MDS develops through a multistep process encompassing an initial deleterious genetic event within a HSC and successive genetic abnormalities, leading to clonal expansion and malignant transformation (27). In recent years, the understanding of the molecular pathogenesis of MDS has been markedly improved by next-generation sequencing (NGS), which has enabled the identification of a large spectrum of new mutations across all MDS subtypes. There are >40 significantly mutated genes in MDS, and these mutations account for nearly 90% of patients with MDS (28). Functionally, the mutations are grouped into several categories based on their prevalence: RNA splicing factors [splicing factor 3B subunit 1 (SF3B1), serine and arginine rich splicing factor 2 (SRSF2), zinc finger CCCH-type, RNA binding motif and serine/arginine rich 2 (ZRSR2) and U2 small nuclear RNA auxiliary factor 1/2 (U2AF1/2)], epigenetic regulators [Tet methylcytosine dioxygenase 2 (TET2), DNA methyltransferase 3α (DNMT3A) and isocitrate dehydrogenase (NADP(+)) 1/2 (IDH1/2)], components of the cohesion complex (stromal antigen 2, CCCTC-binding factor, structural maintenance of chromosomes 1A and RAD21 cohesin complex component), chromatin modifiers [ASXL transcriptional regulator 1 (ASXL1) and enhancer of zeste 2 polycomb repressive complex 2 subunit (EZH2)], transcription factors [tumor protein p53 (TP53), RUNX family transcription factor 1 (RUNX1), ETS variant transcription factor 1 (ETV1) and GATA binding protein 2 (GATA2)], signal transduction molecules [Fms related receptor tyrosine kinase 3 (FLT3), Janus kinase 2 (JAK2), MPL proto-oncogene thrombopoietin receptor (MPL), GNAS complex locus and KIT proto-oncogene receptor tyrosine kinase], RAS pathway [KRAS proto-oncogene GTPase, NRAS proto-oncogene GTPase (NRAS), Cbl proto-oncogene, neurofibromin 1 and protein tyrosine phosphatase non-receptor type 11 (PTPN11)] and DNA repair [ATM serine/threonine kinase, BRCA1/BRCA2-containing complex subunit 3, DNA cross-link repair 1C and FA complementation group L]. Mutations in RNA splicing and DNA methylation genes seem to occur early and are considered founder mutations in >50% of patients with MDS (28). Mutations provide a wide range of prognostic information, from benign to malignant and from good to poor overall survival (OS) time. For example, TP53, EZH2, ETS variant transcription factor 6 (ETV6), RUNX1, ASXL1 and SRSF2 mutations predict shorter survival time. The SF3B1 mutation is strongly associated with ring sideroblasts and thus has been included as a diagnostic criterion in MDS with ring sideroblasts (4).

There are several reports concerning differences in the mutational landscapes between hMDS and NH-MDS (Fig. 1). Nazha et al (29) compared the mutational profiles of 62 genes between patients with hMDS and NH-MDS. Patients with hMDS acquired fewer somatic mutations and had smaller driver clones compared with patients with NH-MDS. Splicing somatic mutations were determined predominantly in patients with NH-MDS, as driver clones were found exclusively in these patients. The study hypothesized that the immune system in patients with hMDS may suppress the driver clone by inhibiting its growth and genetic evolution, thus limiting the acquisition of downstream somatic lesions. Notably, some driver clones, such as SF3B1, SRSF2, TET2, ASXL1 and BCL-6 coreceptor (BCOR), may overcome this inhibitory effect (29). Yao et al (30) detected at least one gene mutation (17 genes) in 35% of patients with hMDS, and the most common mutation was an SF3B1 mutation. Patients with hMDS exhibited significantly lower incidence rates of RUNX1, ASXL1, DNMT3A, EZH2 and TP53 mutations, and a lower number of mutations per subject compared with patients with NH-MDS; however, the number was significantly higher in comparison with the number in patients with AA. Schwartz et al (31) used a whole exome sequencing approach to describe somatic and germline changes in pediatric MDS and found prevalent Ras/MAPK pathway mutations compared with that in adult MDS. Huang et al (6) did not find any difference in the incidence of RAS, acute myeloid leukemia 1 protein, JAK2, PTPN11 or FLT3/internal tandem duplication mutations between hMDS and non-hMDS. Bono et al (5) reported mutational data from a 24-gene panel on a large cohort of hMDS patients (n=93) and detected one or more somatic mutations in 38% of patients with hMDS. In comparison to non-hMDS patients (n=239), the patients with hMDS had a lower number of mutations per subject, but this number was significantly higher than that found in the patients with AA. The prevalence of splicing mutations (SF3B1 and SRSF2) and comutation patterns of TET2, DNMT3A and ASXL1 was lower in hMDS compared with that in non-hMDS. The integration of mutational data into a scoring formula enabled the separation hMDS patients with myeloid neoplasm-like profiles from those with non-malignant profiles. It was suggested that hMDS more likely represents a mixture of entities along a spectrum rather than a homogeneous in-between category (5).

