Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
GeneDosage_scDNAseq_MinZhao - Copy.pptx
1. Gene dosage analysis on the single-cell
transcriptomes linking cotranslational
protein targeting to metastatic triple-
negative breast cancer
Dr. Min Zhao
Senior Lecturer in Genomics
University of The Sunshine Coast
2. Dosage-sensitive CNVs
2
• Copy-number variants (CNVs) reshape gene structure, modulate gene
expression, and contribute to significant phenotypic variation.
• Dosage sensitivity is a major determinant of CNV pathogenicity
• Previous research:
• Copy number loss of tumor suppressors (Zhao 2016)
• Copy number gain of oncogenes (Zhao 2018)
First to explore the dosage effects at the single cell level
3. Two TNBC datasets 3
To study the CNV and mRNA independently
The input CNV and single-cell transcriptome data are required to be derived
from a different technological platform and paired to each other at the
patient level. → Two metastatic TNBC datasets (primary and validating)
Primary dataset GSE118389, there are 1530 cells from six patients.
Karaayvaz M, Cristea S, Gillespie SM, Patel AP et al. Unravelling
subclonal heterogeneity and aggressive disease states in TNBC
through single-cell RNA-seq. Nat Commun 2018 Sep 4;9(1):3588.
PMID: 30181541
The validating study GSE75688 was conducted in 2017 and the number of cells
is comparatively smaller (515 cells).
Chung W, Eum HH, Lee HO, Lee KM et al. Single-cell RNA-seq enables
comprehensive tumour and immune cell profiling in primary
breast cancer. Nat Commun 2017 May 5;8:15081.
PMID: 28474673
5. Technical specification 5
1. To avoid expression bias on those longer genes
The Transcripts Per Million (TPM) → matrix with 21785 genes × 1533 cells
2. Z-score for expression
• 33,396,405 Z-scores
• 21785 × 1533
• Relative expression
• |Z| = |(x- µ)/σ|
• Cutoff 1.95 (P-value 0.05)
3. Gene dosage sensitivities by matching expression to CNVs
89,524 CNV-expression pairs
• Filtering CNV-expression consistency occurred in less than 100 cells.
• Focusing on the gene copy gain
• 47,514 copy number gain-Up-regulations on 94 genes across 1145
cells
7. Mutational and
prognostic features
7
• Focusing on 33 ribosome complex
genes
• Mutational frequency interactome
• Number of connections were found to
be associated with the mutational rates
• Three metastatic breast cancer
cohorts were highly mutated (over
50% of patients in the cohort)
• The 1625 patients with certain genetic
mutations on these 33 genes,
significantly different survival curves
8. Clinical features of 1625 patients
8
• Race category
• Histological grade
• Tumor stage
• Chemotherapy treatment
• Diagnosis age
• Aneuploidy score
• Hypoxia score
9. Heterogeneous indices
9
• These two indices are the classic diversity index in ecology to depict alpha
diversity, which represents the species richness in a plot.
•
• How the number of expressed genes (analogue to species) were expressed in a
cell (analogue to ecological plot).
• Both indices highly correlated to each other, which confirmed the expression
variations in patients.
10. Cell states in pathway level 10
• Breast cancer stemness:
CD44, ITGA6, DNER,
ALDH1A3, and ABCG2
• SRP module (33 genes in this
study)
• Those ribosome proteins
presented positive associations
with cell differentiation, stem-
ness, and EMT/metastasis
• Differences on the patient
level. For example, the SRP
module is not well correlated
with other cell states in PT126
due to a lack of enough
expression data
11. Summary 11
1. The trend in the field of single-cell sequencing will
be multi-omics-based data integration.
2. A reusable computational framework to explore
dosage effect at the single-cell level.
DNAseq >> CNV> scRNAseq >> phenotype
3. The increased ribosome gene copies in TNBC may
associate with their enhanced stemness and
metastatic potential (Chitra et al. Biomedicines 2022, 10(8)).
4. Prioritization of a well-upregulated functional module
confirmed by high copy numbers at the single-cell
level and conferring with patient survival may
indicate the possibility of targeted therapy based on
ribosome proteins for TNBC.
