Molecular Characteristics of High-Dose Melphalan Associated Oral Mucositis in Patients with Multiple Myeloma: A Gene Expression Study on Human Mucosa

Background Toxicity of the oral and gastrointestinal mucosa induced by high-dose melphalan is a clinical challenge with no documented prophylactic interventions or predictive tests. The aim of this study was to describe molecular changes in human oral mucosa and to identify biomarkers correlated with the grade of clinical mucositis. Methods and Findings Ten patients with multiple myeloma (MM) were included. For each patient, we acquired three buccal biopsies, one before, one at 2 days, and one at 20 days after high-dose melphalan administration. We also acquired buccal biopsies from 10 healthy individuals that served as controls. We analyzed the biopsies for global gene expression and performed an immunohistochemical analysis to determine HLA-DRB5 expression. We evaluated associations between clinical mucositis and gene expression profiles. Compared to gene expression levels before and 20 days after therapy, at two days after melphalan treatment, we found gene regulation in the p53 and TNF pathways (MDM2, INPPD5, TIGAR), which favored anti-apoptotic defense, and upregulation of immunoregulatory genes (TREM2, LAMP3) in mucosal dendritic cells. This upregulation was independent of clinical mucositis. HLA-DRB1 and HLA-DRB5 (surface receptors on dendritic cells) were expressed at low levels in all patients with MM, in the subgroup of patients with ulcerative mucositis (UM), and in controls; in contrast, the subgroup with low-grade mucositis (NM) displayed 5–6 fold increases in HLA-DRB1 and HLA-DRB5 expression in the first two biopsies, independent of melphalan treatment. Moreover, different splice variants of HLA-DRB1 were expressed in the UM and NM subgroups. Conclusions Our results revealed that, among patients with MM, immunoregulatory genes and genes involved in defense against apoptosis were affected immediately after melphalan administration, independent of the presence of clinical mucositis. Furthermore, our results suggested that the expression levels of HLA-DRB1 and HLA-DRB5 may serve as potential predictive biomarkers for mucositis severity.


Methods and Findings
Ten patients with multiple myeloma (MM) were included. For each patient, we acquired three buccal biopsies, one before, one at 2 days, and one at 20 days after high-dose melphalan administration. We also acquired buccal biopsies from 10 healthy individuals that served as controls. We analyzed the biopsies for global gene expression and performed an immunohistochemical analysis to determine HLA-DRB5 expression. We evaluated associations between clinical mucositis and gene expression profiles. Compared to gene expression levels before and 20 days after therapy, at two days after melphalan treatment, we found gene regulation in the p53 and TNF pathways (MDM2, INPPD5, TIGAR), which favored anti-apoptotic defense, and upregulation of immunoregulatory genes (TREM2, LAMP3) in mucosal dendritic cells. This upregulation was independent of clinical mucositis. HLA-DRB1 and HLA-DRB5 (surface receptors on dendritic cells) were expressed at low levels in all patients with MM, in the subgroup of patients with ulcerative mucositis (UM), and in controls; in contrast, the subgroup with low-grade mucositis (NM) displayed 5-6 fold increases in HLA-DRB1 and HLA-DRB5 expression in the first two biopsies, independent of melphalan treatment. Moreover, different splice variants of HLA-DRB1 were expressed in the UM and NM subgroups. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 with UM and several known clinical risk factors. The sensitivity of predicting UM was 52% [18]. Apart from the low sensitivity, those studies were limited by their failure to identify phenotypes or causal relationships.
Here, we present a global gene expression study on oral mucosa biopsies and peripheral blood cell samples from consecutive patients with MM that were treated with high-dose melphalan and HSCT. This study aimed to identify new molecular factors that could predict severe oral mucositis.

