dkNET New Investigator Pilot Program in Bioinformatics Awardee Webinar Series
Presenter: Wenting Wu, PhD. Research Assistant Professor, Center for Diabetes and Metabolic Diseases, Department of Medical and Molecular Genetics, Associate Director of Data and Analytics Core for Center for Diabetes and Metabolic Diseases, Indiana University School of Medicine
Abstract
Type 1 diabetes (T1D) is an immune-mediated disease that results in insulin insufficiency and affects 0.3% of the population, including both children and adults. To support clinical trial efforts, there is an urgent need to develop reliable biomarkers capable of predicting T1D risk and guiding therapeutic interventions. Recently, whole blood bulk RNA sequencing has been used to guide T1D clinical trial design and assess response to disease modifying interventions. While the use of bulk RNA sequencing is cost-effective, these datasets provide limited information about cell specific gene expression changes. Here, we aimed to apply computational strategies to deconvolute cell type composition using cell specific gene expression references. Single-cell RNA sequencing (scRNA-seq) was conducted to profile peripheral blood mononuclear cells obtained from youth within recent T1D onset and age- and sex-matched controls and identified 31 distinct cell clusters. Using this pre-defined reference dataset, we ran computational algorithms CIBERSORTx and other deconvolution methods simultaneously to deconvolute cell proportions using public clinical trial data. We focused our initial analysis on data from the TN-20 Rituximab trial, which tested the anti-CD20 monoclonal antibody rituximab vs placebo in recent onset T1D. This talk will introduce recent advances of scRNA-seq techniques and computational deconvolution methods and demonstrate that how we apply different deconvolution approaches for secondary analysis of existing clinical trial data, in the purpose of linking cell specific immune signatures associated with drug responder status.
Upcoming webinars schedule: https://dknet.org/about/webinar
dkNET Webinar: Leveraging Computational Strategies to Identify Type 1 Diabetes Risk and Clinical Trial Responder Status 05/19/2023
1. Leveraging computational strategies to identify type 1
diabetes risk and clinical trial responder status
dkNET Webinar
May19, 2023
Wenting Wu
Center for Diabetes and Metabolic Diseases
Department of Medical and Molecular Genetics
School of Medicine, Indiana University
2. Presentation Overview
• Unravelling alternative splicing patterns in pro-Inflammatory cytokines
treated human islets and new-onset type 1 diabetes (T1D).
• Leveraging single cell RNA-sequencing and deconvolution methods into
heterogeneity of peripheral blood mononuclear cells from youth with
recent onset type 1 diabetes, to identify T1D risk and clinical trial
responder status.
3. Presentation Overview
• Unravelling alternative splicing patterns in pro-Inflammatory cytokines
treated human islets and new-onset type 1 diabetes (T1D).
• Leveraging single cell RNA-sequencing and deconvolution methods into
heterogeneity of peripheral blood mononuclear cells from youth with
recent onset type 1 diabetes, to identify T1D risk and clinical trial
responder status.
4. Genetic
Risk
Immune
Activation
Immune
Response
STAGE 1 STAGE 2 STAGE 3 STAGE 4
Genetic
Risk
Immune
Activation
Immune Response
Development of single
autoantibody
Immune
Response
Immune Activation
Beta cells are attacked STAGE 2
STAGE 1 STAGE 3
Abatacept
Ritixumab
Verapamil
Alefacept
ATG/GCSF
Imatinib
The landscape of disease modifying therapy in T1D
Teplizumab
Golimumab
Teplizumab
Anti-IL-21
Interventions at Stage 3 onset have
shown limited success
• Interventions at Stage 3 onset have not reliably induced a
durable disease remission. Teplizumab delayed progression
from Stage 2.
• Bacteria
• Viruses
• Insulin
• GAD65
• IA2
• IGRP
Normoglycemia
≥ 2 autoantibodies
Dysglycemia
≥ 2 autoantibodies
Clinical Diagnosis
≥ 2 autoantibodies
Established/
Long-
standing T1D
15x increased
risk of T1D in
those with
relatives of
disease
5. STAGE 4
Genetic
Risk
Immune
Activation
Immune
Response
STAGE 1 STAGE 2 STAGE 3
Is there a role for transcriptomic profile in T1D early onset status?
