Transcript 2007_files/Helgason Proteomics ASCO2007

H.H. Helgason1,
J.Y. Engwegen2
M. Zapatka3, M. Kuiper1,
A. Cats1, H. Boot1
J.H. Beijnen2,4
J.H.M. Schellens1,4
Serum proteomic profiling in patients
with advanced gastric cancer receiving
first – line epirubicine, cisplatin and
capecitabine chemotherapy and
identification of disease specific and
predictive biomarkers
1. Dept. of Medical Oncology, Antoni van Leeuwenhoek Hospital/The Netherlands Cancer Institute, Amsterdam, The Netherlands
2. Dept. of Pharmacy, Slotervaart Hospital, Amsterdam, The Netherlands
3. Dept. of Theoretical Bioinformatics, German Cancer Research Center, Heidelberg, Germany
4. Faculty of Pharmaceutical Sciences, Utrecht University, Utrecht, The Netherlands
Introduction



Gastric cancer is a common cause of cancer deaths
Prognosis depends on stage at diagnosis making early
diagnosis mandatory
Specific serum biomarkers of gastric cancer, identified by
SELDI-TOF mass spectrometry, could be of great value
Methods - 1


Patients with advanced gastric cancer receiving 1st line
chemotherapy containing Epirubicine 50 mg/m2,
Cisplatin 60 mg/m2 i.v. on day 1 and Capecitabine 1000
mg/m2, orally bid on day 1 – 14 every 3 weeks
We collected serum prior to each chemotherapy cycle in
gastric cancer patients (GC) and randomly in matched
healthy controls (HC), according to standardized protocol
Methods - 2




Proteomic profiles were generated with SELDI-TOF mass
spectrometry, using CM 10 chips with weak cation
exchange moiety and buffer pH 5
We collected serum prior to each chemotherapy cycle in
gastric cancer patients (GC) and randomly in matched
healthy controls (HC), according to standardized protocol
Serum proteomic profiles of GC were compared to profiles
of matched HC allowing identification of serum
biomarkers gastric cancer
Mass spectrometry data were processed using the tbimass
R-package software (www.r-project.org)
Methods - 3



To assess the classification accuracy a tenfold repetition of
tenfold cross-validation with a nested threefold parameter
optimization loop was conducted
Proteomic profile survival prediction was analyzed with
COXpath, a predictor-corrector method for drawing the L1
regularization path for the Cox proportional hazards
model
Proteomic profile changes according to response and
survival and changes developing during chemotherapy
were analyzed
Results – 1
Patients
82
Age
57 years (range 34 – 74)
Male
57 (70 %)
Lymphadenopathy
87%
Liver metastasis
26%
Peritoneal metastasis
37%
Lung metastasis
9%
Overall survival
11 months (95% CI 9.5 – 12)
Complete response
5
Partial response
20
Stable disease
38 (18 > 6 months)
Not assessable
14
Progressive disease
5
Results – 1



To assess the classification accuracy a tenfold repetition of
tenfold cross-validation with a nested threefold parameter
optimization loop was conducted
We identified proteins that differentiated between gastric
cancer patients and matched healthy controls
Proteins with the highest intensity difference between
patients and healthy controls (Figure 1a and b, table 2)
serve as a candidate biomarkers for gastric cancer
Results – 2
Intensity
3892 Da
3892 Da
 GC
 HC
m/z
Figure 1a: Mean spectra of GC and HC; m/z 3892 Da
m/z
Results – 3
 GC
 HC
m/z
Figure 1b: Mean spectra of the two groups combined; m/z 3892 Da
Results – 3
Peak (kDa)
3.892
40.544
Table 2: Most important peaks
6.623
in the classification model
13.736
4.245
24.024
3.316
4.641
15.625
29.686
9.989
6.674
4.438
3341
Results – 4
Figure 2: Diagram of
frequency of correct
classification by SVM
in cross validation
Frequency
 By selecting the most differentiating proteins we built a
classification model that correctly classified 81% of the gastric
cancer patients and 90% of the healthy controls (Figure 2)
Results – 5
 There was some difference in classification quality according to
measurement day
 No proteins were found to be significantly predictive of survival
in gastric cancer patients
 Few proteins were found to correlate to response to ECC
chemotherapy
 Analysis of changes occurring in proteomic profiles of patients
with advanced gastric cancer during chemotherapy is ongoing
Conclusions




