Role of CSF interleukin-6 in the prognosis of patients with Leptomeningeal metastases of glioblastoma

Abstract

Background

To explore the correlation between cerebrospinal fluid (CSF) and peripheral blood cytokines and survival in patients with Leptomeningeal metastases (LM) of glioblastoma.

Methods

The study retrospectively included 35 patients with glioblastoma LM diagnosed from September 2022 to August 2024 at Beijing Tiantan Hospital. CSF and peripheral blood cytokines were collected from patients and associated with other characteristics with overall survival (OS) using univariate and multivariate analyses. The optimal cutoff values for cytokines were derived using the surv_cutpoint function in the R software.

Results

The median OS for LM patients was 11.8 months (95% confidence intervals (CI) 6.1–17.5 months). After univariate and multivariate analyses, CSF IL-6 (95% CI 0.072–0.683; P = 0.009) and Karnofsky Physical Status (KPS) (95% CI, 0.136–0.876; P = 0.025) remained significantly associated with OS. We further combined CSF interleukin-6 (IL-6) with KPS to construct nomogram to predict survival and demonstrated good predictive performance.

Conclusion

Both CSF IL-6 and KPS are meaningful prognostic biomarkers for patients with LM from glioblastoma.

Introduction

Glioma is the most common primary malignant tumors in the central nervous system with a poor prognosis, and despite standard treatment consisting of maximal tumor resection, concurrent radiotherapy and chemotherapy, treatment efficacy remains poor and the prognosis for patients with gliomas remains unsatisfactory [1, 2]. The highly aggressive nature of gliomas is a major cause of recurrence and poor prognosis [3, 4]. Leptomeningeal metastasis (LM), originally considered a rare complication, arises from the dissemination of glioma cells into the leptomeninges and subarachnoid space. It was first described in 1931 as spinal cord metastasis originating from supratentorial glioblastomas (GBM) [5, 6]. However, in recent years, LM has gained increasing attention. One study found LM in up to 21% of autopsy cases, while another reported spinal subarachnoid seeding in approximately 25% of intracranial GBM patients on autopsy [5, 7]. Additionally, several studies have reported varying incidences of LM [8–10]. For instance, Noh et al. found an LM incidence of 23.4% in GBM patients [10]. This rising incidence may be attributed to improved survival due to emerging therapeutic strategies, such as molecular targeted therapy, immunotherapy, and tumor-treating fields, as well as advances in imaging technology [8, 9, 11]. LM, as a severe complication, is associated with an even worse prognosis than parenchymal tumor progression. Therefore, identifying reliable biomarkers to assess LM prognosis is critical for stratifying high-risk patients.

The growth and invasion of tumor cells often trigger a robust release of cytokines and chemokines into the tumor microenvironment, a hallmark of cancer progression [12]. LM, being a metastatic condition involving the leptomeninges and cerebrospinal fluid (CSF), may also compromise the blood-brain barrier (BBB), leading to elevated levels of cytokines in the CSF and peripheral blood. Detecting these biomarkers is thus crucial for understanding the tumor microenvironment and holds potential as prognostic indicators for LM. Several studies have demonstrated the prognostic value of cytokines in various cancers. For instance, Dan Høgdall et al. reported that serum IL-6 is a prognostic biomarker in biliary tract cancer and that IL-6R may be a potential therapeutic target [13]. Patrick L. Wagner et al. found elevated IL-6 levels in the peritoneal fluid of patients with peritoneal carcinomatosis across multiple tumor types [14]. Furthermore, studies in non-small cell lung cancer and ovarian cancer have shown that high systemic IL-6 levels are associated with poor prognosis [15–17]. In glioma, increasing IL-6 concentrations have also been linked to a higher risk of unfavorable outcomes [18]. However, the prognostic significance of IL-6 in the CSF and peripheral blood of LM patients remains unclear. Therefore, in this study, we collected cytokine data from the CSF and peripheral blood of patients with LM and evaluated their prognostic roles. This investigation may provide valuable insights for clinical decision-making and risk stratification.