Patients
This study included 30 patients, aged 18 years or older, recruited from the Aalborg University Hospital, from September 1 st 2010 to September 1 st 2012. Patients with MM (n = 20) were recruited from the Department of Hematology. Healthy individuals (CON, n = 10) were recruited for a control group from the Department of Maxillofacial Surgery. Of the 20 MM patients, seven withdrew consent before any intervention; one was missed due to earlier start of treatment, which was not communicated to the research unit; and two withdrew after the first biopsy without giving any reason. The remaining 10 patients provided three sequential buccal biopsies and peripheral blood samples. The first biopsy was obtained immediately before they received high-dose melphalan (day0); the second was obtained after the autologous stem cell reinfusion (day2); and the third was obtained during an outpatient control visit (day21). The CON group comprised 10 healthy, non-smoking, age-and gender-matched individuals. Controls provided one buccal biopsy and peripheral blood sample. One CON individual was later diagnosed with the autoimmune disease, psoriasis (CON09), and hence, this subject was not included in the statistical analysis. The North Denmark Region Committee on Health Research Ethics approved the clinical protocol (N-20100022). Informed written consent was obtained from all patients, in accordance with the Declaration of Helsinki.
All patients with MM underwent a comprehensive, initial evaluation, including a medical history and clinical examination. Age, gender, and Eastern Cooperative Oncology Group (ECOG) performance status were recorded at baseline, in addition to the subtype of MM and the time from diagnosis to entering HSCT. The criteria for determining the level of organ involvement at diagnosis was based on the degrees of elevated calcium, renal failure, anemia, and bone lesions (CRAB criteria) [19]. Furthermore, patients were screened for dental infections and, when indicated, these infections were eradicated prior to chemotherapy.
All patients with MM received a standard induction regimen, which consisted of cyclophosphamide 500 mg/m 2 delivered intravenously (i.v.) on days 1 and 8; bortezomib 1.3 mg/m 2 delivered subcutaneously (s.c.) on days 1, 4, 8, and 11; and dexamethasone 20 mg, delivered orally (p.o.) on days 1-2, 4-5, 8-9, and 11-12, repeated every third week, 3 times. After this treatment, patients were primed with cyclophosphamide 2 g/m 2 and recombinant granulocyte stimulating factor (rhG-CSF), before their circulating CD34 + hematopoietic stem cells were harvested with leukapheresis [1]. Only patients without progressive disease were assigned to HSCT. These patients received a high dose of melphalan (200 mg/m 2 ), followed by infusion of autologous hematopoietic stem cells. All patients with MM had received standard antiviral, antifungal, and antibacterial treatment, according to department protocols.

Mucositis and diarrhea assessments
Signs of oral mucositis (OM) were recorded daily for patients with MM during the hospital stay (from administration of chemotherapy to discharge). OM signs were identified according to the WHO oral toxicity assessment worksheet [20], and they included subjective symptoms (pain and ability to eat solid food) and objective findings (erythema, ulceration) in predefined regions of the mouth (lip, check, tongue, floor of the mouth, and soft palate). Grades 0 and 1 (NM) included increasing soreness, with or without erythema, but solid food could be taken. In grades 2 to 4 (UM), food intake gradually declined, due to pain and ulcerations, and parenteral feeding might have become necessary. The maximum OM grade recorded during treatment was considered the patient's general mucositis experience. Diarrhea was estimated according to the Common Terminology Criteria for Adverse Events (CTCAE), issued by The National Cancer Institute of the National Institutes of Health [21]. In grades 1 to 2, vomiting increases from one to two episodes in 24 h to three to five episodes in 24 h. Grade 3 included more than 6 episodes in 24 h, and grades 4 to 5 were considered life-threatening, and could gradually lead to death. Diarrhea data were gathered retrospectively, from medical records.

Biopsy
All biopsies were acquired in a standardized manner. First, the mouth was thoroughly rinsed with chlorhexidine and local anesthesia was applied (0.5 ml Citanest1: prilocain 30 mg/ml + felypressin 0.54 μg/ml; Dentsply, York, PA, US). Then, a 5-mm lens-formed biopsy of the buccal mucosa, approximately 1 cm inferior to the papilla parotidea, was taken with a scalpel. The wound was tightly sutured with resorbable vicryl 4.0 (Ethicon, Sommerville, NJ, US). Patients were instructed to rinse twice daily with chlorhexidine until suture removal, after 10 days. One-half of the biopsy was immediately immersed in RNAlater™ (Ambion, Thermofischer Scientific, Waltham, MA, US) for 24 h; then, it was frozen at -80˚C until analysis. The other half of the biopsy was fixed in 10% neutral-buffered formalin, and within 1½ days, it was embedded in paraffin and maintained at room temperature until further preparation.

