The Kinome of Human Alveolar Type II and Basal Cells, and Its Reprogramming in Lung Cancer

Abstract

The discovery of mutant tyrosine kinases as oncogenic drivers of lung adenocarcinomas has changed the basic understanding of lung cancer development and therapy. Yet, expressed kinases (kinome) in lung cancer progenitor cells, as well as whether kinase expression and the overall kinome changes or is reprogrammed upon transformation, is incompletely understood. We hypothesized that the kinome differs between lung cancer progenitor cells, alveolar type II cells (ATII), and basal cells (BC) and that their respective kinomes undergo distinct lineage-specific reprogramming to adenocarcinomas and squamous cell carcinomas upon transformation. We performed RNA sequencing on freshly isolated human ATII, BC, and lung cancer cell lines to define the kinome in nontransformed cells and transformed cells. Our studies identified a unique kinome for ATII and BC and changes in their kinome upon transformation to their respective carcinomas.

Lung cancer is the leading cause of cancer death in the United States (1, 2). Lung cancer is categorized histologically, with adenocarcinoma of the lung accounting for approximately 40% of diagnoses, whereas squamous cell carcinoma accounts for 25–30% of lung cancers (3). Alveolar type II (ATII) cells, cuboidal cells preferentially located in alveolar corners as solitary cells, comprise 15% of all lung cells (4). These cells have significant functions in the normal lung, serving as the cellular source of pulmonary surfactant, performing transepithelial transport to keep the alveoli relatively free of fluid, repairing and regenerating the alveolar epithelium after lung injury, and assisting in lung innate immunity. Many lines of evidence implicate ATII cells as a lung adenocarcinoma cell of origin (5). The ATII markers thyroid transcription factor 1 and surfactant proteins (6–8) are expressed by the vast majority of adenocarcinomas. In a Jaagsiekte sheep retrovirus model, infected, proliferating ATII cells led to lung adenocarcinomas (9). Prolonged fibroblast growth factor 9 (FGF9) overexpression in the mouse lung resulted in pulmonary adenocarcinomas arising from ATII cells (10). ATII cells in transgenic mouse lines that constitutively overexpress the oncogene c-Myc and a secretable form of epidermal growth factor receptor (EGFR) developed multifocal bronchioloalveolar hyperplasia, adenomas, and carcinomas (11–13). The ATII-specific marker pro–surfactant protein C is produced in a K-Ras conditional mouse model of lung adenocarcinoma (LSL-K-ras G12D) (14). Finally, ATII cells have been shown to be the predominant, if not the only, cell of origin for K-Ras–driven pulmonary adenocarcinomas (13, 15–17).

Basal cells (BC), named for their proximity to the underlying airway basal lamina, are an integral part of the pulmonary airway epithelium and represent 6–30% of the lung cell population, varying along the proximal–distal axis (18). BC are multipotent adult tissue stem cells that can self-renew and generate differentiated progeny in the airways. Squamous cell carcinomas are thought to arise from these cells. Pulmonary squamous cell carcinomas typically arise in the upper airways in the same location as BC and express BC markers. Studies in genetically modified mice show that BC overexpression of Sox2 (sex-determining region Y box 2) in a Cdkn2ab (cyclin-dependent kinase inhibitor 2AB)/Pten (phosphatase and tensin homolog)–null background drives squamous cell carcinoma formation (19). In addition, the BC gene expression signature closely resembles the human squamous cell carcinoma gene signature, further supporting BC as candidate cells of origin of lung squamous cell carcinoma (20).

The discovery of mutant tyrosine kinases as oncogenic drivers of lung adenocarcinomas and the clinical success of molecularly targeted therapy have changed the basic understanding of lung cancer development and therapy. Yet, the expressed kinases (kinome) in lung cancer progenitor cells and whether kinase expression and the overall kinome changes or is reprogrammed in transformed cells are incompletely understood. Understanding these changes may provide further insight into the transformation process, identify chemopreventive and/or therapeutic strategies, and avoid targets that might result in ATII and BC toxicity that could lead to loss of lung homeostasis and regenerative properties.