5. What is the next?
• Integrating Spatial transcriptome
• Epigenetic or microenvironment changes
12. Thank you for your attention.
Questions?
12
Contact: mzhao@usc.edu.au
Editor's Notes
Figure 1. The workflow to explore the global CNV patterns across multiple cell types with gene expression changes at the single-cell level. This workflow starts from the CNV and expression profiling at single-cell level from the same tissue samples. By mapping the CNV and gene expression changes in the same cells, this workflow will identify concordant CNV and expression changes. By running the gene-gene interaction network analysis, the workflow will build highly connected functional modules to prioritize the key genes with significant gene dosage effects. We also recommend validating the whole process by integrating other independent datasets with both CNV and expression data at single-cell level. The final validated functional modules in multiple cells will be highly reliable for further clinical feature evaluation and experimental validation.
Due to the cost-efficient consideration, a lot of CNV were inferred based on single-cell RNA sequencing (scRNASeq) expression data, which is not independent and not suitable for cross-validation. Theoretically, we required the input CNV data, and single-cell transcriptomes should have been derived from a different technological platform and paired to each other at the patient level. For example, the bulk sequencing for all the DNA content from a tumor sample of patient and scRNASeq could be independently applied to thousands of cells from the same sample.
As shown in Figure 1, our pipeline started with a compilation of matched CNV data from DNA sequencing and expression data from single-cell RNA sequencing. Next, we transformed the cell-based RNA expression profile to the Z-score, which will depict how genes were expressed relatively higher or lower in a particular cell by comparing it to the quantitative expression values from other cells. Due to the limitation of RNA content, scRNAseq often has a lot of dropouts, which results in none of the RNA short reads being detected in scRNAseq. To prevent the nonsense Z-score, we masked those cells without expression value. To extract a meaningful Z-score for further mapping to CNV data, we set a threshold for the absolute Z-score over 1.96, which means significant P-values 0.05. Then, the transformed significant Z-score for each gene in a particular cell was mapped to the CNV data from the same patient. The concordant copy number events were screened based on the criteria of whether the copy number gain/loss in the cell correlated with a consistent gene up/down-regulation. By counting how many consistent CNV-expression events occurred across all the cells from the input dataset, we further inferred a list of genes with recurrent CNV-based dosage changes. By incorporating the known gene-gene interaction and functional similarity, the framework would intend to identify functional modules for those recurrent CNV-changed genes detected in hundreds of cells regardless of their cell types.
In summary, the input for our framework is the independent CNV and scRNAseq data from the same patients. The output was a connected functional module that comprised the top mutated genes based on consistent CNV and expression changes. To further reduce the false positives of functional modules inferred from our computational pipeline, we would suggest validating those functional modules in another independent dataset. Once the top prioritized functional modules and genes can be confirmed, any routine computational or experimental design will be adopted to further explore the significance. For instance, our final step of this study was to take advantage of large-scale cancer genomics data to evaluate whether those top-ranked genes were highly mutated and associated with patient survival.
of a gene over the averaged expression across 1533 cells.
Figure 2. Functional analysis of two top-ranked gene lists with concordant Copy Number Gain and UP-regulation (CNG-UP) genes. (A) The bar plot shows the gene ontology (GO) cluster representatives for the 94 CNG-UP genes from the primary scRNAseq dataset. The x-axis indicates the log of corrected P-value. (B) The five functional modules-based gene-gene interactions for the 94 CNG-UP genes from the primary scRNAseq dataset. The different colors represent the five identified functional modules. The top ranked modules related to cotranslational protein targeting membranes were depicted in red nodes. Another four modules are purine ribonucleoside triphosphate biosynthetic process (blue nodes), SCF(Skp2)-mediated degradation of p27/p21 (green nodes), apoptosis (purple nodes), and neutrophil degranulation (orange nodes). (C) The representative GO clusters for 86 CNG-UP genes from the validating dataset. The x-axis indicates the log of corrected P-value. (D) Three functional modules summarized from gene-gene interactions for the 86 CNG-UP genes from the validating dataset. The top three ranked modules are SRP-dependent cotranslational protein targeting to membrane (in red), interferon-gamma-mediated signaling pathway (in blue), and cellular responses to stress (in green). (E) The overlapping of top-ranked modules associated with protein targeting from primary and validating datasets. Seven genes are shared in both modules. (F) The representative GO clusters for 33 CNG-UP genes related to SRP-dependent cotranslational protein targeting to membrane combined from the primary and validating datasets. The x-axis indicates the log of corrected P-value.