Peripheral blood
Within 2 h of taking the biopsy, 15 ml EDTA-mixed venous full blood was drawn. Mononuclear cells (MNCs) were isolated with an in-house standard purification protocol. This protocol follows the manufacturer's guidelines for Ficoll-Paque TM (GE Healthcare, Little Chalfont, Buckinghamshire, UK); density gradient centrifugation in Leukosep R tubes (Greiner Bio-One GmbH, Frickenhausen, Germany). Purified MNCs were suspended in freezing medium containing 10% dimethyl sulfoxide, in units of 5 million, vital frozen at -196˚C in liquid nitrogen, and stored frozen until analysis.

Gene expression
The frozen oral mucosa samples were homogenized with TRIzol R Reagent (Invitrogen, Ther-moFischer Scientific), and total RNA was isolated with the mirVana TM miRNA Isolation Kit (Ambion/Invitrogen, ThermoFischer Scientific) according to the manufacturer's protocol. RNA amplification was performed with the Ambion R WT Expression Kit (Applied Biosystems, ThermoFischer Scientific), according to the manufacturer's instructions, starting with 100 ng total RNA, on a TP Basic Thermocycler, real time PCR instrument (Biometra, Göttingen, Germany). The quality of the RNA product was evaluated on the NanoDrop spectrophotometer and the 2100 Bioanalyzer with the Agilent RNA 6000 Nano Kit (Agilent Technologies, Santa Clara, CA, US). We prepared the RNA samples for hybridization to Affymetrix GeneChip Human Exon 1.0 ST Arrays with the Affymetrix GeneChip WT Terminal Labeling and Controls Kit (P/N 901524) (Affymetrix, Santa Clara, CA, US), according to the manufacturer's instructions. CEL files were generated with Affymetrix GeneChip Command Console Software and deposited at the NCBI Gene Expression Omnibus repository, under number GSE81979. A similar procedure was applied to analyze gene expression in MNCs isolated from blood samples.

Immunohistochemistry
We cut 4-μm-thick biopsy tissue sections and mounted them on glass coverslips. Following an in-house optimized protocol, tissues were stained with an antibody against the HLA class II Histocompatibility antigen, DR beta 5 chain (HLA-DRB5 center region) with a rabbit polyclonal antibody (no. OAAB06426, Aviva Systems Biology, CA, US). Normal tonsil tissue was used as a positive control. The stained slides were then scanned on a Hamamatsu NanoZoomer slide scanner and analyzed with NDP viewer software. To estimate the number of cells that stained positive for HLA-DRB5, each stained slide was searched for a hot spot; then, this spot was framed with a 0.75×0.4 mm (0.3 mm 2 ) rectangle; the area included approximately half lamina epithelialis and half lamina propria. We counted all cells in the frame that were distinctly stained with anti-HLA-DRB5 antibodies. The analyzer was blinded to the mucositis grade.

Statistical analysis
Power estimation of group size. To identify genes that varied more than two-fold between test points with a false discovery rate of less than 0.05% and a power of 90%, we applied the method described by Lee and Whitmore [22], implemented in the R-package, sizepower (Qui 2008). We found that 10 patients in each group were sufficient for detecting significant differences.
Statistical analysis. All statistical analyses were performed with R [23] version 3.2.0 and Bioconductor packages [24]. The p-values adjusted for false discovery rates were controlled with the Benjamini-Hochberg method [25], for each of the above tests. Adjusted p-values below 0.05 were considered significant.
The CEL files produced by the Affymetrix Expression Console and the probes were preprocessed and summarized to gene level with the RMA algorithm in the Bioconductor package 'affy', based on custom CDF files [26]. This preprocessing resulted in the gene expression levels of 38,830 genes for each Exon array each annotated with Ensembl gene (ENSG) identifiers. Patient CON09 was included in the normalizations of the gene expression data, but excluded in the statistical analysis.
With patient ID as a cluster variable, we used the linear model for microarray data (limma package in R), a mixed linear model, and an empirical Bayes approach to test for significant differences in gene expression levels between day2 and day0, and between day21 and day0 [27]. For the peripheral blood samples, we only compared day0 and day21 to baseline, because only two blood samples were analyzable for day2. We performed an unpaired test with the limma package to test for significant differences in gene expression between patients on day0 and controls.
The patients were divided into UM or NM groups, according to their mucositis experience. We used the limma package to detect significant genes that were differentially expressed between the two groups at each time point.
We applied the Mann-Whitney test to test for the relationship between mucositis severity and duration of neutropenia, leukopenia, and thrombocytopenia. We also used the Mann-Whitney test to evaluate differences between groups in the numbers of in-hospital days and years of progression free survival (PFS).

