Biomarkers of immune activation to screen for severe, acute GVHD

Severe acute GVHD (AGVHD) occurring after allogeneic hematopoietic SCT (AHSCT) is often unresponsive to glucocorticoids and, as a result, frequently fatal.1 A preemptive approach, similar to that used for CMV infection could be more effective.1, 2 We conducted a prospective study of plasma markers of immune activation to assess whether the incipient immunologic graft-vs-host response can be detected in a diverse group of transplant recipients before the clinical manifestations of severe AGVHD emerge. Five markers were selected: four markers of T-cell activation, soluble CD8 (sCD8), soluble IL-2 receptor (sIL-2R), soluble CD40 ligand (sCD40L) and soluble CD28 (sCD28) as well as soluble TNF receptor 1 (sTNFR1), an inflammatory marker. These markers are measurable in the blood and have been shown to be increased in clinical AGVHD or other immune-mediated processes.3, 4, 5, 6, 7, 8, 9, 10, 11 The primary hypothesis tested in this study was that by using these markers, preclinical AGVHD can be diagnosed by transplant day 15 with a sensitivity and specificity of at least 80%. Eligible pediatric and adult patients scheduled to undergo AHSCT at Children's Healthcare of Atlanta or the Winship Cancer Institute of Emory University from 1 May 2006 to 30 May 2008 were consented and enrolled before beginning conditioning. Our intent was to develop a test that would be applicable across a wide range of patient ages, diseases, stem cell sources, donor types and conditioning approaches; eligibility, then, was broadly defined. Data were collected weekly through post transplant day 60 to identify patients that would develop AGVHD before beginning the calcineurin inhibitor taper. AGVHD was staged and graded using established criteria.12 This study was approved by the Emory University and Children's Healthcare of Atlanta institutional review boards. Central venous blood samples were collected prospectively at five time points for this analysis: before conditioning, day 0 before transplantation and post transplant days 5, 10 and 15. The samples were centrifuged, and the plasma component was drawn off, aliquoted and stored at −80 °C. All testing was performed according to the manufacturer's specifications. All biomarker values are reported as an increase from day 0 levels to account for interpatient variability. Comparisons of patient characteristics were performed using analysis of variance for continuous variables and Fisher's exact test for categorical variables. Associations between elevated biomarker levels and subsequent AGVHD were identified by performing a univariate Cox proportional hazards analysis to compare the cumulative incidence of grades 3 and 4 AGVHD by day 60 in patients with and without elevated marker levels. Hazard ratios (HR) were calculated using the latter as the reference group. The cutoff point for each marker used to discriminate elevated from nonelevated levels was identified by recursive partitioning analysis (JMP version 8; SAS Institute, Cary, NC, USA). We did not perform a multivariate Cox analysis, as none of the matched related donor graft recipients developed severe AGVHD, preventing us from adjusting for this critical variable. The potential screening test characteristics of the markers were also assessed using a receiver-operating characteristic (ROC) curve analysis. This analysis was limited to the markers for which elevated levels were associated with severe AGVHD. An area under an ROC curve (AUC) was considered statistically significant if the confidence intervals around the estimate did not include 0.5, the value expected for a nondiscriminating test. In selecting cutoffs, sensitivity and specificity were emphasized equally. The ROC curve analysis was conducted to distinguish patients who would develop grade 3 or 4 AGVHD from patients who would develop grade 0–2 disease. Logistic regression was used to create a linear model using multiple biomarkers measured simultaneously (composite model). The statistical analyses were performed using SAS version 9.2 (SAS Institute). A total of 62 patients (43 pediatric and 19 adult) who were scheduled for AHSCT met the eligibility criteria and consented to participate. One patient relapsed before transplant, thus leaving 61 participants available for the analysis. The mean age of patients was 20 years (range: 0–67). Most patients had a malignant disease (64%), received a transplant from an unrelated or mismatched related donor (67%) and received a marrow (46%) or peripheral blood (34%) graft, and 77% received a myeloablative conditioning regimen. AGVHD of any grade was diagnosed at a median of 20 days post transplant (range: 14–32) in 26 patients (43%). 