Progress and applications of mouse models for human lung cancer

Progress in whole genome approaches to detect genetic alterations in human lung cancer has resulted in the identification of a growing number of lung cancer-related genes. Genome-wide association studies, whether they are based on single-nucleotide polymorphism array studies or detecting changes in gene copy numbers via comparative genomic hybridisation arrays, link the occurrence and frequency of mutations in lung cancer-related genes to the well-defined phenotype of high numbers of human lung cancer. These lung cancer-related genes provide great potential as therapeutic targets for lung cancer intervention. Target validation then occurs through in vitro intervention studies of these specific genetic mutations and their respective molecular pathways in human lung cancer cell lines. However, in vitro cell culture studies are limited and cannot fully mimic the more complex in vivo onset of tumorigenesis and response to tumour therapy. Developing lung cancer in mouse models that harbour specific mutations will undoubtedly provide a further and better insight into the mutation-specific effects on lung tumour biology. Moreover, a high degree of pathophysological similarity between lung tumours from mouse models and their human counterparts will make it possible to use these mouse models for pre-clinical tests. Various intervention strategies against specific mutations can then be tested based on better evaluation of both specificity and efficacy in mouse lung tumours of every developing stage. Continuous innovation of techniques to manipulate the mouse genome has enabled us to adjust compound mouse models of lung cancer in such a way that they start to reproduce the more complex human lung cancer in a higher degree. Complementary to this, the number of genetically engineered mouse models for lung cancer is ever expanding.

FIRST MOUSE MODELS FOR LUNG CANCER

The use of mouse models for spontaneous or chemically induced lung tumours has a long history. While susceptibility and incidence of spontaneous lung tumours varies between well-established mouse-inbred strains, all their molecular pathologies share many similarities with human lung cancer 1. This was clearly established in early studies in which defined chemical carcinogens were used to induce lung tumours 1. Incidence of spontaneous and induced lung tumours were both at the highest (61%) in sensitive strains such as A/J and SWR, but very low (6%) in highly resistant strains C57BL6 and DBA for males at 2 yrs 1. Contrary to human lung cancer with its complex molecular genetics and four distinct tumour types that readily metastasise, spontaneous and chemically induced lung lesions in mice often result in pulmonary adenomas 2 and more infrequent adenocarcinomas, which almost never metastasise. Mouse lung tumour development shows initial hyperplastic foci in bronchioles and alveoli, which then become benign adenomas and eventually adenocarcinomas 2. The very reproducible tumour latency obviously depends on strain susceptibility and/or application of carcinogen-induction protocols. Most potent carcinogens are the cigarette smoke carcinogens, such as polycyclic aromatic hydrocarbons, tobacco-specific nitrosamine and benzo[a]pyrene (B[a]P) 3. However, cigarette smoke itself is sufficient to induce reproducibly lung tumours in A/J mice after a 5-month exposure period followed by a crucial 4-month recovery period 4. Exposure of B6C3F1 female mice to life-time (30 months) cigarette smoke resulted in 48% benign and malignant lung tumours through distinct (epi)genetic pathways 5. Therefore, there are clear differences between (spontaneous) murine and human lung tumours. It has been especially hard to replicate the well-characterised pre-malignant lesions in human airway epithelium 6. Nevertheless, major histopathological similarities remain and molecular characterisation of spontaneous and carcinogen-induced murine lung tumours revealed a high degree of genetic lesions compared to their human counterparts 7. A prominent early event is the occurrence of activating Kras mutations 7 in hyperplastic lesions. Besides over expression of cMYC, inactivation of well-known tumour suppressor genes, such as Tumour suppressor protein 53 (Trp53), Fragile histidine triad protein (fhit), adenomatous polyposis coli protein (APC), retinoblastoma protein (RB), mutated in colorectal carcinoma protein (MCC) and p16^INK4A (CDKN2A) 7, readily occur; most frequently adenomas develop which only sporadically progress into adenocarcinomas. Differences in susceptibility to lung cancer development between various mice strain remain, however, very intriguing. Most susceptible strains, such as A/J and BALB/C, do have a polymorphism in intron 2 of Kras 8, 9 and a CDKN2a polymorphism was found in BALB/C, influencing their sensitivity to lung cancer. Due to the genome wide sequence analysis, efforts in more interesting polymorphism information of various mice strains is likely to follow and can then be compared with human syngeneic counterparts found in large panels of human lung cancer genome data. Moreover, as we will see in later in this review, more direct transgenic models will be very suitable to cross into resistant strains in order to facilitate the direct search for modifying genes of lung tumour susceptibility.

TRANSGENIC MICE

First generation

The first generation of transgenic models was based on ectopic transgene expression under control of heterologous promoters. Expression is mainly targeted to specific subsets of lung epithelial cells: surfactant protein-C promoter directs expression primarily to type II alveolar cells whereas Clara cell secreatory protein (CCSP)/CC10 promoters mainly target the nonciliated secretory (Clara) cells along the airways. SV40 Tag (Simian virus large T-antigen) was constitutively expressed behind CCSP 10, 11 or surfactant protein (SP)-C promoters 12. Although each tumour originated from either type II alveolar or Clara cells, they both resulted in quite similar aggressive adenocarcinomas without metastases 13. A similar strategy was used for different oncogenes (such as cRaf and cMyc 14), however, with a milder phenotype, as both transgenic mice mainly ended up in having adenomas and a few progressed adenocarcinomas, again without any metastases.

Many prominent genetic lesions found in human lung cancer clearly link the inactivation of well-known tumour suppressor genes 15 to lung cancer development. Initial attempts to mimic some of these lesions implicated in lung cancer by way of conventional knockout mice studies had rather limited success with respect to the onset of lung cancer. The main reason for this was that germ-line deletion of many essential tumour suppressor genes (such as the retinoblastoma gene (Rb) and wilms tumour-1 homolog gene (Wt-1) 16) lead to embryonal or perinatal lethality. Nonessential tumour suppressor gene knockout mice with a longer life-span often had a very broad tumour spectrum of which lung tumours formed only a minor fraction. Thus, Trp53, p16^INK4A and p19^ARF 17 homozygous null allele mice seldom develop lung adenocarcinomas. However, introducing similar mutations into endogenous Trp53 alleles, such as those prominently found in Li–Fraumeni patients, generated Trp53^R270H/+ and Trp53^R172H/+ which had a different tumour spectrum compared with Trp53^+/- 18, although their mean survival times were identical. Interestingly these mice, but especially Trp53^R270H/+ and Trp53^R270H/-, gave rise to more malignant lung adenocarcinomas, desmoplasia and even metastases which never occur in Trp53^-/- mice. These results suggest that “humanised” Trp53 mutations have a greater impact on lung tumour progression than complete Trp53 loss 18, 19. Targeting genes deleted early in human lung tumorigenesis, such as the complete cluster at chromosome 3p21.3, showed that heterozygous deletion for this 370 kb region showed no obvious predisposition for lung cancer development albeit homozygous deletion caused embryonal lethality 20. A more specific deletion of candidate tumour suppressor genes on chromosome 3 like RassF1a, FHIT and VHL, showed that 31% of RassF1a^-/- mice produced spontaneous mainly lymphomas but also lung adenomas 21. Treatment of RassF1a^-/- mice with B[a]P or urethane resulted in an even higher rate of lung tumours. No spontaneous lung tumours were observed in Fhit^-/- or Vhl^+/- mice, but 44% of Fhit^-/-/Vhl^+/- mice developed adenocarcinomas by age 2 yrs. Again use of mutagens such as dimethylnitrosamine led to 100% adenoma and adenocarcinoma induction in Fhit^-/-/Vhl^+/- mice and even adenomas in 40% of Fhit^-/- mice by age 20 months 22. This showed the usefulness of these knockout mice in recapitulating a pattern of early lung cancer development similar to human pattern.

