Epigenetic Regulation in Myelodysplastic Syndrome as A Consequence of A MicroRNA Dysregulation.

The myelodysplastic syndromes are a group of diseases characterized by impairment in hemopoiesis and variable tendency to progress to acute myeloid leukemia. Recent data regarding the MDS pathogenesis highlight alterations in several compartments of DNA replication, cell cycle control, and gene expression machinery. As previously reported, miRNA families are involved in MDS pathogenesis. Also, different miRNAs have been characterized in AML and MDS and allow the identi�cation of prognostic risk categories independent from the revised international prognostic scoring system (R-IPSS). Herein we report the results and the possible scenarios regarding pathogenesis and disease progression secondary to the evidence of considerable gene network derangement following miRNA evaluation in a cohort of patients.


Introduction
The myelodysplastic syndromes (MDS) are a group of diseases characterized by impairment in hemopoiesis and variable tendency to progress to acute myeloid leukemia (AML).
The World Health Organization (WHO) classi cation divides the MDS into different categories such as: refractory anemia with excess of blast (RAEB), refractory anemia (RA), refractory cytopenia with multilineage dysplasia (RCMD) 1 .Chronic myelomonocytic leukemia (CMML) is a myeloid entity between MDS and myeloproliferative neoplasm; it can have dysplastic features, and it has also tendency to transform in AML 2 .
Several reports showed an abnormal expression of miRNAs in several hematological neoplasms 7 .
Considering that miRNAs act as post-transcriptional regulators of several proteins, their role in the pathological epigenetic regulation of MDS is becoming progressively relevant 8 .
The micro-array-based platform and SNP-A studies showed alteration in miRNA genes (obtained from either methylation of their promoter or genetic loss of material or gene mutation) that allow pathologic expression of miRNA patterns associated with distinct phenotypes of disease.
As previously reported, miRNA families are involved in MDS pathogenesis.Indeed the deletion of miRNA-145 and miRNA-146 located on chromosome 5 leads to the 5q-syndrome 3,9 .Also, different miRNAs have been characterized in AML and MDS and allow the identi cation of prognostic risk categories independent from the revised international prognostic scoring system (R-IPSS) 10 .miRNAs alteration may have oncogenic properties or act as a tumor suppressor and have an active role in the onset of myeloid disorders.
In the last 5-10 years, next-generation sequencing (NGS) has been introduced in most hematologic laboratories, with various myeloid NGS panels now being commercially available.
The detection of these mutation can help to exclude benign causes of cytopenias in cases with nondiagnostic morphology.Considered the relatively high frequency of clonal haematolopoiesis in elderly patients, the presence of mutations in MDS-associated genes should be interpreted with caution and does not establish a diagnosis of MDS in isolation 12 .
A prospective study-based miRNA pro ling was launched at Policlinico San Martino in 2017.
The aim of this paper is to report the results and the possible speculation regarding pathogenesis and disease progression.

Patients selection and bone marrow processing
Twenty-eight patients were considered eligible for this study because they met our inclusion criteria: (i) availability of bone marrow sample at diagnosis, (ii) IR-PSS, and (iii) clinical follow-up.
Patients were allocated to a different type of MDS according to the WHO 2008 classi cation as per investigator choice.
Eleven patients were affected by RAEB, nine by RA, ve by RCMD, and three by CMML.The median age was 66 years old (48 -85).Eight patients evolved into acute myeloid leukemia.
Six patients were alive, 17 died due to disease progression, six were lost to follow up.
The population was also strati ed according to R-IPSS at diagnosis as it follows: 6 low, three intermediate-1, six intermediate-2, 13 high.
NGS analysis was performed on bone marrow aspirate at diagnosis with Myeloid Oncomine Panel Thermo sher; miRNA expression was quanti ed on bone marrow aspirate with TaqMan Advanced miRNA Cards.Data analysis was censored in January 2018.The study was approved by the local ethical committee at Policlinico San Martino; all the procedures were in accordance with local guidelines and regulations.All the patients provided written informed consent regarding the utilization of bone marrow samples for research purposes.
miRNAs selection within the WHO miRNAs signature To identify which genes are in uenced by the miRNAs signature, we proceeded with labeling those miRNAs that are expressed.The average expression of each miRNA was calculated, and the expression value of each miRNA in each patient was compared with its average expression: those miRNAs having expression values above their average were considered expressed.In contrast, the remaining were considered not expressed.
After this labeling phase, for each WHO class, we proceed with selecting those miRNAs that are expressed in most patients.Since the number of patients is not evenly distributed among the four WHO classes, we applied the thresholds shown in Table 1.In the context of a machine learning supervised analysis, signi cant differences in the miRNAs expression among the patients classi ed based on the WHO 2008 classi cation were evaluated.This analysis was conducted using l1l2 13 -penalized regularization method within PALLADIO 14 , a machinelearning framework that can provide robust variable selection in high-dimensional problems.
When a signature of miRNAs able to discriminate the four WHO classes were identi ed, a phase of miRNAs target prediction was performed.We considered six prediction tools and all the target genes resulting from this phase.The lists of genes targeted by the miRNAs signature were characterized through an enrichment analysis using the webtoolkit WebGestalt.Within this tool, we considered the Kyoto Encyclopedia of Genes and Genomes (KEGG), a database of manually curated biological pathways.
Among the pathways found enriched, we selected the pathways characterizing one speci c WHO class and those in common to more classes.Successively we selected those common pathways to all four MDS classes, which were modulated/regulated by most of our signature miRNAs.Considering these two criteria, we highlighted 5 KEGG pathways that played signi cant roles across the four WHO classes of MDS.These pathways are: AMPK signaling pathway 15 , RAS signaling pathway 16 , EGFR tyrosine kinase inhibitor resistance 17 , longevity regulating pathway, RAP1 signaling pathway 18 .
Further experimental procedures are described in the supplemental material.