Results
The clinical characteristics and demographics of the included patients prior to HSCT are shown in Table 1. No signs of infection at the site of biopsy were reported. The clinical data collected during the HSCT and at follow up are shown in Table 2. UM (grades 2-4) was observed in 4 patients, and NM (grades 0-1) was observed in 6 patients. The average mucositis scores were 1.5 (range 0-4) for the whole cohort, 3.3 for the UM group, and 0.3 for the NM group. The average diarrhea scores were 2.2 (range 1-4) for the whole cohort, 3.3 (range 3-4) for the UM group, and 1.5 (range 1-2) for the NM group. The average hospital stays were 22.6  [20]. Patients that experienced mucosal ulcerations during treatment were considered to have ulcerative mucositis, WHO grades 2-4; patients with only soreness or erythema were considered to have none/mild mucositis, WHO grades 0-1. 2 Calculated according to the Common Terminology Criteria for Adverse Events [21] days (range  for the whole cohort, 28.8 days (range 21-41) for the UM group, and 18.5 days (range [16][17][18][19][20][21][22][23][24] for the NM group. The difference in hospital stays was statistically significant (p = 0.020). The duration of neutropenia was not significantly different between UM and NM groups, but thrombocytopenia was significantly prolonged in the UM compared to the NM group (p = 0.047). PFS [28] was not statistically different between the UM and NM groups. The CON and MM groups were comparable in age (CON: age 58 y, range 47-78 vs. MM: age 63.5 y, range 51-69) and gender (CON: females 4/10 vs. MM: females: 6/10).

Analysis of gene expression in mucosa samples
All 40 biopsies (3×10 patients and 1×10 controls) provided gene expression profiles. No statistically significant differences in gene expression were found between the MM group on day0 and the CON group. Patients with MM showed no significant changes in gene expression between day0 and day21. However, 35 genes in patients with MM showed significantly different expression between day0 and day2 ( Table 3). The gene expression levels were independent of clinical mucositis. The dominant gene alterations were observed in apoptosis-related genes, followed by genes related to inflammatory/immunologic response, transcription factors, and members of the Histone Cluster family. We also observed alterations in genes related to metabolism.

Gene expression related to mucositis grade
When we compared unsupervised gene expression profiles between NM and UM, we found that no genes were significantly differentially expressed in the blood. In contrast, in the biopsies, two genes of the major histocompatibility complex (MHC) Class II: HLA-DRB1 and HLA-DRB5 were significantly differentially expressed at the first two time points (Table 4). Patients with UM and CON expressed the same low level of HLA-DRB1 and HLA-DRB5, but patients with NM expressed significantly higher levels (Fig 1). The expression levels of HLA-DRB1 and HLA-DRB5 were independent of melphalan administration. Of the 10 CON subjects, one patient, CON09, showed high levels of HLA-DRB1 and HLA-DRB5 expression, similar to the levels observed in the NM group. We reopened the protocol and returned to the patient to reaffirm his health status. We found that subject CON09 had a mild case of psoriasis that was not reported at the baseline interview. An alternative splicing analysis revealed that patients with NM and the CON09 subject expressed a different isoform of HLA-DRB1 (NM_001243965) than that expressed by patients with UM (NM_002124.1). However, the difference in HLA-DRB5 expression between groups was not due to different isoforms.

