Details of cancer cell lines can be found in Additional file 1. Immortalised human breast luminal epithelial line HB4a 24 was obtained from Professor Mike O'Hare (LICR/UCL Breast Cancer Laboratory, University College, London, UK) and normal male lymphoblastoid line m62 obtained from Dr Chris Tyler-Smith (Wellcome Trust Sanger Institute, Hinxton, UK). Both were cultured in DMEM:F12 (Invitrogen, Paisley, UK).

All tumours were invasive breast carcinomas of small size (91% < 3 cm) and low Nottingham Prognostic Index, collected between 1990 and 1996 as part of the Nottingham/Tenovus series 2526272829. 65% were ER positive. The tissue microarray consisted of over 100 breast tumours. Matching RNA, TRIzol^®-extracted from fresh-frozen tissue, was available for 61 of these tumours.

For the 8p tiling path array, 328 overlapping BAC clones for the length of 8p were selected from the 32 k clone set available from the BACPAC Resource Centre (CHORI, CA, USA) 30. Additional clones to close gaps were obtained from Invitrogen and the Wellcome Trust Sanger Institute. A subset of 6% of these BACs was sequenced to confirm their identity. All except one matched their expected position on the human genome. The remaining clone showed evidence of containing more than one sequence and was substituted on the final array. The selected overlapping BAC clones for 8p provided an average clone spacing of 138 kb. A spot failure rate of around 10% resulted in an actual resolution averaging 155 kb. For the fosmid array, 123 fosmids were obtained from the Sanger Institute, mainly covering the regions of 8p21.3 (27 clones between 21.59 Mb and 22.25 Mb) and 8p22 (43 clones between 12.97 Mb and 15.68 Mb tiling three genes but not intergenic regions), identified by BAC array CGH as containing multiple rearrangements, to a resolution of 0.04 Mb. BACs were amplified by the method of Fiegler et al. 31 and printed with the custom genomic 8p12 array first described in Huang et al. 32 on CodeLink slides (Amersham Biosciences, Amersham, UK). Fosmids were prepared by the same method and printed as a separate array.

Array hybridisations were carried out essentially as described in Garcia et al. 14 and Pole et al. 10. Tumour genomic DNA was labelled with Cy3-dCTP and a pool of normal female genomic DNA with Cy5-dCTP (Amersham Biosciences) using Bioprime labelling kits (Invitrogen). Samples were hybridized to arrays overnight in the presence of Cot1 DNA (Invitrogen), and washed in PBS/0.05% Tween-20 at room temperature, 50% formamide/SSC at 42°C and then again in PBS/0.05% Tween-20. Arrays were scanned on an Axon 4100A scanner with data collected using GenePix Pro 4.1 software (Axon Instruments, CA, USA) and analyzed in Excel. For array painting, chromosomes (provided by Dr Karen Howarth, University of Cambridge, UK) were sorted by flow-cytometry and amplified by GenomiPhi (GE Healthcare, Buckinghamshire, UK) according to the method of Howarth et al. 33. They were hybridised against GenomiPhi-amplified normal female genomic DNA.

The array performance was tested by self-self and male-female hybridisations and using known 8p rearrangements in T47D (Additional file 2). Self-self hybridisation gave a standard deviation (SD) of ± 0.08 and male-female hybridisation ± 0.04 SD. Log₂ ratio shifts in response to copy number changes were tested by hybridisation of T47D against normal female gDNA. Copy number changes were estimated by inspection, taking into account both the shift in log₂ ratio and the level of noise for each sample. In general a change in log₂ ratio of 0.5 or more was scored as a change in copy number and a gain in log₂ ratio of 2 or more was scored as amplification.

The probability of the observed distribution of rearrangements on 8p occurring in the absence of any selection or pre-disposition to breakage at certain sites was calculated by chi-squared test. Chromosome 8 was divided into the major bands (8p11, 12, 21, 22, 23) and the observed number of rearrangements in each band tested against the number expected at random according to the size of each band.