We hypothesized that the kinome differs between ATII and BC and that their respective kinomes are reprogrammed in adenocarcinomas and squamous cell carcinomas. To test our hypothesis, we defined the normal kinome in nontransformed cells and identified kinome changes in transformed cells through RNA sequencing (RNA-seq) of freshly isolated human ATII cells and BC and in lung cancer cell lines. Our studies have identified a unique kinome for ATII and BC, as well as changes in their respective kinome in transformed cells that could provide insight into the pathobiology of transformation and identify potential therapeutic targets.

Methods

Cell Culture

A549, CALU-3, NCI-H1650, NCI-H1975, NCI-H226, NCI-H520, HCC15, SK-MES-1, and SW900 cells were obtained from the University of Colorado Cancer Center Tissue Culture Core and cultured as reported previously (21, 22).

Isolation and Culture of Human ATII Cells

Deidentified human lungs suitable for transplant but for which no recipient match was found were donated for medical research and obtained through the National Disease Research Interchange (Philadelphia, PA) and the International Institute for the Advancement of Medicine (Edison, NJ). Donor lungs were maintained at Po₂ greater than 225 mm Hg with Fi_O2 of 1, with no evidence of infection or consolidation. Lung donors ranged in age from 39 to 66 years old. Two donors were nonsmokers and two were former smokers. Two donors were female and two were male. Cells were isolated as previously published (23–25). There is greater than 90% ATII cell purity using this isolation method (26).

Isolation and Culture of BC

Airway BC were isolated from endobronchial biopsies at lobar carinas collected from individuals undergoing bronchoscopy for research purposes, under a Colorado Multiple Institutional Review Board–approved protocol. Subjects in this study were male former smokers between the ages of 41 and 71 years with sputum atypia and without evidence of airway obstruction by pulmonary function testing. Biopsy tissues were digested with dispase/collagenase/trypsin, and cells were cultured on an irradiated 3T3 fibroblast feeder layer in serum-supplemented EpiCult-B medium (STEMCELL Technologies) (27, 28). Clonal BC were separated from the feeders by differential trypsinization and used to prepare RNA.

RNA Sequencing and Quantitation

Transcriptome libraries were prepared following the Applied Biosystems (ABI) SOLiD Total RNA-Seq protocol and sequenced on an ABI SOLiD 5500 platform using 75-bp by 35-bp paired-end reads.

Definition of Kinase Genes and Their Expression

The list of 531 human kinases used for the analysis was taken as the combination of kinases from the KinHub List of Human Kinases (kinhub.org/kinases.html) and the Cell Signaling Technology list of Kinase-Disease Associations (www.cellsignal.com/common/content/content.jsp?id=science-tables-kinase-disease). The final list of 531 kinases and their annotations mapped to 551 transcripts used in the analysis is shown in Table E1 in the data supplement. A comparison of expression levels between exons and intergenic regions was used to find a threshold for detectable expression above background using the method defined by Ramsköld and colleagues (29) (Figure E1).

Microarray Data Comparison

RNA-seq data from ATII cells and BC were compared with publicly available microarray data available from the Gene Expression Omnibus (accession nos. GSE30723 and GSE24337, respectively) (30–32).

Bioinformatic Analysis

All bioinformatic analyses were performed using packages and custom scripts in the R statistical computing language (version 3.3.2; www.R-project.org). Principal component analysis (PCA) was performed using the prcomp (scale = TRUE) R function and visualized using the scatter3d function of the rgl_0.96.0 and car_2.1-4 R packages. Unsupervised hierarchical clustering was performed on the RNA-seq data using the heatmap_1.0.8 R package with the default options of Euclidean distance, complete linkage clustering, and addition of the row-scaling option.

Differential expression was determined using the DESeq2_1.14.1 package in R (33). Transcripts with fewer than 10 reads were removed. Four separate analyses were performed: 1) ATII cells as the reference against BC, 2) ATII as the reference against adenocarcinoma cell lines, 3) BC as the reference against ATII, and 4) BC as the reference against squamous cell carcinoma cell lines. Selected kinases that were differentially expressed at a significant level had a false discovery rate (FDR) of 0.1 (i.e., adjusted P value) when the Benjamini-Hochberg FDR correction was applied.