We identified a number of proteins that
differentiated between gastric cancer and
healthy controls and could serve as a candidate
biomarker for gastric cancer
We were able to correctly classify 81% of gastric
cancer patients and 90% of healthy controls
We identified few proteins that correlate to
response to 1st line palliative ECC chemotherapy
serving as potential predictive biomarkers
Serial analysis is ongoing but a number of
proteins change during chemotherapy according
to response
Background:
Gastric cancer is the fourth most commonly diagnosed cancer and is the second leading cause of
cancer death worldwide. Prognosis is highly dependent on stage at diagnosis making early diagnosis
mandatory. By using SELDI – TOF mass spectrometry we compared serum protein profiles of gastric
cancer (GC) patients with healthy controls (HC) and those of gastric cancer patients responding to firstline ECC chemotherapy with those with no response or early progressive disease.
Methods:
Serum from patients with advanced gastric cancer was obtained, according to a predefined schedule,
prior to start of first-line epirubicin (50 mg/m2 day 1), cisplatin (60 mg/m2 day 1) and capecitabine (1000
mg/m2 d1-14) chemotherapy (ECC) and serially before each treatment cycle every 3 weeks and
analyzed by SELDI-TOF MS/MS. Samples were drawn according to standardized protocol, centrifuged
within 1 hour and stored at -30°C. After thawing they were analyzed with SELDI-TOF MS/MS on CM10
chips at pH 5. Healthy control subjects were matched according to age, gender and time of serum
collection. All patients had given written consent, had WHO PS ≤ 2 and measurable disease, according
to RECIST criteria, and/or evaluable disease by clinical assessment or tumor marker. Serum proteomic
mass spectrometry data of GC patients and matched HC were processed using the tbimass R-package
(www.r-project.org) and compared. After pre-processing (resampling, baseline correction, normalisation,
alignment correction) support vector machines (SVM) and a specialised variable filtering procedure
based on the relative intensity variance were applied for classification. To assess the classification
accuracy a tenfold repetition of tenfold cross validation with a nested threefold parameter optimisation
loop was conducted. The number of variables used for classification was reduced by recursive feature
elimination (RFE). Furthermore we analyzed proteomic profile changes association with chemotherapy
response, gastric cancer patient survival and for changes developing during chemotherapy (ongoing).
For the serial analysis the time points T: 0, 6, 12 and 18 - 24 weeks were used.
Results:
In total 82 patients with adenocarcinoma of the stomach (mean age 57 years, male 70%) were treated
between 06/2003 and 11/2006 with mean 5 ECC cycles. Response rate was 37% (5 CR, 20 PR) with
additionally 12 patients developing stable disease of more than 6 months duration but 5 patients were
primarily progressive. Seven patients were not assessable because of additional surgery and other 7
patients were not evaluable. The median time to progression was 7 months (95% CI: 6 - 8). and median
overall survival 11 months (95% CI: 9.5 - 12). All patients were previously untreated except respectively
1 and 8 patients who had received chemo-radiotherapy and radiotherapy for proximal gastric carcinoma
(all > 6 months previously). Serum from 77 HC was collected. By comparing GC patients and HC we
identified number of proteins that differentiated between the two groups. By selecting the most
differentiating proteins we built a classification model that correctly classified 81% of the gastric cancer
patients and 90% of the healthy controls. Specific prognostic proteomic profile for gastric cancer
patients was not found. Few proteins were found to correlate to response serving as potential predicting
biomarkers. Proteomic profiling of serial serum collections during chemotherapy in GC is ongoing but
some proteins change according to response during chemotherapy serving as potential biomarker for
treatment monitoring.
Conclusion:
We identified number of proteins that differentiate between gastric cancer and healthy controls by
proteomic profiling with SELDI-TOF MS/MS. By using the most discriminating proteins we were able to
correctly classify 81% of the gastric cancer patients and 90% of the healthy controls serving as
candidate diagnostic biomarkers.