Late p65 nuclear translocation in glioblastoma cells indicates non-canonical TLR4 signaling and activation of DNA repair genes

Glioblastoma (GBM) is the most aggressive brain primary malignancy. Toll-like receptor 4 (TLR4) has a dual role in cell fate, promoting cell survival or death depending on the context. Here, we analyzed TLR4 expression in different grades of astrocytoma, and observed increased expression in tumors, mainly in GBM, compared to non-neoplastic brain tissue. TLR4 role was investigated in U87MG, a GBM mesenchymal subtype cell line, upon LPS stimulation. p65 nuclear translocation was observed in late phase, suggesting TLR4-non-canonical pathway activation. In fact, components of ripoptosome and inflammasome cascades were upregulated and they were significantly correlated in GBMs of the TCGA-RNASeq dataset. Moreover, an increased apoptotic rate was observed when the GBM-derived U87MG cells were co-treated with LPS and Temozolomide (TMZ) in comparison to TMZ alone. Increased TLR4 immunostaining was detected in nuclei of U87MG cells 12 h after LPS treatment, concomitant to activation of DNA repair genes. Time-dependent increased RAD51, FEN1 and UNG expression levels were confirmed after LPS stimulation, which may contribute to tumor cell fitness. Moreover, the combined treatment with the RAD51 inhibitor, Amuvatinib in combination with, TMZ after LPS stimulation reduced tumor cell viability more than with each treatment alone. In conclusion, our results suggest that stimulation of TLR4 combined with pharmacological inhibition of the DNA repair pathway may be an alternative treatment for GBM patients.

Toll-like receptor 4 (TLR4) is part of the receptor family for innate immunity that first recognizes endogenous (damage-associated molecular patterns-DAMPs) and exogenous (pathogen-associated molecular patterns-PAMPs) molecules 1 . This family is composed of 10 known receptors in humans, which are structurally similar (TLR1-10) 2 . Depending on the context, TLR4 activation may induce a pro-inflammatory and pro-survival response, which translates into a proliferative phenotype or may induce an anti-inflammatory response leading to cell death 2,3 .
TLR4 downstream signaling includes two distinct pathways: the myeloid differentiation primary response gene 88 (MyD88; "canonical" pathway) and TIR-domain-containing adapter-inducing interferon-β (TRIF; "noncanonical" pathway). Signaling through the canonical TLR4 leads to the activation of the transcription factor nuclear kappa B (NF-κB) by means of the translocation of its heterodimeric complex p50/p65 to the nucleus 4 . Once in the nucleus, NF-κB induces the transcription of pro-inflammatory genes coding for interleukins 6 (IL6) and 1β (IL-1β), tumor necrosis factor (TNF), adhesion molecules, and chemokines [5][6][7] , as well as genes coding for a proliferative response [8][9][10][11][12] . On the other hand, the non-canonical pathway consists of the TLR4 internalization to the endosome compartment by a phosphoinositide 3-kinase (PI3K)-dependent mechanism 13 . This process results in the activation of TRIF, which promotes an anti-inflammatory response by inducing the expression of interferon type I, interferon regulatory factor 3 and 7 (IRF3/7), and IL-10 2,14 . TRIF may also trigger a cell death pathway by interacting with receptor interacting protein kinase 1 and 3 (RIPK1, RIPK3), and the fas adaptor death domain (FADD), which in turn activate caspase 8 (CASP8), leading to apoptosis. In the absence of CASP8, the necroptosis pathway is activated 15 www.nature.com/scientificreports/ TLR4 stimulation may also lead to NACHT, LRR, and PYD domains-containing protein 3 (NLRP3) inflammasome activation by either MyD88 or the TRIF-dependent pathway 17,18 . The NLRP3 inflammasome complex comprises NLRP3, a sensing molecule with a pyrin domain (PYD), which interacts with an adaptor protein (ASC) with a CARD domain. The resulting combined ASC-CARD domain interacts with pro-CASP1 19 , which cleaves and releases interleukin-1 family members such as IL-1β and interleukin-18 (IL18) 20 .