A different approach to address lung cancer onset was the use of knock-in alleles to activate oncogenes. One example of this is based on the somatic Kras activation via an oncogenic Kras^G12D knock-in allele (Kras^LA2), which is expressed only after a spontaneous recombination event (fig. 1⇓) 23. In this way, sporadic Kras^G12D expression occurred on an endogenous level, which in turn augments efficient development of lung adenocarcinomas. However, these mice also developed other tumour lesions as Kras^G12D expression was not limited to the lung epithelial tissues.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1—

Overview of systems used in most of the genetically engineered mouse models for lung cancer. a) The tetracycline-inducible bitransgenic system. A tissue-specific promoter (TSP) drives the expression of a Tet-inducible transactivator (rtTA). In the presence of doxycycline (DOX), rtTA dimers can specifically bind to seven tetO (7x tetO) sequences and activate oncogene transcription from the minimal TATA promoter. b) The latent activated allele ensures stochastic oncogene activation via spontaneous in vivo recombination of a silent mutant allele. A neomycin resistance (neo) gene separates a wild-type (WT) and a mutated copy of the otherwise identical exons. After recombination, only a WT allele or an allele with an activating mutation is left to exert oncogenic functions. c) Conditional gene inactivation is based on Cre/loxP recombination-mediated deletion of one or more exons. Site-specific Cre recombinase induces mutations through recombination of two direct orientated LoxP sequences. Thus, controlling Cre recombinase activity ensures eventual tumour suppressor (TS) gene inactivation into various tissues of choice. d) Conversely, the conditional LoxStopLox allele makes use of a strong transcription stop, flanked by two lox sites, in front of a coding exon. The resulting null alleles can become an oncogenic mutant allele after removal of the transcription stop element via Cre-mediated recombination. pA: polyadenosine.

MODELS FOR NSCLC

To date, most mouse models generate lung adenocarcinomas with various stages of malignancy. As we have already discussed, a frequently used way of driving lung tumorigenesis is through conditional or spontaneous activating Kras mutations 17. However, it is obvious that more complex mouse models are needed to further our understanding of highly complicated lung cancer genetics (fig. 2⇓). Validation of these mouse models then requires knowledge of how closely mouse models resemble human lung cancer. Several studies used cross-species comparisons of gene expression profiles between murine and human lung tumours to identify similarities in cancer signalling pathways 35, 36. Integration of gene expression data from a Kras^LA2 mouse model and KRAS mutated human lung tumours indeed showed a significant overlap but also revealed a gene-expression signature for Kras mutation in human lung cancer itself 35. These results presented powerful molecular criteria to assess similarities between mouse and human lung cancer, as well as new possibilities to further elucidate important genes that control less well-characterised lung cancer pathways. Since activating Kras mutation models recapitulate some of the human lung tumour phenotypes rather well, closer analyses of early lung tumour initiating events were performed 37, 38. A combination of both CCSP CC10-Cre recombinase and LSL Kras^G12D alleles 37 resulted in a progressive phenotype of cellular atypia, adenoma and finally adenocarcinoma. These lesions originated exclusively from bronchial epithelia but were accompanied with a strong inflammatory response through expression of various chemokines. Kras activation could, therefore, serve as an excellent model to study the complex interactions between transformed lung epithelial and inflammatory cells, chemokines and early tumour development. Another model for generating early, benign lung tumour lesions makes use of a bitransgenic Tet-inducible Kras^G12C allele that can be expressed in both Clara and/or alveolar type II cells 27, 38. Only mild atypical hyperplasia or benign adenomas were found and even after an induced Kras^G12C expression of ≥12 months, no further tumour progression could be detected. Inhibition of Kras^G12C expression through doxycycline withdrawal over 1 month led to a complete reversal of any early proliferative lesion. Thus, both Kras^G12C and Kras^G12D oncogenes are needed for maintaining a tumour phenotype. One explanation of the low tumour penetrance of Kras^G12C alleles could be the absence of activation of Akt and JNK signalling pathways. This latter observation was in strong contrast to the effects of the previously described Kras^G12D models.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2—

Schematic overview of the major signalling pathways that are impaired in most of the nonsmall cell lung cancer models. Activation of the signalling network occurs through activating mutations of the oncogenes (shown in red) or inactivating tumour suppressor genes (shown in blue). At the core of the signalling cascade are RAS (rat sarcoma viral oncogene homologue) and its stimulation of the RAF-MEK-ERK (serine/threonine kinase raf-mitogen-activated protein kinase kinase-extracellular signal-regulated kinase kinase) and PI3K (phosphatidylinositol 3-kinase) pathways. Activating mutations in epidermal growth factor receptor (EGFR) serve as the main models for receptor tyrosine kinases upstream of RAS. As for transforming growth factor (TGF)-β, one has to acknowledge its dual role as a tumour suppressor during the early phase of lung tumorigenesis and a pro-oncogenic factor in more advanced and metastatic lung tumours. Note that RAC1 (Ras-related C3 botulinum toxin substrate 1, small GTPase) exerts oncogenic activity in collaboration with RAS, since inactivating Rac1 models showed an impairment of RAS-driven lung tumorigenesis. EGF: epidermal growth factor; TGFβR: TGF-β receptor; Mekk1: mitogen activated protein kinase kinase kinase 1 or MAP3K1, serine threonine kinase; Mkk3/6: mitogen-activated protein kinase kinase 3 or MAP2K3, serine threonine kinase; p38α-MAPK: mitogen activated protein kinase p38α or MAPK14; PTEN: phosphatase and tensin homolog; AKT: serine/threonine kinase also known as protein kinase B; LKB1: serine/threonine kianse LKB1 also known as serine/threonine protein kinase 11, STK11: mTOR: mammalian target of rapamycin, serine/threonine protein kinase; SMAD 2/3: TGF-β-dependent gene transcription factors.

It is rather puzzling that Kras but not Hras or Nras mutations are predominantly found in lung carcinomas 39. A plausible explanation for this can either be that differences in the Ras protein isoforms render them different oncoprotein activities, or differences in their respective gene regulatory elements confer tissue-specific expression to the mutated Ras gene in its corresponding tumour. Moreover, one has to note that the Kras gene encodes two isoforms, 4A and 4B, with differences in their C terminus. This C terminus of the predominant Kras 4B isoform is prone to exclusive post-translational modifications, such as farnesylation or geranylgeranylation 40. However, Kras 4A has a post-translational modification that includes covalent attachment of a palmitoyl moiety, as do Hras and Nras. The latter modifications give Nras, Hras and Kras 4A distinct signalling activities by facilitating their localisation in specific microdomains in membranes 41. A recent elegant study elucidates Kras mutation specificity in lung cancer by exchanging either a Hras or Kras 4B cDNA into the original Kras gene by generating Hras (Hras^KI) or Kras 4B (Kras^KI) knock-in mice. Surprisingly, the number of urethane-induced lung tumours in Hras^KI/KI was much higher compared to Kras^KI/KI mice and the latter produced even less tumours than the wild-type (WT) control mice 42. Interestingly, when mutations were analysed in Kras^KI/wt heterozygous mice, they were only detected in the WT allele which expresses Kras 4A, as well as Kras 4B. This led to the conclusion that only a mutated Kras 4A allele, which, after all, shares functional C terminus similarity with Hras, can induce lung tumorigenesis and Kras regulatory elements provide its specificity for pulmonary epithelial cells 42, 43. In a different setting, a Kras^LA2 allele was either combined with WT Kras, Hras or Kras^KI alleles. The number of spontaneous lung tumours was high in Kras^LA2/KI, but similarly low in Kras^LA2/WT and Hras^KI; Kras^LA2 mice. This tumour suppressing activity can only be accounted for by an intact WT locus which is capable of encoding an intact Kras 4A isoform and could, therefore, be functionally substituted by a similar Hras^KI allele 42. Thus, there is now compelling evidence that a normal WT Kras allele actually inhibits and that preferentially a mutated Kras 4A allele actively induces lung tumorigenesis, respectively, in a strictly tissue specific manner 42–44. In hindsight, this also explains earlier observations in which vinyl-carbamate treated mice with one Kras KO allele produced a higher number of more advanced adenocarcinoma than control WT Kras mice 45.