Immunohistochemistry
In hematoxylin and eosin-stained specimens, no gross anatomical changes were observed in the epithelial or mesenchymal layers. In general, both the epithelium and stroma were represented in equal amounts. However, two specimens that were cut at a tangential angle that revealed only superficial layers were excluded from the histological analysis (MM15_1 and MM18_2). Generally, when present, cells that stained positive for HLA-DRB5 were localized in the lower part of the epithelial layer, near the basal membrane, around the papillae, and in the upper part of the lamina propria. Faint, diffuse HLA-DRB5 staining of the endothelium was not included in the assessment. Examples of high HLA-DRB5 expression/low-grade mucositis (MM01) and low HLA-DRB5 expression/severe-grade mucositis (MM18) are shown in Fig 2. A dotplot of the numbers of cells that stained positively for HLA-DRB5 is shown in Fig 3, and these findings supported the gene expression analysis.

Gene expression in peripheral blood
Of the blood samples drawn from 10 patients with MM, we successfully performed gene expression profiles in 8 out of 10 drawn on day0 and day21, but only 2 out of 10 drawn on day2 (MM20 and MM05). Consequently, we performed an unsupervised global gene analysis of peripheral MNCs by comparing the MM day0 samples to CON samples (disease vs. healthy) and MM day0 samples versus MM day21 samples (before vs. after treatment). We found that two genes that encoded B-cell surface markers (CD22, CD200) were downregulated in MM day0 samples compared to CON samples, and these genes were further downregulated on day21, though the differences from day0 were not significant. The levels of CD22 and CD200 expression were independent of the mucositis grade.

Discussion
This study described the gene signature of buccal mucosa samples from patients with MM during HSCT. We found that this signature was dominated by altered expression of inflammatory and anti-apoptotic genes, but expression was independent of the presence of clinical mucositis. Furthermore, we identified a specific isoform of the immunomodulatory gene, HLA-DRB1, which may serve as a biomarker for mucositis severity.

