 |
 |
GEO help: Mouse over screen elements for information. |
|
| Status |
Public on Aug 02, 2024 |
| Title |
RPM_10 week tumors_rep2 |
| Sample type |
SRA |
| |
|
| Source name |
Prostate
|
| Organism |
Mus musculus |
| Characteristics |
tissue: Prostate
cell line: RPM RR112
cell type: Prostate epithelial
genotype: Rb1/Trp53 KO; cMyc overexpressing (RPM)
|
| Growth protocol |
RPM (Rb1/Trp53 knockout; cMyc-overexpression) mouse prostate organoids were transplanted into immunocompetent C57BL/6 8-10 week old male mice (dorsal prostatic lobe). Primary and metastatic lymph node tumors were isolated from mice between 2-10 weeks post transplantation.
|
| Extracted molecule |
total RNA |
| Extraction protocol |
We generated Visium data from two adjacent sections as technical replicates, with each slide containing RPM tumor tissues from at least two individual mice. Visium Spatial Gene Expression slides were prepared with FFPE sections by the Molecular Cytology Core at MSKCC. Tumor samples with a target RIN value > 0.5 were processed for spatial transcriptomics.
Probe pairs targeting the whole transcriptome were added to slide capture areas and allowed to hybridize overnight at 50°C. Bound pairs were ligated to one another, then released from the tissue by RNase treatment and permeabilization and captured by oligos bound to the slide. Probe extension and library preparation proceeded using the Visium Spatial for FFPE Gene Expression Kit, Mouse Transcriptome (10X Genomics, 1000337) according to the manufacturer’s protocol. After evaluation by real-time PCR, sequencing libraries were prepared with 11-14 cycles of PCR. Indexed libraries were pooled equimolar and sequenced on a NovaSeq 6000 in a PE28/88 run using the NovaSeq 6000 SP Reagent Kit (100 cycles) (Illumina). FASTQ files from sequencing (NovaSeq) were processed via spaceranger count (version 2.0.0) to align reads to the GRCm38 (mm10) reference genome and generate count matrices for subsequent bioinformatic analyses. An average of 74 million paired reads was generated per sample, corresponding to 37,000 reads per spot.
|
| |
|
| Library strategy |
OTHER |
| Library source |
transcriptomic |
| Library selection |
other |
| Instrument model |
Illumina NovaSeq 6000 |
| |
|
| Description |
RR1123 (10 week tumor) RR1132 (10 week tumors)
|
| Data processing |
Cell type deconvolution using BayesPrism: We deconvolved Visium data from two replicates using the snRNA-seq data collected from the 10-week RPM tumor. We used BayesPrism to perform statistical deconvolution as it has been previously shown to yield accurate results for Visium data and outperformed other methods in multiple settings. More importantly, BayesPrism jointly imputes the posterior of cell type fraction and the cell type-specific gene expression profile in each Visium spot, enabling the cell type-specific gene count matrices by PrismSpot, as described below.
To increase signal to noise ratio for deconvolution, we selected marker genes that are differentially upregulated in each of the 19 cell types, which was defined as the 19 phenograph clusters in the snRNA-seq data. We then took the union of these marker genes and deconvolved over these genes. Specifically, we performed a pairwise t-test using the findMarker function from the scran package70, provided in the wrapper function get.exp.stat from the BayesPrism package. As each cell type is compared against every other cell type, we need to summarize N-1 p-values and log2 fold change statistics for each cell type (N denotes number of total cell types). For each non-tumor cell types (4 in total), i.e., mesenchymal, myeloid, normal L/B (luminal/basal) and endothelial cells, we required both the maximum p value to be less than 0.01, and the minimum log2 fold change to be above 0.1, where the max and min were taken over every other cell type. For each tumor cell type (15 in total), i.e., NE, EMT and tumor L/B, we required the same threshold for p-value and log2 fold change, but the max and min were taken over every other non-tumor cell type. As a result, we avoided the comparison between tumor cell types when taking the max and min to retain the maximum number of tumor-specific genes for PrismSpot analysis. This was achieved by defining a coarse level cell type label, where 15 tumor cell types were grouped into a single tumor cell type and using it as the cell.type.labels argument, while setting the original 19 cell types as cell.state.labels in BayesPrism’s built-in function get.exp.stat. To speed up deconvolution we further excluded genes expressed in less than or equal to 3 spots. This yielded 5,125 genes in total used as the input for BayesPrism.
We defined each of the 19 cell states as individual cell types when constructing reference for deconvolution. We set pseudo.min to 0, the numerical lower bound for genes with zero count in a given cell type, to maximize the contrast between different cell types in the reference. All other parameters of BayesPrism were used as default. We performed the updated Gibbs sampling for a more robust correction of batch effects between the snRNA-seq and the FFPE Visium.
