LATTE for locus-specific quantification of transposable element expression across species

The paper introduces LATTE, a high-accuracy computational framework for locus-specific quantification of transposable element (TE) expression across species, which reveals that TEs possess a distinct regulatory landscape decoupled from host genes and contribute significantly to the genetic architecture of complex traits, including Sjogren's syndrome.

The Gist

The Big Picture: Finding a Needle in a Haystack (That Looks Like a Thousand Other Needles)

Imagine your body's genetic code (DNA) is a massive library. Most of the books in this library are unique stories that tell your cells how to build you. But there's a huge section of the library filled with Transposable Elements (TEs).

Think of TEs as photocopiers that got loose. They are ancient viral sequences that can copy themselves and paste themselves into new spots in the library. Because they copy themselves so much, there are thousands of copies of the same "story" scattered all over the library.

The Problem:
Scientists want to know which specific copy of a story is being read (expressed) right now to understand how your body works or why you get sick. But because there are so many identical copies, it's like trying to figure out which specific photocopier in a room of 1,000 identical machines just printed a page. If you just look at the page, you can't tell which machine made it.

Existing tools were like guessing games. They would either ignore the confusing pages or just pick one machine at random, leading to messy, inaccurate data.

The Solution: Enter "LATTE"

The authors of this paper built a new tool called LATTE (Locus-specific quantification of Transposable Element expression).

Think of LATTE as a super-smart detective or a high-tech barcode scanner. Instead of just guessing which machine printed the page, LATTE looks for tiny, unique imperfections or specific patterns in the "ink" (the DNA sequence) to pinpoint exactly which copy of the story is being read.

How does it work?

1. The Multi-Clue Strategy: If a page could have been printed by Machine A, B, or C, LATTE doesn't just guess. It looks at three clues:

   • The Family: Which family of machines does it belong to?
   • The Wear and Tear: Is the page slightly more worn out in the middle (like a specific part of the machine)?
   • The Location: Where in the library was the machine sitting?
     By combining these clues, it calculates the probability of which machine actually did the work.

2. Filtering the Noise: Sometimes, a page looks like it came from a machine in your library, but it was actually printed by a virus from outside the library (like an exogenous virus). LATTE has a special "spam filter" (using machine learning) to spot these fakeouts and ignore them, so you only count the real internal activity.

What Did They Discover?

Once they had this super-accurate tool, they looked at data from humans, cows, and chickens. Here is what they found, using more analogies:

1. TEs are their own bosses, not just sidekicks.
For a long time, scientists thought TEs were just "passengers" riding along when the main genes (the host) were being read.

• The Discovery: It turns out TEs have their own remote controls. Just because the main gene is loud and active doesn't mean the TE inside it is. They are often regulated independently. It's like a radio station (the gene) playing music, but the TE is a separate DJ in the booth who decides whether to play a remix or stay silent, regardless of what the main station is doing.

2. TEs are the "Special Effects" of the genome.
The researchers found that TEs are responsible for a unique layer of complexity.

• The Discovery: When they looked at complex traits (like height in humans or milk speed in cows), they found that TEs explained about 8.7% of the differences that the main genes couldn't explain.
• The Analogy: If genes are the script of a movie, TEs are the special effects, the lighting, and the sound design. You can have the same script, but different special effects make the movie feel completely different.

3. The "Splicing" Battle (The Sjögren's Syndrome Example).
The most fascinating finding was at a specific location in the human genome related to an autoimmune disease called Sjögren's syndrome.

• The Scenario: There is a genetic risk factor (a typo in the code) that acts like a traffic cop.
• The Twist: This traffic cop tells the cell to build the "Main Character" (the gene IRF5) but stop building the "Special Edition" version that includes the TE.
• The Result: The disease happens because the cell is forced to choose between the "Standard Version" and the "TE Version." The risk factor pushes the cell to make too much of the Standard Version and suppress the TE version, throwing the whole system out of balance. It's a tug-of-war where the genetic typo tips the scale.

Why Does This Matter?

Before LATTE, scientists were trying to study these "photocopiers" with a blurry map. They missed a lot of the action.

• For Medicine: This helps us understand diseases better. Some diseases aren't caused by the main genes breaking, but by the "photocopiers" (TEs) getting out of control or being silenced when they should be active.
• For Farming: Since they tested this on cows and chickens too, farmers can use this to breed animals that are healthier or produce more milk/meat by understanding these hidden genetic levers.
• For Evolution: It shows that these "jumping genes" aren't just junk; they are active, independent players in the story of life, shaping how species evolve and adapt.

In short: LATTE is the high-definition camera that finally lets us see exactly what these chaotic, jumping genetic elements are doing, revealing that they are the secret architects of many complex traits and diseases.

Technical Summary

1. Problem Statement

Transposable elements (TEs) constitute a significant portion of eukaryotic genomes (e.g., ~49% in humans) and are pivotal drivers of evolution and phenotypic diversity. However, quantifying their expression at a locus-specific resolution remains a major bottleneck in functional genomics due to three primary challenges:

• Multi-mapping Ambiguity: High sequence homology among TE families and subfamilies causes short-read RNA-seq reads to map to multiple genomic locations, making it difficult to assign reads to specific loci.
• Annotation and Boundary Issues: Reads spanning exon-TE junctions or intergenic regions often suffer from incomplete or inaccurate genomic coordinates, leading to incorrect identification of TE exonization boundaries.
• Interference and Bias: Sequencing errors, alignment imperfections, and exogenous viral interference (particularly with Endogenous Retroviruses, ERVs) introduce substantial bias in expression estimates.
• Tool Limitations: Existing tools (e.g., SQuIRE, TEtranscripts, SalmonTE) often lack read-specific resolution, rely on single indicators for read assignment, or are restricted to model organisms with pre-defined annotation files.