Gene symbol FC p-value adjusted p-value Qualified GO term Function
The process that leads to mucositis is triggered immediately upon initiation of cancer therapy and before any visible macroscopic damage [10]. Moreover, eventual UM coincides with neutropenia, within 7-10 days of starting chemotherapy. Therefore, to avoid compromising patients with neutropenia, we acquired the second biopsy before the onset of neutropenia. With this approach, the biopsy was unlikely to comprise disintegrated tissue that, presumably, would be dominated by inflammatory mediators. Instead, we aimed to gain insight on the cellular processes that gave rise to the inflammatory state. Mucositis lasts for approximately 7 days, and then, it spontaneously resolves. We acquired the third biopsy on day 21, when the mucosa was fully restored.
Little is known about the effects of melphalan on normal epithelium, but in cancer cells, melphalan induces oxidative stress and upregulates a wide range of apoptosis-related genes [8,29], consistent with our findings of EDA2R upregulation. EDA2R encodes a TNF receptor that mediates the activation of NF-κB and jun-N-terminal (JNK) pathways, which lead to caspaseinitiated apoptosis. Previous studies reported that these pathways were activated in buccal mucosa of patients with various cancers that received HSCT [30][31][32], and in gastrointestinal mucosa of patients treated with 5-fluoruracil [33]. In contrast, we found upregulated expression of INPP5D, which encodes a membrane protein in hematopoietic cells. The INPP5D protein negatively regulates JNK signaling and limits Fas-FasL-induced apoptosis in T-cells found at mucosal surfaces [34]. In addition, we identified five genes involved in suppressing the pivotal p53 apoptotic pathway. Four of these were upregulated: MDM2, CUL9, E2F7, and TIGAR; and one was downregulated: SERPINB10. The MDM2 gene encodes a protein ligase that ubiquitinates p53, and thus, inhibits p53-mediated cell cycle arrest and apoptosis. A previous gene expression study on three patients with acute myeloid leukemia used the same time intervals between biopsies that we used, and they found MDM2 upregulation [35]. In an array of studies, p53 has been identified as a key regulator of apoptosis, which leads to mucositis [11,12,36]. However, the gene alterations associated with apoptosis observed in our study, including EDA2R, did not depend on the level of clinical mucositis.
We found several genes related to transcription that were altered to favor DNA repair. For example, POLH was upregulated; POLH encodes a specialized DNA polymerase that accurately replicates UV-damaged DNA. Conversely, members of the histone cluster family (e.g., HIST1H1A), NR1D1, and NR1D2 were downregulated. These results implied that defense against apoptosis and DNA damage was a central objective in the initial stage of mucositis. Importantly, this objective was independent of the mucositis grade, which implies that other factors must be involved in distinguishing UM and NM.
The immune response was activated at an early stage, through the upregulation of TREM2 and LAMP3 on day2. Both these genes encode DC membrane proteins that contribute to Tcell activation and mucosal inflammation. The LAMP3 gene is specifically expressed in mature DCs [37,38]. TREM2, which is expressed on both DCs and macrophages, can bind and phagocytose yeasts, Gram positive bacteria, and Gram negative bacteria [39], which are commonly present in the oral cavity [40]. Generally, DCs are potent antigen-presenting cells that respond to microbial exposure by secreting abundant cytokines; e.g., IL-12 and type I interferon. In turn, IL-12 mobilizes natural killer (NK) cells. The NCR3LG1 gene, which encodes a ligand that triggers NK cells, was also upregulated [41]. Several studies have shown that an important aspect of mucositis pathology is the thinning of the epithelium, in combination with changes in the composition and concentration of the oral and gastrointestinal microbiota [13,42,43]. Our results confirmed the notion that the host immune response towards the microbiome played a dominant role, early in mucositis pathogenesis; however, these responses were not associated with mucositis severity.
Among several genes associated with metabolism, ABCA12 and CEL were upregulated 4.7-fold and 4.6-fold, respectively, on day2. ABCA12 encodes a membrane transporter protein primarily involved in the keratinocyte lipid-barrier that maintains homeostasis in the epidermis [44]. To the best of our knowledge, no previous study has described a role for ABCA12 in the mucosa, but it most likely performs a similar function of barrier protection. CEL encodes a lipase with multiple functions in lipid metabolism; it is also expressed in macrophages [45]. The expression levels of both these genes were unaffected by clinical mucositis.

