Correspondence
Autoantibodies in Prostate Cancer
N Engl J Med 2005; 353:2815-2817December 29, 2005
- Article
To the Editor:
Wang et al. (Sept. 22 issue)1 constructed a 22-phage-peptide detector for prostate cancer, with 81.6 percent sensitivity and 88.2 percent specificity. Previously, we constructed a decision tree for classifying prostate cancer, using five tumor-associated antigens, with 79 percent sensitivity and 86 percent specificity.2 With a smaller panel, these rates are equivalent to those of Wang et al. and might be improved by the selection of other tumor-associated antigens.
Only four peptides in the panel that was developed by Wang et al. were derived from in-frame coding sequences. Whether the other 18 peptides are mimotopes could be determined, for example, by analyzing the phage-peptide panel with other controls, including serum from patients with autoimmune disorders and other diseases. In our study and in other trials,3 known tumor-associated antigens were used in the antigen panels.
Nevertheless, the principal conclusion of Wang et al. — that their panel could provide a screening test for prostate cancer — is tenuous. Operating characteristics of the authors' test, as well as of our panel of tumor-associated antigens, do not justify adoption in a screening program. Proportionately, too many cancers would be missed, and high false positive rates would be distressing. Evaluation of putative screening tests should include the calculation of likelihood ratios, as was previously advocated.4
4 ReferencesE.M. Tan, M.D.
James A. Koziol, Ph.D.
Scripps Research Institute, La Jolla, CA 920371
Wang X ,Yu J ,Sreekumar A , et al. Autoantibody signatures in prostate cancer. N Engl J Med 2005;353:1224-1235
Full Text | Web of Science | Medline2
Koziol JA ,Zhang JY ,Casiano CA , et al. Recursive partitioning as an approach to selection of immune markers for tumor diagnosis. Clin Cancer Res 2003;9:5120-5126
Web of Science | Medline3
Suzuki H ,Graziano DF ,McKolanis J ,Finn OJ . T cell-dependent antibody responses against aberrantly expressed cyclin B1 protein in patients with cancer and premalignant disease. Clin Cancer Res 2005;11:1521-1526
CrossRef | Web of Science | Medline4
Fagan TJ . Nomogram for Bayes's theorem. N Engl J Med 1975;293:257-257
Web of Science | Medline
To the Editor:
Wang et al. report that autoantibodies against prostate cancer–specific peptides could be useful in a screening test for prostate cancer. However, their autoantibody screening system was positive for prostate cancer in 30 percent of patients with lung adenocarcinoma. Moreover, eukaryotic translation initiation factor 4 gamma 1 (eIF4G1), which was one of the new biomarkers used by Wang and colleagues, is overexpressed in 72 percent of cases of lung adenocarcinomas and is not a specific marker for prostate cancer.
Yujiro Kida, M.D., Ph.D.
Tsurumi University, School of Dental Medicine, Yokohama 230-8501, JapanTo the Editor:
Wang and colleagues have not clarified whether their study controls were thoroughly investigated to rule out cancer so as to avoid any cross-reactivity. They do not mention the number of times reshuffling of their training and validation sets was done during the analysis; at least 59!+70! different combinations of the training set can exist. Ein-Dor et al.1 have conclusively shown that repeating the analysis on several reshuffled training sets takes away the uniqueness of a molecular signature. They demonstrated that in 70 percent of such reshuffled training sets, more than one signature with equal or better predictive ability exists. Therefore, if this experiment is reanalyzed using a few hundred reshuffled training sets, there could be multiples of 22 autoantibody signatures with the same accuracy; these 22 autoantibodies could just be a few of many antibodies that are elevated in prostate and other cancers. We wonder if this is the real reason for an approximate 30 percent detection rate of this signature in patients after prostatectomy, in those with hormone-refractory prostate cancer, and, intriguingly, in those with lung cancer.
1 ReferencesMangesh A. Thorat, M.D., D.N.B.
Rajendra A. Badwe, M.D.
Tata Memorial Hospital, Mumbai 400012, India
thorat_mangesh@hotmail.com 1
Ein-Dor L ,Kela I ,Getz G ,Givol D ,Domany E . Outcome signature genes in breast cancer: is there a unique set? Bioinformatics 2005;21:171-178
CrossRef | Web of Science | Medline
To the Editor:
Wang et al. propose a new marker for the diagnosis of prostate cancer. They studied a panel of peptides in a group of patients with biopsy-proven prostate cancer and in a group of men (defined as controls) without a history of prostate cancer. Nevertheless, the authors do not provide rigorous evidence of the absence of prostate cancer in the so-called control group. To avoid a verification bias, the authors should provide the results of end-of-study biopsies in the control group.
