De novo detection of somatic variants in high-quality long-read single-cell RNA sequencing data

Arthur Dondi, Nico Borgsmüller, Pedro Ferreira, Brian J. Haas, Francis Jacob, Viola Heinzelmann‐Schwarz, Niko Beerenwinkel

TL;DR

This paper introduces LongSom, a novel computational workflow that leverages long-read single-cell RNA sequencing (LR scRNA-seq) to detect various types of somatic genetic variations (SNVs, CNAs, gene fusions) directly from tumor samples without the need for matched normal tissue. This enables a more comprehensive reconstruction of tumor clonal heterogeneity and the identification of subclones with potentially different treatment responses.

What Problem Does This Paper Address?

Cancer is characterized by accumulating genetic and transcriptomic variations that lead to clonal heterogeneity, a major driver of treatment resistance and poor patient outcomes. Understanding this heterogeneity at a single-cell level is crucial for developing effective therapies. While short-read single-cell RNA sequencing (SR scRNA-seq) has been used, it struggles to detect gene fusions and isoforms. Long-read single-cell RNA sequencing (LR scRNA-seq) offers the potential to simultaneously capture genetic and transcriptomic information from full-length RNA molecules. However, accurately calling de novo somatic variants (variants present in the tumor but not in normal tissue) from this data, especially without matched normal samples, presents a significant challenge due to technical noise, ambient RNA contamination, and the inherent complexity of tumor biology. Existing methods often rely on matched normal samples or struggle with the accuracy of cell type annotations, which are critical for distinguishing somatic from germline variants.

Methods

LongSom is a computational workflow designed to address these challenges. Its key innovations include:

  • De Novo Variant Calling: LongSom can identify somatic single-nucleotide variants (SNVs), mitochondrial SNVs (mtSNVs), copy number alterations (CNAs), and gene fusions directly from LR scRNA-seq data without requiring matched normal samples.
  • Cell Type Reannotation: A crucial step involves reannotating cell types based on identified "high-confidence cancer variants." This helps to correct misclassifications that can arise from gene expression data alone, improving the accuracy of downstream variant calling.
  • Extensive Filtering: The workflow employs a multi-stage filtering process, incorporating hard filters and statistical tests (including beta-binomial modeling of error rates) to distinguish true somatic variants from technical noise and germline polymorphisms. This includes specific filters for mtSNVs to account for ambient RNA contamination.
  • Clonal Reconstruction: LongSom integrates the detected variants (SNVs, fusions, and CNAs) to infer the clonal structure of the tumor using Bayesian nonparametric clustering (BnpC) and inferCNV.

Datasets

The primary dataset used for developing and evaluating LongSom consists of high-quality long-read single-cell RNA sequencing (PacBio) data from omentum metastasis samples of three chemo-naive high-grade serous carcinoma (HGSOC) patients (P1, P2, P3). This dataset includes both cancer cells and microenvironment cells. For validation, the study also utilized:

  • Single-cell whole-genome sequencing (scWGS) data from matched tumor samples.
  • Bulk panel DNA sequencing data from the same patients.
  • SR scRNA-seq data from the same patients for comparative analysis.

Key Insights and Ideas

  • LR scRNA-seq is a powerful tool for comprehensive variant detection: The study demonstrates that LR scRNA-seq can simultaneously capture diverse genetic variations (SNVs, CNAs, fusions) and transcriptomic information within the same cell, offering a holistic view of tumor biology.
  • Cell type reannotation based on mutational profiles is critical: Misannotated cells can lead to significant false-negative variant calls. LongSom's approach of reannotating cells based on their mutational burden significantly improves variant detection sensitivity.
  • Mitochondrial SNVs are valuable but require careful handling: While mtSNVs are abundant and can be informative for cell type identification and clonal analysis, they are prone to contamination from ambient RNA, necessitating specialized filtering strategies.
  • De novo variant calling without matched normals is feasible and impactful: LongSom's ability to accurately call somatic variants without normal samples democratizes this type of analysis, especially in clinical settings where matched normals may not be available.
  • Somatic variants can predict treatment response: By identifying subclones with distinct mutational profiles, LongSom can highlight potential differences in therapeutic sensitivity, paving the way for personalized treatment strategies.

