Tumor MK2 transcript levels are associated with improved response to chemotherapy and patient survival in non-small cell lung cancer

Abstract

Non-small cell lung cancers (NSCLCs) demonstrate intrinsic resistance to cell death, even after chemotherapy. Previous work suggested defective nuclear translocation of active caspase-3 in observed resistance to cell death. We have identified mitogen-activated protein kinase-activated protein kinase 2 (MK2; encoded by the gene MAPKAPK2) is required for caspase-3 nuclear translocation in the execution of apoptosis in endothelial cells. The objective was to determine MK2 expression in NSCLCs and the association between MK2 and clinical outcomes in patients with NSCLC. Clinical and MK2 mRNA data were extracted from two demographically distinct NSCLC clinical cohorts, North American (The Cancer Genome Atlas, TCGA) and East Asian (EA). Tumor responses following first round of chemotherapy were dichotomized as clinical response (complete response, partial response, and stable disease) or progression of disease. Multivariable survival analyses were performed using Cox proportional hazard ratios and Kaplan-Meier curves. NSCLC exhibited lower MK2 expression than SCLC cell lines. In patients, lower tumor MK2 transcript levels were observed in those presenting with late-stage NSCLC. Higher MK2 expression was associated with clinical response following initial chemotherapy and independently associated with improved 2-yr survival in two distinct cohorts, 0.52 (0.28–0.98) and 0.1 (0.01–0.81), TCGA and EA, respectively, even after adjusting for common oncogenic driver mutations. Survival benefit of higher MK2 expression was unique to lung adenocarcinoma when comparing across various cancers. This study implicates MK2 in apoptosis resistance in NSCLC and suggests prognostic value of MK2 transcript levels in patients with lung adenocarcinoma.

NEW & NOTEWORTHY MK2, known to promote caspase-3 nuclear translocation in the execution of apoptosis, is reduced in non-small cell lung cancer cells. In adenocarcinomas of patients, higher MK2 expression is associated with early-stage disease, better clinical response following chemotherapy and independently associated with improved survival.

INTRODUCTION

Cytotoxic chemotherapy remains a mainstay of treatment for many cancers (1, 2). However, chemotherapy resistance in lung cancer is a significant problem that potentially limits effective treatment for patients with lung cancers. Since tumor death following chemotherapy administration is thought to proceed, in part, via apoptosis, the molecular basis of apoptosis resistance in tumors is a major focus of investigation (1, 3, 4).

Apoptosis is a highly regulated process of cell death, mediated by caspases. Activation of the apoptotic cascade by external (extrinsic pathway) or internal (intrinsic pathway) stimuli leads to amplification of “death signals” via initiator caspases and ultimately cell dismantling proceeds via the executioner caspases (5). Caspase-3 is the major executioner caspase in eukaryotic cells, and activation of caspase-3 has traditionally been perceived as the terminal step of the apoptotic cascade and has signaled cell death (5–7). This has led to hegemony of the concept that caspase-3 activation is pathognomonic of apoptosis. However, recent reports have started to identify a disconnect between activation of caspase-3 and the execution of apoptosis (8–11).

Non-small cell lung cancer (NSCLC) is typically more resistant to chemotherapy-induced apoptosis compared with small cell lung cancer (SCLC) (3, 4). Previous work had sought to identify differences in activation of the apoptotic cascade between NSCLC and SCLC to explain the inherent differences in response to chemotherapy (3, 4). Joseph et al. (4) showed that NSCLCs activate the apoptotic machinery in response to chemotherapy, even to the point of activating caspase-3; however, active caspase-3 was sequestered in the cytoplasm of NSCLC cell lines. This was in contrast to SCLC cell lines, where active caspase-3 translocated to the nucleus, suggesting defective nuclear translocation of active caspase-3 as a mechanism of chemotherapy resistance in NSCLC(4).

Separately, our group identified circumstances where activation of caspase-3 did not result in apoptosis (9). Specifically, we observed that endotoxin induces caspase-3 activation, endothelial apoptosis, and endothelial permeability. However, with loss of the signaling molecule mitogen-activated protein kinase-activated protein kinase 2 (MK2; encoded by gene MAPKAPK2), there was prevention of both apoptosis and endothelial permeability; yet, to our surprise, caspase-3 activation was preserved. Under MK2-deficient conditions, caspase-3 was activated yet remained in the cytoplasm, preventing endotoxin-induced apoptosis. This cytoplasmic sequestering of caspase-3 under conditions of MK2 depletion mirrored findings of cytoplasmic caspase-3 localization in chemo-resistant NSCLCs.

Thus, we hypothesized NSCLCs have reduced basal expression of MK2 and higher MK2 levels may be associated with improved outcomes in patients with NSCLC.

METHODS AND MATERIALS

Cell Lines

NSCLC cell lines, H23 and A549, and SCLC cell line, H446, were purchased from ATCC (Manassas, VA) and maintained in full culture media as according to ATCC’s recommendations.

