Single-cell phenomics through integrated imaging and molecular profiling

Johannes Bues, Joern Pezoldt, Camille L G Lambert, Benjamin D. Hale, Elisa Bugani, Ramón Viñas, Tiziana Ferrari, Nadia Grenningloh, Vincent Gardeux, Shuo Wen, Caroline Wandinger, Maximilian Kohnen, R. Augustin, Katharina Eckstein, Assia Ouanaya, Jillian Love, S. Saha, Amirhossein Saba, Aviv Huettner, Maria Vittoria Impagliazzo, Jose Antonio Porto Viso Vasquez, Angel De Jesus Corria Osorio, Demetri Psaltis, Wouter R. Karthaus, Berend Snijder, Maria Brbić, Bart De StrooperbioRxiv 2025 · DOI

TL;DR

This paper introduces IRIS, a novel platform that seamlessly integrates high-resolution microscopy with single-cell RNA sequencing. IRIS allows researchers to link detailed cellular morphology directly to molecular profiles, enabling a deeper understanding of cellular function and heterogeneity. This breakthrough opens new avenues for dissecting how cellular form relates to molecular state and function.

What Problem Does This Paper Address?

Understanding cellular identity and function has been revolutionized by single-cell technologies like transcriptomics and microscopy. However, these technologies often operate in isolation, providing either molecular snapshots or morphological views, but rarely both from the same cell. This disconnect limits our ability to fully grasp how a cell's physical characteristics (morphology) influence its molecular behavior and vice versa. Traditional methods for linking morphology to molecular data are often low-resolution, labor-intensive, or lack scalability. This paper addresses the critical need for a high-throughput, deterministic platform that can precisely pair high-resolution cellular images with comprehensive molecular profiles from individual cells.

Methods

The core innovation is the IRIS (Interconnected Robotic Imaging and Single cell transcriptomics) platform. Key methodological components include:

  • Deterministic Microfluidics: IRIS uses a microfluidic chip with Quake valves to precisely control cell movement, allowing for stopped-flow imaging and deterministic cell capture. This ensures that each cell's image is reliably linked to its subsequent molecular data.
  • High-Resolution Imaging: The platform integrates machine vision-guided microscopy to capture detailed brightfield and multi-channel fluorescence images of individual cells. This includes z-stack acquisition for 3D morphological information.
  • Droplet-Based Single-Cell RNA Sequencing: Cells are encapsulated in nanoliter droplets, where their mRNA is captured and reverse transcribed. This droplet-based approach is crucial for scalability.
  • Dual-Indexing Barcoding: IRIS employs a novel dual-indexing strategy. A "Cell Code" (CC) is introduced during reverse transcription within the droplet, and a "Well Code" (WC) is added later via PCR. This CC+WC combination creates a unique barcode pair for each cell, ensuring robust traceability from image to transcriptome.
  • "Droplet Consortia": To enhance throughput, IRIS can process multiple cells (up to 13) within the same well by iteratively introducing different Cell Codes. This "droplet consortia" method allows for native multiplexing while maintaining traceability.
  • Machine Learning for Morphology-Gene Expression Linking: The platform leverages machine learning models (CNNs and MLPs) to predict gene expression from morphological features (nucleus shape, cell shape) and cell cycle phase.

Datasets

  • HEK293T and NIH/3T3 cells: Used for initial validation of the platform's species-mixing capabilities and continuous processing.
  • FUCCI-expressing 3T3 fibroblasts: Used for benchmarking the platform's ability to capture continuous cell cycle dynamics and link it to transcriptomic data. This serves as a new benchmark for integrated cell cycle phenomics.
  • Peripheral Blood Mononuclear Cells (PBMCs): Used to demonstrate IRIS's ability to resolve morphological heterogeneity within immune cell populations.
  • Naïve CD8+ T cells: A specific subset of PBMCs, used to investigate the link between nuclear-ER architecture and transcriptional programs.

