PERREO: An integrated pipeline for repetitive elements analysis enables the repeatome expression profiling in cancer

The paper introduces PERREO, a comprehensive and user-friendly pipeline that overcomes the limitations of standard RNA-seq tools by enabling sensitive, repeat-aware analysis of repetitive element expression across short- and long-read data to facilitate the study of the repeatome's role in cancer and other diseases.

The Gist

Imagine your body's cells are like a bustling, high-tech city. Inside every cell, there is a massive library of instructions (DNA) that tells the city how to run. For decades, scientists have been reading these instructions, but they've mostly focused on the "main streets"—the clear, well-labeled genes that make proteins.

However, about half of this library is filled with "junk" or "repetitive" sections: pages of the same sentence repeated thousands of times, or ancient, forgotten manuals that look like gibberish. Scientists used to ignore these sections, thinking they were just background noise or errors in the library.

Enter PERREO: The New Librarian

This paper introduces PERREO, a new, super-smart tool designed to finally read and understand these "repetitive" sections of the library.

Think of standard computer programs for reading DNA like a librarian who only knows how to read the main streets. If a book has a page with the same sentence repeated 50 times, the librarian gets confused, throws the book away, or says, "I can't read this." This means scientists were missing out on a huge part of the story, especially in diseases like cancer.

PERREO is like a specialized librarian who loves the repetitive sections. It doesn't get confused by the repetition; instead, it counts them, organizes them, and figures out what they mean.

How PERREO Works (The Analogy)

1. The Problem: Cancer is like a city in chaos. The "repetitive" parts of the DNA library, which are usually quiet and asleep, start waking up and shouting. In healthy cells, these sections are silent. In cancer cells, they go wild. But because they are repetitive, old tools couldn't hear them clearly.
2. The Solution: PERREO is a "one-stop-shop" pipeline. You feed it raw data (like a stack of unorganized books), and it does everything automatically:
   • Cleaning: It washes the books (quality control).
   • Reading: It reads the repetitive parts carefully, even if the same sentence appears in multiple places (handling "multi-mapping" reads).
   • Comparing: It compares the "shouting" cancer library against the "quiet" healthy library to see what changed.
   • Predicting: It uses math to guess if a patient has cancer just by looking at these repetitive shouts.

What They Discovered (The Story)

The researchers tested PERREO on many different "cities" (samples) to see if it worked better than the old tools.

• The Blood Test (Liquid Biopsy): They looked at blood plasma from patients with esophageal cancer. Even though the "noise" in the blood is usually very faint, PERREO found specific repetitive signals that were louder in cancer patients than in healthy people. It's like hearing a specific siren in a noisy crowd that tells you exactly where the emergency is.
• The Brain Tumor (Glioblastoma): They compared two different maps of the human library. One was the old, slightly blurry map (GRCh38), and the other was a brand-new, ultra-high-definition map (T2T-CHM13).
  • When using the old map, the librarian got confused by the repetitive sections and thought there were more changes than there really were.
  • When using the new map, the librarian could see exactly where the repetitive sections belonged. This made the results much clearer and more accurate, helping to distinguish between aggressive and less aggressive tumors.
• The Cell Lines: They looked at cancer cells grown in a lab. PERREO found that different types of cancer "shout" different repetitive songs. Some cancers shout loudly with one type of repeat, while others use a different tune. This could help doctors identify exactly what kind of cancer a patient has.

Why This Matters

Before PERREO, studying these repetitive parts was like trying to solve a puzzle while wearing blindfolds. You needed to be a computer expert just to try.

PERREO takes off the blindfolds.

• It's User-Friendly: You don't need to be a coding wizard to use it. It's designed so that biologists and doctors can just click a button and get results.
• It's Fast and Accurate: It works faster than other tools and finds more clues.
• It's Future-Proof: As we get better maps of the human genome, PERREO can just swap in the new map and keep working.

The Big Picture:
This paper suggests that these "repetitive" sections aren't just junk. They are like the city's alarm system. When the alarm goes off (deregulation), it tells us the city is under attack (cancer). By using PERREO to listen to these alarms, doctors might soon be able to detect cancer earlier, predict how aggressive it will be, and even track if a treatment is working, all by looking at these previously ignored parts of our genetic code.

Technical Summary

1. Problem Statement

Repetitive RNA elements (repRNAs), including transposable elements (TEs), are critical drivers of genomic instability, gene regulation, and disease progression, particularly in cancer. However, their analysis is severely hindered by current bioinformatic limitations:

• Standard Pipeline Bias: Most RNA-seq pipelines are optimized for annotated protein-coding genes and systematically discard or undercount repetitive sequences due to high rates of "multi-mapping" (reads aligning to multiple genomic loci).
• Fragmented Tooling: Existing tools (e.g., RepEnrich, TEtranscripts, SQuIRE) often lack seamless integration across different sequencing technologies (short-read vs. long-read), fail to handle heterogeneous data types, and do not integrate downstream predictive modeling or co-expression network analysis.
• Accessibility Barrier: The lack of user-friendly, automated pipelines forces researchers to rely on complex, custom scripting, limiting the exploration of the "repeatome" in clinical and biological contexts.

2. Methodology: The PERREO Pipeline

The authors present PERREO, a comprehensive, modular, and containerized (Conda) pipeline designed to automate the analysis of repetitive RNA from raw sequencing data.