Immunoblot Analyses

Cell cultures were lysed using cell lysis buffer (CST 9803s, Cell Signaling, Boston, MA) supplemented with protease inhibitors cocktail (Sigma P8340), PMSF (1 mM, Thermo Fischer 36978), NaF (1 mM, Sigma, 201154), and NaOV (1 mM, Sigma, S6508). Protein lysates were denatured using Laemmli Sample buffer (Bio-Rad 1610747), 2-Mercaptoethanol (Millipore Sigma M6250), and 100°C heat (5-min exposure). Proteins were separated by SDS-PAGE (Thermo Fisher XP00122BOX) and transferred to PVDF membranes (Bio-Rad 1620177). Membranes were blocked in 5% nonfat dry milk (Bio-Rad 1706404) in TBS (Quality Biological 50983267) with 0.5% Tween-20 (Thermo Fisher BP337-500). Membranes were incubated with primary antibodies at 1:1,000 dilution overnight in 2.5% nonfat dry milk. Antibodies directed at MK2 (CST-3042) and β-tubulin (CST-5346) (Cell Signaling, Boston, MA) were used. PVDF membranes were then incubated with horseradish peroxidase-linked secondary antibodies, anti-rabbit (CST 7074), at 1:5,000 dilution for 1 h in 1% nonfat dry milk. Protein bands were visualized using chemiluminescent detection methods. Band intensities were quantified using ImageJ software.

Quantitative PCR

For quantitative PCR, we isolated total RNA from cells with TRIzol (ThermoFisher Scientific; Waltham, MA) and purified with a commercially available kit (QIAgen; Germantown, MD). Subsequently prepared cDNA (GE First-Strand cDNA Synthesis Kit; Niskayuna, NY) was used as the template for quantitative PCR using the Bio-Rad iQ5 thermal cycler. Primer sets for human Mapkapk2 (Cat. No. PPH01783A) and β-Actin (Cat. No. PPH00073G) were purchased from QIAgen. Background subtracted amplification data were analyzed using open-source software to estimate Ct values and amplification efficiency. Target gene expression was normalized to a reference gene using the comparative Ct method (12).

Clinical and mRNA Expression Accession

Clinical data for The Cancer Genome Atlas (TCGA) data sets were accessed using the TCGABiolinks package (13–15) in R/Bioconductor (16, 17) or the OncoLnc web interface (http://www.oncolnc.org).

Data Filtering/Processing

After downloading the lung adenocarcinoma (LUAD) data set, we parsed entries based on sample type (normal tissue and primary tumors). For tumor samples, we restricted our analyses to tumors identified on histology as adenocarcinoma-type, and to patients who had either a date of death or a date of last follow-up. Since mRNA transcript levels can vary based on processing, filtering, and normalization parameters, we wanted to verify the accuracy of mRNA transcript counts obtained from OncoLnc web interface. So, these values were compared with normalized RSEM-normalized transcript levels for the MAPKAPK2 gene that were downloaded and extracted from the full TCGA-LUAD transcriptome data set using the TCGABiolinks package (Supplemental Fig. S1). Since the downloaded TCGA-LUAD transcript data were already RSEM normalized, no additional normalizations were performed. Furthermore, since the values obtained from OncoLnc and those obtained directly from TCGA were exactly correlated, we chose to continue using transcript levels (for MAPKAPK2 as well as other genes) from OncoLnc given ease of access.

To validate our findings from the TCGA-LUAD, we identified an East Asian (EA) cohort of patients with lung adenocarcinoma (18, 19). For our validation cohort, MAPKAPK2 mRNA transcript levels (RSEM normalized) and clinical data for the validation cohort, EA cohort, were obtained via cBioportal (18, 19).

Proteomic and transcript data of lung adenocarcinoma patient tumors were obtained from Clinical Proteomic Tumor Analysis Consortium (CPTAC) (20). CPTAC proteomic data are already normalized as Z-scores of protein abundance ratio measured by mass spectrometry, calculated over all diploid samples. Similarly, CPTAC mRNA data are already normalized as Z-scores of mRNA abundance (i.e., log2 RPKM), calculated over all samples. No additional normalizations were performed after downloading.

In the TCGA-LUAD data set, we observed that MK2 levels were significantly lower in late-stage tumor samples. Thus, we opted to stratify our analyses by tumor stage (early vs. late stage). Such a baseline imbalance was not noted in the EA cohort; thus, we chose to include stage as a covariate in our multivariable analysis. In both TCGA and EA data sets, we initially modeled MK2 transcript levels as a continuous variable and noted an association between MK2 transcript levels and survival at 1 yr (Supplemental Material). However, we reasoned that examining MK2 transcript levels as “high” versus “low” may have more translational relevance in terms of patients with risk-stratifying early-stage NSCLC. We initially investigated a cutoff of the median to determine the association between “high” MK2 expression and survival. However, we posited a nonlinear relationship between MK2 expression and survival may provide rationale for a more appropriate cutoff (21). Using flexible general additive models using MK2 expression as a continuous variable, there was a nonlinear association between MK2 expression and survival in a combined TCGA and EA cohort. As such, we identified a cutoff at the part of the curve corresponding to the top 1/3 of MK2 expression, which appears to have a relatively constant slope. Thus, for TCGA-LUAD, models were run with 1) stratification by tumor stage and 2) dichotomization of MK2 as “high” (top 1/3) and “low” (bottom 2/3).

