably, we observed an overlapped pathway, namely RAC1/PAK1/P38/MMP2 pathway that is highly associated with angiogenesis, related with miR- 106b- 3p and its target gene FN1 (Additional file 1:Fig. S9 g). This finding also partly explained the adverse prognosis associated with an intensified miR- 106b- 3p and FN1 interaction in GBM patients since angiogenesis, the recruitment of new blood vessels, is an essential component of tumor metastasis. In summary, these results showed that our method also enables the inference of biologically meaningful MiTIs with positive regulation.

Given the crucial role of ligands in mediating downstream target genes, we additionally perform the analysis of LTI-based CCC with NicheNet to further explore the relationship between the LTI-based and MiTI-based CCC. In detail, we first used NicheNet to infer the significant LTIs from malignant cells to endothelial cells and identified multiple ligands that can regulate the target gene FN1, which is also the target gene of miR- 106b with positive regulation inferred by miRTalk. By performing the correlation between the expression of ligands and miR- 106b, a total of 8 active ligands in malignant cells were identified, wherein the expression of POMZP3, MMP14, VEGFA, and FGF1 have a marked positive correlation with the expression of MiR106B (Additional file 1:Fig. S9 h). Based on the PPI network analysis for the inferred active ligands and their receptors as well as the target gene, several shared processes and pathways between the ligand-target interactions and the miR- 106 - 5p-FN1 interaction were significantly enriched, including the angiogenesis, VEGFA-VEGFR2 signaling, and RAC1/PAK1/P38/MMP2 pathway (Additional file 1:Fig. S9i). These findings indicated the regulatory effect of ligands and miRNAs in concert on the same target gene in receivers, highlighting the complementary advantages of LRI-based and MiTI-based CCC inference methods when studying the CCC mechanisms.

Characterization of signal transmissions forming the fibrogenic niche in the kidney

A murine scRNA-seq dataset of kidney fibrosis induced by unilateral ureteral obstruction (UUO) was used for exploring EV-derived MiTI-mediated CCC among vascular smooth muscle cells (VSMCs), injured VSMCs, mesangial cells, pericytes, parietal epithelial cells (PECs), and myofibroblasts [43] (Fig. 6a, Additional file 1: Fig. S10a, Additional file 2). As a result, intricate CCC mediated by EV-derived MiTIs with negative regulation were inferred in the uninjured and injured kidneys (Fig. 6b, Additional file 5), wherein numerous well-established miRNAs such as miR- 132 - 3p [44], miR- 27a- 3p [45], and miR- 425 - 5p [46] associated with renal fibrosis were identified with the significantly differential expression between the uninjured and injured conditions (Fig. 6c). By incorporating a bulk dataset of kidney fibrosis, we identified a total of 79 potential fibrosis-related MiTIs involving 5 miRNAs with the consistent trend of change in scRNA-seq and bulk-seq datasets for the inferred miRNAs (Fig. 6d,e, Additional file 1: Fig. S10b-c). Specifically, three miRNAs, i.e., miR- 132 - 3p, miR- 27a, and miR- 425 - 5p, were significantly overexpressed in the injured VSMCs and kidneys compared to the uninjured VSMCs and kidneys, respectively, mediating the CCC from the uninjured VSMCs to myofibroblasts (Fig. 6f). It is known that renal fibrosis is typified by the accumulation and deposition of extracellular matrix (ECM). While a multitude of cell types can synthesize ECM, it is widely held that myofibroblasts constitute the primary cellular component contributing to fibrosis in afflicted kidneys [47]. Aligning with this, the miR- 132 - 3p with high gene expression in the injury VSMCs was associated with multiple pathways such as the cell cycle and fibroblast proliferation, and its target gene Btg2 with low gene expression in myofibroblasts involved in the negative regulation of cell population proliferation (Fig. 6g,h), serving as a newly identified potential fibrogenic MiTI in the kidney.

As an essential component of resident glomerular cells, PECs have garnered considerable attention in recent years, fostering a more comprehensive understanding and delineation of their potential roles in renal pathology [48]. In this context, we studied the EV-derived MiTI-mediated communication from inferred sender cell types in the fibrotic kidney to PECs, and several miRNAs derived from the injured VSMC, myofibroblasts, and pericytes were inferred (Fig. 6i). Notably, the injured VSMC was the major contributor in negatively regulating the target genes in PECs by specifically secreting multiple EV-derived miRNAs, suggesting the paracrine-based CCC from the injured VSMCs to PECs in the kidney fibrosis. For example, we observed the elevated expression of miR- 27a- 3p particularly in the injured VSMCs, and inhibited expression of Epha4, a target gene of miR- 27a- 3p, in PECs (Fig. 6j). Based on the functional annotation, we found that both miR- 27a- 3p and its target gene Epha4 are related with the epithelial-mesenchymal transition (EMT). Given the negative regulation between miR- 27a- 3p and its target gene Epha4, the significant involvement of Epha4 in the negative regulation of EMT further demonstrated the association of MiTI between miR- 27a- 3p and Epha4 with the EMT for PECs (Fig. 6k,l). To further validate the transition from the PECs to myofibroblasts, a single-cell trajectory analysis employing monocle3 [49] was used to dissect cellular decisions for PECs, culminating in the reconstruction of a branching trajectory from the root to the end states (Fig. 6m). In alignment with this, the EMT scores intensified along the pseudotime axis, transitioning from PECs to myofibroblasts (Fig. 6n), which indicated that PECs contribute to the origin of myofibroblasts upon communication with injured VSMCs via the interaction of EV-derived miR- 27a- 3p and its target gene Epha4, serving as a newly identified potential fibrogenic MiTI in kidney.

Above all, most of the inferred MiTIs by our method for these fibrogenic and anti-fibrogenic miRNAs have not been reported to be related to fibrosis (Additional file 5), serving as the potential novel fibrogenic or anti-fibrogenic MiTIs in kidney by mediating the CCC mainly among the injured VSMCs, pericytes, PECs, and myofibroblasts, thereby promoting the establishment of the fibrosis-related cellular niche in kidney.

Identification of the granulocyte-hepatocyte communication discriminating between normal and fatty liver transplantation

We employed miRTalk on our previously published scRNA-seq data of rat normal and high-fat diet liver transplantation (LT) involving liver ischemia–reperfusion injury (IRI), which contains granulocytes, MPs, monocytes, DCs, hepatocytes, NK cells, Kupffer cells, T cells, Endo, and B cells [50] (Fig. 7a, Additional file 1: Fig. S11a, Additional file 2). Consequently, numerous EV-derived MiTIs among hepatic parenchymal and non-parenchymal cells, especially granulocytes, were inferred with a more intricate communication network observed for fatty livers (Fig. 7b, Additional file 6). Specifically, the normal and fatty transplanted livers shared most inferred EV-derived miRNAs with different miRNA scores under two conditions (Fig. 7c), suggesting the involvement of these miRNAs with the fatty liver IRI since the fatty transplanted livers have demonstrated severer IRI than the normal transplanted livers [50]. By incorporating a bulk dataset of mouse liver transplantation (LT), we identified a total of 8 potential fatty transplanted liver IRI-related MiTIs with a consistent trend of change in scRNA-seq and bulk-seq datasets for the inferred miRNAs, wherein the miR- 21 - 5p in granulocytes was significantly overexpressed in the fatty transplanted and post-LT livers, while the expression of miR- 223 - 3p in granulocytes and MPs, as well as miR- 290 in granulocytes, saw a significant decrease in the fatty transplanted and post-LT livers, compared to the normal transplanted and pre-LT livers, respectively (Fig. 7d-e).

As the predominant cell population in the liver, hepatocytes in normal and fatty transplanted livers preferred to receive EV-derived miRNAs from hepatic non-parenchymal cells including DCs, granulocytes, Kupffer cells, MPs, and Mono rather than secreting signals (Fig. 7f). Since there are two trends of change for miRNAs, we first focused on the significantly overexpressed ones. Notably, we identified a pervasive presence of miR- 21 - 5p within immune cells exerting an inhibitory effect on Smad7 expression of hepatocytes (Additional file 1:Fig. S11b-d), with increased gene expression and miRNA score of miR- 21 - 5p in the granulocytes of fatty transplanted livers. Concordantly, several overlapped pathways such as the TGF-beta signaling pathway were observed between miR- 21 - 5p and its target gene Smad7 (Fig. 7g). It is known that the Smad7 is a molecule known to negatively modulate the TGF-beta signaling cascade [51], and we observed hepatocytes residing in high-fat diet environments displayed a significantly accentuated TGF-beta pathway module score relative to those within normal diet livers (Fig. 7h). Moreover, elevated expression of Col1a1—a gene encoding the pro-alpha1 chains of type I collagen implicated in hepatic fibrosis onset—was detected in high-fat diet hepatocytes, and the gene set enrichment analysis (GSEA) [52] employing differentially expressed genes (DEGs) and the molecular signature database (MSigDB) [53] elucidates the heightened fibrotic response in high-fat livers via the significant enrichment of collagen-containing extracellular matrix (ECM) and ECM-receptor interaction gene signatures (Fig. 7i, Additional file 1: Fig. S11e). These findings revealed that steatotic perturbations in CCC mediated by EV-derived MiTIs play a vital role in shaping the hepatic susceptibility to fibrosis under high-fat diet conditions, in harmony with the fact that fatty transplanted livers exhibited severer IRI than the normal transplanted livers.

Intriguingly, granulocytes emerged as the predominant regulators of hepatocyte target genes in both normal and fatty transplanted livers. Given their status as the most abundant cellular population within the hepatic environment, granulocytes have been implicated in critical processes such as liver IRI and fibrosis [54]. Given the two significantly lowly expressed miRNAs in the fatty transplanted livers, we focused on the miR- 223 - 3p for its greater fold change between transplanted livers in comparison to miR- 290. Notably, miR- 223 - 3p expressed in granulocytes exerted a substantial inhibitory influence on hepatocytes by downregulating its target gene named Vim—a mediator implicated in ischemia–reperfusion injury (Fig. 7f and 7j). Concordantly, several overlapped pathways such as the apoptosis were observed between miR- 223 - 3p and its target gene Vim (Fig. 7k), and fatty transplanted livers exhibited significantly increased gene expression of Vim and significantly decreased module scores of the apoptosis signatures in hepatocytes, suggesting the anti-IRI and protective role of miR- 223 - 3p by inhibiting its target gene Vim of hepatocytes in the process of LT (Fig. 7l).

