Human cerebrospinal fluid affects chemoradiotherapy sensitivities in tumor cells from patients with glioblastoma

Abstract

Cancers in the central nervous system resist therapies effective in other cancers, possibly due to the unique biochemistry of the human brain microenvironment composed of cerebrospinal fluid (CSF). However, the impact of CSF on cancer cells and therapeutic efficacy is unknown. Here, we examined the effect of human CSF on glioblastoma (GBM) tumors from 25 patients. We found that CSF induces tumor cell plasticity and resistance to standard GBM treatments (temozolomide and irradiation). We identified nuclear protein 1 (NUPR1), a transcription factor hampering ferroptosis, as a mediator of therapeutic resistance in CSF. NUPR1 inhibition with a repurposed antipsychotic, trifluoperazine, enhanced the killing of GBM cells resistant to chemoradiation in CSF. The same chemo-effective doses of trifluoperazine were safe for human neurons and astrocytes derived from pluripotent stem cells. These findings reveal that chemoradiation efficacy decreases in human CSF and suggest that combining trifluoperazine with standard care may improve the survival of patients with GBM.

Fig. 1.. GBM cells are larger and elongated in CSF.

(A to E) Morphology analysis of patient-derived GBM cells cultured in CSF or standard GM for 3 or 7 days. (A) Representative images of GBM cells (SB2b) cultured in GM and after 3 days’ exposure to CSF. Cells are stained with DAPI, phalloidin iFluor 488, and CellTracker Deep Red. (B) Cell area and (D) cell length:width determined using Harmony software analysis of CellTracker Deep Red–stained cells (n > 5000 per cell line). Individual data points represent the mean of six replicates for each of 25 GBM cell lines. Bar graphs represent the means ± SEM of all 25 GBM cell lines. Significance was determined using paired, two-way Wilcoxon tests. (C) Cell area and (E) cell length:width normalized to GM. Graphs represent the means ± SEM of all 25 GBM cell lines. Significance was determined using paired, two-way Wilcoxon tests. ns, not significant.

Fig. 2.. Patient-derived GBM cell lines display heterogeneous responses to standard therapeutics.

(A to J) Cytotoxicity analysis of patient-derived GBM cells cultured in CSF or GM for 3 days before treatment. Representative images of (A) cell lines untreated or treated with 1:2000 dimethyl sulfoxide (DMSO) control and 25 and 100 μM TMZ for 7 days (SANTB00497); (E) untreated or treated with five fractions of 2-Gy irradiation over 5 days (SANTB00469); (H) treated with 1:2000 DMSO or five fractions of 2-Gy irradiation and 100 μM TMZ for 5 and 7 days, respectively (SANTB00469). Cells are stained with DAPI. Responsiveness of 25 patient-derived GBM cell lines exposed to (B) 1:2000 DMSO or 100 μM TMZ for 7 days, (F) five fractions of 2-Gy irradiation for 5 days, and (I) 1:2000 DMSO or five fractions of 2-Gy irradiation and 100 μM TMZ. Individual data points represent six replicates for each GBM cell line normalized to the control. Bar graphs represent the means ± SEM. Significance determined using two-way analysis of variance (ANOVA). Fold change of (C) TMZ-responsive lines (group 1, <50% cell survival); (G) lines responsive (group 4), moderately responsive (group 5, 50 to 75% cell survival), or unresponsive (group 6, >75% cell survival) to irradiation; and (J) lines responsive (group 7), moderately responsive (group 8), or unresponsive (group 9) to combination treatment. Cell viability in GM and CSF normalized to mean population survival in GM. Individual points represent the mean of six replicates for each GBM cell line. Significance determined using two-way paired, nonparametric Wilcoxon test. (D) Dose response of TMZ responsive lines (group 1) treated with 1:2000 DMSO (0 μM) and 12.5, 25, 50, and 100 μM TMZ. Cell counts normalized to 0 μM GM or CSF. DAPI-stained cells analyzed using Harmony. Significance and IC₅₀ determined by nonlinear regression. ns defined as P > 0.05.

Fig. 3.. GBM cells become less proliferative and more quiescent in CSF.

