dlk, a Putative Mammalian Homeotic Gene Differentially Expressed in Small Cell Lung Carcinoma and Neuroendocrine Tumor Cell Line*

Gastrin releasing peptide is mitogenic for mouse Swiss 3T3 fibroblasts and certain human small cell lung carcinoma (SCLC) cells but not for mouse Balb/c 3T3 fibroblasts. To identify new molecules associated with the gastrin releasing peptide-responsive pheno- type, clones isolated from a differential cDNA library between Swiss and Balb/c 3T3 fibroblasts were used to screen for their expression in human SCLC cell lines. Using this approach, we have isolated and characterized human and mouse cDNA clones encoding a novel protein. This protein is a putative transmembrane protein belonging to the epidermal growth factor-like su- perfamily. In vitro transcription and translation stud-ies detect a 42-kDa protein, in agreement with the size predicted from the translated cDNA sequence. This protein (termed Delta-like or dlk) is highly homologous to invertebrate homeotic proteins, including Delta, and Notch, the products of neurogenic loci involved in normal neural differentiation in Drosophila. dlk is ex- pressed in tumors with neuroendocrine features, such as neuroblastoma, pheochromocytoma, and a subset of SCLC cell lines. However, its expression in normal tissues is restricted to the adrenal gland and placenta. These data suggest that dlk may be involved in neuroendocrine differentiation and, because of

human lung cancers. Several cell lines of SCLC have been established, which differ in their expression of several neuroendocrine markers (1,2). In particular, some SCLC cell lines produce gastrin releasing peptide (GRP), which interacts with a specific G-protein-coupled receptor on the cell membrane (3). GRP has been reported to be a mitogen for normal human epithelial lung cells, for mouse Swiss 3T3 fibroblasts, and for some SCLC (4). For these SCLC cells, GRP seems to act as an autocrine growth factor (4). In contrast, GRP is not mitogenic for some other SCLC cells and mouse Balb/c 3T3 fibroblasts.
The precise genetic differences between GRP-responsive and unresponsive cells are unknown. The immediate purpose of this investigation was the identification of new genes associated with the GRP-responsive phenotype in order to improve our understanding of the autocrine mechanism of GRP action in SCLC. This, in addition, could lead to the development of new therapeutic approaches for some SCLC. Even if the molecules identified were not directly involved with the GRP-response mechanism, their characterization still could provide new information about the differences in gene expression in SCLC associated with the differentiation state of these tumors. Furthermore, these molecules might constitute new and useful markers for the detection and staging of SCLC.

MATERIALS AND METHODS
Library Screening and Hybridization Techniques-The differential library of Swiss 3T3 compared with Balb/c 3T3 fibroblasts was constructed as explained in detail previously (5). RNA isolation, electrophoresis, Northern blots, and hybridization techniques were performed according to standard protocols (6). The cDNA probes were labeled with [32P]dCTP (Amersham Corp.) by the random primer method (7). The mouse Swiss 3T3 fibroblast XZAPII cDNA library used to obtain a mouse clone of dlk was purchased from Stratagene (La Jolla, CA). Screening procedures and plasmid rescue of positive X clones were performed following the manufacturer's protocol. Those plasmids were sequenced with Sequenase (United States Biochemical Corp.) by the chain termination method following the manufacturer's protocol, using [(u-~'S]~ATP (Amersham Corp.) as labeling agent.
To obtain a human dlk clone, a X g t l O human adrenal gland cDNA library was purchased from Clontech (Palo Alto, CA) and the screening procedure was performed following the manufacturer's protocol. Positive X clones were subcloned into pGEM4Z (Promega, Madison, WI) and sequenced as described above.
The study of the presence of dlk gene in different species was done on a dried agarose gel wafer of DNA digested with EcoRI (Clontech). The hybridization with a mouse dlk probe was performed following the manufacturer's protocol. At the same time, 5 pg of Drosophila and Xenopus DNA were digested with EcoRI, and electrophoresed in a 1% agarose gel. The gel was then dried and hybridized following Clontech procedures.
I n Vitro Translation Studies-dlk mRNA was selected by hybridization of 2 pg of poly(A+) Swiss 3T3 RNA with 5 pg of nitrocelluloseimmobilized denatured dlk cDNA. The RNA bound was eluted by boiling. RNA was also prepared in vitro from two different full-length mouse dlk cDNAs cloned in pGEM4Z by using the Riboprobe Gemini System I1 (Promega), following the manufacturer's protocol. Those RNAs were then used for in vitro translation assays. The reticulocyte lysate kit used for those assays was purchased from Du Pont-New England Nuclear. The procedure was performed following the manufacturer's protocol. Labeled proteins were analyzed in a 12% polyacrylamide gel followed by fluorography.
