Protein kinase C isozymes; predictors of progression free survival in NSCLC patients

Background Protein expression is deregulated in cancer, and the proteomic changes observed in lung cancer may be a consequence of mutations in essential genes. The purpose of this study was to identify protein expression associated with prognosis in lung cancers stratified by smoking status, molecular subtypes, and EGFR-, TP53- and KRAS-mutations. Methods We performed profiling of 295 cancer-relevant phosphorylated and non-phosphorylated proteins, using reverse phase protein arrays. Biopsies from 80 patients with operable lung adenocarcinomas were analyzed for protein expression and association with progression free survival (PFS) were studied. Results Spearman rank correlation analysis identified 56 proteins with significant association to PFS (p<0.05). High expression of protein kinase C (PKC)-α and the phosporylated state of PKC-α, PKC-β and PKC-δ, showed the strongest positive correlation to PFS, especially in the wild type samples. This was confirmed in gene expression data from 186 samples. Based on protein expression, unsupervised hierarchical clustering separated the samples into four subclusters enriched with the molecular subtypes TRU, PI or PP (p=0.0001). Subcluster 2 contained a smaller cluster (2a) enriched with samples of the subtype PP, low expression of the PKC isozymes, and associated with poor PFS (p=0.003) compared to the other samples. Subcluster 2a revealed increased expression of neuroendocrine markers, supporting the aggressive behavior. Low expression of the PKC isozymes in the subtype PP and a reduced relapse free survival was confirmed with the TCGA LUAD samples. This study identified different proteins associated with PFS depending on molecular subtype, smoking-and mutational-status, with PKC-α, PKC-β and PKC-δ showing the strongest correlation. Cluster analysis detected a subgroup of samples enriched for samples of the PP subtype and poor PFS, which may benefit from a more aggressive treatment regimen. These results demonstrate that essential mutations in lung carcinomas affect several proteins associated with outcome. Based on our results, expression of PKCα and phosphorylated PKCα, PKCβ, and PKCδ seem to be positively associated with PFS, with different isozymes linked to smoking and mutational status of EGFR , KRAS and TP53 . These results illustrate the need to better understand the biological context in order to further improve targeted therapy in cancer. This study supports that a therapy restoring the level of specific isozymes of PKC activity may be beneficial for subgroups of lung cancer patients based on the genetic background. We identified a subgroup of samples enriched with the molecular subtype PP, recognized with early relapse, increased expression of neuroendocrine markers, and a distinct protein expression pattern, including low levels of PKC isozymes. These patients may benefit from a more aggressive treatment regimen. Proteins associated with PFS among never smokers were strikingly different compared to the other investigated subgroups. This is not surprising, but underscores the need for a more stratified therapy in order to improve clinical outcome.


