Identifying differential gene expression in monoterpene-treated mammary carcinomas using subtractive display.

Monoterpene-induced/repressed genes were identified in regressing rat mammary carcinomas treated with dietary limonene using a newly developed method termed subtractive display. The subtractive display screen identified 42 monoterpene-induced genes comprising 9 known genes and 33 unidentified genes, as well as 58 monoterpene-repressed genes comprising 1 known gene and 57 unidentified genes. Several of the identified differentially expressed genes are involved in the mitoinhibitory transforming growth factor β signal tranduction pathway, as demonstrated by isolation of the mannose 6-phosphate/insulin-like growth factor II receptor and the transforming growth factor β type II receptor. The monoterpene-induced/repressed genes indicate that apoptosis and differentiation act in concert to effect carcinoma regression. Apoptosis is suggested by the cloning of a marker of programmed cell death, lipocortin 1. Consistent with a differentiation/remodeling process occurring during tumor regression, subtractive display identified YWK-II and neuroligin 1. Thus far, of the cDNAs putatively identified as differentially expressed in this complex in situ carcinoma model, 5 were tested, and each one has been confirmed to be differentially expressed. Additionally, many of the identified known genes are expressed as rare transcripts and exhibit small but significant changes in abundance. Together, these points demonstrate the unique utility of this new gene expression screen to identify altered gene expression in a complex in vivo environment.

The monoterpenes limonene and perillyl alcohol, an emerging class of naturally occurring anti-cancer compounds, are highly effective against a variety of rodent organ-specific cancer models (reviewed in Refs. 1 and 2). Dietary administration of monoterpenes is effective for chemoprevention and chemotherapy of both 7,12-dimethylbenz[a]anthracene (DMBA) 1 -in-duced and N-methyl-N-nitrosourea-induced rat mammary carcinomas (3)(4)(5)(6)(7)(8)(9)(10). As a chemotherapeutic agent, dietary 10% limonene caused ϳ66% of advanced DMBA-or N-methyl-Nnitrosourea-induced carcinomas to completely regress and an additional ϳ23% to partially regress (6). Moreover, a 2% perillyl alcohol diet resulted in 50% complete and an additional 25% partial regression of advanced DMBA-induced mammary carcinomas (7). Treatment with both monoterpenes at these anti-cancer doses did not cause systemic toxicity. Monoterpenes are currently being tested in Phase I clinical trials on advanced cancer patients in the United Kingdom (11) and the United States. 2 Several observations suggest that monoterpene-mediated mammary carcinoma regression may involve a differentiation/ remodeling process. Histopathology of monoterpene-treated, actively regressing tumors displays regions of dense anaplastic epithelium characteristic of mammary carcinomas and regions of a remodeled epithelial compartment. Regressing carcinomas do not show increased levels of lymphocyte infiltration, inflammation, or necrosis (6). Furthermore, monoterpene-treatment of neuro2A neuroblastoma cells causes morphologic differentiation within 4 h as characterized by neurite outgrowths (12).
Monoterpenes inhibit enzymes in the mevalonate-lipid metabolism pathway, including a selective inhibition of isoprenylation of 21-26 kDa small G proteins (13)(14)(15) and inhibition of ubiquinone and cholesterol synthesis (16). Inhibition of isoprenylation may affect signal transduction pathways because unprocessed or non-prenylated small G proteins are not properly localized within the cell and are thus non-functional (17). Furthermore, investigations into gene expression and protein level alterations associated with monoterpene-mediated tumor regression revealed a large increase of the mitoinhibitory transforming growth factor ␤1 (TGF-␤1) and the mannose 6-phosphate/insulin-like growth factor II receptor (Man-6-P/IGF2R), which both facilitates latent-TGF-␤1 activation and degrades the mammary mitogen insulin-like growth factor II (IGF2; Ref. 18). Interestingly, while Man-6-P/IGF2R RNA expression was significantly induced 2-fold in 10% limonene-treated carcinomas, the induced RNA expression of Man-6-P/IGF2R was not observed in the non-responsive 10% limonene-treated carcinomas (19).
We hypothesized that biochemical events in the cytoplasm modify signal transduction leading to altered gene expression. Thus, in order to further elucidate the monoterpenes' mechanism of cancer chemoprevention and chemotherapy, we screened for differentially expressed genes in 10% limonenetreated, DMBA-induced, advanced mammary carcinomas resected at mid-regression.
Complex tissues, such as mammary carcinomas, are composed of large heterogeneous cell populations and hence require sensitive gene expression screening methods. This is because a particular gene may be induced or repressed only in a specific cell type, which would be diluted by all the other cell types. In order to achieve the required level of sensitivity, the subtraction method of Wang and Brown (20) was initially chosen because this approach uses polymerase chain reaction (PCR) amplification to generate a renewable source of cDNA for multiple rounds of subtraction, and the method efficiently removes commonly expressed cDNA from the experimental and control cDNA pools to allow for enrichment of differentially expressed genes. However, cloning the differentially expressed cDNAs involves cycles of classical probe hybridization experiments to isolate only a few clones at a time, followed by subtraction of the newly cloned cDNAs from the library. The cloning cycle is repeated until the library is completely screened, which may take 20 -30 or more cloning cycles and thus renders this methodology inefficient.