Taken together, these results suggest that the mutational profile of hMDS overlaps with the profile of NH-MDS, except for the lower incidence of mutations in splicing factors and in ASXL1 and IDH1/2 genes. Patients with hMDS have fewer somatic mutations, and overall, smaller driver clones.

In AA, the most frequently mutated genes are phosphatidylinositol glycan anchor biosynthesis class A (PIGA), BCOR/BCOR-like 1 (BCORL1), DNMT3A and ASXL1, suggesting mechanisms of clonal selection. Mutations in PIGA and BCOR/BCORL1 are more specific to AA, while DNMT3A and ASXL1 mutations are also found in MDS (Fig. 1). PIGA somatic mutations are found in up to 40% of patients with AA (16,32). PNH clones are detected in a higher proportion of patients with AA (up to 60%) and have been shown to escape T cell-mediated destruction. Blood cells with PIGA mutations are likely less immunogenic and thus may acquire a survival advantage (33). Somatic mutations in JAK2/JAK3, RUNX1, TP53, TET2, and CUB and sushi multiple domains 1 genes are less common in AA, and SRSF2, U2AF1, MPL and Erb-B2 receptor tyrosine kinase 2 mutations are rare (<3% of acquired AA cases) (34). Detected somatic mutations in AA have mostly variant allelic frequencies of <10% (23,34). Patients with AA and PIGA, BCOR or BCORL1 mutations show a better response to IST, as well as improved progression-free survival (PFS) and OS rates, while DNMT3A, ASXL1, JAK2/JAK3 or RUNX1 mutations are associated with a worse IST response and survival rate. Notably, mutations in DNMT3A or ASXL1 increase the risk of developing MDS from AA (23,34,35). Keel et al (36) detected pathological mutations in MPL and TP53 genes in young patients with AA and MDS.

A high incidence of somatic mutations in MDS suggests that mutational profiling of myelodysplasia-related genes may help to distinguish AA from hMDS and may identify patients who are at risk for progression. Kulasekararaj et al (23) used targeted high-throughput DNA sequencing to determine somatic mutations in patients with acquired AA. Somatic mutations (ASXL1, DNMT3A, BCOR) were detected in 19% of patients with AA who had a longer disease duration and a higher risk of MDS transformation than those without mutations. Notably, the detection of ASXL1, DNMTA, BCOR and TET2 mutations in the AA cohort coupled with published expression data provides a role for the potential association and cooperation between mutations in epigenetic regulators and immune-mediated BMF. Similarly, Huang et al (37) focused on a limited number of genes and found mutations in epigenetic regulator genes, including TET2 and ASXL1, in 17.4% of patients with AA. By contrast, Heuser et al (38) identified somatic mutations in only 5.3% of patients with AA and suggested that mutations in myeloid malignancy-related genes are rare in this disease.