Introduction
Proteins are the functional players driving both normal and disease processes. Some of the most important types of mutations in lung cancer occur in EGFR, TP53 and KRAS. Mutations in these genes may lead to changes in many interacting pathways, leading to significantly altered protein expression. They are known to influence treatment response and regarded as essential for progression of lung cancer [ 1 ]. Some of the changes in protein expression observed in lung cancer are a consequence of mutations in essential driver genes, and targeted therapy is usually efficient for these subgroups of patients [ 2 ].
Patients with somatic genomic alterations in the EGFR gene are routinely treated with EGFR inhibitors, which have improved the outcome for this patient group [ 3 ]. Mutations in KRAS leads to constitutively and persistent stimulus-independent activation of downstream pathways affecting tumor growth, proliferation and survival. Developing treatment targeting KRas has proved to be complicated, but ongoing studies investigate inhibition of effectormolecules downstream of KRas, including ERK and MEK [ 4 ]. Nevertheless, MEK inhibitors are associated with early development of resistance due to crosstalk with other signaling pathways which make this approach challenging [ ]. So far, no efficient therapy to re-establish the function of p53is in clinical use, but studies with reactivation of p53 have been performed [ 6 ]. In order to improve outcome for lung cancer patients, stratification based on alterations in essential genes and the affected pathways may lead to better treatment strategies and increased response rate.
To sub-classify non-small cell lung cancer (NSCLC), intrinsic molecular subtypes have been explored based on gene expression profiling. Three subtypes have been identified for the adenocarcinomas, namely the Bronchioid, Magnoid, and Squamoid subtype, later re-named to terminal respiratory unit (TRU), proximal proliferative (PP), and proximal inflammatory (PI), respectively [7][8][9]. The TRU subtype is most common among females and never-smokers, and often harbours EGFR mutations and a less invasive phenotype. Early stages of the TRU subtype have better prognosis. Gene expression profiles related to biological processes involved in excretion, asthma, and surfactants are associated with the TRU subtype. The PP subtype is reported with a high frequency of KRAS and TP53 mutations, over-expression of DNA repair genes and is often found in heavy smokers.
The PI subtype is recognized with over-expression of defense response genes such as CXCL10 and is most common in high grade tumors [ 7 ]. At late stages, the PI subtype is associated with better survival compared to the other subtypes [ 8 6 ]. Chromosomal instability (CIN), copy number alterations, and genomewide DNA methylations are also reported to differ among the three subtypes, with the PP subtype having the highest CIN [ 8 ]. So far, microRNA expression and protein expression are reported to only partially correlate with the molecular subtypes of adenocarcinoma [ 9 ]. In order to treat lung cancer patients more efficiently, groups of patients who share common biological features such as mutations or pathway alterations should be identified and treated with drugs optimized for their subgroup. It has been known for a long time that never-smoking NSCLCs are recognized with a different underlying biology compared to ever-smokers. However, except from a handful of known genetic aberrations such as EGFR mutations and ALK rearrangement, no subgroups of NSCLC are stratified for optimized treatment based on the molecular profile of the tumor.
The purpose of this study of lung adenocarcinoma, was to identify differences in protein expression levels associated with prognosis, stratified on EGFR, TP53 and KRAS mutations status, and smoking status.

Material And Methods
Patients diagnosed with operable NSCLC adenocarcinoma from 2006 to 2011, were included and analyzed for protein expression (n=80) in the Oslo cohort. The patients underwent curatively intended surgical resection at Rikshospitalet, Oslo University Hospital, Norway. Tumor samples were snap frozen in liquid nitrogen and stored at -80°C until protein extraction. Pathological stage, mutation status and smoking status are outlined in Table 1. All the samples were classified as adenocarcinomas. Never-smokers are defined as those who have smoked less than 100 cigarettes.
The study was approved by the Regional Ethics Committee (S-05307), and written informed consent was obtained from all patients.

Reverse phase protein arrays
We have performed profiling of 295 cancer relevant proteins of which 60 were in a phosphorylated state (S1 Table and supplementary material and methods) on the Oslo cohort, using the reverse phase protein array (RPPA) core facility at MD Anderson Cancer Center (Houston, TX). RPPA data (n=131 proteins) and phenotypes from the LUAD cohort were extracted from the cancer genome atlas (TCGA) data generated by the TCGA Research Network: http://cancergenome.nih.gov/ . Samples with no expression subtype assigned and without registered relapse free survival were filtered out. The remaining 181 LUAD samples were utilized as validation set. Time to relapse was extracted for survival analyses. Of note, the median follow-up time for alive patients was 23.9 months for the LUAD cohort, compared to 60 months for the Oslo cohort.

EGFR, TP53 and KRAS analyses
Mutation analyses of EGFR exons 18-21 were performed using the TheraScreen EGFR mutation kit (DxS, Manchester, UK,) designed to detect 28 specific mutations in the EGFR gene. Assays were carried out according to the manufacturer's protocol and with the use of the Roche LightCycler 480 real-time PCR system. Some of the results were previously published by Helland et al [ 10 ]. The

Immunohistochemistry (IHC)
IHC staining for the neuroendocrine markers CD56, synaptophysin, chromogranin A and NSE were performed using the antibodies as outlined in S2 PFS was calculated from the date of surgery until the date of event, defined as relapse, metastases or death from lung cancer. Correlation between protein expression and PFS was calculated using twosided Spearman test. Student t-test was performed to identify differentially expressed proteins between the groups of interests. Survival analysis was performed using cox regression analysis and Kaplan Meyer survival plot. The significance threshold was set to p<0.05.

