Adenylate Kinase-4 Is a Marker of Poor Clinical Outcomes That Promotes Metastasis of Lung Cancer by Downregulating the Transcription Factor ATF3

Lung cancer remains the most common cause of cancer-related death (1). Despite advances in treatment modalities, including surgical resection, radiation therapy, chemotherapy, target therapy, and combinations of these therapies, the prognosis for patients with lung cancer remains poor. Because of the lack of accurate prognostic markers, the 5-year survival rate for all stages combined is only 15%, and only 16% of lung cancers are diagnosed at an early stage (2). Therefore, the identification and characterization of genes involved during the progression of lung cancer may be valuable for discovering novel prognostic markers and therapeutic targets. Analyzing the gene expression profiles of cells with invasive transformation is an attractive approach for identifying putative lung cancer progression biomarkers that can predict survival (3). In 1997, a cell model of invasive transformation was developed to select progressively invasive cell populations from a parental cell line of human lung adenocarcinoma (AdC), CL1. Five progressive subclones namely, CL1-1, CL1-2, CL1-3 CL1-4, and CL1-5 were selected using transwell and displayed increasing invasion potential (4). In 2001, Chen and colleagues published a microarray analysis comparing CL1-0 versus CL1-5 and found that members of the adenylate kinase family were aberrantly upregulated during the process of invasive transformation (5). Similarly, by comparing gene expression profile, adenylate kinase-4 (AK4) was also found to be significantly overexpressed in CL1-5 compared with CL1-0 (6). However, the association between AK4 and human cancer has not been studied thoroughly.

Adenylate kinases catalyze the transfer of phosphate from 1 molecule of ATP or GTP to AMP, thereby generating 2 molecules of ADP, which contribute to maintaining energy homeostasis by balancing cellular adenine nucleotide composition. To date, 7 known isoforms of adenylate kinase have been identified and named AK1 to AK7 according to the order of their discovery (7, 8). AK4 is located in the mitochondrial matrix and has a high degree of sequence homology with AK3 (9–11). Unlike AK3, AK4 has been reported to be enzymatically inactive in vitro, but still retains its nucleotide-binding ability, physically interacts with the mitochondrial ADP/ATP translocator, and acts as a stress-responsive protein to maintain cell survival (12). In this study, we sought to investigate the potential role of AK4 in lung cancer progression. We assessed the correlation between the expression of AK4 and clinicopathological features, as well as patient survival. In addition, both in vitro and in vivo assays were used to study the function of AK4. Our results showed that AK4 is required to promote invasion and metastasis in lung cancer, thereby highlighting a critical role of AK4 in maintaining the invasive phenotype of lung cancer.

The tissues used were obtained from Kaohsiung Medical University Hospital with Institutional Review Board approval (KMUH-IRB-2011-0286). We used formalin-fixed paraffin-embedded surgical specimens for immunohistochemical staining. The histologic diagnosis was made according to WHO standard. Tumor size, local invasion, lymph node involvement, distal metastasis, and final disease stage were determined according to the TNM classification for lung cancer (13). Follow-up of the patients was carried out for up to 200 months.

Human lung AdC cell lines, CL1-0 and CL1-5, and squamous cell carcinoma, H520 were grown in RPMI 1640 supplemented with 10% FBS (Invitrogen). Human lung AdC cell lines (A549, PC13, and PC14) and large cell carcinoma, H1299 were grown in DMEM plus 10% FBS (Invitrogen). All cells were kept under humidified atmosphere containing 5% CO₂ at 37°C. CL1-0 and CL1-5 were established by Chu and colleagues and displayed progressively increasing invasiveness (4). PC13 and PC14 were developed by Lee and colleagues at National Cancer Center Hospital, Tokyo, Japan (14). Other lung cancer cell lines (A549, H520, and H1299) were obtained from the American Type Culture Collection.

Western blot analysis was done with the primary antibodies anti-AK4 (Icon Biotechnology, 1:2,000), anti-ATF3 (Santa Cruz Biotechnology, 1:500), and anti–α-tubulin (Sigma-Aldrich, 1:5,000).

Polycarbonate filters were coated with human fibronectin on the lower side and Matrigel on the upper side. Medium containing 10% FCS was added to each well of the lower compartment of the chamber. Cells were mixed in serum-free medium containing 0.1% bovine serum albumin and loaded to each well of the upper compartment. After 16 hours, cells were then stained and counted on lower side of the membrane under a light microscope (×200, 10 random fields from each well). All experiments were conducted in quadruplicate.

