Genetic demultiplexing and transcript start site identification from nanopore sequencing of 10x Genomics multiome libraries

This study demonstrates that nanopore sequencing of 10x Genomics multiome libraries enables full-length transcript profiling and accurate genetic demultiplexing, successfully recovering a substantial portion of transcription start sites compared to traditional short-read 5' assays.

The Gist

Imagine you are trying to understand a massive library of books (our cells), but you only have a special scanner that can read the very last page of every book. This is how most modern genetic studies work. They use a technology called Illumina sequencing to read the "3' end" (the back) of RNA molecules. While this tells us which books are on the shelves (which genes are active), it doesn't tell us where the story begins (the Transcription Start Site, or TSS). It's like knowing a movie is playing, but not knowing the opening scene.

This paper introduces a new way to read the entire book from cover to cover using a different scanner called Nanopore sequencing. Here is the story of what they found, explained simply:

1. The Experiment: Mixing the Pools

The researchers took muscle tissue from 52 different people. To save time and money, they mixed samples from 20 people together into one "soup" of cells before putting them into the machine. This is called multiplexing.

• The Challenge: When you mix 20 people's cells together, how do you know which cell belongs to which person later? Usually, scientists look for tiny genetic differences (like a unique fingerprint) in the DNA to sort them out.
• The Fear: Nanopore sequencing is great because it reads long strings of text, but it makes more "typos" (errors) than the standard Illumina scanner. The team wondered: If the scanner makes too many typos, will we lose the fingerprints and fail to sort the people?
• The Result: Surprisingly, no! Even with the "typos," the new scanner was able to sort the cells back into their correct owners almost as well as the old scanner. It was like trying to recognize a friend's handwriting even if they were writing with a shaky hand; you could still tell who wrote it.

2. The Big Win: Seeing the "Start" of the Story

The main superpower of this new method is that it reads the full length of the RNA molecule.

• The Old Way: You only see the end of the sentence. You know the topic, but not the beginning.
• The New Way: You see the whole sentence, from the very first word to the last.

This allows the researchers to find the Transcription Start Site (TSS). Think of a gene as a long novel. The TSS is the very first word on page 1. Knowing where the story starts is crucial because it tells you how the gene is being controlled. Sometimes, a gene can start at different pages to create different versions of the story (isoforms), and this new method helps spot those differences.

3. The "Clean-Up" Crew (SCAFE)

Reading the whole book isn't perfect. The machine sometimes picks up "noise" or artifacts—like seeing a word that looks like the start of a sentence but is actually just a glitch in the scanner.

The team used a smart software tool called SCAFE (Single Cell Analysis of Five-prime Ends) to act as an editor.

• The Problem: The raw data had some "fake" starts.
• The Fix: They developed a special "pre-processing" step (a cleaning routine) to remove the noise before running SCAFE.
• The Result: After cleaning, the "starts" they found using the new Nanopore scanner matched up very well with the "starts" found using a specialized, expensive 5' gene test (the gold standard). They managed to catch about 63% of the starts that the gold standard found.

4. The Takeaway

This paper is a proof-of-concept that says: "You don't need two different expensive machines to get two different types of information."

• Before: If you wanted to know which genes were active AND where they started, you had to run two separate, expensive experiments (one for the back of the book, one for the front).
• Now: You can run one experiment using the Nanopore scanner on the standard "back-of-the-book" library, and you get both pieces of information.

Summary Analogy

Imagine you are a detective trying to solve a crime in a city with 52 suspects.

1. Old Method: You have a camera that only takes photos of the suspects' shoes. You can identify them by their shoes, but you can't see their faces or where they came from.
2. New Method: You get a drone that can take a full-body photo.
   • Can you still identify them? Yes! Even though the drone's camera is a bit grainy (noisy), you can still match the shoes to the faces and sort the suspects correctly.
   • What else do you learn? You can now see their faces (gene expression) and exactly where they were standing when the crime started (Transcription Start Site).

The researchers showed that this "grainy drone" is good enough to do the job, saving time and money while unlocking a whole new layer of biological detail.

Technical Summary

1. Problem Statement

Single-cell and single-nucleus multi-omics (specifically 10x Genomics Multiome) typically rely on short-read Illumina sequencing. This workflow has two primary limitations:

• Loss of 5' Information: Standard 3' Gene Expression (3' GEX) libraries capture only the 3' end of transcripts, losing critical information regarding Transcription Start Sites (TSS) and isoform diversity.
• Inefficiency of Separate Assays: Capturing TSS information usually requires a separate 5' GEX assay, increasing cost and sample consumption.
• Nanopore Challenges: While Oxford Nanopore Technologies (ONT) offers long-read sequencing capable of capturing full-length transcripts, its higher error rate raises concerns about the reliability of genetic demultiplexing (assigning cells to donors based on SNPs) and the accurate identification of TSS without introducing artifacts.

