Introduction

The treatment paradigm for advanced non-small cell lung carcinoma (NSCLC) has significantly shifted in the last decade with the incorporation of checkpoint inhibitor immunotherapy and targeted therapies to treat tumors with specific molecular alterations. Despite these significant clinical advances, the median 5-year overall survival (OS) remains low at 25%1. The prognosis for advanced KRAS-mutant NSCLC is even worse and has remained poor for decades, with a median survival of 1.2 years2. KRAS-driven NSCLC is a distinct molecular subtype of NSCLC, that has been a historically challenging to therapeutically target, and even deemed “undruggable”3. Patients with KRAS-mutant NSCLC harboring co-mutations in STK11 and/or KEAP1 have been associated with the use of tobacco products and are notorious for being aggressive and resistant to treatment4. STK11 and KEAP1 mutations have been shown to confer primary resistance to anti-PD-L1/PD-1 immune checkpoint inhibitor (ICI) therapy in the presence of KRAS mutation4, although the combination with CTLA4 blockade shows promising results5. Despite sotorasib and adagrasib being approved for advanced NSCLC with KRAS G12C mutation in patients who have received at least one prior line of therapy6, outcomes remain poor. Therefore, ongoing efforts continue to test for combination therapies to overcome inhibitor resistance and to test for novel KRAS allele-specific inhibitors. The role of STK11 and KEAP1 co-occurring mutations on primary response rates, duration of response and evolution of resistance remains a principal area of ongoing research and an area of unmet need for new therapeutic options7.

We previously reported that NSCLC with STK11 and/or KEAP1 mutations in a KRAS mutant background (but not KRAS mutant cells lacking these co-mutations) are sensitive to inhibition of N-myristoyltransferases (NMTs)8. Preclinical stage inhibitors of NMTs (which target NMT1 and NMT2) have also shown efficacy against experimental models of lymphoma9 and are currently in stage 2 clinical trial10. We and others have shown that elevated mRNA expression of NMT1 (the most abundant NMT enzyme in most tissues) correlates with decreased OS (overall survival) in lung carcinoma11,12.

We reported that NMT inhibition in lung carcinoma cells decreases the expression of the mitochondrial import protein TIM17A, but only those cells that are sensitive to NMT inhibition were also sensitive to loss of TIMM17A expression12. Accordingly, we found a significant correlation between sensitivity to NMT inhibition and TIM17A dependency in lung carcinoma cells, suggesting a link between both genes in lung adenocarcinoma12. We hypothesized that NMT1-TIMM17A is a novel vulnerability axis that may represent a specific target for therapy of lung carcinomas with STK11 and/or KEAP1 mutations in a KRAS mutant background. To further explore the unexpected relationship between NMT1 and TIM17A, we used data from The Cancer Genome Atlas (TCGA) to investigate the association of NMT1 and TIMM17A mRNA transcript expression levels with overall survival (OS), disease-specific survival (DSS) and disease-free survival (DFS) in a lung adenocarcinoma (LUAD) cohort.

Materials and methods

The Cancer Genome Atlas—Lung Adenocarcinoma (TCGA-LUAD) gene expression data and associated clinical data were queried from cBioPortal Pan Cancer Atlas. A total of 486 patients were linked to RNA-sequencing data (units: normalized counts). Demographic information included age, gender, smoking habits, patient performance status (ECOG, Karnofsky Score), and spirometry data. Clinical data included initial tissue weight, status of prior cancer diagnosis, primary tumor location, tumor pathological stage, and treatment administered such as radiation therapy or post-operative adjuvant therapy. The provided hypoxia scores had been calculated from transcriptomic data. The provided aneuploidy score, MSI Mantis Scores, MSI Sensor Score, total mutation, and total mutation burden score had been calculated from DNA mutation data. Comparing the distribution of continuous variables between groups (e.g. age distribution in high versus low TIMM17A) was performed using student t-test. Comparing the distribution of categorical variables between groups (e.g. gender distribution in high versus low TIMM17A) was performed using Fisher’s exact test.

Survival analysis was performed using Cox regression models. Association between clinical endpoints and gene expression data were assessed using hazard ratios (HR) and their 95% confidence intervals (CI). Gene expression data were tested as both a continuous variable (log-transformed and scaled) and a categorical variable (split dichotomously based on assessment as continuous variable). The complete model was adjusted for gender, age, tumor pathological stage, and KRAS/STK11/KEAP1 mutation status. The fit for various steps of the multivariable modeling with terms added versus removed were tested using ANOVA.

