AACR 2005, the sequel:
Yahoo MB managed to lose the message referenced in the last post, so I'm going to archive it here and cross-post it there instead. Here are what I think are the most notable abstracts from AACR:
>>Abstract Number: 4806
Presentation Title: Serum proteomic fingerprintings as a novel reference pattern to distinguish cancer from non-cancer in early stage and monitor cancer prognosis: A multi center three-year study in China
Presentation Start/End Time: Tuesday, Apr 19, 2005, 1:00 PM - 5:00 PM
Category: CH03-07 Proteomic pattern diagnostics and early detection
Author Block: Yang Xu, Shangxian Gao, Xiaohang Zhao, Ning Li, Zhenhua Yang. National Clinical Proteomic Coordination Center, Beijing, China, State Food and Drug Administration, Beijing, China, Chinese Academy of Medical Sciences, Beijing, China, Ciphergen Biosystems Inc., Beijing, China, Chinese Committee for Clinical Laboratory Standards, Beijing, China
Objective: Cancer is becoming the number one killer in China now. The low specificity and sensitivity of the current cancer markers makes it not an ideal biomarker for the detection of cancer. In this study, we have summarized three-year multi-center study of surface-enhanced laser desorption/ionization (SELDI) time-of-flight mass spectrometry (TOF-MS) to differentiate cancer from non-cancer controls in China.
Method: Serum samples were collected from ten types of carcinoma confirmed by pathology under 4oC and distributed into five 100-uL aliquots with 9 M Urea and /or stored frozen at - 80 oC immediately. Samples were thawed, added to ProteinChip® and analyzed by SELDI-TOF-MS. Before each round of the analysis, we routinely performed standard quality control of serum samples to normalize different assays and different instruments so that all proteomic spectra are remained relatively constant like EKG through standardization of SELDI-TOF-MS in all centers (Kang XX, Xu Y, Wu XY et al, Clin. Chem. published November 18, 2004 online).
Result: The analysis using differentially expressed peptides and proteins resulted in a predictive equation which correctly distinguished cancer from all non-cancer patients in blinded tests with a sensitivity and a specificity as following: hepatocellular carcinoma (91% and 89%), lung carcinoma (82% and 95%), gastric carcinoma (91% and 94%), esophageal carcinoma (84% and 91%), prostate carcinoma (83% and 97%), breast carcinoma (93% and 91%), colon carcinoma (83% and 92%), pancreatic carcinoma (82% and 85%), ovarian carcinoma (99% and 99%) and nasopharyngeal carcinoma (92% and 97%).
Conclusion: Based on the review for the three-year (2002-2004 year) studies in China, we found that the specific cancer type using hydrophobic, anionic, cationic, and metal binding ProteinChip has little difference to affect specificity (10-15% variation) for the cancer detection. In addition, anionic ProteinChip has got the best result for the early detection of all types of cancers. Therefore, using one type of anionic ProteinChip as the standard protocol to set up Chinese cancer fingerprinting graphic bank is necessary. Due to existing of regional and race differentiation in China, we are planning to set up national Chinese cancer fingerprinting graphic bank for future diagnosis and prognosis of cancer. The proteomic fingerprintings are also currently under evaluation to determine whether it is suitable for early diagnosis of cancer and/or prognostic follow-up for the cancer. We found that none of the serum proteomic fingerprinting graphics is similar for all types of carcinoma tested. Therefore, serum proteomic fingerprinting graphics could be used as reference range for different types of carcinoma like EKG for cardiovascular disease. Protein identification of these potential markers is in progress.<<
Here we find the team took some pains to handle samples well and uniformly, and to normalize and calibrate the instruments. The national bank of protein fingerprints is interesting, and should help Ciphergen commercially there.
