Immune and cytokine landscape differs between cytopenic and proliferative phenotypes in primary myelofibrosis

Abstract

Background

Cytopenia represents a distinct clinical phenotype in primary myelofibrosis (PMF), often associated with advanced disease and poor prognosis. We propose that while a common driver mutation initiates the disease, the clinical phenotype of PMF is shaped by inter-individual variability in downstream inflammatory responses, contributing to the observed heterogeneity in disease presentation.

Objectives

To characterize inflammatory and immune signatures distinguishing cytopenic from proliferative phenotypes of PMF.

Design

Retrospective, single-center observational study analyzing clinical, molecular, transcriptomic, and cytokine data from patients with pathologically confirmed PMF.

Methods

We analyzed a cohort of 140 patients with PMF, stratified into cytopenic and proliferative phenotypes based on abnormalities in at least one peripheral blood lineage. Bone marrow transcriptomic profiling was performed using the NanoString nCounter PanCancer Immune Panel. Cytokine levels were measured by MILLIPLEX® bead-based multiplex assay, and phosphorylated NFκB p65 expression was evaluated by immunohistochemistry.

Results

Compared with the proliferative phenotype, cytopenic PMF showed more advanced disease features, including higher blast counts (P<0.001), elevated LDH (P<0.001), and more frequent overt fibrosis (P<0.001). Cytopenic patients had fewer JAK2 mutations (P=0.025) but were enriched for high molecular risk mutations such as ASXL1 (P=0.011) and EZH2 (P=0.035). Transcriptomic profiling revealed upregulation of inflammatory pathways, notably TNFα–NFκB (P=0.0097) and IL2–STAT5 (P=0.0399), while cytokine analysis showed elevated IL-8 (P=0.021) and G-CSF (P=0.007) independent of fibrosis severity. Immunohistochemistry confirmed enhanced NFκB p65 activation in cytopenic marrow.

Conclusion

Our findings highlight cytopenic PMF as a biologically distinct and clinically aggressive phenotype characterized by inflammatory pathway activation, particularly involving NFκB signaling and elevated IL-8. These results support the role of immune dysregulation in the pathogenesis of cytopenia in PMF and provide a basis for further investigation into its clinical implications.

Plain language summary

Primary myelofibrosis (PMF) is a blood cancer that affects bone marrow functions. Some patients with PMF develop low blood counts, known as cytopenia, which can include anemia, low platelets, or low white cells. These patients often have more severe symptoms and shorter survival than those with more typical, high-blood-count forms of the disease. However, the underlying causes are multifactorial, and the contribution of inflammatory processes remains unclear. In this study, we looked at 140 patients with PMF and compared those with cytopenia to those without. We studied their bone marrow samples to understand the differences in gene activity and immune signaling. We also measured the levels of cytokines in the bone marrow and performed tissue staining to look for activation of inflammation-related pathways. We found that patients with cytopenia had signs of more advanced disease and were more likely to carry high-risk genetic mutations. Their bone marrow also showed increased activity of inflammatory pathways, especially those controlled by NF-κB. These patients also had higher levels of immune proteins such as IL-8 and G-CSF, even after accounting for the degree of bone marrow fibrosis. Under the microscope, their bone marrow tissue also showed stronger signals of NF-κB activation. Our findings suggest that abnormal immune activity and inflammation may contribute to low blood counts in PMF. This may help explain why these patients do worse overall. Understanding this process could lead to better ways to predict risk and develop new treatments for this specific group of PMF patients.

Introduction

Primary myelofibrosis (PMF) represents the most severe manifestation within the spectrum of Philadelphia chromosome-negative myeloproliferative neoplasms (MPN), characterized by the dual nature of clonal proliferation of bone marrow cells and the systemic features of a chronic inflammation. Central to the pathogenesis of PMF is the aberrant activation of the JAK/STAT signaling pathway. ¹ This critical pathway is responsible for the sustained proliferative signal regardless of the specific driver mutation identified in individual patients.^2–7 Intriguingly, the presence of a single mutation can manifest into three distinct clinical phenotypes of MPN,^3,5,7 suggesting that disease biology is influenced by factors beyond the genetic mutation per se. One plausible hypothesis is that the aberrant activation of the JAK/STAT pathway initiates a cascade of inflammatory responses, which exhibit variability among patients.^8–10

Patients with PMF exhibit a broad spectrum of symptoms, ranging from subjective symptoms such as fatigue and night sweats to measurable findings like weight loss and splenomegaly.^11,12 However these manifestations are often difficult to standardize across clinical settings. In contrast, objective parameters, such as cytopenia, circulating blasts, and bone marrow fibrosis grade better reflect disease severity. ¹³ Among these, stratifying patients by cytopenia provides as integrated measure of disease burden and biological heterogeneity in PMF.

