Somatic Acquisition and Signaling of TGFBR1*6A in Cancer FREE

Transforming growth factor β (TGF-β) is a potent naturally occurring inhibitor of cell growth. It exerts its action by binding to type I (TGFBR1) and type II (TGFBR2) transmembrane receptors located on the cell membrane. Intracellular signaling begins once TGF-β has bound to the TGFBR1/TGFBR2 complex. TGFBR2 activates TGFBR1, which acts as the initiator of intracellular responses. Mothers against decapentaplegic homolog 2 (SMAD2) and SMAD3 are subsequently activated by TGFBR1 and form complexes with SMAD4. Activated SMAD complexes enter the nucleus where they regulate the activity of target genes.¹ There is evidence that TGF-β related proteins activate not only SMADs but also other signaling pathways.²

The TGF-β signaling pathway is regulated by other cellular elements and pathways. The activation of the epidermal growth factor receptor,³ interferon γ signaling through signal transducers and activators of transcriptions,⁴ and tumor necrosis factor α through activation of nuclear factor κB (NFKB1 )⁵ inhibit the TGF-β signaling pathway. Other cancer-related pathways that affect TGF-β signaling include the RAS–mitogen-activated protein kinase pathway, which is able to inhibit SMAD signaling.⁶ TGF-β inhibition of epithelial growth is achieved through the induction of expression of cyclin-dependent kinase inhibitor 2B (CDKN ) (p15^INK4B) ^7- 8 and CDKN1A (p21^CIP1 ).⁹ Other mechanisms that lead to cellular growth arrest include the inhibition of MYC expression, cyclin-dependent protein kinase 4 (CDK4) and cell division cycle 25A (CDC25A ).¹⁰ The inhibitory signals of TGF-β can also induce apoptosis in several cell types.^11- 16

Increased cell growth due to decreased TGF-β growth inhibition may contribute to cancer development. Indeed, transgenic mice that lack 1 copy of the Tgfb1 gene (Tgfb1^+/−) and mice that lack 1 copy of the Tgfbr2 receptor gene (Tgfbr2^+/−), conditions that result in decreased TGF-β signaling, have an increased susceptibility to develop cancer.^17- 18 In humans, inactivating mutations in TGFBR2 have been identified in colon and head and neck cancers, deletions of TGFBR1 in pancreatic and biliary carcinomas and in lymphoma, and TGFBR1 tumor–specific mutations in breast and ovarian cancer.^19- 23 Additionally, restoration of functional receptors reverses the malignant behavior of several human cancer cell lines that lack functional TGF-β receptors.^24- 25 Thus, TGF-β function correlates with susceptibility to the development of cancer.

Although TGF-β is a potent growth inhibitor of normal epithelial cells, cancer cells secrete in general larger amounts of TGF-β than their normal counterparts. The association of TGF-β secretion with cancer is strongest in the most advanced stages of tumor progression.¹ In mouse models increased TGF-β signaling is associated with decreased cancer incidence.^26- 28 However, the growth of established tumors in mice is fueled by increased TGFBR1 levels and by increased TGF-β signaling. The role of TGF-β therefore depends on the disease status of the host. The molecular changes that result in the redirection of TGF-β growth inhibitory signals into growth stimulatory signals during cancer development are essentially unknown.

TGFBR1*6A referred to as *6A in this article is a common polymorphism of TGFBR1 that consists of a deletion of 3 alanines within a 9-alanine (*9A) repeat at the 3′-end of exon 1 coding sequence.²⁹ Four primary studies and 2 meta-analyses have shown that *6A is one of the first candidate tumor susceptibility alleles that is found in a large proportion of the general population (13.7%) and significantly increases cancer risk by approximately 24%.^30- 35 Using a mink lung epithelial cell line devoid of endogenous Tgfbr1, *9A, and *6A cell lines were established for functional studies. Compared with *9A, *6A was a less effective mediator of TGF-β–antiproliferative signals.^30,36 However, the molecular mechanism underlying *6A decreased growth inhibition is unknown as the 9–base pair (bp) deletion that differentiates *6A from *9A could be part of the mature receptor or belong to the signal sequence, a part of the receptor that is cleaved off once synthesis is completed. Genotyping of tumor samples has consistently shown a higher *6A allelic frequency, 0.116, 0.139, and 0.150^22,29,37 than in blood cells from 3451 healthy controls (0.071) and 4399 patients with a diagnosis of cancer (0.090).³⁵ We reasoned that if *6A in the germline predisposes to these cancers, it might be even more common in the tumors themselves through somatic acquisition. We genotyped tumor and germline DNA from consecutive patients with a diagnosis of colon, head and neck, or breast cancer enrolled in protocols approved by institutional review boards in the United States and in Spain. We chose head and neck cancer because of the high *6A frequency reported in tumor tissue and the suspicion that a proportion may be due to somatic acquisition.³⁷ We chose colon and breast cancers because of the previously shown association of these 2 tumor types with *6A.^30- 33,35 This is, to our knowledge, the first study investigating (1) molecular differences between *6A and *9A with respect to signal sequence cleavage, (2) *6A signaling in cancer cells, and (3) *6A somatic acquisition in cancer.

All tumor and germline tissue samples were obtained from patients who had signed informed consent for genetic study of their tumor and germline tissues and were enrolled in investigation review board–approved protocols at 4 different institutions between June 1, 1994, and June 30, 2004.