Inflammatory cells in a solid tumor microenvironment may exhibit a pro-or anti-inflammatory profile depending on the stimuli 21 . Moreover, inflammatory receptors such as TLR4 are also expressed in tumor cells, adding another layer of complexity that results from potentially distinct effects of TLR4 downstream signaling in each cell compartment 22 . We previously demonstrated an increased TLR4 expression in human astrocytoma, particularly glioblastoma (WHO-grade IV astrocytoma). Glioblastoma (GBM) is the most common and the most aggressive primary brain malignancy in adults 23 , and due to its relevant heterogeneity and the invasive feature, the outcomes of current therapeutic strategies have remained almost invariably lethal. In fact, the standard of care, including surgical tumor cytoreduction followed by radiotherapy and adjunct chemotherapy with Temozolomide (TMZ), and several combinatorial rescue trials have not improved the median overall survival time of approximately 15 months 24 . Among the GBM molecular subtypes, the mesenchymal (MES) subtype, which harbors neurofibromin 1 (NF1) and RB transcription corepressor 1 (RB1) mutations, presents the poorest outcome, compared to the proneural (PN) subtype harboring somatic mutations in tumor protein p53 (TP53), platelet-derived growth factor receptor A (PDGFRA), and isocitrate dehydrogenase 1 (IDH1), and to the classical subtype with epidermal growth factor receptor (EGFR) mutations 25,26 .
In this context, we aimed to analyze the impact of TLR4 stimulation in a MES-GBM tumor cell. We worked with the hypothesis that activating the TLR4 downstream cascade might activate a cell death pathway and contribute to a better outcome for GBM patients, mainly with the MES subtype.

Results
TLR4 expression in human astrocytoma. The upregulation of plasmatic membrane TLRs have been previously demonstrated in astrocytoma, particularly in GBM by our group 27 . Here, we first recapitulated TLR4 expression in our cohort of 140 human astrocytoma of different grades of malignancy (26 AGII, 18 AGIII, and 96 GBM compared to 22 non-neoplastic [NN] brain tissue), and we next analyzed TLR4 signaling pathways. TLR4 expression was significantly higher in AGII, AGIII, and GBM when compared to NN (p < 0.05, Kruskal-Wallis and Dunn tests). Interestingly, among GBM molecular subtypes (14 PN, 36 CS, and 16 MES) 26 , TLR4 expression was higher in MES than in PN and CS subtypes, however a statistical significance was not reached in our cohort due to the small number of cases in each subtype. Then, we validated this result in a larger dataset of the TCGA cohort, and a significant difference of TLR4 expression was confirmed among GBM subtypes (p < 0.0001 Kruskal-Wallis test), being higher in MES subtype compared to PN and to CS subtypes (p = 0.001 for both comparisons, Dunn test) (Fig. 1) Lower grade astrocytomas (AGII and AGIII) presented TLR4 higher expression levels when compared to GBM samples (p = 0.001 for both comparisons, Dunn test).
TLR4 canonical signaling pathway in U87MG cells. To access TLR4 role in a GBM cell line, we treated U87MG cells with LPS and observed NF-κB activation aiming to analyze the canonical TLR4 signaling pathway. The NF-κB translocation to the nucleus was assessed by the presence of the p65 subunit in the nucleus. In immune cells, NF-kB translocation to the nucleus has been detected 30-100 min after TNF stimulation 28 . However, in our experiment the nuclear translocation was not observed at 30 min or 2 h after LPS stimulation, the time interval expected for NF-κB translocation to occur through the MyD88/TRAF6 canonical pathway. It   (Fig. 2B). The mRNAs expression alterations at 12 h after LPS stimulation corroborated the activation of TLR4-inflammasome, and -ripoptosome pathways, with increase of SRF, JUN and particularly IL1B (p < 0.05 One-way Anova, p = 0.012 Tukey test) transcripts levels. (Fig. 3A). We next checked the expression level of these targets in the TCGA RNASeq dataset to validate the activation of these pathways in human astrocytoma (Fig. 3B). In fact, significant upregulations of MYD88, SRF, JUN, IL1B, TRIF, RIPK1, NLRP3 were detected in GBM cases compared to lower grade astrocytomas (AGII and AGIII) ( Supplementary Fig. 1). When the expression pattern of these genes was compared among the GBM subtypes, MES subtype presented higher MYD88, IL1B, RIPK3 and NLRP3 expression levels than PN and CS subtypes ( Supplementary Fig. 1), in a similar pattern to TLR4 expression (Fig. 1C). Therefore, these observations of the TCGA dataset were convergent to U87MG expression profile after LPS stimulation, indicating upregulation of inflammasome and ripoptosome pathways in GBM, particularly in MES subtype.
