Serum Cytokine Profiling Differentiates Underlying Diseases in Cytokine Storm Syndrome

Abstract

Objective

Cytokine storm syndrome (CSS), commonly associated with hemophagocytic lymphohistiocytosis (HLH), is a fatal hyperinflammatory syndrome. Differentiating the underlying diseases responsible for CSS is essential for timely therapeutic decisions. This study explored the clinical usefulness of serum cytokine profiling in distinguishing underlying diseases in patients with CSS.

Methods

Serum samples were collected from 143 adult and pediatric patients with CSS and 22 healthy controls. The cohort included patients with various diagnoses of primary and secondary HLH and Kawasaki disease (KD)‐like hyperinflammatory syndromes. Serum levels of 48 cytokines were analyzed in 97 patients using a bead‐based multiplex immunoassay (Luminex assay). Serum levels of interferon alpha (IFN‐α), interleukin‐18 (IL‐18), IL‐6, CXCL9, and soluble tumor necrosis factor receptor II (sTNF‐RII) were measured in 165 participants using enzyme‐linked immunosorbent assay (ELISA).

Results

Luminex assay categorized patients with CSS into five clusters based on serum cytokine patterns. ELISA revealed distinct cytokine patterns, wherein patients with histiocytic necrotizing lymphadenitis‐associated HLH and systemic lupus erythematosus‐associated macrophage activation syndrome (MAS) showed elevated IFN‐α; systemic juvenile idiopathic arthritis–associated and adult‐onset Still's disease–associated MAS, X‐linked inhibitor of apoptosis protein deficiency with HLH, and NLRC4‐associated autoinflammatory disorder exhibited higher IL‐18 levels. Additionally, KD shock syndrome had higher IL‐6 levels than the other groups. CXCL9 was significantly elevated in patients with virus‐associated HLH, familial HLH, malignant lymphoma‐associated HLH, and KD‐MAS. Multisystem inflammatory syndrome in children and toxic shock syndrome also showed moderate elevations of CXCL9 and IL‐6 levels.

Conclusion

Serum cytokine profiling effectively differentiates CSS subtypes, facilitating better diagnosis and personalized treatment strategies based on specific disease backgrounds.

Serum cytokine profiling was performed using Luminex (48 cytokines) and ELISA (5 cytokines). Assessment of key cytokines (IFN‐α, IL‐18, IL‐6, CXCL9) delineated five dominant inflammatory patterns: I, IFN‐α–dominant CSS; II, IL‐18–dominant CSS; III, IL‐6–dominant CSS; IV, IFN‐γ–dominant CSS; and V, IL‐6 & IFN‐γ intermediate CSS. This profiling approach highlights disease‐specific inflammatory signatures that support precision diagnosis in cytokine storm syndrome. CSS, cytokine storm syndrome; ELISA, enzyme‐linked immunosorbent assay; IFN, interferon; IL, interleukin; CXCL9, C‐X‐C motif chemokine ligand 9; HLH, hemophagocytic lymphohistiocytosis; MAS, macrophage activation syndrome; KDSS, Kawasaki disease shock syndrome; AID, autoinflammatory disease.

INTRODUCTION

Cytokine storm syndrome (CSS) is a severe inflammatory condition caused by excessive cytokine production. It represents one of the most crucial inflammatory events for patients and clinicians. Although its definition is vague and does not refer to a single disease, CSS is frequently associated with hemophagocytic lymphohistiocytosis (HLH) and macrophage activation syndrome (MAS), which complicates HLH in rheumatic diseases. ¹ Various factors contribute to CSS development, including genetic backgrounds that predispose individuals to inflammatory conditions, including autoinflammatory diseases, immunodeficiencies, malignant diseases, and infections. ¹ , ² These diseases cause uncontrolled activation and proliferation of immune cells, resulting in severe inflammation. Different aforementioned causes trigger CSS, and the term has come to be used as an umbrella concept encompassing these conditions. ³ Treatment plans must differ based on the underlying diagnosis despite the similarity in clinical and laboratory findings. Treatment with immunosuppressive or cytotoxic agents, such as etoposide or alemtuzumab, and rapid donor search are required to treat patients with primary HLH—a subset of monogenic inborn errors of immunity (IEIs) characterized by impaired cytotoxic lymphocyte function. Current standard treatments include glucocorticoids combined with etoposide and, in selected cases, emapalumab, an anti–interferon (IFN)‐γ monoclonal antibody, ⁴ , ⁵ whereas glucocorticoids, cyclosporine, or cytokine‐targeted therapies are appropriate for treating patients with secondary HLH or MAS. ¹ , ² , ⁶ Furthermore, CSS may be the initial manifestation of undiagnosed autoimmune or autoinflammatory diseases. Early identification of CSS etiologies is crucial for patient survival. Nevertheless, differentiating among these causes at an early stage is difficult, thus emphasizing the need for techniques to make an early diagnosis.