In the last two decades, it has become increasingly evident that ncRNAs are important regulators of biological processes, including blood cell differentiation and immune response. There are several categories of ncRNAs, such as microRNAs (miRNAs), Piwi-interacting RNAs, small nucleolar RNAs and long ncRNAs (lncRNAs) (39). miRNAs are the most prolific class of ncRNAs and have been shown to play a role in the pathogenesis of MDS (40). Comprehensive data are available on expression miRNA profiles associated with MDS subtypes, disease stages and treatment response, as well as on dysregulation of specific miRNAs and their role in pathogenesis (Table II). As MDS originates in HSCs, a number of studies have been performed on CD34⁺ cells. Abundantly expressed miRNAs in CD34⁺ cells of patients with MDS include, but are not limited to, let-7b, miR-10a, miR-25, the miR-26 family, miR-128a, miR-146, miR-155, miR-181a, miR-222 and miR-223 (41). To date, no study has focused on the differential expression of miRNAs between hMDS and NH-MDS. In general, low-risk patients show distinctive expression profiles compared with high-risk patients (42,43). Sokol et al (43) defined a unique signature of 10 miRNAs (miR-181a/b/c/d, miR-221, miR-376b, miR-125b, miR-155, miR-130a and miR-486-5p) that accurately differentiated low-risk patients from high-risk patients. Notably, the 6-miRNA signature may distinguish RA/refractory cytopenias with multilineage dysplasia (RCMD) patients with a normal karyotype from those with trisomy 8, who usually show a good response to IST.

A cluster of 13 miRNAs, including miR143/miR-145, has been mapped in the deletion region del5q31-5q35 (44); these miRNAs are downregulated in a variety of human cancer types, such as colorectal and gastric cancer (45,46). Haploinsufficiency of these miRNAs and miR-146a (adjacent to the commonly deleted region) contributes to the 5q-syndrome phenotype (47,48). Furthermore, miR-145 and miR-146a are implicated in the dysregulation of innate immune signaling in MDS HSPCs (49). miR-146a is a negative regulator of immune cell activation; it represses two targets, tumor necrosis factor (TNF) receptor-associated factor 6 (TRAF6) and interleukin 1 receptor-associated kinase, which are signaling transducers upstream of nuclear factor κB (NF-κB) (40). NF-κB activation is regulated by miR-125a, which also plays a role in the regulation of innate immunity pathways and erythroid differentiation in MDS (50).

Compared with hMDS, acquired AA has more available miRNA data (Table II). Srivastava et al (51) recently found deregulated expression of miR-126, miR-145, miR-155, miR-146 and miR-150 in AA, and determined their target genes, phosphoinositide-3-kinase regulatory subunit 2 (PIK3R2), MYC proto-oncogene (MYC), suppressor of cytokine signaling 1, TRAF6 and MYB proto-oncogene, respectively. In other recent study, Lu et al (52) integrated multiple expression profiles of miRNAs and mRNAs of BM T cells from patients with acquired AA and showed that miR-34a-5p, miR-195-5p and miR-424-5p may modulate T-cell differentiation and plasticity by targeting histone gene expression and histone modification. A similar approach was used in the study by Adhikari and Mandal (53), which identified significant upregulation of miR-1202 in patients with AA compared with controls, and in which its putative targets, rap guanine nucleotide exchange factor 5 and mannosidase endo-α, were predicted. In the plasma of patients with acquired AA, Hosokawa et al (54) identified deregulation of miR-150-5p, miR-146b-5p and miR-1, which target immune pathways related to Toll-like receptors and TNF-α. Notably, the miRNA expression was restored to normal after successful IST. In particular, miR-150-5p showed a correlation with IST response, suggesting that it may serve as a biomarker for therapeutic monitoring (54). The same group analyzed miRNA levels in CD4⁺ and CD8⁺ T cells from patients with AA and detected downregulation of miR-126-3p and miR-223-3p in CD4⁺ T effector memory cells, and of miR-126-3p, miR-145-5p and miR-223-3p in CD8⁺ T effector memory and terminal effector cells. The expression levels of miR-126-3p, miR-145-5p and miR-223-3p became normal after successful IST. MYC and PIK3R2 genes were shown to be targets of miR-145-5p and miR-126-3p, respectively (55). Sun et al (56) demonstrated that overexpression of miR-34a and downregulation of its target gene, diacylglycerol kinase ζ, enhanced T-cell activation in acquired AA. Giudice et al (57) analyzed exosomal miRNAs in severe AA and MDS, and found distinctive signatures between these BMF disorders. In patients with AA, miR-126-5p showed a negative correlation with IST response, and patients with high miRNA levels at diagnosis had the shortest PFS time compared with patients with lower or normal levels. Furthermore, miR-4651 was exclusively present in severe AA responders to IST (57).