As a next step, we checked whether the activation of TLR4 by LPS presented a co-stimulatory effect to TMZ, the alkylating agent used in the standard of care of GBM patients. Interestingly, U87MG cells presented a more significant early cell death after combined stimulation than either TMZ-or LPS-alone treatment after TLR4 localization in nuclei of U87MG cells. We also analyzed TLR4 protein distribution within U87MG cells upon 12 h of LPS stimulation compared to control. Unexpectedly, immunofluorescence analysis showed a granular pattern of TLR4 expression in nuclei in addition to its distribution in cytosol and plasma membrane in these cells ( Increased mRNA expression of DNA repair genes related to TLR4 stimulation. To better understand the role of TLR4 in the nucleus, we performed RNASeq in U87MG cell lines after 12 h of LPS stimulation and compared to non-treated cells. This transcriptomic analysis showed 286 upregulated genes and 232 downregulated genes (logFC >|0.5|, p < 0.05, and adjusted p < 0.1). The enrichment analysis by Hallmark gene sets 33 showed statistical significance for 25 pathways, including the pathway for DNA repair (HALLMARK_DNA_REPAIR), that was upregulated in the treated cells with a logFC of 0.244 (Fig. 6A). Eighteen differentially expressed genes included in this Hallmark group were detected, 16 genes upregulated and 2 genes downregulated after LPS treatment compared to control cells (Fig. 6B). These genes were grouped according to their biological function by GO classification as: DNA repair, DNA dependent DNA replication, regulation of DNA binding, mRNA splicing via spliceosome and transcription by RNA polymerase III. The analysis with the String Consortium tool showed high connectivity among them (Fig. 6C). Additionally, in the TCGA dataset of astrocytoma, their expressions were significantly higher in GBM compared to lower grade astrocytoma (AGII and AGII), (p < 0.0001, Kruskal-Wallis test), except for XRCC1 and POLA1. Among them, MBD4 (methyl-CpG binding domain 4), PLAUR (urokinase plasminogen activator receptor), POLE (DNA polymerase epsilon, catalytic subunit) and RAD51 (RAD51 recombinase) expressions were significantly higher in GBM-MES subtype when compared to CS and PN subtypes (p < 0.05 Dunn test for all comparisons) ( Supplementary Fig. 1). The DNA repair gene expressions correlated positively among themselves, particularly RAD51 with FEN1 (flap structure-specific endonuclease 1)    Considering these correlation findings in the TCGA dataset and our observations of differential expression after LPS stimulus in U87MG cells, we checked the expression level of RAD51, FEN1, and UNG of U87MG cells after 0.5, 12, 24 and 48 h after LPS stimulation. These three genes presented an increase of their expressions in 24 h, with further increase of UNG expression in 48 h (Fig. 6D). Thus, these findings corroborated our transcriptomic data of DNA repair related genes upregulation upon LPS stimulation.

Decreased viability after RAD51 inhibition and LPS stimulation of U87MG cells.
Considering the upregulation of RAD51 expression after TLR4 stimulation with LPS, we chose Amuvatinib (Amb), a RAD51 inhibitor tested in clinical trial for advanced solid tumors 34 to check the impact on U87MG cells (Fig. 7A).
LPS treatment alone did not alter U87MG cell viability, however it significantly decreased with Amb single treatment (compared to non-treated in 48 and 72 h, p < 0.001). TMZ single treatment led to a sharper viability decrease in 72 h (compared to non-treated in 48 and 72 h, p < 0.001; compared to Amb in 72 h, p < 0.001). Moreover, a more pronounced decrease of cell viability was observed when Amb was combined to LPS stimulation (compared to non-treated in 48 and 72 h, p < 0.001; compared to Amb in 72 h, p < 0.001; compared to TMZ in 72 h, p < 0.001), which was comparable to Amb and TMZ combined treatment after LPS stimulation (compared to non-treated in 48 and 72 h, p < 0.001; compared to Amb in 72 h, p < 0.001; compared to TMZ in 72 h, p < 0.001) (Fig. 7B).

Discussion
In view of TLR4 dual role in cellular fate, which is a direct consequence of both the stimuli and microenvironmental context, we aimed to evaluate TLR4 molecular mechanism in human GBM cells. First, we determined TLR4 expression in 140 human astrocytoma samples of different malignant grades. We observed that TLR4 expression was higher in astrocytoma samples when compared to non-neoplastic samples. Moreover, when we focused on GBM samples, the analysis of the TGCA RNASeq dataset revealed that TLR4 expression was higher in the MES subtype, with the poorest outcome when compared to PN and CS subtypes.
The function of TLR4 has been extensively described in immune cells 35 . Nonetheless, TLR4 role in the tumor cell compartment is still lacking. To analyze TLR4 signaling pathways in GBM cells, the U87MG cell line, harboring NF1 mutation 36 , was chosen as a model for the MES-GBM subtype 25 .