A possible alternative method considered was Liang and Pardee's (21) differential display, since it is also a PCR-based technique that selectively amplifies subsets of cDNAs based on primer design applied to two matched but non-subtracted cDNA populations. This technique clones genes quickly but only ϳ10% of the genes are differentially expressed when tested (22)(23)(24). Problems with reproducibility are probably due to the low stringency in annealing of primers, although recent modifications in primer design and experimental protocol have somewhat improved reproducibility (25)(26)(27)(28). In addition, because the primers effectively anneal as 6-or 7-base oligomers of arbitrary sequence, a statistical argument is needed to determine when the libraries are sufficiently screened (21). Furthermore, there exists a high level of background noise in differential display due to the vast majority of competing heterologous RNA. Differential display shows a strong bias toward identifying abundant mRNAs (29). An efficient gene expression screen should identify rare transcripts, which account for ϳ90% of mRNA species in most cells (29).
In order to more efficiently identify differentially expressed monoterpene target genes, we developed and characterized an alternative approach, termed subtractive display (SD), which incorporates the strengths and minimizes the weaknesses of subtractive hybridization and differential display. We have applied SD to identify differentially expressed genes in limonene-treated advanced rat mammary carcinomas.

EXPERIMENTAL PROCEDURES
Summary of SD Methodology-A flowchart of the SD methodology is given in Fig. 1. cDNA was generated from two matched tissue sources, or for this set of experiments, monoterpene-treated, regressing, mammary carcinomas (ϩcDNA) and control, non-treated carcinomas (ϪcDNA). ϮcDNA was fragmented with restriction endonucleases followed by ligation of symmetrical linkers. The sequence at the cDNA fragment ends reflects the methods used to generate the cDNA libraries and restriction endonuclease recognition sites, and hence, are nonrandom sequences. The subtractive process involves multiple rounds of hybridization and PCR amplification, which renews the cDNA source. ϩcDNA was subtracted from ϪcDNA, and concurrently, ϪcDNA was subtracted from ϩcDNA, to derive induced and repressed genes, respectively. Subtracted cDNA was subcloned into a trimming plasmid and subjected to three cycles of precise, unidirectional deletions. The trimming procedure serves to reconfigure the subtracted libraries such that one end of the cDNA is composed of randomized sequence and both ends of the cDNA fragment are flanked by different or asymmetric primers. The subtracted, reconfigured cDNA libraries were PCR-displayed using 16 sets of primers consisting of 1 fixed primer and 1 of 16 subpopulation primers. The subpopulation primers are designed to anneal with a 2-base overlap (3Ј end of primer) of the randomized sequence end of the reconfigured cDNA. The penultimate 2-bases of the 3Ј end of the subpopulation primers were a specific combination of 1 of 16 (2 4 ) possible 2-base combinations, thus selecting for PCR amplification of a subpopulation of subtracted, reconfigured cDNA. Individual cDNA bands were isolated and characterized to confirm differential expression.
Preparation of cDNA from Mammary Carcinomas-Regressing (monoterpene-treated) and non-regressing (control, not treated) carcinomas for all studies were generated by the following protocol. Virgin female Wistar-Furth inbred rats at 50 -55 days of age were given a single dose of DMBA (50 mg/kg body weight) by gastric intubation. Developing carcinomas were followed by palpation until they grew to Ն10 mm in diameter, at which time rats were randomized to either a monoterpene diet of 10% (w/w) d-limonene or a control diet and pair-fed. Carcinomas from rats on monoterpene diet that regressed to 50% of their maximum diameter were resected. All remaining carcinomas were collected by 15 weeks after diet randomization. All animal use was in compliance with NIH guidelines for humane care and was approved by the University of Wisconsin-Madison Medical Center Animal Use Committee. Total tumor RNA was isolated using RNAzol B Reagent (Tel-Test, Friendswood, TX) and poly(A) ϩ RNA was isolated using the PolyATtract System 1000 (Promega, Madison, WI) following the manufacturer's directions. The Superscript Plasmid System (Life Technologies, Inc., Gaithersburg, MD) was used for double-stranded cDNA synthesis from poly(A) ϩ RNA. 5 g of cDNA synthesized from 7 regressing and 7 non-regressing carcinomas were pooled and termed ϩcDNA and ϪcDNA, respectively. A portion of the ϩcDNA and ϪcDNA was reserved for construction of non-fragmented cDNA plasmid libraries with the vector pSport1 and transformed by electroporation into Electromax Efficiency DH12S Escherichia coli host cells (Life Technologies, Inc.).
