Control‑guided nanopore sequencing to map DUS2‑dependent dihydrouridylation in human lung adenocarcinoma cells.

1. Overview

This project will develop the first transcriptome-wide technology to detect and quantify dihydrouridine (D or DHU) in human mRNA using direct RNA sequencing (DRS) and machine learning. Recent studies show that dihydrouridine synthase (DUS) enzymes are overexpressed in lung adenocarcinoma, the leading cause of cancer-related death worldwide, raising the possibility that aberrant mRNA dihydrouridylation could reprogram translation to promote tumor growth. Yet the presence, distribution, and function of D in mRNA remain completely unexplored. The proposed platform addresses this critical gap by combining CRISPR-engineered A549 DUS-knockout cell lines and synthetic RNAs to train and benchmark an algorithm that distinguishes DHU from uridine (U) and pseudouridine (ψ).

Timeline:

3-4 months: Generate knockout cell lines, sequence KO libraries, identify D sites (Aim 1).
2-3 months: Train and validate the machine learning model using synthetic RNAs (Aim 2).

Data sharing commitment: Processed datasets, trained models, and code will be shared via GEO, SRA, and GitHub. Raw POD5 files will be made available upon request.

Own contributions

Our group has extensive experience in studying RNA modifications across diverse human cell types, combining experimental and computational expertise to profile and quantify chemical RNA marks. We previously developed Mod‑p ID¹, a tool that leverages basecalling errors and in vitro transcribed (IVT) controls² to detect RNA modifications transcriptome‑wide while accounting for SNVs and genetic background effects. Using this approach, we successfully identified and mapped pseudouridine in both primary and immortalized human cells^3-5.

Our group has also established a validated framework for differential modification analysis between conditions, which we have applied in multiple published studies^3,6. Beyond IVT controls, we pioneered control‑based approaches using knockdown of pseudouridine synthases combined with machine learning to identify and characterize writer‑dependent modification sites⁴.

Building on this expertise, we are uniquely positioned to adapt these methods to study dihydrouridine (D) in human lung adenocarcinoma. Our established protocols for direct RNA sequencing, machine‑learning‑based signal analysis, and differential modification testing will ensure the robust development and benchmarking of a new platform to detect D in mRNAs and to assign modification sites to their corresponding writer enzymes.

2. Introduction

Among all types of cancer, lung cancer is the leading cause of cancer-related death worldwide, accounting for approximately 1.8 million deaths annually⁷.
In lung adenocarcinoma (LUAD), particularly non-small cell lung carcinoma (NSCLC), research indicates that a specific family of RNA-modifying enzymes responsible for the deposition of dihydrouridine in the RNA (DUS enzymes) is overexpressed and linked to poorer patient prognosis⁸. Increased DUS expression and subsequent elevated dihydrouridine levels in tRNA are associated with enhanced tumor cell proliferation, metastasis, and overall cancer progression⁸.
Despite the critical role of DHU in lung adenocarcinoma, dihydrouridine is mainly studied in tRNA, leaving its presence and function in mRNA largely unknown.

Emerging evidence suggests that DUS2 may mediate D modifications in mRNA⁹, which could lead to new discoveries about how DUS enzymes regulate LUAD, but current methods struggle to resolve these sites, and no available technology can distinguish D from U and ψ transcriptome-wide. This technological gap has limited studies about dihidrouridine in cancer biology.

We recently observed that D, similarly to the U-to-C mismatch used as a proxy for ψ detection, can be detected by calculating the U-to-V basecalling error in Nanopore direct RNA sequencing (DRS). Building on this finding, we propose to develop a novel DRS-based platform combined with machine learning to enable transcriptome-wide mapping and quantification of D in human mRNAs and lncRNAs. By applying this platform to LUAD A549 cells, we will generate the first map of mRNA dihydrouridylation in human samples and provide mechanistic insights into DUS-mediated modifications in cancer.