Starting Point
15x increased
risk of T1D in
those with
relatives of
disease
Immune
Activation
Seroconversion Normoglycemia
≥ 2 autoantibodies
Dysglycemia
≥ 2 autoantibodies
Clinical Diagnosis
≥ 2 autoantibodies
Established/
Long-
standing T1D
Cadaveric Human Islets (n=10)
Gender: 6M/4F
Mean Age: 30 ± 10.25
Mean BMI: 28.9 ± 4.68.
+ Cytokines
- Cytokines
IL-1𝛽 + 𝐼𝐹𝑁 − 𝛾
• What are the molecular
pathways responsible for
immune and β cell dysfunction
in early stage disease? Can
these pathways be leveraged
for therapeutic strategies?
Model 1
6. STAGE 4
Genetic
Risk
Immune
Activation
Immune
Response
STAGE 1 STAGE 2 STAGE 3
Is there a role for transcriptomic profile in T1D early onset status?
Starting Point
15x increased
risk of T1D in
those with
relatives of
disease
Immune
Activation
Seroconversion Normoglycemia
≥ 2 autoantibodies
Dysglycemia
≥ 2 autoantibodies
Clinical Diagnosis
≥ 2 autoantibodies
Established/
Long-
standing T1D
• What are the
transcriptome signals
released from
“stressed” β cells to
circulation that can be
leveraged to identify
disease at earlier
stages?
Model 2 Whole Blood RNA from new-onset T1D (n=48)
Healthy Controls
Gender: 16M/8F
Mean Age: 12 4.29
Mean BMI: 19.8 3.85.
New Onset T1D
Gender: 16M/8F
Mean Age: 12 4.28
Mean BMI: 19.1. 4.78.
±
±
±
±
HC
T1D
7. What are the pathogenic Alternative Splicing events responsible for immune and β cell
dysfunction in early-stage disease?
Input Data
Random Forest
Identification of Differential
Splicing Events
Wu W, et al. Diabetes. 2022
Nonsense-mediated decay
(NMD) Prediction
GWAS and sQTL
co-localization analysis
qPCR smFISH
ExonImpact Prediction
RNA Sequencing
Primary cultures
PC2
PC1
Depletion Inclusion
motif1
motif2
motif3
motif5
motif4
1 2 3 4 5 6 7
K-mer Z-score distribution
-4 -2 0 2 4
0.
0
0.
1
0.
2
0.
3
0.
4
Comparison of t Distributions
x value
Densi
t
y
RNA Binding Protein Motif Prediction
8. RNA-binding Protein Motif Enrichment Analysis
Exonic and intronic
sequences can modulate the
splice site selection by
functioning as splicing
enhancers or silencers.
Kornblihtt AR et al. (2013) Nat Rev Mol Cell Biol
Wu W, et al. Diabetes. 2022
10. Functional Prediction into Alternative Splicing Exons
ASA: soluble accessible surface areas
PTM: post translational modification
In total of 31 curated protein structure features.
Cross
Validation
SPANR
dPSI test
Clinvar
test
GTEx test
RandomForest Model
CV’s
ROC
dPSI correlate
with FIS
Clinvar
benign FIS
GTEx
FIS test
PhyloP
Secondary structure
ASA
Diisorder
Pfam
PTM
Training
Feature type Testing dataset
129 splicing events coupled with multiple
isoform with prediction score > 0.5
Wu W, et al. Diabetes. 2022
Dong, et al. JCO Clinical Cancer Informatics. 2022
Prediction into Functional Impact using ExonImpact
Prediction into Nonsense-mediated decay (NMD)
HLA-DMB Skipped Exon
11. Consistently Increased Splicing of HLA-DMB Exon in Multiple Human Donors
Upregulated Protein Expression
Supplementary Figure 6. Immunofluorescence staining of HLA-DMB protein in human
islets. Immunofluorescence staining of human islets treated with or without pro-inflammatory
cytokines probed for insulin (blue), HLA-DMB (red), LAMP1 (green) and DAPI (white), showing
decreased colocalization of HLA-DMB with LAMP1 in cytokine treated islets shown with yellow
arrow heads.
