In 2016, the World Health Organization updated the diagnostic criteria for pre-PMF, emphasizing the importance of differential diagnosis between ET and pre-PMF. Furthermore, the distinct prognoses of ET and pre-PMF, along with associated molecular genetic data, bolster the view that they are two separate entities, rather than a continuum of the same disease [1]. Currently, there are methods to separate the two based on clinical and pathological features [37, 38]; however, clinical differential diagnosis can sometimes be challenging, highlighting the need for additional methods to aid in distinguishing between them.

Studies have shown that age [14] and BMI [39] are related to MPN progression. Therefore, we ensured that the two patient groups were matched for age and BMI to minimize potential compounding effects. Furthermore, all included patients harbored the JAK2V617F driver gene mutation, and none had received MPN-related cytoreductive therapy since the disease onset. Moreover, they had not undergone treatment with antibiotics, probiotics, prebiotics, or symbiotics in the preceding three months, potentially reducing the influence of confounding factors on BM changes. Consistent with a previous study, patients with pre-PMF had a higher JAK2V617F allele burden than patients with ET [40]. Decreased serum CHO level is a feature of MPN [12]. Research [41] has shown that reduced levels of HDL cholesterol or apolipoprotein A1 correlate with elevated risks of multiple myeloma, MPN, non-Hodgkin lymphoma, and breast, lung, and nervous system cancers. Compared with polycythemia vera or ET, the impact of hypocholesterolemia on the prognosis of MF is more evident [42]. Serum CHO levels have been shown to correlate with the survival rate of PMF [43]. The HDL levels in patients with MPN appear to vary over time; some studies showed that HDL increased at diagnosis [13], and others have found a decrease in disease progression [12, 42]. In our study, we observed significantly lower levels of CHO and HDL in the pre-PMF group compared to the ET group. These findings may have potential implications for prognosis, suggesting further investigation. Our results revealed a significantly higher NLR in patients with pre-PMF than in those with ET. NLR, identified as an inflammatory biomarker in various malignancies [44,45,46], has been associated with adverse clinical outcomes when elevated. As MPNs are neoplastic disorders characterized by malignant clones triggering inflammatory cytokine release, elevated NLR are commonly observed in patients with MPNs. Some studies [47, 48] have also linked higher NLR values to poorer prognoses within this patient cohort. The elevated NLR observed in patients with pre-PMF in our study suggests a possible rise in systemic inflammatory activity, which may be linked to prognosis. Furthermore, this observation might underscore the potential usefulness of this inflammatory indicator in distinguishing between ET and pre-PMF.

In this study, we analyzed the proteomes of patients with pre-PMF and ET to explore potential differences in protein expression that could aid in distinguishing between the two conditions. We focused on the aforementioned differences in metabolic, inflammatory, and immune clinical indicators in the GO and KEGG top items, highlighting 34 proteins out of 177 that were differentially expressed proteins. We constructed and visualized a PPI network using the STRING database and Cytoscape software. The CytoHubba algorithm was employed to identify the eight hub proteins (APOA4, APOC3, FABP4, CFB, C5, CXCR4, CXCR2 and MX1).