Cytopenia, particularly anemia and thrombocytopenia, are common and clinically significant in PMF. Anemia, defined by a hemoglobin level of less than 10g/dL, is present in nearly 40% of patients with PMF at diagnosis. ¹³ Additionally, 10-20% of PMF patients have platelet counts below 100 × 10⁹/L at diagnosis.^13,14 Both are associated with inferior outcomes^13–17 and form essential elements of prognostic models such as the International Prognostic Scoring System (IPSS), ¹⁸ the Dynamic International Prognostic Scoring System Plus (DIPSS+), ¹⁹ and the Myelofibrosis Secondary to PV and ET Prognostic Model (MIPSS70). ²⁰ The pathogenesis of cytopenia in PMF is multifactorial involving marrow fibrosis that disrupts hematopoiesis, splenic sequestration from compensatory extrameduallary hematopoiesis, and inflammation-driven hepcidin overproduction that impairs erythropoiesis.^{10,17,21–23}

In this study, we investigated whether differential activation of inflammatory and immune pathways contributes to the phenotypic heterogeneity of PMF. By integrating transcriptomic profiling of bone marrow tissue with cytokine analysis and immune profiles of the bone marrow microenvironment, we aimed to characterize the immune signatures associated with cytopenic and proliferative presentations. Our goal is to gain deeper insight into the pathophysiology of cytopenia in PMF, thereby informing more targeted therapeutic strategies to alleviate disease-related symptoms.

Methods

Study population and data collection

This study included 140 patients diagnosed with PMF at the National Taiwan University Hospital between 2005 and 2023, all confirmed by hematopathologists according to the 2016 WHO classification. Patients with available diagnostic PBMC and plasma samples were included, whereas those lacking adequate material or excluded due to funding constraints were omitted. Immune gene expression profiling was performed on 74 patients using the Nanostring PanCancer Immune Panel applied to cryopreserved bone marrow–derived mononuclear cells. Cytokine, chemokine, and growth factor profiling were conducted on bone marrow plasma samples from 73 patients using the MILLIPLEX® bead-based immunoassay. Clinical, laboratory, cytogenetic, and mutational data were retrospectively collected from electronic medical records. Mutation profiling included driver mutations (JAK2, CALR, MPL) as well as other myeloid neoplasm–associated genes, and cytogenetic risk was assessed according to established criteria. Cytopenia was defined as follows: leukopenia as a white blood cell count (WBC) <4 × 10⁹/L, sex-adjusted anemia as hemoglobin <11 g/dL in males or <10 g/dL in females, and thrombocytopenia as platelet count <100 × 10⁹/L. ¹⁵ Patients presenting with at least one of these cytopenias were categorized as having a cytopenic phenotype, whereas those without any of the three cytopenias were classified as having a proliferative phenotype.

Immune related gene-expression profiling using the nCounter system

Immune gene expression profiling was performed using the NanoString nCounter® PanCancer Immune Profiling Panel (NanoString Technologies, Seattle, WA), which includes 109 immune-related genes representing 24 immune cell types and functional populations, along with 40 housekeeping genes for sample-to-sample normalization. RNA was extracted from cryopreserved bone marrow mononuclear cells using RNeasy extraction kit (Qiagen). RNA quality was assessed prior to profiling, and samples were required to meet the following criteria for inclusion: RNA concentration ≥20 ng/µL, total RNA yield ≥100 ng, and an RNA fragment size peak ≥300 nucleotides on Bioanalyzer analysis. Gene expression was analyzed using 100 ng of total RNA, which was hybridized to the NanoString probe set at 65°C for 18 hours. Following hybridization, samples were purified and immobilized on a glass cartridge using the nCounter Prep Station with the high-sensitivity protocol. Target-specific barcodes were digitally counted using the nCounter Digital Analyzer. ²⁴ Raw data underwent initial quality control assessment using nSolver Analysis Software (NanoString Technologies), and subsequent normalization and differential expression analysis were conducted using the nCounter Advanced Analysis module, applying housekeeping gene normalization as per NanoString’s recommended workflow. Immune gene expression data were analyzed using the ROSALIND platform (NanoString’s partner) with a generalized linear model (negative binomial distribution) and Benjamini–Hochberg FDR correction. Gene Set Analysis (GSA) was performed to assess global significance, and enrichment was evaluated using curated databases (WikiPathways, REACTOME, MSigDB, Gene Ontology), applying p-Elim pruning to account for hierarchical dependencies. ²⁵

