Upregulation of NFS1 and downregulation of GSDMD as prognostic biomarkers in acute myeloid leukemia: implications for diagnosis and therapeutic strategies Authors: Sara M. Radwan, Dina Mostafa, Rania A. Radwan, Hany A. Elghobary, Wesam R. Elghamry, Amany M. Kamal Published: 2026 ABSTRACT BackgroundDespite advances in the treatment of Acute Myeloid Leukemia (AML), the predictive 5-year overall survival rate is still poor not exceeding 20%. Most patients develop relapse partly due to resistance mechanisms to classical cell death programs. Currently, cytogenetic markers play a crucial role in the early diagnosis, prognostic classification, and therapy selection for AML. Yet, there is a pressing need for the discovery of novel AML biomarkers for a better understanding of the disease’s molecular mechanisms, diagnosis, prognosis assessment, and personalized therapeutic response. The role of cell death-related genes in driving AML progression and spread remains inadequately explored. Among these, ferroptosis and pyroptosis are emerging as critical mechanisms of programmed cell death that are often dysregulated in cancer.MethodsWe recruited 100 newly diagnosed AML patients and 40 healthy matched individuals. Using reverse transcription-quantitative PCR analysis, we assessed the expression levels of NFS1 and Gasdermin D (GSDMD).ResultsOur findings revealed significant NFS1 upregulation (P < 0.0001) and GSDMD downregulation (P < 0.0001) in AML patients compared to healthy controls. Following treatment, we observed a significant downregulation of NFS1 expression (P < 0.001) and upregulation of GSDMD (P < 0.001). Furthermore, higher pretreatment levels of NFS1 and lower pretreatment levels of GSDMD correlated with poor prognosis (P < 0.001) and decreased one-year survival (P < 0.001).ConclusionOur study suggests that NFS1 and GSDMD may serve as early diagnostic and prognostic markers in AML, offering potential targets for improved therapeutic strategies. Background Acute myeloid leukemia (AML) is the most common acute leukemia in adults, accounting for approximately 80% of cases, with a median age at diagnosis of around 68 years. Its global incidence is estimated at 3–5 cases per 100,000 population per year [1]. AML is an aggressive hematological malignancy that is strongly associated with poor clinical outcomes [2]. Most AML patients experience relapse, with an overall survival (OS) rate below 30% [3]. Gaining deeper insights into the molecular mechanisms of AML could help in discovering new diagnostic and prognostic biomarkers, as well as potential therapeutic targets to enhance patient outcomes. AML is defined by the rapid proliferation of abnormal hematopoietic precursor cells, specific to certain lineages, which fail to mature into functional granulocytes or monocytes within the bone marrow (BM) microenvironment [4–7]. Chemotherapy-induced apoptosis is a primary mechanism for eliminating leukemic cells [8]. However, relapsed or refractory AML frequently exhibits resistance to apoptotic processes [9, 10], necessitating the development of innovative strategies for driving leukemic cell death [11]. Recent advancements in cancer biology have identified novel cell death mechanisms, such as ferroptosis and pyroptosis, which play critical roles in AML’s pathogenesis and treatment response. Ferroptosis is an iron-dependent, non-apoptotic form of programmed cell death (PCD) initiated by lipid peroxidation, leading to toxic lipid peroxidation and mitochondrial dysfunction [12]. Ferroptosis is crucial in controlling cancer progression and holds potential for antitumor therapy and prognosis prediction [13, 14]. Inducing ferroptosis presents a promising strategy for triggering tumor cell death, especially in managing treatment-resistant cancers like AML [15, 16]. AML is particularly sensitive to ferroptosis, relying on numerous ferroptosis-associated genes. Despite recent attention on ferroptosis in solid tumors, its prognostic significance and function in AML remain underexplored. Pyroptosis is a strongly pro-inflammatory type of PCD driven by gasdermin proteins, predominantly Gasdermin D (GSDMD), which are activated by inflammatory caspases, including caspase-1, -4, -5, and − 11 [17]. Upon activation, these caspases cleave GSDMD, releasing its N-terminal fragment, which creates pores in the cell membrane. These pores lead to cell swelling, lysis, and pro-inflammatory cytokine release such as IL-1β and IL-18 [18, 19]. Pyroptosis was found to be strongly linked to the occurrence, progression, and prognosis of various cancers, particularly leukemia [20, 21]. However the molecular mechanisms underlying pyroptosis in AML are not well understood [22]. A significant number of genes have been recognized as regulators of ferroptosis and pyroptosis, functioning as either inhibitors or inducers. In this study, we conducted a comprehensive analysis of NFS1 and GSDMD expression, which are related to ferroptosis and pyroptosis, respectively, and their association with the prognosis of AML patients. NFS1 encodes the cysteine desulfurase enzyme, which is involved in iron-sulfur (Fe–S) cluster biogenesis and is associated with reduced ferroptosis. In contrast, GSDMD, a key executor of pyroptosis, is pivotal in regulating inflammatory responses and tumor microenvironment dynamics in AML. Our findings designate that higher expression levels of NFS1 and lower expression levels of GSDMD are correlated with poor prognosis and decreased one-year survival in AML patients. Subjects and methods Ethical approval and participant consent This study followed the principles of the Declaration of Helsinki and carried approval from the Ethical Committee of Research at the Faculty of Medicine, Ain Shams University, under federal assurance number FWA 000017585. Written informed consent was obtained from all participants. Subjects The research involved 100 newly diagnosed AML patients, recruited from the Internal Medicine Department, Clinical Hematology, and Bone Marrow Transplantation Unit at Ain Shams University. These patients were monitored for 12 months following treatment. Additionally, 40 healthy individuals, age and sex- matched, served as the control group. Diagnosis of AML was based on the French-American-British (FAB) classification criteria and World Health Organization guidelines [23]. This process utilized morphological