>>Abstract Number: 4793
Presentation Title: Low molecular weight proteome for biomarker discovery in human serum
Presentation Start/End Time: Tuesday, Apr 19, 2005, 1:00 PM - 5:00 PM
Category: CH03-02 High throughput proteomics
Author Block: Xiaodong Yu, Xavier Thomas, Daniel W. Chan, Jinong Li. Johns Hopkins University, Baltimore, MD, Edouard Herriot Hospital, Lyon, France
One of the revealing features of the plasma/serum proteome is its molecular weight distribution versus relative abundance of proteins. More than half of the proteins in plasma/serum are smaller than the presumed size cutoff of the kidney filtration apparatus (45 kD) and most of the high-abundance proteins have high molecular weights (e.g., albumin, 66 kD; IgG 140 kD; transferrin, 78 kD). Because of this unique distribution of molecular weight and abundance, removal of high molecular weight proteins also removes a large majority of high abundance proteins. We used this feature as a strategy for serum fractionation to facilitate biomarker discovery in low abundance and low molecular weight proteome in human serum.
In the first phase of this study, a standardized preparation of pooled human serum was used to establish optimized conditions using Millipore YM50 centrifugal ultrafiltration filters (50 kDa filter). Applicability of this method on high-throughput clinical proteomics study was evaluated on depletion efficiency of high molecular weight proteins, recovery efficiency of low molecular weight proteins, and most importantly, consistency on performance of different filters. Recovery of low molecular weight proteins was estimated on insulin ( 68.9% by enzyme linked immunosorbent assay , ELISA), and removal of high molecular weight proteins was estimated on albumin (close to 100%, it was undetectable by both ELISA and surface enhanced laser/desorption ionization time-of-flight mass spectrometry, SELDI-TOF MS). Furthermore, coefficient of variability of the peak intensities of seven selected peaks on SELDI-TOF mass spectra was in the range of 12%-21%. Comparing the SELDI-TOF spectrum of whole serum and the fractionated serum, the relative peak intensities of low molecular weight proteins were greatly enhanced.
In the second phase of this study, we applied this method for biomarker discovery in Acute Myeloid Leukemia (AML). A total of 92 serum samples of 54 AML patients were fractionated and analyzed, among them 37 were collected at time of diagnosis, 35 at different stages of disease during follow up and 20 at long term CR (at least 2 years). Comparison of protein profiles of sera samples at diagnosis and at CR revealed several differentially expressed protein peaks, and further characterization of these differentiated protein/peptides are in progress. <<
The use of filters or LC columns to filter out big proteins has been mentioned before. Petricoin/Liotta, and others have noted, however, that some little proteins of interest bind to the big ones, so you're potentially losing some data by this technique. OTOH, I'm not sure how easy it is isolate those pairings.
>>Abstract Number: 399
Presentation Title: Reproducibility of LC-FT-ICR-MS for proteomic analysis of the low molecular weight fraction of serum and plasma.
Presentation Start/End Time: Sunday, Apr 17, 2005, 8:00 AM -12:00 PM
Category: CH03-06 Proteomic analysis
Author Block: Janet E. Olson, Ann L. Oberg, Kenneth L. Johnson, Christopher J. Mason, David C. Muddiman. Mayo Clinic and Foundation, Rochester, MN
BACKGROUND: Mass spectrometry (MS) is being increasingly used as a proteomic tool to search for potential biomarkers in biological fluids. Liquid chromatography coupled with Fourier-transform ion cyclotron resonance MS (LC-FT-ICR-MS) has the potential to be an extraordinary biomarker discovery platform because it provides high mass measurements accuracy (<1 parts-per-million), high resolving power (>100,000), wide dynamic range, and high sensitivity. In preparation for carrying out experiments in ovarian and pancreatic cancer, we examined the overall reproducibility of LC-FT-ICR-MS for analysis of the low molecular weight fraction of serum and plasma.
METHODS: In parallel experiments, plasma and serum from a single female were diluted, separated into 6 aliquots and passed through 50 kDa low molecular weight cutoff filters. 100ul aliquots from each set of six were combined to create pooled samples of plasma and serum, respectively. Pooled and non-pooled samples were each analyzed 6 times via LC-FT-ICR-MS and the number of molecular species common to all 6 replicates was determined.
