Background: Immunotherapy has been the standard first-line therapy in patients with advanced non-small cell lung cancer (NSCLC), and shown a certain efficacy for patients with brain metastases. however, the appropriate predictive biomarkers for intracranial tumor response to immunotherapy are unclear. We conducted this prospective study to explore the immunological cytokines in cerebrospinal fluid (CSF) to predict intracranial tumor response to immunotherapy in NSCLC patients with brain metastases. Methods: We prospectively enrolled treatment-naive, EGFR/ALK wild-type NSCLC patients with brain metastases in this study. Patients received chemotherapy plus camrelizumab (a humanized anti-PD-1 monoclonal antibody) as first-line treatment, and paired plasma and cerebrospinal fluid (CSF) samples were collected at baseline and first treatment evaluation (8-week). All samples were detected for 92 Immuno-Oncology cytokines using the Olink panels, and were explored for predictive biomarkers to immunotherapy. The Mann-Whitney U test was performed to compare the immunological cytokine expression between intracranial response and non-response groups. And logistic LASSO regression model was used to variate selection and develop a CSF immuno-cytokine model to predict intracranial tumor response. Results: Between April 2020 and May 2022, a total of 28 patients were enrolled this study.At baseline, most immunological cytokines were significantly lower in paired CSF than in plasma, whereas a subset including CD83, PTN, TNFRSF21, TWEAK, ICOSLG, DCN, IL-18 and MCP-1 were increasing expressed in CSF samples. For patients with objective intracranial tumor response compared to patients with non-response, the baseline CSF levels of LAMP3 were significantly higher, whereas the CXCL10, IL-12, CXCL11, IL-18, TIE2, HGF and PDCD1 levels were significantly lower. And patients with lower HGF, IL-18 or PDCD1 levels had significantly longer intracranial progression-free survival to immunotherapy. The above CSF immunological cytokines were significantly decreased at first treatment evaluation in patients with intracranial tumor response. Our logistic CSF immuno-cytokine model selected TIE2, IL-18, LAMP3, PDCD1 and IL-12 that were significantly associated with intracranial tumor response, and yielded a mean AUC of 0.91, compared to PD-L1 expression (mean AUC of 0.72). Conclusions: Immunological cytokines in CSF could predict intracranial tumor response to immunotherapy in NSCLC patients with brain metastases, and findings require validation in a larger prospective cohort.