XXIV Riunione MITO Pisa 4 Dicembre 2014 Development of a

Department of Experimental Oncology and Molecular Medicine
Unit of Molecular Therapies
XXIV Riunione MITO
Pisa 4 Dicembre 2014
Development of a molecular predictor of disease
recurrance by MITO2 miRNA profiling
2
MITO2 miRNA profiling
Case material: 179 cases (of 226 profiled); hereafter OC179
Platform: Agilent SurePrint human miRNA arrays (mirBASE17.0)
Main aims:
a) Identification of groups of pts diverging from specific baselines
b) Identification of miRNA-related subgroups of patients
c) development of a prognostic model
Overall design
- training set:
OC179 from MITO2
- validation set1: INT-CRO series (microarray data); hereafter OC263
- validation set2: TCGA (microarray data ); hereafter OC452
2
ID
Case materials profiled for miRNA expression
case
type of
n°
material published
material samples
source
array
processing
miRNA
platform
Array
OC55
INT/CRO
Oncotarget
2011
frozen
55
INT-Milan
Illumina
Illumina
human v2
OC30
INT
Oncotarget
2011
FFPE
30
INT-Milan
Illumina
Illumina
human v2
OC45
INT
Oncotarget
2011
frozen
45
INT-Milan
Illumina
Illumina
human v2
OC133
CRO
no
frozen
133
INT-Milan
Illumina
Illumina
human v2
Agilent
Agilent
OC179
MITO2
no
FFPE
179
INT-Naples
INT-Milan
OC452
TCGA
Nature
2011
frozen
452
TCGA
SurePrint
miRNA 8x60K
16 plus
SurePrint
miRNA 8x15K
n° of
miRNA on miR-BASE
array
1146
(including
12
putative)
1146
(including
12
putative)
1146
(including
12
putative)
1146
(including
12
putative)
1512
17
723
10
2
Case materials analyzed for miRNA expression
Total number of cases profiled at INT-Milan:
- 263 cases on Illumina Platform from INT and CRO
- 179 cases on Agilent Platform from MITO2
In-silico Case Material analyzed
- 452 cases on Agilent Platform from TCGA consortium
Total: 894 cases “the greatest data set available for EOC miRNA profile”
Illumina data
Agilent data –Milan
Agilent data –TCGA
Data merging miRNA re-annotation:
706 probes detected corresponding to 581 miRNA
921 probes detected
661 probes detected
Re-annotation on miRBase21
385 unique miRNA shared among all studies
Batch effect Adjustment: the empirical Bayes (EB) method [Johnson, 2007]
2
Case materials analyzed for miRNA expression
Training set: OC179 from MITO2
Median PFS 22.83 months
Validation set1: OC263 from INT-CRO
Median PFS 16 months
Validation set2: OC452 from TCGA
Median PFS 16.84 months
2
Identification of groups of patients diverging from specific miRNA
baselines
A) Data deconvolution: definition of a
baseline group of patient with similar
clinical characteristics. Definition of
each patient “molecular distance”
from baseline
B) Dimensional shape recognition by topology networking
Colored by deviation from baseline:
Blue similar to baseline
Red different from baseline
2
Approach for baseline definition on OC179 MITO2 dataset
Stage III-IV patients with no residual disease (NED) and long PFS
(no relapse)
ID GF
(Milano)
Age (years)
FIGO
Grading
Histo
Residual
PFS Status
PFS Time
(months)
AP91
74
IV
G3
serous
none
0
73
AQ29