Choice of Model

We explored the association between MK2 transcript levels and survival using Cox Proportional Hazards model of time to death in 1 yr (right censored at 1 yr of follow-up), stratified by tumors stage, adjusted for sex, and smoking status, using MK2 as a dichotomous variable (high/low).

Censoring

Based on our prior work showing MK2 being necessary for the execution of apoptosis (9), we sought to investigate early time points to demonstrate a link between tumor cell death and survival. To that end, we initially chose a 1-yr time point for censoring within the TCGA-LUAD cohort. We then investigated the EA cohort to validate our findings. However, there were too few events, deaths, within the first year in this cohort; likely representing the difference in patients’ demographics and tumor characteristics. Thus, we opted to look additionally at 2-yr mortality so that the same analysis could be performed in both cohorts.

Model Construction and Testing

Cox proportional hazards models were constructed using the survminer and survival packages in R (22). Proportional hazard assumption testing was performed by examining Schoenfield residuals and Q-Q plots. Some survival models (e.g., modeling survival across the entire span of survival time in the validation cohort) failed to meet proportionality hazard assumption testing. Thus, these models were not used.

Model Results

For survival analysis, Kaplan-Meier curves and hazard ratios were calculated. Point estimates are provided with 95% confidence intervals in Forest plots and β coefficients and standard errors are provided in Tables 1, 2 and 3. A P value < 0.05 was accepted as statistically significant.

Table 1.

MK2 expression and death at 1 and 2 yr in TCGA-LUAD (n = 398)

	Tumor Stage
	Early Stage		Late Stage
	HR (95% CI)a	β (SE)a	HR (95% CI)a	β (SE)a
1-yr Mortalityb	0.24 (0.07, 0.80)	−1.42 (0.61)	0.48 (0.14, 1.66)	−0.71 (0.69)
2-yr Mortalityb	0.52 (0.28, 0.98)	−0.64 (0.31)	0.61 (0.27, 1.40)	−0.48 (0.42)

^aCox proportional hazards modeling of time to death, right censored at 2 yr, as a function of high MK2 expression (dichomotous: top 1/3 vs. bottom 2/3 of transcript expression), adjusted for age, sex, and smoking status. ^bTop 1/3 of MK2 transcript levels (reference: bottom 2/3). Bold signifies statistical significance; CI does not cross 1. TCGA-LUAD, The Cancer Genome Atlas Lung Adenocarcinoma.

Table 2.

MK2 expression and 2-yr mortality in EA cohort (n = 169)

Variable	HR (95% CI)a	β (SE)a
High MK2 expressionb	−0.10 (0.01, 0.81)	−2.24 (1.03)

^aCox proportional hazards modeling of time to death, right censored at 2 yr, as a function of high MK2 expression (dichotomous: top 1/3 vs. bottom 2/3 of transcript expression), adjusted for age, sex, smoking status, and tumor stage. ^bTop 1/3 of MK2 transcript levels (reference: bottom 2/3). Bold signifies statistical significance; CI does not cross 1. EA cohort, East Asian cohort.

Table 3.

Impact of oncogene mutational status on the effect of MK2 on 2-yr survival

	TCGA-LUAD (n = 309)
	HR (95% CI)a		β (SE)a		East Asian Cohort (n = 169)
	Early Stage	Late Stage	Early Stage	Late Stage	HR (95% CI)b	β (SE)b
EGFR Statusc	0.55 (0.29, 1.0)	0.60 (0.25, 1.4)	−0.59 (0.32)	−0.49 (0.43)	0.10 (0.01, 0.8)	−2.24 (1.03)
KRAS Statusc	0.53 (0.28, 1.0)	0.63 (0.27, 1.44)	−0.62 (0.31)	−0.45 (0.42)	0.10 (0.01, 0.8)	−2.27 (1.04)
TP53 Statusc	0.54 (0.28, 1.0)	0.61 (0.26, 1.39	−0.51 (0.32)	−0.49 (0.42)	0.11 (0.01, 0.8)	−2.17 (1.03)

Cox proportional hazards modeling of time to death, right censored at 2 yr, as a function of high MK2 expression (dichotomous: top 1/3 vs. bottom 2/3 of transcript expression), adjusted for age, sex, smoking status, and stratified by tumor stage ^a(TCGA-LUAD) or tumor stage ^b(East Asian cohort). ^cOncogene mutational status defined dichotomously as: WT or mutant. Bold signifies statistical significance; CI does not cross 1. TCGA-LUAD, The Cancer Genome Atlas Lung Adenocarcinoma.

Supplemental Data

All code pertaining to TCGA and EA analyses performed herein is available for review in the Supplemental Material. Additional markdowns, raw data, and code are available at https://github.com/suresh-lab/TCGA-casp3/.

RESULTS

Non-Small Cell Lung Cancer Cell Lines Have Lower MK2 Expression

We obtained previously characterized NSCLC and SCLC cell lines and evaluated MK2 expression (4). There is markedly reduced MK2 protein expression in the H23 and A549 NSCLC cell lines compared with the H446 SCLC cell line (Fig. 1, A and B, and Supplemental Fig. S2). To ensure general applicability, we interrogated the Broad Institute’s dependency map Consortium (https://www.depmap.org), which provides proteomic data across variety of cancer cell lines (23). MK2 protein expression in NSCLC cell lines was decreased compared with SCLC cell lines (Fig. 1C).

