2.1 Ethics

All the usage of patient samples in this study is approved by the Scientific Ethics Committee of the First Affiliated Hospital of Xi'an Jiaotong University, Xi'an, China (approve no. 2016-18). All samples have obtained necessary consent and were embedded in paraffin blocks as previously described. Totally, 63 glioma samples and three non-tumour tissue samples (from tumour edge) were collected from patients underwent surgical operations from 2016 to 2020. The basic information about 63 glioma patients was described in Table S1. And the WHO grade information, IDH and 1p19q status for patient samples, which were performed for a series of experiments in vivo and in vitro, were described in Table S2.

2.2 Differential gene expression analysis

Gene expression datasets and relevant clinical data of LGG were respectively extracted from Chinese Glioma Genome Atlas (CGGA, mRNASeq_693, http://cgga.org.cn/) and GDC Data Portal (https://portal.gdc.cancer.gov/). These gene expression profile data were pre-processed by background correction, gene symbol transformation and normalization using R programming (vision 4.0.0). The limma package was utilized for screening the differentially expressed genes (DEGs) in these datasets. The expression difference of individual gene was identified by Log₂ (Fold change) and adjusted P value, in which Log₂ FC > 2 with an adjusted P value < .05 was identified as a DEG.

2.3 Time-dependent receiver operating characteristic (ROC) curve

Time-dependent ROC analysis was performed according to previous reference for further measurement of the predictive performance among the candidate genes.¹⁶ Briefly, survival data and gene expression data were extracted from CGGA_mRNASeq_693 dataset. False positive rate and true positive rate were used as x axis and y axis, respectively, thus the performance of a gene candidate was evaluated by the area under the ROC curve (AUC) in which a higher AUC value indicates a better performance.

2.4 Least absolute shrinkage and selection operator (LASSO) regression

To further narrow the scope of the candidate genes, we adopted the LASSO binary logistic regression model and multivariate Cox regression according to previous reference.¹⁷ The LASSO logistic regression was applied to genome-wide association study (GWAS) to identify genomic locus associated with complex disease, and it can be directly employed in quantitative trait locus mapping for binary traits as shown in our simulation study and real data analysis. A coefficient profile figure was produced against the ln (λ) sequence. Cross-validation (CV) for tuning parameter (λ) selection was analysed via one standard error criteria. Two dotted vertical lines were presented at the optimal values by using minimum criteria and the one standard error of the minimum criteria (1-SE criteria). Genes with ln (λ) > 1.725 were selected, which is 1-SE criteria. Then the LASSO coefficient profile of nine candidate genes was calculated with CGGA cohort.

2.5 Quantitative RT-PCR (qRT-PCR)

qRT-PCR assays were conducted as previously described.¹⁸ In brief, total RNA was extracted using RNeasy mini kits according to the manufacturer's instruction, and the concentration of RNA was determined by Nanodrop 2000. qRT-PCR was performed following the synthesis of cDNA according to the standard protocols. GAPDH was used as an internal control. Relative mRNA expressions were calculated by 2^−ΔΔt method. The primer sequences were described in Table. S3.

2.6 Western blot

Western blot analysis was performed as previously described¹⁸ and each western blot was repeated at least three times. Antibodies used in this study were shown as below: anti-FABP5 primary antibody was purchased from Abcam (ab84028). Anti-β-actin primary antibody was purchased from Abcam and served as an internal control (ab115777). Anti-IKKα primary antibody was purchased from Cell Signaling Technology (#11930). Anti-p-IKKα (Ser176) primary antibody was purchased from Biorbyt (orb544342). Anti-IKβ (ab32518) and anti-NF-κB (ab32360) primary antibodies were purchased from Abcam. Anti-Rabbit IgG (#7074) and Anti-Mouse-IgG (#7076) were purchased from Cell Signaling Technology and served as negative controls. The raw unedited bands of all western blot images were showed in the Figure S7.

2.7 Immunohistochemistry (IHC)

Immunohistochemistry was performed as previously described.¹⁹ Briefly, anti-FABP5 primary antibodies were purchased from Thermo Fisher Scientific (PA5-76920). Goat anti-rabbit IgG was purchased from Abcam (ab97051) and served as secondary antibody. Tissues embedded with paraffin were cut into 4 mm sections followed by deparaffinized, rehydrated and stained with primary antibodies at 4℃ overnight. The slides were then incubated with secondary antibodies and stained with DAB. All glioma samples used in this study had been pathologically diagnosed and the recurrence was confirmed by computed tomography (CT) or magnetic resonance imaging (MRI). Nuclei were counterstained with haematoxylin.

