miR-146a and miR-155 Expression Levels in Acute Graft-Versus-Host Disease Incidence

Allogeneic hematopoietic stem cell transplantation (allo-HSCT) is a curative treatment for numerous hematological malignancies. However, acute graft-versus-host disease (aGVHD) is still the major complication causing mortality. MicroRNAs (miRNAs) play a significant role in inflammation and have potential as prognostic and diagnostic biomarkers. This study investigated the role of two immune-specific miRNAs (miR-146a and miR-155) as biomarkers for aGVHD incidence in the peripheral blood of allo-HSCT patients prior to disease onset. The study showed that miR-146a and its statistical interaction with miR-155 at day +28 were predictive of aGVHD incidence. Interestingly, the expression levels of miR-146a and miR-155 negatively correlated with the transcription factor, SPI1 (PU.1gene) mRNA expression.

inTrODUcTiOn Graft-versus-host disease (GVHD) is a serious complication of allo-HSCT. Understanding the molecular pathophysiology of GVHD, in particular aGVHD, may aid patient stratification for tailored treatments. MicroRNA (miRNA) investigations have identified biomarker signatures for numerous diseases such as cancer and rheumatoid arthritis and more recently for GVHD (1)(2)(3)(4). Since miR-146a and miR-155 are both involved in the adaptive and the innate immune system (5,6), we assessed their role in allo-HSCT. miR-146a is essential for Treg function (7). TNF receptor-associated factor 6 (TRAF6) and interleukin-1 receptor-associated kinase 1 (IRAK1) are miR-146a targets, and their expressions are regulated in a negative feedback-loop via the TLR-4 (Toll-like receptor) signaling pathway and NFκB activation (8,9). IFN regulatory factor 5 (IRF5) and signal transducer and activator of transcription 1 (STAT1) are also known targets of miR-146a (10). Tang et al. (10) showed that overexpression of miR-146a lowered STAT1 and IRF-5 at both the molecular and the protein level. Moreover, it has been shown in mice that miR-146a depleted Treg cells have high levels of Stat1 and this results in IFN-γ mediated autoimmunity (7). In a recent GVHD murine study, miR-146a was under-expressed in the aGVHD group and the expression correlated with disease severity, while its target, Traf6, was overexpressed, suggesting the involvement of NFκB activity (4).
miR-155 is required for normal B and T lymphocyte function in humans (11). miR-155 is necessary for CD8+ T cell proliferation, and its deficiency can lead to the overexpression of STAT1, leading to down regulation of proliferation of T cells (12,13). In a GVHD patient cohort (n = 5), miR-155 was overexpressed in the gut of aGVHD patients while its expression was absent in normal gut (1). Recently, a murine study has shown that miR-155 expression activates inflammasomes via the migration of dendritic cells (14). Thus, inhibiting miR-155 expression may reduce GVHD severity.
In addition to TRAF6, IRAK1, STAT1, and IRF5 the important transcription factor, SPI1 (PU.1), which functions in hematopoietic differentiation (15), also regulates the expression of miR-146a and miR-155 (16). In a murine study, it was demonstrated that Spi1 regulates the expression of both miR-146a and miR-155 by binding to their pre-miRNA loci (17). Hence, the primary focus of this investigation was miR-146a and miR-155 expression levels because of their known functions in both the adaptive and innate immune system. The aims of this investigation were: (1) to assess whether miR-146a and miR-155 were associated with aGVHD incidence and severity as well as with clinical outcomes including overall survival (OS) and relapse, (2) to establish a statistical model for predicting aGVHD incidence and severity using the miRNA expression levels and (3) to examine the correlation between miR-146a and miR-155 expression with their targets (IRAK1, TRAF6, STAT1-α, IRF5, and SPI1).
In summary, this research has shown that miR-146a and miR-155 are lower in patients who go on to develop aGVHD (grade I-III). SPI1 is regulated by both miRNAs during the early stages of inflammation, and there was no correlation between the miRNAs and their targets.

MaTerials anD MeThODs ethics and consent
The project was approved by the Newcastle and North Tyneside I Research Ethics Committee (Stemdiagnostics: 07/H0906/131). All the investigations were conducted in accordance with the Helsinki Declaration. The patients gave their informed written consent for whole blood, serum sample collection, testing, and analyses.