Figure 1.

Reduced expression of MK2 in non-small cell lung carcinoma cell (NSCLC) lines. A: a representative immunoblot of cell lysates from NSCLC cell line, H23, and SCLC cell line, H446, which were probed with antibodies recognizing total MK2. B: quantification of immunoblot analyses of MK2 expression shows increased MK2 expression in SCLC cell line, H446, compared with NSCLC cell lines, H23 and A549. *P < 0.05 vs. all other conditions using post hoc Dunnett’s multiple comparison test. n = 9 individual cultures per group. C: relative protein expression of MK2 using proteomics of lung cancer cell lines examined by the Cancer Dependency Map project (https://www.depmap.org) shows significantly lower MK2 expression in NSCLC cell lines compared with SCLC cell lines using Mann-Whitney test. D: real-time PCR analysis demonstrates significantly reduced MAPKAPK2 gene expression in the NSCLC cell lines, H23 and A549 as compared with the SCLC cell line, H446. *P < 0.05 vs. all other conditions using post hoc Dunnett’s multiple comparison test. n = 3 individual cultures per group. NSCLC cell lines (E) and SCLC cell lines (F) examined by the Cancer Dependency Map project show direct correlation between MK2 mRNA and MK2 protein expression. G: lung adenocarcinoma samples analyzed by The Clinical Proteomic Tumor Analysis Consortium (CPTAC) show direct correlation between MK2 mRNA and MK2 protein expression. E–G: a fitted line and 95% CI are plotted using GAM function.

We next sought to determine an association between MK2 expression and clinical outcomes in patients with NSCLC. However, many clinical data sets contain transcript levels and not protein levels. Therefore, we first assessed the correlation between MK2 transcript levels and protein. We observed significantly reduced MK2 mRNA transcript levels in H23 and A549 NSCLC cell lines compared with the SCLC cell line, H446 (Fig. 1D), concordant with observed MK2 protein expression. Using the Broad Institute’s dependency map Consortium data, we observed a direct correlation between MK2 mRNA and MK2 protein expression in available NSCLC and SCLC cell lines using a generalized additive model (Fig. 1, E and F). Although there was a direct correlation between MK2 transcript levels and protein expression in patient-derived lung cancer cell lines, we wanted to ensure this remained true in tumors of patients. Data from Clinical Proteomic Tumor Analysis Consortium also demonstrated a direct correlation between MK2 transcript levels and protein levels in lung adenocarcinomas from patients using a generalized additive model (Fig. 1G).

Low Tumor MK2 Expression Is Associated with Late-Stage Lung Adenocarcinoma

Given the variance of MK2 expression in NSCLC cells (Fig. 1C) and patients with lung adenocarcinoma (Fig. 1G), we reasoned we could leverage this variance to determine if MK2 expression was associated with outcomes in patients. Using the TCGA-LUAD data set, we observed a significant increase in MK2 transcript levels in tumors compared with adjacent normal tissue (Fig. 2A). Interestingly within tumors, we observed a significantly lower level of MK2 transcript in late-stage (Stages III and IV) compared with early-stage (Stages I and II) tumors (Fig. 2B). This difference in MK2 transcript levels was not apparent when assessing normal adjacent tissue (Fig. 2C).

Figure 2.

Low MK2 expression in tumors of patients with lung adenocarcinoma is associated with late-stage disease. A: using a subset of patients from The Cancer Genome Atlas Lung Adenocarcinoma (TCGA-LUAD) database, where adjacent normal lung tissue was also sampled, there is significantly higher MK2 expression in tumors compared with adjacent normal lung tissue, using Mann-Whitney test. Mean fold change of MK2 expression (±SE) in tumors compared with normal tissue is also plotted. B: within tumors, there is significantly lower MK2 expression in tumors of patients diagnosed with late stage (Stages III and IV) as compared with those diagnosed with early stage (Stages I and II), using Mann-Whitney test. C: there is no difference in MK2 expression in normal lung tissue between patients diagnosed with late-stage and early-stage lung adenocarcinoma, using Mann-Whitney test.

