A recent study analyzing the global disease burden revealed that lung cancer is the second leading cause of death in the United States and in most other high-income countries [1]. In China, lung cancer is the fifth leading cause of death, following stroke, ischemic heart disease, road injuries, and chronic obstructive pulmonary disease [1, 2]. Globally, the annual incidence of lung cancer is approximately 1.8 million, with an annual mortality of approximately 1.6 million [3, 4]. The 5-year overall survival (OS) rate for lung cancer patients in the United States has improved only moderately over the past four decades from 12.2% in 1975 to 17.8% in 2010, despite the use of many therapeutic modalities [5–7]. The 5-year OS rate of lung cancer patients in China is approximately 16.1% [8], which is close to that observed in the United States. Histologically, lung cancer is classified as small cell lung cancer (SCLC, approximately 15%) and non-small cell lung cancer (NSCLC, approximately 85%); the latter includes adenocarcinoma, squamous cell carcinoma, and large cell carcinoma. Tobacco smoking is the major cause for all types of lung cancer, particularly SCLC and squamous cell carcinoma [9]. Air pollution is another major cause of lung cancer, as a substantial number of lung cancer deaths in China and other East Asian countries may be attributed to fine particles in the air [10–12].

Small cell lung cancer is highly aggressive, and most (95%) SCLC patients are current or former smokers [13, 14]. Although the initial response rate to platinum-based chemotherapy and radiotherapy is high, recurrence and resistance occur inevitably in most SCLC patients, leading to 5-year OS rates of approximately 24% for patients with limited-stage SCLC and of less than 3% for those with metastases [5]. Fortunately, the incidence of SCLC in the United States has been declining over the past decades because of a decreasing prevalence of tobacco smoking [15]. For NSCLC, surgical resection is the standard care for patients with stage I disease, whereas surgical resection plus adjuvant therapy is used for patients with stages II–III disease. NSCLC patients at the advanced stages (stages IIIB–IV) are usually treated with chemotherapy, pathway-targeted therapies, and/or supportive medicine. Early diagnosis is the major factor proven to improve outcomes [16]; the 5-year OS rate is approximately 70% for patients with stage I NSCLC and drops to approximately 5% for patients with stage IV lung cancer [17]. Unfortunately, most lung cancers are diagnosed at an advanced stage and thus have a poor prognosis [9]. Early detection by low-dose computed tomography (CT) screening significantly reduces lung cancer mortality; nevertheless, approximately 96% of the positive results detected by such screening are false positives [18]. Moreover, the current estimated cost of screening to avoid one premature lung cancer death is approximately $240,000, thereby adding substantial expenditures to the health care system [19].

Advances in molecular profiling and targeted therapy have shown that subgroups of lung cancer patients are highly sensitive to some small-molecule inhibitors targeting key molecular nodes that drive carcinogenesis in those patients, such as epidermal growth factor receptor (EGFR) and anaplastic lymphoma kinase (ALK). These breakthrough discoveries have not only led to a new paradigm of biomarker-directed precision or personalized therapy but also greatly accelerated the development of novel anticancer drugs. It is now clear that mutations in cancer driver genes and alterations in gene expression and/or posttranscriptional modifications can all drastically affect treatment responses or clinical outcomes [20–22]. Moreover, the success of targeted anticancer therapy largely depends on biomarkers that can identify patient subgroups that may respond to specific therapeutic agents because alterations in a particular cancer driver usually exist in only a small subset of patients. This review discusses recent advances in predictive biomarker-directed precision medicine and drug development for lung cancers.

Genome-wide gene sequencing analyses have found that each lung cancer may have an average of approximately 150 somatic mutations that are expected to alter their protein products; this number is much higher than the average of 30–60 mutations observed in other solid tumors [23]. The high somatic mutation rates observed in lung cancer reflect the mutagenic roles of cigarette smoking in the pathogenesis of lung cancer. Indeed, the total number of point mutations in coding regions identified in lung cancer is approximately 10 times higher in smokers than in non-smokers [24]. Smoking also causes distinct changes in gene mutation and gene expression signatures in cancer tissues and normal lung tissues [25, 26]. Smoking-induced gene expression alterations observed in normal lung tissues can be transient because the majority of changes at the expression level can revert to non-smoker levels following smoking cessation [26]. Nevertheless, some of these expression changes are irreversible and permanent [26].