Sequence Analysis-Most DNA and protein sequence analyses were performed with the software package PC/Gene (8). The open reading 3817 N P W Homeotic-lihc Gcnc in SCIL' and Neuroendocrinc Tumors frames were found hy using the methods of Fickett (9) and Shepherd (10). The transmemhrane domain of d / k was found with the program RAOARGOS ( 1 1 ). The signal peptide was analvzed with the program PSIGNAI,. according to Von Heijne (12). The alignment of the IWFlike repeats of Fig.  3 was done with the program CLIJSTAI,. The sites of potential hiologicnl importance were analyzed with the program PIIOSITE. The hnmoloky searches in the DNA and protein sequence data hases were performed with the program FASTA (13). This program was created hy the IJniversity of Wisconsin Genetic Computer Group (IJWMX). The program was used to search in the following data hases: GenHank release 69.0; EMHI, release 26.0, NHRF-protein release 26.0, and Swissprot release 18.0.
To study the statistical significance of the homologies found, we used the method of Needleman and Wunsch (14). In this method. the optimal alignment score hetween two proteins is compared with the statistical tlistrihution of a number of random alignments, 100 in our case. An optimal alignment score o f more than 5 standard deviations t o t h e right ofthe mean score of the random alignment distrihution is considered significant. particularlv when no functional or structural relationship hetween the proteins compared is known.

RESULTS AND DISCUSSION
T o identify new molecules associated with the gastrinreleasing pept.ide (GRP) responsive phenotype, we searched for cDNAs, which, in addition to being differentially expressed between Swiss (responsive) and Ralb/c (unresponsive) 3T3 fibroblasts, were expressed in GRP-responsive SCLC cell lines. This approach was based on the assumption that gene products related to the GRP-responsive phenot-ype should be missing from Halb/c and unresponsive SCLC cell lines but present in Swiss 8T3 fihroMasts and certain responsive SCLC cell lines. A differential library was constructed that enriched for clones expressed in Swiss 3T3 but not in Ralb/c 3T3 fibroblasts (3). A partial length clone (150 nucleotides long) isolated from this differential library hybridized with a 1.6kb mRNA showing a pattern of expression compat.ible with the two screening requirements. To obtain a full-length cDNA clone, we used t,his partial length clone to screen a commercial oligo(dT)-primed cDNA library of Swiss 3T3 fibroblasts in t h e XZAPII vector. Several clones with inserts around 1.6 kilohase pairs were obtained and sequenced.
Nucleot.ide sequence analysis of the cDNAs defined an open reading frame of 1155 nucleotides. F3ot.h Ficket.t's (9) and Shepherd's (10) methods classified this open reading frame as coding. This open reading frame encodes a putative protein ( d l k ) of 385 amino acids with a molecular mass of 41,320 daltons. To confirm the existence of the protein predicted from t.he dlk cDNA sequence, we performed in vitro translation assays from mouse dlk mRNA using a rabbit reticulocyte lysate system. dlk mRNA was selected by hybridization of poly(A') RNA from Swiss 3T3 fihrohlasts with denatured full-length dlh cDNA immobilized on nitrocellulose filters. Mouse dlh RNA was also prepared by in vitro transcription from two independent, full-length cDNA clones. These three RNAs were used as templat.es for in vitro translat.ion. A protein of around 42 kDa was present in all three samples (Fig. l), in agreement with the molecular mass of dlh predicted from its cDNA sequence.
Structural analysis revealed that dlh is a transmembrane protein which contains six EGF-like repeats in the ext,racellular domain, and a short intracellular domain. A signal peptide is also present. at the amino terminus (12) (Fig. 2). Hased on these structural features, dlh appears to be a new member of the family of EGF-like homeotic proteins, which a r e implicated in differentiation decisions related to t.he commitment of a given class of cells to become part of a fully different,iated struct.ure (15)(16)(17). In fact, a computer search within the Swissprot and NRRF protein sequence data bases (13) showed that. t.here was a high degree of homology bet.ween including D d t n (18). S c w n k (19), and Notch (20,21)  receptors or other neuroendocrine markers. T o reveal possible homologies of other regions of dlk not containing EGF repeats, FASTA searches within Swissprnt and NRRF data bases were performed with a partial dlk protein sequence devoid of the FXF-like regions. lising t h r method of Needleman and Wunsch (14). no significant homologies were found.
A search of DNA nucleotide sequence homologic~s in GenRank and EMRI, data hases (I:{). identified a X1.Y; sequence identity with pC2, a human cDNA cloned from an adrenal gland cDNA library (27). p(;2 is a gene exprrssed in neuroendocrine tumors whose expression in normal t issrrrs seems limited to the adrenal gland and to specific differrntiation stages in the development of neuroendocrine tissue (28. 29). However, the putative protein coded for hv pC2 (27)  wafer was purchased from Clontech, and the hybridization was performed following the manufacturer's protocol. Five pg of Drosophila and Xenopus DNA were digested with EcoRI and run in a 1% agarose gel in TBE buffer, and the gel was treated following Clontech procedures. The hybridizing bands are, for the most part, of a similar size which is consistent with a gene size of 25-30 kb.