Higher Tumor MK2 Expression Is Associated with Clinical Response to Chemotherapy

Because loss of MK2 protects against endotoxin-induced cell death in endothelial cells (9), we reasoned that lower MK2 expression may be associated with less tumor response following chemotherapy. We analyzed TCGA-LUAD data for tumor responses following first round of chemotherapy, dichotomized as clinical response (complete response, partial response, and stable disease) or progression of disease. Tumors with a clinical response to chemotherapy had higher MK2 expression; this trend remained even after stratification by tumor stage (Fig. 3, A and B). To ensure this association of MK2 levels to clinical response following chemotherapy was not just a reflection of upregulation of the apoptotic pathways, we measured caspase-3 levels based on tumor response. As shown in Fig. 3C, there is no difference in tumor caspase-3 transcript levels in patients with a clinical response versus those with progression of disease. Although caspase-3 is the “terminal step” of the apoptotic cascade, we reasoned that it is plausible that other effectors of the apoptotic cascade may be upregulated leading to a clinical response following chemotherapy. Recently a seven-gene apoptosis gene signature was identified within the TCGA-LUAD cohort (24). Using this apoptosis gene set, we performed a principal component analysis to determine if there were any differences appreciable within the apoptosis gene signature in patients with clinical response compared with those with progression of disease. As shown in Fig. 3D, patients with a clinical response and those with progression of disease had similar apoptosis gene signatures. To defend against the subjective nature of choosing this particular apoptosis gene set, we queried the Gene Ontology database for apoptosis-related terms (n = 341; Supplemental Table S1) and subsequently derived apoptosis-related genesets based on these terms (n = 14,899). We then performed GSVA analysis in the TCGA-LUAD data set (GSVA package in R) using genesets with a minimum size of 5 and maximum size of 500 (n = 6,789). There were no significant differences in GSVA scores comparing patients with clinical response to those with tumor progression of disease (Fig. 3F). We next filtered genesets where the strength of evidence supporting an inclusion of a gene in a particular geneset was inferred from experiment; inferred from direct assay, or stated directly by an author (traceable author statement or nontraceable author statement). This filtering produced 57 relevant genesets. Again, there were no significant differences in GSVA scores comparing patients with clinical response to those with tumor progression (Fig. 3G). In total, these data demonstrate that MK2 levels, and not previously described effectors of the apoptotic cascade, are associated with lung adenocarcinoma tumor response following chemotherapy.

Figure 3.

High tumor MK2 expression is associated with clinical response to chemotherapy and improved survival in patients with lung adenocarcinoma. A–C: TCGA-LUAD data of tumor responses following first round of chemotherapy was analyzed. Response to chemotherapy was dichotomized: clinical response (CR; complete response, partial response, and stable disease) or progression of disease (PD). MK2 mRNA levels for patients with unknown response to chemotherapy are also plotted. Clinical Response to chemotherapy was associated with higher MK2 expression compared with tumors with progression of disease (A); this trend remained even after stratification by tumor stage (B and C) using Mann-Whitney test. D: caspase-3 mRNA levels were plotted based on tumor responses following first round of chemotherapy. There is no difference in caspase-3 expression in tumors with clinical response or progression of disease. E: using a seven-gene apoptosis gene signature, principal component analysis was performed on tumor based on responses following first round of chemotherapy. There is overlapping gene signatures in those with clinical response and those with progression of disease. F: GSVA analysis in the TCGA-LUAD data set using genesets from Gene Ontology database for apoptosis-related terms with a minimum size of 5 and maximum size of 500 showed no significant differences in GSVA scores comparing patients with clinical response to those with tumor progression of disease. G: additional filtering of genesets to include only where the strength of evidence supporting an inclusion of a gene in a particular geneset was: inferred from experiment; inferred from direct assay, or stated directly by an author still resulted in no significant differences in GSVA scores comparing patients with clinical response to those with tumor progression. TCGA-LUAD, The Cancer Genome Atlas Lung Adenocarcinoma.

Higher Tumor MK2 Expression Is Associated with Improved Survival in Patients with Lung Adenocarcinoma

We initially investigated a cutoff of the median to determine the association between higher MK2 expression and survival. As shown in Fig. 4A, we note a survival advantage in those with a higher MK2 expression. However, we posited a nonlinear relationship between MK2 expression and survival may provide rationale for a more appropriate cutoff (21). Based on the shape of the relationship observed when employing flexible general additive models using MK2 expression as a continuous variable, there is evidence for nonlinearity of the association between MK2 expression and survival in a combined TCGA and EA cohort (Fig. 4B). Accordingly, we opted to show the association of select subgroups to better demonstrate the extent of the relationship particularly at the part of the curve corresponding to the top 1/3 of MK2 expression, which appears to have a relatively constant slope. As shown in Fig. 4C, there is a survival advantage in patients with tumors in the top 1/3 of MK2 expression.

Figure 4.

High tumor MK2 expression is associated with improved survival in patients with lung adenocarcinoma. A: plot showing HR estimates and 95% CI results for a univariate Cox PH model assessing the effect of high MK2 expression (> median) on overall 2 yr survival in a combined data set (TCGA + East Asian cohort). B: plot showing the relationship between MK2 levels and predicted survival time in a combined data set (TCGA + East Asian cohort) using an adjusted generalized additive model (GAM). C: plot showing HR estimates and 95% CI results for a univariate Cox PH model assessing the effect of MK2 expression (by tertiles) on overall 2 yr survival in a combined data set (TCGA + East Asian cohort). D: plot showing HR estimates and 95% CI results for a multivariable Cox PH model assessing the effect of MK2 expression stratified by stage in TCGA cohort. Model: h(t) ∼ high MK2 level^a + sex + smoking + age. ^aGene level defined dichotomously as 1: top 1/3 and 0: bottom 2/3rd. E: plot showing HR estimates and 95% CI results for a Cox PH model assessing the effect of high MK2 expression^a on overall survival stratified by stage in EA cohort. Model: h(t) ∼ high MK2 level^a + sex + smoking + age + stage. ^aGene level defined dichotomously as 1: top 1/3 and 0: bottom 2/3rd. F and G: Kaplan-Meier curves showing 2-yr survival in patients with early-stage NSCLC in the TCGA-LUAD (F; n = 312) and EA-LUAD (G; n = 169) data set stratified by high (black) and low (gray) MK2 transcript levels. Time is shown in months. EA-LUAD, East Asian Lung Adenocarcinoma; NSCLC, non-small cell lung cancers; TCGA-LUAD, The Cancer Genome Atlas Lung Adenocarcinoma.

