Precision single-cell profiling of Circulating Tumour Cells: novel markers and data-driven characterization by CTCeek

This study analyzes 3,302 single-cell transcriptomes to distinguish true circulating tumor cells (CTCs) from blood contaminants, identifies novel markers for epithelial and mesenchymal CTC subtypes, and introduces CTCeek, a web-based tool for the automated annotation of bona fide CTCs.

The Gist

Imagine your body is a bustling city, and cancer is a group of rebels trying to escape the main district (the tumor) to start new colonies in other parts of the city (metastasis). These rebels are called Circulating Tumour Cells (CTCs). They are the "spies" of the cancer world, traveling through the bloodstream (the city's highways) to find new places to invade.

For years, scientists have tried to catch these spies to study them, hoping to understand how the cancer is evolving and to stop it. But they've been using a very specific, outdated "net" to catch them.

The Problem: The Wrong Net

The current standard method for catching these cancer spies relies on a specific ID card called EPCAM. Think of EPCAM as a "Red Badge" that most cancer cells wear. Scientists use antibodies (like a magnet) to grab anything with a Red Badge.

Here's the catch:

1. The Disguise: Some cancer spies are smart. They take off their Red Badges (a process called EMT) to blend in with the crowd. The old net misses them completely.
2. The Impostors: The net is so sticky that it also grabs innocent bystanders—like blood cells, platelets, and endothelial cells—who happen to be near the cancer. It's like a security guard grabbing everyone wearing a red hat, even if they are just delivery drivers or tourists, not spies.

The result? Scientists have been studying a mix of real spies and innocent civilians, leading to confusing data.

The Solution: A New Detective Agency (CTCeek)

The authors of this paper, led by Professor Stefano Volinia, decided to clean up the mess. They acted like a massive data detective agency.

Step 1: The Great Sorting
They gathered over 3,300 single-cell profiles from public databases. Imagine they had a giant pile of photos of people found in the bloodstream, labeled "Cancer Spy."

• They looked at the DNA blueprints (specifically, looking for "copy number variations," which are like typos or extra pages in a book that only cancer cells have).
• They realized that about half of the people labeled as "spies" were actually just innocent blood cells (diploid, with no typos).
• They threw out the impostors and kept only the 900 genuine spies (the "bona fide" CTCs).

Step 2: Finding New ID Cards
Once they had a clean list of real spies, they asked: "What do these guys actually look like?"
They discovered that the "Red Badge" (EPCAM) isn't the only thing that identifies them. They found new ID cards that the spies wear:

• For the "Regular" Spies (Epithelial): They found badges like CLDN4, CLDN7, and TACSTD2. These are like new, more reliable security tags.
• For the "Disguised" Spies (Mesenchymal): These are the ones who took off their Red Badges. The old net missed them. The new study found they wear badges like AXL, CAV1, and PODXL. Now, scientists can catch the spies in disguise!
• The Universal Badge: They found one badge, TM4SF1, that almost every type of cancer spy wears, regardless of their disguise.

Step 3: The Platelet Bodyguards
They also noticed something interesting about platelets (the blood cells that clot wounds).

• The "Regular" spies often travel with a squad of platelet bodyguards, using them as shields.
• The "Disguised" spies and the "Super Active" spies (those dividing rapidly) travel alone. They don't need the bodyguards. This tells us that different types of cancer cells have different survival strategies.

The Tool: CTCeek

Finally, the team built a free, online tool called CTCeek.

• Think of it as a "Spotify for Cancer Cells."
• If a researcher has a new sample of blood cells and wants to know, "Are these real cancer spies or just innocent bystanders?", they upload the data to CTCeek.
• The tool compares the new data against their massive, clean reference library.
• It instantly says: "This cell is a real spy," or "This is just a platelet," or "This is a disguised spy."

Why This Matters

This study is a game-changer because:

1. It cleans the data: It stops scientists from studying fake cancer cells.
2. It catches the escapees: It gives us new ways to catch the cancer cells that have changed their appearance (EMT), which are often the most dangerous ones.
3. It levels the playing field: By providing a free tool, any scientist in the world can now accurately identify these cells, leading to better early detection and more effective treatments for cancer patients.

In short, the authors didn't just find new ways to catch the bad guys; they built a better police force and a new database to ensure no criminal slips through the cracks.

Technical Summary

1. Problem Statement

Circulating Tumor Cells (CTCs) are critical for liquid biopsy and monitoring cancer evolution, metastasis, and treatment resistance. However, current CTC research faces three major limitations:

• Reliance on EPCAM: Most isolation methods rely on the Epithelial Cell Adhesion Molecule (EPCAM). This misses EPCAM-negative CTCs, particularly those undergoing Epithelial-to-Mesenchymal Transition (EMT) or originating from EPCAM-low tumors.
• High Contamination Rates: Existing purification platforms (especially label-free or microfluidic methods) co-enrich non-tumor blood cells (hematopoietic, endothelial, platelets). Publicly available single-cell RNA sequencing (scRNA-seq) datasets labeled as "CTCs" often contain significant fractions of these contaminants, leading to inaccurate biological conclusions.
• Lack of Computational Tools: There is no standardized, automated tool to distinguish "bona fide" (true) CTCs from blood-borne contaminants within scRNA-seq datasets.