53
III
G3
serous
none
0
94
AR59
63
III
G3
serous
none
0
67
AQ67
58
III
Undif
Undif
none
0
100
AQ98
75
III
G3
serous
none
0
90
AP33
51
III
G3
serous
none
0
90
AR32
64
III
G3
serous
none
0
83
AR52
57
III
G3
serous
none
0
90
2
Baseline definition
Patients NED and with long PFS (no relapse)
Survival function
OC179 MITO2 dataset
A
B
C
P=0.0016
Time (Months)
Patients closer to baseline (A)
have similar good prognosis
2
EOC subtypes validation in independent datasets
OC263 –INT-CRO
OC452 –TGCA
P=1.87E-04
P=4.25E-03
2
Identification of OC Subtypes driven by miRNA expression
patterns on OC179 MITO2 dataset
Consensus matrix
OC179 MITO2 dataset
Silhouette Plot
P=0.000742
Cl1
Cl2
Cl3
Cl4
2
EOC subtypes validation in independent datasets
OC263 –INT-CRO
OC452 –TGCA
P=3.98E-14
P=0.00228
2
Patients prognosis is correctly identified by both miRNA-driven sub classification
Survival function
OC179 MITO2 dataset
OC179 MITO2 dataset
A
Cl1
Cl2
Cl3
Cl4
B
C
P=0.000742
P=0.0016
Time (Months)
ID GF
Age
Baseline Subtypes
(Milano)
(years)
AQ60
ArmC
Cl2
29
AR49
ArmC
Cl4
60
FIGO
Grading
III
III
NA
G3
AR50
ArmC
Cl4
51
III
G3
AR51
AR48
AR53
AR54
AR55
AR56
AR58
ArmC
ArmC
ArmC
ArmC
ArmA
ArmC
ArmC
Cl4
Cl4
Cl4
Cl4
Cl4
Cl4
Cl4
78
67
61
63
68
47
59
III
III
III
IV
III
IV
III
G2
G1
G3
G3
G3
G3
G3
AQ55
ArmC
Cl4
66
III
G2
Histo
Residual
serous
>1cm
serous
none
endomet
<1cm
roid
serous
>1cm
serous
<1cm
serous
<1cm
serous
<1cm
serous
>1cm
serous
<1cm
serous
<1cm
endomet
not operated
roid
0
1
PFS
(months)
36
12
1
12
1
1
1
1
1
1
1
9
4
8
12
35
27
20
1
5
PFS Status
2
miRNAs differentially expressed between arm A vs. Arm C patients
Survival function
OC179 MITO2 dataset
A
B
C
P=0.0016
Time (Months)
gene symbol
fold-change
A vs. C
P-value
FDR
hsa-miR-513b-5p
hsa-miR-200c-3p
hsa-miR-513a-5p
hsa-miR-141-3p
hsa-miR-200b-3p
hsa-miR-193a-3p
hsa-miR-21-5p
hsa-miR-429
hsa-miR-200a-3p
hsa-miR-15a-5p
hsa-miR-142-3p
hsa-miR-374a-5p
hsa-miR-195-5p
hsa-miR-205-5p
hsa-miR-199b-5p
hsa-miR-125a-3p
hsa-miR-148b-3p
hsa-miR-1225-5p
hsa-miR-135b-5p
hsa-miR-374b-5p
hsa-miR-486-5p
hsa-miR-188-5p
hsa-miR-203a-3p
hsa-miR-509-5p
hsa-miR-224-5p
hsa-miR-135a-5p
hsa-miR-514a-3p
24,52
14,02
8,13
20,65
6,7
0,3
0,47
6,37
6,43
0,48
0,21
0,38
0,35
15,08
0,31
2,02
0,43
3,24
3,41
0,45
0,4
2,35
3,66
3,24
0,33
2,91
3,37
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
0,0000001
0,0000002
0,0000005
0,0000011
0,0000044
0,000005
0,000006
0,000013
0,0000275
0,0000292
0,0001271
0,0001931
0,0003138
0,0005574
0,0022837
0,0029016
0,0052356
0,0434162
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
0,00000081
0,00000147
0,00000337
0,00000636
0,000021
0,0000225
0,0000256
0,0000526
0,000103
0,000103
0,000368
0,000539
0,000847
0,00129
0,00451
0,00547
0,00922
0,0596
2
How to develop a clinically useful
classifier?
E’ stato utilizzato un algoritmo che, sulla base dei
dati di PFS della casistica MITO2 (OC179) e della
relativa espressione dei 385 miRNAs rilevati, ha
costruito un modello in grado di stratificare le
pazienti ad alto e basso rischio di ricaduta.
