TL;DR
This paper introduces Blood Cell Painting (BCP), a high-throughput imaging technique to profile the morphology of millions of immune cells from healthy donors. By linking these detailed cellular images to genetic data, they uncovered novel genotype-phenotype associations, revealing how genetic variations influence immune cell shape and function. This work establishes BCP as a powerful tool for understanding immune cell heterogeneity in both health and disease.
What Problem Does This Paper Address?
Understanding the vast diversity of immune cells in healthy individuals is crucial for identifying disease-related cellular changes. However, the detailed morphology of these cells and how it relates to an individual's genetic makeup has been largely uncharacterized. Existing methods like single-cell RNA sequencing can reveal gene expression but often miss the rich morphological information that can be indicative of cellular state and function. This paper aims to bridge this gap by developing a method to capture and analyze immune cell morphology at scale, and then connect it to genetic and clinical data.
Methods
The core of this work is the Blood Cell Painting (BCP) assay, a modified version of the established Cell Painting technique adapted for peripheral blood mononuclear cells (PBMCs). BCP uses a panel of fluorescent dyes to stain various cellular components (nucleus, mitochondria, ER/Golgi, actin, and nucleic acid), allowing for detailed morphological profiling via high-content imaging.
Key methodological aspects include:
- High-Content Imaging: Millions of cells are imaged using a specialized microscope, capturing detailed fluorescent images.
- Image Analysis Pipeline: A sophisticated pipeline is used for image processing, including illumination correction, nucleus and cell segmentation (using deep learning models), and extraction of 186 distinct cellular features (intensity, shape, texture).
- Batch Effect Correction: Given the large scale of the data, methods like Harmony were employed to correct for technical variations arising from different plates, wells, and imaging conditions.
- Cell Typing: A supervised machine learning approach (Support Vector Machine) was trained on expert-annotated cells to classify PBMCs into four categories: monocytes, macrophages, B & T cells, and uncategorized.
- Genome-Wide Association Study (GWAS): BCP features were treated as quantitative traits to identify genetic variants (SNPs) associated with specific cellular morphologies.
- Phenome-Wide Association Study (PheWAS)-like analysis: Associations between cellular morphology and clinical traits from health registries were explored.
Datasets
- Healthy Donor PBMCs: The study utilized a large dataset of approximately 50 million cells from 390 healthy blood donors, collected as part of the FinnGen project.
- FinnGen Genotype Data: Genotype data for these donors was available from the FinnGen R12 release, encompassing ~10.4 million germline variants.
- Health Registry Data: Demographic factors (age, sex) and health registry endpoints were also used for association studies.
- snRNA-seq Data: For validation, snRNA-seq data from 315 donors was used to compare cell type proportions.
- Acute Myeloid Leukemia (AML) Patient Samples: A proof-of-concept study included PBMC samples from six AML patients and one AML cell line (MOLM-13) to demonstrate BCP's applicability in disease.
Key Insights and Ideas
- Morphological Heterogeneity is Genetically Influenced: The study strongly suggests that subtle differences in immune cell shape and internal structure are not random but are significantly influenced by an individual's germline genetics.
- BCP as a Bridge: BCP effectively bridges the gap between genotype and phenotype by providing a detailed, quantitative measure of cellular morphology that can be directly linked to genetic variations.
- Novel Gene-Morphology Links: The GWAS identified 30 new genomic loci associated with immune cell morphology, pointing to genes involved in critical cellular processes like inflammation, immune signaling, and mitochondrial function.
- Clinical Relevance of Morphology: Beyond genetics, cellular morphology was found to correlate with clinical traits, hinting at how cellular states might reflect or influence health conditions.
- BCP for Disease Profiling: The successful differentiation of AML subtypes using BCP demonstrates its potential as a diagnostic or prognostic tool.
Key Results
- BCP Atlas: The creation of the largest Cell Painting dataset of PBMCs to date, comprising 50 million imaged cells from 390 healthy donors.
- 18 Distinct Morphology Clusters: Analysis of the BCP Atlas revealed 18 distinct clusters of immune cells based on their morphology.
- 93 Significant GWAS Associations: A genome-wide association study identified 93 significant associations between BCP-derived cellular features and genetic variants, spanning 30 genomic loci.
- Functional Gene Links: These loci include genes known to be involved in transcriptional regulation, inflammatory signaling, immune signaling, and circadian rhythms.
- Clinical Trait Correlations: Associations were found between cell morphology and clinical traits like respiratory conditions and healthcare visits related to contraceptive management.
- AML Subtype Differentiation: BCP was able to distinguish between different subtypes of acute myeloid leukemia, serving as a proof of concept for clinical application.
Limitations and Future Work
- Cell Type Frequency Discrepancies: BCP-based cell typing showed some differences in cell type frequencies compared to flow cytometry (e.g., higher B/T cell counts at well edges). The authors suggest optimizing well coatings to improve adhesion of non-adherent cell types.
- Batch Effect Correction Challenges: While Harmony performed well, memory limitations prevented the evaluation of all three batch effect factors (plate, well, FOV) simultaneously.
- Uncategorized Cells: A significant portion of cells remained uncategorized by the supervised classifier, though these often formed distinct clusters suggesting they might represent specific cell types, dead cells, or debris.
- Future Work: The authors envision training label-free models on large BCP datasets for more inexpensive morphological profiling. They also highlight the potential of their BCP Atlas as a valuable resource for future research to characterize disease phenotypes and establish a baseline for comparative analyses, particularly in areas like leukemia research.
Summary updated