All animal experiments were conducted in accordance with a protocol approved by the Academia Sinica Institutional Animal Care and Utilization Committee. Age-matched nonobese diabetic severe combined immunodeficient (NOD-SCID) female mice (6 to 8 weeks old) were used. For evaluating metastasis ability, 1 × 10⁶ cells were resuspended in 0.1 mL of PBS and injected into the lateral tail vein. Metastatic lung nodules were quantified after H&E staining using a dissecting microscope at endpoint.

All observations were confirmed by at least 3 independent experiments. The results are presented as the mean ± SD. We used 2-tailed, unpaired Student t tests for all pair-wise comparisons. Survival curves were analyzed by the log-rank Kaplan–Meier method. Cox proportional hazards regression was used to test the prognostic significance of factors in univariate and multivariate models. P values of less than 0.05 were considered to be significant.

First, we examined the expression profile of AK4 in 3 microarray datasets. From GSE10072 (15), AK4 levels in 58 lung AdC patients and 49 normal lung tissues were compared, and AK4 was significantly upregulated in lung AdC (Supplementary Fig. S1A). We next analyzed another dataset from Array Express containing 9 normal lung tissues and 49 lung AdC patients with known disease stage (16). The results showed that patients with an advanced stage tended to express higher AK4 levels compared with patients at stage I and normal lung tissues (Supplementary Fig. S1B). Furthermore, expression profiles for the AK family in 27 matched-pair lung AdC from GSE7670 showed AK4 was uniquely upregulated among other AK family members in tumors compared with their adjacent corresponding normal lung tissues (Supplementary Fig. S1C).

To validate the mRNA expression levels of AK4 in tumor tissues versus normal, we carried out RT-PCR analysis of 7 normal tissues paired with lung cancer specimens. Data showed higher expression of AK4 (2.8-, 3.1-, and 1.5-fold) in tumor with advanced stages (stage III and IV) compared with their corresponding normal samples (Fig. 1A). To examine the expression of AK4 protein in lung cancer patients, we examined 45 sets of matched samples from primary lung tumors and normal adjacent tissues. Strikingly, in 41 of 45 patients, the AK4 levels were significantly upregulated in tumors compared with normal tissues, evaluated by immunohistochemistry (IHC; Fig. 1B–C, P < 0.001). Furthermore, χ² analysis between the levels of AK4 and clinicopathological characteristics showed that overexpression of AK4 in lung cancer was associated with advanced stage (P = 0.028) and disease recurrence (P = 0.005; Supplementary Table S1). Figure 1D illustrated representative images of AK4 expression in lung AdC and SCC which showed an association between AK4 expression and advanced stage with a mean score for stage I lung cancer (1.07) that was significantly lower than that for stage III (1.67) and stage IV (1.73; Fig. 1E). Next, we determined whether AK4 expression in lung cancer was associated with poor survival. Tissues from 133 lung cancer patients were examined by IHC staining. The antibody specificity and scoring criteria were defined (Fig. 1F). AK4 expression was detected exclusively in the cytosol with granular pattern (Supplementary Fig. S2A). Data showed high expression of AK4 (score 2 and score 3) significantly correlated with reduced overall survival (Fig. 1G, P = 0.008) and disease-free survival (Fig. 1H, P = 0.002) compared with patients with low AK4 levels (score 0 and score 1). To confirm, we analyzed UM dataset from Director's Challenge Consortium (17). Survival analysis showed high levels of AK4 significantly predict worse survival compared with low level of AK4 in 204348_s_at probe (Supplementary Fig. S1D, P = 0.01). Furthermore, univariate and multivariate Cox regression analysis showed that AK4 levels, lymph node involvement and distal metastasis were found to be significant predictors of patient outcome (Table 1).

Our clinical findings suggest that AK4 may play a role in lung cancer progression. We then evaluated the functional role of AK4 on the invasiveness of lung cancer cells. Data showed AK4 protein expression was significantly elevated in CL1-5, A549, and H1299 cells, and the expression of AK4 protein was positively correlated with the invasion activity in 7 human lung cancer cell lines (Fig. 2A). To determine whether AK4 modulates lung cancer cell invasion, we silenced AK4 in CL1-5 and A549 cells using AK4-specific lentiviral shRNAs. Data showed that AK4 shRNA1 could significantly reduce AK4 protein with a concomitant inhibition of invasion potentials by approximately 50% (Fig. 2B). Complementarily, overexpression of AK4 in poorly invasive CL1-0 and PC13 cells significantly enhanced their invasive activity by 3- and 2-fold, respectively (Fig. 2C). Surprisingly, both AK4 knockdown and overexpression showed minimum impact on cell proliferation (Fig. 2D).