Cytokine analysis with Milliplex®

Bone marrow plasma cytokine concentrations were measured using the MILLIPLEX^® human cytokine/chemokine/Growth factor panel A (HCYTA-60K-PX38, Merck Millipore, Germany), following the manufacturer’s instructions. ²⁶ This 37-plex magnetic bead–based assay quantifies a broad range of immune mediators, including growth factors: epidermal growth factor (EGF), platelet-derived growth factor-AA (PDGF-AA), PDGF-AB/BB, vascular endothelial growth factor (VEGF); colony-stimulating factors: granulocyte colony-stimulating factor (G-CSF), granulocyte-macrophage colony-stimulating factor (GM-CSF), macrophage colony-stimulating factor (M-CSF);

Interferons: interferon-α2 (IFN-α2), interferon-γ (IFN-γ); interleukins: IL-1α, IL-1β, IL-1 receptor antagonist (IL-1RA), IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-17A, IL-17E (IL-25), IL-17F, IL-18, and IL-22; chemokines: eotaxin (CCL11), inducible protein-10 (IP-10/CXCL10), monocyte chemoattractant protein-1 (MCP-1/CCL2), monokine induced by IFN-γ (MIG/CXCL9), macrophage inflammatory protein-1α (MIP-1α/CCL3), MIP-1β (CCL4), regulated on activation normal T cell expressed and secreted (RANTES/CCL5); tumor necrosis factors: TNF-α and TNF-β. All samples were analyzed in accordance with the manufacturer’s protocols and quality control procedures.

Immunohistochemical detection of NF-κB activation

Representative bone marrow sections from 12 PMF patients —8 with a cytopenic phenotype and 4 with a proliferative phenotype, each exhibiting variable degrees of bone marrow fibrosis—were stained with a phospho-NF-κB p65 (Ser536) antibody (Bioss) to evaluate NF-κB activation. ²⁷

Statistical analysis

This was a retrospective cohort; therefore, the sample size was determined by case availability within the study period. For laboratory analyses, assay-specific cohorts were defined by availability of adequate PBMCs (immune gene profiling) or plasma (cytokines). No priori power analysis was performed. Descriptive statistics were used to summarize baseline characteristics, and comparisons between cytopenic and proliferative PMF groups were performed using the Mann–Whitney U test for continuous variables and the Chi-square or Fisher’s exact test for categorical variables. Overall survival (OS) and leukemia-free survival (LFS) were estimated using the Kaplan–Meier method, and differences between groups were assessed by the log-rank test. Two definitions of LFS were analyzed: (1) LFS, defined as the time from diagnosis to leukemic transformation, and (2) LFS censored at allogeneic hematopoietic stem cell transplantation (HSCT), where patients were censored at the time of transplant to isolate the natural disease course. In both analyses, patients without events were censored at the date of last follow-up.

To assess associations between cytokines and clinical phenotypes, we performed logistic regression analyses. Cytokine concentrations were log-transformed to correct for right-skewed distributions and to approximate normality, thereby improving model stability and interpretability.^28,29 Univariate logistic regression was used to screen for significant cytokines, which were then entered into multivariable models. Model A examined cytokines associated with the cytopenic phenotype after adjusting for fibrosis severity, while Model B evaluated cytokines associated with overt versus early fibrosis. All statistical analyses were performed using SPSS (version 25) and GraphPad Prism (version 10.4.2), with p values < 0.05 considered statistically significant.

The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. ³⁰

Results

Clinical characteristics and outcome associated with cytopenic PMF

Of the 140 PMF patients, 62 were classified as cytopenic, and 78 displayed a proliferative phenotype. Detailed characteristics of these patients at the time of diagnosis are delineated in Table 1. Patients with cytopenic PMF were significantly older (median age 66.4 vs. 59.4 years, p < 0.001). They had a higher mean percentage of circulating peripheral blood blasts (1.4% vs. 0.4%, p < 0.001), more frequent overt bone marrow fibrosis, defined as grade 2 or higher (82.5% vs. 50.7%, p < 0.001), and significantly lower hemoglobin, WBC, and platelet counts (all p < 0.001). In addition, lactate dehydrogenase (LDH) levels were higher in the cytopenic group (median 516 vs. 355 U/L, p < 0.001).

Table 1.

Comparison of clinical and laboratory features between patients with cytopenic and proliferative primary myelofibrosis.