Results

RNA-Seq of Primary Human ATII Cells and BC

ATII cells were isolated from four human donor lungs as previously reported (23–26). Because the lungs were from anonymous donors, complete medical history was unknown. The number of ATII cells recovered after isolation ranged from 100 × 10⁶ to 200 × 10⁶, with no major difference in the number of ATII cells recovered among the donor lungs. Previous studies have shown that these cells maintain ATII markers expressing surfactant protein A, pro–surfactant protein C, epithelial cell adhesion molecule, and cytokeratin (23) and are greater than 90% ATII cells (26).

BC were isolated as outlined in the methods. All four subjects described in this study were male former smokers without evidence of airway obstruction. Clonal BC grown from bronchial biopsies expressed known BC markers, including keratin 5, tumor protein p63, and keratin 14 (27).

Total RNA was isolated, and mRNA was extracted from each sample. Bar-coded libraries were constructed in three to five replicates per sample group to ensure data reproducibility. A total of 38 libraries were run on an ABI SOLiD 5500 platform, and sequences were aligned to the hg19 reference genome and annotated to RefSeq transcripts. Sample results ranged from 27 million to 57 million reads per library with approximately 80% mapping. To correct for transcript length and depth of coverage, raw read counts of transcript expression were converted to reads per kilobase of exon per million mapped reads (RPKM). Of the 24,364 annotated RefSeq transcripts, 22,208 had counts in at least 1 of 38 samples, with a median RPKM of 1.74 for those transcripts. Correlation of transcripts between replicates in a sample group was high over the entire exome with Pearson correlation ranging from 0.80 to 0.96 for ATII samples, from 0.88 to 098 for BC, and from 0.81 to 0.99 for all cell lines studied (Table E2).

As a further control for data validity, a comparison was made between RNA-seq data and publicly available microarray data for ATII cells and BC (Figures 1A and 1B) (31, 32). In considering all transcripts, results were highly concordant between the datasets (ATII, 14,015 genes measured in common, Pearson’s correlation r = 0.53, Spearman’s correlation r = 0.84; BC, 16,320 genes measured in common, Pearson’s correlation r = 0.58, Spearman’s correlation r = 0.86). These correlation values obtained from unrelated samples are consistent with previous studies that compared microarray and RNA-seq data of identical samples and reported Pearson’s correlations of 0.53–0.65 and Spearman’s correlations of 0.76–0.82 (34, 35).

Figure 1.

Comparison of RNA sequencing (RNA-seq) of all transcripts of human primary (A) alveolar type II (ATII) cells and (B) basal cells (BC), as well as of kinase-only transcripts of human primary (C) ATII cells and (D) BC, with publicly available microarray data. Each circle represents an individual (A and B) RNA or (C and D) kinase transcript. Microarray data, as measured by the Affymetrix Human Genome U133 Plus 2.0 Array, were downloaded as a series matrix files for six ATII cell control samples (donors 39, 40, and 46 at 4 h and 24 h; Gene Expression Omnibus [GEO] accession no. GSE30723) and five BC samples (GEO accession no. GSE24337). Probe set identifiers were mapped to the RNA-seq Human Genome Organisation gene symbol identifiers using the hgu133plus2.db package in the R statistical computing language (version 3.3.2; www.R-project.org). Kinases that remained without probe set identifiers were resolved by hand using the NetAffx web resource (www.affymetrix.com/analysis/index.affx). Any string “dup” that existed in a transcript’s identifier indicating duplicate genes was ignored. If multiple transcript identifiers existed for the same gene in either the RNA-seq data or the microarray, correlation was calculated using the transcript with the maximum mean expression value. pcor = Pearson’s correlation; RPKM = reads per kilobase of exon per million mapped reads; scor = Spearman’s correlation.

Identification of a Specific ATII and BC Kinome and Kinase Signature

We next focused on identification of the ATII and BC kinome. The kinome was defined as 531 kinases mapped to 551 transcripts identified in the entire dataset. In considering only kinase transcripts, the Pearson’s correlation of expression values among replicates in a sample group was high, ranging from 0.91 to 0.97 for ATII cells, from 0.93 to 0.98 for BC, and from 0.90 to 0.99 for all cell lines (Table E2). Comparing our kinase RNA-seq data with published microarray data as a second control for kinase data validity (Figures 1C and 1D), we again found highly concordant data, supporting the high quality of our RNA-seq kinase data (ATII, 475 kinases in common, Pearson’s correlation r = 0.63, Spearman’s correlation r = 0.70; BC, 524 kinases in common, Pearson’s correlation r = 0.55, Spearman’s correlation r = 0.79).

