Integrated mPD‐L1 and metabolic analysis identifies new prognostic subgroups in lung cancers with wild‐type EGFR

Dear Editor, Metabolic reprogramming, especially changes in glycolysis and cholesterogenesis pathways, has been reported to affect tumour prognosis.1 Interestingly, there is an intricate relationship between metabolic changes and immune checkpoints in the tumourmicroenvironment (TME),2 but their interaction and effects on the prognosis of lung cancer patients remain poorly understood. In this work, we established a novel stratification framework based on combined analysis of PD-L1 mRNA (mPD-L1) expression and glycolysis/cholesterol metabolic signatures, which stratified epidermal growth factor receptor (EGFR) wild-type lung cancers intometabolic subtypeswith significantly different prognoses. We also created a visualization website called glycolysis/cholesterol metabolism axis and PD-L1 mRNA expression (GCP) (https://www.liqlab.cn/gcp) for this stratification approach. We first investigated the impact of the three metabolic subtypes on prognosis at different PD-L1 expression levels. In the mPD-L1low group, cholesterogenic cases had a significantly worse OS and progression free survival (FPS) (mOS: 2.9 years; mFPS: 3.7 years) compared to glycolytic (mOS and mPFS were not reached) and quiescent cases (mOS: 5.5 years, mPFS was not reached). However, in the mPD-L1high group, cases belonging to the glycolytic subtype had a significantly worse OS and FPS (mOS: 3.3 years; mFPS: 2.3 years) than the cholesterogenic and quiescent subtypes (mOS: 7.1 years, mPFS: 7.2 years). Of note, no difference was observed among metabolic subtypes for the mPD-L1med group (Figure 1C and Table S4). Furthermore, for the cholesterogenic subtype, higher PD-L1 levels were associated with a better prognosis, whereas opposite findings were observed for the glycolytic subtype (Figure 1D). For the mPD-L1high group, univariate cox regression analysis revealed that glycolytic subtype, pT stage and pTNM stage were correlated with OS and PFS. During multivariate Cox regression analysis, glycolytic subtype (hazard ratio [HR], 2.62; 95% confidence interval [CI],