To further validate the protective role of miR- 223 - 3p-Vim interaction experimentally, we induced the liver IRI model by the oxygen–glucose deprivation/reperfusion (OGD/R) in rat BRL- 3 A hepatocytes and included four groups, namely the control group, OGD/R group, the agomir control treated OGD/R group, and the agomir miR- 223 - 3p treated OGD/R group (Fig. 7m). Consequently, we observed the survival rate of BRL- 3 A hepatocytes was significantly increased after treatment with miR- 223 - 3p, compared with that in the agomir negative control group. In addition, miR‐223‐3p exposed BRL- 3 A hepatocytes showed marked reductions in Vim mRNA expression compared with that in the agomir control group after OGD/R treatment, validating the negative regulation between miR- 223 - 3p and Vim with three representative binding sites identified (Fig. 7n). These findings further demonstrated the protective effect of miR- 223 - 3p by inhibiting the expression of Vim in hepatocytes, serving as an anti-IRI MiTI in the normal and fatty LT.

Inference of spatially resolved pro-generative communication in the mouse skeletal muscle injury

We collected a mouse ST dataset of the skeletal muscles under the uninjured and post-injury day 2, 5, and 7 conditions induced by notexin involving the uninjured myofiber, injured locus, and border [55], to explore the CCC mediated by EV-derived MiTIs among different spatial regions after the muscle injury (Fig. 8a, Additional file 2). Consequently, substantial MiTIs mediating the CCC among the uninjured myofiber, injured locus, and border were identified with the significantly differential expression between the uninjured and injured conditions at post-injury days 2, 5, and 7, especially at post-injury day 2 (Fig. 8b, Additional file 7). By referring to the corresponding bulk miRNA-seq dataset of the skeletal muscle, a total of 8 potential skeletal muscle injury/regeneration-related MiTIs involving 5 distinct miRNAs namely miR- 142a- 5p, miR- 145a- 5p, miR- 15b- 5p, miR- 221 - 3p, and miR- 674 - 5p were identified with the consistent trend of change in scRNA-seq and bulk-seq datasets for the inferred miRNAs (Fig. 8c,d, Additional file 1: Fig. S12a-b). Specifically, these miRNAs exhibited significantly increased expression in the different regions of muscles at post-injury day 2 compared to the uninjured region. Although the association of miR- 142a- 5p [56], miR- 145a- 5p [57], miR- 15b- 5p [58], and miR- 221 - 3p [59] with muscle injury or regeneration has been reported, the involvement of the identified MiTIs for these miRNAs with the skeletal muscle injury has not been studied, serving as the newly identified potential injury/regeneration-related MiTIs in the skeletal muscle.

For example, the expression of miR- 142a- 5p and its target gene Nmrk2 were significantly increased and decreased, respectively, in the injury regions especially the injury locus at post-injury day 2 compared to the uninjured muscles at post-injury day 2 (Fig. 8e,f). Based on the functional annotation for the highly expressed miR- 142a- 5p and lowly expressed Nmrk2, multiple pro-generative pathways were observed for the miR- 142a- 5p and its target gene Nmrk2, such as the Wnt signaling pathway, Vegfa-vegfr2 signaling pathway, and Rac1/pak1/p38/mmp2 pathway involved with the miR- 142a- 50 as well as the negative regulation of myoblast differentiation involved with the Nmrk2 negatively regulated by the miR- 142a- 5p. Notably, this MiTI between the miR- 142a- 5p and its target gene Nmrk2 specifically mediated the autocrine communication of the injury locus and the paracrine communication from the injury border to the injury locus (Additional file 1:Fig. S12c).

In contrast, the expression of miR- 145a- 5p and its target gene Ms4a6c were significantly increased and decreased, respectively, in the uninjured regions at post-injury day 2 compared to the injured regions at post-injury day 2 (Fig. 8g,h), with numerous pro-generative pathways observed for the miR- 145a- 5p and its target gene Ms4a6c whose expression showed significant associations with the module scores of the skeletal muscle tissue development and cell differentiation signatures involved with the functional annotation of its negative regulator miR- 145a- 5p (Fig. 8i–k). Different from the MiTI between the miR- 142a- 5p and its target gene Nmrk2, the MiTI between the miR- 145a- 5p and its target gene Ms4a6c specifically mediated the paracrine communication from the uninjured myofiber and injury locus to the injury border (Additional file 1:Fig. S12 d). Collectively, the two identified MiTIs, namely, miR- 142a- 5p-Nmrk2 and miR- 145a- 3p-Ms4a6c interactions, with cooperative and complementary effects in repairing the injury locus and injury border by mediating the communication among the different spatial regions, serving as the potential pro-generative MiTIs at the early stage after the skeletal muscle injury.

The spatial total RNA-sequencing (STRS) approach realizes the quantitation of mature miRNAs without poly-A tails at a spatial resolution [55], providing the potential for accounting for circulating miRNAs and the contribution of miRNAs secreted by other organs for the possible organ-organ communication mediated by EV-derived MiTIs, we included the circulating potential and their source based on their abundance in the blood and different organs, wherein the miRTalkDB contains 562, 195, and 120 potential circulating miRNAs of humans, mice, and rats, respectively, and 90 different organs for miRNAs. We first evaluated the potential circulating miRNAs based on the inferred MiTIs in the mouse skeletal muscle injury. As a result, the inferred miR- 484, miR- 142a- 5p, let- 7e- 5p, miR- 143 - 3p, and miR- 145a- 5p that regulate the target gene expression in receivers (injury locus) have relatively higher circulating scores compared to other miRNAs (Additional file 1:Fig.S12e), wherein all of them have been verified to exist in blood as circulating miRNAs. Notably, the association of circulating miR- 142a- 5p and miR- 145a- 5p with skeletal muscle injury has been widely verified in literature [60, 61], consistent with our findings that both exerted the pro-generative functions by regulating the target genes, i.e., Nmrk2 and Ms4a6c. Next, the main contributing tissues were identified from the aspects of organ-organ communication and circulating miRNAs by analyzing the related tissues for potential circulating miRNAs, the liver made a great contribution to the circulating miRNAs that regulate the muscle (Additional file 1:Fig. S12f), as evidenced by the widely reported liver-muscle axis mediated by non-coding RNAs [62]. These results revealed the crucial role of the inferred miR- 145a- 5p-Ms4a6c interactions in skeletal muscle injury and regeneration from the aspects of paracrine and long-range organ-organ communication via circulating miRNAs.

Discussion

In this study, we introduced a pioneering computational approach called miRTalk to infer EV-derived MiTI-mediated CCC for scRNA-seq or ST data. Incorporating a custom framework and an integrated database—miRTalkDB—containing EV-derived miRNA-target interactions for humans, mice, and rats, we attest to its novelty. Notably, besides the two basic criteriums for the miRNA and its target gene, we also consider the potential of producing EVs in sender cells and the activation of miRNA processing machinery in receiver cells to improve the inference of miRTalk. Assessment against both simulated and real-world datasets demonstrated the superior performance of miRTalk. Moreover, analyzing four independent datasets using miRTalk unveiled the in-depth EV-derived MiTI-mediated communicative mechanisms within the diseased contexts.

CCC, also known as cell–cell interaction, is an essential feature in multicellular organisms. With the widespread adoption of scRNA-seq data, it is customary to combine the abundance of ligands and receptors to infer CCC by reasoning that the simultaneous high expression of the ligand and receptor represents the potential occurrence of LRI-based CCC. Examples include CellChat [63], CellPhoneDB [64], and scCrossTalk [65]. Other approaches use downstream targets stimulated by LRIs in receiver cells to enhance the accuracy of inferred CCC, such as NicheNet [26], CytoTalk [66], CellCall [67], and exFINDER [68]. Furthermore, the spatial constraint has been widely incorporated for the inference of LRI-based CCC with spatially resolved transcriptomics data, such as SpaTalk [69], Giotto [70], SpaOTsc [71], spaCI [72], cell2cell [73], COMMOT [74], CCPLS [75], and SpaCET [76]. In LRI-based inference of CCC, the expression of the receptor is not regulated by the ligand from the biological aspect. Unlike the LRI-based inference methods, the hypothesis in our method is that the high expression of miRNA genes in senders and altered expression of target genes in receivers represents the potential occurrence of MiTI-based CCC, considering the regulatory effect of target gene expression by the EV-derived miRNAs. Therefore, we applied the strategy of the inclusion of highly variable targets to filter out the potentially unregulated target genes for improving the subsequent inference of communication mediated by EV-derived MiTIs.

As the false positive rate (FPR) is the vital index for a computational method, we used the human GBM data with survival rates for evaluating the FPR. Interestingly, there is no significant difference in identifying the survival reverent MiTIs (FPR: 65–75%) with the MiTIs inferred by miRTalk, miRTalk@P, and miRTalk@R, although their FPRs were slightly lower than that with all MiTIs recorded in miRTalkDB (Additional file 1:Fig. S13a). Notably, we also observed a similar ratio of the survival-relevant LRIs (FPR: 65–75%) with all LRIs recorded in NicheNetDB and CellTalkDB as well as the LRIs inferred by NicheNet and scCrossTalk (Additional file 1:Fig. S13b). These results suggested an approximate range of the FPR for the MiTI-based or LRI-based CCC inference methods in identifying the survival-relevant interaction pairs, regardless of the underlying algorithms.

For the popular poly-T-based scRNA-seq technologies such as 10X Genomics [77], the potential failure scenarios of our method might exist for the scRNA-seq data obtained from the poly-T-based technologies, since the sequenced miRNAs mainly refer to the primary miRNA with a poly-A tail. Besides, relying solely on RISC or RITAC machinery could introduce bias in miRNA regulation, but the incorporation of the second-degree target genes (i.e., targets of miRNA target genes) might be a solution. Additionally, since the other signaling processes such as secreted protein ligands on target genes are not considered in our method, the integrated analysis of our MiTI-based method and other LTI-based methods might be an alternative solution.

Given a recent study showing that sequence motifs within miRNAs, namely EXOmotifs including CAUG and CGGGAG, can predict their secretion in EVs [1], we also evaluated this criterium and observed only a few miRNAs in our database contain the EXOmotifs [78, 79] (Additional file 1:Fig. S14, Additional file 8). The reason might be that different EXOmotifs exist in different cell types rather than the two identified CAUG and CGGGAG motifs. Future advancements in the identification of more universal and efficient EXOmotifs will improve the inference of EV-derived miRNAs.