RESULTS: About 2000 components were detected in at least one replicate of the pooled plasma or serum by LC-FT-ICR-MS. However, only 500 components were consistently detected in all 6 replicates underscoring the need to critically evaluate any analytical platform intended for biomarker discovery. Importantly, the number of common species was not related to the absolute ion abundance. Statistical analyses of these data will be presented in terms of the number of technical and biological replicates required for confident elucidation of molecular species, sources of variability and their magnitude.
DISCUSSION: Embarking on a quest for biomarkers is an extremely difficult analytical challenge. The complexity and wide dynamic range of proteins in serum and plasma is daunting especially in the face of significant biological variability. MS based discovery platforms must be fully characterized and understood to be effectively employed for biomarker discovery. Furthermore, informatics and biostatistics are central to this overall process. We will critically evaluate the LC-FT-ICR-MS biomarker platform at its current stage of development and present both accomplishments and remaining challenges.
CONCLUSION: There exist many sources of variability in MS proteomics experiments. Although LC-FT-ICR-MS instrumentation has the highest levels of mass resolving power and mass measurement accuracy of all MS instruments available on the market today, even these qualities cannot guarantee detection of meaningful biomarkers. Proteomics research being conducted on FT-ICR MS and lower mass resolving power instruments (e.g., SELDI-TOF) must consider the impact of analytical and biological variability on the accuracy of results. <<
First cautionary note. If the best instrument available still has reproducibility issues . . . Not a huge commercial threat, as these things are quite expensive.
>>Abstract Number: 4805
Presentation Title: Evaluation of existing and new proteomic patterns for the detection of ovarian cancer
Presentation Start/End Time: Tuesday, Apr 19, 2005, 1:00 PM - 5:00 PM
Category: CH03-07 Proteomic pattern diagnostics and early detection
Author Block: Lee E. Moore, Ruth M. Pfeiffer, Annette Molinaro, Charles Rabkin, Zheng Wang, Jing Wang, Christine Yip, Xiao-Ying Meng, Marielena McGuire, Eric T. Fung. DCEG, NCI, Bethesda, MD, Ciphergen Biosystems, Fremont, CA
Purpose: Ovarian cancer stage I patients have a 95% five-year survival, however, >65% of ovarian cancers are detected at an advanced stage, with resulting five-year survival rates of 35-40%. New biomarkers for early diagnosis are needed. Recently, we identified and validated three biomarkers (transthyretin, apolipoprotein A1, and a fragment of inter-alpha trypsin inhibitor IV using the Protein Chip System® platform (Zhang et al., Cancer Res, 64, 2004). Subsequently, we developed a chromatographic SELDI-TOF based assay to detect unmodified transthyretin and its post-translationally modified forms (+SO2H, +Cys, +Cys-Gly, +Glutathione) and Apo A1. In this study we i) applied these markers in an independent study ii) determined sources of marker variability among controls iii) identified new potential markers. Methods: Sera were collected from hospitalized patients post-diagnosis/pretreatment at the Mayo Clinic (1980-1988) including 45 women with ovarian cancer and 71 with benign tumors. Controls were 130 women hospitalized for digestive disease. Patient age, smoking status, ICD code, draw date and number of freeze-thaws for each sample were known. For cases, histologic information was provided. A chromatographic ProteinChip assay was used to detect transthyretin and ApoA1, arrays were read in a ProteinChip System, 4000. Mass spectra were externally calibrated for mass, internally normalized for intensity using total ion current and baseline subtracted. Peaks were manually selected and the intensity recorded while blinded to disease status. T-tests assessed differences in peak intensity between groups. Linear and quadratic discriminant analysis and nearest neighbor methods were used to build classifiers. Misclassification rates were estimated by cross validation. Results: Unmodified transthyretin (p=0.02), +SO2H (p=0.007), +Cys (p=0.02), +Cys-Gly (p=0.004), +Glut (p=0.007) and Apo A1 (p=0.0001) levels were lower in cases than in controls. CA125 was elevated in cancer patients (p=0.001). A prediction algorithm using the seven markers, age and CA125 had 81% sensitivity and 76% specificity. Misclassified cases had low stage, histologically well differentiated tumors and CA125 levels <35U/ml. Among