Dear Editor, Metabolic reprogramming, especially changes in glycolysis and cholesterogenesis pathways, has been reported to affect tumour prognosis. 1 Interestingly, there is an intricate relationship between metabolic changes and immune checkpoints in the tumour microenvironment (TME), 2 but their interaction and effects on the prognosis of lung cancer patients remain poorly understood. In this work, we established a novel stratification framework based on combined analysis of PD-L1 mRNA (mPD-L1) expression and glycolysis/cholesterol metabolic signatures, which stratified epidermal growth factor receptor (EGFR) wild-type lung cancers into metabolic subtypes with significantly different prognoses. We also created a visualization website called glycolysis/cholesterol metabolism axis and PD-L1 mRNA expression (GCP) (https://www.liqlab.cn/gcp) for this stratification approach.
We first investigated the impact of the three metabolic subtypes on prognosis at different PD-L1 expression levels. In the mPD-L1 low group, cholesterogenic cases had a significantly worse OS and progression free survival (FPS) (mOS: 2.9 years; mFPS: 3.7 years) compared to glycolytic (mOS and mPFS were not reached) and quiescent cases (mOS: 5.5 years, mPFS was not reached). However, in the mPD-L1 high group, cases belonging to the glycolytic subtype had a significantly worse OS and FPS (mOS: 3.3 years; mFPS: 2.3 years) than the cholesterogenic and quiescent subtypes (mOS: 7.1 years, mPFS: 7.2 years). Of note, no difference was observed among metabolic subtypes for the mPD-L1 med group ( Figure 1C and Table S4). Furthermore, for the cholesterogenic subtype, higher PD-L1 levels were associated with a better prognosis, whereas opposite findings were observed for the glycolytic subtype ( Figure 1D).
For the mPD-L1 high group, univariate cox regression analysis revealed that glycolytic subtype, pT stage and pTNM stage were correlated with OS and PFS.  (Table 1). Overall, the above findings suggest that in EGFR wild-type non-small cell lung cancer (NSCLC), different metabolic subtypes have distinct prognostic outcomes based on PD-L1 expression levels. A more aggressive phenotype could be associated with predominantly cholesterogenic tumours than those with predominantly glycolytic phenotype in EGFR wild-type NSCLC poorly expressing PD-L1.
We subsequently focused on the mPD-L1 low group of EGFR wild-type lung cancers. Analysis of clinical characteristics showed that cholesterogenic cases were more likely to have a smoking history and higher T and N scores ( Figure S3A). For immune profiles, significantly lower immune scores and lower tumour infiltration of endothelial cells, macrophages and B cells were found in the cholesterogenic subtype than in the quiescent subtype ( Figure S3B,C). Higher expression levels of CD274 and TIGIT were observed in cholesterogenic cases than glycolytic and quiescent cases ( Figure S3D). We previously found hypermethylation and low expression of multiple tumour suppressor genes (TSGs) in EGFR wild-type NSCLC poorly expressing PD-L1. 3 In the present study, the expression of TSGs ADAMTS8 (Adam metallopeptidase with thrombospondin type 1 motif 8), CDO1 (cysteine dioxygenase type 1), and GATA5 (GATA binding protein 5) varied across metabolic subtypes. This finding suggested possible heterogeneity in carcinogenic mechanisms across metabolic subtypes ( Figure S3E).   Interestingly, mutation analysis revealed that the mutation frequency and types varied significantly across the metabolic subtypes. Multiple genes likely involved included TP53, TTN, ZFHX4, ROS1, DNAH9, PCDH15, ALK and KRAS (Figure 2A-C). Next, cancer hallmarks analysis showed that tumour proliferation signature, G2M checkpoint hallmark, MYC targets hallmark and mRNAsi signatures were more active in cholesterogenic cases than glycolytic cases ( Figure 2D). The association between different cancer hallmarks further suggested that each  Figure 2E). We also explored the expression of some therapeutic targets. We noticed that the mitochondrial pyruvate carrier 1 (MPC 1) gene expression level was profoundly lower in the glycolytic subtype versus the cholesterogenic subtype ( Figure 2F). 4 Recently, methylenetetrahydrofolate dehydrogenase 2 (MTHFD2) was found to promote PD-L1-mediated tumour immune resistance. Proprotein convertase subtilisin/keying type 9 (PCSK9) has been documented to block intratumoural infiltration by T cells. 5,6 In the present study, we found that cholesterogenic cases expressed significantly higher levels of MTHFD2 and PCSK9 than other cases ( Figure 2G). Furthermore, as shown in Figure 2H, 10 pathways were significantly enriched in the cholesterogenic group (false discovery rate (FDR) < 0.05).
We then evaluated the stratification framework in other cancer types. Network topology analysis was used to identify co-expressed pathway-specific genes in 12 cancer types ( Figure 3A). In mPD-L1 high groups of other cancer types, significant differences in survival were observed for bladder cancer patients (log-rank p = 0.024) as glycolytic cases exhibited favorable OS (mOS was not reached). In mPD-L1 low groups, significant differences in survival across the metabolic subtypes were observed in Kidney renal clear cell carcinoma (kidney renal clear cell carcinoma (KIRC) log-rank p = 0.0013) and thyroid carcinoma (THCA) patients (log-rank p = 0.0095). For KIRC, cholesterogenic cases had better OS (mOS was not reached), whereas glycolytic (mOS: 3.5 years) and quiescent cases (mOS: 4.1 years) had worse OS (mOS: 2 years). For THCA, cholesterogenic and quiescent cases had better OS (mOS was not reached), and glycolytic cases (mOS was not reached) had relatively worse OS ( Figure 3A-C). Interestingly, the cholesterogenic subtype was an independent clinical factor only in KIRC (Table S5). Altogether demonstrated that despite the distinct genomic signatures and TME factors specific to each cancer type, tumour metabolic dependencies varied with PD-L1 expression levels.
To make the stratification framework more convenient and user-friendly, we also created an online tool called GCP (https://www.liqlab.cn/gcp), where investigators can submit transcriptome data of their own samples and obtain the stratification results with a single click ( Figure 3D and Figure S4).
In summary, our stratification framework indicates the role of metabolic phenotype in the prognosis of EGFR wildtype lung cancer, which is also related to the expression of PD-L1 ( Figure 3E). Moreover, this can help improve the current management of EGFR wild-type lung cancer patients. Meanwhile, it also provided clues for the selection of candidate drugs for combination treatment strategy using PD-1/PD-L1 inhibitors in these two subgroups of EGFR wild-type patients (mPD-L1 low /cholesterogenic; mPD-L1 high / glycolytic).

A C K N O W L E D G E M E N T S
This work was partially supported by NCI 1R01CA230339-01, 1R37CA255948-01A1 subaward and The Outstanding Clinical Discipline Project of Shanghai Pudong. The authors would like to thank Jianyu Wang and Yun Lu for their valuable help on the statistical advice and Yue Zhang for her advice on the web interface.