Differentiation of acute cutaneous GVHD and non-GVHD vacuolar interface skin eruptions represents a major challenge in the care of patients following HCT. In an attempt to improve diagnostic accuracy in this setting, we discovered potential tissue-based molecular biomarkers, incorporated a selection of these biomarkers into a quantitative RT-PCR assay, and created a clinical-molecular fusion model that confers a ROC AUC of 0.89. This novel diagnostic approach demonstrates the potential to differentiate GVHD from drug reactions with vacuolar interface changes on skin biopsy in the post-HCT period using a limited set of clinical and molecular data.
Early in the history of HCT, it was recognized that GVHD is a syndrome and that the diagnosis of skin GVHD is complex, has no single clinical or histopathologic pathognomonic finding, and requires clinicopathologic correlation [27]. A prior study designed to distinguish GVHD from drug eruptions included a robust analysis of clinical features, with an emphasis on those dermatologic [1]. It was found that facial involvement was more frequently observed in GVHD (59%) than in drug hypersensitivity reaction (24%; p = 0.05), and the addition of palms/soles involvement increased specificity (36% of patients with GVHD; 0% in patients with drug hypersensitivity reaction; p = 0.006). While these clinical findings may be helpful when present, they lack diagnostic sensitivity.
Little progress has been made in the histopathologic diagnosis of GVHD in the last 3 decades. An early attempt to differentiate GVHD from drug reactions on skin biopsy histopathology and immunohistochemistry found no definite differences and concluded, “neither morphological nor immunohistochemical analysis of skin biopsies is of much help in distinguishing [GVHD] reaction, grade I and II, from drug-induced skin reactions in patients developing skin lesions after bone-marrow transplantation” [3]. A widely cited analysis of histopathologic features of GVHD concluded that skin biopsy “is not a useful tool for differentiating between [GVHD] and other entities that lack specific histologic features such as drug eruptions and viral exanthems” [8]. A subsequent study of clinical outcomes based on skin biopsy results, which found no significant correlations, made the following recommendation: “We suggest that the routine practice of subjecting patients to such biopsies could be abandoned without compromising their care” [30]. More recently, a study found that the CD123 stain for plasmacytoid dendritic cells (a marker of type I interferon influence) may have utility in the diagnosis of GVHD of the colon [17]. Based on this finding, as well as the knowledge that plasmacytoid dendritic cells can be found in skin affected by GVHD [18], our group revisited the potential diagnostic utility of skin biopsy in differentiating cutaneous GVHD from drug reactions with vacuolar interface changes in a small study of 24 skin biopsies [10]. While no differences between CD123 patterns were observed, skin biopsies of GVHD did show significantly less spongiosis and fewer dermal eosinophils than drug reactions. Though dermal eosinophil density has been shown to be inconsistent in GVHD in other studies [19, 28], spongiosis had not been well-documented in this setting previously. Though the absence of spongiosis may warrant further study, data are insufficient to regard it as an independently reliable histopathologic feature of GVHD.
Important of molecular investigations of GVHD have focused on prognostic serum biomarkers for this disease [14, 15, 21, 23]. Comparative molecular evaluations of skin biopsies of GVHD vs. drug eruption do not appear to have been attempted previously, however. Prior researchers have evaluated genetic expression profiles in the skin of murine allogeneic GVHD models and recognized that genes that are interferon-induced, are involved in antigen presentation and cellular adhesion, or are acute phase reactants are upregulated [29]. In addition, the importance of type I interferon and plasmacytoid dendritic cells in skin tissue affected by GVHD has been demonstrated [18, 24]. Our group confirmed upregulation of interferon-mediated and other proteins in skin affected by GVHD [11] and in the present study, demonstrated that their quantitative expression can be used to differentiate GVHD from drug eruption with vacuolar interface changes.
Results of the biomarker discovery phase of this study offer new insights on the biology of GVHD. Specifically, genes that were found to be enriched in GVHD included those that are involved in response to cellular stress, mediated by type I interferons, involved in purine ribonucleoside, anion, or carbohydrate derivative binding, or involved in collagen fibril and extracellular matrix organization (www.string-db.org/). These genes and pathogenic mechanisms warrant further study as possible mediators of GVHD or potential therapeutic targets.
Genes that were incorporated into the final model included MX1 (interferon-induced mediator of Th1 immunity), MNDA (mediator of granulocyte/monocyte cell-specific response to interferon), OAS (interferon-induced regulator of osteoblast differentiation), and GSTM5 (catalyst of reduced glutathione conjugation). The single clinical variable that contributed to the model was the presence of diarrhea. Though this feature is not specific to gastrointestinal GVHD, it has been shown previously to be a relatively specific finding for GVHD over drug eruption when coupled with a morbilliform skin eruption in the post-HCT period (73% vs. 12% in patients with GVHD or drug reaction, respectively; p = 0.000) [1].
Strengths of this study included ensuring that every sample representing drug eruption was derived from patients who had previously undergone HCT, thereby attempting to control for post-HCT-related factors. Each case showed vacuolar interface changes, thereby making sure that cases were comparable to GVHD. In addition, the identification of biomarkers using an unbiased approach allowed for recognition of genes of interest that have do not appear to have been studied in GVHD previously.
The primary limitation to this study is the difficulty in assuring accuracy with case classification, due to the lack of certainty inherent in current diagnostic methods. We attempted to mitigate this limitation by carefully examining the electronic medical record for each patient not only for documentation about patient features but also for clinical course after skin biopsy. The study also is limited by the relatively small sample size, particularly with regards to drug reaction cases. The small number of patients in this group is attributable to the high level of stringency used during case selection. That is, only cases with a high degree of diagnostic certainty and vacuolar interface changes on skin biopsy were included. In addition, this study is limited by the single-center, retrospective design, as it is remains unknown whether this assay will be valuable applicable when used prospectively and/or for patients cared for at other institutions.
We anticipate future larger multi-institution studies to validate the utility of this diagnostic tissue-based molecular assay and clinical-molecular fusion models.