Investigation of Protein Signatures Associated with ALS ......2726 2817 3686 0 500 1000 1500 2000...
1
2726 2817 3686 0 500 1000 1500 2000 2500 3000 3500 4000 1 Investigation of Protein Signatures Associated with ALS Disease Progression Using Data Independent Acquisition Mass Spectrometry 1 Department of Chemistry, 2 Department of Statistics, and 3 Department of Biological Sciences North Carolina State University, Raleigh, NC Allyson L. Mellinger 1 , Emily H. Griffith 2 , and Michael S. Bereman 1,3 Clinical Care Hindered by Subjective “Measurements” Salivation 4 Normal 3 Slight, but definite excess of saliva in mouth, may have nighttime drooling 2 Moderately excessive saliva, may have minimal drooling 1 Marked excess of saliva with some drooling 0 Marked drooling, requires constant tissue Dyspnea (shortness of breath) 4 None 3 Occurs when walking 2 Occurs with one or more: eating, bathing, dressing 1 Occurs at rest, either sitting or lying 0 Significant difficulty, considering mechanical support ALS Functional Rating Scale: Current gold standard for evaluating therapeutic efficacy in clinical trials Data Dependent vs Data Independent Acquisition for CSF Analysis Longitudinal Proteomic Analysis of ALS Patient CSF Data Quality Control and Patient Variability Linear Mixed - Effects Modeling of Disease Progression LASSO and Forward Selection for Model Building Conclusions and Future Work References and Acknowledgements Peptide Gene Estimate p-value EHIVDLEMLTVSSIGTFR EFEMP1 5.49 0.098 AGLAASLAGPH SOD3 3.87 0.041 ESNPHCGSDGQTYGNK SPINK6 2.23 0.228 EVDSGNDIYGNPIK TIMP2 2.01 0.322 ISVTQGASK FCGBP 1.72 0.044 VPTADLEDVLPLAEDITNILSK GC 1.63 0.006 YVMLPVADQDQCIR HP -1.65 0.111 IGEWIGR C8orf34 -1.8 0.078 EVEELMEDTQHK DKK3 -2.45 0.412 STTAVVTNPK TTR -3.97 0.160 FTFHADICTLSEK ALB -5.32 0.012 IISVELPGDAK RELN -7.87 0.003 Model Predicted Score 30 40 30 40 20 Model Observed Score Fitted Residuals 0 -2.5 2.5 30 40 300 320 340 360 380 400 0 10 20 30 Minimum AIC # Variables in Model AIC Negative Control Model Number of Peptide IDs Coefficient of Variation (%) 0 25 50 100 75 Q1 Q2 Q3 Q4 DDA DIA DIA- Pan Human Library DIA- CSF Library DDA DDA 53.9% DIA 97.5% Log 2 (Peptide Abundance) 0 30 20 * 10 QC Patients by Color p = 0.038 p = 0.074 p = 0.031 Log(Percent Change from Baseline) Rate ( Change in ALSFRS Score Change in months ) Targeted Global Inflammation Index Pooled disease cohort sample for digestion QC Day 1 Day 2 Day 3 Filter aided tryptic digestion of 50 μg protein per sample Analysis by nLC- HRAM-MS/MS Pooled for chromatogram library 3 μL each digested ALS sample Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease that destroys both upper and lower motor neurons. As a result, patients experience progressive voluntary muscle action loss and fatal respiratory failure within a few years of onset. Only 10% of cases are considered hereditary. There is no known cure for this disease, and the discovery of any disease reversing treatment is hampered by a lack in pathogenic understanding of the disease along with the absence of an objective and linear measure of disease progression. The ALS-Functional Rating Scale (ALS-FRS) (Figure 1) is currently the most widely used clinical measure. This 12-question and 48-point survey assesses one’s ability to perform everyday activities. A quantitative measurement is sorely needed. In this work, we investigate the efficacy of peptide measurements in cerebrospinal fluid (CSF) of ALS to be accurate signatures in modeling disease progression. Figure 1. Left: Example question from the ALS-FRS. Survey results in subjective measurements that are not linearly related to disease severity. Above: Scatterplot of scores from ALS patients in this study. Some scores measured did not decrease linearly over time (circled). Figure 2. Comparison of DDA and DIA for CSF proteomics. A: Bar graph comparing the average # of peptides identified with each method. B: Venn Diagram comparing average peptide identifications of four replicate analyses using each method. C: Boxplots comparing precision of peptide measurements made in each quantile of abundances using each method. A B Figure 4. A: Boxplots of logged peptide abundances measured in every sample to visualize data quality. One outlier sample is clear*, and this patient was excluded from most subsequent analyses. B: Principal component analysis separating peptide abundances of samples. C: Boxplots of variation both within and among patient and QC samples. A Figure 3. Experimental design for analysis of longitudinal CSF. 63 samples and one digestion quality control sample each day were split between three days of digestion and randomized before analysis sequencing within digestion day. Gas-phase fractionated injections for chromatogram library 1 building were collected in the middle of the experiment.. Longitudinal CSF Collection from ALS Patients × 23 × 23 × 17 + + + Figure 