To focus on the TNBC and explore the potential novel target for cancer therapy, we applied this computational pipeline to the TNBC dataset GEO118390 with six patients and an expression profile from 1533 cells. Instead of cell-type identification, we focused on the common mechanisms underlying copy number variants across all the cell types. To avoid expression bias on those longer genes with more abundant sequencing reads, we used the Transcripts Per Million (TPM) as the quantification method. In detail, the sum of all TPMs in all the 1533 cells is the same, which makes the proportion of short reads mapped to a gene comparable in different cells.
By starting from a TPM expression matrix with 21785 genes × 1533 cells, we first generated the same size matrix with 33,396,405 Z-scores (21785 rows and 1533 columns), which depict the relative expression of a gene over the averaged expression rate across all the 1533 cells. The positive Z-score means the gene expression is relatively higher than those expressions from the remaining cells. By running through the described computational pipeline, we first masked 29,931,625 zero expression values. Base on the remaining 3,464,780 meaningful Z-scores, we mapped the CNV data from the whole exome sequencing in the same six patients. In total, 1,440,802 non-redundant CNV-expression events in 1245 cells were obtained.
However, most CNV-expression events (1,351,278) may not be informative since these expression variations are statistically meaningful (e.g. their absolute Z-score < 1.96). On the contrary, we defined 89,524 meaningful CNV-expression events as consistent events with an absolute Z-score cutoff over 1.96 (Table S1). On the other hand, some CNV-expression consistency just occurred in a small number of cells. By pulling out all the cells with gain or loss of gene copies, we set a threshold of 100 or more cells to prioritize the most informative copy number variation events. Notably, we only focused on those consistent Copy Number Gain and gene UP-regulation (CNG-UP) genes because the majority of those Z-scores for down-regulation are higher than -1.96. Finally, these step-by-step filterings entailed a total of 47,514 CNG-UP events associated with 94 genes across 1145 cells. It is worth noting that these huge consistent patterns across patients is highly reliable, which may empower scRNASeq-driven clinical decisions in future.
Figure 3. The overall mutational and prognostic features for 33 genes with increased gene ex-pression induced by CNGs. (A) The gene-gene interaction network to show the mutational fre-quency in 6688 breast cancer samples combined from 12 studies. The size and color of each node is correlated with the number of connections. (B) The mutational summary for twelve breast cancer datasets; (C) The overall survival plot shows median survival months for those patients with or without any mutation on the 33 genes.
Figure 4. Clinical features summary for the 33 ribosome genes on (A) race category; (B) neoplasm histological grade; (C) tumor stage; (D) chemotherapy treatment; (E) diagnosis age; (F) aneuploidy score; (G) ragnum hypoxia score. The P-values are the statistical test between the patients with or without any genetic mutations on the 33 genes.
Strikingly, these mutations were also positively associated with other important clinical features (Figure 4), including the race category, diagnosis age, histological grade, tumor stage, aneuploidy score, hypoxia score, and chemotherapy treatment. For example, there are relatively less Asian compared to White patients in the group with mutations (Figure 4A). For histological grade and tumor stage, those patients with mutations tend to be in the higher grade (Figure 4B) or late stage (Figure 4C). The chemotherapy was applied more to those patients without any mutations for these 33 genes (Figure 4D). In terms of diagnosis age, those patients with mutations on these 33 genes tend to be diagnosed later (Figure 4E), which might explain their higher grade and late stage. The presence of mutations is also indicative of higher aggression of factors such as aneuploidy (Figure 4F) and hypoxia (Figure 4G). To be brief, the further mutational analysis expanded our understanding about the clinical features on those 33 ribosome proteins identified by our computational framework.
Patient PT126 had the lowest Shannon-Wiener index, which means less genes expressed and with lower expression in those cells in PT126.