We know that the occurrence of somatic Kras mutations in lung cancer is prevalent 46, thus making the understanding of the precise role of Kras pathway effectors very important (fig. 2⇑).

Indeed, mutations in candidate effectors BRAF and PI3K of activated Ras downstream pathways have been reported in human NSCLC 47, 48. Activating mutations of BRAF would most likely have a stimulating effect on mitogen-activated protein kinase (MAPK) pathways. This hypothesis was tested in a lung specific, Tet-inducible, expression of a BRAFV600E mutant transgene 46. In this mouse model, CCSP-rtTA;TetO₇-BRAFV600E showed a development of lung adenocarcinoma with bronchioalveolar carcinoma features. The extracellular signal-regulated kinase (ERK)-1/2 (MAPK) pathway was significantly upregulated. Upon doxycycline withdrawal, the BRAF-mutant lung tumours disappeared in combination with a marked decrease in phosphorylation of ERK1/2. Furthermore, the in vivo use of a specific MAPK/ERK kinase (MEK) inhibitor induced lung tumour regression. Interestingly, the latter effect could be reproduced with lung tumours from CCSP-rtTA;TetO₇-Kras^G12D mice 46. All these results showed that both activated BRAF and KRAS signalling convert on the same MAPK pathway, making this pathway a potential target for lung tumour intervention. These results were confirmed by another independent study in which a floxed BRAFV600E allele was activated in lung epithelium via nasal Adeno-Cre delivery 49, which caused a formation of benign adenoma. However, the majority of these latter lesions did not further develop into adenocarcinoma and the abundant nonprogressed adenoma also showed features of oncogene induced senescence. Combining BRAFV600E activation with loss of Trp53 or Ink4a/Arf led, as expected, to marked tumour progression into adenocarcinoma 49. A close investigation during Kras^G12D driven lung tumorigenesis showed an increased phosphorylation of MAPK during which the MAPK antagonist Sprouty-2 (Spry-2) was upregulated. When Cre/lox dependent Spry-2^F/F;LSL Kras^G12D mice were used for lung tumour induction, a clear increase in both tumour load and development was observed in Kras^G12D;Spry-2^-/- tumours. This clearly suggested a tumour suppressor activity for Sprouty-2 during Kras dependent lung tumorigenesis by impairment of the Ras/MAPK signalling cascade 50.

Another approach to understand the role of the Ras downstream effector pathway led to the simultaneous Cre/lox dependent deletion of Rac1^F/F in the presence of activated Kras 51. Results showed a significant delay in Kras dependent adenoma formation in Rac1^-/- versus Rac1^+/- lesions, which suggest that Kras signalling acts, at least in part, through Rac1 during lung tumour development.

Another downstream effector (fig. 2⇑), interlinked with RAS signalling pathway, is phosphoinositide-3-kinase (PI3K) and downstream targets, such as the protein kinase Akt. In human NSCLC somatic mutations that activate PI3K have been identified in the p110α catalytic subunit (encoded by PIK3CA) 52. Furthermore, selective PIK3CA amplification was found in lung squamous cell carcinomas 53. Therefore, mice were generated with a Tet-inducible expression of p110α, with an activating mutation in its kinase domain (H1047R), and then crossed with CC10 promoter-reverse tetracycline transactivator protein (rtTA) to generate CCSP-rtTA;TetO₇ PIK3CA(H1047R) mice. Upon doxycycline dependent induction of PIK3CA(H1047R) for 14 weeks, double transgenic mice developed adenocarcinomas, which subsequently completely disappeared after doxycycline withdrawal for 3 weeks 54. On the contrary, complete ablation of PI3K activity through deletion of both its p85 regulatory subunits (encoded by Pik3r1 and Pik3r2) led to a dramatic decrease in the number of lung tumours in LSL Kras^G12D;Pik3r2^-/-;Pik3r1^-/- mice 54. Many in vitro studies showed that Ras proteins directly interact with the p110α subunit of PI3K and introduction of specific mutations in PIK3CA blocks this interaction 55. In order to study the Ras-p110α interactions in vivo and its effects on tumorigenesis, Pik3ca^-/- mice were generated and crossed with Kras^LA2 alleles. These Pik3ca^-/-;Kras^LA2 mice were highly resistant against lung tumour formation, which suggest that Ras-p110α interaction is needed for Ras-driven tumorigenesis 55.

All these results underline the importance of PI3K signalling not only for lung tumour induction but also maintenance.

Cross talk between Ras-MAPK, transforming growth factor (TGF)-β1 and its type II receptor (TGF-β RII) does exist 56, while TGF-β1 is implicated as a tumour suppressor gene in human lung cancer 57. To test this latter hypothesis in vivo, lung tumorigenesis was followed in Kras^LA2;Tgf-β1^+/- versus Kras^LA2 mice. The Kras^LA2;Tgf-β1^+/- lung tumours appeared faster and with a relatively higher amount of adenocarcinomas compared to those from Kras^LA2 mice. Surprisingly, adenocarcinoma that retained WT levels of TGF-β1 had significantly higher angiogenic activity than Tgf-β1^+/- adenocarcinoma. Therefore, TGF-β1 is a bona fide tumour suppressor during onset and progression of lung tumorigenesis, but also needed at the same time for lung tumour-induced angiogenesis and metastatic potential 58.

Another key player driving human NSCLC is EGFR signalling, which is independent and upstream of Kras (fig. 2⇑). In fact, mutations in EGFR occur mutually exclusive from those found in the Kras gene in human NSCLC 48.

Most mutations in EGFR occur around the region encoding the ATP binding pocket of the receptor's tyrosine kinase domain in exons 18–21 59. More precisely, these mutations occur as in-frame deletion in exon 19, which eliminate a conserved LREA (lysine, arginine, glutamic acid and alanine protein sequence) motif (EGFR^DEL), and an L858R substitution in exon 21 48. When these specific mutations were engineered in mice by use of the Tet-inducible system, two types of bitransgenic mice were generated: CCSP-rtTA;Tet-O₇-hEGFR^L858R and CCSP-rtTA;Tet-O₇-hEGFR^DEL. Both bitransgenic mice expressed mutated human EGFR in lung type II alveolar cells and following a few weeks of doxycycline administration developed bronchioloalveolar carcinomas 60, 61. However, tumour development continued when the doxycycline application period was extended beyond 4 weeks, resulting in the formation of invasive adenocarcinomas. However, when doxycycline was withdrawn the absence of mutated hEGFR resulted in a complete regression of lung cancer 60. Overall, the tumour phenotype showed a remarkable resemblance with human adenocarcinoma. This makes these bitransgenic models excellent tools for the study of EGFR-dependent NSCLC as it confirms that hEGFR^DEL and hEGFR^L858R are dominant oncogenes capable of inducing fully fledged lung cancer. Another distinct EGFR mutation, the variant III (vIII) in-frame deletion of exons 2–7, was found in some 5% of analysed human squamous cell carcinoma but not in adenocarcinoma 62. Bitransgenic CCSP-rtTA;TetO₇-EGFR^vIII mice developed adenocarcinoma after 16 weeks of doxycycline application. Lung tumours depended on EGFRvIII expression as doxycycline withdrawal led to complete tumour regression 62. All EGFR mutant mouse models showed the importance of EGFR activity in NSCLC and are highly penetrant in producing advanced adenocarcinoma.