To visualize how long a patient would survive for a given continuous value of gene expression, restricted mean survival (RMS) curves were used. More specifically, RMS curve plots the Cox-model estimated years of survival for every nth quantile of gene expression. Next, the normalized gene count values were dichotomized into high versus low expression groups using an optimal cutoff determined by log-rank tests evaluated at every 10th percentile. To visualize the percentage of patients surviving over time for the categories of low and high gene expression, Kaplan-Meier curves were used.

Validation dataset used lung adenocarcinoma cases from the Singapore Oncology Data Portal (OncoSG, n = 172 with OS-events = 44). Pre-processed transcriptomic data (STAR-aligned RSEM raw counts), mutational data as well as associated clinical and demographic data such as overall survival, tumor staging, age, gender, and smoking status were queried from their portal. By histological subtype, the majority of OncoSG lung cancers were acinar (87/172–49.3%) with some solid (18/172–9.9%) and papillary (19/172–11.7%) histological types; TCGA, on the other hand, were mostly either mixed (120/515–23.3%) or not specified (320/515–62.1%), leaving few that were acinar (18/515–3.5%) and solid (5/515–1.0%), and none that were labelled papillary. The OncoSG cohort consisted of individuals of East Asian Ancestry with fewer current smokers (36% vs. 85% in TCGA) and less overall mutational burden (92 vs. 312 in TCGA). Thus, for comparability to TCGA dataset, the OncoSG validation cohort analyzed only those who have ever-smoked (n = 62, OS-events = 20). Multivariable cox-regression models adjusted for age (mean ± SD: 64.1 ± 10.0), gender (n: Male = 49, Female = 13), and tumor stage (n: I = 35, II = 11, III = 13, IV = 1).

Results

High NMT1 and TIMM17A were associated with poor overall survival in lung adenocarcinoma

Pursuing the idea that NMT1-TIMM17A is a novel vulnerability axis of lung carcinomas with STK11 and/or KEAP1 mutations in a KRAS mutant background, we tested their association to OS in the TCGA-LUAD cohort using Cox-regression models. These initial models adjusted for age and gender show that the expression of NMT1 and TIMM17A were each associated with reduced OS. Analyzed as a continuous variable, a one standard deviation increase in NMT1 (log-normalized count) was associated with 17% higher mortality rate (P = 0.046, HR = 1.17 [95% CI 1.00, 1.37]) and TIMM17A was associated with 15% higher mortality rate (P = 0.047, HR = 1.15 [95% CI 1.00, 1.33]) (Fig. 1A); the RMS curve shows that the Cox-model estimated years of survival decreases (y-axis) as the percentile of gene expression increases (x-axis).

Fig. 1
figure 1

High expression of NMT1 or TIMM17A was significantly associated with worse overall survival in TCGA-LUAD. TCGA-LUAD (Lung Adenocarcinoma) RNA-seq expression for NMT1 and TIMM17A (normalized count) were queried from the Broad GDAC Firehose standardized pipelines. Hazard ratios (HR), their 95% confidence intervals (CI), and P-values were calculated using a cox regression model adjusted for age and gender. (A) Analyzed as a continuous linear variable: RMS curves (restricted mean survival) shows the cox model estimated overall survival years (y-axis) and 95% confidence band associated with nth quantile expression of reported gene (x-axis). Dashed lines report the percentile cutoffs used and the equivalent count values used to split the into dichotomous categories. (B) Analyzed as a dichotomous categorical variable Kaplan-Meier survival curves comparing the percent surviving (y-axis) over time (years, x-axis) between group with low expression (green) versus high expression (red). Shading in respective colors show the 95% confidence interval.

Full size image

Next, for easier interpretability, normalized count values were grouped into categorizes of high versus low gene expression based on an optimal cutoff determined by log-rank tests evaluated at every 10 percentiles (NMT1 cutoff: 80th percentile or 3062 count; TIMM17A cutoff: 50th percentile or 1234 count). Those in the group with high NMT1 had 52% higher mortality rate compared to the low group (P = 0.017, HR = 1.52 [95% CI 1.08, 2.14]), and those with high TIMM17A had 60% higher mortality rate compared to the low group (P = 0.003, HR = 1.60 [95% CI 1.18, 2.17]) (Fig. 1B).

Individual and combined effect of NMT1 and TIMM17A on OS, DSS, and DFS

Having shown that NMT1 and TIMM17A were each associated with worse OS, further models tested their individual and combined association to additional clinical endpoints such as DSS and DFS. Based on the same dichotomization thresholds as for the OS models above, high TIMM17A was also associated with reduced DSS and DFS (P = 0.023 and 0.030 respectively), whereas NMT1 showed no association to either DSS or DFS (Table 1).

