Computational challenges and opportunities in spatially resolved transcriptomic data analysis

q-Diffusion leverages the full dimensionality of gene coexpression in single-cell transcriptomics

Unlocking the full dimensionality of single-cell RNA sequencing data (scRNAseq) is the next frontier to a richer, fuller understanding of cell biology. We introduce q-diffusion, a framework for capturing the coexpression structure of an entire library of genes, improving on state-of-the-art analysis tools. The method is demonstrated via three case studies. In the first, q-diffusion helps gain statistical significance for differential effects on patient outcomes when analyzing the CALGB/SWOG 80405 randomized phase III clinical trial, suggesting precision guidance for the treatment of metastatic colorectal cancer. Secondly, q-diffusion is benchmarked against existing scRNAseq classification methods using an in vitro PBMC dataset, in which the proposed method discriminates IFN-γ stimulation more accurately. The same case study demonstrates improvements in unsupervised cell clustering with the recent Tabula Sapiens human atlas. Finally, a local distributional segmentation approach for spatial scRNAseq, driven by q-diffusion, yields interpretable structures of human cortical tissue.

The covariance environment defines cellular niches for spatial inference

A key challenge of analyzing data from high-resolution spatial profiling technologies is to suitably represent the features of cellular neighborhoods or niches. Here we introduce the covariance environment (COVET), a representation that leverages the gene-gene covariate structure across cells in the niche to capture the multivariate nature of cellular interactions within it. We define a principled optimal transport-based distance metric between COVET niches that scales to millions of cells. Using COVET to encode spatial context, we developed environmental variational inference (ENVI), a conditional variational autoencoder that jointly embeds spatial and single-cell RNA sequencing data into a latent space. ENVI includes two decoders: one to impute gene expression across the spatial modality and a second to project spatial information onto single-cell data. ENVI can confer spatial context to genomics data from single dissociated cells and outperforms alternatives for imputing gene expression on diverse spatial datasets.

Recent advances in differential expression analysis for single-cell RNA-seq and spatially resolved transcriptomic studies

Differential expression (DE) analysis is a necessary step in the analysis of single-cell RNA sequencing (scRNA-seq) and spatially resolved transcriptomics (SRT) data. Unlike traditional bulk RNA-seq, DE analysis for scRNA-seq or SRT data has unique characteristics that may contribute to the difficulty of detecting DE genes. However, the plethora of DE tools that work with various assumptions makes it difficult to choose an appropriate one. Furthermore, a comprehensive review on detecting DE genes for scRNA-seq data or SRT data from multi-condition, multi-sample experimental designs is lacking. To bridge such a gap, here, we first focus on the challenges of DE detection, then highlight potential opportunities that facilitate further progress in scRNA-seq or SRT analysis, and finally provide insights and guidance in selecting appropriate DE tools or developing new computational DE methods.

Software Tools for 2D Cell Segmentation

Cell segmentation is an important task in the field of image processing, widely used in the life sciences and medical fields. Traditional methods are mainly based on pixel intensity and spatial relationships, but have limitations. In recent years, machine learning and deep learning methods have been widely used, providing more-accurate and efficient solutions for cell segmentation. The effort to develop efficient and accurate segmentation software tools has been one of the major focal points in the field of cell segmentation for years. However, each software tool has unique characteristics and adaptations, and no universal cell-segmentation software can achieve perfect results. In this review, we used three publicly available datasets containing multiple 2D cell-imaging modalities. Common segmentation metrics were used to evaluate the performance of eight segmentation tools to compare their generality and, thus, find the best-performing tool.

ATAT: Automated Tissue Alignment and Traversal in Spatial Transcriptomics with Self-Supervised Learning

Spatial transcriptomics (ST) has enhanced RNA analysis in tissue biopsies, but interpreting these data is challenging without expert input. We present Automated Tissue Alignment and Traversal (ATAT), a novel computational framework designed to enhance ST analysis in the context of multiple and complex tissue architectures and morphologies, such as those found in biopsies of the gastrointestinal tract. ATAT utilizes self-supervised contrastive learning on hematoxylin and eosin (H&E) stained images to automate the alignment and traversal of ST data. This approach addresses a critical gap in current ST analysis methodologies, which rely heavily on manual annotation and pathologist expertise to delineate regions of interest for accurate gene expression modeling. Our framework not only streamlines the alignment of multiple ST samples, but also demonstrates robustness in modeling gene expression transitions across specific regions. Additionally, we highlight the ability of ATAT to traverse complex tissue topologies in real-world cases from various individuals and conditions. Our method successfully elucidates differences in immune infiltration patterns across the intestinal wall, enabling the modeling of transcriptional changes across histological layers. We show that ATAT achieves comparable performance to the state-of-the-art method, while alleviating the burden of manual annotation and enabling alignment of tissue samples with complex morphologies.