Since expression of phosphatase and tensin homologue deleted from chromosome 10 (PTEN) is often down regulated in NSCLC, several mice models were generated in which Pten was inactivated in the bronchial epithelium 63, 64. PTEN is a tumour suppressor gene that acts by blocking the PI3K dependent activation of serine-threonine kinase Akt (fig. 1⇑). Since Pten^-/- mice are embryonal lethal, one had to make use of floxed Pten alleles (Pten^F/F), combined with CCSP-Cre transgene, targeting Pten deletion into bronchial Clara cells. However, these Pten^F/F;CCSP-Cre mice did not show any aberrant pulmonary development or phenotypic abnormalities even when followed for ≥12 months 64. This changed dramatically when the Pten^F/F;CCSP-Cre alleles were combined with LSLKras^G12D. Lung tumorigenesis was markedly accelerated compared to that of single LSLKras^G12D mice. Kras mutant, Pten deficient tumours were often of the more advanced adenocarcinoma type with more vascularity 64, suggesting that Pten loss cooperates with Kras mutations in NSCLC. Contrary to these results were the findings of another study in which Pten inactivation was targeted in bronchioalveolar epithelium with SP-C-rtTA;TetO₇-Cre 63. When doxycycline was applied in utero at E10-16 during embryogenesis, most mice died post-natally from hypoxia. Their lungs showed an impaired alveolar epithelial cell differentiation with an overall lung epithelial cell hyperplasia. The few surviving mice developed spontaneous lung adenocarcinomas. Post-natal doxycycline application during P21–27 resulted in a mild bronchiolar and alveolar cell hyperplasia and increased cell size but no lethality. A majority of these animals developed adenocarcinomas in comparison to WT controls. Prior addition of urethane induced an even higher amount of adenocarcinomas. Interestingly, most Pten^-/- adenocarcinomas (33%), with or without urethane addition, showed spontaneous Kras mutations. The latter observation again indicates the importance of Kras activity in cooperating with Pten loss during NSCLC development.

Recent studies showed that a large fraction of NSCLC cells had germ line mutations and impaired expression of the LKB1 tumour suppressor gene, a serine threonine kinase also known as STK11 65. Mutations in LKB1 are found in Peutz–Jeghers syndrome (PJS) patients and are characterised by intestinal polyps (hamartoma) and increased incidence of epithelial cancers 66. However, somatic LKB1 mutations have been identified in a minor fraction of these sporadic tumours, such as malignant melanomas, ovaria, breast and pancreatic cancer 67. An exception of this is formed by lung cancer in which LKB1 inactivation is a common event for NSCLC 65, 68. The highest numbers of mutations were found in adenocarcinomas 65 and were significantly more frequent in those adenocarcinomas with KRAS mutations 69. Interestingly, lung cancer does not have the highest incidence among sporadic cancers in PJS patients 67. This suggests that loss of LKB1 function might not be the primary genetic event per se which drives lung tumour initiation, but is rather an important facilitating lesion that ensures tumour progression. Results of a study of Lkb1 loss in the background of a conditional LSLKras^G12D dependent in vivo mouse model 70 showed that LKB1 inactivation cooperates with KRAS activation, suggesting a role for LKB1 as an active repressor of the KRAS downstream pathway. Moreover, single Lkb^F/F mice showed no tumour phenotype. Only Lkb1^F/F/LSLKras^G12D mice showed a broad scale of NSCLCs: the majority were adenocarcinoma but unexpectedly SCCs and large cell carcinoma (LCC) also occurred. Furthermore, 61% of adenocarcinoma progressed into metastases but none were detected for SCC and LCC. Compared to p53^F/F/LSLKras^G12D and (Ink4a/Arf)^F/F/LSLKras^G12D, the Lkb1^F/F/LSLKras^G12D mice had a higher tumour penetration and significant higher number of metastases. In addition, no SCC or LCC were detected in p53^F/F/LSLKras^G12D and (Ink4a/Arf)^F/F/LSLKras^G12D mice. Results show that LKB1 loss permits squamous differentiation and facilitates metastases but these are independent events. Adenocarcinoma from Lkb1^F/F/LSLKras^G12D mice had reduced pAMPK (phosphorylated, adenosyl monophosphate-activated protein kinase) and pACCA (phosphorylated, acetyl-CoA carboxylse α-subunit) levels and activated mTOR (mechanistic target of rapamycin protein) pathway (fig. 2⇑). It is probable that LKB1 loss influences differentiation into the NSCLC subtypes via discrete pathways 71. A large panel of human NSCLC showed LKB1 mutations in adenocarcinoma (34%), SCC (19%) and LCC (16%) 70. Concomitant mutations in p53 and LKB1 suggest nonoverlapping roles in NSCLC. Moreover, reconstitution of LKB1 in human NSCLC cell lines showed anti-tumour effects independent of their p53 or (INK4A/ARF) status 70. Finally, loss of LKB1 expression in alveolar adenomatous hyperplasia, precursor lesion for adenocarcinoma, suggests an early role of LKB1 inactivation during adenocarcinoma development 72.

Not only might direct (epi)genetic mutations in oncogenes and tumour suppressor genes affect their respective expression during lung tumorigenesis, but microRNAs (miRNAs) can also perform similar roles. These miRNAs are classes of small non-coding RNAs that post-transcriptionally regulate the expression of target mRNA transcripts. In order to become active small interfering RNA (siRNA), the primary miRNA transcripts undergo catalytic cleavage by the RNase DICER1. In human lung cancer, increased activities of DICER1 and variant regulations of miRNA clusters have been observed. For the latter, a frequent down regulation of the let-7 miRNA family as well as an upregulation of miR-17-92 has been reported 73. This miR-17-92 encodes a cluster of seven miRNAs transcribed as single primary transcript. To date, functional analyses of Dicer1 and let-7 have been performed in the background Kras induced NSCLC models. A conditional deletion of Dicer1 in the background of LSLKras^G12D;Dicer1^F/F mice let to a marked increase of tumour development 74. However, since the 3′ UTR region of Kras transcripts had been shown to be a direct target of let-7 75, it became especially tempting to increase let-7 expression in Kras^G12D lung tumours. And indeed, intranasal application of both adenoviral 76 and lentiviral 77 let-7 miRNA caused a severe decrease of Kras^G12D;p53^-/- lung tumours. These results showed the potential importance of miRNA regulation for the maintenance of NSCLC.