1, 11, 12 and 2 patients were diagnosed with grade 1, grade 2, grade 3 and grade 4 disease, respectively. All 14 patients (23%) who developed severe AGVHD received grafts from unrelated or mismatched related donors. The intraclass correlation coefficients for the test–retest reliability of the sIL-2R, sCD8, sCD40L and sTNFR1 assays using the masked duplicate quality control samples were 0.94, 0.98, 0.93 and 0.92, respectively. The intraclass correlation coefficient for sCD28 was lower. Testing was repeated several times in a subset of patients, each time yielding a coefficient of <0.9. sCD28, therefore, was not included in subsequent analyses. There were no significant differences in preconditioning levels of sIL-2R, sCD8, sCD40L and sTNFR1 in patients who developed grades 0–1, 2 and 3–4 AGVHD (data not shown). Post transplant elevations in sTNFR1 at days 5 (HR=4.05; 95% CI, 1.25–13.08), 10 (HR=7.33; 95% CI, 1.03–52.36) and 15 (HR=3.6; 95% CI, 1.27–10.20) were associated with severe AGVHD. Elevations of sIL-2R on days 10 (HR=3.07; 95% CI, 1.28–7.32) and 15 (HR=3.07; 95% CI, 1.28–7.32) were associated as was an elevation of sCD8 on day 15 (HR=2.95; 95% CI, 1.04–8.37). There was no association with sCD40L levels at any time point. The day 15 results for all four markers are shown in Figure 1. The results of the ROC curve analyses are shown in Table 1. The AUCs for sIL-2R, sCD8 and sTNFR1, the three markers for which elevated levels were associated with severe AGVHD, were greatest on day 15. They were 0.69, 0.66 and 0.66, for sIL-2R, sCD8 and sTNFR1, respectively. Only the result for sIL-2R on day 15 was statistically significant. The positive predictive value (PPV) was less than 60% at all time points for all markers, whereas the negative predictive value (NPV) ranged from 76% (day 5 sIL-2R) to 90% (day 5 sTNFR1). A composite model of the three markers was created using logistic regression. At each of the three time points, the AUC for the composite panel exceeded the AUCs for the individual markers, and as with the individual biomarkers, testing on day 15 yielded the best screening test characteristics. The AUC for the panel on this day was 0.77 (P<0.001). The corresponding sensitivity, specificity, PPV and NPV of the model were 0.64, 0.76, 0.45 and 0.87, respectively. The results of this prospective study show that it is feasible to detect severe AGVHD before signs and symptoms emerge, using three markers we tested, sIL-2R, sCD8 and sTNFR1. The findings of our study substantiate those of a recent, large study conducted by the University of Michigan BMT program showing that an elevated plasma sTNFR1 level on post-transplant day 7 predicted the subsequent development of moderate to severe AGVHD (grades 2–4).13 In the study presented here, we evaluated multiple markers for predicting severe, rather than moderate to severe, disease. We also assessed the screening test characteristics at three, rather than one, time points. Although there is an extensive body of clinical research regarding the test characteristics of blood biomarkers for diagnosing AGVHD, these are the only two studies we are aware of that focus on screening. The best screening test characteristics were achieved with a composite panel of sIL-2R, sTNFR1 and sCD8. The AUC result for the panel at transplant day 15 (0.77) was statistically highly significant; however, none of the individual markers or the composite panel met our prespecified goal of having both a sensitivity and a specificity of at least 80%. Although the small size of our study precludes us from drawing any firm conclusions, these results suggest, we believe, that sTNFR1, SIL-2R or SCD8 are sufficiently predictive of AGVHD, alone or in combination, to be used for screening, Although the University of Michigan investigators observed a strong association between an elevation in the level of sTNFR1 and the subsequent development of AGVHD, its screening test characteristics were poor. They reported a specificity of 0.83, but only a sensitivity of 0.38.13 We disagree with their assertion that this marker is suitable for screening. Body fluids other than blood represent other potential sources of biomarkers, which could be in conjunction with or instead of blood markers for GVHD screening. German investigators, for example, have identified a panel of urinary markers using proteomic analysis, which accurately predicts AGVHD (sensitivity 83%, specificity 76%).14 They are evaluating this panel further in an ongoing multicenter trial of preemptive therapy. Stool biomarkers, such as those being developed for patients with inflammatory bowel disease, may also hold promise for the early detection of severe AGVHD,15, 16, 17 especially because severe AGVHD almost always affects the gut.18, 19 We are currently conducting a pilot study to begin to evaluate the potential of fecal markers for the early identification of AGVHD. The authors declare no conflict of interest. This work was supported by a Friends Grant from the Children's Healthcare of Atlanta.