A machine learning one-class logistic regression model to predict stemness for single cell transcriptomics and spatial omics

Cell annotation is a crucial methodological component to interpreting single cell and spatial omics data. These approaches were developed for single cell analysis but are often biased, manually curated and yet unproven in spatial omics. Here we apply a stemness model for assessing oncogenic states to single cell and spatial omic cancer datasets. This one-class logistic regression machine learning algorithm is used to extract transcriptomic features from non-transformed stem cells to identify dedifferentiated cell states in tumors. We found this method identifies single cell states in metastatic tumor cell populations without the requirement of cell annotation. This machine learning model identified stem-like cell populations not identified in single cell or spatial transcriptomic analysis using existing methods. For the first time, we demonstrate the application of a ML tool across five emerging spatial transcriptomic and proteomic technologies to identify oncogenic stem-like cell types in the tumor microenvironment.

Spatial-Live: A lightweight and versatile tool for single cell spatial-omics data visualization

Single cell spatial-omics data visualization plays a pivotal role in unraveling the intricate spatial organization and heterogeneity of cellular systems. Although various software tools and packages have been developed for this purpose, challenges persist in terms of user-friendly accessibility, data integration, and interactivity. In this study, we introduce Spatial-Live, a lightweight and versatile viewer tool designed for flexible single-cell spatial-omics data visualization. Spatial-Live overcomes the fundamental limitations of two-dimensional (2D) orthographic modes by employing a layer-stacking strategy, enabling efficient rendering of diverse data types with interactive features, and enhancing visualization with richer information in a unified three-dimensional (3D) space.

DeepCCI: a deep learning framework for identifying cell-cell interactions from single-cell RNA sequencing data

MOTIVATION

Cell-cell interactions (CCIs) play critical roles in many biological processes such as cellular differentiation, tissue homeostasis, and immune response. With the rapid development of high throughput single-cell RNA sequencing (scRNA-seq) technologies, it is of high importance to identify CCIs from the ever-increasing scRNA-seq data. However, limited by the algorithmic constraints, current computational methods based on statistical strategies ignore some key latent information contained in scRNA-seq data with high sparsity and heterogeneity.

RESULTS

Here, we developed a deep learning framework named DeepCCI to identify meaningful CCIs from scRNA-seq data. Applications of DeepCCI to a wide range of publicly available datasets from diverse technologies and platforms demonstrate its ability to predict significant CCIs accurately and effectively. Powered by the flexible and easy-to-use software, DeepCCI can provide the one-stop solution to discover meaningful intercellular interactions and build CCI networks from scRNA-seq data.

AVAILABILITY AND IMPLEMENTATION

The source code of DeepCCI is available online at https://github.com/JiangBioLab/DeepCCI.

SMASH: Scalable Method for Analyzing Spatial Heterogeneity of genes in spatial transcriptomics data

In high-throughput spatial transcriptomics (ST) studies, it is of great interest to identify the genes whose level of expression in a tissue covaries with the spatial location of cells/spots. Such genes, also known as spatially variable genes (SVGs), can be crucial to the biological understanding of both structural and functional characteristics of complex tissues. Existing methods for detecting SVGs either suffer from huge computational demand or significantly lack statistical power. We propose a non-parametric method termed SMASH that achieves a balance between the above two problems. We compare SMASH with other existing methods in varying simulation scenarios demonstrating its superior statistical power and robustness. We apply the method to four ST datasets from different platforms uncovering interesting biological insights.

Principles and challenges of modeling temporal and spatial omics data

Studies with temporal or spatial resolution are crucial to understand the molecular dynamics and spatial dependencies underlying a biological process or system. With advances in high-throughput omic technologies, time- and space-resolved molecular measurements at scale are increasingly accessible, providing new opportunities to study the role of timing or structure in a wide range of biological questions. At the same time, analyses of the data being generated in the context of spatiotemporal studies entail new challenges that need to be considered, including the need to account for temporal and spatial dependencies and compare them across different scales, biological samples or conditions. In this Review, we provide an overview of common principles and challenges in the analysis of temporal and spatial omics data. We discuss statistical concepts to model temporal and spatial dependencies and highlight opportunities for adapting existing analysis methods to data with temporal and spatial dimensions.