Metaphase preparation and FISH was carried out as described previously 1034. Chromosome 8 paint was made using flow sorted chromosomes kindly provided by Patricia O'Brien and Professor Ferguson-Smith (University of Cambridge, UK). All BACs used for FISH were tested individually on normal (m62) metaphases for hybridisation to the correct region of chromosome 8.

Tissue microarray slides were prepared from formalin-fixed paraffin embedded tissue and analysed according to the method of Chin et al. 35. Probes for FISH on paraffin embedded tumours were prepared in the same way as for conventional FISH. A pool of three BACs, RP11-529P14 (21.8 Mb; positions are given as BAC midpoints according to NCBI Build 36), RP11-67H12 (21.9 Mb) and RP11-70D12 (22.0 Mb), was used for the test region and a pool of two BACs, RP11-381G11 (19.4 Mb) and RP11-222M11 (19.5 Mb), was used as a reference.

RNA was reverse transcribed using SuperScript III Reverse Transcriptase (Invitrogen) and oligo dT primer. Quantitative real-time PCR was carried out, in triplicate, in 10 μl reactions containing 1× SYBR Green PCR Master Mix (Applied Biosystems, CA, USA), 2.5 pmol both forward and reverse gene specific primer and 1 μl of 10-fold diluted cDNA. Cycling conditions were 50°C for 2 min, 95°C for 10 min then 40 cycles of 95°C for 15 s, 60°C for 1 min and a final dissociation step of 95°C for 15 s, 60°C for 15 s and 95°C for 15 s in an ABI PRISM 7900 HT Sequence Detection System (Applied Biosystems).

Primer sequences are given in Additional file 3. Standard curves for each primer pair were used to calculate amplification efficiency coefficients and melting curve analysis following qPCR confirmed that each primer amplified a single product. Relative expression levels were calculated as:

( Average Sample Ct for gene  −  Average reference Ct for gene ) Amplification Coefficient for gene ( Average Sample Ct for GAPDH  −  Average reference Ct for GAPDH ) Amplification Coefficient for GAPDH MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaqcfa4aaSaaaeaadaqadaqaaiabbgeabjabbAha2jabbwgaLjabbkhaYjabbggaHjabbEgaNjabbwgaLjabbccaGiabbofatjabbggaHjabb2gaTjabbchaWjabbYgaSjabbwgaLjabbccaGiabboeadjabbsha0jabbccaGiabbAgaMjabb+gaVjabbkhaYjabbccaGiabbEgaNjabbwgaLjabb6gaUjabbwgaLjabbccaGGGaaiab=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@15B2@

For primary samples expression levels were compared to the median-expressing tumour. Purified luminal cells, purified basal cells 36 (both kindly provided by Professor Mike O'Hare, UCL, London), a range of commercial normal RNAs (Stratagene (Cambridge, UK), USBiological (Europa Bioproducts Ltd, Wicken, UK), Ambion (Huntingdon, UK), BioChain (AMS Biotechnology, Abingdon, UK), Clontech (Basingstoke, UK)), and normal breast line HB4a were used as normal controls.

As we described previously, most of the cell lines (25/28) with rearrangements of 8p had at least one copy number step in 8p11-12, with loss of most or all of 8p distal to this region. Similarly, six out of the eight amplicons found were in the known region of amplification in 8p11-12. A full analysis of these events is given in Pole et al. 10.