3. Study Objectives and Approach

This study will:

1) identify transcriptome-wide D sites in A549 cells using matched in vitro transcribed RNA controls and DUS1L/DUS2 knockout A549 cell lines
We will generate homozygous knockouts of DUS1L and DUS2, two key dihydrouridine synthases, in A549 lung adenocarcinoma cells. Total RNA from WT and KO cell lines will be extracted and matched by preparing an unmodified IVT control. The WT, KO, and IVT will be sequenced using Nanopore DRS to generate poly(A)+ selected RNA (mRNA and lncRNA) transcriptome-wide libraries. We will then apply Mod-p ID to calculate U→V basecalling signatures and compare the error profiles across these three conditions to identify D-sites de novo. The use of the KO libraries will further enable assignment of each D-site to the corresponding writer enzyme, providing mechanistic insight into how specific enzymes drive these modifications and enabling functional studies and potential therapeutic targeting of DUS enzymes in LUAD.

2) train and benchmark machine learning models to distinguish D from U and ψ using synthetic RNAs
After identifying D-sites in step 1, we will design and synthesize RNA controls containing a single site-specific U, D, or ψ site. These will be sequenced using DRS and used to train a supervised machine learning model using our previously developed Mod-Quant tool¹⁰. Model performance will be evaluated using synthetic spike-in standards and benchmarked against knockout datasets from step 1 to ensure specificity and accuracy in distinguishing these modifications.

4. Methods

Participants/Sample

Target sample: A549 wild-type lung adenocarcinoma cells purchased from ATCC will be compared to homozygous DUS1L and DUS2 CRISPR-mediated knockout A549 cell lines. The two knockouts will be commercially generated through a custom order from CD Biosynsis. Once our sequencing and analysis are complete, identified sites can be assessed in other cell lines (i.e., publicly available direct RNA sequencing data sets) and also assigned as D without the need for knockout cell lines.Inclusion/exclusion criteria: The knockout cell lines will be re-assessed with Sanger sequencing in our lab after purchase to confirm the CRISPR-mediated mutation.

Inclusion/exclusion criteria: The knockout cell lines will be re-assessed with Sanger sequencing in our lab after purchase to confirm the CRISPR-mediated mutation.

Sample size: We will generate one replicate per condition. The use of a single replicate is justified by the use of the new ONT Nanopore RNA004 PromethION flow cells, which generate ~30 million reads per flow cell. In-silico replicates will be produced if needed using computational approaches.

Materials and Procedures

Experimental procedures: the whole RNA extraction and sequencing process has been extensively applied in our lab, and all the steps of the protocol needed to replicate the whole process can be found in the following STAR protocol.

Total RNA will be extracted from A549 WT, DUS2-KO, and DUS1L-KO cells using the TRIzol protocol.
Poly(A) selection using the NEBNext Poly(A) mRNA Magnetic Isolation Module will be performed on the total RNA.
Direct RNA sequencing will be performed using the ONT PromethION P2 Solo device with RNA004 kits and PromethION flow cells.
Synthetic RNAs containing site-specific U, D, or ψ will be synthesized and sequenced with RNA004 PromethION flow cells for machine learning model training.

Data processing:

Raw reads will be basecalled using Dorado basecalling provided by ONT, and aligned to GRCh38 using minimap2.
WT and KO libraries will be analyzed against the IVT control using Mod-p ID, a tool previously developed by our lab for pseudouridine detection. The tool will be tailored for DHU detection by calculating U-to-V basecalling errors, which we assessed as a stronger signature of D-sites than the original U-to-C mismatch used in pseudouridine detection. Mod-p ID will compare the direct libraries to the IVT transcriptome-wide, calculating if the direct sample is statistically different from the unmodified IVT for all the U-sites.
Synthetic strands carrying a U, D, and ψ will be basecalled and aligned as direct raw reads and will be used to extract both ionic current signatures and basecalling features that will be used to train the ModQuant ML tool. The ML model will then be applied to the native RNA (both WT and KO libraries) to quantify the occupancy of D-sites and compare it to the U→V mismatch levels assessed with Mod-p ID. This comparison will highlight how the U→V basecalling error, used as a proxy for D-site detection, relates to the occupancy.

Planned Analyses

Preprocessing: Basecalling, alignment, Mod-p ID U→V detection, feature extraction of ionic current and dwell time signals, read-level correction for quality.