We observed that stimulation of U87MG cells with LPS, a traditional TLR4 agonist that activates pro-inflammatory pathways 37 , led to NF-kB (p65) nuclear translocation after 12 h. This was considered a late response because both tumor necrosis factor-alpha (TNFα) and LPS stimulation have been described to induce NF-κB translocation to the nucleus through the MyD88/TRAF6 canonical pathway within 30 min to 2 h 38 . Additionally, hyaluronic acid, a component of the brain extracellular matrix, has been described to trigger TLR4-NF-κB pathway in GBM stem-like cell differentiation and maintenance in a proliferative astrocytic precursor state 22 . Furthermore, angiogenin has also induced a proliferative phenotype in U87MG cells through NF-κB translocation to the nucleus 39 . In opposition to these previous observations, an unequivocal proliferative effect was not observed in our experiments with U87MG. In another report, TLR4 was able to suppress tumor growth by decreasing the expression of a transcription factor responsible for the maintenance of cancer stem cells, the retinoblastoma binding protein 5 40 .
A late NF-κB translocation to nucleus was observed in U87MG cells stimulated by LPS, and the gene expression profile of our experiments was compatible to activation of inflammasome and ripoptosome pathways. TRIF www.nature.com/scientificreports/ was described to interact with TRAF6, leading to a late-phase NF-κB activation. Moreover, TLR4 signaling activates the TRIF-RIPK3 complex 41 and leads to apoptosis in the presence of CASP8, or to necroptosis in its absence 32 . However, CASP8 activation can be prevented by its inhibitor, c-FLIP, or by blocking TLR4-pathway and inhibiting apoptosis 42 . Following TLR4 internalization, TRIF mediates the activation of inflammasome complexes 43 . Two distinct inflammasome pathways have been described: a pro-proliferative through NLRP3 overexpression 44 and a cell death signaling pathway 45 , a pyroptotic type of cell death triggered by persistent stimulation of IL-1β 46 . Additionally, the TLR4 pathway has been described to trigger apoptosis in glioma in a TRIF-dependent pathway 47 . Our transcriptomic data of U87MG cells 12 h after LPS stimulation and the TCGA RNASeq dataset of astrocytoma showed upregulation of genes related to inflammasome and ripoptosome pathways, namely MYD88, NLRP3, RIPK1 and RIPK3, and also SRF, JUN and IL1B, the downstream regulated genes. In this context, an increase of apoptosis of LPS stimulated U87MG cells was expected, however only an increase of approximately 10% in early apoptosis 48 h after the combined treatment of TMZ and LPS was observed. Moreover, we detected an increase of TLR4 distribution in the nucleus of U87MG, 12 h after LPS stimulation. In fact, portions of the TLR4 protein were predicted by computational analysis to be in the nucleus 48 , and such nuclear distribution has already been reported in pancreatic cancer cells 49 , as well as in non-neoplastic rat cells 50 . Additionally, the combined TLR4 expression in nuclei and in cytosol has been associated with poor prognosis and metastasis 51 .
In an attempt to understand the mechanisms that lead to the low rate of apoptosis, and the role of TLR4 in the nucleus, NGS transcriptomic analysis of U87MG cells after 12 h of LPS stimulation was performed. Interestingly, among the enriched gene sets differentially expressed after TLR4 activation we found the DNA repair-related genes. Particularly, UNG, FEN1 and RAD51 presented a time-dependent pattern of upregulation after LPS stimulation in U87MG cells. Considering that the activation of DNA repair cascade may increase tumoral cell fitness, with consequent enhancement of tumor cell survival, we tested the impact of inhibiting this cascade in this model of TLR4-LPS stimulation. In fact, such strategy has already been proposed as targeting the phosphorylation of H2AX, a histone acting for the assembly of repair foci, through modification of TLR4 pathway 22 . Moreover, the activation of DNA repair signals has been associated with TMZ resistance, including the MGMT (O-6-methylguanine-DNA methyltransferase) repair system, which catalyzes the transference of the methyl group to the cysteine residue of the MGMT protein 24 . Therefore, targeting DNA repair proteins has been pursued as interesting cancer therapeutic strategy.