​	Cytopenic (n=62)	Proliferative (n=78)	P value
Male, n (%)	39 (62.9)	37 (47.4)	0.068
Age (year), median (range)	66.4 (38-88)	59.9 (22-86)	<0.001
PB blast, %, mean (SD)	1.4 (0-11)	0.4 (0-5)	<0.001
Overt fibrosis, n † (%)	47 (82.5)	37 (50.7)	<0.001
Hb (mg/dL), median (range)	8.9 (4.5- 10.8)	13.4 (11.1- 20.3)	<0.001
WBC (x109/L), median (range)	14709 (1420- 96940)	15075 (5390- 43610)	<0.001
Platelet (x109/L), median (range)	265 (7-2256)	530 (107- 2191)	<0.001
LDH, median (range)	516 (117- 2870)	355 (202- 1162)	<0.001
Unfavorable karyotype ‡	7 (14.3)	4 (6.1)	0.113
Normal karyotype	29 (46.8)	51 (65.4)
Complex	4 (6.5)	2 (2.6)
Trisomy 8	3 (4.8)	2 (2.6)
Del 20q	1 (1.6)	5 (6.4)
Del 13q	2 (3.2)	2 (2.6)
Leukemic change	5 (8.1)	6 (7.7)	0.935
MPN drivers †
JAK2 mutated, n (%)	38 (63.3)	62 (80.5)	0.025
JAK2V617F allele burden (%), median (range)	67.4 (29.7- 97.6)	76.2 (20.9-97.4)	0.841
CALR mutated, n (%)	11 (18.3)	9 (1.7)	0.274
MPL mutated, n (%)	5 (8.5)	4 (5.3)	0.458
Triple negative, n (%)	4 (6.8)	2 (2.7)	0.246
JAK2 and CALR double mutated, n (%)	0	1 (1.4)	0.826

Hb: hemoglobin, WBC: white blood cell, MPN: myeloproliferative neoplasm, ns: none significance.

^†112 available fibrosis grading.

^‡115 available karyotype analysis. Unfavorable cytogenetics: complex karyotype (≧3 changes), +8, −7/7q−, i (17q), inv (3), −5/5q−, 12p− or 11q23 rearrangement.

^§137 available driver mutation analysis.

Cytogenetic profiles were comparable between the two groups, without significant difference in the proportion of unfavorable or complex karyotypes, trisomy 8, and 13q/20q deletions. Unfavorable cytogenetics were defined according to the DIPSS-Plus criteria, which include complex karyotype (≥3 abnormalities), +8, −7/7q−, i (17q), inv (3), −5/5q−, 12p−, and 11q23 rearrangements. ¹⁹ Leukemic transformation rates were also similar over a median follow-up of 62 months (Supplementary Figure S1A-B). In terms of MPN driver mutations, JAK2 mutations were more prevalent in the proliferative group (80.5% vs. 63.3%, p = 0.025), whereas CALR and MPL mutations did not differ significantly. JAK2V617F allele burden was comparable between groups. Triple-negative cases were infrequent and evenly distributed between the groups. Cytopenic PMF was associated with a higher prevalence of high molecular risk (HMR) mutations, defined in this study according to the Genetically Inspired Prognostic Scoring System (GIPSS) criteria as the presence of mutations in ASXL1, EZH2, IDH1, IDH2, SRSF2, or U2AF1Q157. Specifically, ASXL1 mutations were significantly more common in the cytopenic group compared to the proliferative group (37.3% vs. 15.8%, p = 0.011), as were EZH2 mutations (11.8% vs. 1.8%, p = 0.035). Furthermore, a greater proportion of patients in the cytopenic cohort harbored multiple HMR mutations (p = 0.024) (Table 2).

Table 2.

Mutational profiles of common myeloid neoplasm-associated genes in cytopenic versus proliferative primary myelofibrosis.

