Identification of Novel and Widely Expressed Cancer/Testis Gene Isoforms That Elicit Spontaneous Cytotoxic T-Lymphocyte Reactivity to Melanoma

INTRODUCTION

Significant progress has been made in identifying antigenic peptides which can be used to stimulate a tumor-specific CTL response (1, 2, 3) . The tumor antigens from which the peptides are derived can broadly be categorized as differentiation antigens, cancer/testis antigens, mutated gene products, widely expressed proteins, and viral antigens (1, 2, 3) . Although initial trials of peptide immunization of malignant melanoma patients have yielded promising results (4, 5, 6) , truly effective immunotherapy is likely to depend on vaccination with multiple tumor-specific peptides that are derived from different proteins and that are presented by different class I MHC-encoded molecules. This is necessary both to provide coverage of the broadest possible segment of the patient population and to minimize the immunoselection of tumor variants that may evade the vaccine-induced CTL response (7, 8, 9, 10) .

To increase the number of antigenic peptides available for the treatment of melanoma, the identification of additional HLA-A3-restricted epitopes was pursued. Using a previously described melanoma patient-derived, tumor-specific CTL line (11) , we identified an additional HLA-A3-restricted peptide. The source protein for this peptide is a novel cancer/testis antigen that is widely expressed in melanoma and may be a good target for immunological therapy.

MATERIALS AND METHODS

Cell Lines.

The melanoma cell lines A375, AVL3-Mel, DM6, DM13, DM14, DM93, DM122, DM281, DM319, DM331, EB81-Mel, HT144, LB373-Mel, Na8-Mel, SK-Mel-2, SK-Mel-5, SK-Mel-28, VMM1, VMM5, VMM12, VMM15, VMM17, VMM18, VMM19, VMM34, VMM39, VMM64, VMM86, VMM105, VMM150, VMM273, and VMM330 were maintained in RPMI 1640 supplemented with 5–10% fetal bovine serum and 2 mm l-glutamine. K562, a myelogenous leukemia cell line (12) , and the B-lymphoblastoid cell lines VMM12-EBV, VMM18-EBV, and JY were maintained in the same medium. C1R-A3 and T2-A3, were maintained in the same medium supplemented with 200 μg/ml G418.

CTL Line.

VMM18-specific CTLs have been described previously (11) . CTLs were expanded in bulk culture using anti-CD3 antibody (13) and cryopreserved in aliquots of 1–5 × 10⁷ cells for use in epitope reconstitution assays.

Isolation of Peptides Associated With the HLA-A3 Molecule.

Immunoaffinity purification of class I MHC molecules from aliquots of 6–8 × 10¹⁰ VMM18 tumor cells was performed as described previously (14) , except that the HLA-A3-specific monoclonal antibody GAP-A3 (15) , bound to protein A-Sepharose, was used to isolate the HLA-A3 molecules.

Peptide Fractionation.

Peptide extracts were fractionated by reverse-phase–high-performance liquid chromatography (RP-HPLC) using an Applied Biosystems model 140B system. The extracts were concentrated by vacuum centrifugation and injected onto a Higgins (Mountain View, CA) C₁₈ HAISIL column (2.1 mm × 4 cm, 300 Å, 5 μm). The peptides were eluted with a gradient of acetonitrile/0.085% trifluoroacetic acid (TFA) in 0.1% TFA/water, with the concentration of acetonitrile increasing from 0 to 9% (0 to 5 min), 9 to 36% (5 to 55 min), and 36 to 60% (55 to 62 min). Second dimension fractionations of selected first-dimension (TFA) fractions were accomplished using the same gradient but with the substitution of heptafluorobutyric acid for TFA. A third dimension of RP-HPLC was achieved using an Eldex (Napa, CA) MicroPro pump, a homemade C₁₈ microcapillary column and an Applied Biosystems model 785A UV absorbance detector. The column was made by packing a 27-cm bed of 10-μm C₁₈ particles in a section of 285-μm outer diameter × 75-μm inner diameter fused silica. Peptides in a selected second-dimension fraction were loaded onto this column and were eluted with a gradient of acetonitrile/0.67% triethylamine acetate/water in 0.1% triethylamine acetate/water, with the concentration of acetonitrile increasing from 0 to 60% in 40 min. The flow rate was ∼300 nl/min, and fractions were collected into 25 μl of 0.1% acetic acid every 30 s. In all of the RP-HPLC experiments, peptides were detected by monitoring UV absorbance at 214 nm.