Decreased colocalization of HLA-DMB with LAMP1 in cytokine
treated islets
Con Cyt Con Cyt Con Cyt
0.0
0.2
0.4
0.6
0.8
1.0
Inclusion
Level
RNA-Seq
(Gonzalez-Duque et al, Cell
Meta,2018)
IncDiff = -14.4%
IncDiff = - 26.4%
RT-PCR
RNA-Seq
(Our data)
IncDiff = -27.6%
Consistently increased splicing of HLA-DMB exon
Genomic structure of the HLA-DMB gene.
Lysosomal targeting
signal domain
Wu W, et al. Diabetes. 2022
12. STAGE 4
Genetic
Risk
Immune
Activation
Immune
Response
STAGE 1 STAGE 2 STAGE 3
Is there a role for transcriptomic profile in T1D biomarker strategies?
Starting Point
15x increased
risk of T1D in
those with
relatives of
disease
Immune
Activation
Seroconversion Normoglycemia
≥ 2 autoantibodies
Dysglycemia
≥ 2 autoantibodies
Clinical Diagnosis
≥ 2 autoantibodies
Established/
Long-
standing T1D
Alternative Splicing Events in Human Blood
(150M Reads)
Healthy Control (n=12) New Onset T1D (n=12)
Validation Set
Alternative Splicing Events in Human Blood
(180M Reads)
Healthy Control (n=12) New Onset T1D (n=12)
Training Set
Boot-strap
Random
Forest Model
13. Machine learning of transcriptome profile to identify signatures that
predictable new-onset T1D
AS RNA vs mRNA Clustering between T1D and healthy controls
Unpublished data
14. Machine learning of transcriptome profile to identify signatures that
predictable new-onset T1D
AS
Categories
# of
Features in
Training
Set
% Validation Set
Overlap with
Training Set
AUC in
Training (95% CI)
AUC in
Validation
SE 4174 83.9% 0.97 (0.88,1) 0.87
RI 1043 97.3% 0.86 (0.65,1) 0.86
A5SS 479 93.1% 0.95 (0.89,1) 0.85
A3SS 706 92.2% 0.97 (0.88,1) 0.69
MXE 1537 35.6% 0.97 (0.88,1) 0.62
Unpublished data
15. Summary– Part I
• Alternative splicing (AS) is a prominent mechanism of gene regulation for the human islet
cells response to cytokine-mediated stress. In the meantime, it is observed within whole
blood of new onset T1D patients, that may unmask circulation biomarkers.
• Using sophisticated bioinformatics methods, these studies uncovered new role for AS in
the regulation of MHC Class II molecules, in the β-cells and to elucidate their underlying
molecular mechanisms.
• Bioinformatics tools provide further functional relevance of these aberrant alternative pre-
mRNA splicing, e.g. machine-learning based approach to help prioritize potentially
pathogenic splicing events and may identify new predicable elements of risk of T1D.
16. Presentation Overview
• Unravelling alternative splicing patterns in pro-Inflammatory cytokines
treated human islets and new-onset T1D.
• Leveraging single cell RNA-sequencing and deconvolution methods into
heterogeneity of peripheral blood mononuclear cells collected from youth
with recent onset T1D, to identify type 1 diabetes risk and clinical trial
responder status.
17. Heterogeneity among peripheral blood cells
A.Kolodziejczyk, et al. (2015). Molecular Cell
Heterogeneous cell populations
Living cells may be considered
noisy or stochastic biochemical
reactors.