C5, complement component 5, is involved in the complement system. It is cleaved into C5a and C5b. While C5a is vital for chemotaxis, C5b initiates formation of the complement membrane attack complex. CFB, complement factor B, is a component of the alternative complement pathway. Once the alternative pathway is activated, CFB is cleaved by complement factor D, which contributes to the immune response regulation. Activation of the complement system is often observed in autoimmune diseases (e.g., antiphospholipid antibody syndrome [49] and infectious diseases). There is limited research on the complement system in MPN. In our study, the level of complement activation in the pre-PMF group was lower than that in the ET group, possibly because of complement depletion caused by chronic inflammation. One study [50] showed that among patients with myelofibrosis, decreased levels of C3A, C3, and C4 were observed in the most severe cases. Therefore, we hypothesized that a reduction in complement levels could be associated with disease severity and the inflammatory status of the BM. However, further research is needed to validate these hypotheses. APOA4 and APOC3 are apolipoproteins, the primary protein components of lipoproteins. APOA4 is composed of chylomicrons, very low-density lipoproteins (VLDL), and HDL. APOC3 is primarily found in lipoproteins rich in triglycerides, such as chylomicrons, VLDL, and remnant cholesterol. In addition to their crucial role in lipid metabolism, their involvement in tumors has garnered increased attention. Jeong et al. [51] identified reduced APOA4 expression as a marker of cervical squamous cell carcinoma. Research [52] indicated that a reduced concentration of serum APOC3 might assist in the diagnosis of gastric cancer. Our study also observed lower levels of APOA4 and APOC3 in pre-PMF than in ET, implying a potential role for these two lipoproteins in aiding the differentiation between these two disease subtypes. Notably, the top two pairs of proteins exhibiting strong interactions were APOA4 with CFB and APOA4 with C5 (Fig. 4D). CFB demonstrated a significant positive correlation with HDL. Additionally, APOC3 exhibited significant positive correlations with HDL and CHO levels and a significant negative correlation with the NLR (Fig. 4E). These findings suggest a possible connection between lipid metabolism and inflammation. Fatty acid binding protein 4 (FABP4), or adipocyte protein 2, is a carrier protein for fatty acids. It is predominantly expressed in adipocytes and macrophages and is secreted extracellularly. FABP4, when bound to the fatty acid ligand, can interact with unphosphorylated Janus kinase 2 (JAK2), thereby attenuating JAK2 signal transduction [53]. CXCR2, also known as the interleukin eight receptor, beta, is a chemokine receptor. CXCR2 is involved in the migration and activation of neutrophils. It has been reported as an essential component for the recruitment of tumor-associated neutrophils in various cancers [54, 55], playing a notable protumor role. Increased CXCR2 expression is also associated with poor prognosis, as shown by studies on acute myeloid leukemia [56], invasive ductal breast cancer [57], and non-small cell lung cancer [58]. Additionally, Dunbar et al. found that CXCL8/CXCR2 signaling was increased in PMF and that deletion Cxcr2 in the hMPLW515L model improved blood parameters and decreased fibrosis. This underscores the importance of CXCL8/CXCR2 in fibrosis progression and suggests that inhibiting this pathway could be a promising therapeutic strategy for managing myelofibrosis [59]. In line with prior research, we observed higher CXCR2 expression levels in pre-PMF than in ET, which may suggest a potential association with a more severe prognosis in pre-PMF. CXCR4, or CD184, functions as a chemokine receptor. It plays a role in regulating the migration of various immune cells, such as T cells, B cells, and myeloid cells by interacting with its ligand CXCL12. CXCR4 is commonly found to be highly expressed in hematologic malignancies [60] and many types of solid tumors, including melanoma, and tumors of the lung, breast and ovary [61]. This overexpression is associated with a poorer prognosis [62]. Although Barosi et al. indicated that in patients with PMF, the expression of CXCR4 on CD34 positive blood cells is reduced, with implications for prognosis. This differs from our findings, which showed higher overall CXCR4 protein expression levels in pre-PMF than in ET samples [63]. The discrepancy may stem from differences in the focus of the study, with Barosi et al. concentrating on specific CD34-positive blood cells. At the same time, our study encompassed a broader examination of overall protein expression across all cell types in the BM. Additionally, CXCR2 showed a significant negative correlation with the clinical indicators of lipid metabolism, HDL, and CHO levels, while CXCR4 was significantly negatively correlated with HDL. MX1, also known as the interferon-induced GTP-binding protein Mx1, is crucial in inflammation and cancer. Several studies revealed that the level of MX1 protein in the lysates of mononuclear cells from the peripheral blood of patients with systemic lupus erythematosus was markedly elevated compared to the normal controls [64]. Additionally, inhibiting the transcription of MX1 could alleviate renal fibrosis in lupus nephritis [65]. In cancers such as breast and glioblastoma, MX1 expression was elevated compared to normal tissues [66, 67]. Research on colorectal carcinomas has revealed that the expression level of MX1 in tumors with lymph node metastases (UICC stage III) is higher than that in tumors without lymph node metastases (UICC stage II) [68]. The findings of this study indicated higher MX1 expression levels in pre-PMF than in ET. This suggests a potentially heightened inflammatory state in pre-PMF and raises the possibility of a poorer prognosis in pre-PMF. However, further research is necessary to confirm these associations conclusively.