Head and Neck Tumors. Tumor DNA was obtained from fresh or frozen tissue from 226 patients with a diagnosis of head and neck cancer. One hundred twenty-five patients were from Chicago, Ill (Northwestern Memorial Hospital and University of Chicago Hospital), and 101 patients were from Columbus, Ohio (The James Cancer Hospital and Solove Research Institute), 30 of whom had been previously examined for tumor-acquired mutations within the TGFBR1 gene.³⁷ Patient-matched, germline DNA was obtained from peripheral blood lymphocytes in the Chicago-patient cohort or from a distant biopsy site (>2-5 cm from the tumor outer edge) in the Columbus cohort study. Thirty-four patients (15%) had received chemotherapy, radiation therapy, or both prior to tissue sampling.

Colorectal Tumors. Tumor DNA was obtained from 107 consecutive patients from New York, NY (Memorial Sloan-Kettering Cancer Center) with previously untreated stage III colorectal cancer and whose germline DNA had been previously examined for TGFBR1 exon 1 tumor-acquired mutations within the TGFBR1 gene.³⁸ Microsatellites are short stretches of tandem repeats of very simple DNA sequence, usually 1 to 4 bp. Microsatellite instability (MSI) testing was performed according to a defined protocol involving testing paired normal and tumor DNA for MSI with the 5 original National Cancer Institute microsatellite panel: BAT25, BAT26, D2S123, D5S346, and D17S250.³⁹ If 2 or more of the 5 microsatellite sequences in the tumor DNA were mutated, the tumor was termed MSI-high (MSI-H). If only 1 of the 5 microsatellite sequences in the tumor DNA was mutated, the tumor was termed MSI-low (MSI-L). If none of the 5 microsatellite sequences in the tumor DNA were mutated, the tumor was termed microsatellite stable (MSS).⁴⁰ One hundred one tumors were MSI negative (MSS or MSI-L), and 6 (5.6%) were MSI positive (MSI-H).

Two distinct mutational pathways have been identified that result in colorectal cancer. The tumor suppressor pathway, also termed the chromosomal instability pathway, accounts for approximately 85% of all colorectal carcinomas and most sporadic colorectal carcinomas.⁴¹ It is characterized by a high frequency of allelic imbalances such as allelic losses, chromosomal amplifications, and translocations. These genetic changes result in the mutational activation of oncogenes coupled with the mutational inactivation of tumor suppressor genes. The second mutational pathway, the mutator phenotype pathway, accounts for about 15% of all colorectal cancers. It is characterized by the inactivation of both alleles of 1 of the DNA mismatch repair (MMR) genes, which results in variations in the length, and, therefore, instability of short tandem DNA sequences termed microsatellites.⁴² In addition to MSI, this pathway also exhibits secondary mutations in genes relevant to cell growth control and apoptosis. Tumors occurring through the mutator pathway seem to have a better overall prognosis than those occurring through the tumor suppressor pathway.⁴³ To determine whether *6A somatic acquisition is associated with MSI, we obtained additional tumor and germline tissues from patients with MSI-H tumors.

Colorectal (MSI-H) Tumors. Tumor and germline DNA (peripheral lymphocytes) from 30 patients with MSI-H colorectal tumors was obtained from Columbus and 20 patients with MSI-H colorectal cancer from Madrid, Spain. None of the patients had received prior treatment for their tumor.

Colorectal Liver Metastases. Tumor DNA was obtained from 35 patients from Columbus and 9 patients from Chicago (Northwestern Memorial Hospital) with biopsy results verified as colorectal cancer metastatic to the liver. DNA from the primary tumor of 13 of these patients was obtained. Normal adjacent liver was also obtained from 15 of these patients. Treatment information was available for the 9 patients from Chicago.

Breast Tumors. Tumor and germline DNA (peripheral lymphocytes) from 104 white women with sporadic breast cancer who had not received prior cancer-related treatment were obtained from Madrid (Hospital Clínico San Carlos).

TGFBR1 exon 1 was amplified by polymerase chain reaction (PCR) using the Advantage-GC genomic polymerase mix (BD Biosciences Clontech, Palo Alto, Calif) and the following primers: 5′-GAGGCGAGGTTTGCTGGGGTGAGGCA-3′ and 5′-CATGTTTGAGAAAGAGCAGGAGCGAG-3′. *6A somatic acquisition was identified in 3 separate laboratories: Madrid (T.C.), Northwestern (B.P.), and Ohio State University (C.M.W.). Polymerase chain reaction amplification was carried in standard buffer, 2.0 mmol/L of magnesium chloride, 0.25 mmol/L deoxyribonucleotide triphosphates, 25 pmol of each primers, 1 unit of Taq polymerase, and 25 to 50 ng of DNA per 25 μL reaction as recently described.⁴⁴ The ABI Prism 310 Genetic Analyzer (Applied Biosystems, Foster City, Calif) was used for data acquisition. A peak at 115 bp corresponded to the TGFBR1 allele (TGFBR1*9A), whereas a peak at 106 bp corresponded to the TGFBR1*6A variant. The rare equivocal results were confirmed by cloning and sequencing the PCR product.