To ensure the accurate inference of CCC within a specific context, it is a common practice to construct a tailored foundational database in the development of a novel inference method. For instance, CellPhoneDB integrates the subunit architecture of both ligands and receptors, representing heteromeric complexes, while CellChatDB encompasses various cofactor types, such as soluble agonists and antagonists, as well as co-stimulatory and co-inhibitory membrane-bound receptors. To accommodate the inference of metabolite-dependent intercellular communication and neurotransmitter-dependent neuron-neuron interactions, MEBOCOST [80] and NeuronChat [81] curated databases of metabolite-sensor partnerships and ligand-target interactions, respectively. For the accurate inference of spatially resolved intercellular communication, the SpaTalk method relies on the foundational CellTalkDB [82] and transcription factor (TF)-target interaction database. Consequently, we constructed an integrated EV-derived MiTI database, termed miRTalkDB, by incorporating experimentally verified information and the functional annotation of miRNAs.

Despite the coverage of miRNAs in datasets produced by prevalent scRNA-seq methodologies being rather sparse, miRTalk enables the inference of EV-derived miRNAs from these scRNA-seq data. Notably, the recently reported snRandom-seq demonstrated enhanced coverage of non-coding genetic elements, such as miRNAs, within their scRNA-seq data by procuring full-length total RNAs via random primers [83]. The STRS approach can also capture coding RNAs and noncoding RNAs. These technologies have realized the quantitation of mature miRNAs without poly-A tails at single-cell or spatial resolution [55], providing the potential for accounting for circulating miRNAs and the contribution of miRNAs secreted by other organs for the possible organ-organ communication mediated by EV-derived MiTIs as described in the case study of the mouse skeletal muscle injury. The ongoing expansion of scRNA-seq data, characterized by augmented accuracy, abundance, and coverage of miRNAs through such innovative technologies, will undoubtedly facilitate the practicality of miRTalk across a wide spectrum of applications in the fields of biology and biomedicine.

Conclusions

In summary, miRTalk represents the pioneering method for inferring EV-derived miRNA-mediated CCC from scRNA-seq or ST data by incorporating a constructed database miRTalkDB including EV-derived miRNA-target associations for humans, mice, and rats, offering new insights into the understanding of intercellular dynamics underlying biological processes from the aspect of CCC mediated by EV-derived miRNAs.

Methods

Data processing

For the analyses of human ovarian cancer investigations, patient participants with platinum-free intervals exceeding 20 days and corresponding peritoneum and omentum samples containing a minimum of 100 epithelial ovarian cancer cells in the scRNA-seq data were retained. The genes were revised in accordance with the National Center for Biotechnology Information gene database (https://www.ncbi.nlm.nih.gov/gene/), in which unmatched and duplicated genes were removed. For scRNA-seq, the normalized data with the log-normalization of Seurat [84] was used for the subsequent analysis, while the log2 normalization was applied for the bulk miRNA-seq and RNA-seq data.

Construction of miRTalkDB

The experimentally validated MiTIs of humans, mice, and rats were collected from the well-known databases named miRTarBase [14] (https://mirtarbase.cuhk.edu.cn/) and TarBase [15] (https://dianalab.e-ce.uth.gr/tarbasev9). EV-derived miRNAs were included by examining their presence in the following established repositories: (1) ExoCarta [11] (http://exocarta.org/), an exosome database furnishing the content discerned in exosomes across various species; (2) Vesiclepedia [12] (http://microvesicles.org/), a meticulously curated repository of molecular information identified in distinct classes of EVs, such as ectosomes, microvesicles, exosomes, or apoptotic bodies; (3) EVmiRNA [13] (http://bioinfo.life.hust.edu.cn/EVmiRNA#!/), a database of miRNA expression patterns in EVs. For miRNAs not present within these databases, we conducted text mining within PubMed employing the use of critical keywords:"[miRNA name or aliases] AND (extracellular vesicles)", followed by manual validation to include EV-derived miRNAs with literature evidence demonstrating their presence in EVs. By integrating the EV-derived miRNAs, known MiTIs, and functional annotations of miRNAs, we constructed the miRTalkDB as the foundational database to infer EV-derived miRNA-mediated CCC. The functional annotations for EV-derived miRNAs involving biological processes (BP) of GO, KEGG, Reactome, and Wikipathways were downloaded from miRPathDB 2.0 [16] (https://mpd.bioinf.uni-sb.de/). The circulating miRNAs and the abundance of miRNAs in different organs were obtained from the sncRNA Zoo database [85], a repository for circulating small noncoding RNAs in animals (https://www.ccb.uni-saarland.de/asra/), and TarBase, respectively.

Algorithm of miRTalk

There are two components in miRTalk, namely the identification of MiTIs and the specificity calculation of identified MiTIs. The former is the core component that aims to infer the EV-derived MiTIs that facilitate CCC from senders to receivers based on the potential of producing EVs and the expression of miRNAs in senders as well as the activation of miRNA processing machinery and the expression of target genes in receivers, while the latter focuses on weighing the regulatory contribution of cell-type-specific miRNAs in senders on the same target gene in receivers.

Identification of MiTIs

Step 1: preprocessing miRNA and target genes

For miRNA genes, given the sparse expression of miRNA genes in scRNA-seq or ST data, the widely expressed miRNA genes expressed in at least 5% of cells with a minimum of 10 expressed cells were included to represent that they are truly expressed. For target genes, to filter out the potentially unregulated target genes for improving the subsequent inference of communication mediated by EV-derived MiTIs, we included the highly variable target genes within a given scRNA-seq data matrix $G[g\times c]$ ($g$ genes and $c$ cells) with $k$ cell types. Let ${G}_{t}$ represent the highly variable target gene set, which can be calculated by the following function:

$${G}_{t}= {G}_{hvg}\cup {G}_{deg}^{k}$$

Herein, ${G}_{hvg}$ refers to the highly variable genes (HVGs) obtained from $G$ utilizing the local polynomial regression (loess) and the variance stabilizing transformation method in Seurat, and ${G}_{deg}^{k}$ denotes the combined differentially expressed genes (DEGs) for $k$ cell types, employing the Wilcoxon rank-sum (WRS) test, along with a minimal log-scale fold change of 0.25 in conjunction with a minimum of 10% expression in cells.

Step 2: scoring MiTIs between pairwise cell types

For a sender cell type $A$ with $a$ cells and $g$ genes and a receiver cell type $B$ with $b$ cells and $g$ genes, we defined the communication score ${S}_{i-j}^{A-B}$ of a given MiTI as follows:

$${S}_{i-j}^{A-B}= {S}_{i}^{A}\bullet {S}_{j}^{B}$$

For ${S}_{i}^{A}$, it represents the score of secreting the EV-derived miRNA $i$ by the sender cell type $A$. In practice, we combined the min–max scaled Seurat module score of EVBS gene ontology signature (GO:0140112 Extracellular Vesicle Biogenesis) and the expression levels of the miRNA gene to calculate ${S}_{i}^{A}$, as detailed below:

$${S}_{i}^{A}=\frac{1}{a}\bullet {\sum }_{\widetilde{a}}{\varphi}_{i}^{m}(m\in \widetilde{a})$$

where $\varphi_{i}^{m}$ means the product value of the module score ${S}_{EV}^{m}$ of EVBS gene ontology signature and the expression level ${E}_{i}^{m}$ of the miRNA $i$ for a given cell $m$, and $\widetilde{a}$ represents the sender cells that have higher product value than the average product value $\overline{\varphi}_{i}$ of other cells. The function for calculating $\varphi_{i}^{m}$ can be written as:

$$\varphi_{i}^{m}={E}_{i}^{m}\bullet {S}_{EV}^{m}$$

For ${S}_{j}^{B}$, it represents the score of regulating the target gene $j$ in the receiver cell type $B$. In practice, we combined the min–max scaled Seurat module score of RISC-related gene signature (AGO1, AGO2, AGO3, AGO4, HSPA4, HSP90 AA1, DICER1, HSP90 AB1, HSPA8, HOPX, DNAJA2, PTGES3), RITAC-related gene signature (AGO2, DHX9, PAF1, LEO1, CTR9, CDC73, RTF1, SKIC8, SUPT4H1, SUPT5H, SUPT16H, SSRP1, TCEA1, TCEA2, TCEA3), and the expression levels of the target gene to calculate ${S}_{j}^{B}$, as detailed below:

$${S}_{j}^{B}=\frac{1}{b}\bullet {\sum }_{\widetilde{b}}{\uptau }_{j}^{n}(n\in \widetilde{b})$$

For the inference of MiTIs with negative regulation, the ${\uptau }_{j}^{n}$ means the product value of the module score ${S}_{RISC}^{n}$ of RISC-related gene signature and the scaled rank value of the target gene $j$ for a given cell $n$. For the inference of MiTIs with positive regulation, the ${\uptau }_{j}^{n}$ means the product value of the module score ${S}_{RITAC}^{n}$ of RITAC-related gene signature and the scaled rank value of the target gene $j$ for a given cell $n$. The $\widetilde{b}$ represents the receiver cells that have higher product value than the average product value ${\overline{\uptau } }_{j}$ of other cells. Considering there are two types of regulation between miRNAs and target genes, we calculate the rank value ${D}_{j}^{n}$ and ${A}_{j}^{n}$ of the target gene $j$ involving negative and positive regulation, respectively, by the following equations:

$${N}_{j}^{n}={D}_{j}^{n}/\text{max}({D}^{n})$$

$${P}_{j}^{n}={A}_{j}^{n}/\text{max}({A}^{n})$$

where ${D}^{n}$ and ${A}^{n}$ represent the rank values by sorting the expression levels of all genes in descending and ascending order, respectively, in a given receiver cell $n$, and the function for calculating ${\uptau }_{j}^{n}$ based on the scaled rank value ${N}_{j}^{n}$ and ${P}_{j}^{n}$ for the target gene $j$ can be written as:

$${\uptau }_{j}^{n}={N}_{j}^{n}\bullet {S}_{RISC}^{n} (Negative regulation)$$

$${\uptau }_{j}^{n}={P}_{j}^{n}\bullet {S}_{RITAC}^{n} (Positive regulation)$$

Next, a permutation test was performed by randomizing the cell labels to re-calculate the communication score. By repeating this step 1000 times, the communication scores ${S}^{per}$ of a given MiTI from the sender cell type $A$ to the receiver cell type $B$ were obtained for comparison with the real communication score ${S}_{i-j}^{A-B}$ to calculate the P value, wherein the MiTIs with a P value less than 0.05 were identified as the significantly enriched ones between the pairwise cell types.