MODELS FOR SCLC

Contrary to NSCLC, neuroendocrine carcinomas are virtually never found in spontaneous or chemically induced murine lung cancer. One reason for this could be that in these murine lung cancers a combination of both p53 and Rb mutations is almost never found, contrary to the majority of human SCLCs. To avoid this, a Cre/lox based deletion of both conditional alleles for Rb and p53 was performed by intratracheal instillation with Adeno-Cre 78. After 3 months, various foci of neuroendocrine hyperplasia developed through the proximal as well as distal bronchi. After a further 3 months, these early lesions progressed into massive lung tumours with typical histological features of SCLC. Interestingly, some early type lesions remained even in the presence of extensive SCLC. Consequently it is of importance to determine if the early neuroendocrine lesions are indeed precursors for SCLC and if so, which additional (epi)genetic events are then needed for them to progress. Immunohistological characterisation of the full-blown tumours revealed that they indeed shared neuroendocrine features with human SCLC. All neuroendocrine differentiation markers, such as calcitonin gene-related protein (CGRP), neuron-specific enolase, synaptophysin, neural cell adhesion molecule and achaete-scute homolog-1 (ASH-1) were not expressed in the same way as in human SCLC. Furthermore, the murine SCLC readily metastasised towards similar organs as found with human SCLC 78. All primary SCLC, as well as their metastases, had all Rb and p53 alleles inactivated. Tumours that retained one WT Rb allele were all invariably adenocarcinomas without any neuroendocrine features. Therefore, the status of Rb most likely determines if tumours do occur with mixed SCLC and NSCLC phenotype, as has also been sometimes observed in patients 79. No lung tumours could be found in Rb^F/F mice, which suggested that loss of Rb alone is not enough to initiate lung tumorigenesis 78 and the additional loss of p53 is needed.

Not only does Rb loss need extra genetic events, the nature of these complementary lesions also determines which type of lung cancer will develop. For instance, RB inactivation and Kras mutations are almost never found together in the same human lung cancer. Moreover, the overall mutation rate of RB in human NSCLC is very low 80. As we have seen already, Adeno-Cre dependent activation of Kras in a broad range of lung epithelial cells leads exclusively to the onset of NSCLC. However, similar application of Adeno-Cre for inactivating RB (and p53) gives completely different neuroendocrine SCLC. Do these results imply then that Rb loss has no function in Kras driven lung tumorigenesis? Certainly not! When LSLKras^G12D transgenes were combined with Rb^F/F and Rb family p130^F/F alleles for Adeno-Cre dependent lung tumour induction, the results clearly showed much more progressed adenocarcinomas of Kras^G12D;Rb^-/-;p130^-/- genotypes compared to single Kras^G12D 81. Both Rb loss and p130, albeit to a lesser extent, contributed to Kras dependent NSCLC. Clearly, in this genetic context Kras^G12D overrules any effect of Rb loss on neuroendocrine differentiation. Another intriguing observation came from CC10-rtTA;tetO₇-Cre;Rb^F/F mice that underwent doxycycline application during early embryogenesis, causing a complete Rb ablation in all bronchial Clara cells. However, only an increase of hyper cellular neuroendocrine lesions was detected in these mice and no effect on Clara cell homeostasis could be observed. Contrary to this, when all three Rb family proteins (Rb, p107 and p130) were inactivated by a truncated SV40 large T-antigen oncoprotein (T121) in CC10-T121 mice, a severe bronchial hyperplasia with complete dedifferentiation of all Clara cells occurred. These results suggested that Rb might be specifically required for determining neuroendocrine cell fate, but only in a strict cellular and genetic context. All this accumulating evidence from mice models makes it unlikely that NSCLC and SCLC do develop from similar target cells. It would be more plausible that separate, non-identical target cells can develop into different lung cancers, although each still depends on specific major genetic pathways.

Apart from the somatic Rb^F/F;p53^F/F model for SCLC, two other lung cancer models have also been associated with pulmonary neuroendocrine tumours. One model made use of bitransgenic CC10-hASH1;CC10-SV40 large T in which progressive neuroendocrine dysplasia and aggressive lung adenocarcinoma develop with both focal neuroendocrine differentiation (through expression of pro-neural ASH-1 transcription factor) and CC10 expression 82. These adenocarcinomas closely resembled human NSCLC with neuroendocrine differentiation 83. The other model made use of inactivating both cyclin-dependent kinase inhibitor p18 (Ink4c) and Men1, a tumour suppressor gene deleted in human multiple endocrine neoplasia. Spontaneous lung tumour formation in p18^-/-;Men1^+/- mice initially led to adenoma formation which steadily grew, after a long latency period of >1 yr, into mixtures of both adenocarcinoma and neuroendocrine carcinomas 84. These mixed lung tumours were, however, distinct from those that developed in Kras^G12D through their neuroendocrine differentiation. And finally, compared with Rb^F/F;p53^F/F mice these lung tumours lacked the typical histology and metastatic pattern of SCLC.

MODELS FOR SCC

Human lung SCC is closely linked with smoking and shows a distinct sequence of pre-malignant changes in the airway epithelium from hyperplasia, metaplasia, dysplasia and carcinoma in situ towards complete invasive SCC 85. However, normal human or mouse lungs do not contain squamous epithelium. Only under pathological conditions does squamous differentiation (accompanied by high expression levels of keratins such as keratin 14 (K14)) occur in epithelium of conducting airways 86. This explains why spontaneous SCC virtually never occurs in mice and only a few mouse models reported the onset of SCC, mostly after carcinogen application. The latter models use either intratracheal intubation of methyl carbamate 87 or extensive topical application of N-nitroso-methyl-bis-chloroethylurea and N-nitroso-tris-chloroethylurea 88, 89. The whole spectrum of abnormal squamous phenotypes were observed, but only in susceptible inbred mice (NIH Swiss, A/J, Balb/cJ, FVB/J and SWR/J) and not in others (C57BL/6J, AKR/J and 129/svJ) 89. Besides, specific loci for SCC susceptibility could be identified through linkage analyses in several mice strains 89. Another study reported a complete different approach for trying to induce SCC through constitutive expression of human K14 via CC10-hK14 mice 90. Although CC10-hK14 mice had a FVB/J background and hK14 was readily expressed in bronchial epithelium, only precursor lesions varying from hyperplasia to squamous metaplasia could be observed 90. Clearly, the increased K14 expression and onset of squamous differentiation alone is not sufficient to generate fully advanced SCC. However, as we have already described, so far only the LSLKras^G12D;Lkb1^F/F somatic mouse model for NSCLC has been able to generate advanced SCC. LKB1 inactivation results in aggressive and metastatic tumours of which up to 60% have squamous or mixed squamous histology 70, even though precursor lesions were not found in the conducting airways.

MOUSE MODELS IN TRANSLATIONAL LUNG CANCER THERAPY RESEARCH

Xenograft models have been extensively used for pre-clinical testing of lung cancer therapeutics. This has mostly been performed on human lung cancer cell lines after they had been subcutaneously injected into immunodeficient mice. The more preferred method, however, would be orthotopical transplantation of human lung tumour cells in their natural pulmonary environment. The results to date show that xenograft models have a poor record of accurately predicting the clinical efficacy of anti-tumour drugs. Therefore, a justified question arises as to whether spontaneous and genetically engineered mouse models for lung cancer would be more useful as tools for pre-clinical drug tests. It is obvious that there are differences in lung physiology between mice and humans, but some of the mouse models that we have already described have a striking phenotypical resemblance, combined with a genetic signature very similar to human NSCLC 35 and SCLC 91. Importantly, genetically engineered mouse model-derived tumours develop in an innate immune environment and, therefore, use all the tumour-stromal interactions, such as angiogenesis and degradation of the tissue matrix.

Indeed, the first reports have now been published that show very promising predictive capacity of murine lung tumour models in therapy research and we will discuss two of these now.