Results
We performed RPPA on biopsies from 80 lung adenocarcinoma biopsies. The tumor samples were tested for mutations in the EGFR-, TP53-, and KRAS genes, where 12.5% had EGFR mutations, 39 % had KRAS mutations and 39 % had mutations in the TP53 gene. In the group of patients with EGFR mutated tumors, four (representing 44%) were never smokers ( Table 1). Out of 80 patients, 36 (45%) patients were registered with PFS > 60 months. Among these, three (8%) of the tumors were EGFR mutated, 15 (42%) were TP53 mutated, 13 (36%) were KRAS mutated and three patients were never smokers.

Hierarchical clustering analysis
Unsupervised hierarchical clustering analysis performed on protein expression grouped the samples into four subclusters as shown in the dendrogram (Fig 1). The parameters EGFR-, TP53-, KRASmutations, stage, event and smoking status were evenly distributed between the sub-clusters (p> 0.05 in Fisher Exact test). The molecular subtypes were significantly associated with the four subclusters (p = 0.0001), of which subcluster 1 (green branches) was mainly associated with the PI subtype, subcluster 2 (red branches) enriched with the PP subtype, and subcluster 3 and 4 (turquoise and blue branches) were associated with the TRU subtype. Subcluster 2 contained four smaller clusters (2a, 2b, 2c and 2d visualized in S1 Fig Table).
To identify the proteins with the highest correlation to PFS within this group, Spearman Rank correlation was utilized. Eight proteins (myosin11, CD26, YAP, DJ1, YAP_pS127, CD171, HSP27_ps82 and hexokinaseII) were significantly associated (p< 0.05) with PFS in subcluster 2a (S4 Table). To check if subcluster 2a contained neuroendocrine-like features, the expression of the genes NCAM1, CHGA, SYP encoding for the neuroendocrine markers CD56, chromogranin A and synaptophysin respectively, were analyzed. All three markers revealed a higher gene expression in samples from subcluster 2a compared to all the other samples, but only SYP was significant (p=0.03). IHC staining for the proteins CD56, chromogranin A, synaptophysin and neuron-specific enolase (NSE) confirmed an expression of synaptophysin, CD56, chromogranin A, and NSE in nine, two, one, and nine of the 11 samples, respectively (S5 Table). Serum levels of the neuroendocrine markers proGRP, chromoganin A and NSE were elevated in one of the 11 samples (data not shown). Protein expression associated with prognosis Using Spearman rank correlation, we identified 56 proteins significantly associated with PFS (p<0.05), of which nine were in a phosphorylated state (S4 Table). High expression of PKC-α, and phosphorylated PKC-α, PKC-βII, and PKC-δ were all highly associated with increased PFS ( Table 2). The proteins c-ABL, MIF, PAI-1, Jak2, vimentin and the phosphorylated state of JNK were also among the top twenty proteins significantly associated with PFS (p<0.05).   Next, we included all the samples with mRNA data (n=186) and performed a Spearman Rank correlation between genes and proteins associated with prognosis, and found 10 proteins/genes significantly associated with PFS in both analysis (Table 3 and    Protein expression with impact on prognosis differ depending on mutational and smoking status A significant association between PFS and expression of 10 proteins (S4 Table) High expression of B7-H3 was associated with poor survival in the total group, but only found to be significant in KRAS and TP53 wild type tumors, smokers, and EGFR mutated samples after stratification (p<0.05, S4 Table).
When comparing similarities in PFS-associated proteins between the mutated samples and the wild type samples two, three and four overlapping proteins were identified in the EGFR, TP53 and KRAS subgroups, respectively. The proteins PKC-α and phosphorylated PKC-α, PKC-βII, and PKC-δ were the most significantly PFS-associated proteins in smokers, TP53wt, EGFRwt, and KRASwt. Among the mutated samples, fewer proteins were associated with PFS, and as expected, no overlapping proteins were found between EGFR and KRAS mutated samples (S4 Table and  encoding PKC-δ, was significantly associated with PFS among smokers, EGFR mut, EGFR wt, and KRAS wt, which confirmed the findings in protein data. A tendency towards worse relapse free survival in patients with low expression of PKC-δ was also seen in the TCGA data (p=0.06, Fig 3). The PP subtype showed a significantly decreased expression of PKCα, and phosphorylated PKC-α and PKC-δ when compared to PI and TRUE subtype. Further, the PP subtype also showed a reduced relapse free survival when compared to non-PP samples (p=0.085, Fig   3).

