| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Editorials |
Department of Pathology and Laboratory Medicine, University of Pennsylvania, 7.103 Founders, 3400 Spruce St., Philadelphia, PA 19104, E-mail srmaster{at}mail.med.upenn.edu
The recent emergence of methods for rapidly profiling large numbers of protein markers by use of mass spectrometry (MS) has raised hopes for the rapid identification of novel cancer biomarkers (1). In particular, MS-based assays using surface-enhanced laser desorption/ionization (SELDI) have been used to compare serum proteomic patterns from healthy and diseased individuals in hopes of finding diagnostic signatures that can be adapted for use in the clinical laboratory. Early investigations in ovarian cancer patients suggested that panels of anonymous markers might substantially advance our current diagnostic capabilities (2)(3), although subsequent criticism has been directed at the primary data from these studies (4)(5) as well as at the SELDI time-of-flight (TOF) profiling approach itself (6)(7) [for summaries of this ongoing controversy, see Refs. (8)(9)]. Specific questions have been raised regarding the reproducibility of SELDI-TOF spectra, possible changes in protocols or inadequate calibration, and the ability of SELDI to detect low-abundance tumor markers. Additionally, observations that spectra can vary based on analytical factors such as the time of processing have been noted during large profiling experiments (10).
Several of these issues have recently been addressed through the combined efforts of investigators in the National Cancer Institute–sponsored Early Detection Research Network (EDRN). This group has undertaken a systematic assessment of the interlaboratory reproducibility of SELDI-TOF measurements along with their potential applicability to prostate cancer diagnosis. Their initial results demonstrated that relevant portions of SELDI-TOF profiles can be measured reproducibly and used to distinguish a reference set of prostate cancer cases from controls (11). This study was an encouraging step toward defining the analytical reproducibility of serum proteome profiling, although it highlighted the need for rigorous calibration of instruments and adherence to standardized technical procedures. Although promising, the efforts to date have focused on parallel profiling of serum aliquots without directly addressing the importance of variation attributable to differences in sample collection and other sources of preanalytical bias that can be expected in routine clinical practice. A more comprehensive list of issues that must be addressed to understand the effects of preanalytical, analytical, and postanalytical factors on SELDI-TOF and matrix-assisted laser desorption/ionization (MALDI)-TOF profiles has recently been proposed (12).
A valuable example of the effects of preanalytical and analytical variability in SELDI-TOF profiling is reported in this issue of Clinical Chemistry. Karsan et al. (13) describe serum profiles obtained from patients undergoing core needle biopsies of breast lumps. The primary question they wished to address is whether a set of mass spectral peaks can distinguish women with breast cancer from those with benign lesions. Unfortunately, although such a diagnostic signature would have represented a valuable step forward, they were unable to reliably separate these patient groups by serum profiling. In contrast, however, they were able to identify robust SELDI-TOF patterns that grouped samples based on the clinic site from which the samples were obtained and on the date of processing. The ease with which these preanalytical and analytical biases could be detected, coupled with a reported inability to replicate previously published results, led the authors to speculate that similar effects may have tainted previous SELDI-TOF–based marker studies. Their data provide evidence that preanalytical and analytical variation can affect profiled markers, and this result must, at a minimum, raise awareness of the strong potential for bias in serum profiling experiments that are not carefully controlled.
It is not entirely surprising—or necessarily problematic—that such bias can be detected within the serum proteome. Given the large number of proteins surveyed, it is reasonable to suspect that a subset of these markers might be susceptible to effects of the method of collection, storage conditions, or time of processing. The data from Karsan et al. (13) demonstrate not only that proteomic profiles are susceptible to these types of influences but also that the effects are sufficiently reproducible to form the basis for a reliable classification scheme. These results confirm—if it was not already apparent—that substantial investigation is warranted to understand the susceptibility of serum proteomic profiling methods to preanalytical and analytical variation.