Historically, excessive production of inflammatory cytokines, including IFN‐γ and tumor necrosis factor alpha (TNF‐α), and their high serum levels have been closely associated with the pathogenesis of CSS. ² , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² Moreover, several studies have emphasized significant interleukin‐18 (IL‐18) hypercytokinemia in patients with MAS, which frequently complicates systemic juvenile idiopathic arthritis (sJIA) in children and adult‐onset Still's disease (AOSD) in adults. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ In addition, patients with IEIs, such as X‐linked inhibitor of apoptosis (XIAP) deficiency and NLRC4‐associated autoinflammatory disorder (NLRC4‐AID), frequently present with HLH and exhibit abnormally high serum IL‐18 levels. ¹⁷ , ¹⁸ Severe cases of COVID‐19 were associated with excessive IL‐6 production during the COVID‐19 pandemic, and treatment with IL‐6 inhibitors has been confirmed to be effective. ¹⁹ , ²⁰ Similarly, IL‐6 inhibitors have demonstrated efficacy in controlling the cytokine release syndrome (CRS) after chimeric antigen receptor–T cell therapy for hematologic malignancies, suggesting that IL‐6 may contribute to CRS pathology. ²¹ Moreover, congenital disorders characterized by excessive type I IFN signaling, termed “interferonopathies,” have been reported in the past few decades. These conditions exhibit MAS‐like symptoms, including fever, cytopenia, hepatosplenomegaly, and hyperferritinemia, and are gaining attention as a novel CSS subgroup. ²² , ²³ , ²⁴ , ²⁵ IFN‐α has also been reported as a central mediator in the pathogenesis of systemic lupus erythematosus (SLE) and other rheumatic diseases that may be complicated with HLH and MAS. ²⁶

Recent evidence supports the clinical efficacy of direct cytokine blockade in primary HLH. A pivotal trial demonstrated that emapalumab, an anti–IFN‐γ monoclonal antibody, effectively controlled hyperinflammation in pediatric patients with primary HLH. ⁵ A comprehensive review by Cron et al further emphasizes the central role of cytokine dysregulation in CSS and the potential of cytokine‐targeted therapies, including IL‐1, IL‐6, and IFN‐γ blockade, as part of precision medicine strategies. ¹

Given the clinical urgency of early diagnosis, recent studies have highlighted the utility of comprehensive cytokine profiling to distinguish among the underlying etiologies of CSSs. For example, Kessel et al systematically identified serum biomarkers, including S100A12 and IL‐18, that could distinguish sJIA‐MAS from primary or secondary HLH using multiplex immunoassays and validated them across independent cohorts. ²⁷ Carol et al further demonstrated, through a real‐time hyperferritinemia screening system, that markedly elevated IL‐18 and IL‐18/CXCL9 ratios were characteristic of MAS and differentiated them from other inflammatory causes including sepsis and immune dysregulation. ²⁸ Gallo et al developed a multiplex cytokine panel that successfully differentiated hyperinflammatory syndromes such as HLH, multisystem inflammatory syndrome in children (MIS‐C), and sepsis in pediatric intensive care unit settings. ²⁹ These findings underscore the potential of targeted cytokine profiling not only as a diagnostic adjunct but also as a tool to stratify hyperinflammatory syndromes by etiology and severity in real time.

We hypothesized that combining cytokine measurements would enable classifying CSS into specific groups and identifying their underlying diseases at earlier stages. In this study, we analyzed a wide range of biomarkers in the serum of patients with CSS using a 48‐multiplex bead‐based immunoassay, including major CSS‐related cytokines such as IFN‐γ, TNF, IL‐18, IL‐6, and IFN‐α, and investigated whether cytokine measurements could help in classifying patients with CSS with diverse underlying conditions. Moreover, to confirm the validity of that classification, we analyzed those key cytokines related to CSS by enzyme‐linked immunosorbent assay (ELISA) in a larger cohort of adult and pediatric patients with CSS. On the basis of those results, we propose an algorithm for a rapid and accurate classification of the underlying conditions.

MATERIALS AND METHODS

Patients and samples

We enrolled 15 adult and 128 pediatric patients with CSS and 22 healthy controls (HCs). Patients with the following conditions were included in this study: IEI‐associated CSS, including familial HLH (FHL), NLRC4‐AID, Griscelli syndrome type 2 (GS2), and XIAP deficiency‐complicated HLH (XIAP‐HLH); mevalonate kinase deficiency‐complicated MAS (MKD‐MAS); secondary CSS, including virus‐associated hemophagocytic syndrome (VAHS), such as Epstein–Barr virus (EBV)‐related HLH, malignant lymphoma‐associated HLH (LAHS), histiocytic necrotizing lymphadenitis complicating HLH (HNL‐HLH), sJIA‐MAS, AOSD‐MAS, SLE (SLE‐MAS), vacuoles, E1 enzyme, X‐linked, autoinflammatory, somatic syndrome (VEXAS) syndrome (VEXAS‐MAS), and Kawasaki disease (KD; KD‐MAS); and CSS with KD‐like hyperinflammatory syndromes, such as MIS‐C, a systemic inflammatory syndrome reported in children during the COVID‐19 pandemic, toxic shock syndrome (TSS), and KD shock syndrome (KDSS). Diagnoses of FHL, GS2, XIAP‐HLH, NLRC4‐AID, MKD‐MAS, VAHS, LAHS, and VEXAS‐MAS were made by clinicians based on the 2004 HLH criteria. ⁴ AOSD‐MAS, sJIA‐MAS, and KD‐MAS were diagnosed based on the 2016 American College of Rheumatology (ACR)/EULAR sJIA‐MAS criteria. ³⁰ SLE‐MAS and HNL‐HLH were diagnosed based on the 2009 and 2021 preliminary diagnostic criteria for SLE‐MAS, ³¹ , ³² and KDSS, MIS‐C, and TSS were diagnosed based on their respective diagnostic criteria. ³³ , ³⁴ , ³⁵ Serum was isolated from blood samples, aliquoted, and stored at −80°C until use. Serum samples were analyzed before initiating therapeutic agents, including glucocorticoids. This study was approved by the institutional review board of the Institute of Science Tokyo, and all specimens were used after obtaining informed consent from the patients. The committee's reference numbers are M2020‐062 and G2019‐004.