lncRNAs represent another important class of ncRNAs whose role in hematopoietic disorders is being explored. Studies on lncRNAs in MDS display heterogeneity in experimental design (size of patient cohort, MDS subtypes, technologies used and analytical approaches), and thus far, no study has focused only on hMDS. The very first study by Benetatos et al (58) revealed hypermethylation of the maternally expressed 3 (MEG3) gene promoter in 34.9% of patients with MDS, which may confer a worse overall prognosis. Next, genome-wide studies defined the gene expression profiles of lncRNAs in various specific groups of patients with MDS, such as those with primary MDS (59,60), refractory anemia (RA) with excess blasts type 2 (RAEB-2) MDS (61), de novo MDS and MDS evolved from AA (62). Recently, Szikszai et al (59) analyzed lncRNA expression across all MDS subtypes and evaluated them in relation to disease subtypes, cytogenetic and mutational aberrations, and risk of progression. Comparative analysis between low- and high-risk patients determined 16 deregulated lncRNAs [e.g., downregulated RP11-897M7.1 and long intergenic non-protein coding RNA 539, and upregulated T cell leukemia/lymphoma 6, long intergenic non-protein coding RNA 1013, LEF1 antisense RNA 1 (LEF1-AS1) and CTC-436K13.2 in low-risk patients] (59). Yao et al (60) attempted to use lncRNA expression for the risk stratification of patients with MDS and integrated four lncRNAs (TC07000551.hg.1, TC08000489.hg.1, TC02004770.hg.1 and TC03000701.hg.1) whose expression levels were associated with OS into a risk-scoring system. Higher lncRNA scores were associated with high-risk MDS, complex karyotype, and RUNX1, ASXL1, TP53, SRSF2 and ZRSR2 mutations. In relation to the skewed T cell repertoire in MDS, pathway analysis of differentially expressed genes between patients with the highest and lowest lncRNA risk scores determined T cell-related pathways [e.g., cytotoxic T-lymphocyte-associated protein 4 signaling in cytotoxic T lymphocytes and CD28 signaling in T helper (Th) cells] to be the most significant (60). Hung et al (63) recently identified an association between higher KIAA0125 expression (BM mononuclear cells) and high-risk MDS, ASXL1 and NRAS mutations, and poorer OS and leukemia-free survival. A recent study by Li et al (64) reported an association between a higher expression level of LOC101928834 and a higher white blood cell count, a higher blast percentage, RAEB subtype and a shorter OS time in MDS. By contrast, LEF1-AS1 expression has been shown to be significantly downregulated in patients with MDS compared with that in healthy controls (65).