18. The average may not represent
the population
Rare events can be lost…
Whole blood RNA sequencing provides limited resolution
19. Current single-cell profiling technologies
FACS CyTOF qPCR Plate-based protocaols
(STRT-seq, SMART-seq,
SMART-seq2)
Fluidigm C1 Pooled
approaches
(CEL-seq,
MARS-seq,
SCRB-seq,
CEL-seq2)
Massively
parallel
approaches
(Drop-seq,
InDrop)
Cell capture
method
Laser Mass
Cytometry
Micropipett
es
FACS Microfluidics FACS Microdroplets
Number of cells
per experiment
Millions Millions 300-1,000 50-500 48-96 500-2,000 5,000-10,000
Cost $0.05
per cell
$35 per cell $1 per cell $3-6 per well $35 per cell $3-6 per well $0.05 per cell
Sensitivity Up to 17
markers
UP to 40
markers
10-30
genes per
cell
7,000 -10,000 genes per
cell for cell lines;
2,000-6,000 genes per cell
for primary cells
6,000-9,000 genes per cell
for cell line; 1,000-5,000
genes per cell for primary
cells
7,000-10,000
genes per cell
for cell lines;
2,000-6,000
genes per cell
for primary
cells
5,000 genes
per cell for cell
lines;1,000-
3,000 genes
per cell for
primary cellss
Papalexi, et al. (2018). Nature Review Immunology
$$$$
20. Deconvolution
• Reduced cost of bulk sequencing compared to single cell strategies.
• Single-Cell RNA Sequencing are not practical in large sample cohorts.
• Traditional immunophenotyping approaches, including flow cytometry and
immunohistochemistry, rely on small combinations of preselected marker
genes. Ability to multiplex is limited.
• Most fixed clinical specimens (.eg formalin-fixed, paraffin embedded
FFPE) samples cannot be dissociated into intact single-cell suspensions.
• The impact of tissue disaggregation on cell type representation is poorly
understood.
Advantages compared with scRNA-seq:
21. Mixtures (X) are a
linear combination of
signature matrix (S)
and concentration
matrix (C)
𝑋! # $ = 𝑆! # % × 𝐶% # $
mixtures
mixtures
genes
cell types
genes
cell
types
Deconvolution strategies can be used to infer cell type specific changes
from bulk RNA sequencing
Newman, et al. (2019). Nature Biotechnology,773-782
Gene Set Enrichment Test
Finotello, et al. (2018). Cancer Immunology,
Immunotherapy
×
22. A variety of deconvolution strategies have been developed
Method
Statistical
approach
Input
Output:
Cell type
proportions
Output: cell
type
specific
gene
expression
Refence
CIBERSORT V-SVR
Reference
profiles
Y N
Newman et
al.(2015)
MuSiC
Least
squares
Reference
profiles
Y N
Wang et
al.(2019)
LLSR
Least
squares
Reference
profiles
Y N
Abbas et al.
(2009)
QP
Quadratic
prog
Reference
profiles
Y N
Gong et al.
(2011)
PSEA
Least
squares
Marker genes N Y
Kunh et al.
(2011)
MMAD
Maximum
likelihood
Marker genes Y Y
Liebner et
al. (2013)
EPIC
Least
squares
Reference
profiles
Y N
Julien, et al.
(2020)
ICTD
Non-negative
matrix
factorization
Reference
free
Y N
Chang et al.
(2019)
TOAST
Least
squares
Reference
free
Y N
Li et al.
(2019)
Cellular deconvolution. In Wikipedia.
23. CIBERSORTx (reference based)
Newman, et al. (2019). Nature Biotechnology,773-782
M be an n × k matrix with n genes and k mixture GEPs.
B be a subset of H containing discriminatory marker genes for each
of the c cell subsets (i.e., signature or basis matrix).
M’ be the subset of M that contains the same marker genes as B.
F, a c × k fractional abundance matrix with columns [f1,f2,...,fk].
Non-negative least squares regression (NNLS)
Supporter vector regression
H be an n × c matrix with n genes and c cell types.
24. Deconvolution methods summary
• Transcriptome deconvolution aims to estimate the cellular composition of an RNA
sample from its gene expression data, which in turn can be used to correct for
composition differences across samples.
• Deconvolution methods fall into two main categories: reference-based deconvolution
and reference-free deconvolution.