CTL Epitope Reconstitution Assay.

Aliquots of each RP-HPLC fraction were tested for the presence of peptides that could sensitize C1R-A3 targets for lysis by VMM18 CTLs in standard 4-h ⁵¹Cr-release assays as described previously (14) .

Mass Spectrometric Analyses.

Active RP-HPLC fractions were screened by online RP-HPLC/electrospray ionization mass spectrometry using a homemade microcapillary column and a Finnigan-MAT TSQ 7000 triple quadrupole mass spectrometer (Finnigan, San Jose, CA). Approximately 1% of the active RP-HPLC fraction was loaded onto a section of 185-μm outer diameter × 75-μm inner diameter fused silica packed with 10–12 cm of 10-μm C₁₈ particles. Peptides were eluted directly into the mass spectrometer using a 10-min 0–60% acetonitrile in 0.1 m acetic acid gradient. Ions were formed by electrospray ionization, and mass spectra were recorded by scanning between mass:charge ratios (m/z) 300 and 1400 every 1.5 s.

Active second-dimension RP-HPLC fractions were analyzed using an effluent splitter on the microcapillary HPLC column. The column (360-μm outer diameter × 100-μm inner diameter with a 25-cm C₁₈ bed) was connected with a zero dead volume tee (Valco, Houston, TX) to two pieces of fused silica of different lengths (25 μm and 40 μm inner diameter). Peptides were eluted with a 34-min gradient of 0–60% acetonitrile in 0.1 m acetic acid. The 25-μm capillary deposited one-fifth of the RP-HPLC effluent into the wells of a microtiter plate for use in a CTL epitope reconstitution assay, and the remaining four-fifths of the effluent was directed into the mass spectrometer, with mass spectra recorded as described above (14) .

Peptide sequences were determined by collision-activated dissociation tandem mass spectrometry using an LCQ (Finnigan) ion trap mass spectrometer and methods as described previously (16 , 17) .

Peptide Synthesis.

Peptides were synthesized using a Gilson (Madison, WI) AMS 422 multiple peptide synthesizer using conventional FMOC [N-(9-fluorenyl)methoxycarbonyl] chemistry. Peptides were purified by RP-HPLC using a 4.6-mm inner diameter × 100-mm long POROS (Perseptive Biosystems, Cambridge, MA) column and a 10-min 0 to 60% acetonitrile in 0.1% TFA gradient.

Total mRNA Isolation.

Total RNA was prepared from 2–10 × 10⁶ cells using the RNeasy Mini kit (Qiagen, Valencia, CA) as per the kit instructions. RNA was quantified by absorbance at 260 nm.

PCR Primers.

The gene-specific primers (GSPs) 1361 and 1362 are specific for GAPDH and the remaining primers are directed toward the TAG gene. The primers are as follows: 1361, 5′-CCACCCATGGCAAATTCCATGGCA-3′; 1362, 5′-TCTAGACGGCAGGTCAGGTCCACC-3′; A52, 5′-AGGAA GGGGCTCCCACAGTGC-3′; A73, 5′-AGCGGCGGGCTGAAGGA-3′; A73.92, 5′-AGCGGCGGGCTGAAGGACTC-3′; C723, 5′-CCCAGGTTAGAACGGTCAGCAGAA-3′; E600, 5′-GAGGGTAGGGTGGTCATTGTGTCA-3′; F473, 5′-CAGCACAACAGGAACATTCAGTGG-3′; G608, 5′-GGGGGATTTTATTGCGGTGAAAGT-3′; RLS-F-A, 5′-CCAGGAAGGGGCTCCCACAGT-3′; RLS-F-B, 5′-CTGTCACGTCTCAGCAATAGA-3′; RLS-F-15, 5′-AAGGACTCCTCAAGTGCCACCAAAG-3′; RLS-F-180, 5′-GGAAGGGGCTCCCACAGT-3′; RLS-F-216, 5′-ACTCCTCAAGTGCCACCAAA-3′; RLS-R-331, 5′-CTGCTTACCTCAAGAGCAGTCT-3′; RLS-R-119, 5′-GCAGTCTATTGCTGAGACGTGACAG-3′.