Subtraction-We used the PCR-based subtractive hybridization method of Wang and Brown (20) described in detail elsewhere. In brief, the subtraction protocol operates by fragmenting the cDNA with restriction endonucleases (i.e. AluI and AluI plus RsaI) and ligating onto their termini 21-bp oligodeoxynucleotide linkers having a 5Ј blunt end and a 4-base 3Ј overhang. The flush end contained an EcoRI site. Following linker ligation, the fragmented cDNAs are subjected to PCR amplification using the 21-bp linkers as primers (sequences given in Ref. 20). It is assumed that cDNA fragments common to both ϩcDNA and ϪcDNA populations were amplified equally, while different cDNA fragments within a population may be differentially amplified. Dependent on the direction of subtraction, whether up-or down-regulated cDNAs were being enriched, either ϩcDNA or ϪcDNA and their derived populations were tracer or driver cDNA. Thus in the following, driver or tracer cDNA is specified but not if ϩcDNA or ϪcDNA was used. 50 g of driver cDNA was completely digested with EcoRI to suppress contaminating driver cDNA from being amplified. Driver cDNA was then photobiotinylated and combined with 2.5 g of non-biotinylated tracer cDNA. The mixture was completely denatured by boiling and then cooled for a long hybridization (20 h). Addition of streptavidin led to formation of streptavidin-biotin-DNA complexes that were removed by several phenol-chloroform extractions. Subtracted tracer cDNA (termed 1cDNA) was again subtracted from 25 g of biotinylated driver cDNA, this time for a short hybridization of 2 h to yield 2cDNA. ϩ2cDNA and Ϫ2cDNA were PCR amplified. 2.5 g of non-biotinylated tracer 2cDNA was subtracted against 50 g of EcoRI treated-biotinylated driver 2cDNA for a long hybridization followed by 25 g of driver 1cDNA for a short hybridization, resulting in 4cDNA, which was then PCR amplified. Another similar cycle was performed giving 6cDNA.
The subtractive hybridization process was monitored by displaying the cDNA libraries after each round of subtraction on a sequencing gel. The cDNA libraries were amplified using a primer derived from the original linker sequence ligated onto the cDNA fragment ends but modified by deletion of the two 5Ј bases and the addition of the 2 mixed bases A/T and G/C on the 3Ј end that overlaps the cDNA insert (5Ј-CTTGCTTGAATTCGGACTA(A/T)(C/G)-3Ј), giving 4 possible primers and 16 possible primer pairs. This mixed primer anneals at both ends of the cDNA fragments and because of the two mixed bases at the 3Ј end that overlap the cDNA insert, the sequence selectively amplifies a representative cDNA subpopulation such that individual bands are visualized (Fig. 2). The PCR temperature profile used was 94°C (5 min), hot start followed by 94°C (1 min), 50°C (1 min), 72°C (2 min) for 30 cycles. The PCR products were resolved on a 5% polyacrylamide gel and visualized by exposure to X-Omat AR film (Kodak, Rochester, NY).
Trimming Cycles-The subtracted cDNA libraries were next reconfigured such that unique primer sites flank the cDNA insert. This was accomplished by cloning the libraries into the trimming plasmid pTRIM14 using the EcoRI site within the libraries' linkers and the EcoRI site located 2 bp downstream of pTRIM14's trimming cassette (30). This procedure resulted in deletion of 9 bp of the cDNA libraries' linker ends, allowing 12 bp to remain. Precise unidirectional deletions were then generated on one end of the subtracted cDNA libraries by three trimming cycles of 14 bp each. We followed the trimming cycle protocol as detailed elsewhere (30). In brief, pTRIM14's trimming cas-sette contains BseRI and BsgI recognition sites configured such that stepwise digestion first with BsgI and then BseRI produces cleavage of DNA 16/14 nucleotides distal and 4/2 nucleotides proximal, respectively, relative to the 3Ј end of the cassette, thus conserving the trimming cassette but cutting into the 5Ј end of the cDNA fragment. The trimming cycle continues by treatment with mung bean nuclease to remove single-stranded ends of the linearized DNA and finally recircularization of the plasmids with T4 DNA ligase. Noteworthy, cDNA fragments that may contain BsgI or BseRI are not lost from the library since a portion of the cDNA fragment is retained. Additionally, this trimming procedure was shown to produce deletions with very high efficiency on a population of plasmids (30).
PCR-All PCR reactions were carried out using the following conditions unless otherwise stated. In a final reaction volume of 50 l, PCR was performed using 200 M of each dNTP, 400 nM of each primer, and 2.5 units of AmpliTaq DNA polymerase (Perkin Elmer, Foster City, CA). The 1 ϫ PCR buffer was composed of 50 mM KCl, 10 mM Tris-HCl (pH 8.3), 1.5 mM MgCl 2 , and 0.001% (w/v) gelatin. All primers were synthesized at the Wisconsin Biotechnology Center (Madison, WI) unless otherwise stated.
Subtractive Display-The subtracted and reconfigured cDNA libraries were displayed using sets of a fixed primer (5Ј-ATTACGAATTCG-GACTA-3Ј) and 1 of 16 possible subpopulation primers (5Ј-GAGGAG-GTGCAGTANN-3Ј), as determined by the specific combination of the two variable penultimate 3Ј bases designated by NN ( Figs. 1 and 3). An explanation of these primer sequences is given under "Results." The nucleotide concentration for PCR used was 20 M of each dNTP plus 10 Ci of [␣-35 S]dATP, and the PCR temperature profile used was 94°C (5 min) hot start followed with 94°C (1 min), 59°C (1 min), 72°C (1 min) for 30 cycles and lastly 72°C (10 min) to ensure double-stranded cDNA. Labeled PCR products were resolved on a 5% polyacrylamide sequencing gel and visualized by a PhosphorImager (Molecular Dynamics, Sunnyvale, CA). These experiments were also performed in parallel by omitting the radiolabel and instead silver staining the gel (31) to visualize the bands directly, which greatly improves efficiency of band isolation. Unique bands were excised, eluted from the gel, and reamplified using the same fixed PCR primer and a modified subpopulation primer, or pan primer, that amplifies the entire cDNA population. The pan primer has two C bases and an EcoRI site (underlined) added to the 5Ј end and the two terminal variable 3Ј bases removed from the subpopulation primer (5Ј-CCGAATTCGAGGAGGTGCAGTA-3Ј). Isolated cDNA fragments were digested with EcoRI, subcloned into the multiple cloning site of the pSport1 plasmid (Life Technologies, Inc.), and transformed by electroporation into Electromax Efficiency DH12S E. coli host cells (Life Technologies, Inc.).