Luminex assay: a bead‐based multiplex immunoassay

Serum levels of cutaneous T cell–attracting chemokine, eotaxin, basic fibroblast growth factor, granulocyte colony‐stimulating factor, granulocyte‐macrophage colony‐stimulating factor, growth‐related oncogene 1α, hepatocyte growth factor, IFN‐α2, IFN‐γ, IL‐1α, IL‐1β, IL‐1RA, IL‐2, IL‐2Rα, IL‐3, IL‐4, IL‐5, IL‐6, IL‐7, IL‐8, IL‐9, IL‐10, IL‐12 (p40), IL‐12 (p70), IL‐13, IL‐15–18, IFN‐γ–induced protein 10 kDa, leukemia inhibitory factor, monocyte chemoattractant protein–1 (MCP‐1), MCP‐3, macrophage colony‐stimulating factor, macrophage migration inhibitory factor, mitogen‐inducible gene, macrophage inflammatory protein–1α (MIP‐1α), MIP‐1β, β‐nerve growth factor, platelet‐derived growth factor–BB, RANTES, stem cell factor, stem cell growth factor–β, stromal cell‐derived factor–1α, TNF, lymphotoxin‐α, TNF‐related apoptosis‐inducing ligand, and vascular endothelial growth factor–A (VEGF‐A) were measured in 88 patients with CSS and 9 HCs using Bio‐Plex Pro Human Cytokine Panel 48‐plex (Bio‐Rad Laboratories, Cat. #12007283) according to the manufacturer's instructions. According to the manufacturer's instructions, the panel shows <10% analyte cross‐reactivity, intra‐assay precision ≤15% coefficient of variation (CV), inter‐assay precision ≤25% CV, and spike‐and‐recovery accuracy of 70% to 130%. Data analyses were conducted using the Bio‐Plex Manager Software version 6.1 (Bio‐Rad Laboratories).

ELISA

Serum levels of IFN‐α, IL‐18, CXCL9, IL‐6, and soluble TNF receptor II (sTNF‐RII) were measured in all 165 participants with commercial ELISA kits (IFN‐α: PBL Assay Science; IL‐6, CXCL9, and sTNF‐RII: R&D Systems, Inc; and IL‐18: MBL) according to the manufacturer's instructions.

Analytical rationale

CXCL9 is induced exclusively by IFN‐γ, ³⁶ , ³⁷ whereas sTNF‐RII reflects TNF bioactivity ³⁸ , ³⁹ ; these assays therefore serve as surrogates for IFN‐γ and TNF activity, respectively.

Precision assessment

All samples were analyzed in duplicate, and the mean was used for downstream analyses. To assess intra‐assay variability, triplicate repeats were performed on a subset of 77 sera with sufficient volume; CVs were <20% for every cytokine (Supplementary Table S1).

Exploratory measurement of IL‐18 binding protein

IL‐18 binding protein (IL‐18BP) was quantified by ELISA (R&D Systems) in a subset of 77 patients for whom sufficient residual serum was available; triplicate measurements yielded CVs <20% (Supplementary Table S2). To further evaluate IL‐18 pathway activity, we also measured serum IL‐18BP and calculated the IL‐18/IL‐18BP ratio in this subset.

Statistical analysis

Statistical analysis was conducted using GraphPad Prism 10 (GraphPad) and R software for macOS version 4.3.3 (R Foundation). For principal component analysis (PCA), we used the “FactoMineR” package, and visualizations were created using “ggplot2.” For hierarchical clustering and heatmap generation, the “pheatmap” package was used, and data transformation was performed using “reshape2.” Excel files were imported using “readxl.” Fisher's exact test, unpaired t‐test with a two‐tailed P value, and the Mann–Whitney U test were conducted to compare variables within two groups. Comparisons among several groups were conducted using the Kruskal–Wallis rank test. One‐way analysis of variance with Dunn's multiple comparison test was performed to compare several groups. For group comparisons, disease categories with fewer than three patients were excluded from formal statistical testing. Clustering analysis was conducted by hierarchical clustering. P < 0.05 indicated statistical significance. The optimal cutoff values for receiver operating characteristic (ROC) curve analysis were identified using Youden's index, a method that maximizes sensitivity and specificity (sensitivity + specificity − 1).

Data availability

The datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.

RESULTS

Clinical characteristics of patients with CSS

Table 1 shows the clinical characteristics of patients in each group. We enrolled 15 adults and 128 pediatric patients, with an age range from 20 days to 82 years. One of each case of MKD‐MAS, GS2, and VEXAS‐MAS was included, whereas other patient groups involved at least three patients. Among the 13 patients registered as having VAHS, 11 were caused by EBV, 1 by human parvovirus B19, and 1 by adenovirus. Of the six FHL cases, three cases were each type 2 and type 3. Remarkably, two patients with FHL type 2 were diagnosed at age 22 years. The TSS group consisted of nine patients, including three caused by Staphylococcus infection and six caused by Yersinia pseudotuberculosis infection.

Table 1.