Reverse Transcription-PCR (RT-PCR).

RT-PCR (Promega, Madison, WI) was used to screen VMM12 and VMM18 mRNA for the expression of a gene coding the RLSNRLLLR sequence. The primer pairs RLS-F-180/RLS-R-331 and RLS-F-216/RLS-R-331 were used to amplify 152 bp and 116 bp fragments, respectively. RT-PCR conditions were 48°C for 45 min; 94°C for 2 min; 35 cycles of 94°C for 30 s, 50°C for 60 s, 68°C for 60 s; and 68°C for 5 min.

For all other PCRs, total RNA was first converted to cDNA by using the SuperScript First-Strand Synthesis System (Invitrogen, Carlsbad, CA). PCR was then performed on 250 ng of cDNA using platinum Taq High Fidelity (Invitrogen). The PCR mixes were heated to 94°C for 2 min, and 30 cycles of amplification were performed, followed by a final extension at 68°C for 5 min. When amplifying the TAG genes, the 30 cycles consisted of 94°C for 30 s, 62°C for 30 s, and 68°C for 60 s. When the GAPDH gene was amplified, the 30 cycles consisted of 94°C for 30 s, 60°C for 30 s, and 68°C for 60 s. The PCR products were visualized on ethidium bromide-stained agarose gels.

DNA Sequencing.

Automated DNA sequencing was performed at the University of Virginia DNA Sequencing Core on either an Applied Biosystems 377 Prism DNA Sequencer or a 3100 Genetic Analyzer, using Big Dye terminator chemistry with TaqDNA polymerase.

Rapid Identification of cDNA Ends (RACE).

The GeneRacer system (Invitrogen) was used to perform both 5′ and 3′ RACE. For the 5′ RACE procedure, the GeneRacer 5′ Primer was used in conjunction with the GSP RLS-R-119 (5′-GCAGTCTATTGCTGAGACGTGACAG-3′). Cycling conditions were: 94°C for 2 min; 5 cycles of 94°C for 30 s, 76°C for 2 min; 5 cycles of 94°C for 30 s, 74°C for 2 min; 5 cycles of 94°C for 30 s, 72°C for 2 min; 15 cycles of 94°C for 30 s, 70°C for 30 s, 72°C for 2 min; 72°C for 5 min).

Nested PCR was used for the 3′ RACE procedure. Outside reactions used the GeneRacer 3′ primer in conjunction with either of the GSP primers RLS-F-A or RLS-F-15. Cycling conditions for the RLS-F-A PCR consisted of 94°C for 2 min; 5 cycles of 94°C for 30 s, 68°C for 2 min; 5 cycles of 94°C for 30 s, 66°C for 2 min; 20 cycles of 94°C for 30 s, 61°C for 30 s, 68°C for 2 min; and 68°C for 10 min. Cycling conditions for the RLS-F-15 PCR consisted of 94°C for 2 min; 5 cycles of 94°C for 30 s, 72°C for 2 min; 5 cycles of 94°C for 30 s, 70°C for 2 min; 20 cycles of 94°C for 30 s, 65°C for 30 s, 68°C for 2 min; and 68°C for 10 min. Inside reactions used the 3′ GeneRacer nested primer with the GSP primer RLS-F-B. Cycling conditions were 94°C for 2 min; 14 cycles of 94°C for 30 s, 76°C (decreasing 0.5°C/cycle) for 2 min; 16 cycles of 94°C for 30 s, 68°C (decreasing 0.5°C/cycle) for 30 s, 68°C for 2 min; and 68°C for 10 min.

The PCR products were visualized on ethidium bromide-stained low-melting agarose gels, and selected bands were purified using the QIAquick (Qiagen) purification system. The purified DNA was cloned into pCR4-TOPO (Invitrogen), transformed into One Shot TOP10 Chemically Competent Escherichia coli (Invitrogen), and selected with 100 μg/ml ampicillin on Luria-Bertani agar. DNA from individual colonies was purified using the Qiagen Plasmid Mini kit.