Gene dosage can influence gene expression,⁴⁵ and for most cancer-related genes, loss or mutation of both copies is required before noticeable changes in expression occur. Similarly, local gene amplification has been shown to be the driving force in some tumors.⁴⁶ We therefore sought to determine whether *6A somatic acquisition is associated with chromosomal gain or losses at the TGFBR1 locus with 2 different and complementary methods, comparative genomic hybridization (CGH), and loss of heterozygosity (LOH) analyses.

CGH Analysis. DNA from tumor tissue and reference DNA from peripheral blood lymphocytes were random prime labeled separately with biotin and digoxigenin. Reference DNA and 600 ng each of tumor DNA were cohybridized to normal metaphase chromosome spreads together with excess unlabeled Cot-1 blocking DNA. Hybridization was detected as described before.⁴⁷

9q22 LOH Analysis. We have previously mapped TGFBR1 to 9q22.²⁹ We screened 5 polymorphic markers (D9S287, D9S180, D9S1851, D9S1786, D9S176) mapping to the 9q22 region using primers obtained from Research Genetics (Invitrogen Corp, Carlsbad, Calif). Sequence map positions for the markers’ bp: D9S287: 90209651-90209826, D9S180: 92393020-92393239, D9S1851: 91314418-91314562, D9S1786: 90783544-90783745, and D9S176: 93801914-93802050. Markers D9S287, D9S1786, D9S1851, and D9S180 are located centromeric of TGFBR1; D9S176 is located telomeric. Data were collected using an ABI 377 automated DNA sequencer, analyzed with Genescan software, and allelic imbalance was determined using GeneScan/Genotyper software (Applied Biosystems). Allelic loss was determined using the method of Canzian et al⁴⁸ which calculates allele pair ratios for normal and tumor samples and designates a change of greater than 40% indicative of a loss of heterozygosity.

To investigate whether the deletion of GCG repeats within TGFBR1 was gene specific or associated with generalized genomic instability, we identified and sequenced other genes containing a similar sequence. BLASTN, basic local alignment search tool (available at http://www.ncbi.nlm.nih.gov/BLAST),⁴⁹ was used to search for genes containing sequential GCG repeats encoding alanines to determine whether the somatically acquired 9-bp deletion observed in the tumors of 5 patients was specific for TGFBR1 or also affected other genes with a similar repeat sequence. Three genes containing either 5 or 10 consecutive GCG codon repeats were identified (human homeo box HB9, HLXB9, gene identification 3110; human serine threonine kinase 39, STK39, gene identification 27347; Human growth differentiation factor 11, GDF11, Gene identification 10220) and amplified by PCR using the Advantage-GC genomic polymerase mix (BD Biosciences Clontech, Palo Alto, Calif) and the following primers: HLXB9: 5′-GCT GCT GCC CAA GCC GGG CTT CCT GG-3′ and 5′-GGA GTT GAA GTC GGG CAT CTT AGG CAG G-3′, STK39: 5′-TCC TGC TCT CCT CCG CAG CAT CAT G-3′ and 5′-CCT GCA GCT CGT ACG CGT CCC TGC A-3′, and GDF11: 5′-CGC TGC TGC TGG GCT TCC TGC TC-3′ and 5′-CGG CTG ATG TTG-GGC GCC TCC TT-3′.

Polymerase chain reaction products were cloned into the pCR 2.1 vector (BD Biosciences Clontech). Automated sequencing of 10 clones was performed to determine the presence or absence of TGFBR1*6A. If 2 or more clones contained the TGFBR1*6A alleles, the sample was considered to harbor TGFBR1*6A.

Briefly, HEK 293 cells (CMT, Phillipsburg, NJ) grown to 60% to 70% confluence were transiently transfected with pCMV5-TGFBR1-HA-Flag and pCMV5-TGFBR1*6A-HA-Flag using FuGENE according to the supplier’s (Roche Diagnostics, Indianapolis, Ind) instructions. Cells were harvested in ice-cold Dulbecco’s phosphate-buffered saline without calcium or magnesium and collected by centrifugation. Following cell lysis, the membrane fraction was collected by centrifugation. The solubilized receptor was purified by affinity-chromatography using Anti-Flag M2 Affinity Gel (Sigma-Aldrich, St Louis, Mo) column. The receptor was eluted with Flag peptide. Eluted fractions were pooled and concentrated. To prepare samples for protein sequence analysis, the partially purified receptor preparations were subjected to sodium dodecyl (lauryl) sulfate-polyacrylamide gel electrophoresis. For liquid chromatography mass spectrometry (LC-MS) analysis, gel slices containing the receptor were cut out of the gel after brief staining with Coomassie brilliant blue (Sigma-Aldrich). For protein sequence analysis by Edman degradation, proteins were transferred to Sequi-Blot PVDF membranes (Bio-Rad Laboratories, Hercules, Calif) and stained with 0.5% Ponceau S (Sigma-Aldrich) in 1% acetic acid. The major band representing the receptor was then cut out, washed extensively with water, and air-dried.

*9A and *6A protein bands were electroblotted to polyvinylidiene difluoride membrane, visualized with Ponceau S stain, excised and subjected to Edman degradation on an Applied Biosystems Procise 494 HS protein sequencer.

