Comprehensive detection of genetic and epigenetic alterations in cancer using long reads with TumorLens

TumorLens is a unified long-read sequencing framework that simultaneously detects diverse genetic and epigenetic alterations, including SNVs, structural variants, copy-number changes, and methylation patterns, to enable comprehensive tumor profiling and advance precision oncology.

The Gist

Imagine your body is a massive, bustling city. The DNA inside your cells is the city's master blueprint, containing all the instructions for how the city should run. In a healthy city, the blueprint is perfect, and the construction crews (your cells) follow the plans exactly.

But in cancer, the blueprint gets corrupted. Some pages get torn out, others are duplicated, some words are misspelled, and the ink colors (chemical tags) change, confusing the construction crews. This leads to a chaotic, runaway construction site that becomes a tumor.

For a long time, scientists trying to read these corrupted blueprints had a major problem: they were using short-read sequencing. Think of this like trying to understand a complex novel by reading it one word at a time, but the words are cut off after just a few letters. You can see the spelling mistakes (single letter changes), but you miss the big picture: where entire paragraphs were deleted, where chapters were swapped around, or where the ink color changed to hide a secret message.

Enter TumorLens.

What is TumorLens?

TumorLens is like a high-tech, all-seeing microscope that reads the entire blueprint in one go, word-for-word, without cutting it up. It uses a technology called Long-Read Sequencing (specifically from Oxford Nanopore).

Instead of reading a few words at a time, TumorLens reads long, continuous sentences. This allows it to see:

1. The Typos: Small spelling mistakes (mutations).
2. The Missing Pages: Large chunks of DNA that are gone.
3. The Duplicated Chapters: Sections of the blueprint that are copied over and over.
4. The Hidden Ink: Chemical tags (methylation) that act like highlighters or erasers, turning genes on or off without changing the letters themselves.

The "Tumor Purity" Problem

One of the biggest headaches in cancer research is that a tumor sample isn't 100% cancer. It's a messy mix of cancer cells and healthy cells (like a bowl of fruit salad where some apples are rotten and some are fresh).

Old tools often got confused by this mix. If you have a 50/50 mix of healthy and cancer cells, a tool might think a "missing page" is just a printing error because half the sample still has the page.

TumorLens is special because it has a built-in "Purity Calculator." It knows exactly how much of the sample is cancer and how much is healthy. It can mathematically "filter out" the healthy noise to see the cancer's true blueprint, even if the sample is only 50% cancerous. It's like having a detective who can look at a muddy crime scene and perfectly reconstruct the original footprint, even if half the print was covered in mud.

The "Immune Shield" (HLA)

Your immune system is the city's police force. To catch criminals (cancer cells), the police need to see the criminals' "wanted posters" (proteins called HLA). Cancer cells are sneaky; they often try to rip up their own wanted posters or paint over them so the police can't see them. This is called immune escape.

Because the "wanted poster" genes (the HLA locus) are incredibly complex and repetitive, short-read tools often couldn't read them at all. TumorLens, with its long reads, can navigate this complex maze. It can tell if a cancer cell has:

• Lost a poster: (Deleted the gene).
• Copied a fake poster: (Duplicated a gene to confuse the police).
• Painted over the poster: (Changed the chemical tags to hide the gene).

Real-World Results

The researchers tested TumorLens on real cancer samples (lung, ovarian, and stomach cancers) and found things previous tools missed:

• The "Silent" Switch: In some lung cancers, they found that the "wanted posters" were still there, but the chemical tags had turned them off (hypermethylation). The police couldn't see them, even though the blueprint was intact.
• The "Double Trouble": In some stomach cancers, they saw that the cancer cells had deleted the "interferon" genes (which usually sound the alarm for the immune system) and simultaneously duplicated the genes that help hide from the police. It was a coordinated, two-pronged attack.
• No Match Needed: Usually, you need a "healthy" sample from the same patient to compare against the tumor. TumorLens is so smart it can often figure out what's wrong even if you only have the tumor sample, by comparing it to a massive database of healthy people.