Specificity calculation of identified MiTIs

For a given miRNA $i$ and a given cell type $A$ with $a$ cells that are recorded in the identified MiTIs between pairwise cell types, we calculate the miRNA score ${S}_{i}^{A}$ by the following function:

$${S}_{i}^{A}={\overline{E} }_{i}^{A}\bullet \frac{\sum_{m=1}^{a}{E}_{i}^{m}}{\sum_{m=1}^{c}{E}_{i}^{m}}$$

where ${\overline{E} }_{i}^{A}$ means the average expression value of the miRNA $i$ in the cell type $A$, and ${E}_{i}^{m}$ represents the expression value of the miRNA $i$ in a given cell $m$. Next, for a given target gene $j$ and a given receiver cell type $B$ that recorded in the identified MiTIs between pairwise cell types, we calculated the specificity ${\widetilde{S}}_{i-j}^{A-B}$ regarding a given target gene in a receiver cell type by the following function:

$${\widetilde{S}}_{i-j}^{A-B}=\frac{{S}_{i}^{A}}{\sum {S}_{x-j}^{X-B}}$$

where $X$ and $x$ mean all sender cell types and all cell-type-specific miRNAs that can regulate the target gene $j$, respectively, involving the identified MiTIs from sender cell types to receiver cell types.

Simulation of scRNA-seq data for benchmarking

The ESCO [25] R package was utilized to simulate the benchmarked datasets by variegating the gene and cell number. By setting parameters such as'nGenes = 1,000','nCells = 1,000','group.prob = c(0.5, 0.5)','deall.prob = 0.3','de.prob = c(0.5, 0.5)','de.facLoc = c(1.9, 2.5)','withcorr = TRUE', and'trials = 1', the initially generated scRNA-seq data matrix of 1000 gene and cell pairings were derived, which is composed of a sender cell type containing 510 cells, and a receiver cell type containing 490 cells. For testing miRTalk, 1000 genes were bifurcated into 500 miRNA genes matching with the remaining 500 target genes. For the inference of MiTIs with negative regulation, the true predictions represent the significantly high expression of miRNA genes and EVBS scores in sender cells, paired with the significantly low expression of target genes and significantly high RISC scores in receiver cells, while other fifteen combinations with the expression of miRNA genes and EVBS scores in sender cells, as well as the expression of target genes and RISC scores in receiver cells, were false predictions. For the inference of MiTIs with positive regulation, the true predictions represent the significantly high expression of miRNA genes and EVBS scores in sender cells, paired with the significantly high expression of target genes and RITAC scores in receiver cells, while other fifteen combinations with the expression of miRNA genes and EVBS scores in sender cells as well as the expression of target genes and RITAC scores in receiver cells were false predictions. The top 25 significantly highly expressed genes in sender cells as well as the top 12 and 15 significantly highly expressed genes in receiver cells were regarded as the EVBS, RISC, and RITAC signature genes, consistent with the numbers in miRTalkDB.

Performance evaluation with simulated datasets for miRTalk

Through the variation of key parameters such as gene number (1000, 800, 600, 400, and 200), ligands/receptors coverage (50%, 40%, 30%, 20%, and 10%), cell number (1000, 800, 600, 400, and 200), and sender/receiver ratio (8:2, 6:4, 5:5, 4:6, and 2:8), we simulated 20 different datasets under four distinct contexts. This allowed us to test the performance of miRTalk with different baseline methods. For the correlation baseline, namely miRTalk@P, we use the significantly correlated MiTIs across sender and receiver cells, whose score is calculated by replacing the product of the miRNA and its target gene with the absolute Pearson coefficient while retaining the other parts same as the that in miRTalk. For the random baseline, namely miRTalk@R, we use the random scores based on the miRTalk predictions for the inference of significantly correlated MiTIs.

Comparison with other methods

We let the simulated datasets with the EVBS, RISC, and RITAC scores equal to 1 for the comparison of miRTalk with other known baseline methods using the same simulated MiTIs databases. For the NicheNet baseline, the default parameters including all inferred ligands were used and the weight was regarded as the probability. For the CSmiR baseline, all parameters were kept default. AUROC and AUPRC were calculated to evaluate the performance of predictions with the precrec [86] R package based on the probability values and true labels for each benchmarked method.

Bulk data analysis

For the bulk RNA-seq datasets about the EVBS as well as receiving EVs, the"AddModuleScore"function in Seurat was used to calculate the module scores of the EVBS signatures and RISC-related gene signatures. For the bulk miRNA-seq datasets, the miRNA score was calculated by the product of the expression and the ratio of the expression against the sum of expression in the corresponding paired samples for a given miRNA according to the principle used in miRTalk for calculating the miRNA score. For the bulk datasets with miRNA-seq and RNA-seq, the EV-derived MiTI score in a given sample was calculated by the product of the expression of a given miRNA, the min–max scaled module score of EVBS gene signatures, the scaled rank value of a given target gene, and the min–max scaled module score of RISC-related gene signatures. Specifically, the scaled rank value of a target gene was obtained with the rank value of the descending ordered gene expression divided by its max rank value for a given sample, representing the negative regulation between a given miRNA and its target gene. For the analysis of the association between the EV-derived MiTI and survival, we separated the patients into two categories based on the median EV-derived MiTI score and analyzed their OS and DFS relevance using the survival and survminer R packages with default parameters. In detail, the Log-rank test was used to evaluate the significant difference between two curves based on the Kaplan–Meier analysis. The correlation analysis was based on the min–max scaled values of two variables. The MiTI score was calculated directly by the product of the miRNA expression and the scaled rank value of target expression in descending order for the comparison of MiTI scores between normal and tumor samples, considering the influence of the module score of EVBS on evaluating the TS-MiTIs.

Benchmarking with false positive rates

For the human GBM scRNA-seq data, the MiTIs and LRIs between pairwise cell types were inferred with default parameters for miRTalk, NicheNet, and scCrossTalk. For the TCGA-GBM cohort data, the MiTI score is based on the product of miRNA expression and the scaled rank value of target expression in descending order, while the LRI score is based on the product of ligand expression and receptor expression. Based on the high-risk and low-risk GBM patients separated by the median MiTI or LRI score, the relevance of a given MiTI or LRI pair with the OS or DFS was obtained using the Log-rank test.

Pathway and biological process enrichment

The GSVA [87] R package and the"AddModuleScore"function in Seurat were utilized to compute the module score of pathways, GO biological processes, and known hallmarks. To obtain the signatures, the Molecular Signatures Database (MSigDB, http://www.gsea-msigdb.org/gsea/msigdb) was accessed through msigdbr. GSEA was conducted by clusterprofiler [52] with the ordered gene list and MSigDB [53]. The STRING [88] web tool (https://string-db.org/) was used to analyze the protein–protein interaction network and the enriched GO processes and KEGG pathways.

Single-cell trajectory analysis

Mouse normal and fibrotic kidney cells underwent pre-processing and analysis with monocle3 [49] using default parameters to generate a single-cell trajectory and dissect cellular decisions, where uninjured VSMCs were considered as the root to perform the pseudo-time analysis.

Cell culture and oxygen–glucose deprivation/reperfusion (OGD/R) model

We obtained the rat BRL- 3 A hepatocytes from the Cell Bank of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences. Cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, USA) supplemented with 10% fetal bovine serum (FBS, Gibco, USA) and a 1% penicillin–streptomycin mixture (100 × Penicillin–Streptomycin Solution, Gibco, USA) at 37 °C with 5% CO₂. BRL- 3 A cells were cultured with DMEM sugar-free medium (Gibco, USA) at 37 °C in a closed hypoxia chamber filled with N₂ in a tri-gas incubator (94% N₂, 5% CO₂, 1% O₂) for 12 h, followed by reoxygenation in normal culture conditions 3 h to establish the OGD/R injury model.

Cell viability assay

Agomir miR‐223‐3p and the negative control (NC) were purchased from Guangzhou RiboBio Co., Ltd. (China). The sequences are listed in Additional file 8. The cell viability was detected by CCK- 8 (Beyotime, China) assay. BRL- 3 A cells were seeded in a 96-well plate at a density of 1 × 10⁴cells/well. Four groups were established as follows: normal culture, OGD/R, OGD/R + agomir control, and OGD/R + agomir miR- 223 - 3p. The control group was cultured with normal conditions, while the other groups were performed following the OGD/R treatment. Among them, the OGD/R + agomir control group and the OGD/R + agomir miR- 223 - 3p group were simultaneously treated with 200 nM agomir control or 200 nM agomir miR- 223 - 3p respectively for 12 h at the beginning of OGD/R. Then, the medium containing 10% CCK- 8 reagent was added to each well. After incubation for 1.5 h. The absorbance of the solution was measured at 450 nm using a fluorescent plate reader (SpectraMax M5, Molecular Devices, USA).

RNA extraction and RT‐qPCR

BRL- 3 A cells were cultured in 6-well plates at a density of 3 × 10⁶cells/well for 24 h. Subsequently, four groups were established as described above. After reperfusion for 3 h, the total RNA from cells was extracted using RNA-Quick Purification Kit (SE Science, China), and the cDNA synthesis was performed with Hifair® AdvanceFast One-step RT-gDNA Digestion SuperMix for qPCR (Yeasen, China) according to the manufacturer’s guidance. We quantified gene‐specific transcripts (Additional File 8; Sango, China) in qPCR assay with the Hieff® qPCR SYBR Green Master Mix (No Rox) (Yeasen, China) and expressed the data relative to β-actin by calculating the fold change with the 2^−ΔΔCt method.

Statistical analysis

R (version 4.1.1 and 4.1.3) and GraphPad Prism 8.0.1 were used for the statistical analyses. Differences between the two groups were determined using the Wilcoxon test or T-test; P < 0.05 was considered to indicate a significant difference.