*6A and *9A bands were excised from the colloidal Coomassie Blue gel, and peptides sequenced by microcapillary high-performance liquid chromatography ion trap tandem mass spectrometry using Finnigan LCQ Deca XP+(Thermo Electron, San Jose, Calif) following proteolytic digestion. To ensure thorough peptide coverage of the amino terminus, each band was split equally, digested separately with trypsin and chymotrypsin, and then recombined for liquid chromatography coupled to tandem mass spectrometry. Tandem mass spectra were acquired on the top 4 ions following each survey scan during the high-performance liquid chromatography separation, with dynamic exclusion and relative collision energy of 30%. Data analyses were facilitated with the SEQUEST algorithm and Sequest Browser (Harvard Microchem), and manually confirmed.

The previously described TGFBR1-HA, TGFBR1*6A-HA, and TGFBR1*10A-HA constructs with an HA epitope at their COOH terminus^29- 30 were cloned into the pCS2 plasmid (Promega, Madison, Wis) with the initiator ATG codon in the context of a Kozak⁵⁰ consensus sequence. pIRES-TGFBR1*6A-HA-Flag and pIRES-TGFBR1-HA-Flag were generated by insertion of each of the HA-Flag–tagged alleles into the pIRES vector (BD Biosciences, Clontech). pIRES-TGFBR1*6AK-HA-Flag was generated by replacing the wild type kinase region of the receptor with that of the TGFBR1-K232R receptor.⁵¹ The proper alignment of the constructs was verified by sequencing.

The pCS2 constructs were transcribed by SP6 RNA polymerase (Promega) and translated in a reticulocyte lysate as described.⁵² Addition of inhibitor acceptor peptide and proteinase K treatment of microsomes was carried out as described.⁵² Translation products were analyzed by sodium dodecyl (lauryl) sulfate–polyacrylamide gel, which were quantified on a Fuji FLA-3000 phosphoimager using Fuji Image Reader 8.1j software (FujiFilm, Tokyo, Japan).

SW48, HCT 116, DLD-1, SW837, SW1417, HT-29, COLO 201, and COLO 320DM colorectal cancer cells were grown in the medium recommended by the supplier (ATCC, Manassas, Va).

MCF-7 breast cancer cells grown in the medium recommended by the supplier (ATCC) were stably transfected by electroporation with pIRES, pIRES-TGFBR1-HA-Flag, and pIRES-TGFBR1*6A-HA-Flag. MCF-7 cells were also stably transfected with a pIRES-TGFBR1*6AK-HA-Flag vector that contains the K232R mutated kinase domain.⁵¹ TGFBR1 expression was assessed by Western immunoblotting of whole-cell extracts using the anti-TGFBR1 sc-398 (Santa Cruz Biotechnology, Santa Cruz, Calif), the 3F10 anti-HA (Roche Diagnostics), and the anti-α-tubulin T6074 (Sigma) antibodies. MCF-7 clones with similar levels of receptor expression were chosen for TGF-β–mediated growth inhibition assays. Cell growth was assessed by ³H-thymidine incorporation assays. Briefly, 2 × 10⁵ cells in medium with 10% heat inactivated fetal bovine serum (Hyclone, Logan, Utah) were seeded into 6 well plates on day 1. On day 2, the medium was replaced with new medium with or without 100 pmol/L TGF-β (R&D Systems, Minneapolis, Minn). On day 3, 18 hours later, the medium was replaced with new medium with or without 100 pmol/L TGF-β and 5 μCi ³H-thymidine (Amersham, Piscataway, NJ). Four hours later, the cells were washed with phosphate-buffered saline at 4°C, then fixed with 95% methanol. Cell lysis was achieved with 0.2 N sodium hydroxide and ³H-thymidine incorporation measured with a scintillation counter (Beckman Coulter, Fullerton, Calif) and expressed in counts per minute.

Total RNA isolation was performed using RNeasy protect mini kit (Qiagen, Valencia, Calif). We used the following primers and Taqman probes for TGFBR1 and glyceraldehyde-3-phosphate dehydrogenase (GAPD): TGFBR1 sense primer (5′-GCTTCGTCTGCATCTCACTCAT-3′), antisense primer (5′-TTGGCACTCGATGGTGAATG-3′), Taqman probe (5′-FAM TTGATGGTCTATATCTGCCACAACCGCA QSY7-3′), GAPD sense primer (5′-GAAGGTGAAGGTCGGAGTC -3′), antisense primer(5′-GAAGATGGTGATGGGATTC-3′), Taqman probe (5′-FAM CAAGCTTCCCGTTCTCAGCC QSY7-3′). Polymerase chain reaction amplification and detection was performed on the ABI PRISM 7700 Sequence Detection System (Applied Biosystems). The primers used recognize both the endogenous and transgenic TGFBR1. The TGFBR1 transcripts were quantitated relative to GAPD by Comparative C_T method following the Applied Biosystems protocol.

We used Fisher exact test to compare *6A somatic acquisition among MSI-H and MSI-L colorectal tumors. We used Pearson χ² test to compare the proportion of *6A in liver metastases to *6A among 3451 healthy controls and 4399 patients with a diagnosis of cancer. One sample 2-sided t test was performed to test the significance of TGF-β–mediated growth inhibition or stimulation for MCF-7 cells stably transfected with either *9A or *6A and for DLD-1 and SW48 cells. We used a 2-stage nested design with clones nested under transfected cells *6A and *9A for the growth inhibition/stimulation assays. There are 2 different *6A clones with 5 and 6 replicates, and 2 different *9A clones with 8 and 6 replicates. All analyses were performed with SAS version 9.0 software (SAS Institute, Cary, NC).