There are limited data on lncRNAs in AA. Recently, Wang et al (66) demonstrated decreased expression of MEG3 in CD4⁺ T cells derived from patients with AA. MEG3 can act as an miRNA sponge that sequesters miR-23a and induces T cell immunoreceptor with Ig and ITIM domains expression in CD4⁺ T cells, leading to the expansion of Th17 and Th1 cells and increased serum TNF-α and interferon-γ (IFN-γ) levels. Jiang et al (67) reported that fibroblast growth factor 1 promoted the proliferation of BM mesenchymal stem cells from patients with AA by acetylation of lncRNA in the testis development related 1 gene promoter. Lu et al (68) recently analyzed differentially expressed lncRNAs and mRNAs between children with acquired AA and healthy controls. The study defined immune- or hematopoietic-related lncRNA/mRNA pairs [AC007556.1/dehydrogenase/reductase 9, AC007922.2/histamine receptor H4, AC147651.1/platelet derived growth factor subunit A, AC111000.4/growth factor independent 1B transcriptional repressor, AC007991.2/indoleamine 2,3-dioxygenase 1 and RHOXF1P1/semaphorin 7A (John Milton Hagen Blood Group)] that may be involved in the pathology of acquired AA (68).

Although there are no studies describing miRNA/lncRNA profiles exclusively in hMDS, there are reports demonstrating that RA and RCMD categories (typical for the majority of hMDS cases) show ncRNA expression patterns distinct from those of other MDS subtypes. Moreover, levels of specific ncRNAs have been successfully used for classification and stratification of patients with MDS (42,43,60). It may be assumed that hMDS is also associated with specific ncRNA profiles that differ from those of AA and could be used for differentiation. The dysregulation of ncRNAs detected in T cells derived from patients with AA and hMDS indicates that these regulators may contribute to the immunopathogenesis of these disorders.

The overlap of immunological features and the responsiveness of a significant proportion of patients with hMDS/AA to IST suggest that these distinctive clinical entities share an immune-mediated pathogenic mechanism. Clinical and experimental studies have provided compelling evidence that HSPCs are damaged by abnormally activated cytotoxic T cells (Fig. 2). Expanded CD4⁺ and CD8⁺ T cell clones have been observed in the BM of both patients with hMDS and those with AA (69,70). Melenhorst et al (71) analyzed the CD4⁺ and CD8⁺ T cell repertoires in patients with MDS by flow cytometry and PCR. Multiple T cell expansions (of both helper and cytotoxic T (Tc) cells) were observed, as well as the functional differentiation in vivo of T cells from memory to effector T cells, in CD8⁺ cells. Similar findings were reported by Fozza et al (72), supporting the involvement of cytotoxic T cells either in antitumor immune surveillance or in autoreactive aggression toward hematopoietic precursors. Moreover, dominant T cell clones persist in patients with MDS that is unresponsive to immunosuppression and regress in responders (73). Strong polarization of BM CD4⁺ cells toward Th1 and of CD8⁺ cells toward Tc1 was observed in low-risk MDS compared with that in AA, suggesting T cell stimulation from clones of malignant hematopoietic cells (74). Regulatory T cells (Tregs) are deficient in quantity and function in patients with early MDS (75). The function of Tregs is to suppress the autoreactivity of other T cell populations to normal tissue; thus, their hypofunction may favour the autoimmune destruction of HSPCs (76).

The antigens that trigger the immune response in MDS are not known, but potential candidates [such as Wilms tumor protein 1 (WT1)] have been suggested. As patients with MDS and trisomy 8 often show a good response to IST, an immunological mechanism underlying BMF has been proposed. Trisomy 8 cells express high levels of WT1, and CD8⁺ T cells are able to recognize WT1 peptides and induce IFN-γ expression in vitro, suggesting that this antigen may contribute to the induction of an immune response (77). Sloand et al (78) further demonstrated that marrow HSCs with trisomy 8 may escape T cell-mediated destruction by overexpression of prosurvival protein cyclin D1 and survivin. Other neoantigens or overexpressed self-antigens [human leukocyte antigen (HLA)-A2-restricted nonameric peptide] may also elicit an immune response (79).