Alignment of spatial genomics data using deep Gaussian processes

Spatially resolved genomic technologies have allowed us to study the physical organization of cells and tissues, and promise an understanding of local interactions between cells. However, it remains difficult to precisely align spatial observations across slices, samples, scales, individuals and technologies. Here, we propose a probabilistic model that aligns spatially-resolved samples onto a known or unknown common coordinate system (CCS) with respect to phenotypic readouts (for example, gene expression). Our method, Gaussian Process Spatial Alignment (GPSA), consists of a two-layer Gaussian process: the first layer maps observed samples' spatial locations onto a CCS, and the second layer maps from the CCS to the observed readouts. Our approach enables complex downstream spatially aware analyses that are impossible or inaccurate with unaligned data, including an analysis of variance, creation of a dense three-dimensional (3D) atlas from sparse two-dimensional (2D) slices or association tests across data modalities.

Understanding tumour endothelial cell heterogeneity and function from single-cell omics

Anti-angiogenic therapies (AATs) are used to treat different types of cancers. However, their success is limited owing to insufficient efficacy and resistance. Recently, single-cell omics studies of tumour endothelial cells (TECs) have provided new mechanistic insight. Here, we overview the heterogeneity of human TECs of all tumour types studied to date, at the single-cell level. Notably, most human tumour types contain varying numbers but only a small population of angiogenic TECs, the presumed targets of AATs, possibly contributing to the limited efficacy of and resistance to AATs. In general, TECs are heterogeneous within and across all tumour types, but comparing TEC phenotypes across tumours is currently challenging, owing to the lack of a uniform nomenclature for endothelial cells and consistent single-cell analysis protocols, urgently raising the need for a more consistent approach. Nonetheless, across most tumour types, universal TEC markers (ACKR1, PLVAP and IGFBP3) can be identified. Besides angiogenesis, biological processes such as immunomodulation and extracellular matrix organization are among the most commonly predicted enriched signatures of TECs across different tumour types. Although angiogenesis and extracellular matrix targets have been considered for AAT (without the hoped success), the immunomodulatory properties of TECs have not been fully considered as a novel anticancer therapeutic approach. Therefore, we also discuss progress, limitations, solutions and novel targets for AAT development.

Transcriptomics-driven drug repositioning for the treatment of diabetic foot ulcer

Diabetic foot ulcers (DFUs) are a common complication of diabetes and can lead to severe disability and even amputation. Despite advances in treatment, there is currently no cure for DFUs and available drugs for treatment are limited. This study aimed to identify new candidate drugs and repurpose existing drugs to treat DFUs based on transcriptomics analysis. A total of 31 differentially expressed genes (DEGs) were identified and used to prioritize the biological risk genes for DFUs. Further investigation using the database DGIdb revealed 12 druggable target genes among 50 biological DFU risk genes, corresponding to 31 drugs. Interestingly, we highlighted that two drugs (urokinase and lidocaine) are under clinical investigation for DFU and 29 drugs are potential candidates to be repurposed for DFU therapy. The top 5 potential biomarkers for DFU from our findings are IL6ST, CXCL9, IL1R1, CXCR2, and IL10. This study highlights IL1R1 as a highly promising biomarker for DFU due to its high systemic score in functional annotations, that can be targeted with an existing drug, Anakinra. Our study proposed that the integration of transcriptomic and bioinformatic-based approaches has the potential to drive drug repurposing for DFUs. Further research will further examine the mechanisms by which targeting IL1R1 can be used to treat DFU.