We next investigated the effects of AK4 expression on in vivo metastasis. We stably silenced AK4 in CL1-5 cells and inoculated these cells intravenously into NOD-SCID mice. Compared with a nonsilencing shRNA control, AK4 knockdown resulted in a significant reduction in the number of pulmonary metastatic foci (Fig. 3A). Quantification of the metastatic nodules in the lung tissues confirmed that the number of lung metastases showed ∼50% reduction in mice carrying AK4-knockdown cells compared with the control (Fig. 3B, P < 0.01). We also intravenously injected CL1-0 cells overexpressing 2 levels of AK4 (Fig. 3C, left) and assessed the presence of pulmonary metastasis by microscopic examination 8 weeks post-injection. Lung foci were observed in 40% (2/5) of mice in the AK4 M.O.I. = 5 group and 66% (4/6) of mice in the AK4 M.O.I. = 10 group, whereas a complete absence of lung foci was observed in the vector control group (Fig. 3C, right). H&E and IHC analysis further confirmed that the lung loci observed in the CL1-0 AK4 group indeed overexpressed AK4 (Fig. 3D).

To elucidate the molecular mechanisms by which AK4 contributes to the invasive phenotype of lung cancer cells, we carried out microarray analysis comparing the gene expression of CL1-5 NS versus CL1-5 shAK4 cells and CL1-0 pLenti 6.3 versus CL1-0 AK4 cells. Differentially expressed genes with at least a 1.5-fold change were identified. We then selected 4,155 genes as putative invasion inducers and 3,654 genes as putative invasion suppressors regulated by AK4 (Fig. 4A–B, top panel). Venn diagram analysis further identified common putative targets affected by AK4 in a complementary fashion (Supplementary Tables S2 and S3). Next, we conducted gene-annotation enrichment analysis using Database for Annotation Bioinformatics Microarray Analysis (DAVID; refs. 18, 19) to determine whether particular gene sets were significantly enriched in differentially expressed genes regulated by AK4. Twenty-eight functional classifications, as annotated by Gene Ontology (GO), were significantly enriched, including transcription, nucleotide binding and metabolic process, etc. (Fig. 4A–B, bottom panel). Next, we decided to focus on the differential expression of transcription regulators. ATF3 was identified as being downregulated 2.15-fold after AK4 overexpression and was upregulated by 2.14-fold after AK4 silencing (Supplementary Table S2). Interestingly, other than transcription activity, ATF3 was categorized in several metabolic processes by GO (Supplementary Table S4). To further identify targets regulated by ATF3, we examined the knowledge-based interactome surrounding the regulation of ATF3 using the Ingenuity Pathway Analysis (IPA) and overlaid with microarray data from CL1-0 AK4 with a 1.5-fold change cut-off. Several targets that were reported to be involved in the processes of invasion and metastasis were differentially regulated, including MMP2, SERPINB5, AKT, PI3K, and CTNNB1 (Fig. 4C).

To validate the expression of ATF3, we carried out Western blotting to detect the endogenous expression of AK4 and ATF3 in CL1-0 and CL1-5. Figure 5A showed that the endogenous expression of ATF3 protein in CL1-0 cells was higher than in CL1-5 cells. Western blotting further validated that ATF3 was upregulated by 2-fold in AK4-silenced CL1-5 cells compared with NS shRNA-transduced cells, whereas overexpression of AK4 in CL1-0 cells could suppress ATF3 expression by 70% compared with empty-vector-transduced CL1-0 cells (Fig. 5B). To confirm whether ATF3 was the pivotal regulator for invasion activity in CL1-0 and CL1-5 cells, we knocked down the expression of ATF3 in CL1-0 by shATF3 lentivirus and transiently overexpressed ATF3 cDNA in CL1-5 cells. The results showed that knockdown of ATF3 in CL1-0 cells could enhance the invasion activity by 2.4-fold, whereas overexpression of ATF3 in CL1-5 cells could suppress it by 37% (Fig. 5C). Furthermore, knockdown of ATF3 following AK4 silencing in CL1-5 cells abolished the inhibitory effects of ATF3 on invasion and subsequently restored the invasion activity to 72% of control. Complementarily, rescue of ATF3 expression following AK4 overexpression in CL1-0 could significantly suppress the invasion activity caused by AK4 (Fig. 5D). To further investigate the downstream targets by which ATF3 contributes to modulate invasion activity, we analyzed the transcriptional levels of ATF3 and MMP2 following AK4 overexpression. RT-PCR showed that increased ectopic expression of AK4 resulted in decreased ATF3 expression with a concurrent induction of MMP2 transcription in a dose-dependent manner (Supplementary Fig. S3A). Furthermore, knockdown of MMP2 expression could also impede the induction of invasion activity by AK4 (Supplementary Fig. S3B).