assessments of Wright-Giemsa-stained bone marrow aspirate smears, supplemented with immunophenotyping analyses of leukemic cells. Inclusion criteria for patient participation were: (i) age over 18 years and (ii) a recent diagnosis of AML suitable for chemotherapy. Exclusion criteria included: (i) secondary AML, (ii) relapsed AML, and (iii) AML of M3 subtype FAB classification due to receiving a different treatment protocol. Patient histories, clinical examinations, and relevant data from medical records were systematically collected. Patients received standard induction chemotherapy as per the National Comprehensive Cancer Network (NCCN) 2019 guidelines for AML. The treatment protocol, known as “3 + 7,” involved anthracycline and cytarabine administration for all AML subtypes [24]. On day 28 post-treatment initiation, surviving patients were evaluated and categorized as either responders, achieving complete remission (CR), or non-responders, classified as refractory to chemotherapy. CR was defined by the following parameters: an absolute neutrophil count exceeding 1,000/µl, platelet count of at least 100,000/µl, BM blasts under 5%, and no evidence of extramedullary disease (EMD). Post achieving complete remission, 29 patients (29%) underwent allogenic bone marrow transplantation (BMT) from fully matched related donor after receiving one to two cycles of high dose cytarabine. Eighteen patients (18%) who either have favorable cytogenetics [positive t(8;21) or inv (16)/ t(16;16), negative flt3 gene] or who has no related matched donor for allogenic BMT, received consolidation treatment with 4 cycles of high dose cytarabine (HIDAC). The rest of the patients (53 patients) (53%) passed away either before starting consolidation treatment or during the first 2 cycles of consolidation. Patients alive at the conclusion of the 12-month follow-up were censored, as the study only assessed one-year survival. Methods Blood sample collection Peripheral blood samples were collected from both AML patients, prior to and after completing 28 days of induction therapy, and control participants. Samples were drawn into Na2 EDTA-containing vacutainer tubes for subsequent analyses, including complete blood count (CBC) determination, cytogenetic analysis, and RNA extraction for molecular analysis. Complete blood count determination Hemoglobin, platelet count, total leukocyte count, absolute neutrophil count and absolute lymphocyte count, peripheral blood blast count, absolute peripheral blood blast count and BM blast count were measured using Z2TM Coulter Counter, Analyzer, Coulter Electronics, Hialeah, Florida. Immunophenotype analysis Flow cytometry technique was used for immunophenotype analysis using diagnostic kits [CD34, CD13, CD33, myeloperoxidase, human leukocyte antigen-DR isotype (HLA-DR) HLA-DR, CD117, CD2 and CD19; Beckman Coulter, Inc.] supplied by Beckman Coulter, Fullerton, California [25]. Cytogenetic analysis Fluorescence in situ hybridization (FISH) for t(8;21) and inv (16)/t(16;16), as well as PCR analysis for FLT3 mutations were utilized to detect cytogenetic abnormalities. FISH employed fluorophore-labeled locus-specific identifier DNA probes (Abbott Molecular, Chicago, Illinois). Hybridization signals were analyzed via fluorescence microscopy, following the method outlined by Fröhling et al. [26]. Gene expression analysis Total RNA was extracted from whole blood using the QIAamp RNA Blood Mini Kit (Qiagen GmbH), as per the manufacturer’s instructions. RNA was then reverse-transcribed into complementary DNA (cDNA) using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Thermo Fisher Scientific, Inc.) and stored at − 20 °C for further analysis. Quantitative PCR was conducted on the Rotor-Gene Q® Real-Time PCR cycler (Qiagen GmbH) using standard thermocycling conditions. All reactions were conducted in technical triplicates, and appropriate negative controls, including no-template controls and no-reverse-transcription controls, were included in each run to exclude contamination and genomic DNA amplification. Gene expression levels were assessed using TaqMan assays specific for NFS1 (catalog no. Hs00221619_m1) and GSDMD (catalog no. Hs00986748_g1), with GAPDH (catalog no. Hs04420566_g1) serving as the internal reference gene. Relative expression levels were normalized and calculated using the 2−∆∆Ct method, where ∆∆Ct=(Cttarget gene-CtGAPDH) sample-(Ct target gene-CtGAPDH) calibrator [27]. Statistical analysis Statistical analysis was conducted using IBM© SPSS© Statistics version 24 (IBM© Corp., Armonk, NY) and MedCalc© version 20.218 (MedCalc Software Ltd, Ostend, Belgium; https://www.medcalc.org; 2023). Skewed numerical data are presented as the median with interquartile range (IQR), and comparisons between two groups were performed using the Mann-Whitney U-test. Categorical data are presented as counts and percentages, with differences assessed using the Pearson chi-squared test. Correlations between numerical variables were evaluated using Spearman’s rank correlation, with Spearman’s rho interpreted as follows: <0.2 = very weak, 0.2–0.39 = weak, 0.4–0.59 = moderate, 0.6–0.79 = strong, and ≥ 0.8 = very strong. The discriminative value of biomarkers was assessed via receiver-operating characteristic (ROC) curve analysis. The area under the ROC curve (AUC) was categorized as follows: <0.6 = fail, 0.6–0.69 = poor, 0.7–0.79 = fair, 0.8–0.89 = good, and ≥ 0.9 = excellent. Internal validation of the diagnostic performance of NFS1 and GSDMD was performed using bootstrap resampling with 1,000 iterations. This procedure allowed estimation of 95% confidence intervals (CIs) for the AUC and assessment of model stability. Survival analysis was performed using the Kaplan–Meier (KM) method, followed by Log-rank test to compare KM curves. Statistical significance was set at P < 0.05. Multivariate Cox proportional hazards regression analysis was performed to identify independent predictors of 1-year overall survival. Finally, time-dependent ROC curve analysis was performed at 3, 6, 9, and 12 months after diagnosis. Results Clinical characteristics of patients with AML Demographic data and baseline characteristics of the 100 AML patients are presented in Table 1. Table 1The clinical and demographic data of the studied groupsCharacteristicsControl subjectsAML patientsDemographic parameters N40100 Gender (M/F)21/1952/48 