Data availability

The previously published scRNA-seq data can be accessed through public data resources. The bulk RNA-seq datasets about the EVBS as well as receiving EVs were downloaded from the Gene Expression Omnibus (GEO) repository: GSE165557 [89], GSE215308 [90], GSE153581 [91], GSE129883 [92], GSE135187 [93], and GSE155881 [94]. The bulk datasets with RNA-seq and miRNA-seq were obtained from GSE125905 [95] and GSE76711 [96]. The bulk miRNA-seq datasets about the quantification of miRNAs in EVs were collected from GSE240294 [97] and GSE121964 [98]. The bulk miRNA-seq datasets involving mouse kidney fibrosis, rat LT, and mouse skeletal muscle injury can be accessed from GSE162794 [99], GSE56071 [100], and GSE200480 [101], respectively. The bulk RNA-seq and miRNA-seq datasets of normal and tumor samples for BLCA, CHOL, GBM, and OV were retrieved from the TCGA cohort with the R package TCGAbiolinks [102]. The GBM datasets containing the miRNA and mRNA expression profiles as well as clinical information such as the OS and DFS were directly downloaded from the cBioPortal (https://www.cbioportal.org/datasets) web resource [103]. The scRNA-seq dataset of cholangiocarcinoma in humans was obtained from GSE138709 [104]. The human scRNA-seq datasets of bladder cancer, glioblastoma multiforme, and ovarian cancer were curated from the TISCH2 [105] database (http://tisch.comp-genomics.org/) with the accession numbers GSE130001 [106], GSE84465 [107], and GSE165897, respectively. The mouse scRNA-seq data of normal and fibrotic kidneys was procured from the Zenodo data archive (https://zenodo.org/record/4059315) [108]. The previously published scRNA-seq data of normal and fatty liver transplantation was deposited in the Genome Sequence Archive (GSA) in the National Genomics Data Center (NGDC) under the accession number CRA004061 [109]. All other relevant data supporting the key findings of this study are available within the article and its Additional files or from the corresponding author upon reasonable request. Source codes for the miRTalk R package, the related scripts, and the miRTalkDB database are available at Github (https://github.com/multitalk/miRTalk) [110] and Zenodo (https://zenodo.org/records/13856217) [111] under GPL- 3.0 license.