Tissues were available for 531 patients enrolled at our institutions between June 1, 1994, and June 30, 2004, and included 226 patients with squamous cell carcinomas of the head and neck, 157 patients with primary colorectal cancer, 104 patients with breast cancer, and 44 patients with liver metastases from colorectal cancer.

Somatic Acquisition of *6A During Cancer Development. First, we genotyped 226 squamous cell carcinomas of the head and neck and identified 46 (20.4%) with a *6A/*9A genotype (Figure 1). We determined the germline genotype in 24 of these individuals using DNA extracted either from peripheral blood lymphocytes or biopsy specimens from normal tissue sampled more than 2 cm away from the tumor. We observed that 4 patients had no evidence of *6A alleles in their normal DNA, 2 from blood, and 2 from histologically normal oral mucosa. To explore the possibility that *6A acquisition might occur in other tumor types, we genotyped tumor tissue from 157 patients with colorectal cancer and identified 30 patients (19.1%) whose tumors carried both the *6A and *9A alleles. Normal tissue from 4 of these patients was confirmed to carry 2 *9A alleles indicating a tumor-specific acquisition of the *6A allele in the course of cancer development (Figure 1). One patient with head and neck cancer and evidence of *6A somatic acquisition had received prior chemotherapy and radiation therapy. The other 3 patients had not received any prior cancer-related therapy. We genotyped tumor tissue from 104 patients with a diagnosis of breast cancer and identified 25 tumors (24.0%) that contained *6A and *9A alleles. We found that all germline and tumor DNA had the same *6A/*9A genotype suggesting that *6A acquisition may not be a common event during breast cancer development (Figure 1).

All results were confirmed by repeat PCR amplification, cloning of the PCR product, and sequencing of at least 10 clones. To assess the possibility of patient sample mispairing as an explanation for *6A acquisition, we tested normal and tumor DNA from 2 patients with head and neck cancer and 2 with colorectal cancer, all of whom were exhibiting evidence of *6A acquisition with several highly polymorphic (>80% heterozygosity) dinucleotide markers. Both colorectal cancer cases were heterozygous for the same alleles in peripheral blood lymphocyte and patient-matched tumor DNA: D18S69, D2S123, and D5S346 for the first patient and D9S180, D9S287, and D9S1786 for the second patient. Similarly, both patients with head and neck cancer were heterozygous for the same alleles in histological normal oral mucosa and patient-matched tumor tissue (D9S180, D9S1786, and D9S1851 for the first patient and D9S287, D9S1786, and D9S1851 for the second patient). These findings rule out sample mix-up as an explanation for our results.

Mosaicism is defined as the presence of genetically different cells derived from a single zygote. To rule out the possibility of circumscribed or even widespread mosaicism as a possible explanation for our results, we genotyped 12 additional head and neck tumors with matching normal tissue and lymph node sampled several centimeters away from the tumors. We found the same TGFBR1 genotype in all 3 tissues, 3 *6A/*9A and 9 *9A/*9A in tissue samples belonging to the same individuals. This shows that *6A is not commonly acquired during embryogenesis.

Somatic Acquisition of *6A in Liver Metastases From Colorectal Cancer. Having demonstrated that *6A is somatically acquired during colorectal and head and neck cancer development, we hypothesized that it may bestow cancer cells with a growth advantage and may be even more commonly acquired among metastases. We obtained 44 liver metastases from colorectal cancer and genotyped them for TGFBR1 exon 1. Twenty-two samples (50%) were *6A/*9A (Figure 1), a proportion more than 3.5-fold higher than that found in the germline of 3451 healthy controls (P<.001) and 3-fold higher than that found among 4399 patients with a diagnosis of cancer (P<.001).³⁵ Similarly to the patient population in our study, the 7850 cases and controls reported previously were predominantly white from the United States and Europe.

To determine whether *6A was somatically acquired in metastases, we extracted DNA from the surrounding normal liver, from the primary tumor of *6A/*9A liver metastases, or both. We were able to retrieve germline DNA from 15 patients and primary tumor from 13 patients of the 22 patients with *6A/*9A liver metastases. We found evidence of *6A somatic acquisition in 13 (29.5%) of 44 metastases. Thus, our results show that the high *6A frequency observed in liver metastases is predominantly due to somatic *6A acquisition either at the site of the primary tumor or, more commonly, during the process of metastasis as the primary tumor of 11 (85%) of 13 *6A/*9A liver metastases had a *9A/*9A genotype. Information on prior chemotherapy treatment was available for 9 patients. The 2 patients with evidence of somatically acquired *6A in the liver metastases and not in the primary tumor had not received any prior cancer treatment. This suggests that chemotherapy is unlikely to contribute to *6A acquisition in liver metastases.

To assess the possibility of patient sample mispairing as an explanation for the high frequency of *6A acquisition among liver metastases, we tested normal liver and tumor DNA from 13 patients with evidence of somatically acquired *6A. We used 9 polymorphic microsatellite markers (D17S250, AGATp53-190, D18S69, D2S123, D5S346, DSS550, D9S283, D4S413, and D9S176) to amplify and compare the normal and tumor DNA samples for each patient. We obtained conclusive results for an average of 4 markers per sample pair (range, 2-6) and found no mismatched normal/tumor pairs. These findings rule out sample mix-up as an explanation for our results.