Clinical features of patients with CSS^*

	HNL‐HLH	SLE‐MAS	AOSD‐MAS	sJIA‐MAS	XIAP‐HLH	NLRC4‐AID	MKD‐MAS	KDSS	VEXAS‐MAS	FHL	GS2	VAHS	LAHS	KD‐MAS	MIS‐C	TSS
Number of patients, n	3	10	6	27	6	3	1	9	1	6	1	13	4	9	35	9
Male	1 (33.3)	1 (10.0)	1 (16.7)	13 (48.1)	1 (16.7)	1 (33.3)	0 (0.0)	5 (55.6)	1 (100.0)	5 (83.3)	0 (0.0)	5 (38.5)	1 (25.0)	5 (55.6)	19 (54.3)	3 (33.3)
Age, y	13 (10–14)	12.5 (10–29)	44.5 (20–82)	7 (0.33–14)	6 (0.08–19)	0.06 (0.04–4)	0.06	2 (0.17–7)	69	0.08 (0.08–22)	11	4 (1.33–22)	13 (11–15)	7 (0.83–14)	8 (0.75–15)	9 (1.16–15)
WBC, 103/μL	1.0 (0.93–1.8)	2.4 (1.8–8.0)	12.7 (5.1–21.1)	8.9 (2.3–52.9)	4.8 (3.2–8.7)	22.9 (19.3–23.8)	10.8	19.5 (6.7–57.4)	1.4	3.6 (1.1–4.7)	2.4	2.1 (0.5–14.2)	2.6 (0.6–3.2)	5.8 (2.1–11.4)	9.9 (3.3–20.3)	10.9 (5.2–22.2)
Hemoglobin, g/dL	13.8 (11.7–13.9)	11.8 (8.9–12.9)	11.6 (8–13.9)	11.5 (7.4–14.1)	11.2 (9.5–13.0)	10.0 (9.4–10.3)	12.1	9.9 (6.8–13)	9.9	7.9 (5.7–11.0)	10.8	11.4 (7.1–14.2)	10.6 (9.7–11.6)	11.1 (9.8–13.7)	12.2 (6.7–14.8)	13.4 (9.4–15.0)
Platelets, 104/μL	12.5 (12.3–13.0)	12.4 (7.1–26.8)	10.9 (8.4–36.9)	16.8 (7.1–32.7)	12.0 (5.2–55.1)	39.8 (12.1–41.2)	11.8	22.3 (7.1–44.8)	10.6	6.9 (1.8–12.3)	2.0	6.6 (2.3–24.4)	13.6 (5.8–35.1)	8.4 (5.1–20.5)	13.6 (2.8–69.9)	23.2 (6.9–32.1)
CRP, mg/dL	0.43 (0.12–0.50)	0.2 (0.03–7.99)	10.4 (0.69–19.9)	2.8 (0.03–20.1)	1.2 (0.14–18.7)	8.4 (5.2–9.0)	1.2	18.7 (4.4–34.0)	0.3	0.9 (0.3–21.0)	6.29	2.1 (0.3–31.2)	2.7 (2.2–3.9)	3.8 (0.47–15.6)	2.0 (9.9–23.3)	11.8 (0.3–22.2)
AST, IU/L	59 (26–320)	76 (22–363)	106 (63–1,082)	136 (36–1,591)	56 (28–214)	24 (16–37)	559	58 (17–498)	97	151 (46–698)	150	170 (43–888)	106 (26–241)	133 (38–2,218)	40 (12–269)	38 (16–159)
LDH, IU/L	728 (710–889)	675 (334–1,578)	439 (811–2,822)	872 (412–4,395)	655 (268–1,241)	341 (311–627)	557	301 (192–1,157)	810	487 (329–1,337)	382	1,168 (457–8,013)	657 (489–1,394)	731 (220–1,452)	306 (117–675)	278 (199–470)
Triglycerides, mg/dL	161 (106–194)	173 (144–520)	165 (98–214)	173 (86–811)	198 (88–405)	509 (120–2,319)	277	97 (62–128)	233	223 (118–342)	271	202 (103‐414)	135 (106–579)	156 (42–408)	121 (46–478)	147 (69–179)
Fibrinogen, mg/dL	336 (252–384)	322 (140–460)	271 (172–644)	258 (73–597)	183 (145–445)	260 (115–406)	223	518 (219–765)	112	199 (109–266)	197	194 (88–267)	253 (218–374)	295 (84–699)	489 (286–766)	539 (238–705)
FDP‐DD, ng/mL	2.7 (1.6–2.9)	3.5 (1.5–88.5)	10.9 (9.4–12.4)	6.2 (0.6–217.6)	29.6 (0.9–97.5)	21.9 (2.6–39.6)	1.1	3.8 (0.7–34.5)	43.5	10.4 (1.2–14.0)	14.2	22.1 (1.6–88.7)	5.1 (1.2–30.9)	7.5 (2.9–54.6)	3.6 (1.1–51.4)	3.4 (0.8–7.1)
sIL‐2R, IU/mL	1,015 (909–1,120)	1,543 (852–2,683)	3,158 (2,090–3,379)	2,214 (933–5,660)	1,688 (756–3,601)	2,108 (1,355–2,861)	4,188	7,099 (2,852–11,345)	1,923	13,618 (4,958–22,699)	8,821	9,886 (1,157–22,975)	1,740 (1,300–2,440)	4,385 (2,052–7,418)	2,423 (800–9,877)	4,750 (2,639–5,467)
Ferritin, ng/mL	794 (378–7,794)	1,519 (599–8,074)	21,614 (801–202,254)	5,140 (963–173,921)	3,468 (337–16,315)	1,025 (80–4,317)	1,591	641 (128–3,497)	5,148	2,151 (239–12,485)	1,282	3,898 (467–39,387)	1,572 (275–7,425)	1,429 (688–6,472)	555 (62–29,058)	251 (76–1,107)
HScore, points (0–337)	159 (122–185)	153 (90–204)	198 (141–235)	198 (135–289)	168 (95–221)	158 (87–235)	106	114 (87–141)	250	225 (106–309)	204	225 (106–309)	179 (106–309)	146 (106–254)	112 (68–170)	79 (49–142)