Sequence Analysis-All subcloned cDNA fragments were sequenced using the M13/pUC forward primer and the PRISM Ready Reaction Dye Deoxy Terminator Cycle Sequencing Kit (Perkin Elmer), following the manufacturer's protocol. The sequencing reaction products were resolved on an ABI PRISM Automated DNA Sequencer (Advanced Biotechnologies Inc., Columbia, MD). The monoterpene-induced gene (MIG) and monoterpene-repressed gene (MRG) sequences were compared to a non-redundant nucleotide sequence data base that includes sequences from the Brookhaven Protein Data Bank, GenBank, Gen-Bank updates, EMBL, and EMBL updates using the BLAST algorithm at the National Center for Biotechnology Information (NCBI) (32).
Competitive Reverse Transcriptase-PCR (RT-PCR)-Relative RNA expression levels between regressing (10% limonene-treated) and nonregressing (control, not treated) mammary carcinomas were determined by competitive RT-PCR (33)(34)(35)(36)(37)(38). The competitive RT-PCR assay involved PCR of a target gene (i.e. a MIG/MRG cDNA) competing for amplification with its exogenously added MIMIC, and amplification of an internal standard (i.e. glyceraldehyde 3-phosphate dehydrogenase (GAPDH)) competing for amplification with its exogenously added MIMIC, for a total of 4 PCR products per sample. A MIMIC is a piece of heterologous DNA with flanking sequences identical to the primer sequences used for the target gene, such that a known amount of MIMIC is exogenously added, which competes for simultaneous coamplification with a target gene. In essence, the competitive RT-PCR assay normalizes the MIG/MRG cDNA to an external control (MIG/ MRG MIMIC) and to an internal GAPDH control, which itself was normalized to an external control (GAPDH MIMIC). After the target MIG/MRG expression for each sample was normalized, its expression level in each sample of the control and regressing carcinoma groups was averaged and compared in order to determine fold-induction or foldrepression. Statistical significance was calculated using the two-tailed Student's t test.
Relative RNA expression levels for lipocortin 1 (LC1), transforming growth factor ␤ type II receptor (TGF-␤IIR), and neuroligin 1 were determined by competitive RT-PCR. RNA was prepared from five individual regressing (10% limonene-treated) and five individual non-regressing (control, not treated) mammary carcinomas as described above. cDNA was generated using the 1st-Strand cDNA Synthesis Kit (Clontech Laboratories, Palo Alto, CA). Primers were designed using Oligo 5.0 Primer Analysis software (National Biosciences, Plymouth, MN), with the exception of the GAPDH primers (Clontech Laboratories). MIMICs were generated using the PCR MIMIC Construction kit (Clontech Laboratories). Generation of MIMICs involves a 1°PCR reaction using a composite primer, consisting of target gene or genespecific primer sequence at the 5Ј end and heterologous DNA primer sequence on the 3Ј end. Next, an aliquot of the 1°PCR reaction was used as template in a 2°PCR reaction using only target gene or gene-specific primers. Primers and dNTPs were removed by passing the 2°PCR reaction over a Chromaspin-100 column (Clontech Laboratories) and the MIMIC was quantified by spectrophotometric analysis. Primer sequences and product sizes are given in Table I

PCR primers and product sizes
Primer sequences for competitive RT-PCR and respective product sizes are shown. In order to achieve sufficient agarose gel resolution of all 4 PCR products for each competitive RT-PCR experiment, GAPDH MIMIC 1 was used in the LC1 competitive RT-PCR reactions with GAPDH MIMIC 2 was used in the TGF␤IIR and neuroligin 1 competitive RT-PCR reactions. MIMIC sequences reflect only the 3Ј end of the composite primer used in the 1°PCR reactions during MIMIC construction to amplify from the heterologous DNA, and the 5Ј end of the composite primers are given by each respective target gene primer sequence. profile used for all competitive RT-PCR experiments was 94°C (5 min) hot start followed by 94°C (1 min), 60°C (1 min), 72°C (2 min) for 30 cycles, with the exception of LC1, which required an annealing temperature of 58°C. PCR products were resolved on a 2% NuSieve GTGagarose gel (FMC BioProducts, Rockland, ME). Primers were 5Ј endlabeled with fluorescein during their synthesis, thereby allowing PCR product quantitation using a FluorImager (Molecular Dynamics).

RESULTS
An overview of the SD methodology is presented in Fig. 1. An evaluation of each aspect of this technology is presented below.