controls, unmodified transthyretin (p=0.05) and +Cys (p=0.02) were negatively associated with sample draw year and +Cys, +Cys-Gly (p=0.05) and +Glut (p=0.01) with control diagnosis. Age was associated with a fragmented form of transthyretin. Discovery analysis revealed additional peaks that discriminated cases from controls in univariate analysis, but did not change classification when included in the algorithm. Conclusions: We confirm the work of Zhang et al. using serum samples stored >15 years, however, misclassification rates particularly for benign ovarian tumor patients were high. Influences of sample storage conditions and subject characteristics on markers require further study.<<
This one is the most alarming to me. This is the only abstract that mentions use of the 4000 system, the 2nd generation instrument that is supposed to be clinically robust. The pattern they are talking about is the one Ciphergen is trying to commercialize. The concluding sentence noting the misclassification issue almost gainsays the clause about confirming the work of Zhang's team. The sensitivity and specificity are both lower, but that might be because they apparently didn't use the alpha trypsin fragment. How they can confirm the work without it is beyond me. Knowing the number of freeze thaw cycles per sample helps some, but other studies show that the more cycles the samples go through, the more degraded they become. What's wrong with just chipping off what you need without thawing the whole thing? Must be something, or they'd do that, I'd think.
>>Abstract Number: 4798
Presentation Title: Stage stratification improves proteomic serum profiling classification of breast cancers
Presentation Start/End Time: Tuesday, Apr 19, 2005, 1:00 PM - 5:00 PM
Category: CH03-07 Proteomic pattern diagnostics and early detection
Author Block: E. Ellen Schwegler, Minetta C. Liu, O. John Semmes, Richard R. Drake. Eastern Virginia Medical School, Norfolk, VA, Georgetown University Medical Center, Washington, DC
The objective of our studies was to assess the use of SELDI-TOF-MS proteomic technologies to evaluate whether stratification of patient sera according to breast cancer staging produced proteomic profiles capable of improving the ability to detect and diagnose differences between normal, early and late stage cancers. Serum samples (n=114) from normal healthy subjects and those with either DCIS, stage 1, stage 2, stage 3 or stage 4 breast cancers were obtained from the Breast Cancer Biomarker Resource at the Lombardi Comprehensive Cancer Center. Diluted serum specimens were processed and applied in duplicate to IMAC3-Cu ProteinChips using automated liquid handling robotics. Clustering and classification analyses were performed on all possible sample pair combinations using Ciphergen Biomarker Wizard and Biomarker Patterns software packages. The protein serum profiles demonstrated distinct patterns with the ability to distinguish DCIS and the different stages from normals. The SELDI-TOF MS analysis generated 61 total protein peak clusters, 34 of which were differentially expressed (p = 0.05). The most robust classifications were obtained for distinguishing stage 1 and stage 2 samples from normal and DCIS groups. The mean intensities of two peaks (m/z 7776 and 9299) were increased at Stage 1 relative to normals, but decreased steadily with stage progression. The mean intensity of another peak (m/z 8944) was elevated primarily in stage 3 and stage 4 samples. These results indicate that stratification of serum samples based on breast cancer disease stage can provide potentially useful diagnostic or prognostic biomarkers for breast cancer, and should be factored in the design of future proteomic studies on larger sample cohorts. Supported by the NCI Early Detection Research Network.<<
This idea could easily be applied to other cancers, and it should be.
Other studies involved using protein fingerprints as surrogate endpoints; the measurement of response to therapy, and some little improvements in specificity/sensitivity in various cancers.
Overall, I was hoping to see reproducibility issues more convincingly addressed, and more showing from the 4000 system. According to the last 10-K, the 4000 now comprises roughly 80% of system sales. That would be great if sales weren't declining overall.
As always, if anyone reaches different conclusions than I do, please chime in.
Ciphergen usually issues an AACR PR a day or so before it starts. They'll spin hard, I'm sure, but nothing here warrants a position, IMO. One could perhaps daytrade it, but be quick.
Cheers, Tuck |
