Patients with low nicotinamide N-methyltransferase expression benefit significantly from bevacizumab treatment in ovarian cancer

The publicly available gene expression data used in our study is described in GEO database (GSE10971, GSE30587, GSE44104, GSE28739, GSE9891, and GSE140082). GSE10971 dataset was used to determine the differential expression of NNMT (probe ID:202237_at and 202238_s_at) between fallopian tube epithelium (FTE) (n = 24) and primary high grade serous tubal cancer/ovarian cancers (n = 13). GSE30587 dataset was used to determine the differential expression of NNMT (probe ID: 7943998) between paired primary ovarian cancer tissues(n = 9) and their omental metastasis counterparts(n = 9). GSE44104 dataset was used to determine the differential expression of NNMT (probe ID:202237_at and 202238_s_at) between ovarian cancer with recurrence (n = 20) and ovarian cancer with non-recurrence(n = 40). GSE28739 dataset was used to determine the differential expression of NNMT (probe ID: A_23_P127584) between drug sensitive ovarian cancer (n = 20) and drug resistant ovarian cancer (n = 30). If there are multiple probes all targeting NNMT, the average value of all probes is taken as an expression value of NNMT. Level 3 gene expression data with lymphovascular space invasion (LVSI) from TCGA dataset was used to explore the differential expression of NNMT between LVSI-positive ovarian cancers and LVSI-negative ovarian cancers.

CSIOVDB [15] (http://​csibio.​nus.​edu.​sg/​CSIOVDB/​CSIOVDB.​html) was used to assess the association of NNMT expression with the clinical and molecular parameters of ovarian cancer.

GSE9891 dataset and TCGA dataset were used to develop and validated a NNMT-related nomograms to predict overall survival (OS) respectively. We only included patients with enough information (including age at initial diagnosis, histological grade, FIGO stage, debulking status, days to death, and survival status) in our final analysis. Totally, 242 patients in GSE9891 dataset and 462 patients in TCGA dataset were included. The expression value of NNMT expression together with the clinical information in GSE9891 dataset and TCGA dataset were extracted from the “curatedOvarianData” Bioconductor package (version 2.12 for R 3.0.3).

Kaplan-Meier analysis of OS for combination therapy of standard treatment and bevacizumab vs. standard treatment alone in ovarian cancer patients stratified by NNMT expression was performed in GSE140082 dataset.

Detailed methods of statistical analysis were described in our previous published paper [16]. The expression of genes in GSE9891 and TCGA dataset were extracted from the curatedOvarianData R package. Since the present study was a single gene-centric study, we retained both the P value generated from student-t test and the adjusted P value generated from Limma R package for comparing gene expression between groups [17]. When using limma R package, the Benjamini-Hochberg multiple comparison adjustment was applied and the adjusted P value < 0.05 was considered statistically significant. “Low” and “High” were classified according to the gene expression level. The optimal cutoff values for NNMT, lactate dehydrogenase A (LDHA), and phosphoglycerate mutase 1 (PGAM1) expression were determined by the maximally selected rank statistics using R package survminer (http://​www.​sthda.​com/​english/​rpkgs/​survminer/​). Correlation analysis was performed by Pearson method. All the statistical analyses were performed using IBM SPSS Statistics and R. P values < 0.05 are considered statistically significant.

The mean expression value of NNMT was elevated in POCs/PTCs compared with that in FTs (Fig. 1a, GSE10971), and was increased in MOCs and ROCs compared with POCs (Fig. 1b, GSE30587; Fig. 1c, GSE44104). But the differences were not statistically significant suggested by adjusted P value generated from Limma R package. However, NNMT expression was significantly elevated in LVSI-positive ovarian cancers compared with LVSI-negative ovarian cancers (Fig. 1d, TCGA dataset, P < 0.05).

Next, we try to explore whether NNMT expression was correlated with clinical and molecular parameters using CSIOVDB database. We showed that increased expression of NNMT was associated with increased tumor stage (Fig. 2a), grade (Fig. 2b), and mesenchymal molecular subtype (Fig. 2c). The mean expression value of NNMT was elevated in drug-resistant ovarian cancers compared with drug-sensitive ovarian cancers (Fig. 2d), but the difference was not statistically significant suggested by adjusted P value generated from Limma R package.

Next, we evaluate the association of NNMT expression with the prognosis of ovarian cancer patients. Interestingly, survival analysis indicated that increased expression of NNMT was associated with a reduced OS in both GSE9891 dataset (Fig. 3a, HR: 2.28, 95%CI: 1.51–3.43, Log-rank P < 0.001) and TCGA dataset (Fig. 3b, HR: 1.55, 95%CI: 1.02–2.36, Log-rank P = 0.039). Multivariate analysis further confirmed the negative impact of NNMT expression on OS in ovarian cancer patients in both GSE9891 dataset (Table 1) and TCGA dataset (Table 2).

To quantitatively predict the survival of ovarian cancer patients, we developed a nomogram using GSE9891 dataset (training dataset) based on age at initial diagnosis, debulking status, FIGO stage, and NNMT expression, and validated it using TCGA dataset (validation dataset). Among the included factors, age at initial diagnosis had the largest contribution to OS, followed by FIGO stage and NNMT expression (Fig. 4). The C-index for predicting OS was 0.70 (95%CI, 0.64–0.75) in GSE9891 dataset (training dataset) and 0.64 (95%CI, 0.60–0.68) in TCGA dataset (validation dataset), respectively. The calibration plots demonstrated optimal agreement between the predicted and observed probability (Fig. 5a-f).

Then, we explore whether NNMT expression was predictive of response to bevacizumab usding GSE140082 dataset. Interestingly, it was found that bevacizumab conferred significant improvements in OS for patients with low NNMT expression (Fig. 6a, HR: 0.56, 95%CI: 0.31–0.99, Log-rank P = 0.049). In contrast, patients with high NNMT expression didn’t benefit from bevacizumab treatment significantly (Fig. 6b, HR: 0.85, 95%CI: 0.48–1.49, Log-rank P = 0.561).

Finally, we explore whether NNMT expression was correlated with the expression of genes (hexokinase 2 (HK2), LDHA, and PGAM1) involved in Warburg effect. Interestingly, our results suggested that NNMT expression was positively correlated with LDHA (Fig. 7a, r = 0.347, P < 0.001) and PGAM1 (Fig. 7b, r = 0.38, P < 0.001) expression, but not with HK2 expression (Fig. 7c, r = − 0.01, P = 0.905). The positive correlations of NNMT expression with the expression of LDHA (Fig. 7d, r = 0.15, P = 0.003) and PGAM1 (Fig. 7e, r = 0.30, P < 0.001) were confirmed in GSE140082 dataset. LDHA expression was not associated with bevacizumab response in ovarian cancer patients (Fig. 8a, Fig. 8b). However, bevacizumab conferred significant improvements in OS for patients with low PGAM1 expression (Fig. 8c, HR: 0.34, 95%CI: 0.12–0.94, Log-rank P = 0.037). In contrast, patients with high PGAM1 expression didn’t benefit from bevacizumab treatment significantly (Fig. 8d, HR: 0.80, 95%CI: 0.52–1.24, Log-rank P = 0.325).