After identifying an appropriate cutoff for MK2 transcript values and survival in a combined TCGA and EA cohort, we then focused specifically on the TCGA cohort. We noted high MK2 transcript level (top 1/3 vs. bottom 2/3) was associated with improved survival at 1-yr (HR 0.24, 95%CI: 0.07–0.80) and 2 yr (HR 0.52, 95%CI: 0.28–0.98) in early-stage, but not late-stage, NSCLC (Fig. 4D and Table 1).

We noted patients were entered into the TCGA data set over a long period of time (1991–2013). The number of patients enrolled (per year) was skewed, with the majority of patients being enrolled after 2005 (Supplemental Fig. S3A). To determine whether time of enrollment could be confounding our results, we performed a jackknife analysis whereby we examined the effect of sequential exclusion of one particular year of data (i.e., excluding all patients diagnosed in that year) on the point estimate results for Cox Proportional Hazards model (Supplemental Fig. S3B). We observed that the MK2 hazard ratio and confidence intervals were very stable even with specific years excluded, suggesting that imbalances related to year of enrollment were likely not playing a significant role.

Using the EA cohort, we observed improved 2-yr survival associated with higher MK2 expression (HR: 0.1, 95%CI: 0.01–0.81; Fig. 4E and Table 2). Kaplan-Meier curves for TCGA and EA cohorts, shown in Fig. 4, F and G, demonstrate a significant survival advantage in patients with high compared with low (top 1/3 vs. bottom 2/3) MK2 expression.

MK2 Expression’s Association with Improved Survival Is Independent of Common Oncogene Mutations

Next, we sought to determine if MK2 expression was merely a reflection of oncogenic driver mutations, known to impact survival (25–27). There were no significant differences in MK2 transcript levels based on oncogenic driver mutation status for EFGR and TP53 across both TCGA and EA cohorts (Fig. 5, A and B). Interestingly, tumors with a KRAS mutation, typically associated with higher mortality, tended to have higher MK2 levels. Using multivariable Cox PH, we observed an independent effect on 2-yr survival associated with higher MK2 expression in both TCGA and EA cohorts even after additional adjustment for oncogenic driver mutation status (Table 3).

Figure 5.

MK2 expression is not associated with common oncogene driver mutations. Using The Cancer Genome Atlas Lung Adenocarcinoma (TCGA-LUAD; A) and East Asian Lung Adenocarcinoma (EA-LUAD; B) databases, MK2 expression is plotted based on oncogenic mutation status for EGFR, KRAS, and TP53. Statistical significance is assessed using Mann-Whitney test.

High Tumor MK2 Expression Is Associated with Survival Only in Lung Adenocarcinoma

To determine whether the marked improvement in survival in patients with higher MK2 expression was truly an outlier, we plotted a histogram for HRs for the multivariable Cox PH model for 1-yr survival for all genes available in the TCGA-LUAD data set (N = 14,899 genes). As shown in Fig. 6A, the HR for MK2 is beyond two standard deviations from the mean HR.

Figure 6.

High tumor MK2 expression is associated with improved survival only in lung adenocarcinoma. A: histogram plot of HR estimates for Cox PH model assessing the effect of high gene expression (defined as top 1/3 of transcript level) on overall 1-yr survival in TCGA-LUAD. Red line represents hazard ratio for MK2. Black vertical lines represent two standard deviations surrounding the mean hazard ratio. N = 14,899 genes. B: plot showing HR estimates and 95% CI results for a Cox PH model assessing the effect of high MK2 expression^a on overall survival at 2 yr. Vertical line represents HR of 1. Model: h(time to death | covariates) ∼ high MK2 level^a + cancer stage^b+ sex+ smoking + age. ^aGene level defined dichotomously as 0: Bottom 2/3rd of log normalized gene transcript levels for that cancer type; 1: Top 1/3. ^bCancer stage defined as early stage (Stages I and II) vs. late stage (Stages III and IV). C: lollipop plots showing Cox PH model metrics by cancer type. Vertical line represents P value of 0.05. Left: dots represent the log P value for goodness of fit (Wald test) for each Cox PH model (P < 0.05 represents significant). Right: dots represents results of proportionality hazard testing P value (P > 0.05 represents no violation of PH assumption). D: within malignancies with a favorable point estimate for the Cox PH model noted in B (BRCA, STAD and PAAD cancers), there is no difference in MK2 expression between early stage (Stages I and II) and late stage (Stages III and IV) within, using Mann-Whitney test. BLCA, bladder urothelial carcinoma; BRCA, breast invasive carcinoma; COAD, colon adenocarcinoma; ESCA, esophageal carcinoma; KIRC, kidney renal cell carcinoma; KIRP, kidney renal papillary cell carcinoma; LIHC, liver hepatocellular carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; PAAD, pancreatic adenocarcinoma; READ, rectum adenocarcinoma; SKCM, skin cutaneous melanoma; STAD, stomach adenocarcinoma; TCGA, The Cancer Genome Atlas.