2. Methodology

The authors developed a workflow to integrate ONT long-read sequencing with standard 10x Multiome libraries.

• Experimental Design:
  • Sample Source: 60 frozen vastus lateralis skeletal muscle biopsies from 52 unique donors.
  • Pooling Strategy: Samples were pooled into three batches (20 donors each) to create multiplexed libraries.
  • Library Generation:
    • Multiome: Generated 3' GEX and ATAC libraries. The intermediate full-length cDNA from the 3' GEX preparation was sequenced on ONT (ONT GEX).
    • Controls: Parallel 5' GEX libraries (short-read Illumina) and standard 3' GEX/ATAC (Illumina) were generated for comparison.
• Data Processing Pipeline:
  • Standard Processing: Used STARsolo, CellBender (ambient RNA removal), and standard 10x pipelines for Illumina data.
  • ONT Specific Processing:
    • Custom NextFlow pipeline for barcode/UMI correction and alignment.
    • Genetic Demultiplexing: Used demuxlet to assign nuclei to donors using SNPs from ONT GEX, 3' GEX, and ATAC reads.
  • TSS Identification (SCAFE):
    • Utilized the SCAFE (Single Cell Analysis of Five-prime Ends) tool to identify Transcribed Cis-Regulatory Elements (tCREs).
    • Optimized Preprocessing: Developed a specific filtering step for ONT BAM files prior to SCAFE to remove artifacts. This included:
      • Filtering reads lacking the expected residual TSO sequence (GGG) at the 5' end.
      • Removing excessive soft-clipping.
      • Resolving discrepancies in PCR duplicates where apparent 5' ends differed.
      • Trimming reads to <100bp to avoid SCAFE read-length errors.

3. Key Contributions

• Validation of ONT Demultiplexing: Demonstrated that despite higher error rates, ONT sequencing of 3' GEX libraries allows for successful genetic demultiplexing with high concordance to short-read data.
• TSS Recovery from 3' Libraries: Proved that full-length cDNA from 3' GEX libraries, when sequenced on ONT, can recover TSS information comparable to dedicated 5' GEX assays.
• Optimized Preprocessing Protocol: Introduced a critical preprocessing workflow for ONT reads to eliminate false-positive tCREs caused by template-switch oligo (TSO) artifacts and alignment errors, significantly improving data quality.
• Open Source Tools: Provided scripts for ONT preprocessing and a custom NextFlow pipeline for analysis.

4. Key Results

• Demultiplexing Concordance:
  • There was 92% concordance in donor assignments and doublet calls between ONT GEX and standard 3' GEX (Illumina).
  • While the number of nuclei identified per donor was slightly lower for ONT, the correlation was strong, confirming the feasibility of multiplexing ONT data.
• Transcriptomic Concordance:
  • Pseudobulk gene expression showed high correlation between ONT GEX and Illumina 3' GEX.
  • Clustering (UMAP) and cell type identification were highly consistent between ONT and Illumina datasets.
• TSS/tCRE Identification:
  • Recovery Rate: ONT GEX captured a median of 63% of the tCREs detected by the dedicated 5' GEX assay.
  • Artifact Reduction: The optimized preprocessing (filtering TSO artifacts and duplicate discrepancies) significantly reduced false-positive tCREs in quiescent chromatin regions and gene bodies, aligning ONT results more closely with 5' GEX and ChromHMM active states.
  • Specific Loci: Successful identification of promoter regions for key muscle genes (e.g., MYL2, VGLL2) and the CA3 locus using ONT data.
• Limitations: A substantial number of TSS detected by 5' GEX were missed by ONT, likely due to differences in transcript capture efficiency between 3' and 5' library kits or library preparation imperfections.

5. Significance

• Cost and Sample Efficiency: This approach allows researchers to obtain full-length transcript information (including TSS and isoforms) and perform genetic demultiplexing from a single 3' Multiome library, eliminating the need for a separate 5' assay.
• Scalability: It validates the use of ONT sequencing for large-scale, multiplexed single-nucleus studies, enabling higher loading concentrations and reduced batch effects through genetic pooling.
• Methodological Advancement: The study provides a robust, reproducible pipeline for handling the specific error profiles of ONT data in single-cell contexts, bridging the gap between long-read capabilities and single-cell resolution.
• Biological Insight: By recovering TSS data from standard 3' libraries, researchers can study transcriptional regulation and isoform usage in tissues where 5' assays are not feasible or too costly.