Age (years)46 (43–50)39 (23–50)Clinical parameters Hemoglobin (g/dl)14.5 (13.9–15.4)8 (6.5–9.2)*** Total leukocyte count (thousands/mm3)5.7 (4.3–7.8)13.5 (5.1–26.2)*** Absolute neutrophil count (thousands /mm3)4.2 (3.6–5.7)2.8 (1.4–5.4)* Absolute lymphocyte count (thousands /mm3)2.4 (1.9–2.6)2.9 (1.4–5.2) Platelets (thousands /mm3)261 (247–328)19 (11–41)*** NFS1 (fold expression)0.96 (0.93–0.99)9.74 (4.51–14.5)*** GSDMD (fold expression)1 (0.98–1.02)0.49 (0.29–0.95)***Extramedullary disease, n +-30 --70FAB classification, n M0-6 M1-24 M2-40 M4-18 M5-12Prognosis according to cytogenetic studies, n Favorable-31 Unfavorable-69Response to treatment, n Responsive-46 Resistant-54Immunotyping results, n CD13 +-67 --33 CD33 +-41 --59 MPO +-70 --30 CD117+-45 --55 HLA-DR +-64 --36 CD34 +-48 --52Results are represented as median (25th quartile–75th quartile)AML Acute myeloid leukemia, FAB French-American-British, CD Cluster of differentiation, HLA-DR Human leukocyte antigen-DR isotype***Significantly different from healthy control group at P < 0.001**Significantly different from healthy control group at P < 0.01*Significantly different from healthy control group at P < 0.02 Table 1The clinical and demographic data of the studied groupsCharacteristicsControl subjectsAML patientsDemographic parameters N40100 Gender (M/F)21/1952/48 Age (years)46 (43–50)39 (23–50)Clinical parameters Hemoglobin (g/dl)14.5 (13.9–15.4)8 (6.5–9.2)*** Total leukocyte count (thousands/mm3)5.7 (4.3–7.8)13.5 (5.1–26.2)*** Absolute neutrophil count (thousands /mm3)4.2 (3.6–5.7)2.8 (1.4–5.4)* Absolute lymphocyte count (thousands /mm3)2.4 (1.9–2.6)2.9 (1.4–5.2) Platelets (thousands /mm3)261 (247–328)19 (11–41)*** NFS1 (fold expression)0.96 (0.93–0.99)9.74 (4.51–14.5)*** GSDMD (fold expression)1 (0.98–1.02)0.49 (0.29–0.95)***Extramedullary disease, n +-30 --70FAB classification, n M0-6 M1-24 M2-40 M4-18 M5-12Prognosis according to cytogenetic studies, n Favorable-31 Unfavorable-69Response to treatment, n Responsive-46 Resistant-54Immunotyping results, n CD13 +-67 --33 CD33 +-41 --59 MPO +-70 --30 CD117+-45 --55 HLA-DR +-64 --36 CD34 +-48 --52Results are represented as median (25th quartile–75th quartile)AML Acute myeloid leukemia, FAB French-American-British, CD Cluster of differentiation, HLA-DR Human leukocyte antigen-DR isotype***Significantly different from healthy control group at P < 0.001**Significantly different from healthy control group at P < 0.01*Significantly different from healthy control group at P < 0.02 The clinical and demographic data of the studied groups Results are represented as median (25th quartile–75th quartile) AML Acute myeloid leukemia, FAB French-American-British, CD Cluster of differentiation, HLA-DR Human leukocyte antigen-DR isotype ***Significantly different from healthy control group at P < 0.001 **Significantly different from healthy control group at P < 0.01 *Significantly different from healthy control group at P < 0.02 Expression of NFS1 and GSDMD in AML patients compared to healthy control group NFS1 expression was significantly elevated in AML patients compared to healthy individuals, with median relative expression levels of 9.74 (IQR, 4.51–14.5) and 0.96 (IQR, 0.93–0.99), respectively (P < 0.0001; Table 1). In contrast, GSDMD expression was significantly reduced in AML patients compared to healthy individuals, with median relative expression levels of 0.49 (IQR, 0.29–0.95) and 1 (IQR, 0.98–1.02), respectively (P < 0.0001; Table 1). Additionally, ROC curve analysis was conducted to establish the optimal cut-off value for NFS1 and GSDMD expression to differentiate between AML patients and healthy individuals. The analysis revealed a strong diagnostic value for NFS1, with an AUC of 1 and an optimal cut-off value of > 1.31, while GSDMD demonstrated a moderate diagnostic value with an AUC of 0.85 and an optimal cut-off value of ≤ 0.96. The diagnostic sensitivity and specificity for NFS1 were 100% and 100%, respectively, whereas for GSDMD, they were 77% and 95%, respectively. Internal validation of the diagnostic performance of NFS1 and GSDMD was performed using bootstrap resampling with 1,000 iterations, allowing estimation of 95% CIs for the AUC (Fig. 1A, B). Fig. 1Receiver-operating characteristic (ROC) curves for evaluating the diagnostic value of (A) NFS1 and (B) GSDMD expression levels Receiver-operating characteristic (ROC) curves for evaluating the diagnostic value of (A) NFS1 and (B) GSDMD expression levels NFS1 and GSDMD expression levels with respect to AML patients’ FAB classification NFS1 expression was significantly higher in patients with the less differentiated subtypes (M0, M1 and M2) compared with M4 subtype (Fig. 2A). On the other hand, GSDMD expression was significantly lower in patients with the less differentiated subtypes (M0 and M1) as compared to M4 subtype (Fig. 2B). Fig. 2Fold expression levels of (A) NFS1 and (B) GSDMD across AML subtypes according to the FAB classification: M0 (n = 6), M1 (n = 24), M2 (n = 40), M4 (n = 18), and M5 (n = 12). Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Fold expression levels of (A) NFS1 and (B) GSDMD across AML subtypes according to the FAB classification: M0 (n = 6), M1 (n = 24), M2 (n = 40), M4 (n = 18), and M5 (n = 12). Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Association between NFS1 and GSDMD pre-treatment expression and CBC parameters in AML patients A strong negative significant correlation was observed between NFS1 and GSDMD (r = -0.766; P < 0.01). To further explore the clinical relevance of elevated NFS1 expression and reduced GSDMD expression in AML patients, we examined the association between their expression levels and various clinical characteristics. Both NFS1 and GSDMD expressions showed significant associations with platelets and BM blast counts in AML patients. NFS1 exhibited a negative correlation with platelets (r = -0.341; P < 0.001) and a positive correlation with BM blasts (r = 0.26; P < 0.01). Conversely, GSDMD demonstrated a positive correlation with platelets (r = 0.352; P < 0.001) and a negative correlation with BM blasts (r = -0.411; P < 0.001). However, no statistically significant associations were observed between their expression and hemoglobin, white blood cell and peripheral blood blast counts (Table 2). Table 2Correlation coefficients (r) of NFS1, GSDMD, and CBC parameters in