2.8 German Immunohistochemistry Score (GIS)

GIS was used to measure the expression of FABP5 as previously described.¹⁹ Immunoreactivity score = positive cell score × staining intensity score. The positive cell score was calculated as below: 0, negative; 1, <10% positive; 2, 11%-50%; 3, 51%-80%; 4, >80%. Staining intensity score was graded as below: 0, negative; 1, weakly positive; 2, moderately positive; 3, strongly positive. Immunoreactivity score > 5 was considered as high expression and ≤ 5 were defined as high and low expression, respectively.

2.9 Cell cultures

LGG-derived primary culture cells 0708, 7419 and 7624 were obtained from LGGs patient samples that underwent surgical resection in the Department of Neurosurgery, the First Affiliated Hospital of Xi'an Jiaotong University. Fresh surgical LGG specimens were rinsed in PBS in order to remove adhering blood and visible necrotic portions. Mechanical and enzymatic tissue dissociation by trypsin solution was carried out in order to obtain single cell suspensions. Single cells were seeded into DMEM-F12 containing 15% FBS, 2% B27 and antibiotics (100 U/ml penicillin and 100 mg/ml streptomycin) and were incubated in sterile conditions at 37°C in a 5% CO₂ humidified atmosphere and the complete medium was totally replaced every 2-3 days. Only the early passages of the LGG primary cultures were used as in vitro cell model and cell viability was determined using the trypan blue methods. NHA cell line was gifted from Dr Jianxin Liu and cultured in DMEM-F12 containing 10% FBS with antibiotics (100 U/ml penicillin and 100 mg/ml streptomycin). Cell Counting Kit-8 (CCK-8) methods were used to evaluate cell proliferation according to the manufacturer's protocol.

2.10 In vitro cell viability assays

Cell viability assays were conducted as previously described.¹⁹ Briefly, cell number was calculated by cell counter with trypan blue, and cells were seeded into 96-cell plates after suspended adequately at a density of 2 × 10³ cells/100uL per well and cultured for 24 hours with 5% CO₂ at 37℃. These cell lines received indicated in vitro chemotherapy using corresponding concentration of TMZ. At last, cell number was counted by Cell Counting Kit-8 (CCK-8) and IC50 was calculated by SPSS 22.0.

2.11 Lentivirus production and transduction

Lentivirus production and transduction were performed as described previously.^{18, 19} The lentiviral overexpression/knock-down of FABP5 was designed and synthesized by Genechem (Shanghai, China). These lentiviruses were introduced into LGG cells according to manufacturer's instruction. Stable clones transduced with FABP5 and shFABP5 were selected for 4 weeks by puromycin. Lentiviral particles packaging the shRNA are targeting FABP5 (shFABP5#1: CCGGATTGAAAGATGGGAAATTAGTCTCGAGACTAATTTCCCATCTTTCAATTTTTTTG and shFABP5#2: CCGGGATGGGAAGGAAAGCACAATACTCGAGTATTGTGCTTTCCTTCCCATCTTTTTG).

2.12 Colony formation assay

Colony formation assays were conducted to detect the tumorigenic potential of LGG cells. 0708 or 7419 cells transduced with FABP5 overexpression lentivirus, shFABP5 lentivirus or empty vector were seeded into 6-well plates under indicating conditions. After cultured for 14 days to form colonies, these cells were fixed with methanol and stained with methylene blue. Number of clones was counted to evaluate the colony forming ability.

2.13 Wound healing assay

The wound healing assay is used to study the migratory ability of 0708 and 7419 cells. Cells were seeded in 6-well plates after adequately suspended. Afterwards, a sterile pipette tip was used to produce a wound line when cells reached the appropriate confluence. Images were taken under inverted microscope at the same location at 7 days after seeding. The leading edges were marked by white lines, and the relative distance of the borders was measured by Image J software.

2.14 Matrigel invasion assays

The matrigel invasion assays were performed to assess the invasion ability of 0708 or 7419 cells, 1 × 10⁵ cells pre-transduced with indicating lentivirus were suspended sufficiently and seeded into the upper chamber of ThinCert cell culture inserts (8 μm pore size, Greiner Bio-One). For cell invasion assays, the inserts were coated with 100 µl of 1 mg/ml growth factor depleted matrigel (Corning) before adding the transfected colon cancer cells. 500 µL serum-free medium was added to the lower chamber and replaced 24 hours later by DMEM-F12 containing 10% FBS. After incubation for 96 hours for invasion, inserts were removed, cells were fixed in 4% paraformaldehyde and methanol, and stained using 0.1% crystal violet. The non-migrated cells were slightly removed by cotton swabs from the upper surface of the filters, and photomicrographs of invaded cells were taken.