Patient study cohort
The cohort consisted of 54 patients who had undergone allo-HSCT (2010-2013) with either Human Leukocyte Antigen (HLA) matched siblings or matched unrelated donors ( Table 1). There were only seven cases of "female donor-to-male recipient" and overall there were 21 patient deaths (events) and 33 patients who were alive at the end of the study (censored cases). This study included only patients who developed no aGvHD or who developed aGVHD within the first 100-days post-transplantation (classic aGVHD). Patients who received donor lymphocyte transfusions, or started immunosuppression withdrawal, before 100 days were excluded. None of the patients had developed GVHD prior to day 28. Classic maximum aGVHD grade [disease onset: median = 62 days (range: 33-100)] was assessed as per the NIH consensus (18) [grade 0: without (n = 26), grade I: mild (n = 17), grade II: moderate (n = 10) and grade III: severe aGVHD (n = 1)]. The incidence of chronic GVHD (cGVHD) was significantly associated with prior aGVHD ( Table 1). In the patient group who did not develop aGVHD (grade 0), 50% had died from disease relapse and 30% due to infections, while in patients with aGVHD grades I-III, 37% had died of relapse and 46% due to infections. In total two patients had died of cGVHD and an additional two due to other complications. Whole blood was collected from patients 7 days before HSCT (n = 35), at 28 days (n = 54) and at 3 months (n = 34) post-transplantation. At day +28, none of the patients were on steroids nor had they received donor lymphocyte infusion, and all patients survived until day 100. All patients had received alemtuzumab.

Total rna extraction from Whole Blood
Peripheral blood was collected in PAXgene Blood RNA Tubes (19). The samples were stored at −20°C prior to extraction. Total RNA was extracted using the PAXgene Blood miRNA kit (Qiagen/ BD Company) as per the manufacturer's guidelines. Total RNA quality was determined using the Nanodrop 1000 (Agilent, USA).

reverse Transcription and real-Time Pcrs for Micrornas and for gene expression
For miRNA expression assays (Life Technologies, USA): 1-10 ng of total RNA was reverse transcribed (RT) in a 15 μl reaction using Taqman specific reverse transcription primers [hsa-miR-146a-5p (Assay ID: 000468), miR-155-5p (Assay ID: 000479), and SNORD48 (Assay ID: 001006)] and the Taqman miRNA Reverse Transcription kit. The RT PCR step was performed as per the manufacturer's guidelines. miRNA-specific cDNAs were used to run qPCR using miRNA Taqman primer-probe sets and TaqMan ® Gene Expression Master Mix. SNORD48 was used for normalization.

statistical analyses
Analyses were performed per time-point (day −7, day +28, and  To obtain the best model fit, and thus the most accurate predictions, aGVHD incidence, which is recorded using a binary scale (0 = no aGVHD or 1 = aGVHD grades I-III), was assessed using a forward step-wise, multiple binary logistic regression procedure (SPSS version 21.0). This uses the logit function to model the probability (or "odds") of a patient having or not having aGVHD, given their clinical background and miRNA results. In this procedure, the significant clinical risk factors were considered as covariates, as well as the effects of miR-146a and miR-155 expression levels, and to give the best model fit, their statistical interaction was also considered (21,22).

Survival Analyses
Univariate Kaplan-Meir and Cox regression were used to determine candidate clinical risk factors for time-dependent outcomes OS and relapse (p ≤ 0.2). Multivariate Forward-LR Cox regression analysis was performed to model these outcomes (SPSS version 21.0).

Correlation of miRNAs with Targets
Spearman correlation was performed to test for strength of association between miRNA expression levels and their targets. The correlation was performed separately for those who never developed aGVHD (grade 0) and those who went on to develop grade I-III aGVHD.
Kruskal-Wallis analysis of variance was performed when there were three or more groups in the comparison, and Dunn's post hoc test was used to safeguard against false positives (GraphPad Prism version 5).