2. Methodology

The authors employed a large-scale, integrative bioinformatics approach to re-analyze public data and develop a new classification tool.

• Data Collection & Integration:
  • Aggregated 3,302 putative CTC scRNA-seq profiles from 27 public datasets (GEO/SRA), primarily from breast cancer but also including prostate, pancreatic, lung, gastric, and colorectal cancers.
  • Integrated these with 10,434 Peripheral Blood Mononuclear Cell (PBMC) profiles to serve as a reference for non-tumor blood cells.
• Pre-processing & Quality Control:
  • Removed cell cycle genes prior to integration to prevent clustering based on mitotic activity rather than cell type.
  • Used Scanorama for dataset integration and Leiden algorithm for clustering.
• Bona Fide CTC Identification Strategy:
  • Marker Scoring: Calculated scores for epithelial markers (e.g., Keratins) vs. hematopoietic markers (e.g., CD45/PTPRC).
  • Copy Number Variation (CNV) Inference: Used CopyKAT and SCEVAN to identify aneuploid cells. The study posits that aneuploidy is a definitive marker of cancer origin, whereas diploid cells in CTC fractions are likely contaminants.
  • Cell Cycle Analysis: Confirmed that true CTCs were actively engaged in the cell cycle, unlike most contaminant blood cells.
• Marker Discovery:
  • Performed differential expression analysis (Wilcoxon test, ROC AUC) comparing pure CTCs vs. PBMCs.
  • Analyzed markers at both single-cell and pseudo-bulk levels to ensure robustness.
• Tool Development:
  • Developed CTCeek, a web-based application using the Shiny framework. It uses a reference mapping algorithm (anchorage) to project query cells onto the established PBMC/CTC reference atlas for automatic annotation.

3. Key Contributions & Results

A. Characterization of CTC Heterogeneity

The study successfully filtered out contaminants, resulting in a high-confidence set of 900 bona fide CTCs. These were classified into distinct subpopulations:

1. Epithelial A: High EPCAM, associated with platelets.
2. Epithelial B: Distinct from Epithelial A, expressing specific transcription factors (SP6, LHX1) and membrane proteins (KCNK15, LY6K).
3. Mesenchymal: Vimentin-positive, EPCAM-negative, expressing EMT genes (ZEB1/2, SNAI1/2).

B. Novel Biomarker Discovery

The authors identified a comprehensive panel of markers superior to or complementary to EPCAM:

• Pan-CTC Markers (Universal):
  • TM4SF1: Expressed in all CTC subclasses (Epithelial A, B, and Mesenchymal) but absent in PBMCs.
  • TACSTD2 (TROP2), CLDN4, CLDN7, SDC4: High AUC values, effective for epithelial CTCs.
• Mesenchymal-Specific Markers:
  • PODXL, AXL, CAV1, TGM2: Highly specific to mesenchymal CTCs. These are crucial for detecting EPCAM-negative cells undergoing EMT.
  • Note: AXL is also expressed in rare dendritic cells (ASDCs), requiring negative selection strategies (e.g., targeting TYROBP or HLA-DRB1) to avoid false positives.
• Contaminant Markers (for Depletion):
  • Identified markers for non-CTC contaminants that are CD45-negative (e.g., SELP, TMEM40, CLEC1B, PEAR1, GP9, ITGA2B). These allow for the negative selection of platelets and endothelial cells without relying on CD45.

C. Biological Insights

• Platelet Association: Platelets are physically associated with Epithelial A CTCs but are notably absent in Epithelial B and Mesenchymal CTCs, suggesting different mechanisms of immune evasion or survival.
• Aneuploidy as a Gold Standard: The study confirms that aneuploidy is a near-perfect discriminator between true CTCs and diploid blood contaminants in scRNA-seq data.
• Tumor Origin Specificity: Markers like KLK2/KLK3 and AR were confirmed as specific to prostate CTCs, validating the method's ability to distinguish tissue of origin.

D. The CTCeek Tool

• Function: An automated web tool that annotates scRNA-seq profiles as CTCs or specific blood cell types.
• Performance:
  • Specificity: >99.9% (tested against large PBMC datasets).
  • False Positive Rate: <0.04%.
  • Validation: Successfully identified bona fide CTCs in independent validation datasets (GSE255889, GSE295441), outperforming or matching aneuploidy detection tools like CopyKAT and SCEVAN.

4. Significance

• Paradigm Shift: Moves CTC research beyond the EPCAM-centric view, enabling the capture of the "missing" mesenchymal and EPCAM-low CTC populations that drive metastasis.
• Improved Isolation Strategies: Provides a molecular blueprint for designing next-generation isolation platforms using multi-target antibody panels (e.g., combining TM4SF1, AXL, and depletion of SELP/ITGB4) to achieve unbiased CTC enrichment.
• Standardization: CTCeek provides the first computational standard for validating CTC datasets, ensuring that future single-cell studies are based on "pure" tumor cells rather than contaminated mixtures.
• Therapeutic Relevance: Identifies targets like TACSTD2 (already used in ADC therapy) and AXL for potential therapeutic intervention in metastatic disease.

In summary, this paper provides a rigorous computational framework to clean existing CTC data, discovers a robust panel of novel markers for comprehensive CTC capture, and releases a public tool (CTCeek) to standardize the field of liquid biopsy analysis.