Il modello contiene 35 miRNAs che dopo crossvalidazione (10-fold) mantengono il loro impatto
prognostico anche se con rilevanza diversa.
miRNAs la cui espressione è associata a prognosi
sfavorevole (score superiore al cut-off di algoritmo)
miRNAs la cui espressione è associata a prognosi
favorevole (score inferiore al cut-off di algoritmo)
Unique id
hsa-miR-193a-5p
hsa-miR-508-3p
hsa-miR-509-5p
hsa-miR-514a-3p
hsa-miR-506-3p
hsa-miR-507
hsa-miR-509-3p
hsa-miR-592
hsa-miR-29c-5p
hsa-miR-513b-5p
hsa-miR-513a-5p
hsa-miR-200c-3p
hsa-miR-141-3p
hsa-miR-200b-3p
hsa-miR-423-5p
hsa-miR-486-5p
hsa-miR-200a-3p
hsa-miR-23a-5p
hsa-miR-330-3p
hsa-miR-30b-3p
hsa-miR-484
hsa-miR-769-5p
hsa-miR-135b-5p
hsa-miR-100-3p
hsa-miR-99b-5p
hsa-miR-143-5p
hsa-miR-429
hsa-miR-151a-3p
hsa-miR-574-5p
hsa-miR-452-5p
hsa-miR-29a-5p
hsa-miR-195-3p
hsa-miR-890
hsa-miR-30d-5p
hsa-miR-193b-5p
p-value % CV Support Hazard Ratio
0,0000177
100
1,977
0,0000311
100
0,747
0,0000474
100
0,684
0,0000478
100
0,811
0,0000507
100
0,635
0,0000572
100
0,588
0,0000713
100
0,783
0,0001548
100
0,255
0,0007134
100
1,595
0,0007233
100
0,817
0,0007357
100
0,766
0,0015449
100
0,793
0,0016807
100
0,819
0,0026893
100
0,786
0,002895
90
1,765
0,0029908
90
1,345
0,0031706
100
0,808
0,0052072
80
1,641
0,0060584
80
1,856
0,0064133
100
1,983
0,0078602
80
1,6
0,008215
70
1,762
0,008942
80
0,851
0,0089818
90
1,958
0,0093801
70
1,35
0,0095842
80
1,674
0,0122341
60
0,835
0,013404
60
1,363
0,0161045
50
1,283
0,0174535
60
1,276
0,0179111
50
1,765
0,0186502
40
1,629
0,0231142
40
0,085
0,0233194
40
1,253
0,0240755
60
1,506
2 OC179 – MITO2 patients’ stratification according to the molecular classifier
P=6.83E-4
Sample size
Median PFS (months)
high risk
89
17.99
low risk
90
37.9
HR= 0.5463
95% CI = 0.3829 to 0.7795
2
Molecular classifier validation on independent datasets
OC452 TCGA
OC263 – INT-CRO
P=0.0045
P=1.33E-14
Sample size
Median PFS (months)
high risk
141
12
low risk
122
34
HR= 0.356
95% CI = 0. 267 to 0.476
Sample size
Median PFS (months)
high risk
283
15.2
low risk
169
18.7
HR= 0.72
95% CI = 0. 58 to 0.899
2
35 miRNA molecular classifier performance in defining patients’ prognosis
OC179 MITO2 dataset
Survival function
OC179 – MITO2
A
B
P=6.83E-4
C
P=0.0016
Time (Months)
Row Labels
High risk
Low risk
Grand Total
ArmA
38
65
103
ArmB
35
23
58
ArmC
10
10
2
35 miRNA molecular classifier performance in defining patients’ prognosis
OC179 MITO2 dataset
OC179 – MITO2
Cl1
Cl2
Cl3
Cl4
P=0.000742
Row Labels
High risk
Low risk
Grand Total
Cl1
55
44
99
P=6.83E-4
Cl2
6
44
50
Cl3
18
2
20
Cl4
10
10
2
miRNAs identified with different strategies
gene symbol
fold-change
A vs. C
P-value
FDR
hsa-miR-513b-5p
hsa-miR-200c-3p
hsa-miR-513a-5p
hsa-miR-141-3p
hsa-miR-200b-3p
hsa-miR-193a-3p
hsa-miR-21-5p
hsa-miR-429
hsa-miR-200a-3p
hsa-miR-15a-5p
hsa-miR-142-3p
hsa-miR-374a-5p
hsa-miR-195-5p
hsa-miR-205-5p
hsa-miR-199b-5p
hsa-miR-125a-3p
hsa-miR-148b-3p
hsa-miR-1225-5p
hsa-miR-135b-5p