Materials and methods

Patients

This study retrospectively analyzed 35 adult patients (> 18 years at initial diagnosis) with pathologically confirmed glioblastoma who developed LM. All patients were treated at Beijing Tiantan Hospital from September 2022 to August 2024. The study protocol was approved by the institutional review board. Inclusion criteria required a confirmed LM diagnosis via magnetic resonance imaging (MRI) or CSF analysis, along with complete records of blood and CSF cytokine levels prior to LM treatment. Exclusion criteria included primary spinal cord gliomas, prior steroid use before LM diagnosis, or a history of chronic inflammatory/autoimmune disorders. LM was defined by radiological evidence of leptomeningeal dissemination (brain or spine MRI), subependymal/ependymal enhancement on brain MRI, or positive CSF cytology verified by pathology reports.

The therapeutic regimen consisted of intrathecal methotrexate (MTX) combined with systemic chemotherapy. The selection of systemic chemotherapy regimens was determined based on the NCCN Guidelines for Central Nervous System Cancers and the Chinese Guidelines for the Diagnosis and Treatment of Glioma. Systemic options included temozolomide (TMZ) alone, an etoposide (VP-16) plus carboplatin (EC) combination, or a TMZ plus cisplatin (TP) combination. Treatment cycles were repeated every 28 days. Intrathecal MTX (10 mg) was administered weekly, totaling 2–3 doses per cycle. Systemic therapy was tailored as follows: TMZ Regimen: Used as the first-line option for patients who had not yet completed 6 cycles of systemic TMZ. Dosage: 150–200 mg/m²/day for 5 days per cycle. EC Regimen: Prescribed for patients who experienced TMZ failure within 6 months. Dosage: Carboplatin (AUC 5, day 1) and VP-16 (100 mg/m²/day for 3 days) per cycle. TP Regimen: Reserved for patients with TMZ failure occurring after more than 6 months. Dosage: Cisplatin (30 mg/m²/day for 3 days) and TMZ (150–200 mg/m²/day for 5 days) per cycle. Additionally, patients with severe cerebral edema refractory to standard management received Bevacizumab (BEV) at 5 mg/kg once per cycle. Dosing schedules were adjusted according to standard chemotherapy guidelines. Disease status was reassessed every two months using MRI, with or without CSF morphology. Patients showing a response continued treatment for up to 8 cycles. In cases of ineffective treatment, the regimen was switched. Therapy continued until disease progression, treatment withdrawal, or death.

Lumbar punctures and cytokine measurement

All lumbar punctures were used for LM diagnosis, and prior to treatment for LM CSF and peripheral blood were extracted from patients for further analysis. We used a Meso Scale Discovery (MSD) to quantify IL-6, IL-8, and tumor necrosis factor-α (TNF-α) in CSF and peripheral blood. The corresponding detection limits in our analysis were IL-6: 2 pg/mL; IL-8: 5 pg/mL; and TNF-α: 4 pg/mL. If the results were below this detection limit, the values were treated as the detection limit in the statistical analysis.

Data collection

We collected patient age at initial glioma diagnosis, gender, date of initial glioma diagnosis, date of LM diagnosis, tumor location, extent of surgical resection, ventricular-peritoneal shunt, Ommaya reservoir implant, MRI results, CSF results, Karnofsky Physical Status (KPS) at the time of LM diagnosis, treatment before and after LM diagnosis, and date of death or last follow-up. All cases were diagnosed according to the 2021 WHO classification. A 1p19q co-deletion was required for the diagnosis of oligodendroglioma. Patients were assessed for mutational status of isocitrate dehydrogenase (IDH) by immunohistochemistry, and IDH mutations were considered to be present when IDH1 or 2 was mutated in next-generation sequencing [19]. OS was defined as the time from LM diagnosis to death or last follow-up.

Statistical analysis

We used the Kaplan-Meier method to determine survival rates, and the significance was determined using the log-rank test. The optimal cutoff values of cytokines associated with OS were determined according to the surv_cutpoint function. We used univariate and multivariate Cox regression to analyze the predictive factors of survival. Prognostic factors with p < 0.05 in univariate analysis were explored in multivariate analysis. Significance was defined as variables with a two-sided P value < 0.05. All analyses were performed using R (Version 4.4.3), SPSS (Version 26.0, IBM), and GraphPad Prism 8.0 software.

