hepa c protease gene expression analysis to predict
TRANSCRIPT
Hepa�c Protease Gene Expression Analysis to Predict Advanced FibrosisSophie Cazanave,1 Maciej Pacula,1 Eric Huang,1 Gabe Kwong,2 Jay Luther,3 Raymond Chung,3 Kathleen E. Corey,3
1Glympse Bio, Cambridge 2Wallace H. Coulter Department of Biomedical Engineering, Georgia Tech College of Engineering and Emory School of Medicine, Atlanta, GA 3Department of Medicine, Massachuse�s General Hospital, Harvard Medical School, Boston, MA, USA.4Transla�onal & Clinical Research Ins�tute, Faculty of Medical Sciences and Newcastle NIHR Biomedical Research Centre, Newcastle upon Tyne Hospitals NHS Founda�on Trust, Newcastle upon Tyne, UK
Glympse Bio, Inc.35 CambridgePark DriveSuite 100Cambridge, MA 02139
Introduc�on♦♦
––
–
Glympse Bio is developing an injectable mixture of protease ac�vity sensors (Glympse Bio Test System (GBTS)-NASH) that:
Proteases are important in the progression of fibrosis in NASH¹,²,³
Specifically detects the ac�vity of proteases dysregulated in human NASH⁴Is safe for human use [for more informa�on, see poster “Safety and Tolerability in Healthy Volunteers of the Glympse Bio Test Sytem-NASH Diagnos�c”, poster 1550]Predicts NASH with fibrosis severity and monitor Drug treatment response in animal models of NASH4,5
Glympse Bio Test System (GBTS)-NASH Diagnos�c
(I) Protease activity sensors are peptides with a mass barcode and protease substrate conjugated to an 8-arm polyeth-ylene glycol (PEG) carrier. Multiple protease sensors are mixed together in a cocktail for i.v. injection; individual protease sensors report on distinct protease activity by pairing each different protease substrate with a specific mass barcode(II) On i.v. injection, the protease sensors are cleaved by liver proteases implicated in various pathways of NASH progres-sion (III) Mass barcodes liberated from the PEG carrier by peptide substrate proteolysis are small and return to the circulation, where they are enriched by the kidneys into the urine. Urine enriched with mass barcodes is quantified by liquid chroma-tography–tandem mass spectrometry (LC-MS/MS) and input into a classifier that reports disease status or regression rate
FAP
PLAU
ST14
CTSK
******
*********
******
FAPMMP9MMP7
MMP19PLAU
MMP2ST14
GZMKMMP14
GZMACTSK
FURINCTSD
0 1.0
Log2 Fold Change (RNA)
2.0
NCL cohortF2+ vs F0-F1
Gen
e
FAP
MMP9MMP7
MMP19PLAU
MMP2
ST14
GZMK
MMP14
GZMA
CTSK
FURIN
CTSD
0.0 0.1
Spearman Correlation with Fibrosis
0.2 0.3 0.4 0.5
Gen
e
p>0.05p≤0.05
MMP2
Log
(1+
coun
ts)
-2-101234
NAFL F0/F1 F2 F3 F4NASH
****
A B C
NAFL F0/F1 F2 F3 F4NASH
Log
(1+
coun
ts)
-2-101234
FAP
*** p<0.001
** p<0.01
****
False Positive Rate0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
True
Pos
itive
Rat
e
Train NCL > Test NCL
RNA F2+ (n=13) AUC=0.85 (0.79- 0.91)FIB4 AUC=0.64 (0.55- 0.72)
0.0
0.5
1.0
Frac
tion
of p
redi
ctio
ns
False positiveFalse negative
Correct
F0-F1
PREDICTED
F0-F1
59
23 82
15
NASH F2+
NASH F2+H
ISTO
LOG
Y
Train classi�er
RNA F2+Classi�er
Training Set
Test Set
NCL cohort
80% training and 20% validation
A B
F0 F2 F3 F4F1Histology Fibrosis Stage
NCL cohort
FIB4or
False Positive Rate0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
True
Pos
itive
Rat
e
n=13
0.0 0.2 0.4 0.6 0.8 1.00.0
0.2
0.4
0.6
0.8
1.0
MGH individual AUCF2+ vs F0-F1
NCL
indi
vidu
al A
UC
F2+
vs F
0-F1
signi�cantnon-signi�cant
Train MGH -> Test NCL AUC=0.86 (0.80-0.91)
Train NCL -> Test MGH AUC=0.90 (0.85-0.94)
r2=0.72p=5.1e-34
MMP9MMP2
FAP
PLAU MMP19
ST14
MMP14MMP7
CTSKCTSD
GZMKGZMA
FURIN
RNA F2+ Classi�er
Train classi�er
RNA F2+Classi�er
NCL cohort
MGH cohort
Training Set