the studied groupsVariableNFS1GSDMDNFS1--0.766**Hemoglobin0.1030.1Platelets0.341**0.352**Total leukocyte count-0.1530.134Absolute neutrophil count-0.157-0.121Absolute lymphocyte count0.093-0.083Peripheral blood blasts0.049-0.066BM blasts0.26**-0.411**Absolute peripheral blood blast count-0.007-0.058**Correlation is significant at the 0.01 level (two-tailed) Table 2Correlation coefficients (r) of NFS1, GSDMD, and CBC parameters in the studied groupsVariableNFS1GSDMDNFS1--0.766**Hemoglobin0.1030.1Platelets0.341**0.352**Total leukocyte count-0.1530.134Absolute neutrophil count-0.157-0.121Absolute lymphocyte count0.093-0.083Peripheral blood blasts0.049-0.066BM blasts0.26**-0.411**Absolute peripheral blood blast count-0.007-0.058**Correlation is significant at the 0.01 level (two-tailed) Correlation coefficients (r) of NFS1, GSDMD, and CBC parameters in the studied groups **Correlation is significant at the 0.01 level (two-tailed) Impact of induction therapy on NFS1 and GSDMD Expression in AML To examine how NFS1 and GSDMD expression levels change in response to therapy, we compared NFS1 and GSDMD expression before and after therapy. Building upon pretreatment correlations between the two factors, our results revealed distinct responses of NFS1 and GSDMD to therapy in AML. Following treatment, NFS1 expression exhibited a significant decrease (P < 0.001), suggesting a potential role in therapeutic response. Conversely, GSDMD expression levels showed a significant increase (P < 0.001) in response to therapy. These contrasting responses underscore the intricate interplay between different molecular pathways in AML and may offer valuable insights into the mechanisms underlying treatment efficacy and resistance. Further elucidation of the dynamics of NFS1 and GSDMD in response to therapy shows potential for refining treatment strategies and enhancing patient outcomes in AML (Fig. 3). Fig. 3Fold expression levels of NFS1 and GSDMD before and after treatment in the whole patient cohort (n = 100). Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Fold expression levels of NFS1 and GSDMD before and after treatment in the whole patient cohort (n = 100). Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test NFS1 and GSDMD pre-treatment expression levels and prognosis in patients with AML AML patients were classified into favorable and unfavorable prognostic groups based on cytogenetic analysis by FISH and in accordance with NCCN guidelines (Version 2.2025). The favorable-risk group included patients with normal karyotype harboring t(8;21) or inv (16)/t(16;16) and negative FLT3 mutation. The high-risk (unfavorable) group included patients with normal karyotype without t(8;21) or inv (16)/t(16;16), those with complex karyotype, monosomal karyotype, or positive FLT3 mutation. Most patients presented with unfavorable cytogenetics (69 patients) as compared to 31 with favorable cytogenetics. Patients in unfavorable prognosis group with intermediate and poor risk cytogenetics showed significantly higher expression levels of NFS1 (median: 12.88 vs. 4.42, P < 0.001) and significantly lower expression levels of GSDMD (median: 0.35 vs. 0.98, P < 0.001) compared to patients with favorable prognosis (Fig. 4). Fig. 4Fold expression levels of NFS1 and GSDMD in patients with favorable (n = 31) or unfavorable (n = 69) cytogenetics. Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Fold expression levels of NFS1 and GSDMD in patients with favorable (n = 31) or unfavorable (n = 69) cytogenetics. Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Correlation between NFS1 and GSDMD pre-treatment expression level and treatment response in AML patients To evaluate the association between NFS1 and GSDMD pre-treatment expression levels and treatment response in the whole patient cohort (n = 100), patients were categorized as either treatment-responsive or treatment-resistant, including patients who died after 28 days of induction therapy. Pretreatment NFS1 expression was significantly higher in the treatment-resistant group compared to the treatment-responsive group (median: 14.18 vs. 4.47, P < 0.001) (Fig. 5). On the contrary, Pretreatment GSDMD expression was significantly lower in the treatment-resistant group compared to the treatment-responsive group (median: 0.29 vs. 0.97, P < 0.001) (Fig. 5). Fig. 5Fold expression levels of NFS1 and GSDMD in the whole patient cohort (n = 100), categorized as responders (n = 46) and non-responders (n = 54) to treatment. Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Fold expression levels of NFS1 and GSDMD in the whole patient cohort (n = 100), categorized as responders (n = 46) and non-responders (n = 54) to treatment. Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test At the end of the induction therapy cycle on Day 28 (D28), the study groups were compared based on their NFS1 and GSDMD expression levels and treatment outcomes. Six patients passed away within 28 days of induction therapy. Those who did not survive by D28 had significantly higher NFS1 expression and significantly lower GSDMD expression (Table 3). The 94 survivors were reassessed for their response to chemotherapy and were classified into 30 responders and 64 non-responders. A comparison of NFS1 and GSDMD expression between responders and non-responders showed that responders had significantly lower NFS1 levels and higher GSDMD levels compared to non-responders (Table 3). The data suggests a potential association between higher NFS1 and lower GSDMD pre-treatment expression and poor outcomes, both in terms of survival on D28 and response to induction chemotherapy. Table 3Association between NFS1 and GSDMD pre-treatment expression and post-induction chemotherapy outcomeDifferent D28 treatment outcomePatients(n)Median NFS1 fold expression P Median GSDMD fold expression P Survival on D28 (n = 100) Alive948.9< 0.0010.5< 0.001 Dead616.040.19Response to induction chemotherapy in survived patients (n = 94) Responder304.2< 0.0010.99< 0.001 Resistant6410.660.36D28, day 28 at the end of induction chemotherapy cycle Table 3Association between NFS1 and GSDMD pre-treatment expression and post-induction chemotherapy outcomeDifferent D28 treatment outcomePatients(n)Median NFS1 fold expression P Median GSDMD fold expression P Survival on D28 (n = 100) Alive948.9< 0.0010.5< 0.001 Dead616.040.19Response to induction