​	Cytopenic (n=51)	Proliferative (n=57)	P value
High molecular risk (HMR) mutation
ASXL1 mutated, n (%)	19 (37.3)	9 (15.8)	0.011
EZH2 mutated, n (%)	6 (11.8)	1 (1.8)	0.035
IDH1 mutated, n (%)	1 (1.9)	1 (1.8)	0.959
IDH2 mutated, n (%)	1 (2)	0	0.288
SRSF2 mutated, n (%)	3 (5.9)	5 (8.8)	0.567
U2AF1 Q157 mutation, n (%)	2 (3.9)	1 (1.8)	0.494
Number of HMR mutations in each patient			0.024
One HMR mutation	17 (32.1)	7 (12.3)	​
Two HMR mutation	4 (7.5)	4 (7.0)	​
Three HMR mutation	2 (3.58)	0	​
Other mutation genes
BCOR mutated, n (%)	2 (3.9)	2 (3.5)	0.910
BCORL1 mutated, n (%)	1 (2)	2 (3.5)	0.625
CUX1 mutated, n (%)	1 (2)	5 (8.8)	0.123
DNMT3A mutated, n (%)	2 (3.9)	4 (7)	0.483
ETV6 mutated, n (%)	0	1 (1.8)	0.342
FBXW7 mutated, n (%)	0	1 (1.8)	0.342
GATA2 mutated, n (%)	0	2 (3.5)	0.177
GNAS mutated, n (%)	1 (2)	1 (1.8)	0.937
KDM6A mutated, n (%)	1 (2)	0	0.288
NOTCH1 mutated, n (%)	0	2 (3.5)	0.177
NRAS mutated, n (%)	1 (2)	1 (1.8)	0.937
PHF6 mutated, n (%)	1 (2)	1 (1.8)	0.937
RUNX1 mutated, n (%)	1 (2)	3 (5.3)	0.364
SETBP1 mutated, n (%)	2 (3.9)	1 (1.8)	0.494
SF3B1 mutated, n (%)	3 (5.9)	1 (1.8)	0.257
STAG2 mutated, n (%)	1 (2)	1 (1.8)	0.937
TET2 mutated, n (%)	7 (13.7)	7 (12.3)	0.823
TP53 mutated, n (%)	5 (9.8)	3 (5.3)	0.368
ZRSR2 mutated, n (%)	2 (3.9)	0	0.131

The commonly used prognostic scoring systems including IPSS, DIPSS+, and MIPSS70 demonstrated significant differences in risk category distribution between cytopenic and proliferative phenotypes. Patients with the cytopenic phenotype had a median IPSS of 3, median DIPSS of 3, and median DIPSS+ of 4, whereas those with the proliferative phenotype had a median IPSS, DIPSS, and DIPSS+ of 1. In contrast, the GIPSS model which is independent of clinical parameters and based solely on molecular and cytogenetic features, also revealed a significant disparity between groups. A greater proportion of cytopenic patients were classified as having higher GIPSS scores compared to proliferative patients (54.9% vs. 26.3%, p = 0.002) (Table 3).

Table 3.

Distribution of patients by prognostic risk category in cytopenic and proliferative primary myelofibrosis.

​	Cytopenic	Proliferative	P value
IPSS risk category (n=111)	​	​	<0.001
Low	2 (3.7)	11 (19.3)	​
Intermediate-1	5 (9.3)	18 (31.6)
Intermediate-2	19 (35.2)	19 (33.3)
High	28 (51.9)	9 (15.8)
DIPSS+ risk category (n=111)	​	​	<0.001
Low	2 (3.8)	14 (25)	​
Intermediate-1	3 (5.7)	21 (37.5)
Intermediate-2	18 (34)	18 (32.1)
High	30 (56.6)	3 (5.4)
MIPSS70 risk category (n=103)	​	​	<0.001
Low	0	6 (11.1)	​
Intermediate	25 (51)	39 (72.2)
High	24 (49)	9 (16.7)
GIPSS risk category (n=108)	​	​	0.023
Low	2 (3.9)	4 (7)	​
Intermediate-1	21 (41.2)	38 (66.7)
Intermediate-2	21 (41.2)	10 (17.5)
High	7 (13.7)	5 (8.8)
Higher GIPSS	28 (54.9)	15 (26.3)	0.002

Sample sizes vary across prognostic models due to differences in available cytogenetic, fibrosis grade and mutation data. MIPSS70 incorporates bone marrow fibrosis grade and high molecular risk mutations, which are not included in IPSS or DIPSS+. GIPSS is solely based on genetic and cytogenetic parameters. Data are presented as number (%).

Treatment was largely supportive, including transfusion support, cytoreductive therapy with hydroxyurea or anagrelide when indicated, and symptom-directed management. With a median follow-up of 62 months, patients with cytopenic PMF exhibited significantly inferior OS compared to those with the proliferative phenotype (Figure 1(A); p < 0.001; HR = 6.579, 95% CI: 3.126–13.848). OS censoring at the time of allogeneic HSCT, performed in 9 patients (7 in the cytopenic group and 2 in the proliferative group), yielded consistent results, with a significant survival difference maintained between groups (HR = 0.162, 95% CI: 0.081–0.323; p < 0.001; Supplemental Figure S1C). Within individual cytopenic components, anemia, thrombocytopenia, and elevated peripheral blood blasts (≥2%) were each associated with significantly shorter survival (Figure 1(B)–(D)). Although patients with leukopenia (WBC <4 × 10⁹/L) also appeared to have worse outcomes, statistical significance was not reached, likely due to the small and unbalanced sample size (n = 7) (Figure 1(E)). Similarly, patients with overt fibrosis had significantly poorer survival compared to those with pre-fibrosis (p = 0.019; Figure 1(F)).