• Among the bulk deconvolution methods, least-squares (ordinary least squares, non-
negative least squares), support-vector regression and robust regression approaches
gave the best results across different datasets and pseudo-bulk cell pool sizes.
• Non-negative matrix factorization (NMF) is complete unsupervised approach, by
incorporation of prior knowledge from cell-specific markers, it would dramatically
improve the results.
Finotello, et al. (2018). Cancer Immunology, Immunotherapy
Avila Cobos, et al. (2020). Nature Communications
25. Workflow
Study Procedures:
• Fresh blood samples were
collected after an 8-10 hr fast
and within 48 hrs of diagnosis
of Stage 3 T1D onset.
• PBMC was immediately
isolated within 2 hrs of blood
drawing and sequencing library
construction were performed
timely.
27. Integrative single-cell analysis of PBMCs collected from
individuals with recent onset type 1 diabetes and healthy
controls
18 Cell subsets, with an additional 13 states identified following sub-clustering.
28. • T cells were detected with high
frequency.
• The proportion of CD4+ T central
memory (TCM) cells was increased
(P = 0.018) in the T1D group.
• The proportion of plasmacytoid
dendritic cells (pDC), hematopoietic
stem and progenitor cells (HSPC)
and platelets were significantly
reduced (P = 0.021, 0.012 and
0.017 respectively) in the T1D
group compared to the non-diabetic
control group.
Cell type proportion changes
Cell
proportions
Cell
proportions
* P <0.05
CD4+ TCM pDC
Cell
propo
* P <0.05
Cell
proportions
* P <0.05
Cell
proportions
* P <0.05
HSPC Platelet Sub Cluster 2
Cell
proportions
29. Platelet
Lymphocytes
R = 0.73, P =0.017
R = 0.77, P =0.009
R = 0.79, P =0.006
Monocytes
Cell type proportions correlate with clinical hematology laboratory testing
• Good correlation was observed
between scRNA-seq and
clinical laboratory testing of the
complete blood count.
30. Nature killer cell showed strongest immune response by differential expressed
genes patterns
UMAP_1
UMAP_2
log10 DEG number
Response to Virus
NES = 0.54, P.adjust = 0.0002
Response to cytokine
NES = 0.47, P.adjust = 0.0002
• NK cells’ role in T1D begins at early stage.
• NK cells have a complex relationship to autoimmunity and might work as effector cells to regulator of
immuno-pathology.
• Explanation might be viral trigger results in altered NK cell gene expression, associated with T1D
development.
Xhonneus, et al. (2021). Sci Transl Med
Flodstrom-Tullberg M, et al. (2009). Curr Opin Immunol.
31. Cell-type Specific Regulation in Early Immune Response of T1D development
Xhonneus, et al. (2021). Sci Transl Med
Distinct cell-specific gene
expression changes
characterize progression
to disease onset in
subgroups of patients with
T1D defined by sequence
of IAbs seroconversion.
Teddy Study
32. Peripheral CD8+ T cells are transcriptional heterogeneous
CD8+
TEM
Sub Cluster 1
GZMA
CX3CR1
FGFBP2
PRF1
GZMH
CCL4
NKG7
CST7
GZMB
CCL5
FCGR3A
GZMM
HLA-DPA1
HLA-DPB1
GNLY
CD8+ TEM
Sub Cluster 2
LTB
CD7
RTKN2
MAP3K1
SOX4
CD27
IL7R
TMEM14C
CD8+
Naïve
T Cell
CCR7
MYC
FOXP1
AIF1
EEF1A1
LDHB
CD8+
Naïve T
CD8+ TEM
Sub Cluster 1
CD8+ TEM
Sub Cluster 2
CD8+ TEM : CD8+ effector memory cell
34. Signature matrix derived from scRNA-seq was used for deconvolution
approaches
Subgroup analysis by
multiple regression models
Longitudinal study by
mixed-effects model
Pre-processing steps
Samples with low quality
RNA-Seq excluded
Intersection
CIBERSORTx ICTD
35. Deconvolution profiling of PBMC from T1D, T2D and healthy controls
Healthy
controls
(n=24)
T1D
(n=43)
T2D
(n=12)
Age, yr
(mean±SD)
11.3±4.6 10.1±3.8 14.0±2.3
Sex(% female) 58 60 58
BMI (mean Z
score±SD)
Unknown 0.03 ±
1.33
2.33 ± 0.32
Initial pH less than
7.3
n/a 37% 17%
Initial HbA1c
(mean ± SD)
n/a 11.8 ±
2.0
12.2 ± 1.5
Adapted from GSE9006.
