single cell sequencing - kidneycan · 10/12/2019 · single cell sequencing a ari hakimi md dept...
TRANSCRIPT
Single Cell SequencingA Ari Hakimi MD
Dept of Surgery, Urology
Immunogenomics and Precision Oncology Platform
MSKCC
Differences within cell type Differentiation trajectories- cell type vs cell state
High-resolution heterogeneity
Integrative analysis- TCR-seq, ATAC-seq, whole-genome
Spatial sequencing is of major interest
Why single-cell RNA-seq?
Types of single cell sequencing
Stuart et al Nat Rev Gen 2019
Stuart et al Nat Rev Gen 2019
Cell of origin studies
Matthew D. Young et al. Science 2018;361:594-599
Copyright © 2018, American Association for the Advancement of Science
Matthew D. Young et al. Science 2018;361:594-599
Copyright © 2018, American Association for the Advancement of Science
Importance of Batch Correction
• Merged Mouse samples• P1: Parental replicate 1
• P2: Parental replicate 2
• P3: Parental replicate 3
• Q1: Experimental replicate 1
T cell cluster enrichmentBefore Batch Correction
After Batch Correction
No batch correction (N = 24):
color by region
No batch correction (N = 24):
color by patient
Batch correct each patient (N = 18):
color by region
Batch correct each patient (N = 18):
color by patient
Batch correct each region (N = 18):
color by region
Batch correct each region (N = 18):
color by patient
Increasing strictness of correction
Once the entire dataset is in- need to quantify this effect more rigorously
Batch correction – human data
UT1: N = 26177 UT2: N = 13213
t1 (Nivo-exposed):
N = 30547
t2 (Ipi/Nivo-resistant):
N = 30547
t3 (Ipi/Nivo-mixed response):
N = 38660
t4 (Ipi/Nivo-complete response):
N = 38660
N = 167283
Following batch correction using mutual nearest
neighbors (MNNCorrect)
Six patients w/ site-matched
scRNA+TCR-seq: 2 untreated (UT1-2),
4 treated (t1-4)
+ exp design schematic, clinical characteristics, any Sounak path data
4 tissue types
Tumor: N = 107806 Lymph Node: N = 3835 Normal Kidney: N = 24096 PBMC: N = 31546
CD45 CA9
CD3D CD14 CD79A
Distribution of immune + malignant cell types
CA9
Cluster 0- Hypoxia/Metastasis
P = 0.003 P = 0.009
Cluster 4- Type 1 Interferon Signaling
P = 0.0 P = 0.0
Multiple CA9 (tumor specific) phenotypes – correlate with unique TMEs
P = 0.0
Cluster 7- Antigen Processing/Presentation + myeloid/CD4 T cell infiltrated
P = 0.0 P = 0.0 P = 0.0 P = 0.0
T/B/NK cell cluster prevalence by tissue
Tumor Normal Kidney PBMC
Myeloid cluster prevalence by tissue
Tumor Normal Kidney PBMC
UT1 UT2 t1
t2 t3 t4
T/B/NK cell cluster prevalence by tumor region
Capturing cell transitional states
• New methods to capture the relationship between tissues + continuity of gene expression underlying immune cell differentiation/development
• PCA: linear embedding- useful for visualization- not useful for capturing underlying structure/differentiation
• Diffusion maps: non-linear embedding that emphasize transitions in the data- typically used when processes are continuous (i.e. T cells, monocytes from blood → tumor)
Diffusion Maps: Non-linear embeddings for scRNA-seq data
UT1 T cells
Associations with response in BMS009
C2: NK cells C23: B cells
C0: CD8+ T cells C10: FOXP3+ Tregs
Binary classification using RECIST criteria: responders defined as CR/PR/S; non-responders as PD
Conclusions
• Single cell sequencing continues to evolve
•Batch correction critical when merging data sets/types• Easy to find spurious associations
• Transition between cell states can be capture with new embedding techniques (ie diffusion maps)
•Unique opportunities to study the TME and TME evolution on treatment