Results

Patients’ characteristics

There were 35 patients incorporated in this study with a median age of 41 years (ranging from 18 to 72), with a median follow-up period of 15.9 (interquartile range [IQR] 11.3–18.3) months. Most patients were male (60.0%). There were 30 (85.7%) patients with tumors in supratentorial and others in infratentorial. These results are summarized in Table 1.

Table 1.

Patient characteristics

Characteristic	Number of patients (%)
Gender
Male	21 (60.0%)
Female	14 (40.0%)
Age at initial glioma diagnosis (years)
< 41 years	14 (40.0%)
≥ 41 years	21 (60.0%)
Location
Supratentorial	30 (85.7%)
Infratentorial	5 (14.3%)
Surgery
Gross total resection	9 (25.7%)
subtotal resection	6 (17.1%)
biopsy	3 (8.6%)
None	17 (48.6%)
Treatment before LM
Radiation therapy	2 (5.7%)
Concurrent chemoradiotherapy	32 (91.4%)
Treatment field	1 (2.9%)
None treatment	3 (8.6%)
Ventricular-Peritoneal shunt
Yes	7 (20.0%)
No	28 (80.0%)
LM diagnosis
Brain MRI	27 (77.1%)
Spine MRI	7 (20.0%)
CSF cytology	13 (37.2%)
Interval between initial glioma and LM diagnoses (months, median)	15.7 (8.4-23.0)
KPS at LM diagnosis
< 70	10 (28.6%)
≥ 70	25 (71.4%)
Treatment after LM
MTX	1 (2.9%)
MTX+TMZ	3 (8.6%)
MTX+VP-16+DDP	22 (62.9%)
MTX+TMZ+DDP	9 (25.7%)
Bevacizumab	20 (57.1%)
Cycles of LM treatment
< 5	25 (71.4%)
≥ 5	10 (28.6%)
Ommaya reservoir implant
Yes	7 (20.0%)
No	28 (80.0%)
The content of total protein in the CSF (mg/dL, median)	124.8 (76.9,244.3)
< 45	2 (5.7%)
≥ 45	33 (94.3%)
CSF
TNF-α
< 5.39	7 (20.0%)
≥ 5.39	28 (80.0%)
IL-6
< 93	29 (82.9%)
≥ 93	6 (17.1%)
IL-8
< 89.3	3 (8.6%)
≥ 89.3	32 (91.4%)
Plasma
TNF-α
< 4.84	3 (8.6%)
≥ 4.84	32 (91.4%)
IL-6
< 2.83	15 (42.9%)
≥ 2.83	20 (57.1%)
IL-8
< 89.1	22 (62.9%)
≥ 89.1	13 (37.1%)

Values are presented as number (%) or median (IQR). IQR interquartile range, IDH isocitrate dehydrogenase, MTX Methotrexate, TMZ Temozolomide, VP-16 Etoposide, DDP Cisplatin, CBP carboplatin, LM Leptomeningeal metastases, MRI Magnetic resonance imaging, KPS Karnofsky Physical Status, CSF Cerebrospinal Fluid

Univariate and multivariate analyses

Follow-up continued until December 2024. Median OS was 11.8 months (range 2.2–28.6 months, 95% confidence intervals (CI) 6.1–17.5 months). The median time to diagnosis of LM from initial glioma was 15.7 months (range 0–32.8 months, IQR 11.4–31.2 months).