To verify the expression of ATF3 in lung cancer, we analyzed the expression of ATF3 in GSE10072. Consistent with our study, ATF3 (202672_s_at) was significantly downregulated in tumor specimens compared with normal tissue (Supplementary Fig. S4A). Furthermore, we examined the endogenous expression of ATF3 and AK4 in CL1-0, CL1-1, CL1-5, and CL1-5 F4 from GSE7670. The results showed an inverse correlation between the expression of AK4 and ATF3 (Supplementary Fig. S4B). To further verify the relative levels of AK4 and ATF3 in N/T paired lung cancers, we examined GSE7670 and the data showed the same trend with an inverse correlation between AK4 and ATF3 expression in 27 paired lung AdC samples (Supplementary Fig. S4C).

To investigate the interplay between AK4 and ATF3 in human lung cancer patients, we carried out IHC on serial sections of lung cancer specimens. Representative IHC staining for AK4 and ATF3 show an inverse correlation in lung AdC tissues (Fig. 6A). IHC analysis of AK4 and ATF3 expression in CL1-0 Vec and CL1-0 AK4 subcutaneous xenograft tumor also confirmed this inverse correlation (Supplementary Fig. S2B). We next investigated the prognostic value of ATF3 by IHC analysis. Figure 6B showed representative immunohistochemical expression of ATF3 in lung cancer specimens and high nuclear ATF3 expression levels (score 2 and score 3) were significantly correlated with better overall and disease-free survival relative to patients with low nuclear ATF3 expression (score 0 and score 1; Fig. 6C–D, middle). The relationship between ATF3 and clinicopathological features is summarized in Supplementary Table S5. The multivariate Cox regression analysis showed that the N status and M status significantly predicted worse overall survival and disease-free survival. However, only low ATF3 scores, as assessed by univariate analysis, significantly predicted poor survival (Supplementary Table S6). Moreover, the combination of AK4 and ATF3 further revealed that patients with high AK4 and low ATF3 exhibited poor outcomes in sharp contrast to patients with low AK4 and high ATF3 expression (Fig. 6C–D, right, P < 0.001).

In addition, we separated AdC and SCC as 2 different entities and carried out survival analysis. The results showed both AK4 alone and ATF3 alone can be a significant prognostic predictor for AdC patients and combination of AK4 and ATF3 is more powerful in predicting survival in AdC (Supplementary Fig. S5A). In SCC, AK4 can also be a significant prognostic predictor but ATF3 cannot and combination of AK4 and ATF3 could not further enhance the prediction power. Therefore, AK4-ATF3 axis as prognostic factor is more accurate in AdC patients (Supplementary Fig. S5B).

To validate our findings, we searched 5 independent patient datasets in PrognoScan database and compared the prognostic value of our biomarkers (AK4 and ATF3) with others using a minimum P value approach for stratifying the patients for survival analysis (20). In the database, it is possible to validate and compare our findings with a number of other cohorts including NCCRI (Japan), Seoul (Korean), UM (USA), Michigan_1994-2000 (USA), and HLM (USA). Expression of 4 genes (AK4, ATF3, DEPDC1, and ERBB3) with 17 probes showed that AK4 level was significant associated with poor outcome in both RFS and OS across different cohorts whereas ATF3 level did not significantly correlate with better outcome (Supplementary Table S7). DEPDC1 was identified as a novel prognostic marker for lung cancer by NCCRI group (21) and it also showed positive association with poor prognosis in 3 of the 5 cohorts (Supplementary Table S7). ERBB3 was identified as a member of the lung AdC gene signatures (17). However, the expression of ERBB3 alone cannot significantly predict patient outcome in most of the selected cohorts (Supplementary Table S7).