Test Set
MGH cohort
NCL cohort
RNA F2+ classi�er-naively applied
or
or
Healthy NASH Steatosis
proteases
Healthy (n=76) NAFL (n=90)NASH F0 (n=74)F1 (n=62)F2 (n=34)F3 (n=13)
F4 (n=6)
NanoString on 76 Healthy + 279 NAFLD
human biopsies(MGH cohort)
MGH human NAFLD cohort
Pooled urine collec�on
Injec�on of GBTS-NASH mul�plexedprotease ac�vity sensors
Mass spec analysis Fibrosis Classifica�onDrug and Diet-induced
regression Predic�on (AUROC>0.9)
Inte
nsity
Retention time
Glympse Bio Diagnos�c Pla�orm in Animal Models
Glympse Bio pipeline focused on transcriptomic analysis of human NAFLD liver biopsies to iden�fy protease targets based on differen�al expression
Glympse Bio has previously trained a mul�-gene protease classifier (with 13 proteases) using NanoS-tring technology predic�ve with high accuracy of fibrosis stage above 2 (F2+) in a large cohort of adults with NAFLD and obesity from Massachuse�s General Hospital (MGH)
♦
♦
False Positive Rate0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
True
Pos
itive
Rat
e
MGH cohort
0 0.5 1.0Log2 Fold Change (RNA)
GZMK
MMP7
CTSK
ST14
MMP2PLAU
CTSDMMP14
MMP9
GZMA
MMP19
FAP
FURIN
Gen
e
1.5 2.0
F2+ vs F0-F1
******
*********
******
******
******
******
RNA F2+ Classi�er
n=13 proteases AUC=0.90 (0.85-0.95)(spe=83.1%, sens=83.0%)
MGH cohort
*** p<0.001
Glympse Bio Test System (GBTS)-NASH
injectable multiplex of protease activity sensors
Selec�on of NASH target proteases
in Pa�ents with NASH: a Cross-valida�on Study
Objec�vesWe sought to cross-validate our previously reported hepa�c protease gene expression signature with an independent cohort of NAFLD pa�ents obtained from Newcastle University in the United Kingdom (NCL). The ini�al pa�ent samples (MGH) were collected from a bariatric surgery clinic while the NCL samples were obtained as part of a study on the natural history of NAFLD. These poten�al differences in popula�on distribu�on and clinical characteris�cs allow us to assess the generalizability of our NASH protease signature
♦
Methods
Results
NanoString protease gene expression quan�ta�on was performed on RNA extracted from formalin fixed and paraffin embedded (FFPE) and frozen liver �ssue from 146 NAFLD pa�ents from NCL includ-ing 33 NAFL and 113 NASH (5 F0, 29 F1, 24 F2, 46 F3, 9 F4). In the NASH group, longitudinal liver sam-ples were collected twice for 21 pa�ents, 3 �mes for 3 pa�ents and 4 �mes for 2 pa�ents with a median �me between biopsies of 5.9 years (min-max range of 1.2 - 14.6 year) for a total of 179 liver samples. Gene expression from 206 proteases that overlapped between NCL and MGH NanoString panels were used for transcriptomic analysis
♦
All biopsies were read by the same blinded pathologist to improve scoring accuracy and histological fi-brosis score was applied to stage disease (Brunt criteria)Binary classifica�on of fibrosis stage F2+ versus F0–F1 with a regularized logis�c regression classifier trained either with RNA F2+ gene classifier (comprising a subset of 13 NASH proteases normalized-gene counts) or calculated FIB4 score using a randomized 80% training and 20% valida�on sets (100 rounds of cross-valida�on)
Compared to the MGH cohort (BMI=46.0 ± 7.5), pa�ents from the NCL cohort have a lower average BMI of 34.1 ± 6.0 (p≤0.001) closer to the expected NASH popula�on weight
♦
♦
Other clinical parameters such as age, ALT and AST were also significantly increased in the NCL cohort (p≤0.001)
Mean ± SEM (Min-Max)
Mean ± SEM (Min-Max)
p value
Age 44.50 0.65 (16-74) 55.71 ± 0.92 (25-75) 1.69e-19BMI 46.08 ± 0.40 (26-69) 34.06 ± 0.50 (22.79-58.82) 1.49e-51
