DBiT-seq_FFPE

This repository aims to share the raw data processing and visualization codes used in DBiT-seq_FFPE project.

This repository includes the main R scripts used for the visualization of the sequencing data, including clustering, Differential expression gene analysis and cell annotation with reference scRNA-seq data.

One MATLAB script to automatic identify pixels on tissue was also included.

Refer to our lastest preprint paper for more details: https://www.biorxiv.org/content/10.1101/2020.10.13.338475v1.abstract

Abstract

DBiT-seq was a spatial multiomic sequencing technique that can also be directly applied for FFPE samples. We have successfully demonstrated spatial transcriptome sequencing of embryonic and adult mouse FFPE tissue sections at cellular level (25μm pixel size) with high coverage (>1,000 genes per pixel). Spatial transcriptome of a E10.5 mouse embryo identified all major anatomical features in the brain and abdominal region.

Data processing

1. Raw Fastq data processing using ST pipeline and generate expression matrix

We did the illumina pair-end 100 sequencing using Hiseq 4000 and pool two samples (tissue sections) for each sequencing lane.

The DBiT-seq Raw fastq file

Read 1: Contains the cDNA sequence

Read 2: Contains the spatial Barcode A, Barcode B and UMIs

Reformat Fastq Read 2 file

To run ST pipeline, the Read 2 sequence needs to be reformated, see following figure. Due to different experimental design, the Read 2 of DBiT-seq is equal to the "Read 1" in ST pipeline, while Read 1 will be the "Read 2".

To reformat the Raw data, run the fastq_process.py in Rawdata_processing folder and gzip the resulted fastq file to save space:

python fastq_process.py
gzip sample_R2_processed.fastq

The reformated data was processed following ST pipeline.

Run ST pipeline

Run st_pipeline.sh to start the ST pipeline: The input is processed_R2.fastq.gz and Raw R1.fastq.gz. It also requires a "spatial_barcodes_index.txt" to decode the spatial location information. Genome references and annotatation files were aslo needed.

#!/bin/bash

# FASTQ reads
FW=PATH_TO_PROCESSED_R2/sample_R2_processed.fastq.gz
RV=PATH_TO_R1/R1.fastq.gz

# References for mapping and annotation 
MAP=PATH_TO_ALIGNMENT_REF/Dropseq_Alignment_References/mm10/
ANN=PATH_TO_ALIGNMENT_REF_GTF/Dropseq_Alignment_References/mm10/mm10.gtf

# Barcodes settings
ID=PATH_TO_BARCODE_INDEX/spatial_barcodes_index.txt 

# Output folder and experiment name
OUTPUT=PATH_TO_OUTPUT/st_pipeline_new/
mkdir -p PATH_TO_OUTPUT/st_pipeline_new/

TMP=PATH_TO_TEMP/st_pipeline_new/tmp
mkdir -p PATH_TO_TEMP/st_pipeline_new/tmp

# Do not add / or \ to the experiment name
EXP=FFPE-2

# Running the pipeline
st_pipeline_run.py \
  --output-folder $OUTPUT \
  --ids $ID \
  --ref-map $MAP \
  --ref-annotation $ANN \
  --expName $EXP \
  --htseq-no-ambiguous \
  --verbose \
  --log-file $OUTPUT/${EXP}_log.txt \
  --allowed-kmer 5 \
  --mapping-threads 20 \
  --temp-folder $TMP \
  --no-clean-up \
  --umi-start-position 16 \
  --umi-end-position 26 \
  --overhang 0 \
  --min-length-qual-trimming 10 \
  $FW $RV

Convert Ensemble to Gene Names

Then, Run converttoname.sh to annotate the resulting FFPE2_stdata.tsv.

#!/bin/bash

tsv_E=FFPE-2_stdata.tsv
path_to_annotation_file=PATH_TO_ALIGNEMNT/Dropseq_Alignment_References/mm10/mm10.gtf
convertEnsemblToNames.py $tsv_E --annotation $path_to_annotation_file --output FFPE-2_exp_matrix.tsv

Now, the expression matrix is successfully generated. The row names are "XxY" location for each pixel, and columne names are Genes.

2. Identify useful pixels (pixel on tissue) from microscope image using Matlab

Useful pixels were generated from the Matlab script. Basically, it divides the real tissue microscope image into 50x50 small sqaures which match with DBiT-seq pixels. The intensity inside each pixel was calculated and only pixels have signals above a threashold will be selected.

There two steps: To run the Matlab script "Pixel_identification.m"

1. Use Photoshop or other photo editing software to crop the microscope image into exactly the size of the DBiT-seq covering area. For example, the upperleft of the image should be the 1x1 pixel of DBiT-seq, and the lowerright is the 50x50. No space is allowed. See "FFPE-2.jpg" for example.

2. Use threashold function under Image->adjustment menu to adjust the image, so that your tissue is black and background is compeletely white.
3. Invert the color of the image. The final image is like "FFPE-2_BW.jpg" in the Example_Data folder.

4. Run the matlab script and a "postion.txt" file will be generated, which contains only the useful pixels.

Data visualization

The data visualization was completed with R language. The package used extensively the functions in Seurat V3.0 and ggplot2.

Common data visualization scripts include:

1. Total_transcripts and Gene_count.R

This R script generates the Filtered_matrix.tsv (expression matrix with useful pixels only), makes the count per pixel distribution plot (UMI.pdf and Gene.pdf) and the spatial heatmap of genes and UMIs(UMI_heatmap.pdf and Gene_heatmap.pdf). See below: See below:

UMI.pdf

Gene.pdf

UMI_heatmap.pdf

Gene_heatmap.pdf

2. Clustering.R

This R script clusters the pixels using Seurat Package, with SCTransform performed for all pixels. It will generate the spatial cluster map (clusters.pdf), and also the UMAP plot (on the Rstudio UI). See examples below:

clusters.pdf

UMAP_clusters.pdf

3. Intergration with scRNA-seq data

Contact

For questions, you can contact Yang Liu ([email protected])