^*Data are n (%) or median (range) unless specified. AOSD, adult‐onset Still's disease; AST, aspartate aminotransferase; CRP, C‐reactive protein; CSS, cytokine storm syndrome; FDP‐DD, fibrin/fibrinogen degradation products‐D dimer; FHL, familial HLH; GS2, Griscelli syndrome type 2; HLH, hemophagocytic lymphohistiocytosis; HNL, histiocytic necrotizing lymphadenitis; KD, Kawasaki disease; KDSS, KD shock syndrome; LAHS, lymphoma‐associated HLH; LDH, lactate dehydrogenase; MAS, macrophage activation syndrome; MIS‐C, multisystem inflammatory syndrome in children; MKD, mevalonate kinase deficiency; NLRC4‐AID, NLRC4‐associated autoinflammatory disorder; sIL‐2R, soluble interleukin receptor type 2; sJIA, systemic juvenile idiopathic arthritis; SLE, systemic lupus erythematosus; TSS, toxic shock syndrome; VAHS, virus‐associated hemophagocytic syndrome; VEXAS, vacuoles, E1 enzyme, X‐linked, autoinflammatory, somatic syndrome; WBC, white blood cell counts; XIAP, X‐linked inhibitor of apoptosis protein.

Patients with CSS and HCs are categorized into six groups based on the levels of 48 cytokines

We measured the levels of 48 cytokines and chemokines in the serum of 88 patients with CSS and 9 HCs using the Luminex assay to evaluate the use of multiple cytokine levels in classifying patients with CSS (Supplementary Table S3). Based on hierarchical clustering analysis, patients with CSS and HCs were broadly categorized into several groups (Figure 1). Specifically, one group included most patients with sJIA‐MAS, AOSD‐MAS, XIAP‐HLH, and NLRC4‐AID (clusters in rows 1–16 and 75–81); another group involved most patients with KDSS (rows 27–30); another cluster (rows 37–61) contained most patients with MIS‐C, TSS, and KD‐MAS; and a final cluster (rows 63–71) consisted of several patients with SLE‐MAS and HNL‐HLH. Nevertheless, specific clusters were not established by patients with VAHS, LAHS, or FHL. PCA of these 97 cases revealed some grouping, with an obscure differentiation (Supplementary Figure S1). Hence, patients with CSS having 15 different clinical diagnoses could be categorized into several patterns based on comprehensive serum cytokine profiles.

Figure 1.

Hierarchical cluster analysis with serum levels of 48 cytokines and chemokines in patients with CSS. Serum biomarker profiles of 97 patients with CSS. Data comprising 48 analytes are presented as a heat map after unsupervised hierarchical clustering of biomarker expression profiles according to correlation distance. Colors indicate column Z score. AOSD, Adult onset Still's disease; FHL, familial hemophagocytic lymphohistiocytosis; GS2, Griscelli syndrome type 2; HC, healthy controls; HLH, hemophagocytic lymphohistiocytosis; HNL, histiocytic necrotizing lymphadenitis; KDSS, Kawasaki disease shock syndrome; LAHS, lymphoma associated HLH; MAS, macrophage activation syndrome; MIS‐C, multisystem inflammatory syndrome in children; MKD, Mevalonate kinase deficiency; NLRC4‐AID, NLR‐family CARD domain‐containing protein 4 associated autoinflammatory disorder; SLE, systemic lupus erythematosus; TSS, toxic shock syndrome; VAHS, virus‐associated hemophagocytic syndrome; XIAP, X‐linked inhibitor of apoptosis.

Differences among assays in the measurement results of five CSS representative cytokines

Mysteriously, no marked increases in IFN‐α2 levels were detected in patients with SLE or HNL, in whom IFN‐α is generally important in disease pathogenesis, compared with those in other CSS groups (Figure 1). Considering that cytokine levels differ depending on the measurement method, we also quantified the levels of CXCL9, sTNF‐RII, IL‐6, IL‐18, and IFN‐α by ELISA in the same serum samples of 97 patients analyzed using the Luminex assay and then compared those results. Serum IFN‐α levels measured using an ELISA kit that detects all subtypes exhibited a significant increase in patients with HNL‐HLH and those with SLE‐MAS; however, there was no significant increase in IFN‐α2 levels measured using the Luminex assay in patients with HNL‐HLH and those with SLE‐MAS (Figure 2A). Significant correlations were confirmed between the Luminex assay and ELISA measurements for the other cytokines, but there was little correlation between IFN‐α and IFN‐α2 levels (Figure 2B, 2C). In the Luminex assay kit we used in the present study, only IFN‐α2 could be quantified of the 13 IFN‐α subtypes. Analyzing only a single subtype may not be sufficient to evaluate the upregulation of the IFN‐α pathway in patients with CSS, and quantifying IFN‐α in all subtypes may be required as in the ELISA kit we used. Therefore, the results of the Luminex assay for IFN‐α2 were confirmed inaccurate as representative values for IFN‐α signaling in patients with CSS.

Figure 2.