To define which kinases were “expressed” in a sample, a comparison between the expression levels of exons and intergenic regions was used to define a threshold value above which there was the highest confidence in the validity of the expression level (29). Through this comparison, transcripts with an RPKM at or above 0.1 were considered to represent expressed transcripts in a sample. Using this threshold, a range of 463–469 kinase transcripts were identified as expressed out of 551 known human kinase transcripts (representing 531 unique kinases) in ATII replicates, with 449 transcripts expressed in all four ATII cell replicates. For expressed kinase transcripts in ATII samples, the median RPKM was 5.925. The top five expressed kinases in ATII cells (mean RPKM) were EFNA1 (146.87), TRIM28 (141.89), TGFBR2 (132.36), DDR1 (102.31), and LRRK2 (96.17).

A range of 447–488 kinase transcripts were identified as expressed in BC replicates, with 439 expressed in all four BC replicates. For expressed kinase transcripts in BC, the median RPKM was 6.185. The top five expressed kinases in BC (mean RPKM) were CSNK1A1 (163.87), TRIM28 (135.60), EPHA2 (117.15), PLK2 (Polo-like kinase 1; 113.78), and MKNK2 (111.84).

PCA (Figure 2A) of the kinome expression data (531 expressed transcripts) was performed comparing ATII with BC. The PCA plot shows that for the two lineages, individual samples of each cell type clustered together and lineages were separable based on their kinome. Unsupervised hierarchical cluster analysis also distinguished ATII from BC by their kinome (Figures 2B and E2).

Figure 2.

Identification of the ATII cell and BC kinome. (A) Principal component analysis (PCA) of kinase expression data for ATII cells (blue) and BC (green). (B) Unsupervised hierarchical clustering of kinase mRNA expression of ATII and BC. Expression of the 531 kinase transcripts with nonzero expression is shown for each transcript (rows), scaled relative to the mean expression of the transcript across samples (columns) (red = above mean; blue = below mean). PC = principal component.

Plotting the mean log RPKM of ATII kinases versus the mean log RPKM of BC kinases identified large expression differences in a select group of kinases (Figure 3). Higher expression of ROS1 and LRRK2 (leucine-rich repeat serine/threonine-protein kinase 2) was seen in ATII than in BC, for example, whereas PDGFRA (platelet-derived growth factor receptor-α) and PDGFRB (platelet-derived growth factor receptor-β) had higher expression in BC than in ATII.

Figure 3.

Plot of mean log RPKM of ATII cells versus mean log RPKM of BC. Each circle represents a kinase. Red circles represent kinase expression that is outside the first and third quartiles of the difference of the means.

To identify kinases that were differentially expressed between ATII and BC in a statistically significant manner, we performed volcano plot filtering between the two cell types (Figure 4). A transcript was considered differentially expressed if it was significantly different (threshold FDR = 0.1), and a minimum twofold change threshold between the groups was required (see Methods). Using these criteria, we identified 87 differentially expressed kinases, with 54 upregulated in ATII with respect to BC and 33 upregulated in BC with respect to ATII (Tables 1 and E3). This defines an ATII-specific kinase signature as a set of 54 transcripts upregulated that were consistently different between ATII and BC, as well as a BC-specific kinase signature as a set of 33 upregulated transcripts consistently different between ATII and BC (Figure E2).

Figure 4.

Volcano plot of ATII cells versus BC. Volcano plot of log₂ fold change and −log₁₀ adjusted P value (p.adj). Each circle represents an expressed kinase. Significant results (red circles) have p.adj less than or equal to 0.1 and at least a twofold change between the cell types. Volcano plots were developed by plotting the maximum pairwise fold change between the regularized log-transformed mean value of the reference and any nonreference sample group in a given comparison (x-axis) against the minimum p.adj over the extracted pairwise contrasts (y-axis).

Table 1.