The observed deletion of 3 GCG codons within the TGFBR1 gene in primary and metastatic tumors could theoretically be the result of a generalized phenomenon affecting all GCG repeat sequences within cancer cells with evidence of *6A somatic acquisition. Alternatively, it could be the result of background mutational activity subsequently fixed in tumor cells by selective growth advantage.

*6A Acquisition and Mutator Phenotype. To investigate whether the deletion of GCG repeats within the TGFBR1 was gene-specific or associated with generalized genomic instability, we identified 3 other genes containing GCG repeats coding for alanine in the protein: GDF11 (10 GCG), STK39 (5 GCG), and HLXB9 (5 GCG). Each gene was amplified by PCR from normal and tumor DNA obtained from 5 patients (3 patients with head and neck cancer, 2 patients with colon cancer) with evidence of *6A somatic acquisition. The PCR products were cloned, and 10 clones were sequenced. No differences in sequence were noted between normal and tumor DNA (Figure 2) demonstrating that *6A acquisition is not associated with a mutator phenotype.

*6A Acquisition of MSI. To examine the association of *6A with MSI, we tested 4 head and neck tumor samples with evidence of *6A somatic acquisition for expansion or contraction of the BAT-25 and BAT-26 mononucleotide repeats that are highly sensitive markers of MSI.⁵³ None of them showed evidence of MSI. To assess a possible relationship between *6A acquisition and MSI, a common finding in colorectal cancer, we genotyped similar numbers of MSI-L and MSI-H tumors. Among these tumors, we found that *6A was acquired de novo in 1 out of 17 MSI-L tumors and in 3 out of 13 MSI-H tumors. The difference between MSI-H (3 of 13) and MSI-L (1 of 17) is not significant: P = .29 (23.08% vs 5.88%) by Fisher exact test. This, together with the BAT marker data, shows that, in 2 different tumor types, MSI is neither necessary nor sufficient for *6A acquisition.

*6A Acquisition and Gains or Losses at 9q22. To explore the possibility of chromosomal gains or losses at 9q22, we performed CGH using tumor DNA from 2 head and neck tumors with evidence of somatic *6A acquisition. We found evidence of chromosomal deletions at 1p36, 2q29, 7p22, 9q34, 16p13.3 and 4p16, 16p11-p12, and 17p13 in the tumor of these 2 patients. In both cases, the 9q22 region had a balanced state indicating that amplification or deletion is not associated with acquisition of *6A in these tumor samples (Figure 3). We further investigated LOH at 9q22 as a possible explanation for this phenomenon by testing several additional polymorphic markers. Five markers were informative in 3 head and neck tumors and 1 colon tumor with somatic *6A acquisition: D9S287 (centromeric), D9S180 (centromeric), D9S1851 (centromeric), D9S1786 (centromeric), and D9S176 (telomeric). There was no instance of LOH in this chromosomal region (Table), thereby providing additional evidence that the emergence of the *6A allele is not associated with 9q22 LOH.

*6A and *9A Signal Sequences. Signal sequences play a key role in targeting and membrane insertion of secretory and membrane proteins.^54- 55 Signal sequences are usually N-terminal extensions directing nascent or completed proteins from the cytosol to translocation sites in the membrane of the endoplasmic reticulum in eukaryotic cells.⁵⁵ After membrane insertion, signal sequences are commonly cleaved from the precursor protein by a membrane-bound signal peptidase.

It has been proposed that the TGFBR1 signal sequence cleavage site is located within the protein's polyalanine tract.⁵⁶ In contrast, computerized prediction using the SignalP program (http://www.cbs.dtu.dk/services/SignalP/)⁵⁷ suggests that the signal sequences of *9A and *6A are likely to be cleaved between Ala33 and Leu34 (*9A) and between Ala30 and Leu31 (*6A). This site has all the hallmarks of a classic signal sequence cleavage site with Ala at positions −1 and −3 and Pro at position −5 relative to the cleavage site.⁵⁸ Theoretically, the deletion of 3 alanines in *6A should not affect the cleavage of *6A and *9A signal sequences. However, differences in TGF-β–mediated growth inhibition^30,36 and epidemiological evidence that *6A acts as a tumor susceptibility allele^34- 35 suggest that the deletion of 3 alanines may have significant functional consequences. To address this crucial question, we transfected human kidney HEK293 cell lines with a HA-Flag epitope-tagged TGFBR1 or TGFBR1*6A. Following lysis of the cells, the membrane fraction was purified by affinity chromatography using anti-Flag agarose.

The *6A and *9A bands were electroblotted to polyvinylidiene difluoride membrane, excised and submitted to automated Edman degradation. The amino-terminal sequencing results confirmed our predictions that the signal sequence cleavage of both proteins is located between positions 30 and 31 for *6A and between positions 33 and 34 for *9A (Figure 4). To confirm these results, equivalent *6A and *9A bands were excised from the gel and subjected to peptide sequencing by ion trap tandem mass spectrometry. Seventy peptides for *6A and 74 peptides for *9A were identified, corroborating the results of Edman analysis. Both sequencing strategies unveiled a minor secondary form of *9A signal sequence cleaved between positions 25 and 26 (Figure 4). Comparison of the phenylthiohydantoin amino acid yields from the Edman analysis indicated that this secondary form of *9A accounts for a few percent of the total amount. The secondary form was not detected in *6A.