Gene alterations associated with clinical mucositis grade
When we compared the gene expression profiles between the six patients with NM and the four patients with UM, we found two genes that were more highly expressed in NM patients: HLA-DRB1 and HLA-DRB5 (Fig 1). The HLA-DRB1 and HLA-DRB5 genes are related members of the MHC Class II family, located on chromosome 6p21.32. They encode surface proteins that are almost exclusively expressed on specialized antigen presenting cells, including macrophages, B-cells, and DCs or Langerhans cells [46,47]. These surface receptors function as a ligand for the T-cell receptor, and their primary function is to capture potentially foreign antigens on the cell surface and to present them for recognition by CD4+ T-cells [48]. Thus, they form a communication between the innate and adaptive immune systems, and determine whether to bring forth resistance or tolerance, in addition to taking up and processing dying cells [49].
Several pharmacogenomic GWAS studies were recently performed on drug toxicity, which showed that HLA Class I and II paralogs were associated with toxicity [50] or inflammatory mucosal conditions [51,52]. Even more interestingly, certain HLA-DRB1 alleles (HLA-DRB1 Ã 15) have been detected in patients with MM that were exposed to bisphosphonates and developed osteonecrosis of the jaw [53].
We conducted a search for alternatively spliced variants of HLA-DRB1 and HLA-DRB5 and found two isoforms of HLA-DRB1 (NM_001243965 and NM_002124). According to the UCSC genome browser, NM_001243965 harbors six exons, and NM_002124 harbors an extended isoform within seven exons. We found that patients with NM expressed the longer transcript variant of HLA-DRB1 (NM_002124), and patients with UM and healthy subjects (CON) expressed the shorter variant (NM_001243965). No splice variant was found for HLA-DRB5; however, that gene was expressed at different levels. We found that HLA-DRB5 was expressed 4.5 to 5 times more frequently in the NM group than in the UM group, in the first two of three biopsies. These findings were confirmed in an immunohistochemical analysis of HLA-DRB5 in the biopsies (Fig 3). Furthermore, cells that stained positively for HLA-DRB5 were localized primarily in the epithelium and submucosa, relatively close to the basal membrane, and these cells displayed a morphology similar to DCs (Langerhans cells).
In contrast to the other healthy subjects, CON09 (a patient with psoriasis) expressed the long transcript variant of HLA-DRB1 and a high level of HLA-DRB5 protein, similar to patients in the NM group. Previous reports have indicated that patients with psoriasis are 77% less likely to develop mucositis [10,54]. Psoriasis is an auto-inflammatory skin disorder with reduced apoptosis. It is known that patients with psoriasis express certain HLA-DRB1 alleles [55]. Our results suggested that mechanisms related to inflammatory and/or apoptotic pathways may be common in psoriasis and low-grade mucositis in MM. In addition, ABCA12 expression was upregulated in CON09 compared to the other healthy subjects. Previous gene expression studies on patients with psoriasis confirmed this finding [56].
The two recent GWAS studies on patients that received HSCT identified SNPs near the locus of MMP and other genes related to inflammation, but none related to HLA-DR [17,18]. Other recent studies found a major role for MMPs in mucositis pathology [57]. We did not find any changes of that nature in our material. The phenotypes described in our study may provide additional information to guide future GWAS studies [50]. Recent studies have shown that induction therapy with immune modulating agents reduced the frequency and severity of mucositis [58]; our results may provide additional knowledge to elucidate the development of those therapies.
In peripheral blood, we did not find any differences in gene expression between NM and UM groups, at any time point. However, among all patients with MM, CD22 and CD200 were downregulated on day0 compared to controls. Both these genes encode cell membrane glycoproteins of the immunoglobulin superfamily. CD200 is expressed in multiple cell types, including B-cells, Tcell subsets, DCs and endothelial cells. In contrast, CD22 is exclusively expressed on mature B-cell lineages [59,60]. Low CD200 expression has been linked to prolonged survival among patients with MM [61]. In our cohort, the lowest CD200 expression levels were observed among patients with UM; however, the levels were not significantly different between UM and NM groups.

Study strengths and limitations
There was some concern that breaking the mucosal barrier by taking a biopsy during chemotherapy might lead to potential fatal infections. We could reject this concern, because none of our patients experienced any infection related to the biopsy; only mild discomfort was reported. This finding was also reported in previous studies [15,35,36,62]. Therefore, we concluded that our method would be safe for patients undergoing HSCT, provided that the second biopsy is taken before the onset of neutropenia. The major limitation of the study was the low number of subjects. We designed the study to identify genes that were altered by more than 2-fold between time points, with a false discovery rate of less than 0.05% and a power of 90%. However, because our method of harvesting human mucosa during high-dose melphalan treatment was controversial, we sought to include the least possible number of patients required to draw valid conclusions. However, we recognize that the power of this study was set to estimate any fold-changes above two, and false negative findings may be concealed. Consequently, we did not expect to elucidate the full, true picture; nevertheless, we brought to light some important biological associations, which have motivated us to continue this research and confirm the results in a larger cohort.

Conclusions
Currently, there is a great need to develop a clinically applicable method for identifying potential susceptibility to toxicity among patients before treatment initiation. We found that patients with NM displayed upregulations of HLA-DR1 and HLA-DRB5 compared to patients with UM and healthy individuals. These genes encode proteins expressed on the surface of antigen presenting cells in mucosa, which suggested that the immune response might play a major role as a primary effector in UM. Indeed, the results suggested that expression of a certain isoform of HLA-DRB1 might diminish the inflammatory response to melphalan toxicity. However, because the levels of HLA-DRB1 and HLA-DRB5 expression were constant throughout treatment, this isoform may serve as a predictor of UM. The findings in this study were based on a small number of samples, and thus, our results must be validated in a larger patient cohort.