Differences and correlations of serum cytokine levels quantified by ELISA and Luminex assay. (A) Comparison of serum IFN‐α2 levels determined by the Luminex assay and serum IFN‐α levels determined by ELISA. (B, C) Correlations of serum cytokine and chemokine levels. The X‐axis shows cytokine levels in serum quantified by the Luminex assay, and the Y‐axis shows cytokine levels in serum quantified by ELISA. The green dotted line indicates that the value of the ELISA/Luminex assay ratio is 10, and the blue line indicates that the ratio is 1. Statistically significant differences among each patient group are shown as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. CSS, cytokine storm syndrome; ELISA, enzyme‐linked immunosorbent assay; HC, healthy control; HNL‐HLH, histiocytic necrotizing lymphadenitis complicating hemophagocytic lymphohistiocytosis; IFN‐α, interferon alpha; IL, interleukin; MIG, monokine induced by IFN‐γ; SLE‐MAS, systemic lupus erythematosus–macrophage activation syndrome; sTNF‐RII, soluble tumor necrosis factor receptor II; TNF, tumor necrosis factor.

Patterns of elevation in serum levels of five representative cytokines differ among groups of patients with CSS

Due to the aforementioned result, we analyzed several key cytokines that are believed crucial for CSS pathogenesis in a larger cohort of patients to more accurately and easily determine the classification patterns. We quantified IFN‐α, IL‐6, IL‐18, CXCL9, and sTNF‐RII levels in the serum of 143 patients with CSS and 22 HCs using ELISA (Figure 3A, Supplementary Table S4). Patients with HNL‐HLH and those with SLE‐MAS showed significantly higher serum IFN‐α levels than other groups of patients with CSS. Patients with AOSD‐MAS and those with sJIA‐MAS exhibited significantly higher IL‐18 levels than other groups of patients with CSS. Furthermore, serum IL‐18 levels in patients with XIAP‐HLH and NLRC4‐AID increased compared with those in the other groups, although this may be due to the small sample size, with no statistically significant difference. Serum IL‐18BP levels showed minimal variation across CSS groups, with no significant differences except in sJIA‐MAS and VAHS. The IL‐18/IL‐18BP ratio was notably higher in sJIA‐MAS and AOSD‐MAS, consistent with the results of IL‐18 levels (Supplementary Figure S2 and Supplementary Table S2). Patients with KDSS showed significantly increased serum IL‐6 levels compared with other patients with CSS. Serum CXCL9 levels were markedly elevated in several patients with CSS compared with HCs, with particularly high levels in patients with FHL, VAHS, LAHS, and KD‐MAS. Moreover, sTNF‐RII levels increased in patients with CSS compared with HCs, but with no distinctive differences among CSS subgroups (Figure 3A).

Figure 3.

Comparison of serum cytokine levels in patients with CSS. (A) Serum cytokine levels of patients with CSS. Bars represent median levels and interquartile ranges. Statistically significant differences among each patient group are shown as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. All units of cytokines are pg/mL. (B) Serum cytokine profiles in each patient group. The blue hexagons show median levels of serum cytokines in patients with each background disease, and the white hexagons show those in HCs. All units of cytokines are pg/mL. GS2, Griscelli syndrome type 2; HNL, histiocytic necrotizing lymphadenitis; KDSS, Kawasaki disease shock syndrome; LAHS, lymphoma associated HLH; MKD, Mevalonate kinase deficiency; NLRC4‐AID, NLR‐family CARD domain‐containing protein 4 associated autoinflammatory disorder; TSS, toxic shock syndrome; VAHS, virus‐associated hemophagocytic syndrome; XIAP, X‐linked inhibitor of apoptosis.

Classification of patients with CSS into five groups based on dominant cytokines

PCA conducted using the aforementioned five cytokines enabled clearer group classification than the 48‐cytokine analysis conducted using the Luminex assay. PCA revealed the following groups: HNL‐HLH and SLE‐MAS (IFN‐α–dominant group); AOSD‐MAS, sJIA‐MAS, XIAP‐HLH, and NLRC4‐AID (IL‐18–dominant group); KDSS (IL‐6–dominant group); VAHS, LAHS, FHL, and KD‐MAS (IFN‐γ–dominant group); MIS‐C and TSS (IL‐6– and IFN‐γ–intermediate group); and HCs (Supplementary Figure S3). Visual representation of these cytokine median values with radar charts enabled a clear differentiation of CSS‐related underlying conditions (Figure 3B). Interestingly, the PCA using a limited panel of five cytokines (IL‐18, CXCL9, IFN‐α, IL‐6, and sTNF‐RII) demonstrated even tighter separation between CSS patients and HCs than that achieved with the full 48‐cytokine dataset (Supplementary Figure S1 and S3). This suggests that a small focused cytokine set may be sufficient for practical disease stratification.

An analysis of variable contributions to principle component 1 (Supplementary Figure S4) of the 48‐cytokine analysis by the Luminex assay revealed that several top‐contributing cytokines were not strongly associated with canonical CSS pathways (eg, CXCL9, IL‐18, IL‐6). This may indicate that inclusion of less relevant cytokines introduced additional variance, thereby obscuring group differentiation. In contrast, PCA based on a restricted set of five cytokines more clearly delineated CSS from controls (Supplementary Figure S3), supporting the utility of a focused panel for practical CSS stratification.