As a further control, we translated pCS2-TGFBR1, pCS2-TGFBR1*6A, and pCS2-TGFBR1*10A in vitro in the presence of rough dog pancreas microsomes under standard conditions.⁵² All 3 proteins were efficiently inserted into the microsomal membrane as evidenced by efficient glycosylation of the unique glycosylation acceptor site in the short extracellular domain of the protein and protease-sensitivity of the large cytoplasmic domain in intact microsomes (Figure 5). Thus, neither the 9 bp deletion in the *6A signal sequence nor the 3 bp insertion in the *10A signal sequence measurably affect either targeting to or translocation across the endoplasmic reticulum membrane.

*6A Acquisition and Growth Advantage to Cancer Cells. The high frequency of *6A somatic acquisition in liver metastases and the absence of LOH suggested that *6A may act as a dominant allele that confers growth advantage to tumor cells and contributes to their clonal expansion. To test this hypothesis, we chose a cancer cell line with preserved TGF-β–mediated growth inhibition, the MCF-7 breast carcinoma cell line. We genotyped MCF-7 cells and found that they contain 2 copies of *9A and no *6A. We assessed the TGF-β–mediated proliferation of MCF-7 cells stably transfected with pIRES, pIRES-*9A, and pIRES-*6A. Both pIRES-*9A and pIRES-*6A clones were chosen based on similar levels of receptor expression compared with the endogenous level of TGFBR1 expression represented by the pIRES clone. We included clones with low (*9A-5 and *6A-5), intermediate (*6A-1) and high (*9A-9) levels of receptor expression as assessed by real-time PCR and confirmed by Western immunoblotting (Figure 6). All experiments were performed at least 4 times in triplicates. As shown in Figure 7, *6A results in growth stimulation (*6A-5 and *6A-1). However, cells transfected either with *9A or with the vector alone are growth inhibited by TGF-β (pIRES, *9A-5, and *9A-9).

To determine whether *6A effects on TGF-β mediated growth inhibition depend on the receptor-signaling capabilities, we stably transfected MCF-7 cells with a pIRES-*6A construct that contains a mutation of lysine 232 to arginine (K232R). This mutation destroys the kinase and signaling abilities of the receptor.⁵¹ We chose 1 clone with high (*6AK10) and 1 with intermediate (*6AK15) levels of receptor expression (Figure 6). Similarly to what was observed with clones *6A-5 and *6A-1, both clones were growth stimulated on exposure to TGF-β (Figure 7). This shows that *6A effects on TGF-β mediated growth inhibition are independent of TGF-β signaling. The growth pattern of *9A and *6A clones in the absence of exogenously added TGF-β was similar (P = .54). In the presence of exogenously added TGF-β, the average growth inhibition rate for *9A clones was 28.47%, while the average growth stimulation rate for *6A and *6AK clones was −26.33% and −30.30%, respectively. The average difference in the proliferation rate between the 2 *9A and the 2 *6A clones was 54.8%.

To further determine whether endogenous *6A could yield similar biological effects in colorectal cancer, we genotyped 8 commonly used colorectal cell lines for TGFBR1 exon 1. One cell line (DLD-1) had a *6A/*9A genotype, the others a *9A/*9A genotype. Cell proliferation assays showed that the growth of the DLD-1 cell line was significantly stimulated by TGF-β. As a matching control, we chose the SW48 colorectal cell line which, similarly to DLD-1, is a MSI-H cell line. The growth of the SW48 cell line was significantly inhibited by TGF-β. Each experiment was performed at least 4 times in triplicate. The average growth stimulation rate for the DLD-1 cells was −31.16% (95% confidence interval, −44.0% to −18.33%, χ²₄; P=.003) and the average growth inhibition rate for the SW48 cells was 29.51% (95% confidence interval, 18.28%-40.75%; χ²₄; P=.002). Hence, these experiments suggest that endogenous *6A may be associated with TGF-β–mediated growth stimulation in colorectal cancer cells.

The mechanism responsible for polyalanine-coding sequence mutations is still debated. Two hypotheses have been proposed: slipped mispairing during DNA synthesis according to the mechanism proposed by Streisinger and colleagues^59- 60 and deletions or duplications of mixed GCG sequences due to unequal crossovers during meiosis.⁶¹ It may seem surprising that this mechanism results in a number of GCG repeats identical to that of *6A, a tumor susceptibility allele found in 13.7% of the general population.^34- 35 None of the other genes containing a similar number of GCG repeats exhibited any triplet deletion, which argues against a new mutator phenotype mechanism affecting GCG repeats. That *6A was the only somatically acquired TGFBR1 allele in primary and metastatic tumors suggests that, even if instability is present at the TGFBR1 locus, the *6A allele provides a selective growth advantage over other TGFBR1 alleles, such as the previously reported *5A, *8A, and *10A alleles.³⁰ That *6A is somatically acquired in a small proportion of primary colorectal and head and neck cancer and in a large proportion of colorectal metastases is consistent with the results from our genetic studies and suggests that *6A acquisition is not due to a specific mechanism but simply to background mutational activity subsequently fixed in malignant cells by selective growth advantage. Whether *6A is acquired before the development of cancer or concomitantly with the emergence of cancer will need to be further studied.