Kinase-Specific Signature of Alveolar Type II Cells and Basal Cells

ATII Kinase Signature (Kinases Different between ATII and BC in Decreasing Order of Fold Change)				BC Kinase Signature (Kinases Different between ATII and BC in Decreasing Order of Fold Change)
1	ROS1	28	ACVR1C	1	PDGFRB	28	MAP2K6
2	LRRK2	29	EPHA7	2	PDGFRA	29	PRKACG
3	NRK	30	DAPK1	3	DDR2	30	FLT1
4	AATK	31	EFNA1	4	LCK	31	ALPK2
5	NPR1	32	CDKL5	5	PLK2	32	PNCK
6	ERBB4	33	ALPK3	6	SPEG	33	SRPK3
7	DAPK2	34	NEK8	7	CAMK2A
8	STK33	35	CDK20	8	EFNA3
9	MGC42105	36	TRIB3	9	AXL
10	TNIK	37	TRPM6	10	TEK
11	HUNK	38	KSR2	11	MYO3B
12	MAPK4	39	PAK7	12	WNK4
13	PKDCC	40	DYRK1B	13	JAK3
14	KDR	41	PDK2	14	MARK1
15	CDKL2	42	TEC	15	EFNB1
16	EPHA10	43	CLK1	16	AKT3
17	ROR1	44	PRKCE	17	CSNK1A1L
18	EPHA4	45	STK31	18	ACVRL1
19	MERTK	46	STK36	19	CDK6
20	FGFR2	47	LMTK2	20	ZAK
21	PRKG1	48	TRIM66	21	HCK
22	SBK1	49	DCLK1	22	MAPK12
23	PKN1	50	C9orf96	23	KALRN
24	RPS6KA2	51	PRKD1	24	EPHB2
25	ADRBK2	52	LOC283070	25	CSF1R
26	PRKCZ	53	KIT	26	MAPK11
27	FLT4	54	MAPK15	27	GRK5

Definition of abbreviations: ATII = alveolar type II cells; BC = basal cells.

To determine the gene functions affected by the cell type–specific differentially expressed kinases, we performed DAVID functional analysis on the set of signature kinases (Table E4). The set of 54 upregulated kinases in ATII cells was enriched by nominal P values in categories such as lung alveolar development (FGFR2, FLT4 [FMS-like tyrosine kinase 4], PKDCC [protein kinase domain containing, cytoplasmic]) and transmembrane receptor protein kinase activity (FGFR2, ERBB4, FLT4, ROR1 [receptor tyrosine kinase–like orphan receptor 1], KIT, MERTK [MER proto-oncogene, tyrosine kinase], ROS1, and KDR [kinase insert domain receptor]). The set of 33 BC upregulated kinases involved functions determined largely by the two key kinases PDGFRA and PDGFRB, positive regulation of PI3K activity and signaling, positive regulation of phospholipase C activity, positive regulation of fibroblast proliferation, and positive regulation of cell proliferation by vascular endothelial growth factor–activated platelet-derived growth factor receptor signaling pathway.

Kinome Reprogramming upon Conversion of ATII and BC to Lung Cancer

With the definition of the ATII and BC kinome, we next determined the expressed kinome in pulmonary adenocarcinoma (A549, NCI-H1650, NCI-H1975, CALU-3) and squamous cell carcinoma cell lines (HCC15, NCI-H520, SW900, SK-MES). Of the adenocarcinoma cell lines, NCI-H1975 and NCI-H1650 have EGFR mutations that function as driver oncogenes, whereas A549 and CALU-3 do not. The kinome of each of the cell lines was identified as previously outlined for ATII and BC. We again performed PCA (Figure 5) to determine whether the expressed kinome of the lung cancer cell lines could be separated from their progenitor cell types. PCA was able to separate primary, nontransformed cells from each of the carcinoma cell lines, with replicates of the individual samples again clustering together by lineage. The squamous cell carcinoma cell line kinome (Figure 5, red dashed line) was separable from the BC kinome (Figure 5, light green circle). The adenocarcinoma kinome (Figure 5, blue dashed line) was also separable from the ATII kinome (Figure 5, dark green circle). In addition, the transformed cells segregated into an overlapping subcluster of adenocarcinoma and squamous cell carcinoma. This separation of the normal ATII and BC kinome from the carcinoma cell lines supports our hypothesis that the expressed kinome of the progenitor cells changes upon conversion to carcinoma and that it changes in a lineage-specific fashion.