Key Insights and Ideas

  • Morphology is a Rich Source of Biological Information: IRIS demonstrates that detailed cellular morphology, beyond simple size or shape, encodes significant biological information that is often missed by transcriptomic-only analyses.
  • Linking Structure to Function: The platform provides a direct link between cellular architecture (e.g., nuclear shape, ER organization) and specific transcriptional programs, revealing previously uncharacterized axes of cellular heterogeneity.
  • Continuous Cell Cycle Dynamics: IRIS can capture continuous cell cycle progression, revealing regulatory associations with progression speed that are difficult to discern with traditional discrete phase annotations.
  • Machine Learning as a Bridge: Machine learning models can effectively learn to predict gene expression from morphological features, highlighting the predictive power of imaging data and enabling inference of molecular states from morphology.
  • Uncovering Hidden Biological Processes: By integrating imaging and transcriptomics, IRIS can reveal subtle, non-discrete cell states and hidden biological processes that are obscured in single-modality analyses.

Key Results

  • High-Resolution Imaging and Transcriptome Pairing: IRIS achieves precise one-to-one pairing of submicron-resolved morphological features with transcriptomes.
  • Scalability and Sensitivity: The platform offers scalable processing (up to 1,150 cells/h) and outperforms existing droplet-based platforms in gene detection sensitivity (2.6-fold higher than Drop-seq).
  • Cell Cycle Dynamics: IRIS successfully profiled cell cycle dynamics in FUCCI-expressing fibroblasts, identifying 670 cycling genes and uncovering relationships between FUCCI reporter intensity and DREAM complex activity.
  • T Cell Heterogeneity: The platform resolved distinct transcriptional programs associated with nuclear-ER architectures in naïve CD8+ T cells (TØ vs. TO cells), identifying novel molecular targets and surface markers for these subsets.
  • Machine Learning Performance: Morphology-based models could predict gene expression with a mean Pearson correlation of 39-47% of the reference model, with cell shape models performing better in early G1 and nuclear models in S phase.

Limitations and Future Work

  • Throughput: While improved, the throughput of ~1,000 cells per hour needs to increase for population-scale studies.
  • Suspension-Based Profiling: IRIS profiles cells in suspension, which removes spatial context present in adherent or tissue environments. This defines a complementary niche to spatial transcriptomics.
  • Imaging Resolution: Further improvements in imaging resolution could reveal deeper layers of cellular heterogeneity.
  • Widespread Adoption: Engineering refinements and community access are needed for broader adoption.

Future work could involve exploring cellular dynamics, monitoring perturbation responses, and developing new capabilities like image-based sorting for prospective isolation of rare morphotypes. The platform's ability to build image-based predictors calibrated against transcriptomic data holds translational promise for applications in diagnostics and therapy monitoring.

Summary updated

Abstract

Abstract Single-cell technologies such as transcriptomics, microscopy, and flow cytometry have revolutionized the study of cellular identity and function. While each of these technologies is powerful on its own, their full potential lies in their integration, enabling multimodal profiling of the same cell and revealing how distinct modalities influence one another. Here, we introduce IRIS (Interconnected Robotic Imaging and Single cell transcriptomics), a deterministic single-cell platform technology that seamlessly couples high-resolution microscopy with droplet-based single-cell RNA sequencing. IRIS enables precise cell positioning, multimode imaging across brightfield and fluorescent channels, and subsequent molecular capture from the same cell, directly linking high-resolution morphological features to matched transcriptomes. We validate IRIS by recovering cell cycle progression states and transcriptional programmes associated with canonical morphologies and demonstrate its discovery power by molecularly resolving two nuclear-ER architectures within naïve CD8 + T cells, each defined by distinct gene expression profiles and functional markers. IRIS establishes an integrative single-cell phenomics framework, opening new avenues for dissecting how cellular form relates to molecular state and function.
BibTeX
@misc{bues2025single,
  author = {Johannes Bues and Joern Pezoldt and Camille L G Lambert and Benjamin D. Hale and Elisa Bugani and Ramón Viñas and Tiziana Ferrari and Nadia Grenningloh and Vincent Gardeux and Shuo Wen and Caroline Wandinger and Maximilian Kohnen and R. Augustin and Katharina Eckstein and Assia Ouanaya and Jillian Love and S. Saha and Amirhossein Saba and Aviv Huettner and Maria Vittoria Impagliazzo and Jose Antonio Porto Viso Vasquez and Angel De Jesus Corria Osorio and Demetri Psaltis and Wouter R. Karthaus and Berend Snijder and Maria Brbić and Bart De Strooper},
  title = {{Single-cell phenomics through integrated imaging and molecular profiling}},
  year = {2025},
  doi = {10.1101/2025.11.28.690954},
  howpublished = {bioRxiv},
}
Figure