However, first we want to briefly address some of the findings from studies in A/J mice susceptible for carcinogen-induced lung tumorigenesis. The use of standardised carcinogen-induction protocols in A/J mice generates a highly reproducible amount of well-defined lung adenocarcinoma 92. Therefore, A/J mice are widely used to test the efficacy of chemo-intervention. For example, a 50–60% tumour mass reduction of adenocarcinomas was obtained after treatment with cis-platinumin combination with metoclopramide and/or indomethacin, while adenoma growth was not affected 93. Since strain A/J lung tumour models have proved their utility for testing chemotherapeutics, they have also been widely used for evaluating chemoprevention protocols 94. A broad range of chemopreventive agents, such as farnesyl transferase inhibitors, isothyocynates, nonsteroidal anti-inflammatory drugs and glucocorticoids, as well as natural products like green tea and carotene are just some examples of what has been tested so far 95. At best, lung tumour growth was impaired by up to 50–60% 94. Introduction of p53 and p16^ink4a/p19^ARF mutant alleles into strain A/J mice caused, as expected, a higher lung tumour penetrance but not a dramatic difference on chemoprevention sensitivity 96. In order to unravel some of the undoubtedly complex genetic traits that cause a chemopreventive effect in the A/J strains, first the genetic signatures for such chemopreventive agents as green tea and budenoside have been established using gene expression arrays 97, 98.

We already described two models for NSCLC in which either the continuous oncogenic activity of Kras 29 or EGFR 61 are prerequisites of tumour maintenance. This not only shows that tumour growth critically depends on the respective active oncogenic pathways, but it also stresses the usefulness of these oncogenic pathways as therapeutic targets. Direct tumour intervention studies with tyrosine kinase inhibitors (TKIs) against EGFR mutations proved to be highly effective in several hEGFR transgenic mouse models. TKIs such as gefitinib, erlotinib and HKI-272 led to complete tumour regression 60–62. In addition, prolonged treatment with humanised anti-hEGFR antibody (cetuximab) caused a strong tumour regression 60. Further studies will be needed to investigate the pathways that determine sensitivity and resistance to EGFR related TKI interventions. Other mouse models for NSCLC were also used for targeted therapies. First, induced over expression of the PI3K p110α catalytic subunit (PIK3CA), mutated in its kinase domain (H1047R) in CCSP-rtTA;TetO₇ PIK3CA(H1047R) mice, induces adenocarcinomas. Treatment of these lung tumours with NVP-BEZ235, a dual pan-PI3K and mammalian target of rapamycin (mTOR) inhibitor, caused a marked lung tumour regression. Interestingly, when this single agent NVP6-BEZ235 was tested on lung tumours in CCSP-rtTA;TetO₇-Kras^G12D mice, no regression was observed. However, after NVP-BEZ235 was combined with MEK inhibitor ARRY-142886, significant regression of Kras^G12D tumours occurred 54. Obviously, two major RAS (rat sarcoma viral oncogene homolog protein) downstream effector pathways (fig. 2⇑) need to be impaired in order to get an irreversible regression in Ras mutated NSCLC.

Direct targeting of RAS has so far been widely unsuccessful for lung cancer therapy. Many small molecules against Ras functions have been tested and farnesyl transferase inhibitors are the most marked examples of these failed attempts 99, 100. An explanation for this can possibly be found in the fact that only KRAS4B is farnesylated but not its isoform KRAS4A. Recent results with lung cancer mouse models, as we already described previously, strongly suggest that actually KRAS4A and not KRAS4B is driving the onset of NSCLC. Although we still have to see if KRAS4A is indeed important in human NSCLC, one can imagine the importance of Kras mouse models in testing functional inhibition of KRAS4A 42.

The use of optimised genetically modified mouse models for lung cancer for therapy research necessitates sophisticated noninvasive tools to follow tumour development and response to therapy. Measurement of tumour size as a function of time is the most obvious way of doing this and existing techniques such as positron emission tomography-computed tomography or magnetic resonance imaging are adapted for use on small animals 54, 61. However, these techniques are time consuming, making them less suitable for high throughput. Other sensitive and reproducible techniques are now being widely used for measuring gene expression or tumour growth in vivo: fluorescence imaging and bioluminescence 101, 102. Expression of fluorescent proteins can be coupled to oncogenes in transgenic mouse models which facilitates fluorescent imaging of tumour growth. In case of bioluminescence, transgenic expression of luciferase allows accurate longitudinal monitoring and good quantification of tumour mass as has been shown in the LSL Kras lung tumour model 103. Feasible imaging techniques will greatly enhance the accuracy and reproducibility of mouse models.

We have shown some of the promising applications of mouse models in therapy research. However, one should not forget that mouse transgenic models are rather simplistic approximations of the human disease, although some of the somatic models do have a striking resemblance to human lung cancer. Lung tumours in mouse models are often driven by limited and defined pathways on which they remain dependent. Whether human lung cancers show similar dependencies on a continued activity of initiating oncogenes remains to be seen. The strength of genetic engineered mouse models is that it can give us insight into the importance of such particular genetic lesions on tumour growth in a therapeutic setting. As such, it can be a valuable first step in more efficient pre-clinical therapy research.

STEM CELLS AT THE ORIGIN OF LUNG CANCER

Initial mutated cells undergo a series of various genetic alterations before they finally acquire full tumorigenic capacity 104. Different early stages of tumour development have been identified for human squamous carcinoma and adenocarcinoma 6, which suggested that transformed bronchial and alveolar cells themselves can serve as cells of origin for lung cancer. As we have shown, this has been confirmed for most, but not all, genetically engineered mouse models for lung cancer. Results from NSCLC as well as SCLC models 70, 78, 81 indicated that both distinct types of genetic mutations and the cell-type in which these lesions occur control initial events of lung cancer development. However, no certainty exists about the nature of cells of origin for lung cancer. Knowledge of these cells can have an important impact, not only on our understanding of lung cancer biology but also by offering us new targets for lung cancer therapies. For many years it has been known that human lung cancer of both adenocarcinoma and SCLC contain small populations of cells (<1%) that are able to form colonies in soft agar 105. Upon transplantation into athymic nude mice, selected soft agar colonies were shown to generate tumours with similar features of the original primary lung cancers 105. Therefore, human lung cancer clearly consists of heterogeneous cell types of which only a small subset can fully renew the tumour. It is tempting to classify the regenerative or clonogenic subset of lung cancer cells as “cancer stem cells” (CSC). Being a CSC would imply that they share features or are directly derived from mutated, transformed normal stem cells. So what are the features of normal stem cells that make them such ideal tumorigenic targets? Stem cells are defined as rare, long-lived cells whose main properties include self-renewal capacity, extensive proliferation via transit-amplifying cells, and multipotent capacity through generation of mature, terminally differentiated cells 106, 107. Self-renewal allows maintenance of an undifferentiated stem-cell pool over the lifetime of the host and together with its longevity makes CSCs ideal candidates for accumulating sufficient genetic lesions needed to initiate tumorigenesis. Multidrug-resistance is an additional property of normal stem cells, which secures their longevity in the face of toxic agents, including many of the well-known drugs used in lung cancer therapy. Functional multidrug-resistance is often mediated by over expression of adenosine triphosphate-binding cassette (ABC) transporter that efflux drugs 108. Indeed, over expression of ABC transporters have been described in clonogenic subsets and are a probable cause for chemoresistance of the treated tumours 108. The latter observation could also be of particular interest for human lung cancer since recurrence after chemotherapy and radiotherapy is common in human lung cancer 109. A defined chemoresistant CSC population could be a potential source of tumour recurrence and thus become in itself a valuable therapeutic target.