Among the DNA repair related genes identified in our transcriptomic data, we selected RAD51 as its inhibition has already shown antitumor activity when combined with other therapeutic agents 34 . RAD51 has a role in the DNA repair pathway of homologous recombination (HR), and it forms a complex in a single strand DNA break, allowing the homology search and the HR process to continue 52 . Amuvatinib (Amb) is a multi-kinase inhibitor, including RAD51, and it has been tested in a clinical trial for advanced solid tumors 34 . Additionally, in several GBM cell lines, Amb had a radiosensitization effect and delayed tumor growth, as in U87MG subcutaneous xenograft mice model 53 . Convergently to this previous finding, we observed a significant reduction of U87MG cell viability with Amb treatment after TLR4-LPS stimulation, in a similar fashion as the combined Amb and TMZ treatment after TLR4-LPS stimulation. Furthermore, TLR4 protective role against tumorigenesis by downregulation of DNA damage repair proteins was also described in a mouse hepatocellular carcinoma 54 . Additionally, the absence of TLR4 was associated with DNA damage resistance after UV irradiation in mice skin cells 55 . These cumulative pieces of evidence suggest TLR4 as a DNA repair pathway regulator due to its capacity to recognize DAMPs and to generate an auto-paracrine signal in the presence of DNA damage 56 . Therefore, we speculate that the reduction of tumor cell viability by TLR4 stimulation combined with the downregulation of DNA repair genes may be an attractive complementary cancer therapeutic strategy, and a treatment alternative for tumor TMZ resistance, as hypermutated TMZ status 57 . Further clarification of the involved mechanisms might enable TLR4-targeted therapies to be combined with the current standard of care to improve GBM patient outcomes.

Materials and methods
TLR4 expression in human astrocytoma and GBM cell line. Biological samples were collected during the neurosurgical procedure by the Neurosurgery Division of the Neurology Department of the Hospital das Clinicas, Faculdade de Medicina, Universidade de Sao Paulo (FMUSP) after informed and written consent from the patients following the Institutional Ethical Committee guidelines (process number: 691/05). The present study was approved by the HCFMUSP (process number: 059/15) and the FMUSP, ethical committee (process number: 278/15). The samples were snap frozen in liquid nitrogen and necrotic, gliotic, and non-neoplastic areas were previously macrodissected to guarantee presence of more than 90% of tumor cells in the processed tumor fragments as described previously 57 .
Biological samples and cell line total RNA were extracted using the RNeasy Mini Kit (Qiagen, Hilden, Germany) following the manufacturer instructions. Purity and concentration were analyzed by NanoDrop (Thermo Fisher Scientific, Carlsbad, CA, USA), and 1.8-2.0 values for the absorbance ratio 260/280 were considered satisfactory. Electrophoresis in agarose gel was done to check RNA quality.
For the reverse transcription, 1 µg of total RNA of each sample was used. Treatment with DNase I (FPLC-puro, GE Healthcare, Uppsala, Sweden) was performed, and RNA was reversely transcribed with random primers, oligodT oligonucleotides RNase inhibitor, and SuperScript III (Thermo Fisher Scientific). Lastly, the cDNA was treated with RNase H (GE Healthcare) and diluted in TE (Tris/EDTA) buffer.
We analyzed 22 non-neoplastic (NN) brain samples obtained from temporal lobectomy of epilepsy surgery, 26 astrocytoma grade II (AGII) samples, 18   Gene expression analyses. Gene expression levels were evaluated by qRT-PCR using the Sybr green approach to in an ABI 7500 (Thermo Fisher Scientific), using the Power Sybr green PCR master mix (Thermo Fisher Scientific). The primers were synthesized by IDT (Coralville, IA, USA) and Thermo Fisher Scientific. The sequence of primers were: TLR4 (F: TTT ATC CAG GTG TGA AAT CCA GAC; R: TCC AGA AAA GGC TCC  CAG  Statistical analysis. The program SPSS version 23.0 (IBM, Armonk, NY, USA) and Graph Pad Prism (Graph-Pad Software Inc, CA, USA) were used for the statistical analysis and graphs. The Normality of the data distribution was analyzed by the Kolmogorov-Smirnov test. Kruskal-Wallis and post hoc Dunn tests were used to analyze the differences among groups when non-parametric, and One-way Anova when parametric and multiple groups comparisons for the cellular treatment experiments, and Tukey as post hoc test (p65 nuclei staining and expression levels). For multiple variables comparison Two-way Anova was used, upon significance of interaction, column and row factor, Bonferroni post hoc test (apoptosis by flow cytometry) and Tukey post hoc (cellular viability assay) were applied. The test Mann Whitney was used for two groups comparison (TLR4 nuclear staining). The correlations were performed by Spearman's rho test. Differences were considered significant for p < 0.05.