Gene ontology (GO) analysis, Gene set enrichment analysis (GSEA) and Gene set variation analysis (GSVA).

The gene expression profiles of previously mentioned datasets were read into R programming and pre-processed with background correction, gene ID transformation and normalization. Furthermore, these data were ordered by the expression level of FABP5 to divide all the LGG samples into two groups (FABP5^High and FABP5^Low) by quartile cutoff. Subsequently, the limma package was used to perform DEGs analysis. Afterwards, the screened DEGs by log₂FC > 2 and adjusted P value < .05 were used to perform gene ontology (GO) annotation using the Database for Annotation, Visualization and Integrated Discovery online tool (DAVID, https://david.ncifcrf.gov/), and GSVA package and GSEA software were used to validate, and visualized results using clusterprofiler package.

Besides, the expression profiles of LGG were uploaded into the GSEA software strictly following the guideline of the software to elucidate the enriched KEGG pathways that were significantly enriched in FABP5 high groups with the number of permutations set at 1000.

2.15 Statistical analysis

All the results in this study were presented as mean ± SD (standard deviation). The number of independent replications is presented in the figure legends. Two-tailed t tests were conducted to evaluate statistical differences and one-way ANOVA following Dunnett's post-test was used to compare the statistics among multiple groups. Kaplan-Meier survival analyses were performed using log-rank test. All statistical analysis was performed by GraphPad Prism 7 or SPSS 22.0 software, and statistical significance was defined as a two-sided P value less than 0.05.

3.1 FABP5 is one of the most differentially expressed genes in malignant LGGs

To further identify the genetic heterogeneity and malignant signatures among LGGs, IDH and 1p/19q status was used as specific classifier of LGGs. After removal of absent expression and clinical data, totally 546 LGGs samples extracted from the CGGA_mRNASeq_693 dataset were divided into three subgroups which exhibited significantly genetic diversity (Figure 1A). Additionally, Kaplan-Meier analysis was performed among the three subgroups and the results indicated that LGGs patients with IDH wild-type and 1p/19q non co-deletion had the most severe overall survivals while patients with IDH mutation and 1p/19q co-deletion exhibited prolonged survivals, which was consistent with the WHO classification (Figure 1B). Furthermore, differential gene expression analysis was performed between IDH^WT/1p19q^{Non co-del} and IDH^Mut/1p19q^Co-del samples to explore the most significant prognosis-related genes (Figure 1C). The results indicated that FABP5 was one of the most remarkably elevated genes in IDH^WT/1p19q^{Non co-del} samples (Figure 1D). Moreover, the prognostic capacity of the all genes was further demonstrated by the AUC of the time-dependent ROC curve. As shown in Figure 1E, FABP5 was one of the most appropriate prognostic genes compared with other candidates. To further narrow the scope of the candidate genes, LASSO logistic regression was performed.Nine selected candidate DEGs featured coefficients (not zero) in a further LASSO logistic regression model in which the selected genes were required to appear 1000 times of 1000 repetitions. Finally, two genes (FABP5 and IGFBP2) were selected as prognostic gene candidates for LGGs (Figure 1F,G).

To further investigate the expression of FABP5 in LGGs, qRT-PCR analysis was performed by using LGG samples (0864, 3247, 0708, 1789, 7419, 6567 and 7624) compared with non-tumour samples deriving from the matched para-cancerous tissues (7419 and 7624). The results demonstrated that FABP5 was significantly enriched in LGGs compared to non-tumour tissues (Figure 1H), which is consistent with the results from western blot analysis (Figure 1I and Figure S1A). Additionally, both qRT-PCR (Figure 1J) and western blot analysis (Figure 1K and Figure S1B) demonstrated that FABP5 was highly expressed in early passages of patient tumour-derived LGGs primary cultures compared to normal human astrocytes. Taken together, these data suggested that FABP5 was one of the most differentially expressed genes in malignant LGGs.