Two cell lines, MDA-MB-134 and BT-20, showed amplification in an overlapping region of 8p21.3. This was confirmed and the amplicon further mapped by fosmid array CGH (Figures 2a and 2b, Additional file 6). Both amplicons contained the genes GFRA2 (GDNF Family Receptor Alpha-2) (21.6 Mb), DOK2 (Docking protein 2) (21.8 Mb), XPO7 (exportin 7) (21.9 Mb), NPM2 (nucleophosmin/nucleoplasmin family member 2) (21.9 Mb) and FGF17 (Fibroblast Growth Factor 17) (22.0 Mb). The centromeric edge of the amplicon in MDA-MB-134 fell within gene EPB49 (Erythrocyte membrane Protein Band 49) (22.0 Mb), whereas the amplicon in BT-20 extended proximally to encompass up to a further eighteen genes and one micro RNA. By both fosmid and BAC array CGH, the amplicon in BT-20 consisted of two levels of amplification with a step up at 22.02 Mb (Figure 2a), placing the highest level of BT-20 amplification outside the MDA-MB-134 amplicon.

FISH was used to confirm these amplifications and to investigate their structure. FISH on BT-20 with BAC RP11-235B11 (23.0 Mb), from the region of highest amplification, and BAC RP11-459H21 (21.2 Mb), from the adjacent region of gain, suggested that these two regions of copy number increase were not co-localised (Figure 2c). There was one apparently normal chromosome 8 with signals from both BACs. All extra copies of the region of highest amplification were found on a single small derivative chromosome. One copy of the region of lower amplification was seen on this derivative and one extra copy was present on a separate derivative chromosome 8.

MDA-MB-134 is known to have an 8p12 amplicon that is part of a marker chromosome containing co-amplified 8 and 11 sequences 17. FISH on MDA-MB-134 was carried out with BAC RP11-135I5 (21.5 Mb), from the 8p21.3 amplicon, and BAC RP11-104D16 (40.2 Mb) from the 8p12 amplicon. This showed that the 8p21.3 amplicon was intermingled with the 8p12 and 11q13 amplicons on the der(?)t(8;11)ins(8;11) (Figure 2d). Incidentally, FISH also showed that the two copies of this marker chromosome are not identical but have undergone further evolution since endoreduplicating. In addition there was one apparently normal copy of chromosome 8.

Dramatic amplification of the region that was both gained in BT-20 and amplified in MDA-MB-134 was found in a single primary breast tumour, out of 98 useable tumour cores, by FISH on a primary breast tumour tissue microarray, with a pool of BACs positioned within the amplicon (21.8 Mb-22.1 Mb) and a pool positioned 2.5 Mb distal to the amplicon (19.3 Mb-19.6 Mb) (Figure 2e).

Expression of the genes within the amplicon, GFRA2, XPO7, NPM2, FGF17 and EPB49, was analysed to see if amplification had an effect on expression levels. Quantitative RT-PCR was performed on cDNA from the two cell lines that had amplification, and a further four cell lines without amplification in this region (Figure 2f). DOK2 was excluded from quantitative expression studies as expression was not detectable by conventional PCR in normal breast line HB4a, BT-20 or MDA-MB-134 (results not shown). MDA-MB-134 showed over-expression of GFRA2, XPO7, NPM2 and FGF17 compared to HB4a but 0.5-fold expression of EPB49, consistent with the edge of the amplicon falling within this gene (Figure 2f). BT-20 showed marginal, up to 3-fold, over-expression of XPO7 and EPB49. Interestingly, one cell line, DU4475, showed huge over-expression of NPM2 and FGF17 in the absence of amplification (Figure 2f).

Since NPM2 and FGF17 were over-expressed in two cancer cell lines, expression levels were analysed in 61 primary tumours, from the primary breast tumour series analysed by FISH, for which cDNA was available (Figures 2g and 2h). However, cDNA was unavailable for the tumour that showed 8p21.3 amplification by FISH. For both genes, cDNA made from five commercial normal breast RNAs showed extreme variation – over 100-fold difference – in expression across the samples, highlighting the problem of identifying suitable normal controls for primary tumour samples. The variation between commercial normal samples may reflect differences in how they are obtained, for example if RNA is extracted from tissue obtained after reduction mammoplasty it may be more representative of adipocytes than breast epithelium. These technical concerns may limit the utility of commercially available normal RNA as a reference. Purified breast luminal and basal cells 36 were included as further controls and tended to be more consistent in their expression levels.