We next sought to determine the effects of higher MK2 expression on patient survival were specific to lung adenocarcinoma. We examined the effect of MK2 on overall survival across multiple TCGA data sets. For this analysis, we prespecified the multivariable Cox PH model as follows: MK2 was specified as high/low (as in our previous analyses), and we included cancer stage (defined as early vs. late), sex, smoking and age as relevant covariates. Similar to our prior analysis, we again censored the data at 2 yr. Since we were applying a Cox PH model across multiple data sets, we also extracted metrics of model fit, including overall goodness of fit (i.e., the P value associated with the Wald test for the Cox PH model) and PH assumption testing (i.e., P value associated proportionality hazard assumption testing, where P < 0.05 signifies violation of hazard assumption). In Fig. 6, B and C, we show not only the HR for high MK2 in the multivariable model across data sets, but also the relative performance of our prespecified model in each data set. As shown in Fig. 6B, the LUAD data set was unique in that MK2 was associated with a significant lower HR for death. We note that the model fit for LUAD was good, as evidenced by 1) the MK2 variable did not violate PH assumption and 2) a significant goodness of fit (Wald P value < 0.05; Fig. 6C). Given the favorable point estimates for the Cox PH model in BRCA, STAD, and PAAD, we investigated if MK2 expression was potentially expressed differentially in early versus late stage, as in LUAD (Fig. 2B). There is no difference in MK2 expression between early stage and late stage within BRCA, STAD and PAAD cancers (Fig. 6D). Interestingly, in kidney renal clear cell cancer (KIRC), high MK2 transcript expression appeared to be associated with worse survival. However, when comparing MK2 mRNA and MK2 protein expression in renal cell carcinoma cell lines, we observed a negative correlation, which was not statistically significant (Supplemental Fig. S4).

DISCUSSION

This study shows reduced MK2 expression in NSCLC cell lines. Among patients presenting with lung adenocarcinoma, tumors had higher MK2 expression than adjacent normal tissue. In tumors, higher MK2 expression was associated with early-stage disease and better clinical response following chemotherapy. Using multivariable models in two separate cohorts, our data show higher MK2 expression is associated with improved 2-yr survival in patients presenting with lung adenocarcinoma, even after adjusting for a variety of clinical parameters, including presence of common oncogenic driver mutations. Furthermore, our data show survival advantage of higher MK2 expression is unique to lung adenocarcinoma.

Our data extend on other published work of MK2 in cancer biology. For example, intestinal carcinogenesis was related to MK2 kinase activity as chemical inhibition of MK2 decreased proliferation and tumor growth (28). In a novel murine model of NSCLC that generates MK2-proficient and MK2-deficient tumors within the same animal, Morandell et al. (29) show that MK2-proficient tumor area is consistently greater than MK2-deficient tumor area. Interestingly, the majority of data showing a role for MK2 in cancers focus on MK2’s kinase activity in tumor development (28, 30–32). Our data show increased MK2 transcript levels in tumors compared with adjacent normal tissue, suggesting MK2 may be important to development of lung adenocarcinoma in patients. In addition, our data suggest once lung adenocarcinoma is diagnosed, higher MK2 expression is associated with improved outcomes (chemotherapy response and survival; Figs. 3 and 4). One explanation for our data is that MK2 may have divergent roles in tumor development and apoptosis resistance. In addition, many previously published works focused on different cancer types (i.e., intestinal tumors) or utilized murine models (28–32). In patient-derived data from TCGA, we note improved survival associated with higher MK2 expression in lung adenocarcinoma. Based on the favorable point estimates for the Cox PH models in LUAD, BRCA, STAD, and PAAD, we initially thought that MK2 may have a different role in adenocarcinomas compared with other types of malignancy. However, many other adenocarcinomas within the TCGA data sets, i.e., READ, COAD, do not have an point estimate that is protective in terms survival (Fig. 6B). We were unable to find a discernible pattern in the malignancies with a favorable estimate point for hazard ratio. Although the point estimate for the Cox PH model in BRCA, STAD, and PAAD appears favorable, it must be noted that the 95% confidence interval does cross one for these malignancies. It is intriguing to consider that increasing the sample size may help discern a true benefit; however, the BRCA data set has a large number of patients already, N = 1,097. Furthermore, there is no difference in MK2 expression between early-stage and late-stage within BRCA, STAD, and PAAD cancers (Fig. 6D). In total, our data suggest that a beneficial role for higher MK2 expression appears to be unique to lung adenocarcinoma.

