A novel DNA damage repair-related signature for predicting prognositc and treatment response in non-small lung cancer

Abstract

DNA damage repair (DDR) is essential for maintaining genome integrity and modulating cancer risk, progression, and therapeutic response. DDR defects are common among non-small lung cancer (NSCLC), resulting in new challenge and promise for NSCLC treatment. Thus, a thorough understanding of the molecular characteristics of DDR in NSCLC is helpful for NSCLC treatment and management. Here, we systematically analyzed the relationship between DDR alterations and NSCLC prognosis, and successfully established and validated a six-DDR gene prognostic model via LASSO Cox regression analysis based on the expression of prognostic related DDR genes, CDC25C, NEIL3, H2AFX, NBN, XRCC5, RAD1. According to this model, NSCLC patients were classified into high-risk subtype and low-risk subtype, each of which has significant differences between the two subtypes in clinical features, molecular features, immune cell components, gene mutations, DDR pathway activation status and clinical outcomes. The high-risk patients was characterized with worse prognosis, lower proportion and number of DDR mutations, unique immune profile and responsive to immunetherapy. And the low-risk patients tend to have superior survival, while being less responsive to immunotherapy and more sensitive to treatment with DNA-damaging chemotherapy drugs. Overall, this molecular classification based on DDR expression profile enables hierarchical management of patients and personalized clinical treatment, and provides potential therapeutic targets for NSCLC.

Keywords: DNA damage repair-based prognostic signature; NSCLC classification; biomarker; non-small lung cancer (NSCLC); tumor microenvironment.

Figure 1

Study flow chart.

Figure 2

Identification of two NSCLC subclasses based on the prognostic DDR-related genes. (A) Forest map of prognostic DDR-related genes. (B) Kaplan–Meier (KM) survival curves with transcriptomic profile of prognostic DDR-related gene. (C, D) Identification of NSCLC subclasses with PAC clustering using 17 prognostic DDR-related genes in the training and testing sets. (E, F) The CDF plot shows a flat middle segment for K = 2 in the training and testing sets. (G, H) OS of both subclasses (C1 and C2) in the training and testing sets. (I, J) PCA showing the distribution of the two NSCLC subclasses in the training and testing sets.

Figure 3

The comparisons of the clinical features between two NSCLC subtypes. (A–H) Analysis of clinical features between NSCLC subclasses by Fisher’s exact test. (I, J)Comparison of mutation alterations among subtypes in the training set. OncoPrint of mutation status of 15 shared genes among the top 20 in C1 (I) and C2 (J). (K) Comparison of the absolute mutated number between NSCLC subclasses. Distribution of somatic copy number alteration in C1 (L) and C2 (M).

Figure 4

Comparison of the molecular characteristics between the NSCLC subtypes. (A) Heatmap of top 100 DEGs expressed in NSCLC subtypes as annotated by clinical features. (B) Venn diagram showing the intersection between the prognosis-related DDR genes and DEGs in the training set. (C) KEGG and (D) GO analysis results of DEGs between the NSCLC subclasses in the training set. (E) Comparison of the DDR pathway and key DDR pathway genes between the NSCLC subclasses. Heatmaps were generated to show the biological processes, in which red indicates activation status and blue indicates inhibition status. (F) Comparison of expression differences of key DDR genes in the two subclasses. *,P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, no significance.

Figure 5

Characterization of the immune infiltration landscape of the two NSCLC subtypes. (A–F) Box plots of immune score, stromal score, and total score for NSCLC subclasses derived from ESTIMATE. The lines within boxes on box plots represent median values. The bottom and top lines represent the minimum and maximum values, respectively. (G) Heatmap showing the proportion of 28 types of infiltrating immune cells in C1 and C2 by ssGSEA algorithms. (H) Comparison of the proportion of 28 types of infiltrating immune cells between the subtypes by the rank sum test. NS denotes P>0.05, * indicates P ≤ 0.05, ** represents P ≤ 0.01, *** indicates P≤ 0.001, and **** indicates P≤0.0001.

Figure 6

Comparison of immune checkpoint molecules expression and TIDE scores between subtypes. Comparison of the expression of immune checkpoint genes in two subtypes in the training set (A) and testing set (B). (C) TIDE analysis. (D) The expression score of the response prediction biomarker, IFNG, was computed using TIDE analysis. NS denotes P>0.05, * indicates P ≤ 0.05, ** represents P ≤ 0.01, *** indicates P≤ 0.001, and **** indicates P≤0.0001.

Figure 7

Construction of a prognostic-related signature by the LASSO regression model. (A) Selection of the optimal λ-value through a 10-fold cross-validation. (B) The fitting processes for LASSO Cox regression models were constructed from the six signature genes. The tuning parameter λ was derived from the partial likelihood deviance with a 10-fold cross-validation, and the coefficient was plotted against Log(λ). The six gene signature was identified based on the best fit profile. (C) LASSO coefficient profiles of the six key prognostic DDR-related genes. (D) Forest plot of prognostic DDR‐related genes based on univariate Cox regression analysis. (E) The heatmap was scaled with the Z-Score using the log2(FPKM+1) expression of signature genes. With the increase in the risk scores, CDC25C, NEIL3, H2AFX, NBN, and XRCC5 expression levels were upregulated and RAD1expression level was decreased.

Figure 8

Validation of six DDR-related signatures by the LASSO regression model. (A) Distribution of risk scores in NSCLC subclasses. (B - D) Kaplan-Meier OS analysis of the prognostic DDR-related signature in NSCLC and NSCLC subtypes. (E) Evaluation of the prognostic values for different clinicopathological characteristics (age, smoking_year, gender, stage, T staging, N staging, M staging, and EGFR_mutation) as well as a risk score using univariate Cox regression analysis. (F) Multivariate Cox regression analysis was used to test the independence of the risk score and other factors for predicting the prognosis of NSCLC.

Figure 9

mRNA and protein expression levels of the prognostic DDR-related signature genes in NSCLC. (A) The mRNA expression of CDC25C, NEIL3, H2AFX, NBN, XRCC5, and RAD1 in normal lung and NSCLC (LUAD and LUSC) tumor tissues via the GEPIA. *p < 0.05; **p < 0.01. (B) Immunohistochemistry analysis showing the protein expression of the signature genes in normal lung and NSCLC tumor tissues obtained from the HPA database (data for NEIL3 were not available). (C) Immunofluorescence images of CDC25C, H2AFX, NBN, XRCC5, and RAD1 in cells with green, blue, and red indicating target proteins, nuclei, and microtubules, respectively.

Figure 10

Effects of CDC25C knockdown on proliferation and chemosensitivity in NSCLC cells. (A) Drug response analysis by Cancer Treatment Response gene signature DataBase (CTR-DB). (B) Immunofluorescence was performed 72 h after transfection. (C) Expression of CDC25C mRNA by qRT-PCR in A549 and NCl-H1299 cells transfected with shRNA-NC and shRNA-CDC25C. (D) Plate colony formation assay (n=3). (E) CCK8 detection of the proliferation of shRNA-CDC25C-transfected A549 and NCl-H1299 cells (n=4). And all data are illustrated as mean ± SD;.NS, no significance; *p<0.05; **p < 0.01; ***p < 0.001; and ****p < 0.0001. (F) Drug sensitivity of Paclitaxel was represented by the half-maximal inhibitory concentration (IC50) (n = 4).