TEsingle enables locus-specific transposable element expression analysis at single-cell resolution

The paper introduces TEsingle, a computational tool that overcomes challenges in repetitive sequence alignment and intron retention to enable accurate, locus-specific analysis of transposable element expression in single-cell transcriptomes, revealing cell-type-specific TE dysregulation in Parkinson's disease.

The Gist

Imagine your genome is a massive, bustling library containing the instructions for building and running a human being. Inside this library, there are two main types of books:

1. The Instruction Manuals (Genes): These are the clear, organized chapters that tell your cells how to function.
2. The "Glitchy" Copy-Paste Stickers (Transposable Elements or TEs): These are ancient, viral-like sequences that make up nearly half of the library. They are like stickers that got stuck all over the walls, the floor, and even inside the instruction manuals. Most of them are old and broken, but some are still active and can copy themselves to new spots.

For a long time, scientists could read the "Instruction Manuals" very well. But trying to read the "Glitchy Stickers" in a single cell was like trying to find a specific, slightly torn sticker in a dark room while wearing foggy glasses. Because these stickers look almost identical to each other, and because they are often stuck inside the instruction manuals, it was incredibly hard to tell:

• Is this sticker actually being read (expressed)?
• Which specific sticker is it?
• Or is it just a piece of an instruction manual that got stuck halfway through (an intron)?

Enter: TEsingle (The Super-Organized Librarian)

The authors of this paper built a new tool called TEsingle. Think of TEsingle as a super-smart, high-tech librarian who has a special set of glasses and a new filing system.

How it works (The Analogy):
Imagine you are trying to count how many times a specific sticker was used in a single cell.

• The Old Way: Previous tools would look at a page and say, "This looks like a sticker, but it also looks like part of a manual. I'll just guess." This led to a lot of confusion and wrong counts.
• The TEsingle Way: TEsingle looks at the "barcode" on every single piece of paper (the UMI) to make sure it's counting unique items, not just photocopies. It then uses a clever math trick (called an Expectation-Maximization algorithm) to play a game of "most likely." It asks: "Given all the clues, is this piece of text more likely to be a sticker or a manual?" It does this for every single cell, one by one.

The result? TEsingle can tell you exactly which specific sticker is active in a specific cell, without getting confused by the messy parts of the instruction manuals.

The Big Discovery: The Parkinson's Disease Mystery

The researchers took this new tool and applied it to brain tissue from patients with Parkinson's Disease (PD). Parkinson's is a disease where specific brain cells (dopamine neurons) die off, causing movement problems.

They wanted to see if the "Glitchy Stickers" were acting up in these sick brains. Here is what they found, using their new super-librarian tool:

1. It's Not Just "General Noise": They discovered that the stickers aren't just randomly active everywhere. They are very specific.

   • In Neurons: Certain young, "fresh" stickers were waking up specifically in the dopamine neurons that are dying in Parkinson's. It's like finding that only the books on the "Motor Control" shelf have been vandalized by a specific type of sticker.
   • In Immune Cells (Microglia & Astrocytes): The brain's immune cells (which act like the library's security guards) were also showing signs of stress. In these cells, the stickers were acting like alarm bells. The more "angry" or "reactive" the immune cell was, the more stickers were active.

2. The "Young" Culprits: The stickers that were causing the most trouble were the "young" ones—versions that haven't been in the human genome for millions of years. They are still intact and capable of moving around. It's as if the library's security system failed specifically against the newest, most aggressive intruders.

Why This Matters

Before this tool, scientists were like people trying to hear a whisper in a storm; they knew something was happening, but they couldn't make out the words.

TEsingle cleared the storm. It showed us that:

• Parkinson's isn't just about dying neurons; it involves a chaotic reaction from the brain's immune system and specific "glitches" in the genetic code.
• Specific "bad actors" (specific stickers) are linked to specific cell types. This gives scientists new targets to look at. Maybe if we can calm down these specific stickers, we can help the brain cells survive.

In short: The authors built a better microscope (TEsingle) that finally lets us see the messy, chaotic "sticker" part of our DNA. When they used it to look at Parkinson's brains, they found that these stickers are waking up in very specific, dangerous ways, giving us a new clue about how the disease works and how we might stop it.

Technical Summary

1. Problem Statement

Transposable elements (TEs) constitute nearly 45% of the human genome. While most are dormant, specific subsets are transcriptionally active in aging and neurodegenerative diseases like Parkinson's Disease (PD). However, accurately quantifying TE expression in single-cell RNA sequencing (scRNA-seq) and single-nucleus RNA sequencing (snRNA-seq) data remains a significant challenge due to:

• Repetitive Nature: TEs exist in multiple copies across the genome, leading to multi-mapping reads that are difficult to assign to a specific locus.
• Intron Retention: In snRNA-seq, unprocessed pre-mRNA containing introns constitutes up to 40% of transcripts. Since many TEs reside within introns, distinguishing between true TE expression and intron retention events is difficult.
• Locus Specificity: Existing tools often aggregate TE expression at the subfamily level, losing the resolution required to identify specific, potentially active, or disease-relevant TE loci.
• UMI Handling: Standard single-cell pipelines often fail to properly resolve Unique Molecular Identifiers (UMIs) when dealing with ambiguous TE alignments, leading to PCR duplicate inflation or loss of data.

2. Methodology

The authors developed TEsingle, a computational pipeline designed to quantify both gene and locus-specific TE expression in single-cell and single-nucleus data.