Our hypothesis of reduced MK2 expression in NSCLCs as a plausible mechanism for apoptosis resistance was rooted in the similarities between cytoplasmic sequestration of caspase 3 in MK2-deficient conditions of our previous work in endothelial cells (9) and that of Joseph et al. (4) using NSCLC cell lines. However, we struggled to find an appropriate comparator cell type which to compare NSCLC cells’ MK2 protein expression against because of the known varied protein expression across tissue types, cell types (proteinatlas.org and genecards.org) and the numerous publications implicating MK2 in the development of tumors, suggesting increased expression in tumors compared with normal tissues (28–32). We ultimately settled on using SCLC cells as a comparator because they were both from lung tissue and carcinomas. Our data clearly show NSCLC cell lines have significantly reduced MK2 expression compared with SCLC cell lines (Fig. 1). Although we have previously demonstrated knocking down MK2 expression via siRNA, reduced endotoxin-induced apoptosis in endothelial cells (9), we do not suggest that sensitivity of SCLC to chemotherapy (3, 4) is mediated by higher MK2 expression. Carcinomas, of all kinds, have many prosurvival and antiapoptotic molecular drivers that ultimately result in sensitivity to chemotherapy. Thus, we limited our studies to focus on the association between MK2 expression and clinical outcomes of patients with NSCLC.

A major strength of our study is providing a mechanistic link of MK2 expression, known to promote nuclear translocation of caspase-3 and the execution of apoptosis (9, 10) in NSCLCs cells (having reduced MK2 expression) and patient level data demonstrating increased expression of MK2 is associated with early-stage lung adenocarcinoma (Fig. 2), improved clinical response following chemotherapy, i.e., increased tumor death (Fig. 3) and improved survival (Fig. 4). Our data provide rationale for MK2 as an independent prognostic biomarker, where higher MK2 expression is associated with improved outcomes.

Our studies have several limitations. First, as is common with retrospective studies, our genomic analyses of two clinical cohorts are post hoc and thus susceptible to residual confounding. Second, MK2 mRNA and not protein levels were used in analyses. This is somewhat offset by direct correlation between mRNA and protein levels for MK2 in lung adenocarcinomas of patients and patient-derived cell lines (Fig. 1). Third, only initial tumor MK2 transcript levels were available for analyses. Therefore, it is unclear if lower MK2 transcript level associated with late-stage disease is a result of tumor progression. Serial analyses of tumors are challenging as repeated biopsies are rarely clinically indicated. However, it is unlikely lower MK2 transcript predisposes to late-stage lung adenocarcinoma as we find similar MK2 expression in normal tissue of patients with early and late-stage disease (Fig. 2C). Fourth, although higher MK2 levels were associated with clinical response to chemotherapy, it is possible that different chemotherapy regimens may have differential responses based on MK2 expression. Our finding of stable and protective HR estimates through 1991–2013 using jackknife analyses, when first-line chemotherapy regimens varied, suggest an independent effect of MK2 (Supplemental Fig. S3).

In summary, this study shows lung adenocarcinoma MK2 transcript levels at time of diagnosis are independently associated with improved response to chemotherapy and survival in patients, a finding unique to lung adenocarcinoma. Although additional longitudinal and confirmatory proteomics-based studies are needed, our observations provide potential prognostic value of MK2 in patients diagnosed with lung adenocarcinoma and offer rationale on how MK2 may be leveraged to overcome inherent resistance to chemotherapy.

DATA AVAILABILITY

The results shown here are in whole or part based on data generated by The Cancer Genome Atlas (TCGA) Research Network: https://www.cancer.gov/tcga. Additional markdowns, raw data, and code are available at https://github.com/suresh-lab/TCGA-casp3/.

SUPPLEMENTAL DATA

Supplemental Figs. S1–S4, Supplemental Table S1, and MK2 Analysis Code: https://doi.org/10.6084/m9.figshare.21466686.

GRANTS

This study was supported by National Institutes of Health Grants R01HL133413 (to M.D.), R01HL159906 (to L.A.S. and M.D.), R01HL151530 (to K.S.), and K08HL132055 (to K.S.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

K.S. and M.D. conceived and designed research; K.S., O.D.R., and M.D. performed experiments; K.S., O.D.R., N.M.P., T.M.K. and M.D. analyzed data; K.S., O.D.R., M.K., G.S., A.S., L.Z., X.Y., N.M.P., N.P., M.B.M., H.J., F.D., M.S., A.B., L.A.S., M.M., M.J.R., C.E.M., J.M., R.H., A.L.K., N.M.P., T.M.K., and M.D. interpreted results of experiments; K.S. and M.D. prepared figures; K.S. and M.D. drafted manuscript; K.S., O.D.R., M.K., G.S., A.S., L.Z., X.Y., N.M.P., N.P., M.B.M., H.J., F.D., M.S., A.B., L.A.S., M.M., M.J.R., C.E.M., J.M., R.H., A.L.K., N.M.P., T.M.K., and M.D. edited and revised manuscript; K.S., M.K., O.D.R., G.S., A.S., L.Z., X.Y., N.M.P., N.P., M.B.M., H.J., F.D., M.S., A.B., L.A.S., M.M., M.J.R., C.E.M., J.M., R.H., A.L.K., N.M.P., T.M.K., and M.D. approved final version of manuscript.