The results from genome-wide sequencing analyses for primary lung adenocarcinoma [24, 25, 27, 28], squamous cell cancer [29], SCLC [30, 31], and carcinoids or neuroendocrine tumors [32] have been reported recently by several groups. Information regarding genetic alterations (mutations and copy number changes), mRNA expression data, and protein/protein phosphorylation levels in lung adenocarcinoma, squamous cell cancer, and SCLC can be retrieved from The Cancer Genome Atlas (TCGA) databases and publically available datasets at the website http://www.cbioportal.org. Figure 1 shows the frequencies of cancer driver genes that are frequently mutated (including copy number changes) in lung adenocarcinoma, squamous cell cancer, and SCLC. Based on data retrieved from cBioPortal, tumor protein p53 (TP53) is the most frequently mutated cancer driver gene for all three histological types of lung cancer, with frequencies varying from 46% in adenocarcinoma to 86% in SCLC. Additionally, Kirsten rat sarcoma viral oncogene homolog (KRAS), cyclin-dependent kinase inhibitor 2A (CDKN2A), mixed-lineage leukemia 3 (MLL3), serine/threonine kinase 11 (STK11), kelch-like ECH-associated protein 1 (KEAP1), and EGFR are the top frequently mutated genes detected in lung adenocarcinomas. Phosphatidylinositol-4,5-biphosphate 3-kinase, catalytic subunit alpha (PIK3CA), sex-determining region Y-related gene family 2 (SOX2), CDKN2A, TP63, fibroblast growth factor receptor 1(FGFR1), and MLL2 are the top mutant genes in squamous cell cancer; retinoblastoma 1 (RB1), E1A-binding protein p300 (EP300), MLL2, smoothened (SMO), and PIK3CA are the top mutant genes in SCLC. Our recent sequencing analysis of the exomes of 200 cancer-related genes found that mutations in TP53, KRAS, MLL3, SET domain-containing 2 (SETD2), AT rich interactive domain 1A (ARID1A), PIK3CA, and ALK were frequently detected in primary tumors and corresponding patient-derived xenografts of NSCLC [33]. Most of the mutations (93%) detected in the primary tumors were also detected in their corresponding patient-derived xenografts. Nevertheless, the numbers of mutations detected in each primary tumor varied greatly, and most tumors had mutations in more than two genes. The results from these molecular profiling studies clearly demonstrated heterogeneity in the molecular pathogenesis of lung cancer. As shown in Figure 2a, the mutational status of the top seven frequently mutated cancer driver genes varied greatly in 230 lung adenocarcinomas [28]. Moreover, mutations in those genes, particularly for tumor suppressor genes such as TP53 (Figure 2b), are often widely distributed in the entire coding region. Evidence has shown that mutations in TP53 can lead to either loss or gain of functions [34–38], and both may promote tumorigenesis through different mechanisms. Similarly, different mutations in KRAS, even if in the same codon such as G12C, G12V, and G12D mutations in the KRAS gene, may lead to different conformational changes in KRAS and have different effects on clinical outcomes and molecular pathway activation [39]. Ultimately, this molecular heterogeneity may affect treatment responses to therapeutics targeting different pathways.

Notably, however, current TCGA data are derived mostly from patients in Western countries. The mutational status in other ethnic populations may differ [40–42]. For example, activated EGFR mutations are detected in approximately 10–17% of lung adenocarcinoma patients in the United States and Europe [27, 43–46] but in approximately 30–65% of lung adenocarcinoma patients in Asia [47–50]. In contrast, KRAS mutations were detected in 35–50% of lung adenocarcinomas in Caucasian patients [28, 51] but in less than 5% of lung adenocarcinomas in Chinese patients [48, 49, 52]. Nevertheless, changes in molecular signaling can occur in the absence of gene mutations. Indeed, the activation of RAS signaling pathways can be detected in substantial numbers of primary tumors with wild-type RAS genes [53]. Increased expression of the RAS genes [54, 55], increased upstream signaling from tyrosine kinase growth factor receptors such as v-erbB2 avian erythroblastic leukemia viral oncogene homolog 2 (HER2) and EGFR [56, 57], and altered expression of microRNAs such as let-7 [58, 59] may all contributes to the activation of RAS signaling pathways.