Figure 1.

Kaplan-Meier estimates of overall survival in patients with PMF. (A) Cytopenic versus proliferative phenotype (B) Leukopenia (WBC< 4000/μL) versus normal/elevated WBC (≥4000/μL) (C) Anemia defined as hemoglobin <11 g/dL in males or <10 g/dL in females. (D) Thrombocytopenia (platelet count <100 × 10⁹/L) versus platelet count ≥100 × 10⁹/L. (E) Circulating peripheral blood blasts ≥2% versus <2%. (F) Overt bone marrow fibrosis (MF-2 or MF-3) versus early fibrosis (MF-0 or MF-1).

Immune profiling and pathway analysis

In the differential expression analysis comparing cytopenic to proliferative PMF patients, using Nanostring PanCancer Immune gene expression panel, we were able to identify 12 genes that were up-regulated, and 10 genes down-regulated in cytopenic patients by setting a threshold of a log fold change greater than 1.5 and a P-value of less than 0.05 (Figure 2). The upregulated genes are: TPSAB1, KIT, CCL4, MS4A2, CSF1, IL-2RA, CXCL3, PTGDR2, IFNL1, PDGFRB, ALCAM and IL1RL1. Down-regulated genes are: SIGLEC1, EGR2, CXCL5, JAM3, F13A1, CXCL1, CD86, CD1C and CAMP. Pathway enrichment analysis using the Hallmark gene sets identified significant dysregulation in TNFα-NFκB signaling (p = 0.0097) and IL2-STAT5 signaling (p = 0.0399) pathways in the cytopenic group. In alignment with the Hallmark results, the Bioplanet database analysis also indicated a preferential upregulation of TNF-α (Figure 3).

Figure 2.

Volcano plot shows 12 upregulated genes and 10 downregulated genes in cytopenic versus proliferative primary myelofibrosis. Genes with |log₂ fold change| >1.5 and p < 0.05 are highlighted. Data analysis and figure generation were performed using the ROSALIND® platform.

Figure 3.

Pathway enrichment in cytopenic versus proliferative primary myelofibrosis (PMF). Bubble plots illustrate significantly enriched signaling pathways identified through pathway enrichment analyses in cytopenic versus proliferative PMF patients, using (A) Hallmark and (B) BioPlanet databases. The x-axis denotes statistical significance represented as -log10 (p-value), with pathways toward the right indicating greater enrichment significance. The bubble size reflects the total number of differentially expressed genes within each pathway. Bubble color denotes the directionality of gene regulation, with red indicating predominantly upregulated genes, blue representing predominantly downregulated genes, and intermediate shades (white/pale colors) signifying balanced or neutral dysregulation.

To further delineate the contribution of inflammation other than fibrosis severity, we focused on the patients with overt PMF, stratified by cytopenic versus proliferative phenotype. In this subgroup, several genes remained upregulated in the cytopenic group, including PLA2G6, RUNX1, BMI1, IL3RA, CSF1, IL2RA, KIT, IL1RL1, CXCL12, and FLT3, further supporting inflammatory and stem cell regulation signature (Supplemental Table S1). Gene set enrichment analysis again identified IL-2/STAT5 signaling as the most significantly enriched pathway (adjusted p = 7.14 × 10^-7), followed by IL-6/JAK/STAT3 signaling (Supplemental Figure S2).

BM plasma cytokine, chemokine and growth factors

To validate the immune dysregulation identified by transcriptomic profiling, we measured the concentrations of 42 cytokines, chemokines, and growth factors in bone marrow plasma from 73 PMF patients. Patients with cytopenic PMF showed significantly elevated levels of multiple pro-inflammatory cytokines and myeloid growth factors, including IL-6 (p = 0.0004), G-CSF (p = 0.0016), IL-15 (p = 0.0029), M-CSF (p = 0.0024), TNFα (p = 0.0429), MIP-1α (p = 0.0323), MIG/CXCL9 (p = 0.0186), IL-5 (p = 0.012), IL-10 (p = 0.0381), and IL-12 (p40) (p = 0.0381) (Figure 4). IL-8 (p=0.0513) also trended higher in the cytopenic group. These findings are consistent with enhanced inflammatory signaling and monocytic activation in the cytopenic subset. In contrast, cytokines significantly reduced in the cytopenic group included PDGF-AA (p < 0.0001), PDGF-AB/BB (p = 0.0002), VEGF-A (p = 0.0003), EGF (p = 0.007), and Eotaxin (p = 0.0164) (Figure 4).

Figure 4.