Deciphering tissue heterogeneity from spatially resolved transcriptomics by the autoencoder-assisted graph convolutional neural network

Spatially resolved transcriptomics (SRT) provides an unprecedented opportunity to investigate the complex and heterogeneous tissue organization. However, it is challenging for a single model to learn an effective representation within and across spatial contexts. To solve the issue, we develop a novel ensemble model, AE-GCN (autoencoder-assisted graph convolutional neural network), which combines the autoencoder (AE) and graph convolutional neural network (GCN), to identify accurate and fine-grained spatial domains. AE-GCN transfers the AE-specific representations to the corresponding GCN-specific layers and unifies these two types of deep neural networks for spatial clustering via the clustering-aware contrastive mechanism. In this way, AE-GCN accommodates the strengths of both AE and GCN for learning an effective representation. We validate the effectiveness of AE-GCN on spatial domain identification and data denoising using multiple SRT datasets generated from ST, 10x Visium, and Slide-seqV2 platforms. Particularly, in cancer datasets, AE-GCN identifies disease-related spatial domains, which reveal more heterogeneity than histological annotations, and facilitates the discovery of novel differentially expressed genes of high prognostic relevance. These results demonstrate the capacity of AE-GCN to unveil complex spatial patterns from SRT data.

Unraveling non-genetic heterogeneity in cancer with dynamical models and computational tools

Individual cells within an otherwise genetically homogenous population constantly undergo fluctuations in their molecular state, giving rise to non-genetic heterogeneity. Such diversity is being increasingly implicated in cancer therapy resistance and metastasis. Identifying the origins of non-genetic heterogeneity is therefore crucial for making clinical breakthroughs. We discuss with examples how dynamical models and computational tools have provided critical multiscale insights into the nature and consequences of non-genetic heterogeneity in cancer. We demonstrate how mechanistic modeling has been pivotal in establishing key concepts underlying non-genetic diversity at various biological scales, from population dynamics to gene regulatory networks. We discuss advances in single-cell longitudinal profiling techniques to reveal patterns of non-genetic heterogeneity, highlighting the ongoing efforts and challenges in statistical frameworks to robustly interpret such multimodal datasets. Moving forward, we stress the need for data-driven statistical and mechanistically motivated dynamical frameworks to come together to develop predictive cancer models and inform therapeutic strategies.

signifinder enables the identification of tumor cell states and cancer expression signatures in bulk, single-cell and spatial transcriptomic data

Over the last decade, many studies and some clinical trials have proposed gene expression signatures as a valuable tool for understanding cancer mechanisms, defining subtypes, monitoring patient prognosis, and therapy efficacy. However, technical and biological concerns about reproducibility have been raised. Technical reproducibility is a major concern: we currently lack a computational implementation of the proposed signatures, which would provide detailed signature definition and assure reproducibility, dissemination, and usability of the classifier. Another concern regards intratumor heterogeneity, which has never been addressed when studying these types of biomarkers using bulk transcriptomics. With the aim of providing a tool able to improve the reproducibility and usability of gene expression signatures, we propose signifinder, an R package that provides the infrastructure to collect, implement, and compare expression-based signatures from cancer literature. The included signatures cover a wide range of biological processes from metabolism and programmed cell death, to morphological changes, such as quantification of epithelial or mesenchymal-like status. Collected signatures can score tumor cell characteristics, such as the predicted response to therapy or the survival association, and can quantify microenvironmental information, including hypoxia and immune response activity. signifinder has been used to characterize tumor samples and to investigate intra-tumor heterogeneity, extending its application to single-cell and spatial transcriptomic data. Through these higher-resolution technologies, it has become increasingly apparent that the single-sample score assessment obtained by transcriptional signatures is conditioned by the phenotypic and genetic intratumor heterogeneity of tumor masses. Since the characteristics of the most abundant cell type or clone might not necessarily predict the properties of mixed populations, signature prediction efficacy is lowered, thus impeding effective clinical diagnostics. Through signifinder, we offer general principles for interpreting and comparing transcriptional signatures, as well as suggestions for additional signatures that would allow for more complete and robust data inferences. We consider signifinder a useful tool to pave the way for reproducibility and comparison of transcriptional signatures in oncology.

Alternative normalization and analysis pipeline to address systematic bias in NanoString GeoMx Digital Spatial Profiling data

Spatial transcriptomics is a novel technique that provides RNA-expression data with tissue-contextual annotations. Quality assessments of such techniques using end-user generated data are often lacking. Here, we evaluated data from the NanoString GeoMx Digital Spatial Profiling (DSP) platform and standard processing pipelines. We queried 72 ROIs from 12 glioma samples, performed replicate experiments of eight samples for validation, and evaluated five external datasets. The data consistently showed vastly different signal intensities between samples and experimental conditions that resulted in biased analysis. We evaluated the performance of alternative normalization strategies and show that quantile normalization can adequately address the technical issues related to the differences in data distributions. Compared to bulk RNA sequencing, NanoString DSP data show a limited dynamic range which underestimates differences between conditions. Weighted gene co-expression network analysis allowed extraction of gene signatures associated with tissue phenotypes from ROI annotations. Nanostring GeoMx DSP data therefore require alternative normalization methods and analysis pipelines.

