Abstract
Single-cell RNA sequencing (scRNA-seq) has revolutionized our understanding of cell types and tissues. However, empty droplets and poor quality cells are often captured in single cell genomics experiments and need to be removed to avoid cell type interpretation errors. Many automated and manual methods exist to identify poor quality cells or empty droplets, such as minimum RNA count thresholds and comparing the gene expression profile of an individual cell to the overall background RNA expression of the experiment. A versatile approach is to use unbalanced overall RNA splice ratios of cells to identify poor quality cells or empty droplets. However, this approach is computationally intensive, requiring a detailed search through all sequence reads in the experiment to quantify spliced and unspliced reads. We found that the expression level of MALAT1, a non-coding RNA retained in the nucleus and ubiquitously expressed across cell types, is strongly correlated with this splice ratio measure and thus can be used to similarly identify low quality cells in scRNA-seq data. Since it is easy to visualize the expression of a single gene in single-cell maps, MALAT1 expression is a simple cell quality measure that can be quickly used during the cell annotation process to improve the interpretation of cells in tissues of human, mouse and other species with a conserved MALAT1 function.
Introduction
Single-cell RNA sequencing (scRNA-seq) is a transformative tool for biological discovery, yet the resulting data is inherently noisy. In droplet-based methods, such as 10x Genomics, thousands of oil droplets are intended to capture individual cells. However, many droplets capture only "ambient RNA"—genetic debris from ruptured cells—or remain empty.
To address this, researchers utilize quality control (QC) metrics. One highly accurate metric is the nuclear fraction, defined as the ratio of unspliced (intron-retaining) RNA to spliced mRNA. Because unspliced RNA is primarily localized within the nucleus, a low nuclear fraction indicates a droplet likely lacks a nucleus and contains only debris. Despite its accuracy, calculating this ratio is computationally intensive, requiring access to raw sequencing files (BAM files) and significant processing time to quantify intronic reads.
Methods
This study proposes that MALAT1, a long non-coding RNA (lncRNA) that is highly abundant and strictly retained in the nucleus, can serve as a biological proxy for the presence of a nucleus. To validate this, the researchers performed correlation analyses between normalized MALAT1 expression and the nuclear fraction (calculated via DropletQC) across 16 datasets comprising over 215,000 cells from human and mouse tissues.
Additionally, the authors compared MALAT1 levels in droplets identified as "empty" versus those containing intact cells. They also developed an automated thresholding heuristic that fits a quadratic function over the MALAT1 expression distribution to identify optimal filtering cutoffs.
Results
Analysis across diverse tissues—including blood, liver, kidney, and the Tabula Muris Senis atlas—demonstrated that MALAT1 is a robust indicator of cell quality. MALAT1 was the single most correlated gene with the nuclear fraction in 81% of the samples tested. In 10x Genomics samples, MALAT1 was consistently the top differentially expressed gene between empty droplets and intact cells (p < 1.73e-84).
Key findings include:
- Universal Application: MALAT1 is expressed across nearly all mammalian cell types, making it a near-universal "nucleus detector."
- Biological Consistency: The strongest negative correlations were observed in erythrocytes. As mature red blood cells lack a nucleus, they correctly exhibited both low MALAT1 levels and low nuclear fractions.
- Efficiency: Because MALAT1 is included in standard gene expression matrices, researchers can assess quality instantly without raw BAM files.
Discussion and Limitations
While MALAT1 provides a "computationally free" QC metric, certain limitations exist. Some cell types, such as hepatocytes or specific myocytes, are naturally fragile or exhibit variable MALAT1 levels, requiring caution to avoid filtering out valid biological signals. Furthermore, technical batch effects can shift the baseline of MALAT1 expression, necessitating sample-specific thresholding rather than a global cutoff.
Despite these caveats, MALAT1 thresholding offers a significant improvement over traditional mitochondrial DNA metrics, which can conflate high metabolic activity with cell damage. Integrating MALAT1 analysis into standard pipelines like Seurat or Scanpy will make high-quality data filtering more accessible to the broader research community.
Summary updated