Results
miRNA expression and WHO subtypes.
Each patient carried a distinctive miRNA signature composed of a median of 650 miRNAs.
The supervised analysis identi ed a signature of 24 miRNAs (Table 2) able to discriminate the 4 WHO classes with 62% of accuracy and p-value < < 0.001.

Target Gene Prediction
Based on the miRNA expression result, prediction tools for miRNA target gene prediction were used as explained in the supplemental material.
The target genes identi ed were compared to the complete list of genes known to be involved in MDS from the related page of the Online Mendelian Inheritance in Man (OMIM).
This latter list is composed of the following three groups of genes: TET2, SF3B1, ASXL1, and GNB1 belong to the group of genes characterized by somatic mutations; GATA2, TERC, and TERT belong to the group of genes characterized by heterozygous germline mutations that predispose patients to MDS and other myeloid disorders and the last group of genes, which are DNMT3A, U2AF1, and JAK2, are characterized by heterozygous somatic mutations.
When we compared the lists of targets, coming from the miRNAs signature identi ed in the multiclass task, with the above-described OMIM list, we found the rst and the last group of genes be involved in disease expression in our cohort of patients.
Our miRNA signature did not identify the genes belonging to germline predisposition.

Functional characterization of the target genes list
Aiming to identify which gene pathways were involved (mostly silenced) by the miRNA signature detected, a functional analysis using the webtoolkit WebGestalt was used 19 .
Within this tool, we considered KEGG a database of pathways 20 .
The miRNA enrichment analysis allowed the identi cations of genes involved in (i) pathways (some in common and some speci c for a WHO class) likely involved in the onset of MDS, (ii) in pathways of diseases related to MDS and disease evolution like the acute myeloid leukemia, as summarized in gure 3.
Through the miRNA signature and its enrichment, it was possible to detect some known gene pathways involved in disease pathogenesis and progression, such as: cAMP signaling, mTOR, hedgehog, TGF-beta signaling, hypoxia-inducible factor, and phospholipase D signaling.
It was also possible to identify other gene pathways that may contribute to some of the symptoms reported by MDS patient; the miRNA signature is suggesting likely the silencing of different genes involved in the T-cell receptor signaling 21 (i.e., AKT serine/threonine kinase 1-2-3, CD3e, CD4, CD8a, CD20, and CD40 ligand), FC gamma receptor, calcium signaling, adherens junction, and CD46 22 .