Challenges and considerations for single-cell and spatially resolved transcriptomics sample collection during spaceflight

Single-cell RNA sequencing (scRNA-seq) and spatially resolved transcriptomics (SRT) have experienced rapid development in recent years. The findings of spaceflight-based scRNA-seq and SRT investigations are likely to improve our understanding of life in space and our comprehension of gene expression in various cell systems and tissue dynamics. However, compared to their Earth-based counterparts, gene expression experiments conducted in spaceflight have not experienced the same pace of development. Out of the hundreds of spaceflight gene expression datasets available, only a few used scRNA-seq and SRT. In this perspective piece, we explore the growing importance of scRNA-seq and SRT in space biology and discuss the challenges and considerations relevant to robust experimental design to enable growth of these methods in the field.

Single-cell and single-nuclei RNA sequencing as powerful tools to decipher cellular heterogeneity and dysregulation in neurodegenerative diseases

Neurodegenerative diseases affect millions of people worldwide and there are currently no cures. Two types of common neurodegenerative diseases are Alzheimer's (AD) and Parkinson's disease (PD). Single-cell and single-nuclei RNA sequencing (scRNA-seq and snRNA-seq) have become powerful tools to elucidate the inherent complexity and dynamics of the central nervous system at cellular resolution. This technology has allowed the identification of cell types and states, providing new insights into cellular susceptibilities and molecular mechanisms underlying neurodegenerative conditions. Exciting research using high throughput scRNA-seq and snRNA-seq technologies to study AD and PD is emerging. Herein we review the recent progress in understanding these neurodegenerative diseases using these state-of-the-art technologies. We discuss the fundamental principles and implications of single-cell sequencing of the human brain. Moreover, we review some examples of the computational and analytical tools required to interpret the extensive amount of data generated from these assays. We conclude by highlighting challenges and limitations in the application of these technologies in the study of AD and PD.

Belayer: Modeling discrete and continuous spatial variation in gene expression from spatially resolved transcriptomics

Spatially resolved transcriptomics (SRT) technologies measure gene expression at known locations in a tissue slice, enabling the identification of spatially varying genes or cell types. Current approaches for these tasks assume either that gene expression varies continuously across a tissue or that a tissue contains a small number of regions with distinct cellular composition. We propose a model for SRT data from layered tissues that includes both continuous and discrete spatial variation in expression and an algorithm, Belayer, to learn the parameters of this model. Belayer models gene expression as a piecewise linear function of the relative depth of a tissue layer with possible discontinuities at layer boundaries. We use conformal maps to model relative depth and derive a dynamic programming algorithm to infer layer boundaries and gene expression functions. Belayer accurately identifies tissue layers and biologically meaningful spatially varying genes in SRT data from the brain and skin.

Computational Methods for Single-cell Multi-omics Integration and Alignment

Recently developed technologies to generate single-cell genomic data have made a revolutionary impact in the field of biology. Multi-omics assays offer even greater opportunities to understand cellular states and biological processes. The problem of integrating different omics data with very different dimensionality and statistical properties remains, however, quite challenging. A growing body of computational tools is being developed for this task, leveraging ideas ranging from machine translation to the theory of networks, and represents another frontier on the interface of biology and data science. Our goal in this review is to provide a comprehensive, up-to-date survey of computational techniques for the integration of single-cell multi-omics data, while making the concepts behind each algorithm approachable to a non-expert audience.

Machine learning-based prediction and inverse design of 2D metamaterial structures with tunable deformation-dependent Poisson's ratio