Abnormal overproduction of proinflammatory cytokines [such as TNF-α, IFN-γ and interleukin 17 (IL-17)] has been reported in patients with MDS and contributes also to ineffective hematopoiesis (80,81). In patients with low-risk MDS, the G/A polymorphism in the TNF-α promoter is associated with high levels of TNF-α produced by CD4⁺ and CD8⁺ T lymphocytes (82), suggesting its role in anemia. In a previous study, an increased frequency of CD4⁺ T cells producing IFN-γ was detected in hMDS, and in vitro decrease of interferon by cyclosporine led to improved hematopoiesis (10). The production of IFN-γ and TNF-α in low-risk MDS may be further enhanced by high levels of IL-17 (83). Based on the clinical/immunological/molecular features, Fattizzo et al (84) recently defined two hMDS phenotypes, namely, AA-like and MDS-like hMDS. The first is characterized by prevailing inflammation and immune activation, and a response to IST, and the second is characterized by genetic lesions, clonal selection and an increased risk of leukemic evolution.

Up to 80% of patients with AA show a response to T cell-directed IST, supporting involvement of aberrant T cell populations in the pathogenesis. As in hMDS, BM T cells are also skewed toward oligo/polyclonal patterns in acquired AA (16,70,85). Giudice et al (85) observed oligoclonal characteristics in CD8⁺CD57⁺ cells, as well as in total CD8⁺ T cells from patients with AA. de Latour et al (86) found an increased population of CD3⁺CD4⁺IL-17-producing T cells in patients with AA at presentation compared with that in controls, and this correlated with disease activity. Abnormally activated T cells destroy HSPCs through apoptosis [via Fas cell surface death receptor (Fas)/Fas ligand, granzyme, perforin] and the overproduction of proinflammatory cytokines. Extensive apoptosis of BM HSPCs has been observed in patients with AA, indicating that apoptosis is a major mechanism of cell destruction (87). BM CD34⁺ progenitor cells and lymphocytes of patients with AA overexpress Fas, which is involved in triggering the Fas-mediated apoptotic pathway (88). By contrast, normal expression of Fas has been observed in patients with AA in remission (89). Overproduction of cytokines may upregulate the expression of Fas (77).

As in hMDS, AA Tregs exhibit a decreased quantity and ability to suppress the proliferation of autologous T cells. Deep phenotyping of AA Tregs defined two specific Treg subpopulations, Treg A and Treg B, that may predict the response to IST. The Treg B subpopulation with a memory/activated phenotype was overrepresented in IST responders, while the Treg A subpopulation was significantly higher in non-responders. Furthermore, Tregs from patients with AA were IL-2-sensitive and could be expanded in vitro (90).

AA is strongly associated with PNH. PNH clones deficient in glycosyl-phosphatidylinositol (GPI)-anchored proteins appear to be spared by the immune attack mediated by T cells in BMF syndromes. PNH clones are frequently found in acquired AA (≤60%) and are also observed in MDS (10-20%, more common in low-risk cases) (91). The mechanism of this escape is not clear. It has been suggested that antigen targets of T cell attack or coregulators are GPI-linked proteins. Gargiulo et al (92) demonstrated that CD1d-restricted, GPI-specific CD8⁺ T cells are expanded in patients with PNH, suggesting that the GPI may be targeted by autoreactive T cells and that these T cell clones are responsible for the BMF in PNH. Hanaoka et al (93) suggested that immunoselection of PIGA mutant cells is due to a deficiency in the stress-inducible GPI-linked membrane proteins UL16 binding protein 1 and 2, which activate natural killer and T cells. Furthermore, PIGA clones may acquire additional somatic mutations (TET2, SUZ12 polycomb repressive complex 2 subunit, U2AF1 and JAK2), resulting in a proliferative advantage (94). Mechanisms and factors implicated in the immunopathogenesis of AA and hMDS are summarized in Table III.

In addition to the T cell-mediated immune response, aging, which is associated with numerous changes in the immune system, including chronic low-grade inflammation (known as inflammaging), may be involved in the development of AA and hMDS in elderly patients (95).