Repeats Mimic Pathogen-Associated Patterns Across a Vast Evolutionary Landscape

Petr Šulc, Andrea Di Gioacchino, Alexander Solovyov, Sajid A. Marhon, Siyu Sun, Håvard T. Lindholm, Raymond Chen, Amir Hosseini, Hua Jiang, Bao-Han Ly, Parinaz Mehdipour, Omar Abdel‐Wahab, Nicolas Vabret, John LaCava, Daniel D. De Carvalho, Rémi Monasson, Simona Cocco, Benjamin D. GreenbaumbioRxiv 2021 · DOI

TL;DR

This paper introduces a novel theoretical framework to quantify "viral mimicry" – repetitive elements in our genome that resemble pathogen patterns – and how natural selection has shaped these features. They demonstrate that this mimicry is a widespread evolutionary strategy, likely serving as a genome-wide quality control system against transcriptional errors.

What Problem Does This Paper Address?

Our genomes are filled with repetitive elements, often remnants of ancient viral infections. While usually silenced, these repeats can become active and mimic patterns found on pathogens (Pathogen-Associated Molecular Patterns or PAMPs). This "viral mimicry" can trigger our innate immune system, contributing to various diseases like cancer and autoimmune disorders. However, a significant gap exists in understanding how this viral mimicry evolves, how widespread it is across genomes, and what selective pressures maintain it. Quantifying these PAMP-like features and their evolutionary dynamics is crucial for understanding both genome evolution and disease.

Methods

The authors employ a Maximum Entropy (MaxEnt) framework, a powerful tool from statistical physics, to quantify "selective forces" on specific sequence features. These forces represent deviations from what's expected by random chance, indicating enrichment or depletion of a pattern. They specifically calculate:

  • CpG force ($x_{CpG}$): The selective pressure on CpG dinucleotides, which are known PAMPs recognized by innate immune receptors like TLR9 and ZAP.
  • Double-stranded RNA (dsRNA) force ($x_{ds}$): A novel application of their framework to quantify the tendency of RNA sequences to form double-stranded structures, another PAMP recognized by receptors like MDA5 and TLR3.

They also utilize a Kimura-based model to simulate neutral evolution and compare it with observed patterns, helping to distinguish selection from random drift. For dsRNA, they developed a "direct approach" for intuitive calculation and an "equivalent direct approach" for computational efficiency, both yielding analogous results to the MaxEnt method.

Datasets

  • Human Genome (hg38): Used extensively for analyzing repeat families, CpG content, and dsRNA formation.
  • Viral Genomes: Used as reference for comparing nucleotide biases.
  • DFAM Database: Provided annotations for repeat families.
  • RepeatMasker Database: Used for repeat annotations.
  • Published Datasets:
    • MDA5 receptor ligand datasets (for validating dsRNA predictions).
  • J2 monoclonal antibody binding data (for validating dsRNA predictions).
  • RNA-Seq data from splice inhibitor experiments (H3B-8800 treated vs. DMSO treated SF3B1-K700 mutant cells).
  • Newly Generated Data:
    • RIP-seq data for L1 ORF1p and ORF2p, and ZCCHC3, to investigate interactions with high CpG L1 RNA.
  • J2 antibody immunoprecipitation RNA-seq data.

Key Insights and Ideas

  • "Viral mimicry" is a conserved evolutionary strategy: Repetitive elements have evolved to mimic pathogen patterns, suggesting a functional advantage.
  • Selection actively maintains PAMPs: Certain repeats, like HSATII and LINE-1, show evidence of selection to retain CpG motifs, and inverted Alu repeats are selected to form dsRNA.
  • dsRNA formation is linked to repeat structure: Inverted repeats, particularly AluS, are major sources of self-RNA that can form dsRNA structures recognized by innate immune sensors.
  • Genome-wide quality control: The authors propose that viral mimicry acts as a general mechanism for genomes to co-opt pathogen-associated features, serving as a quality control system against transcriptional dysregulation and loss of epigenetic control.
  • Computational framework is powerful: The MaxEnt approach, extended to dsRNA, provides a robust way to quantify these evolutionary forces across diverse genomic elements and species.

Key Results

  • Quantified viral mimicry landscape: The study provides the first quantification of viral mimicry across multiple species, revealing its breadth and conservation.
  • HSATII and LINE-1 under selection for CpG: These repeats are actively maintained to retain CpG motifs, suggesting a role in innate immune sensing.
  • Alu repeats are key dsRNA sources: Inverted Alu repeats, especially AluS, are identified as major contributors to dsRNA formation, validated by experimental data.
  • Predictive power: The framework successfully predicts dsRNA-forming regions and their association with MDA5 ligands, validated by experimental assays.
  • Cross-species conservation: High $x_{CpG}$ and $x_{ds}$ regions show conservation across primates, indicating selective pressure.

Limitations and Future Work

  • Complexity of dsRNA recognition: While the framework predicts dsRNA formation, it doesn't fully capture the nuances of all PRR recognition (e.g., triphosphate modifications for RIG-I).
  • Distinguishing selection vs. byproduct: While the study strongly supports selection, the authors acknowledge that in some cases, high RNA concentration from dysregulation might non-specifically engage PAMPs, making it challenging to definitively separate selection from a byproduct.
  • Future directions: The authors suggest extending the framework to more complex patterns and machine learning approaches. They also highlight the potential for therapeutic applications by manipulating these repeat-mediated immune responses and the possibility of discovering new "self-agonist" patterns within the non-coding genome.

Summary updated

Abstract

ABSTRACT An emerging hallmark across human diseases – such as cancer, autoimmune and neurodegenerative disorders – is the aberrant transcription of typically silenced repetitive elements. Once active, a subset of repeats may be capable of “viral mimicry”: the display of pathogen-associated molecular patterns (PAMPs) that can, in principle, bind pattern recognition receptors (PRRs) of the innate immune system and trigger inflammation. Yet how to quantify the landscape of viral mimicry and how it is shaped by natural selection remains a critical gap in our understanding of both genome evolution and the immunological basis of disease. We propose a theoretical framework to quantify selective forces on virus-like features as the entropic cost a sequence pays to hold a non-self PAMP and show our approach can predict classes of viral-mimicry within the human genome and across eukaryotes. We quantify the breadth and conservation of viral mimicry across multiple species for the first time and integrate selective forces into predictive evolutionary models. We show HSATII and intact LINE-1 (L1) are under selection to maintain CpG motifs, and specific Alu families likewise maintain the proximal presence of inverted copies to form double-stranded RNA (dsRNA). We validate our approach by predicting high CpG L1 ligands of L1 proteins and the innate receptor ZCCHC3 , and dsRNA present both intracellularly and as MDA5 ligands. We conclude viral mimicry is a general evolutionary mechanism whereby genomes co-opt pathogen-associated features generated by prone repetitive sequences, likely offering an advantage as a quality control system against transcriptional dysregulation.
BibTeX
@misc{sulc2021repeats,
  author = {Petr Šulc and Andrea Di Gioacchino and Alexander Solovyov and Sajid A. Marhon and Siyu Sun and Håvard T. Lindholm and Raymond Chen and Amir Hosseini and Hua Jiang and Bao-Han Ly and Parinaz Mehdipour and Omar Abdel‐Wahab and Nicolas Vabret and John LaCava and Daniel D. De Carvalho and Rémi Monasson and Simona Cocco and Benjamin D. Greenbaum},
  title = {{Repeats Mimic Pathogen-Associated Patterns Across a Vast Evolutionary Landscape}},
  year = {2021},
  doi = {10.1101/2021.11.04.467016},
  howpublished = {bioRxiv},
}
Figure