• Algorithmic Core:

  • Input: Accepts STARsolo-aligned BAM/SAM files, gene GTF annotations, and TE GTF annotations.
  • UMI Graph Construction: Groups reads by cell barcode and constructs a UMI graph where nodes represent UMIs and edges connect UMIs differing by one Hamming distance (accounting for sequencing/PCR errors). Each subgraph represents a single originating transcript.
  • Ambiguity Resolution:
    • Uniquely Mapping Reads: Used as anchors to restrict the list of possible origins for a transcript.
    • Expectation-Maximization (EM): An iterative EM algorithm is applied only to ambiguous reads with potential TE annotations to probabilistically assign them to the most likely locus. Crucially, uniquely mapping gene reads are excluded from the EM step to prevent bias toward highly expressed genes.
  • Intron Handling: Models gene annotations as full-length transcripts (including introns) rather than excluding intronic reads, improving the disambiguation of intron retention vs. intronic TE expression.
  • Output: Generates a count matrix in Matrix Market Exchange (MEX) format compatible with standard downstream tools like Seurat.

• Benchmarking Strategy:

  • Synthetic Data Generation: Used FluxSimulator to generate realistic scRNA-seq and snRNA-seq datasets with known ground truth. Parameters were calibrated based on real data to reflect specific intron retention rates (20% for whole-cell, 40% for snRNA-seq).
  • Comparison: TEsingle was benchmarked against STARsolo, CellRanger, scTE, and soloTE.
  • Metrics: Accuracy was measured using F1 scores (harmonic mean of precision and recall), with strict definitions for "accurate" (within ±15% of ground truth), "over/underestimated," "false positive," and "false negative."

3. Key Contributions

• Novel Software (TEsingle): The first tool to simultaneously provide accurate gene expression estimates and locus-specific TE expression quantification for single-cell and single-nucleus data.
• Improved Accuracy: Demonstrated that TEsingle outperforms existing tools (including modified versions of STARsolo and CellRanger) in both gene and TE quantification, particularly in snRNA-seq data where intron retention is high.
• Methodological Rigor: Introduced a robust benchmarking framework using synthetic data with calibrated intron retention rates to evaluate TE analysis tools.
• Biological Discovery in PD: Applied TEsingle to post-mortem substantia nigra (SNpc) data from Parkinson's Disease patients, revealing cell-type and state-specific TE dysregulation that was previously undetectable.

4. Results

A. Benchmarking Performance

• Gene Expression: TEsingle achieved an F1 score of 0.87 for gene expression (whole-cell model), outperforming STARsolo-TE (0.81), CellRanger-TE (0.79), and scTE (0.66). It showed that including TEs in the analysis does not degrade gene quantification; in fact, it improves it.
• TE Expression (Locus-Specific): TEsingle achieved an F1 score of 0.74 for locus-specific TE quantification, significantly higher than STARsolo-TE (0.51). Other tools (CellRanger, soloTE, scTE) performed poorly at the locus level, often aggregating to subfamily levels or generating high false-positive rates.
• snRNA-seq Specifics: TEsingle maintained high accuracy in single-nucleus simulations (40% intron retention), whereas other tools showed increased false positives and negatives due to misinterpreting intronic reads as TE expression.

B. Application to Parkinson's Disease (PD)
Using a public snRNA-seq dataset of the SNpc from PD patients and controls:

• Validation: TEsingle successfully recapitulated 84.9% of the differentially expressed genes (DEGs) found in the original gene-only study, confirming its reliability for gene expression.
• TE Dysregulation in Neurons:
  • Dopaminergic Neurons (DA): Showed elevation of young, intact, and human-specific TEs, specifically SVA-F-dup252 and SVA-E-dup596.
  • Excitatory Neurons: Exhibited the largest set of upregulated TEs, including young LINE-1 (L1PA2-dup4380) and ERV families (HERV9, LTR12B).
• TE Dysregulation in Glia:
  • Reactive Astrocytes: A specific subpopulation of reactive astrocytes (marked by HSPH1) showed a global elevation of TEs, including the human-specific HERVK11-dup192.
  • Cytokine-Responding Microglia: This inflammatory microglial state showed the highest TE upregulation among microglia, enriched for young LINE-1 (L1HS-dup497) and HERVK9 elements.
• Specificity: TE expression was highly cell-type and cellular-state specific. Most dysregulated TEs were unique to a single subpopulation, suggesting TEs act as markers for specific disease-associated cellular states.

5. Significance

• Technical Advancement: TEsingle solves the "multi-mapping" and "intron retention" problems that have historically prevented accurate TE analysis in single-cell genomics. It enables researchers to move beyond subfamily-level analysis to the specific genomic locus, which is critical for understanding the functional impact of TEs.
• Biological Insight into Neurodegeneration: The study provides the first high-resolution map of TE activity in PD. It reveals that TE reactivation is not a global phenomenon but is tightly linked to specific pro-inflammatory cellular states (reactive astrocytes, cytokine-responding microglia) and vulnerable dopaminergic neurons.
• Mechanistic Hypothesis: The findings suggest that specific young TEs (like SVA and HERVK families) may act as inflammatory sensors or drivers in PD, potentially contributing to neuroinflammation and neuronal stress. This opens new avenues for investigating TE-targeted therapies in neurodegenerative diseases.
• Resource Availability: The authors provide open-source code, simulated benchmarking datasets, and pre-processed count matrices, facilitating immediate adoption by the community.