SELDI-TOF-Based Proteomic Analysis of Plasma Samples as a Diagnostic Tool to Predict Acute Graft-Versus-Host Disease after Allogeneic Stem Cell Transplantation

Allogeneic hematopoietic stem cell transplantation (SCT) offers a curative treatment option for high-risk hematological tumors. Graft-versus-host disease (GvHD) is a severe and potentially life-threatening complication of allogeneic SCT. GvHD is caused by the activation of transplanted alloreactive T-cells and results in severe inflammation of host tissues in skin, liver, and gastrointestinal tract. Manifest GvHD can be diagnosed by clinical criteria and histology only. So far, no diagnostic blood test is able to detect imminent GvHD. GvHD is prevented and treated by systemic immunosuppression which is associated with opportunistic infections and higher relapse rates. Therefore, reliable detection of incipient GvHD would allow to investigate preemptive immunosuppressive therapy tailored to individual patients at appropriate time points. Proteomic profiling is based on the premise that alterations related to pathological states (e.g., GvHD) are accompanied by quantitative and qualitative alterations in plasma protein profiles. The surface-enhanced laser desorption/ionization (SELDI) technology relies on time-of-flight mass spectrometry for accurate measurement of the mass-to-charge ratio (m/z) of proteins that have been preselected on appropriate functional surfaces. We conducted a prospective case-control study based on >3000 serial plasma samples of 236 consecutive patients undergoing allogeneic SCT. 43 patients developed acute GvHD grade I-IV during their sampling period and were matched to control patients remaining GvHD-free. Matching criteria were prior antithymocyte globulin treatment, donor type (related versus unrelated), conditioning protocol, remission status at SCT, diagnosis, age, and sex. These factors were ranked in the listed order. Plasma protein patterns were obtained by SELDI-TOF on day -2 prior to the onset of clinical GvHD and compared to 43 plasma samples obtained on the corresponding day after SCT from the matched GvHD-free patients. In order to enhance the number of analyzable proteins, plasma samples were fractionated to yield 5 different sets of SELDI spectra. Protein peak positions and calibrated peak heights were determined according to Yasui et al (J Biomed Biotechnol, 2003). To identify a limited number of predictive peaks, a sparse logistic regression model was fitted by a boosting approach (Tutz & Binder, Computational Statistics & Data Analysis, 2007), resulting in 4 peaks at 11963, 22363, 22479, and 30977 Da. Prediction performance of a signature derived from this approach on new data was evaluated using bootstrap samples, resulting in an estimated misclassification rate of 38.5%, an estimated sensitivity of 61.0%, and specificity of 60.5% In addition, the identified predictive peaks could be recovered in most bootstrap samples, indicating sufficient stability of the signature. While the performance of this assay strategy is insufficient for use in clinical practice at present, our data nevertheless demonstrate the potential of proteomic plasma analysis in combination with a tailored statistical analysis to identify GvHD prior to its clinically recognizable onset. In contrast to the clinical routine of immunosuppressive GvHD prophylaxis for all patients after allogeneic SCT with treatment escalation in the case of clinically relevant GvHD, this diagnostic tool therefore offers the perspective for preemptive GvHD therapy administered only to patients with an imminent GvH reaction. Such a tailored GvHD strategy may represent a cornerstone to reduce the risk of infectious complications and tumor relapse while maintaining effective GvHD control in patients at risk. Current work addresses the identity of the index peaks, the temporal evolution of the signature, and refinement of the proteomic analysis.