Benchmarking of BayesPrism. We benchmarked both the accuracy and robustness of the cell type fraction deconvolved by BayesPrism. A big challenge in the deconvolution field is the lack of ground truth for benchmarking. To estimate some ground truth of the cell type fractions from the Visium data we used 1) histology with H&E, and 2) a marker genes-based approach. The comparison with histologically defined NEPC regions showed strong concordance with the fraction of NEPC cells inferred by BayesPrism. Additionally, we benchmarked the robustness of cell type fraction inferred by BayesPrism by assessing the concordance across 5 histologically defined regions between two technical replicates, which includes 3 neuroendocrine regions and 2 non-neuroendocrine regions. As tissues are slightly shifted between technical replicates, we manually adjusted the selected regions to achieve better spatial concordance on the x-y plane. For each region, we compute the average fractions for each cell type. We then computed both cell type-level and region-level Pearson’s correlation coefficients.
BayesPrism’s deconvolution infers cell type-specific gene expression matrices for each spot. For the Hotspot analysis, we use spot specific tumor gene expression as the input. Specifically, to generate tumor-specific gene expression profiles, we summed up the posterior mean of cell type-specific gene expression, Z, across all tumor clusters (n=15) outputted by BayesPrism. As Hotspot models the raw count data using negative binomial distribution, we rounded up the posterior mean of Z. An important step in Hotspot analysis is to define neighborhood structure over which the spatial co-expression is defined. In our Visium data, we have two adjacent slides, referred to as technical replicates. Each technical replicate contains tumor samples from two different mice, referred to as biological replicates. To leverage the full set of Visium data points to yield higher statistical power while accurately reflecting the neighborhood structure, we performed a single Hotspot analysis for all Visium spots while restricting neighborhoods to be only within each mouse and each technical replicate. We achieved this by shifting the coordinates of Visium data points for each tumor sample in each replicate by a large numeric number, e.g.,1000, such that coordinates have no overlap within the Kth (K=6, discussed below) nearest neighbor for Visium spots from different tumor samples. We excluded i) genes with zero count in all spots, and ii) spots containing fewer than 1000 genes or 1000 UMIs from downstream Hotspot analysis. We define n_neighbors=6 in the create_knn_graph in Hotspot to include only Visium spots that are adjacent to each other and disabled weighted_graph arguments to treat all adjacent spots and the self-spot equally. We selected genes with spatial autocorrelation FDR less than 0.01 and transcription factors within the top 500 autocorrelation Z score, followed by re-computing the pairwise local correlation between the top 500 transcription factors. To cluster transcription factors into modules, we used the “create_modules” function provided by the Hotspot package, with parameters min_gene_threshold=15, core_only=True, fdr_threshold=0.05, which performs a bottom-up clustering procedure by iteratively merging two genes/modules with the highest pairwise Z-score. We analyzed the Hotspot local correlation statistics over a more comprehensive gene set without pre-filtering genes based on autocorrelation to demonstrate the behavior of the statistics of different types of markers. We call the modules resulting from this combination of BayesPrism and Hotspot, PrismSpot.
Assembly: mm10/GRC38
Supplementary files format and content: .pdf file contains lowres tissue positions and tumor IDs
Supplementary files format and content: .tif files contain highres tumor slide scan
Supplementary files format and content: .cloupe are processed cloupe datafiles
Supplementary files format and content: barcodes.tsv.gz are processed spaceranger barcodes
Supplementary files format and content: features.tsv.gz are processed spaceranger features
Supplementary files format and content: matrix.mtx.gz are processed spaceranger feature-barcode matrices
Library strategy: Spatial Transcriptomics
|
| |
|
| Submission date |
Nov 01, 2023 |
| Last update date |
Aug 02, 2024 |
| Contact name |
Rodrigo Romero |
| Organization name |
Memorial Sloan Kettering
|
| Street address |
1275 York Avenue
|
| City |
New York |
| State/province |
NY |
| ZIP/Postal code |
10065 |
| Country |
USA |
| |
|
| Platform ID |
GPL24247 |
| Series (2) |
| GSE246762 |
Single nuclear RNA sequencing and 10X spatial transcriptomic profiling of murine tumors initiated by transplantation of engineered prostate organoids [visium] |
| GSE246770 |
Profiling of murine tumors initiated by transplantation of engineered prostate organoids |
|
| Relations |
| BioSample |
SAMN38061798 |
| SRA |
SRX22329443 |
| Supplementary file |
Size |
Download |
File type/resource |
| GSM7876412_RR3_B1.tif.gz |
112.1 Mb |
(ftp)(http) |
TIFF |
| GSM7876412_RR3_B1_barcodes.tsv.gz |
28.2 Kb |
(ftp)(http) |
TSV |
| GSM7876412_RR3_B1_features.tsv.gz |
256.3 Kb |
(ftp)(http) |
TSV |
| GSM7876412_RR3_B1_matrix.mtx.gz |
79.6 Mb |
(ftp)(http) |
MTX |
| GSM7876412_Sample_RR3_B1_IGO_13263_6.cloupe.gz |
200.6 Mb |
(ftp)(http) |
CLOUPE |
SRA Run Selector |
| Raw data are available in SRA |
|
|
|
|
 |