Functional characterization in RAEB patients
A speci c focus was directed to those patients affected by RAEB because they had detectable blasts and likely had more DNA damage.
Among the miRNAs expressed in most of the patients belonging to the RAEB class, miR-1270 and miR-1179 emerged because the functional characterization of their target genes achieved good enrichment results.Both are targeting TET2, likely as suppressors of its expression.
The enrichment in KEGG gave a statistically relevant association of 16 genes pathways targeted by miR-1270, whereas miR1179 can in uence 19 genes pathways (table 3 and table 4 show which gene pathways are controlled).Among these 16 genes pathways, NF1, ABL1, and TP53 were also detected with our NGS analysis (not all patients with RAEB).
Since the TET2 gene is essential in the pathogenesis of MDS and can be prognostic in terms of therapy 23 , it was assessed the possible interactions of this gene with those belonging to some interesting KEGG pathways listed in Table 3 and 4 using the web tool STRING.
It was possible to detect 85 protein-protein interactions between TET2 and genes of the "Signaling pathways regulating pluripotency of stem cells" (in uenced by miR1270 and miR1179) (Figure 4).
This network shows that the interaction between TET2 and TCF3 is experimentally known 24 (pink edge), while the interactions between TET2 and the other three proteins, that are JARID2, KRAS, and JAK2, derived from co-expression experiments in homologs proteins in other species.
JARID2 encodes for a transcriptional repressor that interacts with the Polycomb repressive complex 2 (PRC2), which plays an essential role in regulating gene expression during embryonic development.This protein facilitates the recruitment of the PRC2 complex to target genes.Mutations in this gene are associated with myeloid malignancies 25 .
Shared miRNAs and early steps in MDS pathogenesis.
The miRNA signature and its enrichment also allowed to identify of a list of miRNAs whose targets are involved in regulating some genetic pathways that are in common between the different classes.
This result suggests that several genes in different gene pathways may be affected by these eight miRNAs.Table 5 is summarizing which biological processes are in uenced.Since these miRNAs are present in all patients, it is reasonable to suggest that these may be the ones to be deranged likely at the onset of disease.Also, they are likely working as post-transcriptional suppressor in gene pathways that have been described to be fundamental for normal haemopoiesis 26,27,28 .
STRING tool allowed to detect protein-protein interactions between the genes controlled by the miRNA signature and the genes mutated detected in NGS ( gure 5).
Three genes are outliers in STRING: RAPGEF5, CSDE1, MSFD1.This result is suggesting that even if the miRNA signature may negatively in uence the expression of these genes, they could be not involved neither in MDS pathogenesis or disease progression.

Discussion
MDS are a group of diseases with different clinical behavior arising from a complex pathogenesis.
Several mechanisms of DNA-damages can be detected through SNP karyotyping, NGS, and miRNA assessment.
The present study evaluated the presence miRNA expression in a cohort of patients affected by MDS.
As described by other groups, we identi ed different mutations via NGS in known genes involved in pathogenesis and disease progression 29,30 ; anyway, a 43% rate of gene mutations underestimates the actual mutation rate in MDS.Our result may be a consequence of old samples tested for NGS; indeed, DNA degradation is a well-known mechanism for failure of NGS analysis 31 .
We decided to focus on miRNA analysis for several reasons: (i) miRNA are well known mechanism of post-transcriptional regulation, (ii) some miRNAs have been associated with some speci c type of MDS.
Our analysis identi ed a signature able to discriminate different types of MDS according to the WHO classi cation (p-value < < 0.001).
It's worth to highlight that the system of MDS classi cation is based on cells morphology, degree of pancytopenia and karyotyping.For the exception of SF3B1 and deletion chromosome 5q, which are ndings de ning a speci c subgroup of MDS, there is no test in molecular biology that could support a speci c type of MDS diagnosis.This analysis gives a different contribution in de ning speci c types of MDS.
Also, the miRNA enrichment allowed a deepest analysis of miRNA de-arrangement in our population: the gene target prediction of our miRNA signature was able to detect genes involved in RNA splicing and epigenetic mutation but not those involved in germ line predisposition to myeloid malignancies.This nding suggests that MDS arising in patient with the latter group of mutations may have a different pathogenesis even if some mechanism of disease onset and progression are shared with cases de novo.
Further, the miRNA enrichment was able to identify in our population genes involved in MDS pathogenesis and disease evolution to AML.It is interesting that the miRNA signature in this study in uences genes involved in the immune system; this result may be the starting point to speculate about infective risk, opportunistic infection and autoimmune events in MDS.Anyway, this aspect was not explored due to the lack of clinical data about infective events and because our study population is relatively restricted.The patient's functional characterization affected by RAEB suggested a disarranged gene expression in this population.We detected higher rate of NGS mutation in our population, but we identi ed two miRNAs (mir-1270 and mir-1179) with an interesting enrichment: it suggests a possible broad genetic lesion as suggested in Fig. 5.
The STRING analysis showed several protein-protein interactions likely in uenced by our miRNA signature.
Even if there is no mutation all genes illustrated in Fig. 5 we think they are silenced the miRNA signature identi ed in our study contributing to the onset of RAEB.
It is well known that mutation in TET2 are causing loss of function; this study showed the miR-1270 and miR-1179 are targeting TET2, causing its silencing.
It is reasonable to suggest that this miRNA expression and gene mutations contribute to the clinical/morphological aspect of the disease.
The availability of a broad miRNA signature allowed also a different point of view: it was possible to identify 8 miRNAs that focus on speci c gene pathways common to all patients.
The involvement of AMPK, EGFR, RAS, longevity, RAP1 pathways is in speci c genes (Table 6) and some of them (i.e.IGF1R 32 , PRKAA2 33 , TSC1 34 ) have key role in all the pathways mentioned above; this suggests a redundant mechanism of disease progression in keeping with the effect of miRNAs on different gene pathways.
Table 6 -lists regulated by 7 or 8 miRNAs involved in ve pathways common to all the WHO classes.The 8-focused miRNAs are targeting common genes within this common list of pathways, suggesting a redundant mechanism of disease progression Also, these genes and genetic pathways have been previously described to be fundamental for normal hemopoiesis and cell maturation; it is plausible that the 8-focusing miRNAs are the one likely to be involved in early phase of the disease and are likely the rst mechanism of cell alteration/disease progression.Anyway, our results are not showing if the 8-focusing miRNAs are the cause of altered epigenetic in MDS or are consequence of it.
It is reasonable to think that these early miRNAs are in uencing genes fundamental in normal cell processes and in regulation of other miRNAs; this would a domino effect in MDS patient, where there is common miRNAs and genes that become patient speci c with disease progression due the involvement of other genes and miRNAs.This is an interesting theory that needs further in vitro validation, but a similar mechanism was already described in pediatric acute lymphoblastic leukemia 35 .
Anyway, some limitations can be detected in this study: small and retrospective nature of the study and non-selection of CD34 + cells.Considering the presence of difference clones in MDS population, our miRNA signature may not represent the expression of only one clone 36 .The dominant one may in uence the miRNA signature hiding the ones from minor clones or the latter may "contaminate" the miRNA signature detected.Also, the small population is not allowing the identi cation of other miRNAs WHO speci c and it is not possible to predict a miRNA signature that can be prognostic in term of AML transformation or sensitiveness to hypomethylating therapy.
In conclusion, the miRNA signature and its enrichment may become potential tools for de ning the diagnosis of each MDS subtypes; the integration of this strategy with NGS needs a larger and homogenous population to further de ne which genetic pathways are involved and which miRNAs may be useful to further understand the pathogenesis of MDS.
Also, this study is suggesting that the epigenetic regulation in MDS is not only a consequence of altered methylation but it is also a consequence of a miRNAs dysregulation causing a broad genetic damage.STRING network of the Hematopoietic stem cell self-renewal ability highlighted in red.We obtained this network using STRING database.In this context TET2 interact with TCF3, JARID2, KRAS and JAK2 proteins.In red are highlighted those proteins that are mentioned together in one article (PMID: 23675967) focused on the therapeutic potential of the mechanism underneath the HSC self-renewal ability which could be targeted to prevent leukemic stem cell self-renewal in myeloid malignancies.STRING network between the genes common to all the 5 pathways and the genes measured in our NGS experiment.In this PPI analysis we want to observe the possible connections between the genes we identi ed to be regulated by the 8 miRNAs involved in the 5 pathways common to all the WHO classes and the panel gene list measured in our NGS experiment.The PPI enrichment p-value of this network is signi cant (< 1.0e-16).The pink and the blue sky edges represents the strongest connections because