resUlTs
The statistical interaction between mir-146a and mir-155 Was significantly associated with agVhD incidence at Day +28 Post-allo-hscT A hierarchical model was built using a forward stepwise procedure including significant clinical risk factors, miR-146a and miR-155 expression levels, and their interaction, separately for each time-point. Only day +28 data produced a significant model fit. The clinical risk factors initially determined to be of possible relevance (p ≤ 0.2) in predicting aGVHD incidence, and which were included in the step-wise procedure were: patient gender, underlying disease, HLA class II mismatches, conditioning regimen, and transplant protocol ( Table 1). Results showed that the clinical factors were not significant once both miRNA expression levels at day +28 were included in the model. The final model that resulted in the best overall fit and predictive ability included a significant statistical interaction between the two miRNAs ( Table 2: the "main effect" of each miRNA was included to enforce model hierarchy; only the main effect of miR-146a was statistically significant). By simple graphical or tabular examination of the raw data, we can see that this statistical interaction represents a biological synergy: if both miR-146a and miR-155 were expressed at low levels, this was associated with a higher incidence of aGVHD (Figure 1; Table 3). The classification matrix illustrating the predictive ability of the model showed that 69% of the patients were accurately classified as without aGVHD (grade 0) and 75% as with aGVHD incidence (grades I-III).
As the clinical risk factors were not significant in the final model, the equation that can be used for predicting the probability of a patient having aGVHD as a function of their miR expression levels is: probability (patient has aGVHD given miRNA levels) , β0 = intercept coefficient and β1-3 = regression coefficients, Log146a = −log2(miR-146a expression), Log155 = −log2(miR-155 expression).
In this study the probability (patient has aGVHD, given miRNA levels) In contrast, aGVHD severity (grades II-III) was not predictable from miRNA expression levels for this cohort, possibly because the majority of the patients had either none or grade I aGVHD (79.6%) post day +28.

mir-146a and mir-155 expression in Whole Blood Was not associated with relapse
Patient gender, transplant protocol, donor cytomegalovirus status, aGVHD incidence, and mismatches in HLA Class I were determined as significant clinical risk factors for inclusion in the model building. The multivariate Cox regression model showed that only mismatches in HLA class I was marginally predictive of patient risk of relapse [p = 0.048, HR = 4.744 (95% CI: 1.016−22.154)], and that the expression levels of both miR-146a and miR-155 were not significantly associated with relapse.
a Positive correlation Was Observed between the mir-146a Targets at the mrna level The step-wise logistic regression results, discussed previously, indicated that miR-146a was best for predicting aGVHD incidence; thus, to further understand its role in GVHD, its established targets (TRAF6, IRAK1, STAT1-α and IRF-5) were   Table 4).
In addition, correlation between miR-146a and its targets was also assessed in the aGVHD patient cohort (grades I-III). Measurements at day +28 showed that there was no relationship between miR-146a and its targets, but there was statistically significant (p ≤ 0.02) positive correlation observed between all the targets (data not shown). At 3 months post-transplantation, there was a significant negative correlation (rs = −0.69, p = 0.019) between miR-146a and IRF5 mRNA expression. At the same time-point, TRAF6 and IRAK1 mRNA expression positively correlated with each other (rs = 0.86, p = 0.001).

irF5 mrna expression Positively correlated with Protein levels Measured in Whole Blood at Day +28 in agVhD Patients
The target protein levels were quantified in a representative subset of the cohort (grade 0, n = 5; grades I-III, n = 10). Only STAT1-α and IRF5 proteins were detectable by ELISA. It may be that the TRAF6 and IRAK1 protein levels were below the sensitivity of the assay [detection range (picogram per milliliter), for TRAF6: 39-2500 and IRAK1: 312-2000].
The pre-transplant STAT1-α and IRF5 protein levels neither correlated with their mRNA expression nor with miR-146a-5p expression levels (p > 0.05) (data not shown). The day +28 protein levels of STAT1-α and IRF5 in the aGVHD grade 0 cohort significantly positively correlated with each other (rs = 0.90, p = 0.037). At 3 months post-transplantation, there was no correlation between miR-146a and its protein targets. In the aGVHD cohort (grades I-III), at Day +28, there was a trend for positive correlation between miR-146a-5p and STAT1-α protein levels (rs = 0.53, p = 0.092). A similar correlation was observed between STAT1-α mRNA and its serum protein level (rs = 0.57, p = 0.068). There was also a significant positive correlation (rs = 0.71, p = 0.009) between IRF5 mRNA and its serum protein level. At 3 months post-transplantation, there was a significant negative correlation (rs = −0.69, p = 0.019) between miR-146a-5p and IRF5. Some patients at 3 months were receiving steroids so this result must be interpreted with caution.
mir-155 and mir-146a expression correlated with the Transcription Factor, sPi1 In the aGVHD patient cohort (grades I-III) only, miR-155-5p expression was significantly negatively correlated with its target SPI1 in whole blood samples at day +28 (Figure 2A). In the same cohort, there was a significant negative correlation between miR-146a-5p and SPI1 ( Figure 2B).