With the aid of recent efficient and prior knowledge-free machine learning (ML) algorithms, extraordinary mechanical properties such as negative Poisson's ratio have extensively promoted the diverse designs of metamaterials with distinctive cellular structures. However, most existing ML approaches applied to the design of metamaterials are primarily based on a single property value with the assumption that the Poisson's ratio of a material is stationary, neglecting the dynamic variability of Poisson's ratio, termed deformation-dependent Poisson's ratio, during the loading process that a metamaterial other than conventional materials may experience. This paper first proposes a crystallographic symmetry-based methodology to build 2D metamaterials with complex but patterned topological structures, and then converts them into computational models suitable for molecular dynamics simulations. Then, we employ an integrated approach, consisting of molecular dynamics simulations capable of generating and collecting a large dataset for training/validation, in addition to ML algorithms (CNN and Cycle-GAN) able to predict the dynamic characteristics of Poisson's ratio and offer the inverse design of a metamaterial structure based on a target quasi-continuous Poisson's ratio-strain curve, to eventually unravel the underlying mechanism and design principles of 2D metamaterial structures with tunable Poisson's ratio. The close match between the predefined Poisson's ratio response and that from the generated structure validates the feasibility of the proposed ML model. Owing to the high efficiency and complete independence from prior knowledge, our proposed approach offers a novel and robust technique for the prediction and inverse design of metamaterial structures with tailored deformation-dependent Poisson's ratio, an unprecedented property attractive in flexible electronics, soft robotics, and nanodevices.

Scalable and model-free detection of spatial patterns and colocalization

The expeditious growth in spatial omics technologies enables the profiling of genome-wide molecular events at molecular and single-cell resolution, highlighting a need for fast and reliable methods to characterize spatial patterns. We developed SpaGene, a model-free method to discover spatial patterns rapidly in large-scale spatial omics studies. Analyzing simulation and a variety of spatially resolved transcriptomics data showed that SpaGene is more powerful and scalable than existing methods. Spatial expression patterns identified by SpaGene reconstruct unobserved tissue structures. SpaGene also successfully discovers ligand-receptor interactions through their colocalization.

Emerging Artificial Intelligence Applications in Spatial Transcriptomics Analysis

Spatial transcriptomics (ST) has advanced significantly in the last few years. Such advancement comes with the urgent need for novel computational methods to handle the unique challenges of ST data analysis. Many artificial intelligence (AI) methods have been developed to utilize various machine learning and deep learning techniques for computational ST analysis. This review provides a comprehensive and up-to-date survey of current AI methods for ST analysis.

Metastatic Transition of Pancreatic Ductal Cell Adenocarcinoma Is Accompanied by the Emergence of Pro-Invasive Cancer-Associated Fibroblasts

Cancer-associated fibroblasts (CAFs) are now appreciated as key regulators of cancer metastasis, particularly in cancers with high stromal content, e.g., pancreatic ductal cell carcinoma (PDAC). However, it is not yet well understood if fibroblasts are always primed to be cooperative in PDAC transition to metastasis, if they undergo transformation which ensures their cooperativity, and if such transformations are cancer-driven or intrinsic to fibroblasts. We performed a fibroblast-centric analysis of PDAC cancer, as it transitioned from the primary site to trespass stromal compartment reaching the lymph node using published single-cell RNA sequencing data by Peng et al. We have characterized the change in fibroblast response to cancer from a normal wound healing response in the initial stages to the emergence of subclasses with myofibroblast and inflammatory fibroblasts such as signatures. We have previously posited "Evolved Levels of Invasibility (ELI)", a framework describing the evolution of stromal invasability as a selected phenotype, which explains the large and correlated reduction in stromal invasion by placental trophoblasts and cancer cells in certain mammals. Within PDAC samples, we found large changes in fibroblast subclasses at succeeding stages of PDAC progression, with the emergence of specific subclasses when cancer trespasses stroma to metastasize to proximal lymph nodes (stage IIA to IIB). Surprisingly, we found that the initial metastatic transition is accompanied by downregulation of ELI-predicted pro-resistive genes, and the emergence of a subclass of fibroblasts with ELI-predicted increased invasibility. Interestingly, this trend was also observed in stellate cells. Using a larger cohort of bulk RNAseq data from The Cancer Genome Atlas for PDAC cancers, we confirmed that genes describing this emergent fibroblast subclass are also correlated with lymph node metastasis of cancer cells. Experimental testing of selected genes characterizing pro-resistive and pro-invasive fibroblast clusters confirmed their contribution in regulating stromal invasability as a phenotype. Our data confirm that the complexity of stromal response to cancer is really a function of stage-wise emergence of distinct fibroblast clusters, characterized by distinct gene sets which confer initially a predominantly pro-resistive and then a pro-invasive property to the stroma. Stromal response therefore transitions from being tumor-limiting to a pro-metastatic state, facilitating stromal trespass and the onset of metastasis.