Bernardo Rocco, M.D.
Istituto Europeo di Oncologia, 20141 Milan, Italy
bernardo.rocco@ieo. it Bob Djavan, M.D.
University of Vienna, 1030 Vienna, AustriaOttavio de Cobelli, M.D.
Istituto Europeo di Oncologia, 20141 Milan, ItalyAuthor/Editor Response
In response to Tan and Koziol: In the study by Koziol et al.,1 which differs considerably from ours, seven tumor-associated antigens were arbitrarily selected and used in a multiplex enzyme-linked immunosorbent assay to monitor autoantibody levels in patients with cancer. In our study, we identified new peptides that are immunogenic in patients with prostate cancer, as compared with control subjects. The panel could be cut from 22 to 10 phage peptides with equivalent performance characteristics.
The assays in both trials had approximately 80 to 90 percent sensitivity and specificity for the detection of prostate cancer.1 The recursive partitioning algorithm used by Koziol et al. suffers from the limitation that the described trees are unstable (i.e., if the data are perturbed by the addition of stochastic noise, it will affect the variables and cutoff points chosen for the trees). We agree that both approaches could be improved by the inclusion of additional tumor antigens and that neither method has yet achieved the performance characteristics or additional validation necessary for adoption in screening programs. Regarding the likelihood ratio, it is equivalent to the risk score we have developed by Bayes's theorem.2
In response to Kida: Owing to the high prevalence of prostate cancer, a substantial fraction of the patients with lung adenocarcinoma, especially those 55 years of age or older, might very well be expected to have prostate cancer.3 Thus, some of the autoantibodies that were identified for prostate cancer may be shared by patients with lung cancer.4 Although autoantibodies to eIF4G1 may not be specific to prostate cancer, in the context of a 22-peptide panel, they may be useful in detecting prostate cancer.
In response to Thorat and Badwe: The possible number of combinations of training and test sets is actually [129! ÷ (70! × 59!)] × [128! ÷ (68! × 60!)]. The point of the randomization analysis is to determine whether the sensitivity and specificity of the optimal classifier change on the basis of the training and test sets used, not to determine the identity of the optimal classifier. Thorat and Badwe are confusing this issue with the issue of the uniqueness of the signature.
In response to Rocco et al.: We agree that a limitation of our study is that the control group did not undergo biopsy to confirm that they did not have incidental prostate cancer. That said, a negative biopsy result does not completely rule out the presence of prostate cancer. In a follow-up prospective study, we are evaluating patients with negative biopsy results.
Xiaoju Wang, Ph.D.
Debashis Ghosh, Ph.D.
Arul M. Chinnaiyan, M.D., Ph.D.
University of Michigan, Ann Arbor, MI 48109
arul@umich.edu Drs. Wang and Chinnaiyan report being listed as inventors on a patent on the findings of this study filed by the University of Michigan.
4 References1
Koziol JA ,Zhang JY ,Casiano CA , et al. Recursive partitioning as an approach to selection of immune markers for tumor diagnosis. Clin Cancer Res 2003;9:5120-5126
Web of Science | Medline2
McIntosh MW ,Pepe MS . Combining several screening tests: optimality of the risk score. Biometrics 2002;58:657-664
CrossRef | Web of Science | Medline3
Thompson IM ,Pauler DK ,Goodman PJ , et al. Prevalence of prostate cancer among men with a prostate-specific antigen level ≤4.0 ng per milliliter. N Engl J Med 2004;350:2239-2246[Erratum, N Engl J Med 2004;351:1470.]
Full Text | Web of Science | Medline4
Brass N ,Heckel D ,Sahin U ,Pfreundschuh M ,Sybrecht GW ,Meese E . Translation initiation factor eIF-4gamma is encoded by an amplified gene and induces an immune response in squamous cell lung carcinoma. Hum Mol Genet 1997;6:33-39
CrossRef | Web of Science | Medline
- Citing Articles (1)
Citing Articles
1
C. Desmetz, A. Mange, T. Maudelonde, J. Solassol. (2011) Autoantibody signatures: progress and perspectives for early cancer detection. Journal of Cellular and Molecular Medicine 15:10, 2013-2024
CrossRef