Subtraction of cDNA Libraries-The subtractive hybridization procedure developed by Wang and Brown (20) served to remove commonly expressed cDNA fragments and enrich those that are differentially expressed. The subtraction process involved initial fragmentation of the cDNA libraries with restriction endonucleases and ligation of 21-bp oligodeoxynucleotide linkers onto the cDNA fragment ends, thus allowing for a renewable source of cDNA by PCR. The monoterpene-treated derived cDNA (ϩcDNA) was subtracted against control, nontreated derived cDNA (ϪcDNA) to enrich for MIG. The inverse subtraction, control, non-treated cDNA subtracted against monoterpene-treated cDNA, was also performed to enrich for MRG. The cDNA libraries were subjected to cycles of alternating long (20 h) and short (2 h) hybridizations for a total of 6 hybridizations. We monitored the subtractive process by PCR amplification of the cDNA libraries after each round of subtrac-tion, before the libraries were reconfigured using pTRIM14, and displayed them on a sequencing gel. A representative cDNA subpopulation was PCR amplified by modifying the 21-bp primer, which anneals to the linkers initially ligated onto the fragment ends, such that the two 5Ј bases were removed and two mixed bases were added to the 3Ј end, thus overlaping the cDNA insert, as described in detail under "Experimental Procedures." Fig. 2 demonstrates the efficiency of the subtractions. The unsubtracted ϩcDNA and ϪcDNA lanes showed smears, while with each additional cycle of subtraction, individual bands became increasingly prominent. Furthermore, the banding patterns between the subtracted ϩcDNA and the subtracted ϪcDNA lanes were reproducible, indicating that specific cDNA fragments were increasingly enriched while others were driven out of the population. Efficiency of subtraction was further tested by isolating 2 prominent bands from both the ϩ6cDNA and Ϫ6cDNA lanes, followed by subcloning and sequencing the cDNAs by similar methods of clone characterization discussed below. One of the derived ϩ6cDNA fragments was identified by a nucleotide data base search using the BLAST algorithm (32) as the YWK-II gene, which was again identified at the end of the SD process (indicated by an arrow; compare Figs. 2 and 5) and shown to be differentially expressed as discussed below.

Reconfiguration and Display of the Subtracted Libraries-
The subtracted cDNA libraries were next reconfigured for the display step. The libraries were reconfigured because during their construction both ends of the cDNA fragments were ligated to the same linker, i.e. the ends are symmetrical. Symmetrical primer sites limit the ability to display the subtracted cDNA libraries because a primer designed to overlap the cDNA insert by 2 bases will anneal on both ends of the cDNA fragment, thus not permitting all 2-base independent combinations at both cDNA insert ends for complete screening of the libraries. In addition, the first few bases at both ends of the cDNA fragments in the subtracted libraries are not random but reflect both the methods used to make the cDNA (priming the poly(A) tail with a sequence containing a NotI site) and that used to fragment the cDNA (restriction digestion with two 4-base cutters). One end of each cDNA fragment needs to be a random sequence, such that a subpopulation primer will selectively amplify a small number of distinct bands during the PCR-display step, as discussed below. Reconfiguration of the subtracted cDNA libraries was accomplished by cloning the subtracted libraries into pTRIM14 (30) and processing the cDNA libraries through 3 cycles of precise, unidirectional, 14 bp trimming, as discussed under "Experimental Procedures." pTRIM14 was constructed with a trimming cassette composed of BsgI and BseRI recognition sites. A trimming cycle operates by sequential digestion with BsgI and then BseRI to produce cuts 16/14 nucleotides distal and 4/2 nucleotides proximal, respectively, relative to the 3Ј end of the trimming cassette, thereby producing a deletion onto the 5Ј end of the cDNA fragment but retaining the trimming cassette. Single-stranded DNA ends are removed by digestion with mung bean nuclease followed by ligation with T4 DNA ligase. The subtracted cDNA libraries were subjected to three trimming cycles, generating ϩ6⌬3 and Ϫ6⌬3 cDNA (6 rounds of subtraction and 3 cycles of trimming). Accounting for subcloning the subtracted libraries into pTRIM14 (deletion of 9 bp) and 3 trimming cycles, the entire 21-bp linker plus an additional 30 bp on the left end of the cDNA libraries was completely removed, leaving one cDNA end with randomized sequence, as described in greater detail under "Experimental Procedures." The reconfigured, subtracted cDNA libraries were displayed by consecutive amplification of cDNA subpopulations using 16 sets of PCR primers (Figs. 3 and 5). In each primer set, one primer is always the same or fixed and anneals at the unmodified or right (3Ј) end of the cDNA fragment. This fixed primer is a 17-mer with a sequence of 5Ј-ATTACGAATTCGGACTA-3Ј, and comprised in the 5Ј-3Ј direction of pTRIM14 sequence, an EcoRI site (underlined), and original linker sequence. The second primer in each primer set anneals on the randomized or left (5Ј) end of the cDNA fragment and is termed the subpopulation primer. The 5Ј end of the subpopulation primer anneals to 14 bp of pTRIM14 sequence, and its 3Ј end overlaps the cDNA fragment by 2 bp. These two 3Ј bases in each individual subpopulation primer will consist of 1 of the 16 possible 2-base combinations, resulting in a subpopulation primer sequence of 5Ј-GAGGAGGTGCAGTANN-3Ј (Fig. 2). Therefore, 1 of a total of 16 sets of PCR primers results in selective annealing and amplification of a subpopulation of the cDNA, and consecutive amplification with each of the 16 sets of primers allows complete screening of the libraries.
