A Biomarker Panel for Chronic Graft-Versus-Host Disease

Background. Chronic GVHD (cGVHD) remains the major contributor to morbidity and mortality for survivors of allogeneic hematopoietic cell transplant (HCT), but it remains a clinical diagnosis.

Methods. We used a proteomics discovery approach comparing plasma pools from onset of de novo cGVHD (N=17), progressive cGVHD (N=18), and matched time-point samples from 19 patients without GVHD. Of 105 proteins that showed at least 1.3-fold change in the quantification ratio, we further selected 24 proteins based on their involvement in relevant pathway networks, and the availability of ELISA. In addition, two markers (CXCL9 and ST2) were measured based on previously noted associations with cGVHD or refractory acute GVHD. 

Levels of these 26 proteins were measured by ELISA in plasma from an independent set of 178 patients with cGVHD, and from 33 controls without cGVHD. Logistic regression was used to evaluate the association between cGVHD and biomarkers after log transformation. All analyses were adjusted for significant clinical variables considering age, sex, stem cell source, conditioning (nonmyeloablative vs. others), donor (matched sibling vs. others), and time from HCT to sample collection. To determine the best combination model, we used forward selection with a 0.05 significance threshold, confirmed by backward selection. ROC curves were generated for the best single biomarker, and the combination model. The analysis of nonrelapse mortality (NRM) divided the panel weighted sum on the median value among cGVHD cases (N = 178), and compared cases above and below the median.

Results. Of the 26 proteins tested, 9 were associated with cGVHD with p-values < 0.05 (Table 1). Together ST2, CXCL9, MMP3, and OPN compose a biomarker panel for diagnosis of cGVHD with an AUC of 0.89 (Fig 1). We next compared the biomarker panel between groups with different cGVHD severity (none, mild/moderate, severe) (Fig 2). Severity of cGVHD was associated with the biomarker panel [(p<0.0001 compared to control, and p=0.007 compared to each other (adjusted)]. NRM was compared between groups with high vs. low biomarker levels, showing that the biomarker panel predicted NRM [Hazard Ratio=11.4, p=0.002 (adjusted)] (Fig 3).

We conclude that ST2, CXCL9, MMP3, and OPN represent a biomarker panel for reliable, non-invasive diagnosis of cGVHD as well as for correlation with cGVHD severity, and prediction of NRM. Once further validated, these biomarkers could help identify patients with cGVHD at risk for poor outcomes with current treatment approaches.

Table 1.

	Area under ROC	P-value	Adjusted P-value	Combination
ST2	0.832	<0.0001	<0.0001	Yes
MMP3	0.787	<0.0001	<0.0001	Yes
TRAILR3	0.716	0.0002	0.0002
OPN	0.708	<0.0001	<0.0001	Yes
SELP	0.661	0.008	0.005
CKIT	0.654	0.007	0.01
COMP	0.643	0.009	0.002
CXCL9	0.628	0.02	0.01	Yes
CD146	0.616	0.03	0.04
Combination	0.885	<0.0001	<0.0001

Fig. 1

           

Fig. 2

Fig. 3