Discussion
In this study we examined the expression of 235 proteins and 60 phospho-proteins in tumors from 80 patients diagnosed with lung adenocarcinomas. We identified 56 proteins and phospho-proteins likely to impact the patient outcome. When stratifying the samples according to mutations in the genes TP53, EGFR or KRAS, more proteins were associated with PFS in samples without mutations. This was also seen when analyzing the corresponding mRNA data. When we compared proteins associated with progression in never-smokers and former/current smokers, no overlapping proteins were found, confirming the differences between these groups of lung cancer. Cluster analysis identified a small subcluster containing 11 patients enriched with tumors of the PP subtype, recognized with early relapse, low expression of PKC isozymes, and increased expression of neuroendocrine markers.

Cluster analysis
Based on the expression of 295 proteins, unsupervised hierarchical clustering separated the samples into four subclusters. However, pathological stage, smoking status, or mutations in the genes TP53, KRAS or EGFR did not seem to impact the clustering. Interestingly, the four subclusters were significantly correlated with the molecular subtypes TRU, PP and PI. Previous work using protein expression identified six subgroups of adenocarcinoma, where the subgroups partially overlapped with the three mRNA-derived subtypes. The PP subtype was further divided into two groups [ 9 ], which also was seen in our data set. Subcluster 2a, enriched with PP subtype samples, was recognized with lower expression of members of the mTOR pathway and the MAPK pathway. This is in line with previous finding where TRU samples are associated with higher expression of these proteins when compared to PP samples [ 15 ]. Interestingly, cluster analysis performed on mRNAs corresponding to the proteins, grouped the samples with PP subtype from the protein-derived subcluster 2a together, but not the non-PP subtypes. However, the distinct pattern with very low expressed proteins within subcluster 2a, was not reflected using mRNA data. This can be explained with a lower correlation between The hierarchical clustering performed on 186 mRNA samples resulted in a significantly different distribution of the molecular subtypes between the clusters, although the PI samples did not cluster together. Of note, the overlap between the genes used for the original sub-typing [ 8 ] and our clustering based on protein expression was sparse (n=18), indicating that the proteins included in our analysis seem to be important for the subtypes. Both the TRU-and PPsubtype clustered based on our mRNA data, and we suggest that the TRU-and PP-subtypes are more distinct subtypes compared to PI. In a large meta-study, the TRU subtype was identified as the most prognostically important subtype compared to the non-TRU subtype, arguing for the need to identify additional classifiers [ 18 ].