Declarations Figures
on bone marrow sample at diagnosis

Figure 2 Frequency
Figure 2

Table 1 -
-miRNAs thresholds and number of subjects in the 4 WHO classes.For each WHO class it is shown the number of subjects belonging to that class and the thresholds applied for considering a miRNA expressed in that class.

Table 2 -
-Signature of the twenty-four most frequent miRNAs.Freq is the frequency that indicates the number of times each miRNAs was selected over the number of repetitions of regular and permutation experiments (see Materials and Methods).Here we chose 100 repetitions subdivided in 50 regular and 50 permutation experiments.

Table 3 -
-Result of the enrichment analysis in KEGG of the miR-1270 target genes list.The genes predicted to be regulated by miR-1270 were functionally characterized by enrichment analysis in KEGG database.The sixteen pathways shown in this table derived from a selection of the most relevant in the MDS context.FDR, False Discovery Rate.#Genes, number of genes belonging to our list of genes predicted to be regulated by miR-1270.

Table 4 -
-Result of the enrichment analysis in KEGG of the miR-1179 target genes list.The genes predicted to be regulated by miR-1179 were functionally characterized by enrichment analysis in KEGG database.The nineteen pathways shown in this table derived from a selection of the most relevant in the MDS context.The pathways in bold are in common with the ones found enriched in the miR-1270 functional analysis.FDR, False Discovery Rate.#Genes, number of genes belonging to our list of genes predicted to be regulated by miR-1179.

Table -
List of signi cant GO terms belonging to biological process and molecular function and of signi cant KEGG pathways related to STRING network shown in Figure5..This table show the enrichment in Gene Ontology (GO) and in KEGG related to the gene list in input to STRING in order to identify the network shown in gure 5. FDR, False Discovery Rate.