Statystica analyses:

The detection of D-sites with Mod-p ID will be followed by a filtering of significant sites (p-value<0.001) and a differential analysis that will retain only sites with sufficient coverage and basecalling errors (#reads>20, U-to-V mismatch >10) that have a significant decrease in U→V mismatch in the KO sample compared to the WT (ΔU→V_(WT-KO)> 10). The rationale of this differential approach, described in Fanari et al.⁴, is that the absence of the dihydrouridine enzyme in the knockout library will lower the modification rates, introducing a decrease in the U-to-V basecalling error in the KO library vs the WT.
The D-sites will be assigned to a specific modifying enzyme after assessing which KO library between DUS2-KO and DUS1L-KO had a decrease in U→V compared to the WT. This will allow us to provide mechanistic insights into how specific enzymes drive these modifications and enable functional studies and potential therapeutic targeting of DUS enzymes.
ModQuant ML model training with cross-validation for each synthetic strand generated from the native data.

Ethics and Data Management

Ethics: This project exclusively uses immortalized human cell lines (A549 wild-type and CRISPR-engineered knockouts) obtained from commercial sources (ATCC, CD Biosynsis). As such, it does not constitute human subjects research, and Institutional Review Board (IRB) approval is not required. No identifiable human data will be collected, and no patient-derived materials will be used, ensuring no confidentiality risks.

Data sharing: Processed datasets, code, and trained models will be deposited in GEO/SRA and GitHub upon publication. Raw sequencing pod5 data will be shared upon request.
We commit to providing regular updates on ResearchHub and to uploading a preprint version of these findings to ResearchHub and bioRxiv when results are available.

5. Pilot Data

Pilot observation: We identified a reproducible U-to-V basecalling error in D-containing RNAs in nanopore direct RNA sequencing, supporting the feasibility of detecting D.

Preliminary model performance: Synthetic RNAs demonstrate that Mod-Quant can distinguish pseudouridine from canonical uridine based on ionic current features, justifying its use for D, another uridine modification.

6. Budget

Sequencing: PromethION flow cells ($3,980.00)
Sequencing kit for RNA004 ($690)
Cell lines: DUS2-KO A549 cell line ($3,720)
DUS1L-KO A549 cell line ($3,655)
Total: $12,045

7. References

S. Tavakoli, et al., Semi-quantitative detection of pseudouridine modifications and type I/II hypermodifications in human mRNAs using direct long-read sequencing. Nat. Commun. 14, 334 (2023).
McCormick, C. A. et al. Multicellular, IVT-derived, unmodified human transcriptome for nanopore-direct RNA analysis. GigaByte 2024, gigabyte129 (2024).
Fanari, O. et al. Pseudouridine reprogramming in the human T cell epitranscriptome: from primary to immortalized states. RNA (2025) doi:10.1261/rna.080633.125.
Fanari, O. et al. Probing enzyme-dependent pseudouridylation using direct RNA sequencing to assess epitranscriptome plasticity in a neuronal cell line. Cell Syst. 101238 (2025).
C. A. McCormick, et al., mRNA psi profiling using nanopore DRS reveals cell type-specific pseudouridylation. bioRxiv (2024).
Fanari, O., Meseonznik, M., Bloch, D. & Rouhanifard, S. H. Protocol for differential analysis of pseudouridine modifications using nanopore DRS and unmodified transcriptome control. STAR Protoc 6, 103948 (2025).
Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 71, 209–249 (2021).
Kato, T. et al. A novel human tRNA-dihydrouridine synthase involved in pulmonary carcinogenesis. Cancer Res 65, 5638–5646 (2005).
Finet, O., Yague-Sanz, C., Marchand, F. & Hermand, D. The Dihydrouridine landscape from tRNA to mRNA: a perspective on synthesis, structural impact and function. RNA Biol 19, 735–750 (2022).
A. Makhamreh, et al., Nanopore signal deviations from pseudouridine modifications in RNA are sequence-specific: quantification requires dedicated synthetic controls. Scientific Reports 14, 1–13 (2024).

Control‑guided nanopore sequencing to map DUS2‑dependent dihydrouridylation in human lung adenocarcinoma cells.

Control‑guided nanopore sequencing to map DUS2‑dependent dihydrouridylation in human lung adenocarcinoma cells.

Funding Progress

Peer Reviews

Related Funding

Request for Proposals - Cancer Biology Research Project

Nonprofit Recipient

Topics

DOI