Key Results

  • LongSom successfully detected somatic SNVs, mtSNVs, CNAs, and gene fusions in LR scRNA-seq data.
  • Cell type reannotation improved variant detection, identifying up to 31% more somatic SNVs in one patient.
  • Validation against scWGS data showed that 80% of LongSom's somatic SNV calls were supported as somatic, with only 11% supported as germline.
  • LongSom outperformed SComatic (a SR scRNA-seq variant caller) in terms of the number of somatic SNVs detected and their concordance with scWGS data, particularly in challenging samples.
  • The workflow identified distinct cancer subclones in patient P1, with genetic and transcriptomic differences suggesting differential sensitivity to therapies like cisplatin and PARP inhibitors.
  • Comparison with SR scRNA-seq revealed that LR scRNA-seq detected significantly more somatic SNVs with better validation support, and was crucial for reconstructing clonal structures.

Limitations and Future Work

  • Data Quality Dependence: LongSom's performance is highly dependent on the quality of the LR scRNA-seq data (e.g., PacBio's Q20 threshold). Further testing on data generated with newer Nanopore chemistries is needed.
  • Ambient RNA Contamination: While LongSom has strategies to mitigate ambient RNA contamination (especially for mtSNVs), it remains a challenge that could lead to false positives or negatives. The biological implications of this contamination (technical artifact vs. biological mechanism) require further investigation.
  • Coverage Limitations: Like all scRNA-seq methods, LongSom is limited by sequencing coverage. Low coverage in noncancer cells can lead to potential false negatives, and LongSom filters sites with low coverage in noncancer cells.
  • Intronic/Intergenic Variants: While LongSom can detect variants in intronic and intergenic regions, their functional significance may require further investigation.
  • Future Directions: The authors envision LR scRNA-seq becoming more common for cancer research, and LongSom can serve as a foundational tool for guiding these analyses. Future work could involve testing LongSom on a wider range of cancer types and integrating it with other omics data for a more comprehensive understanding of cancer evolution and treatment resistance.

Summary updated

Abstract

In cancer, genetic and transcriptomic variations generate clonal heterogeneity, leading to treatment resistance. Long-read single-cell RNA sequencing (LR scRNA-seq) has the potential to detect genetic and transcriptomic variations simultaneously. Here, we present LongSom, a computational workflow leveraging high-quality LR scRNA-seq data to call de novo somatic single-nucleotide variants (SNVs), including in mitochondria (mtSNVs), copy number alterations (CNAs), and gene fusions, to reconstruct the tumor clonal heterogeneity. Before somatic variant calling, LongSom reannotates marker gene-based cell types using cell mutational profiles. LongSom distinguishes somatic SNVs from noise and germline polymorphisms by applying an extensive set of hard filters and statistical tests. Applying LongSom to human ovarian cancer samples, we detected clinically relevant somatic SNVs that were validated against matched DNA samples. Leveraging somatic SNVs and fusions, LongSom found subclones with different predicted treatment outcomes. In summary, LongSom enables de novo variant detection without the need for normal samples, facilitating the study of cancer evolution, clonal heterogeneity, and treatment resistance.
BibTeX
@inproceedings{dondi2025de,
  author = {Arthur Dondi and Nico Borgsmüller and Pedro Ferreira and Brian J. Haas and Francis Jacob and Viola Heinzelmann‐Schwarz and Niko Beerenwinkel},
  title = {{De novo detection of somatic variants in high-quality long-read single-cell RNA sequencing data}},
  booktitle = {Genome},
  year = {2025},
  doi = {10.1101/gr.279281.124},
}
Figure