References

Garcia-Martin R, Wang G, Brandao BB, Zanotto TM, Shah S, Kumar Patel S, Schilling B, Kahn CR. MicroRNA sequence codes for small extracellular vesicle release and cellular retention. Nature. 2022;601:446–51.
Article CAS PubMed Google Scholar
Makarova J, Turchinovich A, Shkurnikov M, Tonevitsky A. Extracellular miRNAs and cell-cell communication: problems and prospects. Trends Biochem Sci. 2021;46:640–51.
Article CAS PubMed Google Scholar
Mori MA, Ludwig RG, Garcia-Martin R, Brandao BB, Kahn CR. Extracellular miRNAs: from biomarkers to mediators of physiology and disease. Cell Metab. 2019;30:656–73.
Article CAS PubMed PubMed Central Google Scholar
Isaac R, Reis FCG, Ying W, Olefsky JM. Exosomes as mediators of intercellular crosstalk in metabolism. Cell Metab. 2021;33:1744–62.
Article CAS PubMed PubMed Central Google Scholar
Wang J, Li L, Zhang Z, Zhang X, Zhu Y, Zhang C, Bi Y. Extracellular vesicles mediate the communication of adipose tissue with brain and promote cognitive impairment associated with insulin resistance. Cell Metab. 2022;34(1264–1279): e1268.
Google Scholar
He Y, Rodrigues RM, Wang X, Seo W, Ma J, Hwang S, Fu Y, Trojnar E, Matyas C, Zhao S, et al: Neutrophil-to-hepatocyte communication via LDLR-dependent miR-223-enriched extracellular vesicle transfer ameliorates nonalcoholic steatohepatitis. J Clin Invest. 2021;131:e141513.
Sun F, Li H, Sun D, Fu S, Gu L, Shao X, Wang Q, Dong X, Duan B, Xing F, et al. Single-cell omics: experimental workflow, data analyses and applications. Sci China Life Sci. 2024;24:5–102.
Google Scholar
Isakova A, Neff N, Quake SR. Single-cell quantification of a broad RNA spectrum reveals unique noncoding patterns associated with cell types and states. Proc Natl Acad Sci U S A. 2021;118:e2113568118.
Article CAS PubMed PubMed Central Google Scholar
Ramanujam D, Schon AP, Beck C, Vaccarello P, Felician G, Dueck A, Esfandyari D, Meister G, Meitinger T, Schulz C, Engelhardt S. MicroRNA-21-dependent macrophage-to-fibroblast signaling determines the cardiac response to pressure overload. Circulation. 2021;143:1513–25.
Article CAS PubMed PubMed Central Google Scholar
Zhang M, Yang H, Wan L, Wang Z, Wang H, Ge C, Liu Y, Hao Y, Zhang D, Shi G, et al. Single-cell transcriptomic architecture and intercellular crosstalk of human intrahepatic cholangiocarcinoma. J Hepatol. 2020;73:1118–30.
Article CAS PubMed Google Scholar
Keerthikumar S, Chisanga D, Ariyaratne D, Al Saffar H, Anand S, Zhao K, Samuel M, Pathan M, Jois M, Chilamkurti N, et al. ExoCarta: a web-based compendium of exosomal cargo. J Mol Biol. 2016;428:688–92.
Article CAS PubMed Google Scholar
Pathan M, Fonseka P, Chitti SV, Kang T, Sanwlani R, Van Deun J, Hendrix A, Mathivanan S. Vesiclepedia 2019: a compendium of RNA, proteins, lipids and metabolites in extracellular vesicles. Nucleic Acids Res. 2019;47:D516–9.
Article CAS PubMed Google Scholar
Liu T, Zhang Q, Zhang J, Li C, Miao YR, Lei Q, Li Q, Guo AY. EVmiRNA: a database of miRNA profiling in extracellular vesicles. Nucleic Acids Res. 2019;47:D89–93.
Article CAS PubMed Google Scholar
Huang HY, Lin YC, Cui S, Huang Y, Tang Y, Xu J, Bao J, Li Y, Wen J, Zuo H, et al. miRTarBase update 2022: an informative resource for experimentally validated miRNA-target interactions. Nucleic Acids Res. 2022;50:D222–30.
Article CAS PubMed Google Scholar
Skoufos G, Kakoulidis P, Tastsoglou S, Zacharopoulou E, Kotsira V, Miliotis M, Mavromati G, Grigoriadis D, Zioga M, Velli A, et al: TarBase-v9.0 extends experimentally supported miRNA-gene interactions to cell-types and virally encoded miRNAs. Nucleic Acids Res. 2024;52:D304-D310.
Kehl T, Kern F, Backes C, Fehlmann T, Stockel D, Meese E, Lenhof HP, Keller A. miRPathDB 2.0: a novel release of the miRNA pathway dictionary database. Nucleic Acids Res. 2020;48:D142–7.
Article CAS PubMed Google Scholar
Iwakawa HO, Tomari Y. Life of RISC: formation, action, and degradation of RNA-induced silencing complex. Mol Cell. 2022;82:30–43.
Article CAS PubMed Google Scholar
Koo D, Cheng X, Udani S, Baghdasarian S, Zhu D, Li J, Hall B, Tsubamoto N, Hu S, Ko J, et al. Optimizing cell therapy by sorting cells with high extracellular vesicle secretion. Nat Commun. 2024;15:4870.
Article CAS PubMed PubMed Central Google Scholar
Morimoto Y, Yamashita N, Daimon T, Hirose H, Yamano S, Haratake N, Ishikawa S, Bhattacharya A, Fushimi A, Ahmad R, et al. MUC1-C is a master regulator of MICA/B NKG2D ligand and exosome secretion in human cancer cells. J Immunother Cancer. 2023;11:e006238.
Article PubMed PubMed Central Google Scholar
Umezu T, Tsuneyama K, Kanekura K, Hayakawa M, Tanahashi T, Kawano M, Taguchi YH, Toyoda H, Tamori A, Kuroda M, Murakami Y. Comprehensive analysis of liver and blood miRNA in precancerous conditions. Sci Rep. 2020;10:21766.
Article CAS PubMed PubMed Central Google Scholar
Garcia-Silva S, Benito-Martin A, Nogues L, Hernandez-Barranco A, Mazariegos MS, Santos V, Hergueta-Redondo M, Ximenez-Embun P, Kataru RP, Lopez AA, et al. Melanoma-derived small extracellular vesicles induce lymphangiogenesis and metastasis through an NGFR-dependent mechanism. Nat Cancer. 2021;2:1387–405.
Article CAS PubMed PubMed Central Google Scholar
Yin X, Zeng W, Wu B, Wang L, Wang Z, Tian H, Wang L, Jiang Y, Clay R, Wei X, et al. PPARalpha inhibition overcomes tumor-derived exosomal lipid-induced dendritic cell dysfunction. Cell Rep. 2020;33: 108278.
Article CAS PubMed PubMed Central Google Scholar
Portnoy V, Lin SH, Li KH, Burlingame A, Hu ZH, Li H, Li LC. saRNA-guided Ago2 targets the RITA complex to promoters to stimulate transcription. Cell Res. 2016;26:320–35.
Article CAS PubMed PubMed Central Google Scholar
Xie Y, Zheng M, Chu X, Chen Y, Xu H, Wang J, Zhou H, Long J. Paf1 and Ctr9 subcomplex formation is essential for Paf1 complex assembly and functional regulation. Nat Commun. 2018;9:3795.
Article PubMed PubMed Central Google Scholar
Tian J, Wang J, Roeder K. ESCO: single cell expression simulation incorporating gene co-expression. Bioinformatics. 2021;37:2374–81.
Article CAS PubMed PubMed Central Google Scholar
Browaeys R, Saelens W, Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods. 2020;17:159–62.
Article CAS PubMed Google Scholar
Zhang J, Liu L, Xu T, Zhang W, Zhao C, Li S, Li J, Rao N, Le TD. Exploring cell-specific miRNA regulation with single-cell miRNA-mRNA co-sequencing data. BMC Bioinformatics. 2021;22:578.
Article CAS PubMed PubMed Central Google Scholar
Jeppesen DK, Fenix AM, Franklin JL, Higginbotham JN, Zhang Q, Zimmerman LJ, Liebler DC, Ping J, Liu Q, Evans R, et al. Reassessment of exosome composition. Cell. 2019;177:428–445.e418.
Article CAS PubMed PubMed Central Google Scholar
Wang L, Sebra RP, Sfakianos JP, Allette K, Wang W, Yoo S, Bhardwaj N, Schadt EE, Yao X, Galsky MD, Zhu J. A reference profile-free deconvolution method to infer cancer cell-intrinsic subtypes and tumor-type-specific stromal profiles. Genome Med. 2020;12:24.
Article CAS PubMed PubMed Central Google Scholar
Zhang K, Erkan EP, Jamalzadeh S, Dai J, Andersson N, Kaipio K, Lamminen T, Mansuri N, Huhtinen K, Carpen O, et al. Longitudinal single-cell RNA-seq analysis reveals stress-promoted chemoresistance in metastatic ovarian cancer. Sci Adv. 2022;8:eabm1831.
Article CAS PubMed PubMed Central Google Scholar
Yu G, Jia Z, Dou Z. miR-24-3p regulates bladder cancer cell proliferation, migration, invasion and autophagy by targeting DEDD. Oncol Rep. 2017;37:1123–31.
Article CAS PubMed Google Scholar
Liao B, Chen S, Li Y, Yang Z, Yang Y, Deng X, Ke S. LncRNA BLACAT1 promotes proliferation, migration and invasion of prostate cancer cells via regulating miR-29a-3p/DVL3 axis. Technol Cancer Res Treat. 2021;20: 1533033820972342.
Article CAS PubMed PubMed Central Google Scholar
Xu Z, Chen Y, Ma L, Chen Y, Liu J, Guo Y, Yu T, Zhang L, Zhu L, Shu Y. Role of exosomal non-coding RNAs from tumor cells and tumor-associated macrophages in the tumor microenvironment. Mol Ther. 2022;30:3133–54.
Article CAS PubMed PubMed Central Google Scholar
Ehrlich L, Hall C, Venter J, Dostal D, Bernuzzi F, Invernizzi P, Meng F, Trzeciakowski JP, Zhou T, Standeford H, et al. miR-24 inhibition increases menin expression and decreases cholangiocarcinoma proliferation. Am J Pathol. 2017;187:570–80.
Article CAS PubMed PubMed Central Google Scholar
Han C, Zhang C, Wang H, Zhao L. Exosome-mediated communication between tumor cells and tumor-associated macrophages: implications for tumor microenvironment. Oncoimmunology. 2021;10:1887552.
Article PubMed PubMed Central Google Scholar
Gu Y, Li C, Xiao L, Li J, Pei H, Xu D, Jiang Y, Zhang X, Zhang L, Li K, et al. High expression of long non-coding RNA NNT-AS1 facilitates progression of cholangiocarcinoma through promoting epithelial-mesenchymal transition. Am J Transl Res. 2019;11:5438–56.
CAS PubMed PubMed Central Google Scholar
Lin H, Xu X, Chen K, Fu Z, Wang S, Chen Y, Zhang H, Niu Y, Chen H, Yu H, et al. LncRNA CASC15, MiR-23b cluster and SMAD3 form a novel positive feedback loop to promote epithelial-mesenchymal transition and metastasis in ovarian cancer. Int J Biol Sci. 2022;18:1989–2002.
Article CAS PubMed PubMed Central Google Scholar
Darmanis S, Sloan SA, Croote D, Mignardi M, Chernikova S, Samghababi P, Zhang Y, Neff N, Kowarsky M, Caneda C, et al. Single-Cell RNA-Seq analysis of infiltrating neoplastic cells at the migrating front of human glioblastoma. Cell Rep. 2017;21:1399–410.
Article CAS PubMed PubMed Central Google Scholar
Nieland L, Morsett LM, Broekman MLD, Breakefield XO, Abels ER. Extracellular vesicle-mediated bilateral communication between glioblastoma and astrocytes. Trends Neurosci. 2021;44:215–26.
Article CAS PubMed Google Scholar
Suman S, Mishra A, Kulshrestha A. A systems approach for the elucidation of crucial genes and network constituents of cervical intraepithelial neoplasia 1 (CIN1). Mol Biosyst. 2017;13:549–55.
Article CAS PubMed Google Scholar
Bader JE, Voss K, Rathmell JC. Targeting metabolism to improve the tumor microenvironment for cancer immunotherapy. Mol Cell. 2020;78:1019–33.
Article CAS PubMed PubMed Central Google Scholar
Liu H, Lei C, He Q, Pan Z, Xiao D, Tao Y. Nuclear functions of mammalian MicroRNAs in gene regulation, immunity and cancer. Mol Cancer. 2018;17:64.
Article PubMed PubMed Central Google Scholar
Kuppe C, Ibrahim MM, Kranz J, Zhang X, Ziegler S, Perales-Paton J, Jansen J, Reimer KC, Smith JR, Dobie R, et al. Decoding myofibroblast origins in human kidney fibrosis. Nature. 2021;589:281–6.
Article CAS PubMed Google Scholar
Bijkerk R, de Bruin RG, van Solingen C, van Gils JM, Duijs JM, van der Veer EP, Rabelink TJ, Humphreys BD, van Zonneveld AJ. Silencing of microRNA-132 reduces renal fibrosis by selectively inhibiting myofibroblast proliferation. Kidney Int. 2016;89:1268–80.
Article CAS PubMed Google Scholar
Wu L, Wang Q, Guo F, Ma X, Wang J, Zhao Y, Yan Y, Qin G. Involvement of miR-27a-3p in diabetic nephropathy via affecting renal fibrosis, mitochondrial dysfunction, and endoplasmic reticulum stress. J Cell Physiol. 2021;236:1454–68.
Article CAS PubMed Google Scholar
Duan ZY, Bu R, Liang S, Chen XZ, Zhang C, Zhang QY, Li JJ, Chen XM, Cai GY. Urinary miR-185-5p is a biomarker of renal tubulointerstitial fibrosis in IgA nephropathy. Front Immunol. 2024;15:1326026.
Article CAS PubMed PubMed Central Google Scholar
Yuan Q, Tan RJ, Liu Y. Myofibroblast in kidney fibrosis: origin, activation, and regulation. Adv Exp Med Biol. 2019;1165:253–83.
Article CAS PubMed Google Scholar
Shankland SJ, Smeets B, Pippin JW, Moeller MJ. The emergence of the glomerular parietal epithelial cell. Nat Rev Nephrol. 2014;10:158–73.
Article CAS PubMed Google Scholar
Qiu X, Mao Q, Tang Y, Wang L, Chawla R, Pliner HA, Trapnell C. Reversed graph embedding resolves complex single-cell trajectories. Nat Methods. 2017;14:979–82.
Article CAS PubMed PubMed Central Google Scholar
Yang X, Lu D, Wang R, Lian Z, Lin Z, Zhuo J, Chen H, Yang M, Tan W, Yang M, et al. Single-cell profiling reveals distinct immune phenotypes that contribute to ischaemia-reperfusion injury after steatotic liver transplantation. Cell Prolif. 2021;54: e13116.
Article CAS PubMed PubMed Central Google Scholar
Topper JN, Cai J, Qiu Y, Anderson KR, Xu YY, Deeds JD, Feeley R, Gimeno CJ, Woolf EA, Tayber O, et al. Vascular MADs: two novel MAD-related genes selectively inducible by flow in human vascular endothelium. Proc Natl Acad Sci U S A. 1997;94:9314–9.
Article CAS PubMed PubMed Central Google Scholar
Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z, Feng T, Zhou L, Tang W, Zhan L, et al. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation (Camb). 2021;2:100141.
CAS PubMed Google Scholar
Liberzon A, Birger C, Thorvaldsdottir H, Ghandi M, Mesirov JP, Tamayo P. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 2015;1:417–25.
Article CAS PubMed PubMed Central Google Scholar
Wang H, Mehal W, Nagy LE, Rotman Y. Immunological mechanisms and therapeutic targets of fatty liver diseases. Cell Mol Immunol. 2021;18:73–91.
Article CAS PubMed Google Scholar
McKellar DW, Mantri M, Hinchman MM, Parker JSL, Sethupathy P, Cosgrove BD, De Vlaminck I. Spatial mapping of the total transcriptome by in situ polyadenylation. Nat Biotechnol. 2023;41:513–20.
Article CAS PubMed Google Scholar