We used the surv_cutpoint function of the R package survminer in the R programming language to determine the optimal cut-off value of cytokine associated with OS. And the optimal cut-off values were as follows: TNF-α for CSF was 5.39 pg/mL and for peripheral blood was 4.84; IL-6 for CSF was 93 pg/mL and for peripheral blood was 2.83; and IL-8 for CSF was 89.3 pg/mL and for peripheral blood was 89.1. As shown in Table 2, on univariate Cox analysis of the entire cohort, KPS (P = 0.049) and CSF IL-6 < 93 pg/mL (P = 0.018) were associated with longer OS; on multivariate Cox analysis, CSF IL-6 < 93 pg/mL (vs. ≥ 93 pg/mL; hazard ratio (HR), 0.222; 95% CI, 0.072–0.683; P = 0.009) and KPS ≥ 70 (vs. < 70; HR, 0.346; 95% CI, 0.136–0.876; P = 0.025) were positively significant factors correlated with OS. Figure 1A-C show the Kaplan–Meier OS curves stratified according to all patients, KPS, and CSF IL-6 levels. The median OS was significantly different in the subgroup of patients with KPS (“≥70” vs. “<70”: 6.7 vs. 28.6 months, P = 0.04). And the difference in the subgroup of patients with CSF IL-6 was also statistically significant (“< 93 pg/mL” vs. “≥ 93 pg/mL”: 13.1 vs. 5.6 months, P = 0.01).

Table 2.

Univariable and multivariable Cox analyses of potential prognosistic factors for overall survival of glioma patients with LM

Characteristic	Univariable analysis		Multivariable analysis
	HR (95% CI)	p	HR (95% CI)	p
Gender
Fmale	1.344 (0.509-3.547)	0.551
Male	Ref
Age at initial glioma diagnosis
< 41 years	0.493 (0.188-1.288)	0.149
≥ 41 years	Ref
Location
Infratentorial	1.892 (0.436-8.206)	0.394
Supratentorial	Ref
Ventricular-Peritoneal shunt
Yes	0.502 (0.146-1.732)	0.275
No	Ref
KPS at LM diagnosis
≥ 70	0.402 (0.162-0.997)	0.049	0.346 (0.136-0.876)	0.025
< 70	Ref		Ref
Cycles of LM treatment
≥ 5	1.223 (0.462-3.235)	0.685
< 5	Ref
Ommaya reservoir implant
Yes	0.993 (0.329-3.000)	0.990
No	Ref
The content of total protein in the CSF (mg/dL)
< 45	0.960 (0.126-7.280)	0.968
≥ 45	Ref
CSF
TNF-α
< 5.39	0.550 (0.158-1.910)	0.346
≥ 5.39	Ref
IL-6
< 93	0.270 (0.091-0.801)	0.018	0.222 (0.072-0.683)	0.009
≥ 93	Ref		Ref
IL-8
< 89.3	0.035 (0.000-7.759)	0.224
≥ 89.3	Ref
Plasma
TNF-α
< 4.84	2.640 (0.757-9.213)	0.128
≥ 4.84	Ref
IL-6
< 2.83	2.043 (0.825-5.054)	0.122
≥ 2.83	Ref
IL-8
< 89.1	1.772 (0.669-4.690)	0.249
≥ 89.1	Ref

IDH isocitrate dehydrogenase, MTX Methotrexate, TMZ Temozolomide, VP-16 Etoposide, DDP Cisplatin, LM Leptomeningeal metastases, MRI Magnetic resonance imaging, KPS Karnofsky Physical Status, CSF Cerebrospinal Fluid, HR Hazard Ratio

Fig. 1.

Kaplan–Meier estimates of overall survival for the entire cohort and stratified by CSF IL-6 and KPS. Kaplan-Meier (KM) curves for OS of All patients (A), CSF IL6, P =0.04 (B), and KPS, P =0.01 (C)

CSF IL-6 is a useful diagnostic biomarker for LM patients of GBM

Figure 2A and B display the receiver-operating characteristic curves according to the CSF IL-6 and KPS. The areas under the curves for CSF IL-6 to predict 3, 5, and 7 months survival are 0.771 (95% CI : 0.494–1.048), 0.593 (95% CI : 0.392–0.793), and 0.634 (95% CI : 0.476–0.792), respectively. The areas under the curves for KPS to predict 3, 5, and 7 months of survival are 0.526 (95% CI: 0.244–0.808), 0.626 (95% CI : 0.408–0.845), and 0.609 (95% CI : 0.433–0.785), respectively. As shown in Fig. 2C, based on the previous multivariate Cox model, we integrated CSF IL-6 and KPS into the nomogram to predict 3-, 5-, and 7-month OS.

Fig. 2.