Figure 5.

PCA of kinase expression for all cell types. Primary cell samples (ATII cells and BC) are shown in green shades, adenocarcinomas (ADENO) in blue shades, and squamous cell carcinomas in red shades.

Unsupervised hierarchical clustering confirmed that in individual samples, the kinome of replicates of each cell type (ATII, BC, adenocarcinoma, squamous cell carcinoma) clustered together (Figure 6). Two major clusters were identified; nontransformed cells (ATII, BC) clustered separately from transformed cell lines. In addition, the kinome of squamous cell cancers generally segregated from adenocarcinomas, supporting our PCA.

Figure 6.

Hierarchical clustering of kinome data for all cell types. Expression of the 531 kinase transcripts with nonzero expression is shown for each transcript (rows), scaled relative to the mean expression of the transcript across samples (columns) (expression scale, red = above mean; blue = below mean). Ad = adenocarcinoma; Sq = squamous cell carcinoma.

To identify kinases that were differentially expressed in transformed cells versus nontransformed precursor cells, a differential expression analysis was performed with differential expression defined as a transcript significantly different in at least one of the cancer cell lines compared with nontransformed precursor cells, using a threshold FDR of 0.1, and also had a minimum twofold change (see Methods). Using these criteria, for ATII, 304 differentially expressed kinases were identified, with 40 upregulated and 20 downregulated with respect to all four adenocarcinoma cell lines (Tables 2 and E5). Functional enrichment analysis of the ATII kinases upregulated with respect to the adenocarcinoma cell lines highlighted categories of organ morphogenesis (FGFR2, NEK8, and SYK) and mitogen-activated protein kinase activity (MAPK13, MAPK4, MAPK15, and MAPK10). Among the downregulated kinases, significant categories include a number of functions driven by the kinases PLK1, NEK2, BUB1, BRSK2, PKMYT1, BUB1B, AURKA (Aurora kinase A), and AURKB (Aurora kinase B) and related to mitosis, such as spindle organization, G₂/M transition of mitotic cell cycle, and sister chromatid cohesion. Other functions include cell division (NEK2, BUB1, BRSK2, BUB1B, AURKA) and cell proliferation (ZAK, PLK1, BUB1 [budding uninhibited by benzimidazoles 1], BUB1B, AURKB), which are expected to be upregulated in cancer cells.

Table 2.

Differential Expression of Kinases between Progenitor Cells and Lung Cancer

Differential Expression of Kinases between ATII and Lung Adenocarcinoma Cell Lines				Differential Expression of Kinases between BC and Lung Squamous Cell Carcinoma Cell Lines
ATII Upregulated Kinases vs. Adenocarcinomas (n = 40)		ATII Downregulated Kinases vs. Adenocarcinomas (n = 20)		BC Upregulated Kinases vs. Squamous Cell Carcinomas (n = 27)			BC Downregulated Kinases vs. Squamous Cell Carcinomas (n = 5)
1	LRRK2	1	FGFR4	1	EPHA1	1	PKDCC
2	ROS1	2	EFNA2	2	ERN2	2	PKN1
3	ERN2	3	MAST1	3	RIPK3	3	CAMK4
4	RPS6KA2	4	MAP3K14	4	PTK6	4	TTK
5	NRK	5	MYLK2	5	MAST4	5	CDK4
6	RIPK3	6	AURKA	6	MAPK13
7	SYK	7	BRSK1	7	MAPK10
8	AATK	8	MAPK12	8	LMTK3
9	FGFR2	9	BRSK2	9	FGFR3
10	IRAK3	10	NEK2	10	ERBB2
11	DAPK2	11	BUB1	11	GUCY2D
12	PRKG1	12	AURKB	12	FGFR2
13	MAPK13	23	CHEK1	13	EFNB1
14	STK32A	24	GSG2	14	ERBB3
15	GUCY2D	25	PLK1	15	ANKK1
16	NPR1	26	PLK4	16	ACVRL1
17	MAPK10	27	TTK	17	FGR
18	PRKG2	28	PKMYT1	18	MYO3B
19	EPHA4	29	ZAK	19	EFNA3
20	MAPK4	30	BUB1B	20	CDC42BPG
21	KDR			21	CSNK1A1L
22	MGC42105			22	PRKCD
23	MAPK15			23	PRKACG
24	CDKL2			24	LIMK2
25	TNNI3K			25	SRMS
26	FGFR3			26	PAK6
27	MAST4			27	WEE2
28	PRKCZ
29	FES
30	PAK7
31	WNK3
32	FGR
33	EFNA1
34	YSK4
35	TTN
36	C9orf96
37	ERN1
38	CLK1
39	NEK8
40	CDKL4