In order to maximize subpopulation selective amplification and minimize redundancy of clone isolation, the dNTP concentration and annealing temperature of the SD reactions were optimized. Optimization of dNTP concentration and annealing temperature increases the number of unique bands in one subpopulation and reduces the number of bands that appear multiple times across various subpopulations due to mispriming of the two terminal 3Ј bases of the subpopulation primer. Initial SD optimization experiments used representative sets of SD primers including the fixed primer and the subpopulation primers CC, AC, AG, and GA, as specified by the variable bases of the subpopulation primer (or the two 3Ј penultimate bases of the sequence 5Ј-GAGGAGGTGCAGTANN-3Ј), at dNTP concentrations of 200, 20, 10, and 2 M, all at an annealing temperature of 59°C (data not shown). The 20 M SD reaction amplified the cDNA fragments efficiently with moderate band redundancy, while the 10 M SD reaction amplified the cDNA fragments with lower efficiency but less band redundancy. SD conditions were further refined using the subpopulation primer sets AC, AG, and GA because of their similar sequence. These three primer sets were used in SD experiments under annealing conditions of 57, 59, 61, and 63°C using 10 and 20 M dNTP (Fig. 4). SD reaction conditions of 20 M dNTP and an annealing temperature of 59°C resulted in reproducibly efficient amplification with relatively low redundancy of bands.
SD reactions were performed with all 16 sets of subpopulation primers using both 10 M dNTP (data not shown) and 20 M dNTP (Fig. 5). Both experiments were performed in parallel, with radiolabel and without radiolabel in order to isolate bands by silver staining (31). Unique bands from the SD experiment using 20 M dNTP were excised, eluted from the gel, reamplified using the same fixed PCR primer and a pan primer that amplifies all cDNA fragments (sequence given under "Experimental Procedures"), and PCR products were subcloned into pSport1. Individual subcloned cDNAs were designated as a MIG or MRG by virtue of the cloning process (i.e. whether the cDNA originated from the monoterpene-treated ϩcDNA or the untreated control ϪcDNA, respectively).
Characterization of Differentially Expressed Clones-We isolated 42 MIG and 58 MRG rat cDNAs (Table II), all of which were sequenced and compared against sequence data bases using the BLAST algorithm for possible identification. A MIG/ MRG clone was defined to be a known gene if it shows Ն95% sequence identity over the entire cDNA fragment with a known FIG. 3. Schematic of a cDNA fragment and primers used in subtractive display. The schematic represents a cDNA fragment processed through 6 rounds of subtractive hybridization and 3 cycles of trimming (Ϯ6⌬3 cDNA). The unique linker sequences at each end of the cDNA fragment are boxed; the bold sequence in the right linker is an EcoRI recognition site. The shaded region of the cDNA insert represents 2 randomized base pairs, after trimming, which overlap the 2 bases (NN) at the 3Ј end of the subpopulation primer. gene from the sequence data base. The cDNA insert sizes ranged from 45 to 312 bp due to the methods used for library generation, modification, and cloning. Considering the MIG cDNAs as one group, 9 were known, of which 3 were identified by more than 1 MIG cDNA. Additionally, 33 MIG cDNAs were not identified in the sequence data base. Considering the MRG cDNAs as one group, 1 was a known gene and 57 MRG cDNAs were not identified in the sequence data base. The identification of genes was limited due to our strict criteria for sequence identity plus the fact that our sequences were from rat whose representation in the data bases is limited.
More than 1 cDNA fragment, varying in molecular weight, identified cytochrome c oxidase subunit II, sperm membrane protein YWK-II (39), and Y89 (5Ј end similar to YWK-II) ( Table  II). The multiple cDNA fragments that identified each of these genes had common sequences but differed in the size of the cDNA fragment; for example, of the 6 cDNAs that identified cytochrome c oxidase subunit II, all the clones shared 70 bp while the largest clone (MIG-27) had an additional 32 bp not found in the smallest clone (MIG-3).
The potential of the SD screen to identify differentially expressed genes was suggested by the isolation of Man-6-P/ IGF2R as MIG-42. We have previously shown that the Man-6-P/IGF2R was induced 2.0-fold (p Յ 0.002) at the RNA level in 10% limonene-treated, regressing mammary carcinomas using an RNase Protection Assay (Table III; Ref. 19).
We next tested the differential expression of other identified MIG/MRG clones. We initially confirmed that YWK-II, a gene identified by 7 separate MIG clones (Table II), was induced 2.9-fold (p Յ 0.00006) in a panel of 10% limonene-treated regressing carcinomas relative to control non-regressing carcinomas (Table III). Expression was quantitated by Northern blot analysis and a PhosphorImager (Molecular Dynamics). YWK-II was also found to be most highly expressed in rat brain, moderately expressed in normal mammary gland, liver, and kidney, and at low levels in the spleen (data not shown).
However, Northern blots and RNase Protection Assay use ϳ1 g of poly(A) ϩ RNA or 10 g of total RNA per sample. In order to conserve regressing tumor RNA, which is quite limited, we used competitive RT-PCR (48 -53) to determine relative fold-induction or fold-repression of the remaining MIG/ MRG cDNAs.
We tested the differential expression of MIG-12, identified as lipocortin 1 (LC1; Ref. 40), using the competitive RT-PCR assay described under "Experimental Procedures" (Fig. 6A). LC1 was induced 2.9-fold (p Յ 0.00003) in 10% limonene-treated regressing carcinomas relative to control non-regressing carcinomas (Fig. 6B, Table III). We then examined LC1 RNA expression in the normal rat mammary gland by the same assay and found that LC1 was induced 3.1-fold (p Ͻ 0.0005) in the involuting mammary gland relative to the virgin mammary gland (data not shown). LC1 has been shown to be a marker for apoptosis in the involuting rat mammary gland (40).
