DOTSeq enables genome-wide detection of differential ORF usage

The paper introduces DOTSeq, a comprehensive statistical framework that enables genome-wide detection of differential ORF usage and translation efficiency in both bulk and single-cell ribosome profiling experiments, overcoming the limitations of gene-level analysis to uncover translational control events at the ORF level.

The Gist

Imagine your DNA is a massive library of instruction manuals (genes) for building the body's machines (proteins). For a long time, scientists thought they could just read the whole manual to understand how much of a machine was being built. They assumed that if the manual was read more often, more machines were made.

But scientists recently discovered that these manuals are tricky. Inside one big manual, there are often smaller, hidden notes (called ORFs or "Open Reading Frames") that act like traffic lights or volume knobs. Sometimes, the cell reads the main instruction (the "main ORF") to build the machine. Other times, it gets distracted by the hidden notes (like "upstream ORFs" or uORFs), which can tell the factory to slow down, stop, or change how the machine is built.

The problem? The old tools scientists used to study protein production were like a blurry camera. They could see the whole library getting busy, but they couldn't see which specific notes inside the manual were being read. They missed the traffic lights.

Enter DOTSeq. Think of DOTSeq as a high-definition, super-smart microscope that can zoom in on every single note in the instruction manual.

Here is how it works, broken down with simple analogies:

1. The Two Main Superpowers

DOTSeq has two different "modes" to catch different types of changes:

• Mode A: The "Traffic Light" Detector (DOU)

  • The Analogy: Imagine a busy intersection. Sometimes, the traffic light turns red for the main road (the main protein) and green for a side street (a small note). The total number of cars (RNA) might stay the same, but where they go changes.
  • What it does: This mode, called Differential ORF Usage (DOU), looks at the ratio. It asks: "Is the cell spending more time reading the hidden notes and less time on the main instruction?" It's perfect for spotting when a cell switches strategies, like during cell division, even if the total amount of "reading" hasn't changed much.
  • Why it's special: Old tools would say, "Nothing changed here!" because the total traffic is the same. DOTSeq says, "Wait! The traffic light just flipped!"

• Mode B: The "Volume Knob" Detector (DTE)

  • The Analogy: Imagine someone just turns the volume up on the radio. The song is the same, but it's much louder.
  • What it does: This mode, called Differential Translation Efficiency (DTE), looks for cases where the cell is just reading the entire manual much faster or slower than usual. It measures how efficiently the ribosome (the factory worker) is turning the instructions into products.

2. The Real-World Test: The Cell Cycle

The researchers tested DOTSeq on cells going through their daily routine (the cell cycle). They looked at cells in three states: resting, dividing, and preparing to divide.

• What they found: They discovered that when cells prepare to divide, they start reading the "hidden notes" (uORFs) much more often. It's like the cell is hitting the "pause" button on building new proteins to make sure everything is ready for the big split.
• The "Aha!" Moment: Old tools missed this completely. They only saw that the total protein production went down. DOTSeq revealed why: the cell was actively switching its focus to the hidden notes to control the process. It's like realizing the factory didn't just slow down; it changed its assembly line entirely.

3. Zooming In: Single-Cell Vision

Most tools look at a "smoothie" made of millions of cells mixed together. You can taste the fruit, but you can't see the individual seeds.
DOTSeq can also look at single cells. It can take a snapshot of one tiny cell and say, "This specific cell is reading the hidden notes, while its neighbor is not." This helps scientists understand how individual cells make different decisions, which is crucial for understanding diseases like cancer.

4. Why DOTSeq is the New Champion

The authors ran a "race" against other popular tools.

• The Race: They created fake data with known changes to see who could find them best.
• The Result: DOTSeq's "Traffic Light" mode (DOU) was the fastest and most accurate at finding the subtle shifts in how cells read their manuals. It rarely made false alarms (saying something changed when it didn't) and caught changes that other tools completely missed.

The Bottom Line

DOTSeq is like upgrading from a black-and-white TV to a 4K HDR screen.

Before, scientists could see that the "protein factory" was busy or quiet. Now, with DOTSeq, they can see exactly which instructions are being followed, how the cell is prioritizing different parts of the manual, and how it uses hidden switches to control life's most complex processes. This helps us understand how cells divide, how they react to stress, and potentially how to fix them when they go wrong in diseases.

Technical Summary

1. Problem Statement

Current differential translation analysis methods primarily operate at the gene level, assuming uniform changes in translation across all Open Reading Frames (ORFs) within a transcript. This approach fails to resolve cis-regulatory events that occur between different ORFs (e.g., upstream ORFs/uORFs, downstream ORFs/dORFs, and main ORFs/mORFs) within the same gene.

• Limitation: Gene-level methods cannot detect shifts in the relative usage of specific ORFs (e.g., a uORF becoming more active while the mORF is repressed), which is a critical mechanism for translational control (e.g., uORF-mediated repression of ATF4).
• Need: A statistical framework capable of detecting Differential ORF Usage (DOU) at genome-wide scale in bulk Ribosome Profiling (Ribo-seq) data, while also providing ORF-aware read summarization for single-cell Ribo-seq (scRibo-seq).

2. Methodology: The DOTSeq Framework

DOTSeq is a Bioconductor-based statistical framework designed to analyze Ribo-seq and matched RNA-seq data. It offers two primary statistical modules and a unified workflow for data processing.