Human Subjects.

All of the research involving human subjects was approved by the University of Virginia Human Investigation Committee in accordance with an assurance filed with and approved by the Department of Health and Human Services. Informed consent was obtained from all of the study participants.

RESULTS

VMM18 CTL Recognize Three Distinct HLA-A3-Restricted Epitopes.

The peptides bound to HLA-A3 molecules on 8 × 10¹⁰ VMM18 tumor cells were purified as described in “Materials and Methods,” and fractionated by RP-HPLC using TFA as the organic modifier. A CTL epitope reconstitution assay was performed using 2.5% of each RP-HPLC fraction (2 × 10⁹ cell equivalents), and three peaks of activity were observed (Fig. 1) ⇓ . Peak A activity (fractions 15–17) corresponds to the previously identified SQNFPGSQK peptide which is derived from an unidentified gene (18) , and peak B activity (fractions 26–28) corresponds to the previously described ALLAVGATK peptide from Pmel17/gp100 (11) .

Identification of the Antigenic Peptide in Peak C.

Active fraction 38 (Fig. 1 ⇓ , peak C) was further fractionated by RP-HPLC using heptafluorobutyric acid as the organic modifier. In CTL epitope reconstitution assays fractions 66 and 67 of a second fractionation of peak C contained the active peptide (Fig. 2) ⇓ . The peptides were further fractionated by a third round of RP-HPLC, using triethylamine acetate as the organic modifier. In CTL epitope reconstitution assays, peak C antigenic peptide was present primarily in fractions 32–34 (Fig. 3) ⇓ .

Analysis of the active third-dimension peak C fractions showed that the biological activity in epitope reconstitution assays correlated with the abundance of the m/z 571 ion. Analysis of the collision-activated dissociation mass spectra suggested that the peptide sequence was RXSNRXXXR, where “X” is either a leucine or isoleucine residue (these residues are indistinguishable by low-energy collision-activated dissociation). A mixture of sixteen peptides with leucine and isoleucine incorporated at each of four positions in the sequence RXSNRXXXR was, therefore, synthesized, and this peptide cocktail was shown to have potent epitope reconstituting activity (Fig. 4A) ⇓ . Each of the 16 peptides was next individually synthesized and tested in epitope reconstitution assays. A range of activities was observed, with most of the sequences sensitizing C1R-A3 targets for at least some lysis by VMM18 CTLs, and with no one sequence being significantly and reproducibly superior to all of the others (data not shown). Subsequent RP-HPLC coelution studies clearly demonstrated, however, that the unknown m/z 571 in the active fractions was RLSNRLLLR, and the epitope reconstitution assay showed that this peptide is active at concentrations as low as 10 pm (Fig. 4B) ⇓ . Mass spectrometric analysis of RP-HPLC-fractionated peptides eluted from immunoaffinity-purified HLA-A3 molecules from the melanoma cell line VMM12 demonstrated that the RLSNRLLLR peptide was present in the expected fractions (data not shown) and, thus, represents a novel shared melanoma antigen. The abundance of an additional peptide, FLSNRILLR, also correlated with the activity observed in epitope reconstitution assays, but the synthetic peptide did not reconstitute the epitope recognized by VMM18 CTLs (Fig. 4A) ⇓ .

Basic Local Alignment Search Tool (BLAST) Search Results for the Gene(s) Coding for the RLSNRLLLR Peptide.

A homology search of the RLSNRLLLR peptide yielded three exact matches: (a) AE003619, a Drosophila melanogaster genomic scaffold gene; (b) AC106771, Homo sapiens chromosome 5 clone RP11-308B16; and (c) AC106790, Homo sapiens chromosome 5 clone RP11–376E20. The human sequences are overlapping clones, and in both cases, the sequence coding for the RLSNRLLLR peptide is immediately followed by a stop codon, suggesting that the peptide might occur at the COOH-terminal end of a protein expressed from a gene coded for in these two clones. To determine whether such a gene was expressed in VMM18, PCR primers were designed to amplify a region of DNA that would encompass that coding for the RLSNRLLLR peptide, as well as sequence immediately 5′ to that region. Two primer sets (RLS-F-180/RLS-R-331 and RLS-F-216/RLS-R-331) respectively amplified the predicted 152-bp and 116-bp fragments from both VMM12 and VMM18 cDNA (data not shown), thus confirming that a gene encompassing this region was expressed in melanoma cell lines known to express the peptide.