The transforming growth factor ␤ type II receptor (TGF-␤IIR; 41-44) was identified by MIG-33 and was induced 3.1fold (p Ͻ 0.0002) in 10% limonene-treated regressing carcino- were PCR-amplified in the presence of [␣-35 S]dATP using the fixed primer along with the AC, AG, or GA (sequence specifies the 2 bases at the 3Ј end) subpopulation primers, as indicated. Lanes alternate monoterpene-treated regressing (ϩ) and control non-regressing (Ϫ) carcinoma-derived cDNA. Annealing of primers was carried out at 63, 61, 59, and 57°C using dNTP concentrations of either 10 or 20 M, as indicated. Labeled PCR products were resolved on a 5% polyacrylamide sequencing gel and visualized on a PhosphorImager. mas relative to control non-regressing carcinomas (Table III) by the competitive RT-PCR assay.
The SD screen identified neuroligin 1 as the clone MRG-31 (45). The competitive RT-PCR assay demonstrated neuroligin 1 expression to be repressed 8.8-fold in one of five or not detectable in four of five of 10% limonene-treated regressing carcinomas as compared to control carcinomas (Table III). DISCUSSION The Subtractive Display Gene Expression Screen-In order to better understand the mechanism by which monoterpenes mediate tumor regression, we sought to identify differential gene expression patterns of actively regressing, advanced DMBAinduced rat mammary carcinomas treated with limonene relative to control (not treated) carcinomas. Due to the complexity of regressing tumor tissue, we required a very sensitive gene expression screening method that could detect relatively small alterations in gene expression levels and/or genes which may be expressed in low abundance. Because of these concerns, we integrated the strengths of subtractive hybridization and differential display into a new methodology termed SD. Wang and Brown's (20) subtractive hybridization method generates cDNA for subtraction using PCR; however, clones are isolated by classical probe-hybridization techniques that require an undefined number of probing cycles. In contrast, SD cloning of enriched cDNAs is PCR-based and is therefore faster, more efficient, and more sensitive. Unlike Liang and Pardee's (21) differential display where the 5Ј primer preferentially anneals upstream on the cDNA at a low temperature, the SD method uses a 16-base subpopulation primer and a 17-base fixed primer, each annealing at defined left and right linkers, respectively, on the cDNA fragment ends. In addition, annealing is performed at the highest temperature that still permits PCR to proceed (59°C). The combination of larger primers annealing at defined sites and a very high annealing temperature both contribute to high specificity and stringency, leading to high reproducibility between identical reactions. In addition, by primer design, the libraries are completely screened in 16 sets of reactions. Furthermore, because the cDNA libraries are subtracted before the display step, background noise is significantly reduced, thereby increasing reproducibility and detection of rare cDNAs. Moreover, there exists a potential for noise with differential display because differential display depends on precise registry of cDNA fragments in paired lanes such that unique bands are isolated; yet identical cDNA fragments in treated and non-treated populations may both be amplified but differ in size by a few base pairs. Alternatively, with SD, since all cDNA fragments are putatively differentially expressed because of the subtraction step, the precise registry of cDNA fragments in paired gel lanes is not necessary to identify the differentially expressed genes.
The SD screen identified the Man-6-P/IGF2R, which was already known to be induced 2-fold at an RNA level in the epithelium of regressing carcinomas (19). This initially suggested that SD had high sensitivity. Additionally, every MIG/ FIG. 5. Subtractive display using all 16 subpopulation primer sets. cDNA libraries processed through 6 rounds of subtraction and 3 cycles of trimming (6⌬3 subtracted cDNA) were PCR-amplified in the presence of [␣-35 S]dATP using the fixed primer along with each of the 16 subpopulation primers; the subpopulation primer used, specified by the 2-base sequence at the 3Ј end of the primer, is indicated above each lane. Lanes alternate monoterpene-treated regressing (ϩ) and control non-regressing (Ϫ) carcinoma-derived cDNA. Annealing of primers was carried out at 59°C using 20 M dNTP. The arrow indicates one of the bands that identified the YWK-II gene. Labeled PCR products were resolved on a 5% polyacrylamide sequencing gel and visualized on a PhosphorImager. MRG clone tested for differential gene expression exhibited consistent differential expression across a panel of regressing carcinomas relative to non-regressing carcinomas, thus demonstrating the ability of SD to identify a high percentage of true differentially expressed clones. It should be noted that identification of a single gene by multiple clones versus one clone does not indicate the degree of differential expression; YWK-II was identified by 7 separate bands varying in molecular weight but showed a degree of induction similar to Man-6-P/IGF2R, TGF-␤IIR, and LC1, all of which were identified by one clone (Table  II). However, YWK-II may be more abundantly expressed. The multiple isolation of cDNA fragments that identify the same gene may have resulted from restriction digestion of the original cDNA before subtraction with AluI and AluI plus RsaI, and/or cloning artifacts. Furthermore, band intensity does not indicate degree of differential expression, as there was no discernible difference in band intensity which identified neuroligin 1 and bands that identified the other known clones.