Li Z, Abdalla BA, Zheng M, He X, Cai B, Han P, Ouyang H, Chen B, Nie Q, Zhang X. Systematic transcriptome-wide analysis of mRNA-miRNA interactions reveals the involvement of miR-142-5p and its target (FOXO3) in skeletal muscle growth in chickens. Mol Genet Genomics. 2018;293:69–80.
Article CAS PubMed Google Scholar
Zhang Y, Yao Y, Wang Z, Lu D, Zhang Y, Adetula AA, Liu S, Zhu M, Yang Y, Fan X, et al. MiR-743a-5p regulates differentiation of myoblast by targeting Mob1b in skeletal muscle development and regeneration. Genes Dis. 2022;9:1038–48.
Article CAS PubMed Google Scholar
Li Z, Cai B, Abdalla BA, Zhu X, Zheng M, Han P, Nie Q, Zhang X. LncIRS1 controls muscle atrophy via sponging miR-15 family to activate IGF1-PI3K/AKT pathway. J Cachexia Sarcopenia Muscle. 2019;10:391–410.
Article PubMed PubMed Central Google Scholar
Faraldi M, Sansoni V, Vitale J, Perego S, Gomarasca M, Verdelli C, Messina C, Sconfienza LM, Banfi G, Corbetta S, Lombardi G. Plasma microRNA signature associated with skeletal muscle wasting in post-menopausal osteoporotic women. J Cachexia Sarcopenia Muscle. 2024;15:690–701.
Article PubMed PubMed Central Google Scholar
Muroya S, Ogasawara H, Nohara K, Oe M, Ojima K, Hojito M. Coordinated alteration of mRNA-microRNA transcriptomes associated with exosomes and fatty acid metabolism in adipose tissue and skeletal muscle in grazing cattle. Asian-Australas J Anim Sci. 2020;33:1824–36.
Article CAS PubMed Google Scholar
Margolis LM, Hatch-McChesney A, Allen JT, DiBella MN, Carrigan CT, Murphy NE, Karl JP, Gwin JA, Hennigar SR, McClung JP, Pasiakos SM. Circulating and skeletal muscle microRNA profiles are more sensitive to sustained aerobic exercise than energy balance in males. J Physiol. 2022;600:3951–63.
Article CAS PubMed Google Scholar
Hu G, Do DN, Davoudi P, Miar Y. Emerging roles of non-coding RNAs in the feed efficiency of livestock species. Genes (Basel). 2022;13:13.
Article Google Scholar
Jin S, Guerrero-Juarez CF, Zhang L, Chang I, Ramos R, Kuan CH, Myung P, Plikus MV, Nie Q. Inference and analysis of cell-cell communication using cell chat. Nat Commun. 2021;12:1088.
Article CAS PubMed PubMed Central Google Scholar
Garcia-Alonso L, Handfield LF, Roberts K, Nikolakopoulou K, Fernando RC, Gardner L, Woodhams B, Arutyunyan A, Polanski K, Hoo R, et al. Mapping the temporal and spatial dynamics of the human endometrium in vivo and in vitro. Nat Genet. 2021;53:1698–711.
Article CAS PubMed PubMed Central Google Scholar
Shao X, Wang Z, Wang K, Lu X, Zhang P, Guo R, Liao J, Yang P, Xu X, Fan X: A Single-Cell Landscape of Human Liver Transplantation Reveals a Pathogenic Immune Niche Associated with Early Allograft Dysfunction. Engineering 2024.
Hu Y, Peng T, Gao L, Tan K. CytoTalk: de novo construction of signal transduction networks using single-cell transcriptomic data. Sci Adv. 2021;7:eabf1356.
Article CAS PubMed PubMed Central Google Scholar
Zhang Y, Liu T, Hu X, Wang M, Wang J, Zou B, Tan P, Cui T, Dou Y, Ning L, et al. Cell Call: integrating paired ligand-receptor and transcription factor activities for cell-cell communication. Nucleic Acids Res. 2021;49:8520–34.
Article CAS PubMed PubMed Central Google Scholar
He C, Zhou P, Nie Q. exFINDER: identify external communication signals using single-cell transcriptomics data. Nucleic Acids Res. 2023;51: e58.
Article CAS PubMed PubMed Central Google Scholar
Shao X, Li C, Yang H, Lu X, Liao J, Qian J, Wang K, Cheng J, Yang P, Chen H, et al. Knowledge-graph-based cell-cell communication inference for spatially resolved transcriptomic data with SpaTalk. Nat Commun. 2022;13:4429.
Article CAS PubMed PubMed Central Google Scholar
Dries R, Zhu Q, Dong R, Eng CL, Li H, Liu K, Fu Y, Zhao T, Sarkar A, Bao F, et al. Giotto: a toolbox for integrative analysis and visualization of spatial expression data. Genome Biol. 2021;22:78.
Article CAS PubMed PubMed Central Google Scholar
Cang Z, Nie Q. Inferring spatial and signaling relationships between cells from single cell transcriptomic data. Nat Commun. 2020;11:2084.
Article CAS PubMed PubMed Central Google Scholar
Tang Z, Zhang T, Yang B, Su J, Song Q. spaCI: deciphering spatial cellular communications through adaptive graph model. Brief Bioinform. 2023;24:bbac563.
Article PubMed Google Scholar
Armingol E, Ghaddar A, Joshi CJ, Baghdassarian H, Shamie I, Chan J, Her HL, Berhanu S, Dar A, Rodriguez-Armstrong F, et al. Inferring a spatial code of cell-cell interactions across a whole animal body. PLoS Comput Biol. 2022;18: e1010715.
Article CAS PubMed PubMed Central Google Scholar
Cang Z, Zhao Y, Almet AA, Stabell A, Ramos R, Plikus MV, Atwood SX, Nie Q. Screening cell-cell communication in spatial transcriptomics via collective optimal transport. Nat Methods. 2023;20:218–28.
Article CAS PubMed PubMed Central Google Scholar
Tsuchiya T, Hori H, Ozaki H. CCPLS reveals cell-type-specific spatial dependence of transcriptomes in single cells. Bioinformatics. 2022;38:4868–77.
Article CAS PubMed PubMed Central Google Scholar
Ru B, Huang J, Zhang Y, Aldape K, Jiang P. Estimation of cell lineages in tumors from spatial transcriptomics data. Nat Commun. 2023;14:568.
Article CAS PubMed PubMed Central Google Scholar
Zheng GX, Terry JM, Belgrader P, Ryvkin P, Bent ZW, Wilson R, Ziraldo SB, Wheeler TD, McDermott GP, Zhu J, et al. Massively parallel digital transcriptional profiling of single cells. Nat Commun. 2017;8:14049.
Article CAS PubMed PubMed Central Google Scholar
Nitzsche A, Paszkowski-Rogacz M, Matarese F, Janssen-Megens EM, Hubner NC, Schulz H, de Vries I, Ding L, Huebner N, Mann M, et al. RAD21 cooperates with pluripotency transcription factors in the maintenance of embryonic stem cell identity. PLoS ONE. 2011;6: e19470.
Article CAS PubMed PubMed Central Google Scholar
Hinger SA, Cha DJ, Franklin JL, Higginbotham JN, Dou Y, Ping J, Shu L, Prasad N, Levy S, Zhang B, et al. Diverse long RNAs Are Differentially Sorted into Extracellular Vesicles Secreted by Colorectal Cancer Cells. Cell Rep. 2018;25(715–725): e714.
Google Scholar
Zheng R, Zhang Y, Tsuji T, Zhang L, Tseng Y-H, Chen K: MEBOCOST: metabolic cell-cell communication modeling by single cell transcriptome. bioRxiv. 2022. https://doiorg.publicaciones.saludcastillayleon.es/10.1101/2022.05.30.494067.
Zhao W, Johnston KG, Ren H, Xu X, Nie Q. Inferring neuron-neuron communications from single-cell transcriptomics through NeuronChat. Nat Commun. 2023;14:1128.
Article CAS PubMed PubMed Central Google Scholar
Shao X, Liao J, Li C, Lu X, Cheng J, Fan X. CellTalkDB: a manually curated database of ligand-receptor interactions in humans and mice. Brief Bioinform. 2021;22:bbaa269.
Article PubMed Google Scholar
Xu Z, Zhang T, Chen H, Zhu Y, Lv Y, Zhang S, Chen J, Chen H, Yang L, Jiang W, et al. High-throughput single nucleus total RNA sequencing of formalin-fixed paraffin-embedded tissues by snRandom-seq. Nat Commun. 2023;14:2734.
Article CAS PubMed PubMed Central Google Scholar
Hao Y, Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, Srivastava A, Molla G, Madad S, Fernandez-Granda C, Satija R. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol. 2024;42:293–304.
Article CAS PubMed Google Scholar
Fehlmann T, Backes C, Pirritano M, Laufer T, Galata V, Kern F, Kahraman M, Gasparoni G, Ludwig N, Lenhof HP, et al. The sncRNA Zoo: a repository for circulating small noncoding RNAs in animals. Nucleic Acids Res. 2019;47:4431–41.
Article CAS PubMed PubMed Central Google Scholar
Saito T, Rehmsmeier M. Precrec: fast and accurate precision-recall and ROC curve calculations in R. Bioinformatics. 2017;33:145–7.
Article CAS PubMed Google Scholar
Hanzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics. 2013;14: 7.
Article PubMed PubMed Central Google Scholar
Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J, Simonovic M, Doncheva NT, Morris JH, Bork P, et al. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res. 2019;47:D607–13.
Article CAS PubMed Google Scholar
Yamamoto T, Hattori Y, Yamamoto Y, Ochiya T: RNA-seq of human multiple myeloma cell lines and cell lines with long-term exposure to lenalidomide. Datasets Gene Expression Omnibus 2021, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE165557.
Morimoto Y: MUC1-C is a master regulator of MICA/B NKG2D ligand and exosome secretion in human cancer cells. Datasets gene expression omnibus. 2022. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE215308.
Umezu T, Tsuneyama K, Kanekura K, Hayakawa M, Tanahashi T, Kawano M, Taguchi Y, Toyoda H, Tamori A, Kuroda M: Integrated gene expression analysis of blood and liver tissue in precancerous condition [miRNA]. Datasets gene expression omnibus. 2020. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE153581.
Hill N, Michell D, Pusey C, Vickers K, Woollard K: RNA-seq of podocytes treated with glomerular endothelial cell vesicles. Datasets Gene Expression Omnibus 2020, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE129883.
Silva S, Graña O, Peinado H: RNA profiling of human lymphatic endothelial cells exposed to melanoma exosomes. Datasets Gene Expression Omnibus. 2021 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE135187.
Yin X, Zeng W: Next generation sequencing facilitates quantitative analysis of transcriptomes of BMDCs treated without or with exosomes from tumor cells or normal cells. Datasets gene expression omnibus. 2021. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE155881.
Jeppesen D, Coffey R: Reassessment of exosome composition. Datasets gene expression omnibus. 2019. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE125905.
Stik G, Durand C: Stromal cells secrete exosomes carrying specific molecular signature involved in the support of Hematopoietic Stem and Progenitor Cells. Datasets Gene Expression Omnibus. 2016. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE76711.
Santiago J, Natu A, Ramelow C, Rayaprolu S, Xiao H, Kumar V, Seyfried N, Rangaraju S: Identification of state-specific proteomic and transcriptomic signatures of microglia-derived extracellular vesicles [mirna-seq]. Datasets gene expression omnibus. 2023. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE240294.
Hinger S, Cha D, Franklin J, Higginbotham J, Dou Y, Ping J, Shu L, Prasad N, Zhang B, Liu Q, et al: Diverse long RNAs are differentially sorted into extracellular vesicles secreted by colorectal cancer cells. Datasets gene expression omnibus. 2018. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE121964.
Kaneko S: Expressions of MicroRNA in Mouse Kidney with Unilateral Ureteral Obstruction. Datasets Gene Expression Omnibus 2021, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162794.
Lin H, Lai I: Rat liver tissue: IR vs. iPoC. Datasets gene expression omnibus. 2019. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE56071.
McKellar D, Mantri M, De Vlaminck I, Cosgrove B: Small RNA sequencing of regenerating mouse hindlimb muscle and Type 1-Lang reovirus-infected mouse heart. Datasets gene expression omnibus. 2022. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE200480.
Colaprico A, Silva TC, Olsen C, Garofano L, Cava C, Garolini D, Sabedot TS, Malta TM, Pagnotta SM, Castiglioni I, et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. Nucleic Acids Res. 2016;44: e71.
Article PubMed Google Scholar
Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, Sun Y, Jacobsen A, Sinha R, Larsson E, et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal. 2013;6:pl1.
Article PubMed PubMed Central Google Scholar
Zhang M, Yang H, Liu B, Yan X: Dissecting the transcriptomic landscape of the human intrahepatic cholangiocarcinoma by single cell RNA-sequencing. Datasets gene expression omnibus. 2020. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE138709.
Han Y, Wang Y, Dong X, Sun D, Liu Z, Yue J, Wang H, Li T, Wang C. TISCH2: expanded datasets and new tools for single-cell transcriptome analyses of the tumor microenvironment. Nucleic Acids Res. 2023;51:D1425–31.
Article PubMed Google Scholar
Sebra R, Wang L: Single cell RNA-seq profiling of two bladder cancer specimen. Datasets gene expression omnibus. 2019. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE130001.
Darmanis S, Quake S: Single-Cell RNAseq analysis of diffuse neoplastic infiltrating cells at the migrating front of human glioblastoma. Datasets Gene Expression Omnibus 2017, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE84465.
Kuppe C, Ibrahim M, Kranz J, Zhang X, Ziegler S, Jansen J, Reimer K, Perales-Paton J, Smith J, Dobie R, et al: Decoding myofibroblast origins in human kidney fibrosis. Zenodo. 2020. https://doiorg.publicaciones.saludcastillayleon.es/10.5281/zenodo.4059315.
Yang X: The scRNA-Seq atlas of fatty liver graft injury after liver transplantation. Datasets genome sequence archive. 2021. https://ngdc.cncb.ac.cn/gsa/browse/CRA004061.
Shao X, Yu L, Li C, Qian J, Yang X, Yang H, Liao J, Fan X, Xu X, Fan X: Extracellular vesicle-derived miRNA-mediated cell-cell communication inference for single-cell transcriptomic data with miRTalk. Github. 2025. https://github.com/multitalk/miRTalk.
Shao X, Yu L, Li C, Qian J, Yang X, Yang H, Liao J, Fan X, Xu X, Fan X: Extracellular vesicle-derived miRNA-mediated cell-cell communication inference for single-cell transcriptomic data with miRTalk. Zenodo. 2025. https://doiorg.publicaciones.saludcastillayleon.es/10.5281/zenodo.13856217.