Table 1 Association between NMT1, TIMM17A, and the combination of the two on LUAD outcome data. Hazard ratios and confidence intervals for overall survival, disease specific survival, and disease-Free survival for the various categorical splits (rows). Whereas TIMM17A was independently associated with overall survival, disease specific survival, and disease-Free survival, NMT1 was only associated with overall survival. The presence of high NMT1 together with high TIMM17A expression significantly had a significant association with worse OS (P = 0.020 as compared to TIMM17A alone).
Full size table

In combination, the categories of NMT1 and TIMM17A expression had no significant interaction effect on clinical endpoints, meaning their effect on outcome is not modified by or dependent on each other (P = 0.87, 0.43, 0.67 for OS, DSS, and DFS respectively). Testing NMT1 alongside TIMM17A in a Cox proportional hazards model showed improve fit for OS (P = 0.020), indicating that the two genes were independently associated with worse OS (Table 1). To explore their combination further, we tested NMT1 effect on survival across low versus high stratification of TIMM17A (Supplementary Fig. S1). We found that NMT1 was significantly associated with worse OS in the subgroup with also high TIMM17A (P = 0.016; HR = 1.71 [95% CI 1.10, 2.64]), but not in the group with low TIMM17A (P = 0.33; HR = 1.32 [95% CI 0.75, 2.33]). The effect of NMT1 on survival was pronounced in the high expression stratum of TIMM17A, where it was significantly associated with worse survival. However, the effect size was not different between the two strata, thus the non-statistically significant interaction test results above. Significant association in only one of the strata could be attributed to either loss of statistical power with stratification, confounding variables within one stratum (e.g. KRAS/KEAP1/STK11 mutations with TIMM17A), or possibly that NMT1 association with worse overall survival also depends on having high TIMM17A. Overall, this suggests that high NMT1 and high TIMM17A each contribute additively to the risk of death; NMT1 association with worse OS is pronounced when TIMM17A is also elevated.

Next, we leveraged the improved fit from combining NMT1 and TIMM17A, assessing their effect together on OS, DSS, and DFS. Those with high expression of both NMT1 and TIMM17A (High–High) had worse OS, DSS, and DFS compared to those with low expression of both (Low–Low) (Fig. 2) (P: <0.001, 0.023, and 0.049 respectively) (HR and 95% CI: 2.43 [1.52, 3.89], 2.10 [1.11, 3.99], and 1.71 [1.00, 2.91] respectively). Thus, patients with high co-expression of NMT1 and TIMM17A had worse OS, DSS, and DFS when compared to those with low co-expression of NMT1 and TIMM17A.

Fig. 2
figure 2

Elevated co-expression of TIMM17A and NMT1 associates with worse OS, DSS and DFS. Compared to the group with both Low TIMM17A and Low NMT1 (Black), those with both High TIMM17A and High NMT1 (Yellow) showed worse overall survival (Left), disease-specific survival (Middle), and disease-free survival (Right) in LUAD. Shading with respective coloring displays 95% confidence interval for percent surviving (y-axis) across time (x-axis: years). P-values are calculated from initial cox regression models adjusted for age and gender with hazard rates detailed in Table 1. NMT1 was split at the 80th percentile and TIMM17A was split at the 50th percentile into dichotomous categorical variables based on an optimal cutoff tested at every 10 percentiles and their assessment as continuous variables in Fig. 1.

Full size image

Building multivariable cox regression model

We analyzed the composition of the TCGA-LUAD cohort to determine whether there were any factors confounding the TIMM17A-NMT1 effect on outcome. We explored the distribution of demographics, mutation profiles, and clinical factors across low versus high levels of TIMM17A and NMT1, and also assessed how these variables might independently be associated with OS (Table 2). Potentially confounding variables would be those unevenly distributed between TIMM17A-NMT1, and also associated with OS. A higher percentage of women had low TIMM17A expression versus high (61.1% vs. 46.5%, P = 0.001), though the gender variable was not significantly associated with overall survival. Smoking shows trend towards association with TIMM17A (P = 0.093) and was associated with NMT1 (P = 0.018); however, smoking was not significantly associated with overall survival. Fraction mutation altered (percentage of genome affected by copy number gains or losses) was associated with high TIMM17A and NMT1 (P = 0.022 and 0.002 respectively); however, it was not associated with overall survival (P = 0.53). High TIMM17A was associated with higher incidence of KEAP1 mutation (P = 0.001), STK11 mutation (P = 0.009), and associated with having KEAP1 and/or STK11 along with KRAS mutation (P = 0.009). STK11 mutation trended with poor OS (P = 0.090). Consistent with previous studies, STK11 mutation in combination with KRAS mutation was associated with worse OS (P = 0.047). Hypoxia score (gene signature score calculated from transcriptomic data) may be a confounder, as it was higher with high TIMM17A and high NMT1 in addition to being associated with worse overall survival (all P < 0.001) (Fig. 3). Interestingly, a higher Karnofsky Performance Scale at diagnosis was associated with improved survival but was unexpectedly associated with high NMT1.