2. Methodology: The LATTE Framework

The authors developed LATTE (Locus-specific quantification of Transposable element expression), a computational framework designed to overcome these limitations.

• Core Algorithm (Multi-Indicator EM):

  • LATTE employs an innovative Expectation-Maximization (EM) algorithm that utilizes a tiered, multi-indicator strategy to reassign multi-mapped reads.
  • Three Indicators: The algorithm iteratively assigns reads based on:
    1. Subfamily Identity: Aggregating uniquely mapped reads to estimate subfamily abundance.
    2. Base Coverage Distribution: Using the depth and coverage of uniquely mapped reads within TE consensus sequences to calculate probabilities via definite integration.
    3. Genomic Annotation Region: Assigning reads to specific annotated loci based on unique mapping evidence.
  • The process is iterative: if a read's assignment probability exceeds a threshold, it is assigned; otherwise, it proceeds to the next indicator or a new cycle.

• Denoising and Filtering Modules:

  • Long-Read Integration: To mitigate homology bias, LATTE integrates long-read RNA-seq data (from 29 samples across 5 livestock/poultry species) to curate a high-confidence library of transcriptionally active TE subfamilies. This filters out silent genomic copies, reducing the search space for multi-mapping.
  • Viral Interference Detection: A machine learning module using DBSCAN (Density-Based Spatial Clustering of Applications with Noise) identifies anomalous ERV expression. It calculates the ratio of expression level to genomic insertion length; outliers (likely caused by exogenous viral cross-mapping) are flagged and removed.

• Implementation:

  • Written in Shell and Python, LATTE is a lightweight, UNIX-based tool requiring no formal installation.
  • Inputs: Coordinate-sorted BAM files and TE annotation files (GFF format).
  • Outputs: A comprehensive record of TE-derived reads with annotations and a matrix of TPM/read counts for locus-specific TEs.

3. Key Contributions

• Locus-Specific Resolution: LATTE is the first tool to provide exact coordinates for exon-TE junctions and assign multi-mapped reads to specific genomic loci with high precision, a capability previously unattainable by existing tools.
• Cross-Species Applicability: Unlike tools restricted to model organisms, LATTE supports customized GFF annotations, enabling robust TE quantification in non-model species (e.g., cattle, chickens, pigs, goats, sheep).
• Integrated Denoising: The framework uniquely combines long-read data filtering and machine learning-based outlier detection to address sequence homology and exogenous viral interference simultaneously.
• Scalability: The tool processes ~1 million TE-derived reads in ~10 minutes per CPU core, making it suitable for large-scale population studies.

4. Key Results

• Benchmarking Performance:

  • Against simulated and real-world human RNA-seq data, LATTE significantly outperformed state-of-the-art tools (SQuIRE, TEtranscripts, SalmonTE).
  • Accuracy: Achieved an Intra-class Correlation Coefficient (ICC) of 0.998 at the subfamily level and 0.839 at the locus-specific level, surpassing all other methods.
  • Recall: Maintained 100% recall for subfamily capture, whereas other tools missed up to 45% of subfamilies due to annotation limitations.

• Biological Insights (Tissue Specificity):

  • Analysis of 250 RNA-seq samples across five species revealed that 74.6% of locus-specific TE loci are expressed in only one tissue type, exhibiting significantly higher tissue specificity than host genes (where only 13.3% of TEs were ubiquitous vs. 55.3% of genes).
  • Decoupled Regulation: TE expression showed weak correlation ($R^2 < 0.23$) with host gene expression, indicating that TEs are regulated by independent cis-regulatory mechanisms rather than passive co-transcription.

• Genetic Architecture (TE-eQTL):

  • Mapping cis-TE-eQTLs across humans, cattle, and chickens revealed that only ~22% of TE-eQTLs colocalize with host gene eQTLs after accounting for linkage disequilibrium.
  • TE-eQTLs were significantly enriched at splice donor sites, suggesting a functional link between genetic variation, TE expression, and alternative splicing.

• Complex Trait Associations (TWAS):

  • Transcriptome-wide association studies (TWAS) on 3,746 traits showed that TEs contribute 204 (8.7%) exclusive associations with complex traits that are not captured by gene-eQTLs.
  • Case Study (Sjögren's Syndrome): The variant rs10954213 (associated with Sjögren's syndrome) acts as a TE-eQTL for the MSTB1 TE within the IRF5 locus. The risk allele upregulates IRF5 canonical transcripts while downregulating IRF5 transcripts containing the MSTB1 exon, demonstrating an antagonistic regulatory relationship mediated by alternative splicing.

5. Significance

• Redefining TE Biology: The study establishes TEs not merely as "junk DNA" or passive passengers, but as active, independently regulated components of the genetic architecture underlying complex phenotypes.
• Methodological Advancement: LATTE provides a standardized, high-resolution pipeline for TE analysis, bridging the gap between repetitive element biology and complex trait genetics.
• Clinical and Agricultural Impact: By identifying TE-specific regulatory variants and their impact on splicing and disease (e.g., SLE, Huntington's), LATTE opens new avenues for understanding the genetic basis of diseases and improving breeding strategies in livestock.
• FarmGTEx Integration: As a core tool for the FarmGTEx project, LATTE enables the systematic exploration of TE regulation across diverse domesticated species, unlocking the "dark matter" of the genome.