DiscUssiOn
miR-146a and miR-155 have been shown to be indispensable in regulating immune-related pathways (6,7,23,24). It is known in HSCT that miR-155-5p is overexpressed when target organ damage has already occurred in the patient (1). It is also known that graft-versus-tumor effect is required to prevent disease relapse (25). Thus, tailoring treatments to maintain a balance between GVHD and graft-versus-tumor effect are important post-HSCT. This study aimed at assessing whether the expression of the two miRNAs could be used to predict aGVHD prior to disease onset. This knowledge could make clinical interventions possible prior to disease onset and, therefore, prevent the complications associated with aGVHD. Understanding the miRNA regulation involved in the pathophysiology of aGVHD could also be reflective of the GVH reaction and in particular the essential graft-versus-tumor effect, directing new therapies for improved HSCT disease-outcome.
The data models used in this study looked at both the main effects and the statistical interaction between miR-146a and miR-155 in predicting aGVHD incidence. In many biological investigations, in particular in the field of allo-HSCT, statistical interactions are often overlooked even though they are one of the keys to better fitting models (and thus better predictions) and to the better understanding of biological relationships. This investigation is one of the few studies where this method has been utilized for aGVHD incidence prediction.
The statistical interaction of miR-146a and miR-155 expression levels at day +28 post-transplant was significant in predicting aGVHD incidence, and investigation of the data revealed the conditional regulation mechanism wherein when both miRNAs were expressed at lower levels, there was a higher incidence of aGVHD. The statistical interaction observed is in support of the differential "check-point" mechanism shown by Schulte et al. (26). Using various immune system stimulators such as lipopolysaccharides that trigger the expression of miR-146a (8), the authors showed that miR-146a was upregulated first, thus initiating the inflammatory response. miR-155 expression was only altered when miR-146a levels of inflammatory tolerance had been surpassed (26). This may explain why higher miR-155 expression levels have been observed in both the serum and gut of patients at the time of aGVHD onset, which is the effector phase of the GVHD cycle (1,27,28). Our results support the very recent aGVHD study in the mouse that showed miR-146a expression levels were lower in severe aGVHD cases (4). This may suggest that overexpression of miR-146a prevents an inflammatory response and may have a protective role in innate immunity (23). miR-146a expression affects the expression of TRAF6 and IRAK1 at the protein level rather than at the mRNA-level (29). But in this study using ELISAs, the protein level of IRAK1 and TRAF6 were undetectable in the sera of allo-HSCT patients. IRF5 and STAT1-α proteins were detected in the allo-HSCT patients, but they did not correlate with miR-146a expression levels. IRF5 mRNA expression positively correlated with its protein levels at day +28 in patients who went on to develop aGVHD but not in patients who did not develop aGVHD. Not surprisingly, a positive correlation was present among miR-146a-5p targets at the mRNA level at both pre-and post-allo-HSCT time-points. It is at the augmented inflammatory stage that the miRNA expression reaches a level that impacts the expression levels of their targets. This may be the reason why no correlation was seen between miR-146a-5p and its targets pre-aGVHD onset, while the same miRNA has been shown to regulate Traf6 expression at aGVHD onset (4). Mice deficient in miR-146a develop autoimmune disorders and die prematurely (29). Low miR-146a expression could result in an inflammatory phenotype and impact survival (30). This may suggest that patients with lower miR-146a-5p expression would have more graft-versus tumor effect due to exacerbated GVH reaction, but this could lead to fatal GVHD. When the statistically significant clinical risk factors were included in the study, results showed that only HLA class I mismatches were associated with risk of death (data not shown). This finding may suggest that miR-146a expression was not associated with OS when clinical risk factors were included in the analysis. However, there were only three patient-donor pairs with HLA class I mismatch and thus this finding is preliminary and must be considered with caution.
The master transcription factor SPI1 (31) controls miR-155 expression (17). Both miRNAs are predicted to target SPI1 according to microrna.org (32). SPI1 controls the expression of both miRNAs in HSCs (17). Spi1 binds to pre-miR-146a locus permanently to regulate its expression and to miR-155 locus only during hematopoietic stem cell differentiation (17). Our results may suggest that during GVH reaction both miRNAs are triggered by chemokines and cytokines, which lead to their regulation of SPI1 rather than their activation by it (17,33).  (2,9,34). During the early stages of inflammation in GVHD, the TLR and Janus Kinase (JAK)-STAT1 pathway are activated by LPS and IFN-γ, respectively. Under such conditions, miR-146a-5p and miR-155-5p have reduced expression levels. This results in minor alterations of the mRNA target levels and may therefore represent subclinical early inflammatory GVHD.