Table 2 Demographic, mutation profiles, and clinical factors associated with NMT1, TIMM17A, and overall survival. Distribution of demographics, mutation profiles, and clinical factors across low versus high levels of TIMM17A and NMT1, and their association to overall survival. Comparing the distribution of continuous variables between groups (e.g. age distribution in high versus low TIMM17A) was performed using student t-test. Comparing the distribution of categorical variables between groups (e.g. gender distribution in high versus low TIMM17A) was performed using Fisher’s exact test. Highlighted in bold are P < 0.10 and in Italic are P < 0.05.
Full size table
Fig. 3
figure 3

Association between different factors of interest with TIMM17A and NMT1. Potentially confounding variables would be those unevenly distributed between TIMM17A-NMT1, and also associated with OS. High TIMM17A was associated with higher incidence of KEAP1 mutation (P < 0.001), STK11 mutation (P = 0.002), and Buffa hypoxia score (P < 0.001). NMT1 was also associated with higher hypoxia score (P < 0.001). Buffa hypoxia score was associated with worse OS (P < 0.001). KEAP1 and STK11 shows trend towards association to OS (P = 0.114 and 0.090 respectively).

Full size image

To investigate the effect of these various factors, we iteratively added and removed terms from our regression model to assess how each may relate to OS (Table 3). Our initial model included both gender and age based on common practice, though neither significantly contributed to a better model fit (P = 0.78 and 0.45 respectively). The next set of models included tumor pathological staging (known confounder) to evaluate NMT1 and TIMM17A as independent factors to predict survival. We found that whether stage was added as an ordinal variable or a categorical variable (I/II vs. III/IV), high expression of both TIMM17A and NMT1 remained associated with reduced OS. These findings suggest that the prognostic effects of TIMM17A and NMT1 on survival are not explained by tumor stage and may serve as independent markers for poor prognosis. The fourth model with hypoxia score confounded with TIMM17A and NMT1 with some multicollinearity, and the two genes of interest were no longer statistically significant. Adding hypoxia score resulted in redundancy in the information each variable provides to the model, suggesting that TIMM17A and NMT1 are either related to or influenced by hypoxia score. Much of the prognostic effect of TIMM17A and NMT1 is explained by the hypoxic status of the tumor. Finally, the complete model included KRAS/STK11/KEAP1 mutation to isolate the effect of TIMM17A and NMT1 on survival. From Table 2, the prevalence of KRAS co-mutation with KEAP1/STK11 were associated with high TIMM17A (P = 0.009) as well as worse OS (P = 0.047), and therefore is a potential confounder. The complete model showed that both high TIMM17A and high NMT1 were still associated with worse OS (TIMM17A: P = 0.005, HR = 1.56 [95% CI 1.14, 2.14] and NMT1: P = 0.023, HR = 1.51 [95% CI 1.06, 1.14]). KRAS, STK11, and KEAP1 mutations, on the other hand, were not associated with survival. While STK11 mutation did not reach statistical significance in the complete model (P = 0.069), it showed a trend towards worse OS, indicating that STK11 mutation may still have a potential role in influencing survival outcomes. Due to low frequency of these mutations (e.g. n = 36 for STK11), the analysis lacks statistical power to determine whether these mutations or the combination of them mediates the association between TIMM17A, NMT1, and survival. Ultimately, the full model adjusting for potential confounders showed that high expression of TIMM17A and of NMT1 were independently linked to worse OS. The combination of high-high expression of TIMM17A and NMT1 had 240% higher rate of death compared to those with low-low expression (P < 0.001, HR = 2.40 [95% CI 1.51, 3.85], not shown).