However, only recently have similarities between cancer cells and normal stem cells led to the idea of the presence inside epithelial tumours of a small population of CSC or “tumour-initiating cells” 108. CSCs are defined as a rare population of undifferentiated tumorigenic cells responsible for tumour initiation, maintenance and spreading 106, 108. These cells have been identified in many types of cancers by sorting cell subpopulations based on surface marker expression patterns, and by transplantation studies into animal models using prospectively isolated tumour cells. Their existence was first documented in the context of leukaemia 110, 111. Following this, they have been identified in breast cancer 112 and glioblastoma 113. Lately, “tumour initiating cells” have been identified in a wide variety of tumours including prostate, colon, pancreatic and hepatic carcinomas, melanoma and several other tumour types 114. These studies have shown that these cells present an unlimited renewal potential and that they can undergo differentiation leading to the formation of tumours in immunodeficient mice that recapitulate the heterogeneity of the original tumour at high efficiency 115. Nevertheless, at present it is still not clear whether CSCs exist inside a solid tumour. However, another possibility for the ontogeny of cancer is the so-called stochastic model. This model foresees the onset and progression of cancer from multiple transformed progenitor cells leading to a polyclonal tumour, in which each individual subclone can, in principle, reconstitute a complete tumour 116. Thus, the latter model predicts that each transformed, undifferentiated cell has the same tumorigenic capacity and the tumour will comprise of a selection of the most dominant subclones 117. A relative high percentage of cells from a single tumour should, therefore, be able develop into full-blown cancer after xenotransplantation in immunodeficient mice 118.

Thus, there is evidence of distinct lung cancer cell populations and this necessitates a better understanding of how normal bronchial and alveolar tissues are organised in order to identify lung stem cells.

Formation and renewal of lung tissue is largely endogenous as it is now generally accepted that the formation of lung epithelium by circulating mesenchymal progenitor cells is infrequent and not of physiological relevance 119. Stem-cell research has progressed rather slowly in the lung compared to other organs due to the anatomical and functional complexities associated with cellular heterogeneity and the low turnover rate of its epithelium. Only after being challenged by environmental and nutritional factors does the lung epithelium show a dramatic increase in its turnover rate. This is in contradiction to the constant rapidly renewing tissues of intestinal and haematopoietic origin. These fast renewing tissues have a classical hierarchical organisation of stem cells, and transit-amplifying progenitors and their differentiated progeny. The pulmonary system, however, contains a variety of nonclassical organised transit-amplifying cells, (facultative) progenitor cells and terminal differentiated cells that are anatomically separated in the conducting airway epithelium and gas-exchange alveolar regions 120. These different pulmonary epithelial cells were mainly identified by airway injury experiments. Various toxic agents were used to trigger the selective damage response of particular lung epithelial cells after which their increased proliferation rates were followed in vivo through [³H] thymidine or bromodesoxyuridine (BrdU). In pulse/chase experiments, only self-renewing stem cells retain the label for an extended period due to slow turnover and are so-called label retaining cells (LRCs).

Although the distal epithelium of the lung is composed of Clara, ciliated and neuroendocrine cells, only [³H] labelled Clara-like cells were found to be able to proliferate and generate de novo differentiated Clara cells and ciliated cells after exposure with ozone gas had initially damaged all ciliated cells 121. This provided evidence that Clara cells are facultative progenitor cells, but it did not provide evidence for a stem cell hierarchy within the Clara-cell population. Another toxic agent, naphthalene, induces ablation of almost all nonciliated bronchiolar epithelial Clara cells due to selective metabolic activation of the toxin. This treatment results in the proliferation of two different populations of LRCs in regions localised to neuroendocrine bodies (NEBs) in the distal airway, namely Clara-like cells expressing CC10 and pulmonary neuroendocrine cells (PNECs) expressing CGRP 122. Both naphthalene-resistant Clara-like cells (named variant or vClara cells) and the PNECs are able to maintain epithelial renewal and their localisation in NEBs is believed to represent a stem-cell niche 123–125. vClara cells are resistant to naphthalene injury because they do not express the enzyme Cyp2f2, which is a member of the cytochrome P450 family that converts naphthalene into a toxic compound. Besides this, vClara cells are also able to efflux drug via multidrug-resistant receptors and present a side population phenotype 126.

A transgenic mouse model in which the herpes simplex virus thymidine kinase (TK) gene is expressed under the control of the CC10 promoter (CC10-TK) was used to test whether PNECs or vClara populations represent the stem cell pool with pluripotent capacity for airway epithelial repair 127. In this model the CC10-expressing (e.g. both vClara and normal Clara) cells can be selectively ablated in the airways of mice through acute administration of ganciclovir. Studies on the regenerating epithelium showed that the ablation of CC10-expressing cells led to PNEC proliferation and hyperplasia but they are unable to effectively repopulate a differentiated ciliated epithelium 127. PNEC hyperplasia independent of NEB-associated vClara cells may represent a compensatory mechanism to repopulate the airway in the setting of total stem-cell ablation 128. Hence, these studies lead to the conclusion that either a subpopulation of CC10-expressing cells, but not PNECs, may represent stem cells required for repair of Clara cell injury in the mouse distal airway or that these cells may be required to support the stem cells that give rise to new Clara cells 127.

A second epithelial stem-cell niche has been identified after similar naphthalene treatment 123, 129. An LRC subpopulation of Clara cells was found near terminal bronchi, at the bronchioalveolar duct junction (BADJs) where airways terminate and form alveoli and only very few NEBs are observed 123, 124. These cells have been termed bronchioalveolar stem cells (BASCs) 129. They co-express CC10 and SP-C, which is a major secreted compound of type II alveolar cells but expressed at low levels by early embryonic lung epithelial cells 130. However, dual expressing CC10/SP-C cells were not found in late embryonal lungs, so it is likely that these BASCs develop after birth in rapidly expanding pulmonary epithelial tissue.

This double positive (CC10 and SP-C) population is quiescent in normal lung 131. BASCs were found to express cell surface marker Sca-1 and CD34, which are established stem-cell surface markers in other mouse tissues 132. Viable (Sca-1+, CD34+) BASCs were able to be isolated by flow cytometry after excluding haematopoietic (CD45+) and endothelial (CD31+) cells 129. They appear to be crucial for the regeneration of epithelial components of terminal bronchioles and alveoli. Moreover, BASCs are capable of several passages in vitro while either maintaining their (CC10+/SP-C+) status or developing into Clara (CC10+/SP-C-), alveolar type II (CC10-/SP-C+) or alveolar type I (SP-C+/aquaporin5 +) cells. Thus, BASCs were shown to be capable of multipotent differentiation as well as self-renewal. At present, we still do not whether these cells function as true stem cells as the evidence for their self-renewal and multipotency in vivo is, as yet, preliminary. Clearly better characterisation of BASCs is needed to identify more unique surface markers, as has been shown by several isolation studies 133, 134 but also to investigate how similar or identical vClara cells and BASCs are to each other. Indeed, all the evidence so far indicates that both vClara and BASCs are candidates for bronchial stem cells and normal Clara cells are facultative transit-amplifying cells.