ROC-curve analysis and nomogram of KPS and CSF IL6 in LM from GBM for survival prediction of 3, 5, 7 months. ROC curves of CSF IL-6 (A) to predict survival of 3 moths (blue curve), 5 moths (orange curve), 7 moths (green curve). ROC curves of CSF IL-6 (B) to predict survival of 3 moths (blue curve), 5 moths (orange curve), 7 moths (green curve). The visual nomogram combined CSF IL-6 with KPS (C)

Discussion

In this study, we investigated prognostic factors in LM patients of glioblastoma, and we found that KPS and CSF IL-6 were associated with the prognosis in LM patients of glioma treated with intra-CSF chemotherapy.

LM had advanced and life-threatening cancer complications with a poor prognosis. The incidence of LM may be estimated to be as high as 10% in patients with metastatic cancers and even more than 20% in gliomas, and may still be underestimated [5, 7, 20]. Treatment of LM usually includes systemic chemotherapy, intrathecal chemotherapy, local radiotherapy, and targeted therapy to improve the quality of survival, prevent and delay neurological destruction, and prolong survival [6, 21]. Although these treatment modalities have shown promising results, not all patients benefit. The majority of patients with meningeal metastases are currently poorly treated and still rely heavily on clinicians to detect evidence of meningeal metastases early through adjuvant testing to improve patient survival outcomes [22, 23]. Therefore, it is important to be able to identify which patients are most likely to benefit from treatment. In addition, with the increased incidence of LM in patients with glioma and increasing long-term survival, the search for meaningful predictors of survival is critical to guide therapeutic decisions. The relationships between biomarkers and prognosis in glioma patients have received increasing attention in recent years. Bunevicius [18] found in a study of 163 patients with glioma and meningioma receiving elective cranial surgery that serum IL-6 concentration is associated with greater unfavorable outcome risk in brain tumor patients and with greater mortality in high-grade glioma patients. Liu et al. [24] found in a study of 205 treatment-naïve patients with glioma that peripheral blood IL-2, IL-15, and IL-21 conditioning may be instrumental in predicting survival among patients with GBM. Pawel analyzed 50 patients with newly diagnosed high-grade gliomas and 40 healthy individuals and found that IL-1β, IL-6, IL-8, and IL-10 were associated with prognosis [25]. Shan [26]collected 86 glioma tissues, and used ELISA to measure IL-6 in the serum and CSF of these patients, finding that IL-6 in the glioma, CSF, and serum increased remarkably and increased with the elevation of glioma grade compared with normal controls. And Tatsuo [12] collected 75 CSF samples of glioma and found that the concentration of CSF IL-6 is associated with tumor-associated macrophages (TAMs)’ infiltration level, and higher CSF IL-6 levels were associated with poorer prognosis in GBM patients. In our study, we reached some similar and some dissimilar conclusions as well.

To the best of our knowledge, this is the first study on cytokines and prognosis in LM patients with glioma. In our study, we collected CSF and peripheral blood samples from 35 LM patients with GBM and measured concentrations of TNF-α, IL-6, and IL-8, in addition to clinical information of the patients, and performed univariate and multivariate Cox analyses. We found that high levels of CSF IL-6 (> 93 pg/mL) and KPS were associated with poor prognosis, which was similar to previous studies [12, 26]. Also similar to our previous study [27], KPS is associated with a favorable prognosis for LM patients. Therefore, clinical assessment can be conducted for patients with good performance status and whose wishes align with those of their family members, active treatment for LM may be considered [8, 28]. Interestingly, the critical value of CSF IL-6 was higher in our study compared to other studies, which may be due to the seeding of glioma cells into the CSF, resulting in a large release of pro-inflammatory factors. However, our study did not find any correlation between peripheral blood cytokines and prognosis, which is similar to the findings of Camilla et al. [29] who analyzed 158 patients with glioma WHO grade II-IV and found that plasma IL-6, YKL-40, and genetic variation in YKL-40 did not associate with survival in newly diagnosed GBM perioperative, which may be related to the treatment of steroid. Therefore this may be related to previous treatment prior to diagnosis in patients with LM. Importantly, since patients in our cohort routinely received systemic corticosteroids, the sustained prognostic value of CSF IL-6 despite this ubiquitous treatment highlights its robust “real-world” clinical utility and its resistance to systemic immunosuppression.