Table 3 Building multivariable Cox regression with TIMM17A and NMT1 on overall survival. The complete model adjusted for gender, age, tumoral stage, and presence of KRAS, STK11, KEAP1 mutations showed that both high TIMM17A and high NMT1 were associated with worse OS (TIMM17A: P = 0.005, HR = 1.56 [95% CI: 1.14, 2.14] and NMT1: P = 0.023, HR = 1.51 [95% CI: 1.06, 1.14]).
Full size table

Validating results in OncoSG ever-smokers

Since the TCGA cohort comprise of 85% ever-smokers, results were validated using the ever-smokers of the OncoSG database. Similar to the findings in the TCGA, the OncoSG dataset also showed that higher NMT1 and higher TIMM17A were each associated with worse overall survival after adjusting for age, gender, and tumor stage (Supplementary Fig. S2; NMT1: P = 0.033, HR = 2.89 [95% CI 1.09, 7.67] and TIMM17A: P = 0.023, HR = 3.67 [95% CI 1.20, 11.20]). Likewise, when we leveraged the improved fit from combining NMT1 and TIMM17A, high-high compared to low-low expression of NMT1 and TIMM17A together was associated with worse OS for ever-smokers in both the OncoSG and TCGA datasets, even after adjusting for age, gender, and tumor stage (Fig. 4; OncoSG: P = 0.031, HR = 4.66 [95% CI 1.15, 18.93] and TCGA: P = 0.033, HR = 2.89 [1.09, 7.67]).

Fig. 4
figure 4

Elevated co-expression of TIMM17A and NMT1 associates with worse OS in the smoking subset of the OncoSG and TCGA dataset.

Full size image

Discussion

Here, we report that the combination of both high NMT1 and high TIMM17A was associated with poor prognosis (worse OS, DSS, and DFS) in lung adenocarcinoma. We validate previous reports that NMT111,13 and TIMM17A13 are each independently associated with reduced OS. We additionally found that TIMM17A expression is associated with reduced DSS and DFS. Interestingly, NMT1 was associated with reduced OS, but only in the subset with also high TIMM17A (which is also the subset associated with higher prevalence of STK11 and KEAP1 mutations). This is consistent with our previous observations that NMT1 inhibition is particularly effective against lung carcinomas with STK11 and/or KEAP1 mutations in a KRAS mutant background12, and that the sensitivity of lung carcinoma cells to NMT inhibition correlates with TIM17A dependency12. We found that assessing both NMT1 and TIMM17A together revealed a pronounced reduction in OS, DSS, and DFS. Co-overexpression of NMT1 and TIMM17A is a predictor of poor survival and a potentially targetable pathway to treat KRAS STK11 and KEAP1-mutant NSCLC.

Given the TCGA-LUAD cohort predates and is not confounded by the incorporation of ICI in treatment of NSCLC, our survival analysis could not investigate how immunotherapy might alter the association between NMT1 and TIMM17A and survival. Further limitation is that the cohort mostly consists of Caucasian from United States, who are more likely to have a history of current or former tobacco use (representing 85% of the total cohort). As such, the TCGA cohort’s tumor samples are expected to have higher tumor mutational burden and the total number of mutations in the tumor are strongly associated with smoking. This and other population differences may contribute to differences at the mRNA expression levels, which are relative to each study. The results were validated with the smoking subset of the OncoSG dataset, which comprised of individuals from East Asia; however, studying a larger non-smoking Caucasian cohort as well as studying other ethnic groups will help widen the scope and generalizability of these results. Lastly, the number of mutations of KRAS, STK11, and KEAP1 in lung cancer were few, even in the TCGA cohort, which hinders the ability to detect the effect of these mutations on survival. An even larger sample size would be necessary to robustly assess how these mutations may mediate TIMM17A, NMT1, and outcome.

A strength of this study was in using an updated dataset as opposed to a provisional version13, thus having a larger sample size (n = 486 versus 258) with longer follow-up periods. Additionally, we show that NMT1 was associated to overall survival not only as a categorical variable, but also as a continuous variable; splitting gene expression into dichotomous high versus low simplifies the relationship, but analysis as a continuous variable preserves the data and reveals a linear rise in risk with rise in NMT1 expression level. Furthermore, we illustrate how assessing both NMT1 and TIMM17A together shows a pronounced difference in not only OS, but also DSS and DFS. This is surprising given the fact that TIM17A is not a myristoylation target of NMT1. Lastly, our statistical model highlights the value of considering both NMT1 and TIMM17A expression as independent predictors of survival that are beyond what could be modeled using pathological tumor staging. Our findings warrant further studies to understand the relevance of the novel NMT1-TIMM17A axis in the progression and treatment of KRAS/STK11/KEAP1 NSCLC.