Aberrant activation of EGFR, KRAS, and ALK defines 3 major populations in lung AdC. To elucidate whether AK4 is affected by EGFR, KRAS, and ALK mutation, we analyzed GSE31210, which contains EGFR mutation (n = 127), KRAS mutation (n = 20), ALK mutation (n = 11), and EGFR/KRAS/ALK-negative (n = 68) lung AdCs. Data showed AK4 levels were significantly higher in triple-negative patients compared with patients with EGFR mutation and AK4 levels were significantly upregulated in ALK mutation patients compared with triple-negative patients whereas AK4 levels in KRAS mutation patients did not. However, ATF3 levels did not show significantly variations among these patient populations except 202672_s_at was significantly downregulated in ALK mutation patients compared with triple-negative (Supplementary Fig. S6).

In this study, we propose that AK4 acts as a metastasis-associated marker in the context of lung cancer. Lung cancer patients with high AK4 expression had shorter overall survival and disease-free survival times. Knockdown of AK4 inhibited invasion activity in lung cancer cells, whereas ectopic expression of AK4 induced this activity. Microarray analysis further identified ATF3 and several putative targets for lung cancer invasion and metastasis that were altered by AK4. Complementary expression and knockdown of ATF3 further confirmed that AK4 regulates invasion/metastasis via modulating ATF3 expression. A combination of AK4 and ATF3 expression could serve as a more accurate prognostic indicator for lung cancer patients.

Although an early study reported that adenylate kinase was an oncodevelopmental marker in an animal model of prostate cancer (22), the direct link between AK4 and cancer progression has not been explored thoroughly. Recent studies also described the potential oncogenic roles of AK4 in human cancer. Méndez and colleagues reported that AK4 was significantly upregulated, as assessed by microarray, in response to hypoxia and was associated with HIF-1α-mediated invasion and migration in human glioma (23). A kinome-wide RNAi screening identified AK4 as one of the vulnerable kinase targets in human multiple myeloma (24), and systemic protein profiling of clear-cell renal cell carcinoma also found that AK4 was differentially expressed and considered as a potential biomarker (25). In this study, we showed that overexpression of AK4 resulted in the specific induction of invasion activity without a significant impact on cell growth and proliferation. This may explain why patients with high AK4 levels had reduced survival times and higher chances of recurrence. It is known that metastatic cells colonized at distal sites are under considerable stress. To survive, metastatic cells must gain the ability to overcome this stress and adapt to the hostile microenvironment (26). In our in vivo metastasis model, we observed that AK4 could significantly promote the formation of lung foci following i.v. injection, suggesting a role for AK4 in promoting metastatic survival in the lung. Previously, it has been shown that AK4 interacts with mitochondrial ADP/ATP translocase to maintain energy homeostasis and protects cells from H₂O₂-induced cell death, which supports the hypothesis that AK4 plays a role in helping cancer cells to adapt the hostile environment (12).

To address the hypothesis that AK4 has critical roles in promoting invasion/metastasis, and other possible cellular processes required for metastatic development, we analyzed microarray data and identified differentially expressed genes affected by AK4. Interestingly, consensus genes that were differentially regulated following AK4 alteration were significantly linked to transcriptional regulation and several cellular metabolic processes. Notably, other than transcription regulator activity, RNA, hexose, glucose and monosaccharide metabolic process were also enriched, suggesting that AK4 may be involved in the regulation of cellular metabolism (Supplementary Table S4). Cancer metabolism has recently begun to be considered a hallmark of cancer because cancer cells are able to produce and use energy from a variety of sources under aerobic and anaerobic conditions more efficiently than normal cells (27, 28). However, the differences between metastatic cancer cells and their primary tumor cells in terms of metabolic requirements are not completely understood. By measuring levels of ATP, glucose and lactate, we found that overexpression of AK4 in CL1-0 cells resulted in increased ATP and glucose consumption and increased lactate production; this pattern resembled that observed in their invasive counterpart, CL1-5 cells (data not shown). Therefore, we hypothesize that whether cancer cells can efficiently produce and use energy to maintain an invasive phenotype may be the key to the development of metastasis.