5. Global inflammation changes were studied by taking the median peak of peptides from inflammatory proteins and calculating the percent change in those medians from each patient’s baseline collection visit. This was plotted against the rate of disease progression in each patient, defined as the ratio of the change in score over the change in time between collections. • We have shown the efficacy of proteomics in longitudinal sampling to quantitatively model disease progression in ALS patients. • A targeted study of peptides used in the model within an independent cohort of ALS patients should be investigated for further validation of this study. Chromatogram Library and DIA data search Strenuous signal filtering criteria for eligible peptides Filter colinear peptides LASSO for variable selection Multivariate Mixed Effects Longitudinal Model Forward selection to minimize AIC Peptide Filtering Final Model: ALSFRS Score=βX i +Z i b i +ε i A Years from Symptom Onset ALSFRS Score 20 30 40 3 6 9 12 Subjective Non-linear Figure 6. A: Summary of model building workflow B: Forward selection to build model and minimize AIC. C: Diagnostic plots for model fit and variance assumption Figure 7. A: 12 peptides with largest coefficients in model. B: Interaction network of proteins represented by peptides in the model. Peptide CV (%) 75 100 50 25 0 QC We gratefully acknowledge the NEALS Biorepository for providing all the biofluids from ALS patients used in this study. We also acknowledge the ALS Association (grant #19-SI-458) for the funding of this project. We acknowledge the MacCoss laboratory, University of Washington, for guidance and provision of biofluids for quality control. All mass spectrometry measurements were made in the Molecular Education, Technology, and Research Innovation Center (METRIC) at NCSU. (1) Searle, B., et al., Nat. Comm., 2018, 9, 5128. B A C C B A Log(Centered Abundance) Patient 1 PC1 (8.47%) PC2 (4.91%) QC Samples B 28 30 32 Patient 1 C DDA DIA
Transcript of Investigation of Protein Signatures Associated with ALS ......2726 2817 3686 0 500 1000 1500 2000...
2726 2817 36860
500
1000
1500
2000
2500
3000
3500
4000
1
Investigation of Protein Signatures Associated with ALS Disease Progression Using Data Independent Acquisition Mass Spectrometry
1Department of Chemistry, 2Department of Statistics, and 3Department of Biological Sciences North Carolina State University, Raleigh, NC
Allyson L. Mellinger1, Emily H. Griffith2, and Michael S. Bereman1,3
Clinical Care Hindered by Subjective “Measurements”
Salivation
4 Normal
3 Slight, but definite excess of saliva in mouth,
may have nighttime drooling
2 Moderately excessive saliva, may have
minimal drooling
1 Marked excess of saliva with some drooling
0 Marked drooling, requires constant tissue
Dyspnea (shortness of breath)4 None
3 Occurs when walking
2 Occurs with one or more: eating, bathing,
dressing
1 Occurs at rest, either sitting or lying
0 Significant difficulty, considering mechanical
support
ALS Functional Rating Scale: Current gold standard
for evaluating therapeutic efficacy in clinical trials
Data Dependent vs Data Independent Acquisition for CSF Analysis
Longitudinal Proteomic Analysis of ALS Patient CSF
Data Quality Control and Patient Variability
Linear Mixed-Effects Modeling of Disease Progression
LASSO and Forward Selection for Model Building
Conclusions and Future Work
References and Acknowledgements
Peptide Gene Estimate p-value
EHIVDLEMLTVSSIGTFR EFEMP1 5.49 0.098
AGLAASLAGPH SOD3 3.87 0.041
ESNPHCGSDGQTYGNK SPINK6 2.23 0.228
EVDSGNDIYGNPIK TIMP2 2.01 0.322
ISVTQGASK FCGBP 1.72 0.044
VPTADLEDVLPLAEDITNILSK GC 1.63 0.006
YVMLPVADQDQCIR HP -1.65 0.111
IGEWIGR C8orf34 -1.8 0.078
EVEELMEDTQHK DKK3 -2.45 0.412
STTAVVTNPK TTR -3.97 0.160
FTFHADICTLSEK ALB -5.32 0.012
IISVELPGDAK RELN -7.87 0.003
Model Predicted Score30 40
30
40
20
Mo
de
l O
bs
erv
ed
Sc
ore
Fitted
Re
sid
ua
ls
0
-2.5
2.5
30 40
300
320
340
360
380
400
0 10 20 30
Minimum AIC
# Variables in Model
AIC
Negative Control
Model
Nu
mb
er
of
Pe
pti
de
ID
sC
oe
ffic
ien
t o
f V
ari
ati
on
(%
)
0
25
50
100
75
Q1 Q2 Q3 Q4
DDA
DIA
DIA-
Pan
Human
Library
DIA-
CSF
Library
DDADDA
53.9%
DIA
97.5%
Lo
g2(P
ep
tid
e A
bu
nd
an
ce
)
0
30
20
*
10 QC
Patients by Color
p = 0.038
p = 0.074
p = 0.031
Log(Percent Change from Baseline)R
ate
(C
ha
ng
ein
AL
SF
RS
Sc
ore
Ch
an
ge
inm
on
ths
)
Targeted Global Inflammation Index
Pooled
disease
cohort
sample
for
digestion
QC
Day 1
Day 2
Day 3
Filter aided
tryptic
digestion of
50 µg
protein per
sample
Analysis by nLC-
HRAM-MS/MS
Pooled for chromatogram library3 µL each digested ALS sample
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease that destroys both upper and
lower motor neurons. As a result, patients experience progressive voluntary muscle action loss and
fatal respiratory failure within a few years of onset. Only 10% of cases are considered hereditary.
