1 of 17

Re-analysis of the public dataset:

comparison and evaluation of T cells in paired tumor and normal lung single cell samples

Student:

Fedor Grigoryev

Moscow Institute of Physics and Technology

Supervisor:

Ekaterina Esaulova,

BioNTech US

2 of 17

2

Introduction

Due to numerous mutations in the tumor genome, the neoplasm expresses mutation-associated neoantigens (MANA) that can trigger an immune response against cancer in patients. However, as the tumor evolves, malignant cells often acquire the ability to evade the immune response by expressing checkpoint proteins. Immunotherapy using antibodies against these checkpoints has been developed, but a significant number of patients do not respond to this treatment. Biomarkers, such as the expression of checkpoint proteins, tumor mutational load, and neoantigenic load, are used to determine the suitability of immunotherapy. These biomarkers are particularly relevant for highly mutated cancers like melanoma and lung cancer. However, they are not comprehensive, highlighting the need for further research on the immune component of tumors and also dysfunctional programs in MANA specific tumor-infiltrating lymphocytes (TILs).

Our project is based on open data from the article “Transcriptional programs of neoantigen-specific TIL in anti-PD-1-treated lung cancers” (Caushi J. X. et al., 2021. doi: 10.1038/s41586-021-03752-4).

3 of 17

3

Aim and objectives

Aim

Re-processing and replication of the analysis for published single-cell RNAseq and matched TCRseq data: T cells from normal and cancerous tissues from multiple donors

Objectives

  • Obtain post-cellranger data from GEO
  • Perform QC, including identification of mitochondrial gene content, UMI content to filter out dead/damaged cells
  • Filter genes for the downstream analysis to remove noise: removed Ig, TCR genes and IFN response genes
  • Identify and annotate T cells clusters by the expression of marker genes of these subsets
  • Evaluate amount of cells with TCR and compare repertoires between normal and tumor matched samples  
  • Evaluate diversity of the TCR repertoire

4 of 17

4

Approaches

+ TCRseq

single-cell

RNAseq

5 of 17

5

Results

Low-quality cells were filtered out if:

  • the number of detected genes was below 250 or above 2500
  • the proportion of mitochondrial gene counts was higher than 10%
  • the proportion of ribosomal gene counts was lower than 10%
  • doublets were removed with with the help of Scrublet

Mitochondrial genes, and genes associated with poorly supported transcriptional patterns were also removed from the analysis

 

Patient

Type of tissue

Number sequencing cell

Number sequencing cell after filtration (%)

MD043-006

normal

10316

7049 (80.5)

tumor 

11958

9633 (82.0)

MD043-011

normal

3633

2816 (77.5)

tumor 

7264

4373 (65.3)

MD01-010

normal

3502

2901 (82.8)

tumor 

6327

4373 (69.1)

MD043-008

normal

2854

2130 (77.5)  

tumor 

2115

1735 (65.3)

Total

47969

35385 (73.8)

Table 1. Number sequencing cell before and after filtration

/ QC and filtration

6 of 17

6

Results

/ Need for batch correction

Fig. 1 UMAP after QC filtration

TCR, immunoglobulin and mitochondrial genes, as well as features that constitute Interferon I mediated pathway, were excluded from clustering to make sure that clustering result will not be influenced by their variability.

PCA was performed based on the 3,000 most variable genes. UMAP on PCA results have shown that cells group by samples of their origin, indicating the need for a batch effect correction.

7 of 17

7

Results

/ UMAP after batch correction

After harmonization, cell clusters are more evenly distributed among patients, however, the biological difference of the tumor/normal immune environment is not lost

Fig. 2 UMAP after batch correction

8 of 17

8

Results

/ Clustering and annotation

Leiden clustering resulted in 14 separate clusters, that were annotated using combination of general CD4/CD8 markers with common subset specific markers (as FOXP3 for Tregs; MKI67 for proliferating cells; CXCL13 for Follicular helpers; GZMA, GZMB, GZMK for effector cells; ZNF683 and ITGAE for memory cells; KLRC1 for NK cells; SLC4A10 for MAIT cells and so on)

Fig. 3 Clustering results

Fig. 4 Cell subtype marker expression

9 of 17

9

Results

Fig. 5. Dotplot of cell markers, differentially expressed genes and T cell checkpoint associated genes

Differentially expressed genes were found using wilcoxon test for each cell type vs all other cells

Fig. 3 Clustering results

/ Cell annotation

10 of 17

10

Results

Fig. 6. Cell type composition by type of tissue

As expected, cellular distribution in tumor and normal sample is highly different.

  • CD8+ effector2 cells are enriched in Tumor, while CD8+ effector1 – in Normal tissue.
  • Both CD4+ helper subtypes are enriched in Normal, although CD4+ Follicular helpers and CD4+ T reg are more specific for Tumor.
  • CD8+ mem1 is enriched in Tumor

/ Cell type enrichment by tissue

Different subsets of CD8+ cells are expanded to different pathogenic sources. Whereas influenza-specific cells should be the most abundant in normal lung, MANA-specific CD8 cells should be more numerous in the tumour.

Indeed, query to VDJdb showed that proportion of CD8+ TCRs specific to common lung pathogens (InfluenzaA, SARS-CoV-2, M.tuberculosis) was 1.4 fold higher in normal subset.

11 of 17

11

Results

/ Matched TCR seq merge

Fig. 7. Fraction of cells with matched TCR by sample

Fig. 8. Fraction of cells with matched TCR by patient

Fig. 9. UMAP for matched TCRs

12 of 17

12

Results

Fig. 10.  Clonal expansion per sample

/ Matched TCR seq merge

Fig. 11. Clonal expansion per cell type

CD8+ cell clones are highly more expanded than CD4+

13 of 17

13

Results

/ Matched TCR seq merge

Fig. 12. Cell type composition across 10 most populated clonotypes in Tumor and Normal

Tumor and Normal samples do not share clonal compositions

14 of 17

14

Results

Construction of a clonotype network by computing distances between CDR3 sequences reveals a single public clonotype - clonotype cluster shared by all of the patients, illustrating possible specificity to a common pathogen

The query to VDJdb has shown that TCRs from one of public clonotype – 1238 – networks were indeed specific for antigens from single pathogenic species - CMV.

/ Clonotype network construction

Fig. 13. Clonotype networks

15 of 17

15

Results

Query to VDJdb has revealed specificity of some TCRs to epitopes of common pathogens. However, as the neoantigens are tumor specific, it is impossible to mine TCR specificity from public databases. In fact, exploring transcriptional programs of MANA specific cells would be of greatest interest.

/ Epitope specificity

Fig. 14. Map by TCR sequence specificity to corresponding pathogenic species queried from VDJdb

16 of 17

16

Summary and Further plans

Conclusion:

We have explored the possibilities of paired RNAseq+TCRseq on a public dataset of matched tumor and normal tissues in resectable non-small cell lung cancer.

  • The T-cells were clustered and annotated into subgroups. Cell type enrichment was explored for Tumor and Normal samples.
  • Matched TCR sequencing has allowed for researching clonal expansion independently by subtypes.
  • Construction of clonotype networks has revealed a specificity to a public epitope.

It would be of great interest to explore and compare transcriptional programs of tumor associated antigen specific clones in the subsequent works.

17 of 17

17

GitHub

to be announced!