3.2 Elevated FABP5 expression revealed poor prognosis in LGGs

To gain more insight into clinical relevance of FABP5 expression in LGGs, CGGA and TCGA glioma databases were utilized to investigate the expression pattern of FABP5 in different pathological subtypes. The result indicated that FABP5 was increased in LGGs with higher WHO grade, particularly in GBM, compared to those with lower WHO grade (Figure 2A). Additionally, FABP5 was highly expressed in IDH^WT and 1p/19q^{Non co-del} LGGs, which was considered as the most lethal subtype of LGGs (Figure 2B). We then explored the correlation between FABP5 expression and tumour recurrence. FABP5 was overexpressed in recurrent gliomas (Figure 2C), consistently, was significantly enriched in recurrent LGGs (Figure 2D). IHC staining was performed by using samples derived from patients underwent surgical resection in our institution. The results indicated that FABP5 was predominantly enriched in the cytoplasm of glioma cells, while no significant staining was observed in the non-tumour tissues (Figure 2E). In addition, LGGs were divided into two groups according to the GIS of FABP5 expression. The data indicated that FABP5 were likely to be enriched in IDH^WT and 1p/19q^{Non co-del} LGGs compared to other two subgroups (IDH^WT and 1p/19q^{Non co-del} samples accounted for 44.44% in FABP5^High samples and accounted for 23.08% in FABP5^Low samples) (Figure 2F). Consistent results could be addressed when analysing CGGA dataset (Figure 2G). Moreover, increased FABP5 expression could be observed along with the malignancy elevation of gliomas in CGGA dataset (Figure 2H). Altogether, FABP5 elevation was closely correlated with higher histological grade and malignant subtypes of LGGs.

As previously described, FABP5 was a valid indicator for pathological classification and molecular grading of LGGs. To further investigate the prognostic characters of FABP5 in LGGs, Kaplan-Meier analysis was performed by using our LGG samples as well as open access datasets. Consistent with other previous studies, IDH^WT and 1p/19q^Non-co-del samples showed more severe outcomes compared with samples exhibiting other molecular subtypes (Figure 3A). As shown in Figure 3B, LGG patients with lower expression of FABP5 represented a favourable outcome when compared to those with higher FABP5 expression. Similar results were observed when using data extracted from TCGA and CGGA (Figure 3C,D). Then we combined FABP5 expression with IDH or 1p/19q status as predictors for survival in LGG patients. The results demonstrated that elevated FABP5 combined with either IDH wild-type or 1p/19q non co-deletion indicated poor prognosis in LGG patients (Figure 3E,F), which is consistent with the results from CGGA dataset (Figure 3G,H). Furthermore, increased FABP5 was closely correlated with severe prognosis in either primary or recurrence gliomas, particularly among LGGs (Figure S3A,B & Figure 3I,J). Lastly, down-regulated FABP5 indicated prolonged survival in WHO grade III gliomas and GBM; however, there was no significant correlation between FABP5 and prognosis in WHO grade II gliomas (Figure 3K,L and Figure S3C). Taken together, these findings demonstrated that FABP5 overexpression revealed poor prognosis in LGGs, particularly in malignant LGGs exhibiting high pathological grade or severe subtype based on IDH and 1p/19q status, indicating that FABP5 could serve as a prognostic biomarker and supporting indicator for malignancies in LGGs.

3.3 FABP5 was functionally required for multiple malignant behaviours in LGGs

To investigate the function of FABP5 in malignancies of LGGs, an exogenous silencing of FABP5 was performed in early passages of LGGs primary cultures 0708 and 7419 cells via lentivirus infection. The efficiency of transfection was roughly evaluated by fluoresces (Figure 4A,B). qRT-PCR and western blotting analysis confirmed that FABP5 expression was markedly decreased in FABP5 knock-down 0708 cells and 7419 cells (Figure 4C,D and Figure S1C). Furthermore, in vitro cell growth assay indicated that the proliferation of 0708 and 7419 cells was obviously reduced by FABP5 silencing (Figure 4E). Moreover, colony formation assays, matrigel invasion assays and wound healing assays were carried out to further study the function of FABP5 on tumour malignant biological behaviour in LGGs. The results indicated that self-renewal ability, immigration and invasion of LGG cells were remarkably reduced by exogenous suppression of FABP5 (Figure 4F-I, Figure S1D,E).

To further confirm the correlation between FABP5 and chemoresistance in LGGs, FABP5 overexpressed 0708 and 7419 cells were constructed before cell viability assays. The efficiency of transfection was roughly evaluated by fluoresces (Figure 5A,B). Results of qRT-PCR and western blot assays indicated that FABP5 was significantly up-regulated after lentivirus infection (Figure 5C,D and Figure S1F).