chemotherapy in survived patients (n = 94) Responder304.2< 0.0010.99< 0.001 Resistant6410.660.36D28, day 28 at the end of induction chemotherapy cycle Association between NFS1 and GSDMD pre-treatment expression and post-induction chemotherapy outcome D28, day 28 at the end of induction chemotherapy cycle NFS1 and GSDMD pre-treatment expression levels and survival in patients with AML Patients were monitored for 12 months, and by the end of the follow-up period, only 30 of the 100 enrolled patients remained alive. Patients who ultimately died exhibited significantly higher pre-treatment NFS1 expression levels compared to those who survived (median: 11.77 vs. 4.20, P < 0.001). This suggests a potential association between higher initial NFS1 expression and poorer survival outcomes. Following treatment, there was a decrease in NFS1 expression for both groups. Similar to NFS1, patients who died had significantly less GSDMD expression levels compared to survivors before treatment (median: 0.35 vs. 0.99, P < 0.001). This indicates a potential correlation between lower pre-treatment GSDMD expression and worse survival outcomes. After treatment, both groups showed an increase in GSDMD expression. The findings suggest that pre-treatment levels of NFS1 and GSDMD may serve as predictive biomarkers for one-year survival in AML patients (Fig. 6). Fig. 6Fold expression levels of NFS1 and GSDMD in the whole patient cohort (n = 100), categorized as survivors (n = 30) and non-survivors (n = 70). Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test Fold expression levels of NFS1 and GSDMD in the whole patient cohort (n = 100), categorized as survivors (n = 30) and non-survivors (n = 70). Box plots represent the interquartile range, the line inside the box represents the median, and the bars represent the minimum and maximum values. Shown p-values are for Mann-Whitney U-test The Kaplan–Meier survival analysis revealed that patients with high NFS1 expression had a significantly lower overall one-year survival rate compared to those with low NFS1 expression (P < 0.0001; Fig. 7A). Conversely, patients with lower GSDMD expression exhibited a significantly reduced overall one-year survival rate compared to those with higher GSDMD expression (P < 0.0001; Fig. 7B). Fig. 7Kaplan–Meier estimates of one-year survival according to low or high (A) NFS1 and (B) GSDMD expression levels Kaplan–Meier estimates of one-year survival according to low or high (A) NFS1 and (B) GSDMD expression levels Moreover, to determine whether NFS1 and GSDMD expression levels independently predict survival outcomes, multivariate Cox proportional hazards regression analysis was performed to identify independent predictors of 1-year overall survival. Variables that showed clinical relevance or statistical significance in univariate analysis were entered into the model using enter, backward likelihood, and forward likelihood methods. Across all three models, elevated NFS1 expression was consistently associated with poor survival, whereas higher GSDMD expression independently predicted improved survival. Female sex and FAB subtype were also retained as independent predictors in the final forward model, while age lost statistical significance after adjustment (Tables 4 and 5). Table 4Cox proportional hazards regression models for the prediction of 1-year survival as based on the combined expression levels of NFS1 and GSDMDModel 1 (Enter Method)95.0% CI for Exp(B)VariableBSEWald(df = 1)P valueExp(B)LowerUpperAge (years)-0.0420.0139.8940.0020.9590.9350.984Female sex (= 1)0.6950.3364.2880.0382.0031.0383.867FAB 0 (= 1)1.1660.5993.7880.0523.2090.99210.383Unfavorable cytogenetics (= 1)1.0640.6342.8140.0932.8970.83610.037Extramedullary involvement (= 1)0.1150.4130.0770.7811.1220.5002.518Non-Response to treatment0.3470.9750.1260.7221.4140.2099.554NFS1 (fold expression)0.2600.07312.631< 0.0011.2971.1241.498GSDMD (fold expression)-10.0091.99125.258< 0.0010.000050.0000010.002Model 2 (Backward Likelihood Method)95% CI for Exp(B)VariableBSEWald(df = 1)P valueExp(B)LowerUpperAge (years)-0.0410.0139.9610.0020.9590.9350.984Female sex (= 1)0.7190.3065.5260.0192.0521.1273.738FAB 0 (= 1)1.1460.5724.0130.0453.1461.0259.653Unfavorable cytogenetics (= 1)1.0900.6243.0540.0812.9750.87610.105NFS1 (fold expression)0.2650.07213.582< 0.0011.3031.1321.500GSDMD (fold expression)-10.0201.93326.876< 0.0010.000040.0000010.002Model 3 (Forward Likelihood Method)95% CI for Exp(B)VariableBSEWald(df = 1)P valueExp(B)LowerUpperAge (years)-0.0390.0139.0940.0030.9620.9380.987Female sex (= 1)0.7170.2995.7520.0162.0491.1403.681FAB 0 (= 1)1.2320.5674.7310.0303.4301.13010.412NFS1 (fold expression)0.2770.07115.237< 0.0011.3191.1481.516GSDMD (fold expression)-10.5751.90330.889< 0.0010.000030.0000010.001 Table 4Cox proportional hazards regression models for the prediction of 1-year survival as based on the combined expression levels of NFS1 and GSDMDModel 1 (Enter Method)95.0% CI for Exp(B)VariableBSEWald(df = 1)P valueExp(B)LowerUpperAge (years)-0.0420.0139.8940.0020.9590.9350.984Female sex (= 1)0.6950.3364.2880.0382.0031.0383.867FAB 0 (= 1)1.1660.5993.7880.0523.2090.99210.383Unfavorable cytogenetics (= 1)1.0640.6342.8140.0932.8970.83610.037Extramedullary involvement (= 1)0.1150.4130.0770.7811.1220.5002.518Non-Response to treatment0.3470.9750.1260.7221.4140.2099.554NFS1 (fold expression)0.2600.07312.631< 0.0011.2971.1241.498GSDMD (fold expression)-10.0091.99125.258< 0.0010.000050.0000010.002Model 2 (Backward Likelihood Method)95% CI for Exp(B)VariableBSEWald(df = 1)P valueExp(B)LowerUpperAge (years)-0.0410.0139.9610.0020.9590.9350.984Female sex (= 1)0.7190.3065.5260.0192.0521.1273.738FAB 0 (= 1)1.1460.5724.0130.0453.1461.0259.653Unfavorable cytogenetics (= 1)1.0900.6243.0540.0812.9750.87610.105NFS1 (fold expression)0.2650.07213.582< 0.0011.3031.1321.500GSDMD (fold expression)-10.0201.93326.876< 0.0010.000040.0000010.002Model 3 (Forward Likelihood Method)95% CI for Exp(B)VariableBSEWald(df = 1)P valueExp(B)LowerUpperAge (years)-0.0390.0139.0940.0030.9620.9380.987Female sex (= 1)0.7170.2995.7520.0162.0491.1403.681FAB 0 (= 1)1.2320.5674.7310.0303.4301.13010.412NFS1 (fold expression)0.2770.07115.237< 0.0011.3191.1481.516GSDMD (fold expression)-10.5751.90330.889< 0.0010.000030.0000010.001 Cox proportional hazards regression models for the prediction of 1-year survival as based on the combined