Differential levels of cytokines, chemokines, and growth factors in bone marrow plasma from patients with cytopenic versus proliferative primary myelofibrosis (PMF). (A) Dot plots showing cytokines significantly reduced in patients with cytopenic PMF. (B) Dot plots showing cytokines significantly elevated in cytopenic PMF. Results were calculated after the removal of outliers identified by Grubb’s test. Asterisks indicate levels of statistical significance: *P<0.05, **P<0.01, ***P<0.001, ****P≤0.0001.

Distinguished cytokine signatures from fibrosis-driven effects

To avoid the confounding effects from the severity of marrow fibrosis, we may restrict our analysis on overt fibrosis patients, but this approach would substantially reduce sample size and statistical power. Therefore, we employed regression models to adjust for fibrosis severity across the full cohort to allow a more comprehensive evaluation of cytokine effects on the cytopenic phenotype.

Two logistic regression models were constructed for this purpose. Model A evaluated cytokines associated with the cytopenic phenotype after adjusting for fibrosis grade. In univariate analysis using log-transformed cytokine levels, several inflammatory and myeloid-related factors—including G-CSF, IL-6, IL-8, M-CSF, IL-15, IL-10, IL-12 (p40), MIG/CXCL9, and TNFα—showed significant associations (Supplemental Table S3). In multivariable analysis, only G-CSF (OR 8.09, p = 0.007) and IL-8 (OR 3.95, p = 0.021) remained independently associated with the cytopenic phenotype. Model B examined cytokine levels in patients with overt versus early fibrosis, regardless of cytopenic status (Supplemental Table S4). Several cytokines overlapped with those identified in Model A, but in multivariable analysis, only MCP-1 remained independently associated with overt fibrosis (p = 0.049; OR = 0.052, 95% CI: 0.003–0.987).

Histologic evidence of NFκB activation in cytopenic PMF

Finally, we performed immunohistochemical staining for phosphorylated NFκB p65 (phospho-p65) on bone marrow sections to directly evaluate inflammatory activation. In the cytopenic group, 5 out of 8 sections showed positive staining for NFκB p65, whereas none of the 4 sections in the proliferative group were positive (P = 0.045), indicating higher NFκB phospho-p65 activation in the cytopenic group. Representative images from six cases, three cytopenic and three proliferative—are shown in Figure 5. Marrow from cytopenic PMF patients exhibited variable degrees of nuclear phospho-p65 positivity, indicating downstream NFκB activation, whereas proliferative PMF samples showed minimal to absent staining.

Figure 5.

Phospho-NFκB p65 staining in bone marrow sections from cytopenic and proliferative primary myelofibrosis (PMF) patients. Cytopenic PMF bone marrow sections (A–C) showed variable degrees of nuclear phospho-p65 staining (brown), reflecting activation of the NFκB signaling pathway. In contrast, nuclear staining was largely absent in proliferative PMF samples (D–F), suggesting limited NFκB pathway activation.

Discussion

Our findings align with previous reports that the cytopenic phenotype of PMF is associated with more aggressive clinical features and inferior survival.^14,15,31 Consistent with the Italian study, patients with cytopenic PMF exhibited higher peripheral blood blast percentage, elevated LDH, and more frequent overt fibrosis. ¹⁵ At the molecular level, cytopenic PMF had a lower frequency of JAK2 mutations and was enriched for HMR mutations, particularly ASXL1 and EZH2, both of which are associated with adverse outcomes. Our data further supports that cytopenia defines a molecularly distinct and clinically aggressive subset of PMF.

Cytopenia in PMF is multifactorial, arising from a combination of marrow fibrosis, ineffective hematopoiesis, splenic sequestration, and clonal evolution.^17,31,32 Beyond these established mechanism, our data suggest that immune dysregulation plays an important role in this phenotype. Transcriptomic profiling revealed differential expression of immune and inflammatory genes between cytopenic and proliferative PMF. Genes such as TPSAB1, KIT, CCL4, CSF1, IL2RA, and IL1RL1 were upregulated in cytopenic PMF, implicating cytokine signaling and myeloid activation, whereas SIGLEC1, CD86, and CD1C were downregulated, reflecting altered myeloid and antigen-presentation pathways. Pathway enrichment analysis identified significant activation of TNFα–NFκB and IL-2/STAT5 signaling, highlighting persistent inflammatory signaling in cytopenic PMF.