Anticancer drug development has a high failure or attrition rate, primarily because of a lack of efficacy, inability to identify responders, and intolerable toxic effects [152, 153]. Only approximately 5% of anticancer agents evaluated in human studies between 1991 and 2000 were successfully approved for clinical applications [152, 154]. The majority (80%) of phase III failures were caused by a lack of efficacy [152, 154–156] that resulted from a lack of efficacy proof-of-concept in humans and a lack of objective biomarkers capable of reporting such efficacy [152, 154]. The development of the anti-EGFR drugs gefitinib and erlotinib and the anti-ALK/ROS1 drugs crizotinib and ceritinib has provided excellent examples of using predictive biomarkers and patient stratifications in clinical trials to increase the chance of success or to reduce the attrition rate in anticancer drug development. Notably, both gefitinib [64, 157, 158] and erlotinib [159, 160] failed to show a benefit in randomized phase III trials with unselected patient populations. Similarly, the successes of the v-Raf murine sarcoma viral oncogene homolog B1 (BRAF) inhibitor vemurafenib [161, 162] and the poly(ADP-ribose) polymerase (PARP) inhibitor olaparib [163–165] in anticancer therapy relied on biomarker-based patient stratifications in clinical trials. In fact, the inability to identify cancers that are most likely to respond to a treatment is a major challenge in anticancer drug development, not only leading to failure to demonstrate the potential benefit of a promising anticancer agent [64, 166] but also exposing patients to the risks of ineffective treatments. Indeed, a lack of predictive biomarkers for patient stratification has halted further development of many other investigational drugs [167–170]. Thus, reliable predictive biomarkers are crucial to the success of anticancer drug development, and the simultaneous development of predictive biomarkers will be highly desirable for promising anticancer drug candidates.

The identification of predictive biomarkers can be relatively straightforward if the therapeutic targets are themselves the cancer drivers and if the cancer cells are addicted to the presence of those targets for survival. As in the cases of EGFR, ALK, and ROS1, the constitutive activation of such targets is the driving cause of a subset of NSCLC and is required for the survival of those cancer cells; therefore, the presence of constitutive activating mutations or gene rearrangement can be explored as predictive biomarkers for patient stratification. Nevertheless, cancer cells can be killed indirectly through synthetic lethality between the cancer drivers and their synthetic lethal partners, as has been demonstrated using the PARP inhibitor olaparib to eliminate breast cancer early onset (BRCA)-mutant ovarian and breast cancer cells [165, 171, 172]. In this case, loss-of-function mutations in the BRCA1 and BRCA2 genes or other functional deficiencies in the DNA repair pathways will serve as predictive biomarkers for response to olaparib and other PARP inhibitors [173]. Synthetic lethality is a lethal phenotype caused by the combined effects of mutations in two or more genes. This concept has recently been exploited for the development of novel genotype-selective anticancer agents and/or the identification of novel therapeutic targets [173–176]. This approach could be useful for the indirect targeting of cancer drivers such as genetic changes in TP53, STK11, RB1, and RAS genes that are nondruggable or difficult to target using small molecules. Notably, a study conducting short hairpin RNA (shRNA) library screening for genes that may have synthetic lethal interactions in the oncogenic KRAS gene in the colon cancer cell line DLD-1 and its isogenic derivatives led to the identification of approximately 370–1,600 KRAS synthetic lethal genes, depending on the stringencies of statistical analyses [177], demonstrating the diversity of biological processes or pathways regulated by KRAS. The molecular pathology of RAS-mediated oncogenesis is further complicated by the fact that cancer cells with different mutations in KRAS, even if in the same codon (such as G12C, G12V, and G12D mutations), may lead to different conformational changes in KRAS, have different effects on clinical outcomes, and activate different molecular and metabolic pathways [39, 178]. Moreover, a substantial number of KRAS-mutant tumors have mutations in other cancer driver genes such as TP53, CDKN2A, and STK11. Studies in transgenic mice have found that activating mutations in KRAS predispose mice to early-onset lung adenocarcinoma [179, 180]. Nevertheless, such tumors do not have an invasive or metastatic phenotype [181]. Metastasis will occur only when additional genetic changes, such as Tp53 mutations or Stk11 deletion, are introduced [182, 183]. Finally, the RAS activation signature is observed in approximately 25% of KRAS wild-type cancers [53] because increased RAS activity in human cancers can result from RAS gene amplification [184, 185] or overexpression [55], an increase in upstream signals from tyrosine kinase growth factor receptors such as HER2 and EGFR [56, 57], or a decrease in the expression of microRNAs such as let-7 family members [58].