Why Does This Matter?

Think of cancer treatment like fixing a broken machine. If you only look at the broken gears (mutations), you might miss the fact that the wiring (epigenetics) is also shorted out.

TumorLens gives doctors a complete, 3D map of the tumor's chaos.

• It helps explain why a patient isn't responding to immunotherapy (maybe their "wanted posters" are hidden).
• It helps find new targets for drugs (maybe we can turn the "hidden" genes back on).
• It works fast (about 6 hours from sample to answer), which could eventually help doctors make life-saving decisions while a patient is still in the hospital.

In short, TumorLens is the ultimate detective tool that finally lets us read the whole story of a cancer's evolution, not just the highlights, helping us understand how the disease hides and how we can finally catch it.

Technical Summary

1. Problem Statement

Current cancer genomics relies heavily on short-read sequencing, which creates significant blind spots:

• Incomplete Variant Detection: Short-read methods prioritize Single Nucleotide Variants (SNVs) and Copy Number Variants (CNVs) but often miss complex Structural Variants (SVs), large CNVs, and Loss of Heterozygosity (LoH), particularly in repetitive regions.
• Epigenetic Blindness: Standard DNA sequencing does not capture methylation states without separate bisulfite conversion, preventing the simultaneous analysis of genetic and epigenetic drivers.
• HLA Complexity: The Human Leukocyte Antigen (HLA) locus is highly polymorphic and structurally complex. Short reads struggle to resolve haplotypes, allele-specific methylation, and somatic LoH in this region, which is critical for understanding immune escape.
• Tumor Purity Ignorance: Existing long-read SV callers (e.g., Sniffles) often fail to account for tumor purity (the proportion of malignant cells), leading to inaccurate variant calling in heterogeneous samples.
• Lack of Unified Framework: There is no existing pipeline that jointly analyzes SNVs, indels, SVs, CNVs, LoH, and DNA methylation from a single long-read assay, especially for tumor-only (no matched normal) scenarios.

2. Methodology: TumorLens Pipeline

TumorLens is a unified computational framework designed for Oxford Nanopore Technologies (ONT) long-read data. It operates on raw FASTQ or BAM files and integrates multiple analysis modules:

• Input & Alignment:

  • Accepts tumor-normal pairs or tumor-only samples.
  • Uses Minimap2 for alignment to GRCh38 (excluding alternative contigs).
  • Extracts methylation signals directly from basecalling (MM/ML tags) without bisulfite conversion.

• Variant Calling Modules:

  • SNVs/Indels: Uses Clair3 (ONT model) with a 10% allele fraction filter.
  • Structural Variants (SVs): Uses Sniffles2 (including mosaic mode for low-frequency variants).
  • CNVs & LoH: Implements a novel adaptation of the Spectre caller. Crucially, this module explicitly models tumor purity (user-provided or estimated) to distinguish somatic events from germline noise and to adjust copy number calculations.
  • Methylation Profiling: Uses modkit to summarize methylation into 10kb windows. It identifies Cancer Differentially Methylated Regions (cDMRs) based on a threshold of ±33 percentage points between tumor and normal.

• Specialized HLA & Antigen Presentation Machinery (APM) Analysis:

  • Targets 74 immune-related genes (including 13 HLA genes and 61 APM genes).
  • Personalized Reference Reconstruction: Uses specHLA to determine HLA types, then re-maps reads to a personalized reference genome containing the specific HLA alleles. This resolves complex haplotypes and enables allele-specific methylation analysis even in LoH regions.

• Tumor-Only Mode:

  • Lacks a matched normal, so it employs strict filtering:
    • SVs: Uses STIX (a long-read annotation tool) to filter out variants present in population databases (e.g., 1000 Genomes), classifying absent variants as "pseudo-somatic."
    • CNVs/LoH: Applies strict size filters (>1Mb) to prioritize large, likely somatic events.
    • Methylation: Uses a "Panel of Normals" (18 healthy blood samples) to establish baseline methylation distributions.