Identification of the Gene Coding for the Source Protein Containing the RLSNRLLLR Peptide.

The GeneRacer method of 5′ prime RACE was chosen as it ensures the amplification of full-length mRNA by directing the ligation of GeneRacer RNA Oligo to mRNA that has not been truncated at the 5′ end. PCR was performed with the GeneRacer 5′ Primer and the 3′ reverse primer, RLS-R-119, that was designed to overlap partially the nucleotide sequence coding for the RLSNRLLLR peptide. An ∼200 bp fragment was obtained, cloned into pCR4-TOPO, and sequenced. A BLAST search of the obtained sequence demonstrated that it was completely homologous to AC106771 and overlapped with the sequence obtained from the 152- and 116-bp fragments. The 5′ end of the insert read directly into the complete GeneRacer RNA Oligo sequence, thus confirming that the complete 5′ end of the gene had been obtained.

3′ RACE was then used to obtain 3′ sequence information for the RLSNRLLLR-coding gene. The two sets of primers yielded two dominant fragments each, and the difference in the size of the fragments between the two primer sets corresponded to the predicted size difference based on the location of the 5′ GSP. The fragments were cloned into pCR4-TOPO and sequenced. A total of four different sequences were obtained for the 3′ end of the gene. The 3′ end of the sequences corresponded to the GeneRacer Oligo(dT) primer, thus indicating that the 3′ primer end of the genes had been obtained.

Gene Structure.

By combining the 5′ and 3′ sequence information, a total of four different isoforms of the gene could be constructed, TAG-1, TAG-2a, TAG-2b, and TAG-2c (Fig. 5) ⇓ . These sequences were further confirmed by sequencing clones obtained after RT-PCR with primers specific for the 5′ and 3′ end of each isoform. The isoforms are composed of three to four exons each, with each having the α1 and α2 exons in common at the 5′ end of the gene. BLAST searches indicate that the genes are coded for on the short arm of Chromosome 5, and have 100% identity to sequences in clones AC106771, AC106790, and AC119151. The seven identified exons span ∼230,000 nucleotides in the genomic sequence (Fig. 6) ⇓ . Appropriate splice sites exist at each of the intron/exon boundaries to allow splicing of the exons to occur. During the course of sequencing clones that corresponded to TAG-2c, an additional isoform (TAG-3) was isolated that lacked the α2 exon and was composed of the α1, α4, and α7 exons. The splicing of the α1 exon to the α4 exon changed the nucleotide sequence such that the COOH-terminal Arg in the RLSNRLLLR peptide was replaced by a Ser (RLSNRLLLS). Although no significant open reading frame initiated from an AUG codon exists within the sequences, there are three nonstandard initiation codons (two CUG and one ACG), all of which are in frame with one another, and all of which initiate an open reading frame that would code for the RLSNRLLLR peptide (Fig. 5) ⇓ . The sequence coding for the RLSNLLLR peptide spans the junction between the first two exons, with the first 26 nucleotides coming from the α1 exon and the 27th nucleotide coming from the α2 exon.

Protein Structure.

Depending on the initiation codon used, the TAG-1 gene potentially codes for a 99-amino-acid (TAG-1α), a 63-amino-acid (TAG-1β), and/or a 59-amino-acid peptide (TAG-1γ), with respective M_r of 10,615, 6,945, and 6,577. Whereas the TAG-2a, TAG-2b, and TAG-3b genes differ from one another in their fourth exon, all of them potentially express identical proteins because the stop codon is located in the third exon. These genes would use the same initiation codons as in the TAG-1 gene, but would differ from the TAG-1 gene at their 3′ end. The expressed proteins would be 93 (TAG-2α), 57 (TAG-2β), and 53 (TAG-2γ) amino acids in length, with M_r of 9,727, 6,057, and 5,689. The TAG-1 protein isoforms, but not the TAG-2 protein isoforms, contain the sequence Asn-Ser-Thr and, thus, could potentially exist in a glycosylated form. The TAG-1 isoforms have three cysteines and TAG-2 isoforms have four cysteines, which could lead to interchain or intrachain disulfide bond formation. A BLAST search of the TAG-1 and TAG-2 protein isoforms does not reveal any significant homology with known proteins.