Signature Matrix
Our Data signatures
PBMC
10X chromium
LM22
ICTD
Cibersortx
36. Cell specific signatures derived from our scRNA-Seq matrix
Dataset to be analyzed
1. Rank-1 module detection
Core markers Rank-1 modules
2. Inferring identifiable cell types
Dataset specific markers
Bi-Cross
Validation test
the rank-1 modules with
genes largely overlap with
the core marker list of one
and only one cell type Cell type level performance
!
𝑥!: 𝑒𝑥𝑝𝑙𝑎𝑖𝑛𝑎𝑏𝑙𝑒 𝑒𝑥𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛
𝑏𝑦 𝑐𝑒𝑙𝑙 𝑝𝑟𝑜𝑝𝑜𝑟𝑡𝑖𝑜𝑛
𝐸 − 𝑠𝑐𝑜𝑟𝑒 = 1 −
∑!"#
$
𝑥!
∗
− 7
𝑥!
&
∑!"#
$
𝑥!
∗ &
𝑥!
∗
: 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 𝑒𝑥𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛 𝑜𝑓
𝑚𝑎𝑟𝑘𝑒𝑟 𝑔𝑒𝑛𝑒 𝑥 𝑖𝑛 𝑠𝑎𝑚𝑝𝑙𝑒 𝑗
Wan, et al. (2019). bioRxiv, p.426593
NK Cell Markers CD8+ TEM Markers cDC2 Markers
Sample-level performance
37. Consistent cellular components between methods
Comparison1: Cibersortx,our signatures vs LM22 Comparison2: Cibersortx (reference based) vs
ICTD (referene free)
38. Cell type composition changes among T1D onset
*: p < 0.05, **: p < 0.01, ***: p < 0.001.
• In T1D onset vs Healthy
controls comparison:
consistently observed with
similar trend with our
study.
• Reduced pDC, NK,
Platelet cell proportion
trend.
• Increased CD8+ TEM,
Treg cell.
39. Increased proportion for B naive cell during T1D early onset
*: p < 0.05, **: p < 0.01, ***: p < 0.001.
T
1
D
4
m
o
n
t
h
s
(
n
=
1
9
)
T
1
D
1
m
o
n
t
h
(
n
=
1
9
)
T
1
D
O
n
s
e
t
(
n
=
4
3
)
H
C
(
n
=
2
4
)
Cell
Type
Proportion
H
C
(
n
=
5
)
T
1
D
(
n
=
5
)
0.00
0.05
0.10
0.15
Cell
Type
Proportion scRNA-Seq
Cell
Type
Proportion
H
C
(
n
=
2
4
)
T
1
D
(
n
=
8
1
)
T
2
D
(
n
=
1
2
)
HC vs T1D vs T2D HC vs T1D onset vs T1D
development
Cell
Type
Proportion
• Use of the anti-CD20 monoclonal antibody, rituximab, to deplete B cells in patients with newly
diagnosed T1DM preserved β cell function and delayed the requirement for insulin
administration following 1 year of treatment.
40. Moving toward clinics: Rituximab, B-Lymphocyte Depletion, and Preservation of β Cell
Function
• Randomized, double-blind clinical trial of anti-
CD20 monoclonal antibody, rituximab
• 87 participants (age 8-40 years of age) with
recent onset T1D (within 100 days of Stage 3
onset)
• Infusions of rituximab or placebo on days 1, 8,
15, and 22 of the study.
• Study endpoint: geometric mean area of the
AUC for the serum C-peptide level during the
first 2 hours of a mixed-meal tolerance test.