expression levels of NFS1 and GSDMD Table 5Diagnostics for the alternative cox proportional hazards models for prediction of 1-year survivalModelCovariates included (n)–2 Log Likelihoodχ² (df)P valueAICΔAIC*Null–595.580––––Model 1 (Enter)8435.616159.964 (8)< 0.0001451.6163.789Model 2 (Backward LR)6435.827159.753 (6)< 0.0001447.8270.000Model 3 (Forward LR)5438.857156.723 (5)< 0.0001448.8571.030AIC Akaike Information Criterion, df Degrees of freedom*ΔAIC = AIC – AIC of the best-fitting model (Model 2). Lower ΔAIC indicates stronger relative support Table 5Diagnostics for the alternative cox proportional hazards models for prediction of 1-year survivalModelCovariates included (n)–2 Log Likelihoodχ² (df)P valueAICΔAIC*Null–595.580––––Model 1 (Enter)8435.616159.964 (8)< 0.0001451.6163.789Model 2 (Backward LR)6435.827159.753 (6)< 0.0001447.8270.000Model 3 (Forward LR)5438.857156.723 (5)< 0.0001448.8571.030AIC Akaike Information Criterion, df Degrees of freedom*ΔAIC = AIC – AIC of the best-fitting model (Model 2). Lower ΔAIC indicates stronger relative support Diagnostics for the alternative cox proportional hazards models for prediction of 1-year survival AIC Akaike Information Criterion, df Degrees of freedom *ΔAIC = AIC – AIC of the best-fitting model (Model 2). Lower ΔAIC indicates stronger relative support Finally, to appropriately evaluate the time-specific discriminative performance of the multivariable Cox proportional hazards model, time-dependent ROC curve analysis was performed at 3, 6, 9, and 12 months after diagnosis. The prognostic index derived from the final Cox model was used as a continuous predictor, appropriately accounting for censored survival data. The area under the time-dependent ROC curve demonstrated excellent discrimination across all time points (Table 6; Fig. 8A-D), with values ranging from 0.958 (95% CI: 0.92–0.98) at 3 months to 0.927 (95% CI: 0.88–0.96) at 12 months. Although a gradual decline in AUC was observed over time, the model maintained strong predictive accuracy throughout the first year following diagnosis. These findings indicate robust time-dependent discrimination ability of the proposed prognostic model. Table 6Summary of the time-dependent ROC curves for survival at 3 months, 6 months, 9 months and 12 monthsTime point (months)AUCOptimal cutpoint (xb)*Predicted survival at cutpoint (%)†Sensitivity (%)Specificity (%)Youden index30.958–0.87692.187.888.30.76160.956–2.91084.783.991.70.75690.935–3.68675.881.990.30.722120.927–7.82868.479.891.70.715*xb = linear indicator (prognostic index as calculated from the Cox model). Optimal cutpoint defined by the Youden index (maximising sensitivity + specificity – 1)†Predicted probability of surviving to the specified time point for a patient with a linear predictor equal to the optimal cutpoint. Calculated from the baseline survivor function of the Cox model Table 6Summary of the time-dependent ROC curves for survival at 3 months, 6 months, 9 months and 12 monthsTime point (months)AUCOptimal cutpoint (xb)*Predicted survival at cutpoint (%)†Sensitivity (%)Specificity (%)Youden index30.958–0.87692.187.888.30.76160.956–2.91084.783.991.70.75690.935–3.68675.881.990.30.722120.927–7.82868.479.891.70.715*xb = linear indicator (prognostic index as calculated from the Cox model). Optimal cutpoint defined by the Youden index (maximising sensitivity + specificity – 1)†Predicted probability of surviving to the specified time point for a patient with a linear predictor equal to the optimal cutpoint. Calculated from the baseline survivor function of the Cox model Summary of the time-dependent ROC curves for survival at 3 months, 6 months, 9 months and 12 months *xb = linear indicator (prognostic index as calculated from the Cox model). Optimal cutpoint defined by the Youden index (maximising sensitivity + specificity – 1) †Predicted probability of surviving to the specified time point for a patient with a linear predictor equal to the optimal cutpoint. Calculated from the baseline survivor function of the Cox model Fig. 8Time-dependent ROC curve for survival at (A) 3 months, (B) 6 months, (C) 9 months and (D) 12 months as estimated from the Cox regression model. AUC = area under the ROC curve, Se = sensitivity, Sp = specificity Time-dependent ROC curve for survival at (A) 3 months, (B) 6 months, (C) 9 months and (D) 12 months as estimated from the Cox regression model. AUC = area under the ROC curve, Se = sensitivity, Sp = specificity Discussion AML often exhibits significant resistance to apoptosis, which poses a major challenge for effective treatment [28]. This resistance underscores the need to explore alternative cell death mechanisms that may be relevant to AML. Recent advancements have highlighted the potential of alternative pathways, such as ferroptosis and pyroptosis, in inducing leukemic cell death [29]. Both are complex processes regulated by various genes. Numerous studies have identified novel ferroptosis-related gene signatures that can predict the prognosis of hepatocellular carcinoma [30, 31], lung adenocarcinoma [32], ovarian carcinoma [33], uveal melanoma [34], colorectal cancer [35], papillary thyroid carcinoma [36], cervical cancer [37] and glioma [38–40]. Additionally, recent research has identified pyroptosis-related gene signatures and revealed their correlation with ovarian cancer [41] and AML [20] prognosis. Consequently, there is significant interest in examining the expression patterns and clinical associations of these cell death related genes in AML. However, clinical studies that directly address the associations between the ferroptosis- and pyroptosis-related gene expressions and patient outcomes in AML patients are still lacking. Here, our study focuses on the expression patterns and prognostic association of NFS1 and GSDMD in AML patients. NFS1, involved in Fe–S cluster biogenesis, is linked to reduced ferroptosis, while GSDMD, a key executor of pyroptosis, plays a role in modulating inflammatory responses and tumor microenvironment dynamics [42, 43]. The function of NFS1 in AML pathogenesis was confirmed by in vitro analysis of knockdown (KD) cell lines where NFS1 KD significantly restrained cell proliferation and promoted cell death [44]. Moreover, in mouse myeloid cells, activation of the caspase-1 signaling pathway through the inflammasome sensor Nlrp1b resulted in the cleavage and activation