Download references

Acknowledgements

The authors thank the High Performance Computing Cluster of Zhejiang University Innovation Center of Yangtze River Delta for their technical support.

Review history

The review history is available as Additional File 9.

Peer review information

Tim Sands was the primary editor of this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.

Funding

This work was supported by the"Pioneer"and"Leading Goose"R&D Program of Zhejiang (2024 C03106, Xiaohui F.), the National Natural Science Foundation of China (82200725, X.S.; 81930016, X.X.), the Key Research and Development Program of China (2021YFA1100500, X.X.), and the Fundamental Research Funds for the Central Universities (226–2024 - 00001, Xiaohui F.).

National Natural Science Foundation of China,82200725,Xin Shao,81930016,Xiao Xu,Innovation Team and Talents Cultivation Program of National Administration of Traditional Chinese Medicine,ZYYCXTD-D- 202002,Xiaohui Fan,Fundamental Research Funds for the Central Universities,226 - 2023 - 00059,Xiaohui Fan,226 - 2023 - 00114,Xiaohui Fan,Key Research and Development Program of China,2021YFA1100500,Xiao Xu,Major Research Plan of the National Natural Science Foundation of China,92159202,Xiao Xu

Author information

Xin Shao and Lingqi Yu should be regarded as joint first authors.

Authors and Affiliations

Zhejiang Key Laboratory of Precision Diagnosis and Therapy for Major Gynecological Diseases, Women’S Hospital, Zhejiang University School of Medicine, Hangzhou, 310006, China
Xin Shao, Lingqi Yu & Xiaohui Fan
State Key Laboratory of Chinese Medicine Modernization, Innovation Center of Yangtze River Delta, Zhejiang University, Jiaxing, 314102, China
Xin Shao, Lingqi Yu, Chengyu Li, Jingyang Qian, Jie Liao, Xueru Fan & Xiaohui Fan
Pharmaceutical Informatics Institute, College of Pharmaceutical Sciences, Zhejiang University, Hangzhou, 310058, China
Xin Shao, Lingqi Yu, Chengyu Li, Jingyang Qian, Jie Liao, Xueru Fan & Xiaohui Fan
The Joint-Laboratory of Clinical Multi-Omics Research Between, Zhejiang University and Ningbo Municipal Hospital of TCM, Ningbo Municipal Hospital of TCM, Ningbo, 315012, China
Xin Shao & Xiaohui Fan
The Center for Integrated Oncology and Precision Medicine, School of Medicine, Affiliated Hangzhou First People’S Hospital, Westlake University, Hangzhou, 310006, China
Xinyu Yang
Zhejiang University School of Medicine, Hangzhou, 310006, China
Lingqi Yu & Xinyu Yang
College of Computer Science and Technology, Zhejiang University, Hangzhou, 310027, China
Haihong Yang
Department of Hepatobiliary & Pancreatic Surgery and Minimally Invasive Surgery, Zhejiang Provincial People’S Hospital (Affiliated People’s Hospital), Hangzhou Medical College, Hangzhou, 310024, China
Xiao Xu
NHC Key Laboratory of Combined Multi-Organ Transplantation, Hangzhou, 310003, China
Xiao Xu

Authors

Xin Shao
View author publications
You can also search for this author inPubMed Google Scholar
Lingqi Yu
View author publications
You can also search for this author inPubMed Google Scholar
Chengyu Li
View author publications
You can also search for this author inPubMed Google Scholar
Jingyang Qian
View author publications
You can also search for this author inPubMed Google Scholar
Xinyu Yang
View author publications
You can also search for this author inPubMed Google Scholar
Haihong Yang
View author publications
You can also search for this author inPubMed Google Scholar
Jie Liao
View author publications
You can also search for this author inPubMed Google Scholar
Xueru Fan
View author publications
You can also search for this author inPubMed Google Scholar
Xiao Xu
View author publications
You can also search for this author inPubMed Google Scholar
Xiaohui Fan
View author publications
You can also search for this author inPubMed Google Scholar

Contributions

Xiaohui F., X.X., and X.S. conceived and designed the study. X.S., C.L., and H.Y. developed the miRTalk algorithm. C.L., J.Q., X.Y., and J.L. collected and processed the real-world data. X.S. performed the benchmark test and analyzed the application cases with miRTalk and drew the main figures. L.Y. and Xueru F. performed the in vitro experiments. X.S. wrote the manuscript. X.S., L.Y., X.X., and Xiaohui F. revised the manuscript. All the authors read and approved the final manuscript.

Corresponding authors

Correspondence to Xin Shao, Xiao Xu or Xiaohui Fan.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

13059_2025_3566_MOESM1_ESM.docx

Additional file 1: Supplementary Figures. This file provides additional details and results from our study, including Figs. S1-S14.

13059_2025_3566_MOESM2_ESM.xlsx

Additional file 2: Supplementary Tables. This file provides additional details about the scRNA-seq and spatial transcriptomics datasets used in cases.

13059_2025_3566_MOESM3_ESM.xlsx

Additional file 3: Supplementary Tables. This file provides detailed MiTIs inferred by miRTalk about the human BLCA, CHOL, and OV scRNA-seq datasets.

13059_2025_3566_MOESM4_ESM.xlsx

Additional file 4: Supplementary Tables. This file provides detailed MiTIs inferred by miRTalk and other supporting information about the case of human GBM.

13059_2025_3566_MOESM5_ESM.xlsx

Additional file 5: Supplementary Tables. This file provides detailed MiTIs inferred by miRTalk and other supporting information about the case of mouse kidney fibrosis.

13059_2025_3566_MOESM6_ESM.xlsx

Additional file 6: Supplementary Tables. This file provides detailed MiTIs inferred by miRTalk and other supporting information about the case of rat normal and fatty LT.

13059_2025_3566_MOESM7_ESM.xlsx

Additional file 7: Supplementary Tables. This file provides detailed MiTIs inferred by miRTalk and other supporting information about the case of mouse skeletal muscle injury.

Additional file 8: Supplementary Tables. This file provides detailed information about the EXOmotif and primer.

Additional file 9: Review History. This file contains peer review reports.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Shao, X., Yu, L., Li, C. et al. Extracellular vesicle-derived miRNA-mediated cell–cell communication inference for single-cell transcriptomic data with miRTalk. Genome Biol 26, 95 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13059-025-03566-x

Download citation

Received: 26 June 2023
Accepted: 02 April 2025
Published: 14 April 2025
DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13059-025-03566-x

Extracellular vesicle-derived miRNA-mediated cell–cell communication inference for single-cell transcriptomic data with miRTalk

Abstract

Background

Results

Overview of the miRTalk method

Performance evaluation on simulated benchmarking datasets

Prediction of miRTalk consistent with the literature evidence on real datasets

Oncogenic modulation of glioblastoma cells on the TME sensed by nonmalignant cells

Characterization of signal transmissions forming the fibrogenic niche in the kidney

Identification of the granulocyte-hepatocyte communication discriminating between normal and fatty liver transplantation

Inference of spatially resolved pro-generative communication in the mouse skeletal muscle injury

Discussion

Conclusions

Methods

Data processing

Construction of miRTalkDB

Algorithm of miRTalk

Identification of MiTIs

Step 1: preprocessing miRNA and target genes

Step 2: scoring MiTIs between pairwise cell types

Specificity calculation of identified MiTIs

Simulation of scRNA-seq data for benchmarking

Performance evaluation with simulated datasets for miRTalk

Comparison with other methods

Bulk data analysis

Benchmarking with false positive rates

Pathway and biological process enrichment

Single-cell trajectory analysis

Cell culture and oxygen–glucose deprivation/reperfusion (OGD/R) model

Cell viability assay

RNA extraction and RT‐qPCR

Statistical analysis

Data availability

References

Acknowledgements

Review history

Peer review information

Funding

Author information

Authors and Affiliations

Contributions

Corresponding authors

Ethics declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Additional information

Publisher's Note

Supplementary Information

Rights and permissions

About this article

Cite this article

Share this article

Keywords

Genome Biology

Contact us