Protein kinase C levels associated with survival
The protein kinase C is a group of enzymes known to be involved in diverse cellular functions, including cell proliferation, apoptosis and cell migration, and has been regarded as an onco-protein.
The members of the PKC-family are encoded from nine different genes which have several known splice variants. Recent work has demonstrated that these proteins may have a more complex role than first assumed, which is supported by the many failed clinical trials for cancer using PKCinhibitors. In addition, mutational studies have revealed that most cancers have loss of function (LOF) mutations in genes belonging to the PKC-family, suggesting a tumor suppressor role for the proteins [ 19 , 20 ]. A meta-study on the use of PKC-inhibitors combined with chemotherapy in lung cancer patients reported decreased response rate and disease control, compared to chemotherapy alone [ 21 ]. In our study, low expression of PKC-α, and phophorylated PKC-α, PKC-βII and PKC-δ were associated with poor PFS. In addition, the levels of the PKC isozymes were strikingly lower in subcluster 2a which also contained the samples with the overall poorest PFS. An association with low expression of PKC-δ and decreased relapse free survival was confirmed in the LUAD TCGA samples. In the EGFR mutated samples, PKC-δ levels positively correlated to patient survival. It has been reported that an activation of PKC-δ can be promoted by an activated EGFR [ 22 ]. Further, activation of PKC-δ may induce apoptosis and growth arrest resulting in reduced tumorigenesis [ 23 ]. As a response to DNA damage, it has been shown that over-expression of p53 increases the transcription of PKC-δ resulting in apoptosis. This may explain the significant correlation we discovered between expression of PKC-δ and PFS in TP53 wild type samples, but not in the TP53 mutated samples. These results were also reflected by gene expression data. It's been demonstrated that PKC can phosphorylate many oncoproteins to suppress their activity, including KRas, PI3K and several tyrosine kinase receptors [ 24 ]. The oncoprotein KRas, recognized with activating mutations in cancer, can be suppressed by activated PKC, which is proposed as a novel approach to target KRas [ 25 ]. A negative correlation between KRas and PKC-α was recently described in colorectal cancer. Further, low expression of PKC-α was also associated with poor prognosis [ 26 ]. This is in concordance with our results, where high levels of PKC-α was associated with better PFS in KRAS mutated samples, whereas PKC-δ showed a higher correlation to PFS in those with KRAS wild type. Interestingly, no isozymes of PKC did influence on survival in never-smoking patients, further supporting this group as a distinct lung cancer disease driven by other mechanisms.
This leads to the hypothesis that PKC may also have an essential role keeping oncoproteins in check [ 24 ]. Based on our results, we suggest that the association to PFS for the different PKCisozymes is connected to mutational and smoking status. Results from gene expression analysis performed on 186 NSCLC samples strengthen these observations. Proteins associated with PFS in subgroups of NSCLC Interestingly, high expression of B7-H3, a molecule involved in immune checkpoint signaling, was correlated to poor outcome in our study, especially in smokers, those without any detected mutations in KRAS or TP53, and in those harboring an EGFR mutation. B7-H3 is a molecule known to inhibit T-cell activation in an immune suppressive manner. This protein has been shown to be linked to poor survival in cancer, and have been suggested as a new immune checkpoint target [ 27 ]. In a recent study of lung cancer patients, expression of B7-H3 was associated with overall survival only in smokers [ 28 ]. This indicates that future anti-B7-H3 therapy may have higher success rate among ever smoking lung cancer patients with KRAS or TP53 wild type tumors, or an EGFR mutation.
Within subcluster 2a, eight proteins were significantly associated with PFS, where low expression of myosin II showed the highest correlation with better PFS. This is also supported by a protein study on early stage lung cancer where myosin IIa was reported to be upregulated in stage Ia/Ib lung cancer patients with early relapse [ 29 ]. Interestingly, low expression of YAP, phosphorylated YAP(s127) and phosphorylated HSP27(s82) were associated with increased PFS in subcluster 2a. It has been shown that high expression of HSP27 leads to less phosphorylated YAP(s127). Further, phosphorylation of YAP on S127 decreased the activity of YAP since this prevent its translocation to the nucleus [ 30 ]. This also means that un-phosporylated YAP promotes tumor aggressiveness and is related to poor prognosis which is in line with our study. These finding highlight the central role HSP27 has in several pathways, including the Hippo pathway. . This is in line with our study, where these proteins showed a high correlation value (Rho > 0.6). Interestingly, HER2 was highly correlated in the breast cancer study across all subtypes, but this protein was poorly correlated with mRNA expression in our lung study (Rho =0.17). On the other hand, EGFR revealed Rho=0.72 in our study, while in breast cancer a correlation between R=0.15 -0.3 was found. This indicates that genes known to be deregulated in a specific cancer type may be regulated by other mechanisms. Proteins such as p53, CDKN1B, and MAPK14 showed very low correlation with the mRNA expression both in our study on lung cancer and in the breast cancer study    Table 2. Top twenty proteins associated with PFS (p<0.05) in the total group of samples (n=80).    Table. List of proteins and the types of antibodies used in the RPPA analysis S2

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.