Expression of the TAG-1, -2a, -2b, and -2c Genes in Melanoma Cell Lines.

PCR specific for each of the TAG-1, -2a, -2b, and -2c genes was performed on cDNA obtained from 32 established melanoma cell lines (Table 1) ⇓ . In most samples, TAG-1 and -2a were expressed at the highest levels, TAG-2b was poorly expressed, and TAG-2c was expressed at an intermediate level (Fig. 7) ⇓ . Of the cell lines tested, only four were negative for the expression of all four isoforms, even after 40 cycles of PCR amplification. With the exception of EB81-Mel, each tumor line expressed all four genes or none at all. Overall, TAG-1, 2a, and -2b are expressed in 88% of the melanoma cell lines tested, and TAG-2c is expressed in 84% of the melanoma cell lines tested. To ensure that the expression of the TAG gene family was not an artifact of in vitro culture conditions, mRNA was prepared from a cryopreserved aliquot of the original tumor sample from which the VMM12 tumor line was established. RT-PCR was positive for each of the TAG genes, thus establishing that TAG is expressed in uncultured melanoma cells (data not shown).

Expression of the TAG-1, -2a, -2b, and -2c Genes in Transformed and Malignant Leukocyte-Derived Cell Lines.

RT-PCR of mRNA derived from the multiple B-lymphoblastoid cell line and from the hybrid T-B-lymphoblastoid cell line, T2-A3, demonstrates that the TAG gene family is not expressed in transformed B or T cells (data not shown). All four TAG isoforms were, however, expressed in K562, a myelogenous leukemia cell line (data not shown).

Expression of the TAG-1, -2a, -2b, and -2c Genes in Normal Tissue.

The expression of the TAG family of genes was determined in mRNA derived from normal tissues (Table 2) ⇓ . The results demonstrated that with the exception of the testis and placenta, the TAG genes are not expressed in normal tissue. The expression of TAG-1 can be seen in the placenta after 30 cycles of amplification, and TAG-2a is weakly detectable. On 40 cycles of amplification, TAG-1, -2a, and -2b are easily detected in placenta, but TAG-3b is not visualized. TAG-1 and -2a expression is readily observed in the testis after 30 cycles of amplification, and all four genes are detectable after 40 cycles of amplification. The expression of the TAG genes in testis and placenta but not in other normal tissues indicates that they share expression profiles with other cancer/testis antigens.

BLAST Search of the TAG Genes in the Human Expressed Sequence Tags (EST) Database.

A BLAST search of the TAG genes against the GenBank Human EST database yielded homology with two chronic myelogenous leukemia sequences and one hepatocellular carcinoma sequence. Chronic myelogenous leukemia clone, BF210037, has 96% identity of a 603-bp sequence with the α1, α2, and α3 exons of TAG-1, whereas chronic myelogenous leukemia clone, BF240333, has a 191-bp region among 741 bp that is 94% identical to the α1 exon of TAG-1, -2a, -2b, and -2c through the first 18 nucleotides coding the RLSNRLLLR peptide, after which the sequences diverge. The hepatocellular carcinoma clone, AV695059, has a 272-bp region among 421 bp that is 98% identical with the TAG α1 exon and all but the last 3 bp of the α2 domain, after which the sequence diverges. These results demonstrate that the TAG genes may be expressed in a variety of tumors, and that there may be additional isoforms that we have not yet identified.