It has been previously shown that intense immunostaining for TGF-β in the primary tumor of patients with colorectal cancer correlates with disease progression to metastasis.⁶² Similarly, colon carcinoma progression is associated with gradual and significant increases in expression of TGF-β messenger RNA and protein.⁶³ The acquisition of *6A by the colorectal primary tumors and the subsequent selective clonal expansion of *6A clones sheds a new light on the role of TGF-β during cancer development and progression. It also indicates that individuals who carry the *6A allele, either in the germline or somatically acquired by the tumor, may have a greater likelihood of developing metastases than individuals who do not carry this allele. *6A may therefore serve as a useful biomarker in cancer. The high frequency of TGFBR1*6A somatic acquisition observed in liver metastases and the growth advantage it confers to cancer cells in a TGF-β-rich environment provide a rationale for targeting TGFBR1*6A in cancer. Given that germline tissue was available for only 24 of the 46 *9A/*6A head and neck cancer, and 15 of 22 *9A/*6A liver metastases, *6A somatic acquisition in these tumors may be underestimated.

Since 13.7% of the general population and 17.1% of patients with a diagnosis of cancer carry at least 1 copy of the *6A allele,³⁵ our findings may have substantial public health importance. The high frequency of *6A carriers in the general population and the moderately increased risk of breast, colon, and ovarian cancer that it confers implies that the dominant effects of *6A have an incomplete penetrance. Additional studies are needed to determine which environmental and genetic factors may modify the penetrance of *6A in these tumor types.

From a functional point of view, we have previously shown that although *6A transduces less effectively TGF-β mediated growth inhibition than *9A, other rare variants such as *10A do not result in impaired TGF-β signaling.^30,36 The fact that *6A and *9A differ through their signal sequences and not their mature receptors provides an explanation for the observed lack of differences in TGF-β binding and receptor turnover between *6A and *9A.³⁰ The presence of a minor alternatively cleaved form of the *9A signal sequence may result in small functional differences between the *6A and *9A forms. Hence, the biological effects of *6A may theoretically either result from its 30 residue long signal peptide or from the proportional decrease of the minor species of mature TGFBR1 starting with the amino acid sequence ALLPG. The findings that the overwhelming majority of the mature *6A and *9A receptors are identical, together with the evidence that *6A switches TGF-β growth inhibition into growth stimulation independently of TGF-β signaling in cancer cells, suggest that the molecular differences between *6A and *9A are signal sequence and not receptor mediated. Indeed, the only difference between MCF-7 cells transfected with *6A and MCF-7 cells transfected with *9A is the 30 amino acid long signal sequence cleaved off from *6A.

We previously found that *6A–mediated growth inhibition was lower than *9A–mediated growth inhibition at every TGF-β concentration,³⁰ a pattern incompatible with a receptor-mediated mechanism. Taken together with the discovery of a probable signal sequence–mediated mechanism, our data suggest that altered *6A mediated–growth inhibition results either from direct transcriptional inactivation or from alteration of pathways that regulate TGF-β signaling. One may predict that the secondary signals generated by *6A signal sequence modulate the gene expression of known effectors of TGF-β induced cell cycle arrest such as CDKN2B (p15^INK4B),^7- 8CDKN1A (p21^CIP1),⁹MYC , CDK4, and CDC25A.⁶⁴ It could also affect the expression levels of TGF-β receptors and SMADs because these mechanisms have been shown to underlie cancer cells’ loss of TGF-β–mediated growth inhibition.⁶⁵ A direct interaction between part of *6A signal sequence and regulatory elements of these genes constitutes a plausible hypothesis. Other potential explanations for *6A biological effects are its potential role as an oncogene and modulation of cell migration and invasion.

The tumor suppressor role of TGF-β in carcinogenesis stems from its ability to maintain homeostatic control of growth in premalignant cells and cells progressing through the early stages of carcinogenesis.³⁰ Inactivating mutations of the TGF-β signaling pathway receptors^30,66 and intracellular messengers^30,67 have established them as bona fide tumor suppressors. Our results provide experimental evidence that TGFBR1*6A acquisition may confer a growth advantage to cancer cells by switching TGF-β growth inhibitory signals into growth stimulatory signals. This highlights a novel role for TGFBR1*6A within the TGF-β signaling pathway with significant functional consequences.

In summary, this study adds pivotal knowledge to the molecular understanding of *6A, a candidate tumor susceptibility allele likely to contribute to a large proportion of some common human cancers. First, these data demonstrate that *6A is somatically acquired at high frequency in colorectal cancer metastases and by a small proportion of primary colorectal and head and neck primary tumors. Second, these findings resolve the underlying molecular differences between *6A and *9A and show that the shorter signal sequence may bestow cancer cells with a significant growth advantage, which, to our knowledge, is the first report of a tumor-susceptibility gene exerting its biological actions by means of its signal sequence. Third, these results provide a plausible mechanism of action that reconciles the apparent discrepancy between the epidemiological data associating *6A with cancer and the relatively modest impairment in TGF-β–mediated growth inhibition previously reported in normal epithelial cells. Finally, these observations suggest that TGFBR1*6A may act as an oncogene regulating cellular pathways that may be amenable to therapeutic intervention.