(A to J) Proliferation analysis of patient-derived GBM cells cultured in CSF or GM for 3 or 7 days. (A) Representative images of GBM cells (SANTB00442) cultured in GM and after 3 days’ exposure to CSF. Cells are stained with DAPI and for Ki67. (B) Percentage of proliferating cells determined using Harmony software analysis of Ki67⁺ and DAPI⁺ cells. Individual data points represent six replicates for each GBM cell line. Bar graphs represent the means ± SEM of all 25 GBM cell lines. Significance determined using paired, two-way Wilcoxon tests. (C) Percentage of proliferating cells in GM and CSF normalized to GM. (D) to (J) Single-cell RNA sequencing (scRNA-seq) analysis of 10 patient-derived GBM cell lines (n = 18,733 cells in GM-CSF) cultured in GM or CSF for 3 days. (D) and (E) Scatter plots representing the percentage of cycling (G₁S or G₂M > 0) and noncycling cells (G₁S and G₂M < 0) in GM and CSF. G₁S and G₂M scores were calculated for each individual cell in the dataset using gene lists by Tirosh et al. (31). Density plots showing (F) proliferation and (I) quiescence scores of cells in GM and CSF. The proliferation score represents an average of G₁S and G₂M scores. Quiescence score was calculated using gene lists by Atkins et al. (30). Significance was calculated using a two-way, unpaired Mann-Whitney test. Gene set enrichment analyses (GSEAs) of (G) and (H) G₁S and G₂M genes and (J) quiescence genes. Genes differentially expressed in CSF and GM were scored as sign(log₂FC)* − log₁₀(adjusted P). GSEA was performed using preranked GSEA with default settings (permutations; 1000, enrichment statistic; weighted). ns defined as P > 0.05. FDR, false discovery rate.

Fig. 4.. GBM cells shift toward an MES-like state in CSF.

(A to H) scRNA-seq analysis of changes in transcriptomic state using 10 patient-derived GBM cell lines cultured in GM or CSF for 3 days (n = 18,733 cells in GM-CSF). Density plots showing (A) MES-like, (E) MES-like 1, and (F) MES-like 2 scores of cells in GM and CSF. MES-like, MES-like 1, and MES-like 2 states were defined using gene lists by Neftel et al. (34). GSEA of (B) MES-like, (G) MES-like 1, and (H) MES-like 2 genes. Differential expression (CSF versus GM) was performed on a downsampled dataset (n = 1400 cells per patient, 700 cells per condition). Relatively up-regulated genes in CSF and GM were scored as sign(log₂FC)* − log₁₀(adjusted P). GSEA was performed using preranked GSEA with default settings (permutations; 1000, enrichment statistic; weighted). (C and D) Scatter plots represent the percentage of proliferative, nonproliferative, MES-like, and non–MES-like cells in GM and CSF. Scatter plots represent the percentage of MES-like, AC-like, NPC-like, and OPC-like cells in (I) nine patient tumors (n = 8043 cells), (J) GM, and (K) CSF. AC-like, NPC-like, and OPC-like scores were defined using gene lists by Neftel et al. (34). (L) Proportion of MES-like, AC-like, NPC-like, and OPC-like cells in GM and CSF. Individual points represent the 10 patient cell lines with error bars representing the means ± SEM of the population. Significance was determined using two-way, paired Wilcoxon tests. (M) Pie charts show the proportion of MES-like, AC-like, NPC-like, and OPC-like cells in GM and CSF for individual cell lines. ns defined as P > 0.05.

Fig. 5.. NUPR1 and other potential therapeutic targets are up-regulated upon exposure to CSF.

(A and B) Differential gene expression analysis for GBM cell lines cultured in GM and CSF for 3 days (n = 7000 cells in GM-CSF). (A) Heatmap shows the average log₂ fold change of the top 20 highly expressed genes in CSF and GM. Differential gene expression analysis between CSF and GM was performed for individual cell lines and all cell lines together. Positive (red) average log₂ fold change represents genes highly expressed in CSF or relatively down-regulated in GM. Negative (blue) average log₂ fold change represents genes highly expressed in GM or relatively down-regulated in CSF. Dot plot represents the average expression and proportion of cells expressing the top 20 highly expressed genes in GM and CSF. The table represents the functional role of the top genes in GM and CSF. (B) Volcano plot showing genes that are differentially expressed (in red: −log₁₀(adjusted P) > 100, log₂(FC) > 0.5; in blue: −log₁₀(adjusted P) > 100, log₂(FC) < −0.5) between GBM cells in CSF relative to GM. P value adjustment was performed using Bonferroni correction. The top 20 relatively up-regulated (red) and down-regulated (blue) genes are labeled. (C) Chord plot shows association of the top differentially expressed genes with GO biological processes. GO overrepresentation analysis was performed using clusterProfiler. Louvain clustering was then applied to significant GO terms (adjusted P < 0.05) using the simplifyEnrichment package. New biological process categories were applied for the eight major clusters based on the enriched GO terms. ER, endoplasmic reticulum.

Fig. 6.. Targeting NUPR1 with repurposed TFP reduces cell survival in CSF.