A. Data Preprocessing & Annotation

• Input: Preprocessed BAM files from adapter-trimmed and splice-aware aligned Ribo-seq and RNA-seq reads (e.g., via STAR or HISAT2).
• ORF Annotation: DOTSeq generates discrete, non-overlapping ORF bins by parsing reference GTF and transcript FASTA files. It offers two paths:
  1. Bioconductor Path: Uses DOTSeq::getORFs to generate GRanges objects.
  2. External Path: Uses orf2gtf.py (RIBOSS engine) to flatten annotations, compatible with tools like featureCounts.
• Read Summarization:
  • Bulk: countReads generates ORF-level count matrices.
  • Single-cell: countReadsSingleCell aggregates counts into sparse matrices suitable for Seurat objects.

B. Statistical Modules

DOTSeq employs two complementary modules to model translational regulation:

1. Differential ORF Usage (DOU) Module:

   • Goal: Detects changes in the relative contribution of an ORF to a gene's total translation output.
   • Model: Uses a Beta-Binomial Generalized Linear Model (GLM) implemented via glmmTMB.
   • Mechanism: Models the expected proportion of Ribo-seq counts for a specific ORF relative to the total counts of other ORFs in the same gene.
   • Key Feature: Includes an interaction term (condition:strategy) to isolate translation-specific effects. It tests if the ratio of Ribo-seq to RNA-seq reads for an ORF changes differently between conditions compared to other ORFs in the same gene.
   • Dispersion: Models dispersion separately for Ribo-seq and RNA-seq to account for distinct noise sources.
   • Output: Estimates marginal means (EMMs) and applies adaptive shrinkage (via ashr) to effect sizes.

2. Differential Translation Efficiency (DTE) Module:

   • Goal: Detects monotonic changes in ribosome loading relative to RNA abundance (standard TE analysis) at the ORF level.
   • Model: Uses a Negative Binomial GLM implemented via DESeq2.
   • Mechanism: Models raw ORF-level counts directly, testing the interaction term (condition:strategy) to see if the Ribo-seq/RNA-seq ratio changes.
   • Use Case: Effective for single-ORF genes or ORFs with monotonic changes that DOU might not capture.

C. Single-Cell Analysis

• DOTSeq enables dimensionality reduction and clustering on scRibo-seq data using ORF-level counts.
• It models the per-cell proportion of uORF footprints ($u/(u+m)$) using a Beta-Binomial GLMM (Generalized Linear Mixed Model) with batch effects as random intercepts.

3. Key Contributions

• Novel Statistical Framework: Introduces the first genome-wide method specifically designed to detect Differential ORF Usage (DOU), moving beyond gene-level assumptions.
• Beta-Binomial GLM for Proportions: Adapts the beta-binomial distribution to model the proportion of reads mapping to specific ORFs within a gene, allowing for the detection of competitive translation dynamics (e.g., uORF vs. mORF).
• Single-Cell Integration: Provides the first ORF-aware read summarization and statistical modeling pipeline for scRibo-seq, enabling the study of translational heterogeneity at single-cell resolution.
• Comprehensive Workflow: Delivers an end-to-end solution including annotation, read summarization, contrast estimation, and visualization.

4. Results

The authors validated DOTSeq using HeLa cell cycle datasets (bulk and single-cell) and extensive simulations.

• Biological Discovery (Bulk Data):

  • Applied to synchronized HeLa cells (mitotic cycling, mitotic arrest, interphase), DOU revealed significant shifts in uORF usage that were missed by DTE.
  • Key Finding: A global shift toward increased uORF usage during mitotic cycling, leading to repression of mORFs for key regulators like RPTOR (mTOR pathway), MAPK6, and CSDE1.
  • Specific Example: In CSDE1, uORFs were highly utilized during mitosis, while the mORF was preferred during interphase, suggesting cell-cycle-dependent translational control.
  • Comparison: DOU detected 764 differentially translated ORFs missed by DTE, while DTE detected 2,754 unique to it. The overlap was partial (639 shared), confirming the modules provide complementary insights.

• Biological Discovery (Single-Cell Data):

  • Analysis of hTERT-RPE-1 cells showed increased per-cell uORF occupancy in G0 and mitosis compared to interphase.
  • Joint UMAP analysis (mORF + uORF) revealed clearer treatment enrichment and cell-state separation than single-modality maps.

• Benchmarking & Performance:

  • Simulation: DOTSeq (DOU module) was benchmarked against anota2seq, deltaTE, RiboDiff, Riborex, and Xtail.
  • Sensitivity: DOU achieved superior sensitivity across various effect sizes.
  • Error Control: DOU was the only method to maintain a near-nominal False Discovery Rate (FDR) of 0.1 across all settings. Other methods (including DTE) showed p-value inflation (U-shaped distributions) when applied to DOU-type regulatory events.
  • Computation: While DOU is computationally heavier (~7 mins for 35k ORFs on a single thread) due to complex GLM fitting, it supports parallel processing via OpenMP.

5. Significance

• Uncovering Hidden Regulation: DOTSeq reveals a layer of translational control (cis-regulatory ORF switching) that is invisible to standard gene-level analyses. This is crucial for understanding how cells regulate protein synthesis during stress, development, and disease (e.g., cancer).
• Methodological Rigor: By using a beta-binomial framework, DOTSeq correctly models the compositional nature of ORF usage data, avoiding the statistical pitfalls (inflated false positives) seen when applying standard count-based methods (like DESeq2) to relative usage problems.
• Scalability: The framework bridges bulk and single-cell technologies, allowing researchers to dissect translational heterogeneity in complex tissues and dynamic processes like the cell cycle.
• Resource: The availability of DOTSeq as a Bioconductor package with reproducible workflows lowers the barrier for the community to perform high-resolution translational profiling.