hsa-miR-374b-5p
hsa-miR-486-5p
hsa-miR-188-5p
hsa-miR-203a-3p
hsa-miR-509-5p
hsa-miR-224-5p
hsa-miR-135a-5p
hsa-miR-514a-3p
24,52
14,02
8,13
20,65
6,7
0,3
0,47
6,37
6,43
0,48
0,21
0,38
0,35
15,08
0,31
2,02
0,43
3,24
3,41
0,45
0,4
2,35
3,66
3,24
0,33
2,91
3,37
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
0,0000001
0,0000002
0,0000005
0,0000011
0,0000044
0,000005
0,000006
0,000013
0,0000275
0,0000292
0,0001271
0,0001931
0,0003138
0,0005574
0,0022837
0,0029016
0,0052356
0,0434162
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
< 1e-07
0,00000081
0,00000147
0,00000337
0,00000636
0,000021
0,0000225
0,0000256
0,0000526
0,000103
0,000103
0,000368
0,000539
0,000847
0,00129
0,00451
0,00547
0,00922
0,0596
Unique id
hsa-miR-193a-5p
hsa-miR-508-3p
hsa-miR-509-5p
hsa-miR-514a-3p
hsa-miR-506-3p
hsa-miR-507
hsa-miR-509-3p
hsa-miR-592
hsa-miR-29c-5p
hsa-miR-513b-5p
hsa-miR-513a-5p
hsa-miR-200c-3p
hsa-miR-141-3p
hsa-miR-200b-3p
hsa-miR-423-5p
hsa-miR-486-5p
hsa-miR-200a-3p
hsa-miR-23a-5p
hsa-miR-330-3p
hsa-miR-30b-3p
hsa-miR-484
hsa-miR-769-5p
hsa-miR-135b-5p
hsa-miR-100-3p
hsa-miR-99b-5p
hsa-miR-143-5p
hsa-miR-429
hsa-miR-151a-3p
hsa-miR-574-5p
hsa-miR-452-5p
hsa-miR-29a-5p
hsa-miR-195-3p
hsa-miR-890
hsa-miR-30d-5p
hsa-miR-193b-5p
p-value % CV Support Hazard Ratio
0,0000177
100
1,977
0,0000311
100
0,747
0,0000474
100
0,684
0,0000478
100
0,811
0,0000507
100
0,635
0,0000572
100
0,588
0,0000713
100
0,783
0,0001548
100
0,255
0,0007134
100
1,595
0,0007233
100
0,817
0,0007357
100
0,766
0,0015449
100
0,793
0,0016807
100
0,819
0,0026893
100
0,786
0,002895
90
1,765
0,0029908
90
1,345
0,0031706
100
0,808
0,0052072
80
1,641
0,0060584
80
1,856
0,0064133
100
1,983
0,0078602
80
1,6
0,008215
70
1,762
0,008942
80
0,851
0,0089818
90
1,958
0,0093801
70
1,35
0,0095842
80
1,674
0,0122341
60
0,835
0,013404
60
1,363
0,0161045
50
1,283
0,0174535
60
1,276
0,0179111
50
1,765
0,0186502
40
1,629
0,0231142
40
0,085
0,0233194
40
1,253
0,0240755
60
1,506
2
The miRNA molecular classifier is an independent prognostic marker
Covariates:
35 miRNA model: above threshold cut-off vs. below
threshold cut-off
Stage:
III-IV vs. I-II
Grade:
3 vs. 1,2
Histology:
serous vs. others
Residual disease: >1cm vs. <1cm
2
MITO2 miRNA profiling: conclusions
Punti di forza:
 prima meta-analysis di miRNAs su EOC
 EOC dataset al momento più numeroso (n=894; OC179 da MITO2; OC263 da INT-CRO; OC452 da TCGA)
 meta-analysis su diverse piattaforme
 gli approcci “baseline” e subtyping individuano un gruppo di tumori (cluster4/Cl4) con prognosi molto
sfavorevole
 questo cluster si ritrova nei validation sets
Punti critici:
 annotazioni diverse tra piattaforme
 riduzione dei miRNA comuni
 continuo aggiornamento miRBASE
Take home message:
 analisi dei casi del clusterC/Cl4
 individuazione di miRNA per validazione funzionale/biologica ClusterC/Cl4
 quale gruppi di pazienti vanno utilizzati per la costruzione di un corretto modello
prognostico?????