(A) UMAP showing NUPR1 expression in 10 GBM cell lines (n = 18,733). (B) NUPR1 expression in cell lines (n = 8) cultured in GM or CSF for 3 days measured by reverse transcription quantitative polymerase chain reaction (PCR). Individual data points, replicate measurements. Bar graphs, means ± SEM. Significance determined using two-way ANOVA. (C) Fold change in NUPR1 expression shown in (B). NUPR1 expression in CSF normalized to GM. (D) Kaplan-Meier survival curves showing overall survival versus NUPR1 expression for GBM. Datasets from GlioVis (70). “NUPR1 high” and “NUPR1 low” defined by optimum cutoff. (E) NUPR1 expression in primary and recurrent GBM. CGGA RNA-seq dataset from GlioVis. Scatter plots, means ± SEM. Significance determined using unpaired t test with Welch’s correction. Responsiveness of 25 GBM cell lines to (F) Dulbecco’s phosphate-buffered saline (DPBS) or TFP for 24 hours or (K) irradiation and TMZ or irradiation, TMZ and TFP. Fold change of cell lines (G) responsive (group 10), moderately responsive (group 11), or unresponsive (group 12) to TFP and (L) moderately responsive (group 13) or unresponsive (group 14) to combination treatment with TFP. Cell viability in GM and CSF normalized to mean population survival in GM. (H) Bar graph showing expression of TFP targets from the scRNA-seq dataset. Individual data points, mean expression (n = 10). Bar graphs, means ± SEM. Significance determined using two-way ANOVA. (J) Representative images of DPBS- and TFP-treated cells (SANTB00468). (I) Dose response of TFP-responsive lines (group 10) to TMZ. (F and K) Individual data points represent six replicates of each cell line normalized to control. Bar graphs, means ± SEM. Significance determined using two-way ANOVA. (I) Individual points and error bars, means ± SEM. Significance and IC₅₀ determined by nonlinear regression. ns indicates P > 0.05.

Fig. 7.. Preclinical neurotoxicity assays of TFP and TMZ on healthy human neuronal cultures.

Representative images of (A) cortical and (F) midbrain neuronal cultures treated with 1:2000 DMSO or 100 μM TMZ and stained for glial fibrillary acidic protein (GFAP) (ACs), MAP2 (neurons), and with DAPI (all cells). The percentage of GFAP⁺ ACs and MAP2⁺ neurons in (B) cortical and (G) midbrain neuronal cultures with 1:2000 DMSO and 100 μM TMZ treatment. DAPI, DAPI/GFAP⁺, or DAPI/MAP2⁺ cells determined using Harmony software. Bar graphs represent the means ± SEM of six replicates. Significance was determined using two-way, unpaired Mann-Whitney tests. (C) Cortical and (H) midbrain: Raster plots depict 60 s of per-neuron firing activity in two individual wells during (day 5) and after (day 10) treatment for cortical and midbrain cultures, respectively. Each line indicates one action potential. Network firing line plots indicate synchronous firing events (arbitrary scale). (D) Cortical and (I) midbrain: Time series of single-cell spike-sorted neuronal electrophysiology data measured by multielectrode array (MEA) over 10 days. Solid lines represent the mean of data from single neurons averaged per well smoothed with a two-point moving average algorithm; the shaded area represents ±SEM. Dashed vertical lines indicate addition and removal of treatment. Six to 9 wells analyzed per condition. Area under the curve of the treatment and recovery sections were calculated in GraphPad Prism, and significance was determined by unpaired t test. (E) Cortical and (J) midbrain: Heatmaps represent the percent change from pretreatment for each parameter across all time points normalized to the percent change at each time point in the DMSO (CTRL). Neuronal cultures were matured for 195 days (cortical; C to E) or 100 days (midbrain; H to J) at the time of treatment. ns defined as P > 0.05.

Fig. 8.. TFP has minimal effect on the survival and functionality of healthy human neuronal cultures.

Representative images of (A) cortical neuronal cultures treated with DPBS and 4 and 10 μM TFP and (F) midbrain neuronal cultures treated with DPBS and 4, 7, and 10 μM TMZ. Neuronal cultures were stained for GFAP (ACs), MAP2 (neurons), and with DAPI (all cells). Percentage of GFAP⁺ ACs and MAP2⁺ neurons in (B) cortical and (G) midbrain neuronal cultures treated with DPBS or TFP. DAPI, DAPI/GFAP⁺, or DAPI/MAP2⁺ cells were determined using Harmony software. Bar graphs represent the means ± SEM of six replicates. Significance was determined using two-way, unpaired Kruskal-Wallis tests. (C) Cortical and (H) midbrain: Raster plots depict 60 s of per-neuron firing activity in two individual wells during (day 5) and after (day 10) treatment for cortical and midbrain cultures, respectively. Each line indicates one action potential. Network firing line plots indicate synchronous firing events (arbitrary scale). (D) Cortical and (I) midbrain: Time series of single-cell spike-sorted neuronal electrophysiology data measured by MEA over 10 days. Solid lines represent mean of data from single neurons averaged per well smoothed with a two-point moving average algorithm; shaded area represents ±SEM. Dashed vertical lines indicate addition and removal of treatment. Six to nine wells analyzed per condition. Area under the curve of the treatment and recovery sections were calculated in GraphPad Prism, and significance was determined by unpaired t test. (E) Cortical and (J) midbrain: Heatmaps represent percent change from pretreatment for each parameter across all time points normalized to percent change at each time point in the DPBS (CTRL). Neuronal cultures were matured in culture for 108 days (cortical; C to E) or 87 days (midbrain; H to J) at the time of treatment. ns defined as P > 0.05.