of GSDMD, leading to pore formation in the plasma membrane, which induced cell death and suppressed AML progression [45–47]. Increased expression of GSDMD caused the induction of pyroptosis [48], while chemical disruption of GSDMD inhibited cell death [49]. However, their clinical relevance in AML patients was not previously investigated. By examining these molecules, we aimed to assess their clinical relevance as biomarkers for diagnosis, treatment response and patient survival. In this study we observed that NFS1 gene expression levels were markedly higher in AML patients compared to controls and significantly decreased after 28 days of treatment. Conversely, GSDMD levels were initially low in AML patients but increased following treatment. Gene expression was carried out in whole peripheral blood rather than purified leukemic blasts; therefore, variations in blast percentage may be taken into consideration. These opposing expression patterns suggest differential regulation of NFS1 and GSDMD during disease course and treatment. Notably, in a previous study, NFS1 expression was found to be upregulated in AML cell line compared to normal BM mononuclear cells [44]. Our finding also come in agreement with previous reports showing that NFS1 plays a role in a variety of cancers. Alvarez et al. (2017) found that NFS1 was highly expressed in lung cancer tissues and cell lines. Further experiments demonstrated that NFS1 increased sulfur extraction from cysteine to form Fe–S clusters, which decreased iron release from cells and significantly mitigated hyperoxia-induced ferroptosis, whereas insufficient Fe–S clusters with decreased NFS1 leads to robust activation of the iron-starvation response which triggers ferroptosis [50]. In animal studies, KD of NFS1 led to a significant delay in tumor formation, indicating that NFS1 is positively selected during lung cancer development to counteract cell ferroptosis [50–52]. In breast cancer, the inhibition of NFS1 leads to sustained DNA damage and the death of tumor cells [53]. As for GSDMD, its expression was decreased in gastric cancer [54] and colorectal cancer [55] where low expression levels were associated with tumor progression and poor prognosis by impairing pyroptotic cell death and promoting tumor survival. In contrast, GSDMD expression is elevated in other cancers as: non-small-cell lung cancer [56] and pancreatic cancer [57] where it is linked to poor prognosis highlighting the complex role of GSDMD plays in cancer. We additionally assessed the diagnostic significance of NFS1 and GSDMD gene expression levels, using ROC analysis. In 100 AML patients, NFS1 individually revealed strong diagnostic value whereas GSDMD revealed moderate diagnostic value. A previous study underscored the significance of increased expression of NFS1 as a diagnostic marker of AML via public data set, TCGA, analysis [44]. Nevertheless, the high AUC values should be interpreted cautiously given the relatively small sample size and lack of external validation, which may result in optimism bias. Notably, the AML samples in this study represented various subtypes (M0-M4) according to the FAB classification, which is based on the origin and maturity of the leukemic cells [58]. Minimally differentiated AML (AML-M0) is a rare subtype known for its very poor prognosis and adverse outcomes [59]. In our study, NFS1 levels were significantly elevated, while GSDMD levels were notably decreased in AML-M0 patients compared to those with the more differentiated AML-M4 subtype. These results indicate that NFS1 and GSDMD expression are regulated differently in AML patients and may be influenced by the cell differentiation stage. Given that there is no established optimal therapy for AML-M0 patients, exploring alternative cancer immunotherapy options for this group may be warranted [59]. Furthermore, we wanted to examine the correlation between NFS1 and GSDMD in AML. Interestingly, we observed inverse correlation between NFS1 and GSDMD that further supports their antagonistic roles in AML. High NFS1 expression is associated with AML pathogenesis, whereas GSDMD appears to play a protective role, possibly through its involvement in pyroptosis. NFS1 exhibited an inverse correlation with platelet count, while GSDMD showed a direct correlation. This suggests that higher NFS1 levels may contribute to thrombocytopenia commonly observed in AML, whereas GSDMD might be involved in mechanisms that help restore platelet counts during treatment. NFS1 levels were directly correlated with the percentage of BM blasts, a key indicator of disease severity, whereas GSDMD levels were inversely correlated. This reinforces the idea that NFS1 promotes leukemic cell proliferation, while GSDMD may counteract this process. Remarkably, high NFS1 expression was predominantly found in patients with unfavorable cytogenetics and was similarly associated with a poorer clinicopathological stage in prostate cancer [60]. In contrast, GSDMD was downregulated in patients with unfavorable cytogenetics consistent with previous studies linking low GSDMD expression to poor prognosis in gastric cancer [54] and colorectal cancer [55]. This differential expression of NFS1 and GSDMD could be linked to the genetic instability and poor prognosis associated with unfavorable cytogenetic profiles in AML. Next, we investigated the effect of NFS1 and GSDMD expression on response to chemotherapy. Our study demonstrated that elevated NFS1 expression was associated with a decreased response to therapy, while high GSDMD expression was linked to better treatment outcomes. This suggests that NFS1 may be used as a biomarker for treatment resistance, whereas GSDMD could be indicative of therapeutic responsiveness. This can be explained in the light that NFS1 may influence cell sensitivity to ferroptosis by regulating Fe–S cluster synthesis in iron metabolism, leading to poor chemotherapy response as previously investigated in colon cancer patients with high NFS1 expression [35, 61]. For GSDMD, there has been ample of evidence suggesting that elevated levels of GSDMD are associated with a favorable response to antileukemia therapy. Notably, GSDMD-expressing leukemia cell lines were more sensitive to curcumin, employed as an anti-leukemia therapy, while leukemia cell lines without GSDMD were not sensitive to curcumin [43]. Moreover, in a study