TL;DR
Dandelion is a new computational pipeline that integrates single-cell gene expression and adaptive immune receptor sequencing (scVDJ-seq) data. It offers improved analysis of immune cell development by creating a novel "V(D)J feature space" for better trajectory inference and uncovering new insights into T and B cell development, as well as innate lymphoid cells.
What Problem Does This Paper Address?
Studying lymphocyte development and function has been revolutionized by single-cell technologies like scRNA-seq and scVDJ-seq. However, integrating these two data modalities presents challenges, particularly in analyzing the complex, categorical nature of adaptive immune receptor (AIR) repertoire data. Existing tools often struggle with accurate contig annotation, identifying non-productive or partially spliced transcripts, and effectively leveraging AIR data for trajectory inference. This paper aims to bridge this gap by providing a comprehensive computational framework, Dandelion, that streamlines scVDJ-seq analysis and enables deeper insights into immune cell development.
Methods
Dandelion offers a holistic pipeline for scVDJ-seq analysis, building upon existing AIRR standards. Key methodological contributions include:
- Refined Contig Annotation: Dandelion improves V(D)J contig annotation by re-annotating with
igblastnandblastn, specifically addressing challenges like identifying non-productive contigs (those that don't translate to functional proteins) and partially spliced transcripts (multi-J mapping). It also enhances γδTCR annotation compared to existing tools. - Novel V(D)J Feature Space: A core innovation is the creation of a "V(D)J feature space." This transforms categorical V(D)J gene usage data into a continuous format by creating pseudo-bulks of cells and computing the frequency of gene usage within these pseudo-bulks. This space can then be used for visualization (e.g., UMAP) and downstream analyses.
- Pseudotime Trajectory Inference: Dandelion leverages the V(D)J feature space for pseudotime trajectory inference. By treating AIR repertoire as a "time-keeper" for cell development, it allows for more accurate modeling of differentiation pathways, particularly for T cells, overcoming limitations of transcriptomic similarity-based methods.
- Integration with scRNA-seq: The pipeline is designed to integrate seamlessly with scRNA-seq data, enabling joint analysis and the exploration of relationships between immune receptor usage and gene expression.
Datasets
The study utilizes data from two primary sources:
- Fetal Human Lymphoid Cells: Data from Suo et al. 2022, which includes scRNA-seq and scVDJ-seq from various fetal immune cell subsets. This dataset was used for analyzing T cell development, B1 cell origins, and ILC/NK cell development.
- Adult Human T Cells: Data from Conde et al. 2022, used to demonstrate the utility of the V(D)J feature space for differential V(D)J usage analysis and pseudotime inference in T cell subsets.
- PBMCs: Newly generated data from control and cycloheximide-treated human PBMCs was used to investigate factors associated with multi-J mapping.
Key Insights and Ideas
- Non-productive contigs are biologically meaningful: Contrary to common practice, Dandelion highlights that non-productive TCR and BCR contigs, often discarded, can provide valuable "fossil records" of a cell's developmental history and origin.
- B1 cell origins might be more direct: The presence of non-productive TRB contigs in B1 cells suggests they may arise directly from pre-pro B cells, bypassing some conventional B cell differentiation stages.
- T and ILC/NK cell development share early stages: Non-productive TRB contigs were also found in ILC/NK cells, supporting the "aborted DN" theory, suggesting a shared early developmental trajectory with T cells before diverging.
- V(D)J data offers a more robust "time-keeper" for cell development: The V(D)J feature space provides a more accurate pseudotime trajectory inference for T cell development compared to gene expression alone, especially in capturing early lineage commitment decisions.
- Multi-J mapping is linked to NMD: The phenomenon of multi-J mapping (multiple J genes on a single contig) is associated with V gene presence and potentially nonsense-mediated decay (NMD), an RNA degradation pathway.
Key Results
- Dandelion improved the alignment of human thymic development trajectories, leading to new predictions about factors regulating T cell lineage commitment.
- Analysis of other cell compartments provided novel insights into the origins of human B1 cells and ILC/NK cell development.
- The V(D)J feature space successfully clustered MAIT cells separately from other T cell types, highlighting its unique V(D)J usage.
- Dandelion demonstrated a higher recovery rate of high-confidence γδTCR contigs compared to 10X Genomics' cellranger.
- The V(D)J-based pseudotime inference accurately ordered DP T cells and revealed early CD4/CD8 lineage decision genes, with better alignment to known biology than GEX-based pseudotime.
Limitations and Future Work
- BCR Trajectory Optimization: The current approach is not yet fully optimized for BCR trajectories due to limitations in the number of B progenitors in existing datasets and the complex rearrangement rules for BCRs (e.g., somatic hypermutation, light chain editing). Future work aims to improve this.
- Multi-J Mapping Implications: The biological implications of multi-J mapping are still unclear and require further experimental validation.
- pDC BCR Analysis: The analysis of pDC BCR expression is limited by cell numbers in the current dataset, suggesting future investigations with larger datasets are warranted.
- Human vs. Mouse DN Development: Discordances in transcription factor expression patterns between human and mouse DN development were observed, suggesting further comparative studies are needed.
- Exploration of Novel Associations: The identification of novel associations between trajectories and transcription factors for CD4/CD8 lineage commitment warrants further investigation.
Summary updated