ALT (UI/L) 40.08 ± 1.63 (8-288) 68.45 ± 4.20 (16.00-328.00) 9.12e-14
AST (UI/L 27.15 ± 1.16 (0-202) 47.33 ± 2.39 (10.00-171.00) 2.05e-16Number of pa�ents Number of pa�ents
Weight Loss Surgery (Yes:No) 339:16 0:146NAS (0:1:2:3:4:5:6:7) 77:74:49:37:34:44:28:12 0:6:10:22:29:41:24:14
Fibrosis (Sirius Red) (0:1:2:3:4) 240:62:34:13:6 33:34:24:46:9
Steatosis (0:1:2:3) 77:150:113:15 0:41:53:52
Lobular Inflamma�on (0:1:2:3) 210:84:49:12 16:75:54:1
Ballooning (0:1:2) 167:124:64 28:70:48
Diabete (Yes:No:Unknown) 117:238:0 59:84:3Gender (Female:Male:Unknown) 260:95:0 95:60:1Ethnicity (Hispanic :Non-Hispanic:Unknown) 26:291:38 0:146:0Sample type (Core Needle Biopsy:Wedge Biopsy) 265:90 146:0
♦
♦
Comparison of Massachuse�s General Hospital (MGH) and Newcastle University (NCL) pa�ent clinical and biochemical characteris�cs
Hepa�c transcript levels of a 13-protease panel classify NASH F2+ pa�ents with high accuracy across two independent data sets despite significant differences in clinical features including BMI, age, ALT and ASTThe clinical efficacy of proteases for fibrosis staging in NASH warrants further studies
♦
♦
NASH proteases gene expression is also dysregulated in NCL NASH F2+ pa�ents and correlates with fibrosis severity
Expression levels of 7 out of the 13 NASH proteases originally iden�fied as upregulated in the MGH cohort were also significantly increased in F2+ NASH pa�ents vs F0-F1 in the NCL cohort (panel A)Expression levels of 11 NASH proteases posi�vely correlated with fibrosis stage (Spearman correla-�on r=0.25-0.5, padj ≤0.05) (panel B) with the top 2 most correlated proteases MMP9 and FAP gradu-ally increasing with fibrosis score (panel C)
A B
RNA F2+ classifier trained on NCL pa�ents discriminated advanced fibrosis (F2+) with high accuracy
A binary classifier trained on the same 13-protease panel iden�fied from MGH cohort accurately pre-dicted F2+ from F0–F1 in the NCL cohort (AUROC=0.85; 95% CI 0.79-0.91) (Sens=78.1%, Spe=79.7%) (panel A) and correctly assigned the histological score of 141 out of 179 samples (panel B) FIB-4 index underperformed the protease RNA classifier to classify NASH F2+ with an AUROC of 0.64 (95% CI 0.55-0.72, p≤0.0001) (panel A) (Low cut-off value of 1.45: Sens=55.4%, Spe=64.2%, high cut-off value of 3.25: Sens=7.9%, Spe=100%)RNA F2+ classifier can discriminate F2+ from F0-F1 across two independent data sets
The ability to predict F2+ pa�ents did not depend on which dataset was used for RNA classifier train-ing, as a classifier that was trained on the MGH and tested on the NCL cohort (panel B, green line, AUROC=0.86, 95% CI 0.80-0.91, Sens= 81%, Spe= 78.4%) was sta�s�cally equivalent to a classifier trained on the NCL and tested on the MGH cohort (black line, AUROC=0.90, 95% CI 0.85-0.94, Sens= 83%, Spe= 86%; p=0.49–0.54).
The performance of single proteases to dis�nguish F2+ vs F0–F1, as quan�fied by AUROC analysis, strongly correlated between the two cohorts (Pearson r2=0.72, p=5.1e-34) (panel A)
Ann Daly,4 Quen�n M. Anstee,4 Olivier Govaere,4 and Wendy Winckler1
BMI: Body Mass Index; NAS: NAFLD Ac�vity Score
Conclusions
References: 1Jiao et al. Curr Opin Gastroenterol 2009; 2Rockey et al. N Engl J Med 2015; 3Pellicoro et al. Nat Rev Immunol 2014; 4Cazanave S, et al. J Hepatol 2019;70(Suppl):e760; 5Cazanave S, et al. Hepatology 2019; 70, S1:1028A Acknowledgement: This study was funded by Glympse Bio, Inc.The Digital American Associa�on for the Study of Liver Disease, 13-16 November 2020
±
♦
♦
♦
♦
♦
♦
1516
MGH NCL
Massbarcode
Proteasesubstrate
×
PEGcarrier
(I)
Protease sensorcocktail
Blood vessel Disease site(II)
(III)
m/zMass-encoded signature
Inte
nsity