Stepwise algorithm for the differential diagnosis of CSS

A stepwise ROC curve analysis was confirmed effective for differential diagnosis (Figure 4). Patients with SLE‐MAS and HNL‐HLH showed a significant increase in serum IFN‐α levels compared with other groups, with a cutoff of 18.0 pg/mL, area under the ROC curve (AUC) of 0.9964, sensitivity of 1.00, and specificity of 0.9764. Moreover, among the remaining groups, significant increases in serum IL‐18 levels were observed in AOSD‐MAS, sJIA‐MAS, XIAP‐HLH, and NLRC4‐AID groups compared with that in the other groups, with a cutoff of 13,420 pg/mL, AUC of 0.9989, sensitivity of 1.00, and specificity of 0.9882. In addition, the abnormal increase in serum IL‐6 levels was detected only in the KDSS group among the remaining groups and was useful for distinguishing the KDSS group from the other groups, with a cutoff of 185.5 pg/mL, AUC of 0.9770, sensitivity of 1.00, and specificity of 0.8816. Finally, serum CXCL9 levels were significantly higher in patients with FHL, VAHS, LAHS, and KD‐MAS than in patients with MIS‐C and TSS, with a cutoff of 10,337 pg/mL, AUC of 0.9212, sensitivity of 0.9688, and specificity of 0.8636. Patients with MKD‐MAS, VEXAS‐MAS, and GS2 were represented by single cases and were excluded from statistical comparisons.

Figure 4.

Stepwise ROC curve analysis and classification algorithm for patients with CSS. For the 165 cases, the high group was defined as “IFN‐α–dominant CSS” with a cutoff of 18.0 pg/mL for IFN‐α, and for the remaining groups with an IFN‐α level ≤18.0 pg/mL, the high group was defined as “IL‐18–dominant CSS” with a cutoff of 13,420 pg/mL for IL‐18. For the remaining groups with an IL‐18 level ≤13,420 pg/mL, the group with an IL‐6 level >185.5 pg/mL was defined as “IL‐6–dominant CSS,” and the remaining groups were defined as “IFN‐γ–dominant CSS” with a CXCL9 level of 10,337 pg/mL as the cutoff. The group with levels <10,337 pg/mL was classified as “IL‐6– and IFN‐γ–intermediate CSS.” HNL, histiocytic necrotizing lymphadenitis; KDSS, Kawasaki disease shock syndrome; LAHS, lymphoma associated HLH; NLRC4‐AID, NLR‐family CARD domain‐containing protein 4 associated autoinflammatory disorder; TSS, toxic shock syndrome; VAHS, virus‐associated hemophagocytic syndrome; XIAP, X‐linked inhibitor of apoptosis.

We defined these groups as “IFN‐α–dominant CSS,” “IL‐18–dominant CSS,” “IL‐6–dominant CSS,” “IFN‐γ–dominant CSS,” and “IL‐6– and IFN‐γ–intermediate CSS,” respectively. We summarized these data and proposed a classification algorithm for CSS using multiple serum cytokine levels (Figure 5).

Figure 5.

Classification algorithm for CSS using multiple serum cytokine levels. A stepwise classification algorithm for the background diseases of CSS using five cytokines. CSS, cytokine storm syndrome; IFN, interferon; IL, interleukin.

DISCUSSION

This study demonstrated that clinically similar patients with CSS with various background diseases can be categorized into several groups through a comprehensive analysis of serum cytokines despite their extremely similar clinical characteristics. We also proposed an algorithm that classifies CSS with high accuracy by evaluating several important CSS cytokines and chemokines in combination. Assuming the background disease early on will significantly contribute toward making decisions regarding subsequent examinations and treatments.

Inflammatory cytokines, such as IFN‐γ and TNF, have long been recognized to be deeply involved in the pathology of CSS, and we detected a high increase in CXCL9 levels, which reflects IFN‐γ production, ³⁶ , ³⁷ and sTNF‐RII levels, which reflects TNF production, ³⁸ , ³⁹ in most patients in this study. Conversely, a marked increase in IFN‐α levels was uniquely observed in the CSS group in patients with abnormal autoimmune diseases such as HNL‐HLH and SLE‐MAS. This finding suggests the overactivation of plasmacytoid dendritic cells and some monocytes, which are the primary sources of IFN‐α production. ⁴⁰ , ⁴¹ We also detected a highly distinctive increase in IL‐18 levels in “IL‐18–dominant CSS,” including in patients with AOSD‐MAS, sJIA‐MAS, XIAP‐HLH, and NLRC4‐AID. Patients with these diseases are known to show extremely elevated serum IL‐18 levels, and we again confirmed that these levels are distinctively elevated compared with those in other CSS groups, which is useful in diagnosis and classification. In these diseases, excessive production of IL‐18 may be caused by hyperactivation of the inflammasome and the innate immune system; hence, future therapies targeting the inflammasome are also anticipated. ⁴² , ⁴³ Moreover, the patient with MKD‐MAS demonstrated a highly elevated serum IL‐18 level, although milder than those in patients with NLRC4‐AID and XIAP‐HLH. The pathogenesis of MKD is closely related to the activation of the inflammasome, and it is also an IEI that sometimes complicates MAS. ⁴⁴ There have been no studies on the serum cytokines of patients with MKD‐MAS, and our results must be confirmed with a larger cohort study.