To further explore the differences in BM immune status between pre-PMF and ET, we utilized CIBERSORT software for the immune infiltration analysis. Our findings showed that pre-PMF has a higher percentage of innate immune cells than ET, with higher proportions of activated mast cells and eosinophils were higher in the pre-PMF group. An increase in the number of eosinophilic granulocytes is occasionally observed in MPNs, and research has also explored their potential protumor effects in other tumors. In preclinical models of oral squamous cell carcinoma, inhibiting eosinophil infiltration led to obstructed carcinoma growth [69]. Similarly, a cervical cancer model indicated that eosinophils activated by tumor thymic stromal lymphopoietin could promote tumor growth [70]. However, the mechanism of eosinophils in MPN requires further investigation. Studies have shown that mast cells play a crucial role in the regulation of inflammatory processes and fibrosis [71]. Activated mast cells produce various substances, including histamines, heparin, tryptases, cytokines, and matrix metalloproteinases, all implicated in fibrogenesis. The proliferation and activation of mast cells are associated with fibrosis in various clinical diseases [72,73,74]. An increased number of mast cells is observed in MPN samples [75]. Consistently, the increased presence of mast cells in pre-PMF observed in our study may imply a potentially more challenging prognosis for pre-PMF than for ET. Furthermore, we conducted ROC curve analysis for the eight selected hub proteins. Each protein exhibited an AUC value exceeding 0.75, suggesting potential value as biomarkers for differentiating between ET and pre-PMF.

In addition to proteomics, we used 2bRAD-M to characterize BM microbiota in pre-PMF and ET. This method is especially suitable for samples with a low biomass. We found that pre-PMF had higher microbiota diversity than ET. At the family level, the top five microbiota in the bone marrow were Beijerinckiaceae, Burkholderiaceae, and Xanthobacteraceae, whereas the most abundant gut microbiota of MPN [26] encompassed Ruminococcaceae, Lachnospiraceae, and Bacteroidaceae. These results showed that the BM and gut microbiota were different in patients with MPN. Additionally, Andrew Oliver et al. found that the fecal microbial community composition in MPN was related to inflammatory states [26]. This supports the notion that variations in the BM microbiome could potentially serve as markers for differentiating pre-PMF from ET. Using LEfSe and random forest analysis, we identified six genera (Sphingomonas, Brevibacillus, Pseudomonas_E, Mycobacterium, Xanthobacter, and L1I39) that distinguished pre-PMF from ET.