Futhermore, this disparity can be largely interpreted through the lens of Paget’s “seed and soil” hypothesis within the anatomically isolated central nervous system (CNS) [30]. The CNS represents a highly specialized anatomical sanctuary for metastatic dissemination, characterized by its distinct neuro-immune and cellular landscape. Notably, the remodeling of the metastatic tumor microenvironment (TME) likely precedes the physical arrival and colonization of disseminated tumor cells. This process is initiated by factors secreted from the primary tumor, which prime a localized “pre-metastatic niche” within the subarachnoid space [31]. In this context, the subarachnoid space represents a distinct “soil.” Disseminated glioblastoma cells (the “seeds”) entering the relatively nutrient-deprived CSF must interact with resident cells, such as leptomeningeal cells, macrophages, and microglia, to establish a metastasis-permissive niche [32, 33]. The localized overproduction of IL-6 within this enclosed space drives a vicious cycle: it not only promotes the survival, proliferation, and invasion of tumor cells but also reshapes the local immune landscape into an immunosuppressive state, allowing the tumor to evade immune clearance [34]. Therefore, CSF IL-6 serves as a direct readout of the hostile microenvironment within the subarachnoid space, driving rapid disease progression and poor survival independently of the patient’s systemic tumor burden or plasma cytokine levels.

IL 6 is a pro-inflammatory cytokine associated with glioma growth. It is also associated with many known cancer hallmarks that constitute the malignant phenotype of glioblastoma, including proliferation, angiogenesis, heterogeneity, promotion of inflammation, invasion, chemoresistance, and immunosuppression in tumors, which makes IL-6 attractive for targeted therapy or immunotherapy. In preclinical models, it has been found that IL-6 gene deletion or IL-6-targeted therapy is effective in blocking glioma growth, and clinical trials have confirmed that targeting IL-6 for GBM is effective in prolonging survival [35]. Our study further combined CSF IL-6 with KPS to predict survival and demonstrated good predictive performance, suggesting that CSF IL-6 can be used as a reference for prognostic assessment in LM patients of glioma and is expected to be integrated into daily clinical practice to guide treatment decisions. However, this should be evaluated with caution when simply using CSF IL-6 to guide treatment decisions at the individual patient level.

Our study has several limitations. On the one hand, the study is limited by its retrospective nature and the relatively small sample size. A larger subgroup of patients from more institutions is needed to further validate these findings. On the other hand, our study showed that CSF IL-6 levels correlate with OS. However, the mechanisms behind this correlation were not explored in this study. Further studies are currently underway at our institution that will help to fully understand and utilize these parameters and explore the biological mechanisms behind them.

In conclusion, the results of this study suggest that both baseline CSF IL-6 at the time of LM diagnosis and KPS are useful and usable prognostic biomarkers for LM patients of glioma, which can help to facilitate the stratification of high-risk patients and to provide clinically applicable value for clinical decision-making.

Abbreviations

CSF

Cerebrospinal fluid

LM

Leptomeningeal metastases

OS

Overall survival

IL-6

Interleukin-6

GBM

Glioblastomas

BBB

Blood-brain barrier

MRI

Magnetic resonance imaging

TNF-α

Tumor necrosis factor-α

KPS

Karnofsky physical status

IDH

Isocitrate dehydrogenase

CI

Confidence intervals

IQR

Interquartile range

HR

Hazard ratio

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Meng Zhang, and Xun Kang. The first draft of the manuscript was written by Meng Zhang and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by Talent Introduction Fund of Beijing Tiantan Hospital (RCYJ-2020-2025-LWB) and the Beijing Clinical Key Specialty Project (2-1-2-038).

Data availability

The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All methods were carried out in accordance with the relevant guidelines and regulations of IRB of Beijing Tiantan Hospital, Capital Medical Uniersity. The study was approved by the Institutional Review Board (approval number: (No. KY2020-151-02). Informed consent was obtained from all participants in accordance with the ethical guidelines of Beijing Tiantan Hospital. For minors under 16, consent was obtained from their legal guardians.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.