DISCUSSION

A novel gene with multiple isoforms coding for the shared cancer/testis antigens TAG-1 and TAG-2 has been identified. TAG-1 and TAG-2 differ from previously identified cancer/testis antigens in three clinically important aspects. First, the genes are expressed in 84–88% of 32 melanoma cell lines tested, whereas most other cancer/testis genes are generally expressed in about 20–50% of melanoma samples (19 , 20) . The high frequency of expression of the TAG genes directly translates to a high percentage of melanoma patients for whom a TAG-based vaccine may benefit. Second, the TAG-derived peptide antigen, RLSNRLLLR, is naturally immunogenic because the RLSNRLLLR-specific CTLs used were obtained from a melanoma patient with spontaneous reactivity to this peptide (21) . This contrasts with the fact that CTL-recognized epitopes have not been identified from some cancer/testis antigens, and those that have been identified are frequently poorly immunogenic (20 , 22) . Third, the TAG genes are expressed in the cell line K562, a myelogenous leukemia, and they are homologous with chronic myelogenous leukemia-derived clones in the human EST database. With the exception of the PRAME, SART-3, and SPAN-Xb antigens, cancer/testis antigens are rarely expressed in leukemia and myeloma cells (22 , 23) . The TAG genes are also homologous with a hepatocellular carcinoma-derived clone in the human EST database. Thus, TAG-derived antigens may be useful in the treatment of cancers other than melanoma and, particularly, in cancers for which few antigens have been described.

TAG-1, -2a, -2b, and -2c genes differ from other genes that code cancer/testis antigens in that they are the first to be localized to chromosome 5, whereas the majority of cancer/testis antigens localize to the X chromosome (20 , 22 , 24) . The TAG genes appear to be authentic in that they are capped at the 5′ end, have an initiation codon (albeit nonstandard), exon/intron splicing, a stop codon, and a polyA tail. Initiation codons other than AUG have previously been described in eukaryotes (25) , and two such codons (CUG is present twice, ACG is present once) are located upstream and in-frame with the sequence coding for the RLSNRLLLR peptide. Like the AUG initiation codon, nonstandard initiation codons are almost always found in concert with purines (A or G) at position −3 and/or +4 relative to the initiation site (25) . The second in-frame CUG codon is accompanied by a G at position −3, thus, suggesting that it might be the authentic start site for the protein encompassing the RLSNRLLLR peptide.

Whereas most epitopes recognized by CTLs are derived from proteins coded for by mRNA with primary open reading frames initiated with an AUG codon, a number of epitopes have been described that arise from nontraditional transcription (26 , 27) . These include peptides that are coded for in alternative open reading frames, the 5′ untranslated region, exon-intron junctions, introns, frame-shifted regions, and from proteins with non-AUG initiation. Like the TAG genes, epitopes from the α-tubulin mRNA (28 , 29) , intestinal carboxyl esterase mRNA (30) , and HB-1 mRNA (31) , all arise from proteins coded for in an open reading frame initiated with a nonstandard initiation codon. In common with HB-1, TAG appears to be only the second description of an antigen derived from a non-AUG initiated translation product that is the primary open reading frame, and not secondary to another open reading frame initiated by AUG.

Although epitopes have previously been described that were derived from an intron-exon junction in an incompletely spliced mRNA (32) , and from the introns of N-acetylglucosaminyltransferase V (33) , gp100 (34) , and tyrosinase-related protein-2 (35) , this appears to be the first report of an epitope derived from a mRNA sequence that spans two exons. That the peptide coding sequence should span two exons is not of great biological significance, although, practically, it points out the potential difficulty of identifying epitope coding sequences directly from genomic sequences.

In summary, the TAG cancer/testis antigens described here represent shared, naturally immunogenic antigens that are expressed on a high percentage of human melanoma cell lines. Thus, they fulfill the requirements of a good vaccine candidate for the treatment of melanoma: (a) they are expressed on tumor cells but not on healthy tissue that can be recognized by CTLs; (b) they are expressed by a high percentage of melanoma cell lines from multiple different individuals indicating that they can be used in a vaccine to treat a population of individuals rather than a specific individual; and (c) they are naturally immunogenic and would likely stimulate a strong CTL response in vaccine recipients. The evidence also suggest that the antigens may be expressed in myelogenous leukemia and hepatocellular carcinoma and, thus, may be useful in the treatment of cancers beyond melanoma.