3. Key Contributions

• First Unified Long-Read Framework: TumorLens is the first tool to jointly detect SNVs, indels, SVs, large CNVs, LoH, and CpG methylation in a single assay.
• Purity-Aware Modeling: It introduces the first long-read implementation of somatic CNV/LoH detection that explicitly incorporates tumor purity estimates, significantly improving accuracy in heterogeneous samples.
• Personalized HLA Reconstruction: By building a sample-specific reference, it overcomes the limitations of standard references in the highly polymorphic HLA locus, enabling precise detection of HLA LoH and allele-specific methylation.
• Tumor-Only Capability: It provides a robust strategy for retrospective or clinical "tumor-only" analysis by leveraging population databases and panel-of-normals approaches to filter germline variants.

4. Results & Validation

A. Benchmarking on Cell Lines:

• GIAB HG008 (Pancreatic): Detected 28/37 known somatic CNVs (75.7% recall) and 1,717 SVs.
• NCI-H2009 (Lung): Successfully detected all 20 previously reported LoH regions with breakpoints within 1kb. Correctly identified homozygous HLA alleles (A03:01, B07:02, C*07:02).
• AOCS21 (Ovarian): Identified complex events including a 1.3Mb duplication affecting HLA and Interferon loci with CN=4, validated by coverage and SNV data.

B. Purity Sensitivity:

• In silico titration (90% to 10% tumor purity) showed TumorLens maintains high sensitivity down to 50% purity.
• At 50% purity, it still detected 82.7% of HLA variants and correctly adjusted copy number calls (e.g., distinguishing CN=3 in a 50% purity sample derived from a CN=4 event).

C. Clinical Cohort Analysis (13 Lung Cancer Pairs):

• Identified a "constitutive" hypermethylation signature in 137 regions across all tumors, including WRAP53 and TP53BP1 (DNA repair).
• Case Study (Lung_5): Detected a 41Mb LoH on Chr6 affecting the entire HLA locus and antigen presentation genes (TAP1, TAP2), suggesting immune evasion.
• Case Study (Lung_B): Revealed a complex landscape with whole-chromosome duplications (Chr6) and LoH in antigen processing genes (RFX5, CTSS), alongside CDKN2A hypermethylation.

D. Tumor-Only Analysis (5 Stomach Cancer Samples):

• Successfully identified high-confidence somatic events without matched normals.
• Detected a 928kb deletion in Stomach_2 affecting five Type I Interferon genes.
• Identified "constitutive" hypermethylation in GPR21 across all samples, a gene linked to reduced expression in various tumors.
• HLA typing revealed frequent homozygosity in Class I and II genes, hinting at somatic LoH.

5. Significance

• Holistic Tumor Profiling: TumorLens bridges the gap between genetic and epigenetic analysis, revealing how mutations and methylation co-evolve to drive tumor progression and immune escape (e.g., simultaneous HLA LoH and interferon gene silencing).
• Clinical Translation: The pipeline processes 30x ONT tumor/normal data in ~6 hours, making it viable for rapid, intraoperative, or real-time clinical decision-making.
• Overcoming Clinical Barriers: By enabling accurate analysis of tumor-only samples and low-purity tissues, it unlocks the potential of long-read sequencing for retrospective studies and clinical settings where matched normals are unavailable.
• Immune Escape Mechanisms: The tool provides unprecedented resolution into the HLA locus, offering new insights into how tumors evade T-cell recognition through combined genetic (deletions/LoH) and epigenetic (methylation) mechanisms.

In conclusion, TumorLens establishes a new standard for comprehensive tumor profiling, demonstrating that long-read sequencing, when coupled with advanced purity-aware analytics, can decode the full spectrum of cancer's genomic and epigenomic architecture.