In BC, 330 differentially expressed kinases were identified, with 27 upregulated and 5 downregulated with respect to all five squamous cell carcinoma cell lines (Tables 2 and E6). There was minimal overlap of differentially expressed kinases between adenocarcinomas and squamous cell carcinomas. Of the 67 kinases making up both signatures, only 9 kinases were present in both (13%). Functional enrichment analysis of the upregulated kinases identified significant categories of cell–cell signaling (FGFR2, FGFR3, EFNB1, EFNA3), phosphatidylinositol-4,5-bisphosphate 3-kinase activity (FGFR2, FGFR3, ERBB3, ERBB2), ERBB2 signaling (ERBB3, PTK6, ERBB2), and negative regulation of cell adhesion (ACVRL1, ERBB3). Owing to the small number of BC downregulated kinases, no significant functional categories were found.

PLK1, AURKA, and AURKB were highly differentially expressed in adenocarcinoma cell lines compared with ATII cells (Table 2 and Figure 7A). PKDCC, PKN1 (protein kinase N1), CAMK4 (calcium/calmodulin-dependent protein kinase type IV), TTK (TTK protein kinase), and CDK4 (cyclin-dependent kinase 4) were highly differentially expressed in the squamous cell carcinoma cell lines versus BC (Table 2 and Figure 7B). This differential expression points to kinome reprogramming on malignant conversion and potential drug targets.

Figure 7.

Kinases significantly differentially expressed in (A) pulmonary adenocarcinomas versus ATII cells and in (B) pulmonary squamous cell carcinomas versus BC. rlog = regularized log-transformed value.

Discussion

In these studies, we hypothesized that the kinome has plasticity and that in transformation of a normal lung cell into a lung cancer, the kinome is reprogrammed. To test this hypothesis, we defined the normal kinome in nontransformed primary cells that are precursors of lung cancer (ATII and BC) and the kinome of the respective lung cancers, adenocarcinomas, and squamous cell carcinomas, studying the entire kinome in a lineage-specific fashion.

The kinome of nontransformed cells was surprisingly distinct between ATII and BC. In addition, a unique signature could be developed to classify and distinguish these lineages. The kinome was also relatively uniform between genetically different individual samples because PCA found their individual kinomes to cluster together.

To determine whether the kinome was reprogrammed upon transition from nontransformed to transformed cells, transcript levels of human kinases expressed in ATII, BC, and transformed tissue derived from these precursor lineages (pulmonary adenocarcinoma and squamous cell carcinoma) were defined and compared. Comparing the expressed kinome of ATII cells with the expressed kinome of adenocarcinomas, we found that the normal ATII kinome is distinct from its malignant derivative. The same is true comparing the BC kinome with the squamous cell carcinoma kinome. Thus, on the basis of definition of the transcriptionally active kinomes, we were able to distinguish two broad categories of tissue, transformed and nontransformed, which conformed to their known biology—ATII cells, BC, adenocarcinoma, and squamous cell carcinoma. These could be further separated into distinct tissue clusters based on their expressed kinome. The results suggest that the transcriptional activities of kinases define tissue types in a biologically meaningful way, despite the fact that the activity of kinases is strongly regulated by post-translational events.