The proliferation assays were then performed to investigate the function of FABP5 overexpression in LGG cells. The result showed that elevated FABP5 significantly enhanced the proliferation and TMZ resistance of 0708 and 7419 cells (Figure 5E,F). Moreover, in vitro cell viability assays indicated that IC50 to TMZ was markedly increased by exogenous FABP5, indicating that overexpression of FABP5 increased the TMZ resistance in LGG cells (Figure 5G). We compare the survival between patients underwent TMZ or not to identify whether FABP5 enhanced the resistance of TMZ. The results showed the high expression of FABP5 with TMZ treatment had a poor prognosis compared to the low expression of FABP5 with TMZ treatment (Figure S2A,B). Altogether, these data demonstrated that FABP5 was functionally required for multiple malignant characters including cell proliferation, colony formation ability, immigration, invasion and chemotherapy resistance in LGGs.

3.4 FABP5 induced epithelial-mesenchymal transition (EMT) in LGGs

As previously illuminated, FABP5 was enriched in LGGs and promoted malignancies of LGGs, raising the possibility that unknown molecular mechanisms for therapy resistance besides pathological characters need to be clarified. To this end, expression profiles of LGG samples in CGGA and TCGA were grouped into two groups, respectively, depending on FABP5 expression. A subset of DEGs was identified through differential gene expression analysis (Figure 6A and Figure S4A). Then the Database for Annotation, Visualization and Integrated Discovery online tool (DAVID, https://david.ncifcrf.gov/) was used to conduct the GO annotation to investigate FABP5-related candidate pathways and data were visualized by using clusterprofiler package while GSVA package and GSEA analysis were used as validation. The results indicated that the EMT of GO terms were enriched when analysing either CGGA or TCGA datasets (Figure 6B and Figure S4B). Similarly, GSEA analysis demonstrated that EMT was simultaneously enriched in FABP5^High LGGs in both CGGA and TCGA database (Figure 6C and Figure S4C). Furthermore, qRT-PCR results verified that mesenchymal-related biomarkers were remarkably elevated by FABP5 overexpression; however, epithelial phenotype was significantly reduced (Figure 6D). In summary, these data revealed that FABP5 induced EMT process, which was possibly responsible for malignant transition of LGGs.

3.5 FABP5 enhanced malignancies in LGG via activation of NF-κB signalling

To intensively study the potential downstream molecular pathways of FABP5 in LGGs, GSEA pathway analysis was performed using DEGs from CGGA and TCGA datasets according to the expression of FABP5 (Figure 7A and Figure S5A). Consistently, the TNFα/NF-κB pathway was simultaneously enriched in FABP5^High LGGs in both CGGA and TCGA databases (Figure 7B,C, Figure S5B,C). Moreover, to gain more insights on the potential mechanisms, a bunch of downstream targets correlated to NF-κB signalling were extracted from CGGA and TCGA databases. Hierarchical bi-clustering analysis showed that NF-κB targeting genes were likely to be up-regulated in FABP5^High samples, indicating that NF-κB signalling could be regulated by FABP5 in LGGs (Figure 8A and Figure S6).

To further validate bioinformatics results and elucidate the effects of FABP5 on regulation of NF-κB signalling, lentiviral suppression of FABP5 followed with or without TNFα treatment was performed in 0708 LGG cells. In vitro cell growth assays indicated that FABP5 silencing significantly reduced cell proliferation of LGG cells while the cell growth kinetics was partially rescued by the treatment of TNFα (Figure 8B). Similarly, as expected, the ability for colony formation, invasion and migration were also decreased by FABP5 knock-down and subsequent TNFα treatment exhibited apparent rescue effects against FABP5 silencing (Figure 8C-E and Figure S1G-I). In addition, qRT-PCR analysis showed that FABP5 knock-down reduced the expression of a wide range of mesenchymal biomarkers; however, TNFα could partially reverse the inhibition effects (Figure 8F). As shown in Figure 8G and Figure S1J, western blot analysis showed that FABP5 knock-down significantly decreased expression of IKKα, particularly, reduced the phosphorylation level of IKKα thus reduced NF-κB expression through increased releasing of IKβ. Moreover, FABP5 inhibition notably presented coordinated tendency which reduced downstream targets of NF-κB pathway, and by contrast, these landmark genes were partially improved on transcriptome and translation levels after TNFα treatment was used (Figure 8H). Altogether, these results suggested that FABP5 induced EMT thus promoted malignancies of LGGs via TNFα-dependent canonical NF-κB activation.