Key Technical Features:

• Multi-Platform Compatibility: PERREO supports three distinct analysis modes:
  • SR-PE/SR-SE: Short-read (Illumina) paired-end and single-end data using the STAR aligner.
  • LR: Long-read (Oxford Nanopore) direct RNA data using minimap2 with settings optimized for spliced alignment.
• Repeat-Aware Alignment & Quantification:
  • Unlike standard pipelines that discard multi-mapping reads, PERREO retains them.
  • It uses featureCounts with a fractional assignment strategy ($1/n$), distributing counts across all genomic loci a read aligns to. This ensures that repetitive families are quantified accurately without losing signal.
  • It is organism-agnostic, accepting user-supplied reference genomes (e.g., GRCh38, T2T-CHM13) and annotations (RepeatMasker GTF), allowing application to humans, mice, zebrafish, and non-model organisms.
• Integrated Downstream Analysis:
  • Differential Expression (DEA): Supports edgeR and DESeq2 with automated batch effect correction (e.g., hospital of origin).
  • Transcriptome Assembly: Integrates StringTie2 to discover novel repeat-containing isoforms.
  • Network Analysis: Implements WGCNA for co-expression network inference.
  • Predictive Modeling: Includes Random Forest and GLMnet algorithms to evaluate the diagnostic/prognostic potential of repRNAs.
• User Interface: Offers a command-line interface and a graphical interface for non-programmers, generating publication-ready visualizations (volcano plots, heatmaps, PCA) and QC reports (MultiQC/FastQC).

3. Key Results & Validation

The authors validated PERREO across 330 samples from public repositories (GEO, Singapore Nanopore Expression Project) spanning humans, dogs, and mice.

A. Plasma repRNAs in Esophageal Cancer (ESCA)

• Batch Effect Correction: Initial analysis of 45 plasma samples (ESCA vs. Healthy Controls) showed minimal differential expression. However, after including "hospital of origin" as a batch covariate in PERREO, the sensitivity increased dramatically, identifying 48 upregulated and 1 downregulated repRNAs. This highlights the pipeline's ability to handle technical confounders automatically.
• Biomarker Potential: Specific repeats (e.g., SST1, HSATII) were identified as potential biomarkers.

B. Impact of Reference Genomes (GRCh38 vs. T2T-CHM13) in Glioblastoma (GBM)

• Reduced Ambiguity: Aligning short-read data against the complete T2T-CHM13 assembly reduced multimapping reads by 45% compared to GRCh38 (6.28% vs. 11.30%), proving that improved reference assemblies significantly enhance mapping accuracy for repeats.
• Biological Insight: While T2T-CHM13 detected fewer total Differential Expressed Repetitive elements (DERs) than GRCh38 (due to fewer spurious assignments), it provided clearer separation between GBM, Low-Grade Glioma (LGG), and healthy controls in PCA plots.
• Predictive Performance: Random Forest models achieved high accuracy (0.91) and AUC (0.97) in classifying tumor stages using T2T-aligned data.

C. Extracellular Vesicles (EVs) and Liquid Biopsies

• Analysis of serum-derived EVs from GBM patients revealed distinct repRNA signatures (18 downregulated in GBM).
• GLMnet outperformed Random Forest in this context (AUC 0.80 vs. 0.75), demonstrating the pipeline's flexibility in selecting optimal algorithms for specific data types.
• A specific simple repeat, (TGTTTT)n, was identified as a top feature in both tissue and EV models, suggesting a shared systemic signature.

D. Long-Read Sequencing in Cell Lines

• Applied to Nanopore direct RNA data from 4 cancer cell lines (HCT116, HepG2, MCF7, K562) and H9 stem cells.
• K562 (leukemia) showed the highest mean repRNA expression.
• H9 vs. K562: Identified 122 DERs, with simple repeats being the most altered class. Notably, LINE-1 elements were upregulated in H9 (consistent with hypomethylation in stem cells), while simple repeats were upregulated in cancer lines.
• Shared Cancer Signature: 23 repeat features were consistently differentially expressed across all cancer cell lines compared to the non-cancerous H9 line.

E. Benchmarking

• vs. Salmon: PERREO (with duplicate removal) showed high concordance with Salmon but retained genomic coordinate information lost in pseudoalignment.
• vs. TEtranscripts: PERREO identified 262 DERs vs. 47 by TEtranscripts in the same dataset. PERREO was significantly faster (<4 hours vs. ~27 hours) due to avoiding iterative Expectation-Maximization (EM) steps, while maintaining robust statistical power.

4. Key Contributions

1. Unified Workflow: First pipeline to seamlessly integrate short-read and long-read data analysis for repetitive elements with a single command.
2. Reference Flexibility: Enables immediate adoption of next-generation reference genomes (like T2T) without code modification, directly translating assembly improvements into biological insights.
3. Accessibility: Lowers the bioinformatic barrier by automating complex steps (multi-mapping handling, batch correction, model training) and providing a GUI.
4. Clinical Relevance: Demonstrates that repRNAs are viable biomarkers in liquid biopsies (plasma, EVs) and tissue, capable of stratifying cancer stages and predicting outcomes.

5. Significance

This work fundamentally shifts the paradigm of RNA-seq analysis by treating repetitive elements not as noise to be discarded, but as a rich source of biological and clinical information. PERREO enables researchers to:

• Discover Novel Biomarkers: Identify repeat-derived transcripts for early cancer detection and prognosis.
• Understand Disease Mechanisms: Investigate the role of repeat activation in genomic instability and tumor microenvironment crosstalk.
• Leverage Existing Data: Re-analyze the vast archive of existing RNA-seq data (including short-read datasets) to uncover previously hidden repeatome dynamics.

By removing technical barriers and integrating predictive modeling, PERREO accelerates the translation of repeatome research into clinical oncology and fundamental biology.