The idea of kinome reprogramming has been seen in breast cancers undergoing chemotherapy (36, 37), in non–small cell lung cancer (NSCLC), and in small cell lung cancer (SCLC). In NSCLC, reprogramming of tyrosine kinases to drive alternate pathways to support tumor cell growth and survival has been reported (38). After exposure to EGFR tyrosine kinase inhibitors, studies have identified Met (hepatocyte growth factor receptor) activation with increased MET expression through gene amplification and increased expression leading to EGFR tyrosine kinase inhibitor resistance as a kinome change driving chemoresistance (39–41). In SCLC, kinase reprogramming is associated with MYC expression whereby MYC drives an SCLC histologic subtype that is vulnerable to Aurora kinase inhibition, whereas the classic phenotype with low MYC expression is not (42). Thus, molecular characterization and kinase vulnerability can define subgroups of cancers. The reprogramming of the kinase environment might be used to define cancer subgroups with distinct therapeutic targets. By defining kinases that are highly expressed in lung cancers but not in their respective precursor cell, it is interesting to consider whether therapeutic targets might be identified and avoid toxicity to normal lung cells.

Kinases highly expressed in adenocarcinoma cell lines, but not in ATII cells, included PLK1, AurkA, and AurkB. In vitro PLK1 affects lung cancer as a synthetic lethal partner of the mutant RAS pathway (43) and decreases tyrosine kinase inhibitor resistance in NSCLC with an EGFR T790M mutation (44). AURKB also is being considered as a therapeutic target in NSCLC with responses seen in vitro (45–47). In addition, kinome reprogramming as a mechanism of resistance suggests that cotargeting multiple pathways may be required for effective therapy.

Similarly, looking for differentially expressed kinases in squamous cell carcinomas versus BC also identified potentially novel therapeutic targets: PKDCC, PKN1, CAMK4, TTK, and CDK4. PKDCC is a surprising finding because its function as a protein kinase occurs in bone growth. However, members of the PKDCC family are involved in the regulation of the JNK (Janus kinase)-dependent Wnt/planar cell polarity signaling pathway. PKN1 also impacts the Wnt signaling pathway by inhibiting the Wnt/β-catenin pathway through inhibition of WNT3A-dependent phosphorylation of LRP6 (low-density lipoprotein receptor-related protein 6). The dysregulation of the Wnt/β-catenin pathway occurs in many cancers, including lung cancer, with altered β-catenin expression, or WNT1, WNT3A, and WNT5A overexpression, associated with poor prognosis in patients with NSCLC (48). CAMK4 is an upstream regulator of the oncogenic mTOR/ribosomal protein S6 kinase pathway, an essential pathway for cancer cell growth and protein synthesis (49–52). Therefore, this strategy may identify logical therapeutic targets in lung cancer. However, potential targets were identified through transcriptional differences, not functional importance, which must be confirmed through further testing.

In these analyses, we have compared nontransformed cells with fully transformed cells, representing the two ends of the spectrum of the transformation process. We have not defined the stepwise transition that likely occurs during the transformation process as the kinome reprograms. This would require additional analysis of dysplastic lesions or adenocarcinoma in situ to define kinome reprogramming longitudinally. Intrinsic to this discussion is whether there is a causal role for the kinome in tumorigenesis. This clearly is the case in kinases that are individual driver oncogenes, such as mutant EGFR, ALK, ROS1, and ERBB2. However, whether there are nonmutated kinases that play a role in cell survival and could lead to tumorigenesis is unknown.

Our analysis has limitations in that we have used cells in culture to determine patterns of gene expression and extrapolated that to the in vivo kinome. Though a logical starting point, there are caveats to consider in extrapolating the data because there are many events controlling gene expression that differ between in vitro and in vivo growth, such as growth media, adherence, and microenvironment. In addition, the studied samples were not matched for smoking history. Smoking did affect the BC transcriptome when active smokers were compared with nonsmokers (53). However, it is not clear whether former smoking status, as in our samples, continues to affect the kinome. Considering how the ATII and BC samples clustered together in the PCA without stratifying for smoking status, improving homogeneity of the study population may improve the clustering further.

The transcriptional levels of the kinase genes as well as their predicted functional associations are likely to be essential when exploring the role of kinases in disease. We have defined a lineage-specific kinase signature for ATII and BC. By also defining the kinome in adenocarcinomas and squamous cell carcinomas, we have shown that the kinome changes, or reprograms upon transformation. The comparison of these signatures might guide a rational prediction for therapy. Systematic analysis of kinase transcriptional activities across all healthy tissues could help to prioritize those kinases whose activation is most disease specific for further study and whose inhibition would theoretically cause fewer side effects.