The Process of Monoterpene-mediated Mammary Carcinoma Regression-Based upon histopathological analysis of monoterpene-treated regressing mammary carcinomas, the phenomenon of tumor regression was postulated to involve a differentiation/remodeling process (6). The identified MIG/MRG cDNAs are consistent with such a mechanism of regression. The present study is consistent with the hypothesized involvement of the TGF-␤ signaling pathway with tumor regression. The Man-6-P/IGF2R and the TGF-␤IIR were identified as up-regulated. The Man-6-P/IGF2R facilitates latent-TGF-␤1 activation (18,46) and trafficks IGF2, a potent mammary carcinoma mitogen, into lysosomes for degradation (reviewed in Ref. 47). Also the Man-6-P/IGF2R has been reported to be a tumor suppressor gene in both liver and breast cancer (48,49). Interestingly, the SD screen identified the TGF-␤IIR, since it is postulated that breast tumors may not respond to TGF-␤ because the tumors express low levels TGF-␤II receptors (50, reviewed in Ref. 51).
The SD screen isolated LC1, which is a marker for apoptosis during mammary gland involution. At the protein level, LC1 was induced 10-fold only in the alveolar epithelium, but remained at basal levels in ductal epithelium and stroma during involution, as determined by immunohistochemistry (40). Because RNA expression of LC1 in the context of the entire involuting mammary gland was not reported, we determined LC1 RNA expression during involution relative to the virgin mammary gland and found LC1 was induced 3.1-fold.
It should be noted that most overexpression of genes thus far analyzed was in the 2-3-fold range. In some cases this could be an underestimate due to the fact that our analysis methods averaged gene expression over cell types in a very heterogenous tissue. For example, LC1 was 3.1-fold overexpressed in total involuting mammary gland (RNA) but was 10-fold (protein) overexpressed in the alveolar cell lineage and not overexpressed in the ductal cell lineage. In other cases it is likely that a 2-fold increase could be biologically meaningful in vivo. For  6. LC1 (MIG-12) RNA expression using competitive RT-PCR. A, the differential expression of LC1 between a panel of control non-regressing (lanes labeled CON1-CON5) and 10% limonene-treated regressing (lanes labeled LIM1-LIM5) carcinomas was demonstrated by competitive RT-PCR. PCR products corresponding to GAPDH, LC1, GAPDH MIMIC, and LC1 MIMIC, and their respective sizes (in bp) are indicated. PCR primers were 5Ј-labeled with fluorescein; products were resolved on a 2% agarose gel and visualized on a Fluor-Imager. B, LC1 was induced 2.9-fold in 10% limonene-treated regressing carcinomas compared to control non-regressing carcinomas. The LC1 expression was quantified from the gel in panel A by first normalizing the LC1 and GAPDH bands to their respective MIMICs and second by normalizing LC1 to GAPDH. example, we detected a 2.0-fold increase in the Man-6-P/ IGF2R, an imprinted gene in rodents (52), suggesting that a 2-fold difference in its expression has been evolutionarly selected and conserved.
Both the induced YWK-II expression and repressed neuroligin 1 expression could play a role in tumor regression through a differentiation process. YWK-II, a transmembrane protein, is expressed in the mammary gland (data not shown) and has been shown to be a marker for differentiation of spermatogonia (53). It modulates cell-cell adhesion, as in sperm-egg adhesion during fertilization (54). Also, the polypeptide sequences of the YWK-II transmembrane and cytoplasmic domains are 70.6% homologous to the same domains of the human A4 amyloid protein found in brain plaques of Alzheimer disease patients (39). Differential gene expression of the A4 amyloid gene is associated with retinoic acid-induced morphologic differentiation of neurite processes in two model systems; the A4 amyloid mRNA is induced 34-fold in retinoid-treated P19 embryonal carcinoma cells (55) and 10-fold in retinoid-treated SH-SY5Y neuroblastoma cells (56). Neuroligin 1 is a neuronal cell surface protein found enriched in synaptic plasma membranes and binds to brain-specific ␤-neurexins as a ligand. Expression of nueroligin 1 was tested in various tissues, but was restricted to brain tissue (45). Therefore, neuroligin 1 expression in mammary carcinomas may reflect deregulated gene expression. Interestingly as carcinomas regressed, neuroligin 1 expression was repressed or turned off, which is consistent with regulated gene expression in most normal tissue.
The SD screen was applied to monoterpene-treated mammary carcinomas resected at mid-regression. Therefore, the identified MIG/MRG cDNAs reflect differential gene expression patterns consistent with apoptosis and differentiation of actively regressing carcinomas. In order to better understand the process by which monoterpenes first initiate tumor regression, we are currently identifying early MIG/MRG cDNAs using SD on mammary carcinomas treated with monoterpenes for 24 h.
In summary, the observed alterations in gene expression help to better define the process associated with monoterpenemediated tumor regression. Although the data presented correlate monoterpene-treatment, induction of TGF-␤ signaling components, and tumor regression, the data do not show that TGF-␤ signaling is a causal event in monoterpene-mediated tumor regression. It is, however, consistent with our working hypothesis that monoterpenes promote the up-regulation of the Man-6-P/IGF2R, thereby causing increased levels of activated TGF-␤1 available for ligand binding to the up-regulated TGF-␤IIR and initiating mitoinhibitory and apoptotic signaling. We are currently investigating this hypothesis by evaluating whether the Man-6-P/IGF2R up-regulation is the central causal event that triggers tumor regression.