There is no known cure for this disease, and the discovery of any disease reversing treatment is
hampered by a lack in pathogenic understanding of the disease along with the absence of an
objective and linear measure of disease progression. The ALS-Functional Rating Scale (ALS-FRS)
(Figure 1) is currently the most widely used clinical measure. This 12-question and 48-point survey
assesses one’s ability to perform everyday activities. A quantitative measurement is sorely needed.
In this work, we investigate the efficacy of peptide measurements in cerebrospinal fluid (CSF) of
ALS to be accurate signatures in modeling disease progression.
Figure 1. Left: Example question from the
ALS-FRS. Survey results in subjective
measurements that are not linearly related
to disease severity. Above: Scatterplot of
scores from ALS patients in this study.
Some scores measured did not decrease
linearly over time (circled).
Figure 2. Comparison of DDA
and DIA for CSF proteomics.
A: Bar graph comparing the
average # of peptides identified
with each method. B: Venn
Diagram comparing average
peptide identifications of four
replicate analyses using each
method. C: Boxplots comparing
precision of peptide
measurements made in each
quantile of abundances using
each method.
A B
Figure 4. A: Boxplots of logged peptide abundances measured in every sample to visualize data quality.
One outlier sample is clear*, and this patient was excluded from most subsequent analyses. B: Principal
component analysis separating peptide abundances of samples. C: Boxplots of variation both within and
among patient and QC samples.
A
Figure 3. Experimental design for analysis of longitudinal CSF. 63 samples and one digestion quality
control sample each day were split between three days of digestion and randomized before analysis
sequencing within digestion day. Gas-phase fractionated injections for chromatogram library1 building were
collected in the middle of the experiment..
Longitudinal
CSF Collection
from ALS
Patients
× 23
× 23
× 17
+
+
+
Figure 5. Global inflammation changes
were studied by taking the median peak of
peptides from inflammatory proteins and
calculating the percent change in those
medians from each patient’s baseline
collection visit. This was plotted against
the rate of disease progression in each
patient, defined as the ratio of the change
in score over the change in time between
collections.
• We have shown the efficacy of proteomics in longitudinal sampling to quantitatively model
disease progression in ALS patients.
• A targeted study of peptides used in the model within an independent cohort of ALS patients
should be investigated for further validation of this study.
Chromatogram Library and DIA
data search
Strenuous signal filtering criteria
for eligible peptides
Filter colinear peptides LASSO for variable selection
Multivariate Mixed Effects
Longitudinal Model
Forward selection to
minimize AIC
Peptide Filtering
Final Model: ALSFRS Score=βXi+Zibi+εi
A
Years from Symptom Onset
AL
SF
RS
Sc
ore
20
30
40
3 6 9 12
Subjective
Non-linear
Figure 6. A: Summary of model building
workflow B: Forward selection to build
model and minimize AIC. C: Diagnostic
plots for model fit and variance
assumption
Figure 7. A: 12 peptides with largest coefficients in
model. B: Interaction network of proteins
represented by peptides in the model.
Pe
pti
de
CV
(%
)
75
100
50
25
0QC
We gratefully acknowledge the NEALS Biorepository for providing all the biofluids from ALS patients used in this study. We also
acknowledge the ALS Association (grant #19-SI-458) for the funding of this project. We acknowledge the MacCoss laboratory,
University of Washington, for guidance and provision of biofluids for quality control. All mass spectrometry measurements were
made in the Molecular Education, Technology, and Research Innovation Center (METRIC) at NCSU.
(1) Searle, B., et al., Nat. Comm., 2018, 9, 5128.
B
A
C
C
BA
Log(Centered Abundance)
Patient 1
PC1 (8.47%)
PC
2 (
4.9
1%
)
QC Samples
B
28 30 32
Patient 1
C
DDA
DIA
![Waters Expression High Definition Proteomics System · [application note] Determination of Quantitative Protein SignatureS for Ductal carcinoma (BreaSt cancer) By lc/mS Proteome analy](https://static.fdocuments.us/doc/165x107/5fa6ef0c7dde475dd1110ce8/waters-expression-high-definition-proteomics-system-application-note-determination.jpg)