The above-discussed situations impose considerable challenges on the identification of predictive biomarkers for anti-RAS therapeutics developed through synthetic lethality screening. Substantial efforts will be required to identify predictive biomarkers for such therapeutic agents. Through synthetic lethality screening using a chemical library for isogenic cell lines with and without oncogenic KRAS, we previously reported a small molecule, designated oncrasin-1, that induces apoptosis in several KRAS-mutant lung cancer cell lines [186]. The in vitro and in vivo activities were optimized by the synthesis and evaluation of analogues [187–190]. The mechanistic characterization revealed that these compounds induce oxidative stress, activate c-Jun N-terminal kinase (JNK), and inhibit RNA polymerase II, STAT3, and cyclin D1 [187, 188, 191–194]. However, a predictive biomarker for oncrasin compounds remained elusive until our recent discovery that such compounds require the expression of a sulfotransferase (SULT), SULT1A1, in cancer cells for their anticancer activity and that the expression of SULT1A1 is capable of predicting the responses to oncrasin compounds [195]. SULT1A1 is a biotransformation enzyme that bioactivates several pro-carcinogens [196–202] and some anticancer drugs such as tamoxifen [203]. The identification of SULT1A1 as a predictive biomarker for oncrasin compounds shed light on mechanisms of the selectivity and, possibly, the toxicity of these compounds [204]. The process of identifying this predictive biomarker underscores the importance of activity characterization in a large set of molecularly annotated cancer cell lines and of collaborations between biostatisticians and molecular biologists. Statistical analysis of associations between the activity of NSC-743380 (oncrasin-72) and the gene expression profile in the NCI-60 cell line panel led to the identification of a number of candidate genes, and subsequent molecular characterizations demonstrated the causal relationship between the expression of SULT1A1 and the anticancer activity of NSC-743380, validating the feasibility of using SULT1A1 as a predictive marker for those compounds [195]. Larger panels of molecularly annotated cancer cell lines have also been used by other groups for the identification of predictive biomarkers for larger sets of anticancer drugs [150, 205–209]. Ultimately, rigorous validation of causal relationships between the sensitivity and the biomarker will be essential for successful clinical translations.

The diversity found in the molecular pathogenesis of lung cancer demonstrates that conventional histology-based classifications of the disease are not sufficient to provide guidance for treatment decisions in the clinic. Classifications based on molecular pathology are required for precision therapy. The successful development of anti-EGFR and anti-ALK/ROS1 therapeutics has proven the critical role of predictive biomarkers in anticancer drug development because the ability to identify patient subpopulations who may benefit from a drug candidate will directly affect the design and patient selection of clinical trials, dramatically reduce the sample size required to reveal potential benefits, and increase the chance of success. It is expected that more efforts will be devoted to the identification of predictive biomarkers and the development of effective assays for biomarkers during drug development. These efforts will likely be rewarded by increased success rates when the drug candidates enter clinical trials. Advances in technologies, such as next-generation sequencing and digital polymerase chain reaction, will also facilitate the detection of changes in genotypes and/or in DNA or RNA levels and the development of effective assays for biomarkers. On the other hand, biomarker-directed patient stratification has changed the clinical practice of anticancer therapy and has substantially improved treatment efficacy by providing guidance for the selection of optimal treatment for some lung cancer patients. However, biomarker-directed personalized therapy can currently benefit only a limited number of patients because most of the currently available predictive biomarkers occur infrequently [24, 27, 210] and because response-predictive biomarkers are not available for most first-line anticancer therapeutics. Thus, alternative approaches will be required to identify the appropriate therapeutics for a particular patient. Such patient-oriented approaches may require direct testing of the drug sensitivity of a patient’s cancer cells to various available anticancer therapeutics rather than predicting drug sensitivity based on the presence or absence of predictive biomarkers. Some recent studies have shown the feasibility of testing for the treatment responses of patient-derived tumor xenografts [211, 212] and/or primary cell cultures [213] to identify effective treatment regimens. Nevertheless, technology improvements and rigorous validations are required before such techniques can be used commonly in clinical practice.