To further validate inflammatory activation at the tissue level, we examined phosphorylated NFκB p65 (phospho-p65) expression by immunohistochemistry. Nuclear phospho-p65 staining was observed in the majority of cytopenic samples but was absent in proliferative PMF, confirming enhanced NFκB activation in the cytopenic phenotype. NFκB is a key transcriptional regulator of pro-inflammatory cytokines, including IL-6, IL-1β, IL-8, TNFα, and TGFβ, all of which were elevated in our cytokine profiling. ¹⁰ The concordance between transcriptomic, cytokine, and histologic findings strongly supports the presence of a distinct NFκB-driven inflammatory process in cytopenic PMF.

To distinguish the contribution of cytokine dysregulation from fibrosis-driven effects, we employed a modeling strategy that adjusted for fibrosis severity, thereby preserving statistical power while accounting for a key confounding factor. Our findings indicate that while several cytokines were associated with both cytopenia and overt fibrosis in univariate analyses, only IL-8 and G-CSF remained independently linked to the cytopenic phenotype after adjustment. Among these, IL-8 (CXCL8) emerged as particularly relevant. It was significantly elevated in cytopenic PMF and independently associated with this phenotype after adjusting for fibrosis severity. This aligns with recent data by Tefferi et al., who identified IL-8 as an independent prognostic marker in PMF, ³³ and with mechanistic insights from Dunbar et al.’s study, who showed that CXCL8/CXCR2 signaling promotes bone marrow fibrosis and HSPC dysregulation. ³⁴ Our data suggest that IL-8 upregulation may occur independently of fibrosis severity, indicating additional pathogenic roles driving cytopenia. G-CSF elevation likely reflects a reactive response to marrow failure and ineffective hematopoiesis. Collectively, these findings reveal that immune activation—particularly NFκB signaling and IL-8-mediated inflammation—plays a major role in shaping the cytopenic phenotype of PMF.

These findings carry important therapeutic implications. Current JAK inhibitors, while effective in symptom control and spleen reduction, are often limited by treatment-related anemia and thrombocytopenia.^24,25 Next-generation agents that modulate inflammatory pathways may therefore be particularly beneficial for this subgroup. Pacritinib, a dual JAK2/IRAK1 inhibitor, can be administered to patients with severe thrombocytopenia (<50 × 10⁹/L) and attenuates NFκB-mediated inflammation through IRAK1 blockade.^26,27 Momelotinib, which inhibits JAK1, JAK2, and ACVR1, has shown the ability to improve anemia by reducing hepcidin production and enhancing erythropoiesis. ²⁸ The molecular profile of cytopenic PMF, marked by inflammatory activation and cytokine dysregulation, suggests that therapeutic strategies combining JAK inhibition with modulation of NFκB or cytokine signaling may hold promise in this challenging clinical phenotype.

This study has several limitations. First, due to sample availability, transcriptomic and cytokine analyses were performed on non-overlapping patient cohorts, which precluded direct correlation between gene expression and cytokine levels in individual patients. Nevertheless, both analyses were derived from the same institutional cohort and were stratified using consistent clinical criteria, allowing for complementary insights into immune dysregulation across cytopenic and proliferative PMF phenotypes. Second, no prospective power calculation or formal sensitivity analyses was undertaken because the study was retrospective and sample size was constrained by available materials; consequently, some analyses may be underpowered, and results should be interpreted in conjunction with effect sizes. Third, analysis of early-stage disease was constrained by the predominance of overt fibrosis in the cohort, as bone marrow biopsy was frequently deferred in patients with less severe clinical manifestations. Finally, the phospho-p65 IHC was performed on a limited number of samples, which, while supportive, warrants validation in larger cohorts. Future studies in stage-stratified and longitudinally followed cohorts are needed to validate these findings and explore whether targeting IL-8 or NFκB-driven inflammation may offer therapeutic benefit, particularly for patients with cytopenic PMF who face limited treatment options and inferior outcomes.

Conclusion

In summary, our findings demonstrate that cytopenic PMF represents a molecularly distinct and clinically aggressive phenotype characterized by NFκB-driven inflammation and IL-8 upregulation. These highlight the potential of therapies targeting inflammatory and cytokine pathways, alongside JAK inhibition, to improve outcomes in this high-risk subgroup.

ORCID iDs

Sze-Hwei Lee https://orcid.org/0000-0001-7160-2285

Yu-Sung Chang https://orcid.org/0009-0000-4571-3328

Chang-Tsu Yuan https://orcid.org/0000-0003-1438-9187

Ethical considerations

This study was approved by the Institutional Review Board of National Taiwan University Hospital (IRB No. 202207050RINB).

Consent to participate

Written informed consent for specimen collection and research use was obtained from all participants in accordance with institutional and ethical guidelines.

Data Availability Statement

All data supporting the findings of this study are available from the corresponding author upon reasonable request and completion of a data use agreement.*