• C-peptide AUC was significantly higher, A1c
and insulin doses were significantly lower at
12 months in rituximab treated individuals
Pescovitz et al and the Type 1 Diabetes TrialNet Anti-CD20 Study Group. NEJM. 2009
41. Elevated T cell transcriptome levels predict poor clinical response following rituximab.
Peter S.Linsley, et al. (2019). Genes Immun.
Whole blood transcriptional signatures are being used to inform
clinical trial design
Based on this analysis,
TrialNet is initiating a trial
testing Rituximab followed by
treatment with CTLA4-Ig,
which is a co-stimulatory
modulatory drug.
CD19.mod GZMK.mod
CD2.mod CHD3.mod
Time (days)
Fraction
subjects
progression
42. • Consistent cellular components between methods.
• One phase II clinical trial testing the efficacy of the anti-
CD20 monoclonal antibody rituximab in T1D patients.
Deconvolution into existing clinical trials transcriptome profiles
Variable Rituximab Placebo
p-value
(difference with
previous study)
No. of subjects 37 17 ns
No. of samples 135 60
Gender (M/F) 25/12 11/6 ns
Age 22.2 + 8.3 21.3 + 9.0 ns
Initial C-peptide 0.80 + 0.43 0.89 + 0.40 ns
Responder (N/%) 24 (65%) 8 (47%)
43. Higher resolution in cell proportion changes after Rituximab treatment
Cell
Type
Proportion
Visit (weeks)
Based on the change in the
AUC of the C-peptide
response from baseline to 6
months, each participant was
designated as a C-peptide
responder or non-responder.
Infusion
44. • B cell subsets were reduced by
rituximab as expected.
• Linear mixed-effects mode
adjusted by age, gender
covariates:
• Non-responders has higher
proportion of CD4+ TCM, dnT
Treg cells.
• Responders had higher
proportion of neutrophils cell.
Cell
Type
Proportion
Visit (weeks)
Cell proportion changes after Rituximab treatment
*: p < 0.05, **: p < 0.01, ***: p < 0.001.
∗∗∗ ∗ ∗
∗
∗
45. Summary – Part II
• ScRNA-seq of PBMCs identified 31 distinct cell clusters. Heterogeneity was observed within each cell type. In
children with T1D, the proportion of CD4+ T central memory (TCM) cells was increased while the proportion of pDC,
platelets, and hematopoietic stem and progenitor cells (HSPC) were significantly reduced.
• NK cell, dendritic cell have shown early immune response in T1D onset, by differential expression pattern.
• By utilizing our scRNA-seq signature to deconvolute subpopulation from existing bulk RNA-seq, strong correlation
was observed between the methods CIBERSORTx and reference-free method ICTD for cell proportion estimates (P
< 2.2×10-16).
• The estimates into rituximab clinical trial data showed that B cells, CD4+TCM, Treg and neutrophils cell percentages
were significantly associated with response to rituximab.
• This approach allows for secondary analysis of clinical trial data to deconvolute bulk RNA sequencing data in
complex tissues in order to link cell-type specific immune signatures associated with drug responder status.
46. Future directions
• Refinement of single cell RNA-seq analysis with deeper interrogation
of the sub clusters within each cell type.
• Validation with flow-cytometry assays.
• Optimize deconvolution pipeline with investigation of additional clinical
trial datasets and validation of key findings.
- Characterize a baseline (pre-existing) immune cell proportion profile as a
prediction factor for diabetes onset or response to disease treatment.
- Determine longitudinal trajectory, and tracing how they (C-peptide AUC) change
over time.
47. Acknowledgements
Funding
NIDDK Information Network (dkNET)
New Investigator Pilot Program in
Bioinformatics
NIH Center for Diabetes and Metabolic Diseases
Pilot and Feasibility program
Indiana University
Carmella Evans-Molina
Yunlong Liu
Chuanpeng Dong
Chi Zhang
Tingbo Guo
Jing Liu
Farooq Syed
Chih-Chun Lee
University of Chicago
Raghavendra G. Mirmira
Pacific Northwest National Laboratory
Webb-Robertson, Bobbie-Jo M
Thank the donors and their families!