structural characteristics similar to those reported for pG2. Sequence data from several human clones isolated indicated that these cDNAs coded for the human counterpart of the mouse dlk protein (Fig. 2). The human dlk protein consists of 383 amino acids and has a molecular mass of 41,154 daltons. Mouse and human dlk protein sequences are 86.2% identical and share many potential sites of biological importance, including six EGF-like repeats, a transmembrane region, and a signal peptide domain. Our inability to clone pG2 cDNA precludes clarification of the precise relationship between dlk and pG2. Based on published observations, it appears that the pg of total RNA or 2 pg of poly(A+) RNA from SCLC cell lines and Swiss and Balb/c 3T3 fibroblasts were run in a 1% agarose gel and then blotted on a nitrocellulose filter. A 1.6-kb band corresponding to dlk can be observed only in the SCLC cell lines NCI-N592, NCI-H69, and NCI-H510, as well as in Swiss 3T3 fibroblasts. In B, a human multiple tissue Northern blot (Clontech), containing 2 pg of poly(A+) of each tissue, was used for hybridization, following the manufacturer's specifications. Only placenta shows high levels of expression; the rest of the tissues studied are negative. However, despite the manufacturer's assertion that equal amounts of RNA were loaded per lane, actin binding shows that this is not the case. This variation certainly affects the relative sensitivity for each tissue, particularly for kidney, which shows the lowest level of actin binding. two cDNAs may belong to the same gene superfamily but their protein products show significant structural and, presumably, functional divergence. dlk shows a high degree of homology with the EGF-like proteins of Drosophila and other invertebrates that are involved in the differentiation of different tissues and structures (18-24). Fig. 3 shows the alignment of mouse or human dlk EGF-like repeat consensus sequence with consensus sequences of the EGF repeats of several proteins; residues well conserved among homeotic genes are also conserved in dlk, New Homeotic-like Gene in SCLC and Neuroendocrine Tumors confirming dlk as a new member of the family of EGF-like homeotic proteins. The amino acid sequence and structure of the EGF-like repeats, as well as the overall structure of dlk, are more closely related to the products encoded by invertebrate homeotic genes than to other vertebrate non-homeotic EGF-like proteins such as the EGF-precursor, transforming growth factor a, the a, pl, and /32 chains of laminin, coagulation factors, or complement proteins, previously thought to be the mammalian counterparts of the invertebrate homeotic genes (18, [20][21][22]30). For that reason, it was interesting to study the species distribution of the dlk gene. Fig. 4 shows that the dlk gene is present in species ranging from birds to human. However, despite its structural homology with invertebrate proteins, dlk is absent from invertebrates and low vertebrates. To our knowledge, this is the first protein homologous to invertebrate homeotic products that is exclusively present in higher animals. This suggests that dlk may possess a function specific to this class of animals for which the conservation of the invertebrate homeotic EGF-like repeats and protein structure is important. In this regard, it is noteworthy that proteins belonging to the EGF-like superfamily are implicated in protein-protein interactions (31). Furthermore, it has been described that the interactions through specific EGF-repeats between homeotic proteins play a role in the transduction of differentiation signals (32,33). This suggests that, to fulfill its function, dlk may interact with other hitherto unknown proteins through its EGF-like repeats.
Expression of dlk can be detected by Northern blot or RNAse protection analysis in the SCLC cell lines NCI-H510, NCI-H69, and NCI-N592 (Fig. 5 A ) , in SCLC NCI-H146, in the human neuroblastoma SK-N-SH, and in the rat pheochromocytoma PC-12 cell lines (data not shown). The data of dlk expression in the different SCLC celI lines shows no correlation between dlk expression and GRP responsiveness, despite this being the initial criterium for selection of differentially expressed genes in a limited number of SCLC cell lines. The Ewing's sarcoma cell lines SK-ES-1, A4573, and TC106, the breast cancer cell lines MCF7, MDA 231, MDA 453, and MDA 468, the prostate cancer cell lines PC3 and LNCAP, and the monocytic cell line U937 do not express dlk (data not shown). Mouse Swiss 3T3 fibroblast RNA also showed a high degree of expression of dlk, but Balb/c 3T3 fibroblast RNA is negative for dlk expression. In normal tissues of hamster (data not shown) and human origin, dlk expression can be detected exclusively in the placenta and adrenal gland, from which it was cloned (Fig. 5B). These data, together with the homology of dlk with homeotic proteins, suggest that dlk may play a role in the differentiation of the neuroendocrine phenotype. SCLC and neuroblastoma are the only tumors known to express dlk, and this expression is probably associated with some differentiation stages, at least in the case of SCLC. This offers the potential prognostic use of dlk expression for the staging of SCLC and neuroblastomas. Moreover, considering that dlk is a transmembrane protein, the expression of which in normal tissues is restricted to the adrenal gland, dlk may be a readily accessible target for antibody imaging or therapy of SCLC and neuroblastoma tumors.