conducted by Wang et al. (2021), GSDMD expression clearly improved colorectal cancer patients response to chemotherapy [62]. Accordingly, GSDMD can be used as a biomarker to evaluate sensitivity to the leukemia therapy [43]. Interestingly, antileukemia therapies are strongly linked to both ferroptosis and pyroptosis. We found that induction therapy consisting to Anthracyclines and Cytarabine caused a significant down regulation of NFS1 and upregulation of GSDMD. Anthracyclines exert part of their cytotoxic effect through ferroptosis via increasing the pool of labile iron in the cell [63]. This together with our findings suggest that Anthracyclines have possibly enhanced ferroptosis in our study cohort via down regulating the expression of NFS1. Additionally, induction of ferroptosis enhances anticancer activity of Cytarabine and Doxorubicin in HL-60 cells, human leukemia cell line [64]. Doxorubicin, a common antileukemia therapy among the class Anthracyclines, was previously found to be a strong inducer of pyroptosis in previous studies [65, 66]. Moreover, in vitro experiments have demonstrated that anthracyclines can induce pyroptosis in breast cancer cells [67, 68]. The correlation of high NFS1 expression with lower survival rates, and high GSDMD expression with higher survival rates, highlights their potential as prognostic markers. Interestingly, their stability across multiple multivariate modeling strategies and the consistency of these findings across enter, backward, and forward models reduces the likelihood of model-dependent bias and reinforces their role as strong independent prognostic biomarkers. Thus, the persistent adverse impact of NFS1 may suggest a potential role in leukemic progression, while the strong protective association of GSDMD may reflect its involvement in pyroptotic pathways. Additionally, the time-dependent ROC analysis confirmed the excellent prognostic performance of the proposed Cox model, with AUC values consistently exceeding 0.90 across all evaluated time points. This high discriminative ability suggests that the model can reliably stratify patients according to short- and intermediate-term survival risk. The slight decline in AUC over time likely reflects the increasing complexity of long-term outcome prediction, as post-diagnosis variables such as treatment response, disease progression, and supportive care factors may increasingly influence survival beyond baseline characteristics. From previous results, patients with elevated NFS1 may have a more aggressive disease course, whereas those with higher GSDMD expression may experience better overall survival. These results align with survival analyses performed using TCGA and public datasets, revealing that NFS1 is positively associated with negative prognosis and cancer-related mortality in AML [44]. Also, colon cancer patients with high NFS1 expression showed poor survival outcomes [35, 61]. Regarding GSDMD, there has been discrepancy in whether high GSDMD correlates with better prognosis. In lung adenocarcinoma, osteosarcoma and glioma, high GSDMD is linked to worse prognosis [56, 69, 70]. Conversely, high GSDMD expression was associated with better prognosis in breast cancer. Patients with elevated GSDMD expression exhibited significantly longer overall survival than those with lower expression. Also, the expression of GSDMD was higher in non-metastatic compared to metastatic breast cancer patients [71]. This discrepancy can be explained by possibly discrepant subcellular localization of GSDMD in membrane, cytoplasm and nucleus that was not investigated in these studies. Wang et al. revealed that high cytoplasmic GSDMD expression was an independent positive prognostic factor and enhanced the effectiveness of chemotherapy in colorectal cancer that was associated with lower possibility of distant metastasis [62]. Yet nuclear GSDMD expression predicted deeper infiltration depth [62]. The role of GSDMD in cancer remains complex and varies depending on the context. Therefore, additional studies are required to elucidate the specific roles and mechanisms of GSDMD in different cancers. While our findings provide important insights, some limitations merit consideration. First, the results are based on gene expression (RNA) data, so validation at the protein level, as well as functional studies, will be necessary to further elucidate the biological and clinical relevance of NFS1 and GSDMD in AML. In addition, gene expression was assessed in whole peripheral blood rather than purified leukemic blasts; therefore, variations in blast percentage and immune cell composition may have influenced the measured expression levels. Future studies using isolated blast populations or BM mononuclear cells are warranted to validate the cellular origin of these markers and strengthen the interpretability of the findings. Furthermore, the absence of comprehensive genomic profiling precluded full classification according to contemporary systems such as the 2022 WHO classification, the International Consensus Classification (ICC), and the European LeukemiaNet (ELN 2022) risk stratification. Future studies incorporating comprehensive cytogenetic and molecular characterization in line with these frameworks are warranted to further validate NFS1 and GSDMD expression as independent prognostic markers. Collectively, these findings support an association between NFS1 and GSDMD expression patterns and clinical outcomes in AML, highlighting their potential relevance as prognostic and predictive biomarkers. However, given the observational nature of this study, future research should aim to clarify the exact mechanisms by which NFS1 promotes AML progression and how GSDMD-mediated pathways can be harnessed to improve patient outcomes. Conclusion In conclusion, our exploratory findings suggest that NFS1 and GSDMD may represent potential prognostic biomarkers for evaluating clinical outcomes in AML patients. However, external validation in larger, independent cohorts, including bone marrow-derived cells is necessary to confirm these associations. While inducing pyroptosis through GSDMD pathways and targeting NFS1 may present a possible avenue for overcoming apoptosis resistance, further mechanistic investigation is required to fully elucidate their roles in AML pathogenesis and to determine their true clinical utility as therapeutic targets.