Furthermore, a marked increase in serum IL‐6 levels was observed in the “IL‐6–dominant CSS” group, including KDSS, compared with that in the other CSS groups, and this was a pattern resembling the extreme type that directly exacerbated the inflammatory pathology of KD. In addition, an excessive inflammatory pathology centered on IL‐6 may be important in KDSS. Moreover, a particularly prominent increase in CXCL9 levels was detected in the “IFN‐γ–dominant CSS” group, such as in patients with FHL, VAHS, LAHS, and KD‐MAS, compared with that in the other CSS groups. The primary pathology and cause of these diseases is generally the runaway of excessive cellular immunity, centering on cytotoxic T cells, and our results are consistent with previous reports. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Furthermore, toxins produced by pathogens such as Staphylococcus, and Y. pseudotuberculosis act as superantigens, causing excessive activation of T cells in “IL‐6– and IFN‐γ–intermediate CSS,” such as in patients with TSS. Although MIS‐C shares clinical features with toxin‐mediated syndromes such as TSS, current evidence suggests that it is primarily driven by postinfectious immune dysregulation and may involve superantigen‐like responses rather than classical bacterial toxins. ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ Therefore, this result comprehensively presented an intermediate pattern involving therapy for the increase in IL‐6 levels associated with infection and the elevation in IFN‐γ levels due to T cell activation.

There has been some research on developing biomarkers for pediatric patients with primary and secondary HLH in the past decade. For instance, Kessel et al conducted a comprehensive analysis of serum cytokines and chemokines using the Luminex assay and reported IL‐18, CXCL9, and S100A12 as useful markers for distinguishing primary HLH, including FHL, and sJIA‐MAS. ²⁷ More recently, Carol et al used a proximity extension assay (PEA) to evaluate serum protein levels and reported IL‐18, ANGPT1, and VEGF receptor 2 as promising biomarkers for early identification of hyperferritinemic conditions, including MAS and sepsis‐related HLH. ²⁸ These studies, although valuable, focused primarily on pediatric populations and a relatively limited disease spectrum, mainly involving sJIA‐MAS and infection‐associated CSS. In a broader clinical context, Gallo et al reported the real‐world implementation of a rapid multiplex cytokine panel in pediatric practice. Their large‐scale analysis demonstrated that cytokine profiles reflected the disease severity and enabled differentiation between FHL or CRS and bacterial sepsis using specific ratios, such as [IFN‐γ + IL‐10]/[IL‐6 + IL‐8]. However, their in‐house panel did not include key cytokines, such as IL‐18 or CXCL9, which are essential for evaluating IL‐18–dominant or IFN‐γ–dominant CSS subtypes. ²⁹

Our study included both pediatric and adult patients and encompassed a broader spectrum of CSS etiologies—including rare conditions such as HNL‐HLH, SLE‐MAS, KD‐MAS, and IEI‐associated CSS (eg, XIAP‐HLH and NLRC4‐AID), as well as adult‐onset CSS like AOSD‐MAS. Furthermore, we assessed a wider array of cytokines, including IL‐18, CXCL9, IFN‐α, and sTNF‐RII, allowing more nuanced classification of CSS phenotypes. The stepwise classification algorithm according to the combination of multiple cytokine levels developed from our results provides a more accurate and rapid method for determining the underlying conditions of CSS compared with that using individual biomarkers. Although the stepwise cytokine‐based approach presented here offers a potential framework for CSS classification, its clinical utility remains to be validated in larger multicenter and prospective cohorts. Future studies will be necessary to assess its generalizability across age groups and clinical settings.

We also confirmed the differences in values between the two representative cytokine quantification methods, ELISA and the Luminex assay (Figure 2C). In particular, for IL‐18, the levels determined by ELISA were approximately 4.7 times higher than those determined by the Luminex assay on median (Figure 2C). Carol et al ²⁸ also confirmed the correlation and differences between the Luminex assay and PEA and indicated that there are unique limitations for each measurement method, such as the “hook effect” in PEA. Therefore, to apply serum cytokine profiling to clinical practice in the future, it is necessary to establish a universally standard cytokine measurement method.

In addition to its analytical performance, the five‐cytokine panel offers notable logistical advantages. These cytokines can be quantified by standard ELISA, which provides a broader dynamic range, requires less specialized equipment, and is considerably more accessible and cost‐effective than the Luminex assay. These features make the five‐cytokine panel a promising candidate for translation into routine clinical use.

Our study has several limitations. First, we included patients with CSS with diverse underlying diseases; however, the sample size is small, and certain conditions that can cause CSS, such as metabolic diseases (eg, HLH associated with lysinuric protein intolerance), were not included, although it may be impossible to enroll all patients with CSS. Furthermore, prospective validation of the predefined thresholds in independent, disease‐matched cohorts remains essential before clinical implementation. Second, there were limitations in terms of ethnic diversity because most of the analyzed patients were Japanese. Third, the retrospective design and limited sample size for several subgroups preclude the immediate clinical application of the proposed classification algorithm. Prospective validation in larger and more diverse cohorts is needed to confirm its utility in practice. Although data from rare disorders such as GS2, MKD‐MAS and VEXAS‐MAS were included in the analyses, these cases were not used to define dominant cytokine‐based groups owing to the small sample size. Their cytokine profiles are shown for descriptive purposes only and should be interpreted with caution.

In conclusion, patients with CSS can be categorized into several groups according to their underlying diseases. Serum cytokine profiling was confirmed as an effective method for their differentiation.

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Shimizu confirms that all authors have provided the final approval of the version to be published and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

ROLE OF THE STUDY SPONSOR

Medical and Biological Laboratories Co, Ltd had no role in the study design or in the collection, analysis, or interpretation of the data, the writing of the manuscript, or the decision to submit the manuscript for publication. Publication of this article was not contingent upon approval by Medical and Biological Laboratories Co, Ltd.

Supporting information

Appendix S1: Supplementary Information

Disclosure form.