Sphingomonas is a bacterial genus that was subclassified from Pseudomonas approximately 30 years ago. Although it is a Gram-negative bacterium, it contains sphingolipids instead of lipopolysaccharides [76]. It was consistently present in all of our study samples, with abundance notably lower in pre-PMF than ET. In a study on the microbiomics of breast tumors, it was discovered that bacteria belonging to the genera Sphingomonas and the species Sphingomonas yanoikuyae were more abundant in healthy breast tissues than in breast tumor tissues [77]. Brevibacillus is a facultative anaerobic, Gram-positive, endospore-forming bacterium that produces a variety of antimicrobial agents including short-sequence microbial peptides as well as glycopeptides, bacilysins, and bacteriocins. Peptides such as Brevilaterin B, extracted from Brevibacillus laterosporus S62–9, are believed to possess anticancer, antibacterial, and antifungal activities [78]. Bogorol is a peptide isolated from B. laterosporus JX-5 with potent antibacterial and anticancer activities [79]. According to the Genome Taxonomy Database (GTDB), Pseudomonas_E refers to a specific taxonomic classification within the Pseudomonas genus. The GTDB utilizes genomic data to provide a consistent phylogenetic taxonomy for bacterial and archaeal genomes. Network analysis of 10,000 genomes revealed a phylogenetic tree of Pseudomonas, uncovering at least 14 Pseudomonas groups [80]. Furthermore, Barco et al. used a genomic index (average nucleotide identities and genome alignment fractions) to describe bacterial genera, where Pseudomonas is considered a known polyphyletic genus [81]. Several studies investigating the relationship between Pseudomonas aeruginosa and cancer have shown that the blue cupredoxin azurin, secreted by this opportunistic pathogen, enters human cancer cells, leading to apoptosis [82]. Moreover, the presence of detectable Pseudomonas aeruginosa and azurin in the tumors of cancer patients has been linked to an increased overall survival rate [83]. Our results indicated that levels of Sphingomonas, Brevibacillus, and Pseudomonas_E were notably lower in pre-PMF than in ET, suggesting a potential contribution to the adverse features in patients with pre-PMF, pending further confirmation. According to the GTDB, both Xanthobacter and L1I39 belong to the bacterial genera of the family Xanthobacteraceae and the order Rhizobiales. Current research on the role of these two bacterial genera in tumors is limited. High-throughput sequencing in microbiome studies has identified a mixed community of Rhizobiales in the lungs and blood of patients with fatal pulmonary illness. Mycobacterium is a genus of over 190 species belonging to the Mycobacteriaceae family. An increasing number of studies have suggested that chronic mycobacterial infections may be associated with an increased cancer risk. This association is primarily attributed to persistent inflammation induced by pathogens. Lung tissues infected with Mycobacterium tuberculosis have been shown to undergo multiple cycles of inflammation and tissue repair, creating a favorable environment for tumor formation and increasing the risk of lung cancer [84]. Our study observed an enrichment of the genera Xanthobacter, L1I39, and Mycobacterium in pre-PMF samples. This implies that these bacterial genera are likely associated with pre-PMF pathogenesis, pending further confirmation. Furthermore, we performed ROC curve analysis for the six selected genera, with all AUC values exceeding 0.75. This suggests the potential value of these genera as biomarkers for differentiating between pre-PMF and ET.

The significance of this study lies in its multi-omics analysis of the differences between ET and pre-PMF BM. First, research on the BM proteome and microbiome in these two disease subtypes is rare, and our findings bridge this knowledge gap. Given the growing interest in gut microbiome research, our study underscores the potential importance of investigating the BM microbiome. It provides preliminary insights that could contribute to the broader understanding of the BM microbiome in patients with MPN, suggesting avenues for further research. Second, we utilized the 4D direct DIA and 2bRAD-M techniques for proteomic and microbiomic analyses. Both techniques are particularly suitable for low biomass samples, especially the FFPE of the BM. Finally, our research provided initial insights into variations in the bone marrow proteomics and microbiomes of patients with ET and pre-PMF, pinpointing distinct proteins and bacterial genera that merit additional exploration as potential diagnostic markers.

However, this study has some limitations. First, the sample size is relatively small. While our findings provide promising preliminary insights, they primarily serve as a reference for future research. We hope that subsequent studies can build upon these findings. Second, a prospective study with a more robust methodological design is required to validate and extend our results. Finally, additional comprehensive experiments are necessary to explore the complex interactions between the BM proteome and microbiome, enhancing our understanding of their roles in the pathogenesis and progression of myeloproliferative neoplasms.