Questions surrounding the nature of this susceptibility are particularly pressing because at least 2 models (with very different implications) can account for the reported results. In the first model, a relatively small number of proteins or posttranslational modifications may be susceptible to preanalytical factors such as those associated in this study (13) with different clinic sites. As long as differences in these "badly behaved" peaks are of sufficient magnitude and reproducibility, they will reliably distinguish samples on the basis of, for example, their clinic of origin. In fact, Karsan et al. (13) identify 2 peaks (m/z 2992 and 5643) with just this property. Additionally, such sets of peaks may well change more as a result of preanalytical factors than other, comparable groups of peaks change as a result of disease state; global clustering approaches would therefore group samples on the basis of this preanalytical variation, and any inadvertent, systematic correlation between preanalytic factors and disease state would likely lead to the spurious incorporation of bias-related peaks into the resulting diagnostic signature. As long as these effects are primarily limited to a specified set of peaks, however, eliminating this set from consideration when searching for diagnostic patterns might reduce or even eliminate the problem. We can envision a systematic study, similar to that proposed by Hortin (12), that would examine a range of typical clinical variables precisely to generate a database of badly behaved peaks for exclusion from future data mining exercises. By eliminating these peaks from consideration when building diagnostic signatures, future studies might also reduce the risk that inadvertent bias will influence the results (although it bears repeating that this problem is best solved by adhering to rigorous standards for study design and implementation).
In the second model, however, sources of preanalytical and analytical bias would have a pervasive effect across many or all peaks in the spectrum. This second case would be much more serious, as it would suggest the necessity for different types of calibration [perhaps based on internal reference ions (14)] and/or other technical developments to ensure reproducibility. Even in this latter case, however, MS-based profiling might still be useful for biomarker discovery under certain circumstances using methods such as principal component analysis to separate systematic effects reflecting preanalytical bias (e.g., effect of clinical site) from more subtle patterns that correlate with disease state [see Alter et al. (15) for an analogous issue in microarray data sets]. Regardless of whether such analytical methods are useful for discovery, the "pervasive effects" model has more profound implications for the feasibility of implementing clinical tests using current SELDI-TOF technology.
Granted that careful study design, standardization/calibration as demonstrated by the EDRN, and a systematic characterization of preanalytical and analytical effects will be required to comprehensively address the problems raised by Karsan et al. (13) and others(10), are there additional steps that might reduce the risk of inadvertent bias in proposed diagnostic profiles? One solution may be to rely on biological mechanisms and/or plausibility as an independent check on profiling results. It is not, strictly speaking, necessary to know the identity of biomarkers (14), and they are not necessarily less effective if they reflect epiphenomena of a disease. However, several alternative ways of sampling the serum proteome can provide primary sequence identification, and these may be of some use in determining whether a credible case can be made for a relationship to the disease. These methods include liquid chromatography–tandem MS analyses of peptides derived from tryptic digests of proteins or of low–molecular-weight proteins bound to albumin (16). It will be interesting to see whether these sequence-based approaches can avoid bias-related pitfalls by identifying the biological reasons for changes in putative biomarker profiles.
The clinical adoption of quantitative, multianalyte assays is still in its infancy. Although the steps necessary for both validation and routine quality control of individual biomarkers—including assessment of linearity, susceptibility to a range of preanalytical variables, matrix effects, and other steps—are well understood, it should come as no surprise that things are more complex at both the discovery and diagnostic stages with the advent of highly multiplexed genomic and proteomic assays. Despite this complexity, the basic data necessary to understand the behavior of diagnostic assays, including an understanding of the effects of a range of preanalytical influences, remain the same. Karsan et al. (13) have furnished clear examples of bias effects in serum SELDI-TOF profiles. An improved understanding of the nature and causes of these effects will be required before such assays are ready for adoption in the clinical laboratory.
References
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  M. Brulliard, D. Lorphelin, O. Collignon, W. Lorphelin, B. Thouvenot, E. Gothie, S. Jacquenet, V. Ogier, O. Roitel, J.-M. Monnez, et al. Nonrandom variations in human cancer ESTs indicate that mRNA heterogeneity increases during carcinogenesis PNAS, May 1, 2007; 104(18): 7522 - 7527. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. G.W. Norden, P. Rodriguez-Cutillas, and R. J. Unwin Clinical Urinary Peptidomics: Learning to Walk Before We Can Run Clin. Chem., March 1, 2007; 53(3): 375 - 376. [Full Text] [PDF]
|
|
|
|
|
|
G. L. Hortin, S. A. Jortani, J. C. Ritchie Jr, R. Valdes Jr, and D. W. Chan Proteomics: A New Diagnostic Frontier Clin. Chem., July 1, 2006; 52(7): 1218 - 1222. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. P. Diamandis Serum Proteomic Profiling by Matrix-Assisted Laser Desorption-Ionization Time-of-Flight Mass Spectrometry for Cancer Diagnosis: Next Steps Cancer Res., June 1, 2006; 66(11): 5540 - 5541. [Full Text] [PDF]
|
|
|
|
|
|
E. P. Diamandis Validation of breast cancer biomarkers identified by mass spectrometry. Clin. Chem., April 1, 2006; 52(4): 771 - 772. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