Another putative lung stem-cell population was found through the use of a serum-free culture system for primary neonatal pulmonary cells. The arising colonies were positive for CC10 and Sca-1 but negative for SP-C and CD34. In addition, cells were found to express octamer-binding transcription factor 4 (Oct4) and stage-specific embryonic antigen-1. BrdU pulse-chase experiments indicated that slow-cycling label retaining Oct4 positive cells are localised at the BADJ 135. These Oct4 positive cells were able to differentiate into both alveolar type 2 or 1 cells. Again it remains to be seen which similarities this candidate bronchial stem cell shares with the other two candidates. Studies on the finding of CSC in human lung cancer were mostly directed on isolating distinct cell populations from primary tumours. To date, no clear robust markers have been characterised that could be exclusively linked with a clonogenic CSC populations. Some of these results have been reviewed elsewhere 136. Furthermore, since there is no known human homolog of Sca-1, it becomes obvious that BASCs but also vClara cells need further characterisation in order to find their human counterpart. Recently, it has also been demonstrated that both human SCLC and NSCLC contain small populations of undifferentiated cells expressing the cell surface marker prominin-1 (CD133) 137. Cd133 has been found to be expressed in cancer-primitive cells as well as normal cells from neural, endothelial, epithelial and haematopoietic lineages 138. Lung cancer CD133 positive cells are able to grow indefinitely as tumour spheres in particular culture conditions and they can generate tumour xenografts in immunocompromised mice, phenotypically identical to the original tumour, while CD133 negative cells do not 137. Lung cancer contains a rare population of CD133 positive cancer stem-like cells that are able to self-renew and generate an unlimited progeny of nontumorigenic cells 137. Chemoresistance was observed in the tumour spheres and was associated with increased expression of ABC transporter and embryonal stem cell markers Oct4 and NANOG. Interestingly, after naphthalene treatment in normal mice, a rare population of CD133 positive cells increased. Therefore, it would be very tempting to suggest that those expanding CD133 positive cells are BASCs. Follow-up studies are needed to substantiate this and to translate some the other findings from the mouse models into a better characterisation of human lung CSC.

The bronchial stem cell population could be an ideal candidate for lung cancer progenitors. As these lung stem cells are already endowed with the capacity for self-renewal, they are excellent candidates for the accumulation of deleterious mutations. In fact, accumulating evidence already exist for BASCs that this is indeed the case.

Cells expressing both SP-C and CC10 were first described in the precursor lesions of murine lung tumours initiated by oncogenic Kras 33. Expression of the oncogenic protein KRAS in these double positive cells induces their proliferation in vivo 33. Moreover, induction of tumorigenesis in naphthalene-treated LSL-Kras^G12D mice induced an increase in the number and size of tumours 129. Furthermore, Kras^G12D expression causes BASCs in vitro expansion and generates hyperproliferation at the BADJ in vivo 129. Similarly, BASCs from p38α MAPK^-/-mice were also highly sensitised to Kras^G12D-induced lung tumorigenesis, leading to an immature and hyperproliferative lung epithelium, thus permitting earlier induction and faster progression to adenocarcinoma 139. Abrogations of major signalling pathways were shown to be important for the proliferation and maintenance of BASCs. Conditional deletion of PTEN or PI3K in mice result in increased numbers of BASCs and increased propensity to lung tumour development 63, 140. In p27^-/- mice, BASCs expand first at the BADJ and progress from hyperplasia to adenocarcinomas through dysplasia and adenomas 141. In contrast, deletion of Bmi1 inhibits tumorigenesis through inhibition of BASC expansion 142. Taken together, these studies show that BASCs may play a role in the initiation of tumorigenesis but there is still no direct proof of being the initial target cell per se. Indeed CSC may arise from bronchial stem cells but we cannot exclude the fact that they develop from transit-amplifying cells, such as PNECS and Clara cells with each having acquired the ability to self renew as a result of oncogenic mutations. In addition to the acquisition of genetic and epigenetic mutations, interactions between tumour cells and their microenvironment (also called niches, which are composed by the stroma, inflammatory cells and recruited vasculature), has a profound influence on the tumorigenic process. More and more evidence shows that tumour growth can be sustained not only by rare CSC populations but also through dominant (sub) clones or a mixture of both 117. Although the importance of characterising CSC of lung cancer is of obvious importance, it would be too optimistic as to say that targeting of CSC would cure lung cancer. Most likely a combined targeting of all the different populations is needed to impair tumour growth and prevent long-term recurrence.

PERSPECTIVES AND NEW DIRECTIONS

The use of mouse models is helping us to understand lung cancer biology by uncovering the critical molecular pathways that govern each stage of tumour formation and progression. It will be important to adjust mouse models for each different human lung cancer type. Progress on genome-wide expression profiling and genomic hybridisation will be helpful in cross-species analysis of similarities in cancer signalling pathways that govern both mouse and human lung tumorigenesis. These studies will be very useful for the design of inhibitors, which can impair lung tumour growth and can be tested in pre-clinical mouse models. New, better adapted mouse models with complete human genes could lead to the development of humanised mice that recapitulate human lung cancer at en even higher degree 32. Rapid advances of lung cancer genome analysis will reveal many additional novel mutations that could be complemented into mouse models for lung cancer.

Metastatic mouse models for both NSCLC and SCLC are now well-established and it will be a challenge to use these models to unravel the mechanism by which lung cancer is often refractory to chemo- and/or radiotherapy. Defined early stages of lung cancer development in combination with noninvasive imaging technologies in these mouse models will be very helpful for testing chemoprevention protocols. Another very interesting aspect of the conditional lung tumour models is that they can yield promising biomarkers for lung cancer. Advanced mass spectromy-based proteomics will allow scrutinising analysis of the mouse serum proteome, under both normal and cancerous conditions. There is a clear need for clinically relevant biomarkers of early lung cancer and both somatic mouse models for NSCLC and SCLC show distinct stages in lung tumour development. Combining gene expression profiling with proteomics of these early murine lung tumours will hopefully lead to the discovery of new, relevant biomarkers that can be translated into diagnostic use. A major challenge will not be the discovery but the evaluation of candidate biomarkers through proper validation on patient samples.

The extra-ordinary power of mouse genetics will continue to help us identify lung tumour precursor cells by selective targeting of lung epithelial cell compartments. Candidate precursor cells could then be tagged with reporter genes so one can follow whether they behave like cancer stem cells during tumour growth.

Much progress has been achieved on mouse models for human lung cancer (table 1⇓, fig. 3⇓), but further characterisation is still needed for most NSCLC and SCLC models. It is also important to note that the development of mouse models for SCC clearly lacks behind. A better knowledge of the cell of origin of lung SCC, together with identifying the most suited genetic lesions that cause this lung cancer type, might facilitate proper design of a mouse model.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3—

Representative lung tumours from mouse models for advanced nonsmall cell lung cancer and small cell lung cancer. The progressed adenocarcinoma from Trp53^F/F;LSLKras^G12V mice 3 months after Adeno-Cre treatment are shown (a–c; unpublished data). a) Adenocarcinoma in which some tumour cells have irregular nuclei (arrows). b, c) Further advanced adenocarcinoma showing tumour cells with pleiomorphic nuclei (white arrows) and giant cells (black arrows). d) Early neuroendocrine hyperplastic lesions inside bronchi of Rb^F/F;Trp53^F/F mice 3 months after Adeno-Cre treatment which, after a further 3 months, progressed into small cell lung cancer (e, f). Note the characteristic moulding of nuclei of adjacent tumour cells and the sparse amount of cytoplasm. Haematoxin and eosin stain. a, b, d, e) Scale bars = 25 μm. c, f) Scale bars = 10 μm.

Finally, we have seen that mouse models for lung cancer have already helped us to further our insight in basic lung cancer biology and test lung cancer therapies. These models still hold many promises within these fields and should be expanded by finding new markers for lung cancer diagnosis. Results will undoubtedly fulfil some of these promises and be importance in the fight against human lung cancer.

Support statement

S. de Seranno was supported by a Postdoctorate grant (Instiut National du Cancer, Paris, France) and R. Meuwissen was supported by a Jeune Chercheur fellowship (INSERM, Paris).

Statement of interest

None declared.