One of the major transcription regulators that were affected following AK4 overexpression and knockdown is ATF3, a stress-inducible transcription factor from the ATF/CREB family that is known to both induce and repress gene expression via the p38 signaling pathway (29, 30). We therefore hypothesized that genes regulated downstream of ATF3 may be responsible for its anti-invasion/metastasis effects. Several genes downstream of ATF3 that have been linked to cancer progression include gadd153/Chop10, which is associated with cell growth (31), and MMP2, which is associated with invasion (32–34). A recent study also reported that ATF3 acts as a metastatic suppressor by transcriptionally repressing Id-1 after MCAM/MUC18 silencing, which resulted in negative regulation of MMP2 expression (35). Consistent with their data, we also found that MCAM/MUC18, Id-1, and MMP2 were upregulated, whereas ATF3 was downregulated, following AK4 overexpression in microarray data (Fig. 4C). We also confirmed MMP2 levels were indeed increased, in a dose-dependent manner, with a concurrent decrease in ATF3 transcripts after AK4 overexpression (Supplementary Fig. S4A). In addition, stable suppression of MMP2 abolished invasion activity induced by AK4, further showing the critical role of AK4-ATF3-MMP2 axis in lung cancer (Supplementary Fig. S4B). The possibility that other ATF3 targets or other mechanisms were involved in AK4-induced invasion is currently being investigated. However, recent reports have indicated a dichotomous role for ATF3 in different types of cancer (36, 37). In breast cancer, it has been reported that ATF3 could enhance cell motility but induced apoptosis in nontumorigenic mammary epithelial cells (38). In prostate cancer, ATF3 transcription was reported to be repressed by the metastasis suppressor Drg-1 (39). On the contrary, ATF3 downregulation in colon cancer induces tumor growth, as well as hepatic metastasis, and ATF3 expression is significantly lower in human colon cancer specimens compared with normal surrounding tissues (40). In addition, treatment with COX-2 inhibitors has been reported to block neoplastic and invasive activity effectively via ATF3 induction (41). Our data showed that AK4 silencing increased the expression of ATF3 and, more importantly, high ATF3 expression was significantly correlated with better overall and disease-free survival compared with lung cancer patients with low levels of ATF3 expression. These data further support the role of ATF3 as an invasion/metastasis suppressor in lung cancer. However, the specific cellular processes that are directly affected by alterations in AK4 expression and subsequent ATF3 regulation remain to be elucidated.

By comparing the expression of AK4 and ATF3 with other's biomarkers in associated with survival, we found AK4 remains an independent marker for poor prognosis across 5 different cohorts, however, ATF3 level did not significantly correlated with better survival as we predicted (Table S7). This may due to ATF3 is a transcription factor and gene transcript levels cannot totally reflect its transcriptional activity. In our study, we scored ATF3 protein expression based on staining intensity in the nucleus. Therefore, whether the subcellular localization of ATF3 is a determinant factor for prognostic prediction remains to be validated on other lung cancer cohorts. Notably, AK4 is a poor prognosis marker for both Asian and non-Asian populations (Table S7) but these results need to be validated at protein level on non-Asian patients.

Accumulating studies have shown that a notable fraction of lung AdCs develops from mutations in EGFR, KRAS, or ALK in a mutually exclusive manner (42). By analyzing GSE31210 datasets, we found that AK4 levels were significantly higher in triple-negative patients compared with patients with EGFR mutation. Therefore, AK4 may be useful for prognostic applications for triple-negative lung cancer patients. Interestingly, AK4 level showed significant upregulation in ALK mutation patients but whether AK4 could also be a prognostic factor for ALK mutation patients remains to be determined.

In conclusion, we showed that AK4 expression enhanced the invasion ability of lung cancer cells by repressing ATF3 expression and concomitantly relieving the expression of the downstream effector, MMP2, thereby promoting the invasion step of the invasion-metastasis cascade. In addition, AK4 levels were correlated with reduced overall survival and disease-free survival, potentially indicating that AK4 can be used as an independent prognostic factor. Moreover, combing measurements of AK4 levels and ATF3 nuclear levels could provide a more powerful predictor of lung cancer outcome.

No potential conflicts of interest were disclosed.

Conception and design: Y.-H. Jan, H.-Y. Tsai, J.-L. Su, M. Hsiao

Development of methodology: Y.-H. Jan, H.-Y. Tsai, Y.-F. Yang, T.-C. Lai, M. Hsiao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y.-H. Jan, H.-Y. Tsai, M.-S. Huang, Y.-F. Yang, Y.-M. Jeng, M. Hsiao

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-H. Jan, H.-Y. Tsai, J.-L. Su, M. Hsiao

Writing, review, and/or revision of the manuscript: Y.-H. Jan, H.-Y. Tsai, C.-J. Yang, C.-Y. F. Huang, Y.-J. Chuang, M. Hsiao

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.-C. Lai, C.-H. Lee

Study supervision: C.-Y. F. Huang, M. Hsiao

We thank Miss Tracy Tsai, Mr. Tzan Fang, Miss Victoria Lee, Miss Yu-Hua Lin, and Mr. Chia-Te Lin for technical support.

This study was supported by Academia Sinica and grants from the National Science Council to Dr. Michael Hsiao.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.