None of the tumours expressed FGF17 above the range of the control groups (Figure 2g). Three tumours expressed NPM2 at a higher level than any of the commercial controls, although at a lower level than the normal luminal and basal cells (Figure 2h). Although tumours do not express either gene outside the range of normal controls, there are four outliers/probable outliers for each gene, which have expression at significantly higher levels than the rest of the tumours (Figures 2g and 2h). Due to the variability of the normal controls the most significant observation may be that there were outliers over-expressing both genes within the tumour group.

A cluster of copy number steps that might affect a common target gene were found between 12.9 Mb and 15.6 Mb in 8p22 (Figure 3a, Additional file 5). Unbalanced changes in this region were seen in four cell lines: HCC1806, SUM44, DU4475, and pancreatic line MIA PaCa-2, which has a homozygous deletion previously reported by Bashyam et al. (4. In addition, we recently found a reciprocal translocation breakpoint at 14.7 Mb in HCC1187 by array painting, i.e. hybridisation of individual chromosomes to arrays 33 (Figure 3a). These rearrangements were fine-mapped to a resolution of 0.04 Mb, by array CGH on a fosmid array covering the three genes in this region, 13.0 Mb-13.4 Mb (DLC1), 14.0 Mb-14.5 Mb (SGCZ – sarcoglycan zeta) and 15.4 Mb-15.7 Mb (TUSC3) (Figure 3a, Additional file 6).

In SUM44 there were two copy number steps resulting in an extra copy of a retained fragment of the region between 13.6 Mb, within the second intron of DLC1 by fosmid array CGH, and 14.6 Mb, within SGCZ. DU4475 had a deletion between 15.4 Mb and 15.6 Mb, virtually the same region as the homozygous deletion in MIA PaCa-2. The copy number shift observed in DU4475 was relatively small and FISH, with BAC RP11-314P10 (15.6 Mb) within the deletion and BAC RP11-480M18 (15.0 Mb) just outside it, showed that the deletion was only present on one out of the four copies of chromosome 8 present in DU4475 cells (Figure 3b).

In HCC1806 two copy number steps fell in this region (Figure 3a). Comparison of array painting results 33 and BAC array CGH allowed us to assign these to two separate copies of chromosome 8. There was a small deletion between 12.6 Mb and 13.9 Mb, including all of candidate cancer gene DLC1, on an otherwise normal copy of chromosome 8, and a much larger interstitial deletion between 14.2 Mb and 31.5 Mb on a del(8)(8p12-22). Hybridisation of flow-sorted del(8)(8p12-22) to the fosmid array pinpointed the distal edge of the deletion to between 14.19 Mb and 14.24 Mb, within the gene SGCZ (Figure 3a).

Overall DLC1, the most well characterised tumour suppressor candidate in 8p22, was only affected in a single cell line by a copy number decrease that included two other genes. This result is perhaps not surprising as it is known that DLC1 expression does not correlate with copy number 39 and, in several tumour types, down-regulation is caused by promoter methylation 40. In support of this, DLC1 expression was decreased in 62% (8/13) cell lines analysed by quantitative PCR, including four lines without detectable genomic changes in this region (Figure 3c).

In contrast, TUSC3 was affected by two small deletions that did not include any other gene. TUSC3 showed decreased expression in these two lines with deletions but not in any other line (Figure 3c). Anecdotally, TUSC3 also showed decreased expression in a panel of 61 primary breast tumours (Figure 3d). This identified six tumours that had completely lost expression of TUSC3, and a further thirteen that expressed it at a lower level than the range observed in normal breast luminal and basal cells.

Neither the 5' nor the 3' end of SGCZ showed expression in the lines with copy number changes within it (HCC1187, SUM44), or model normal breast line HB4a (results not shown), so it is neither a likely target of deletion nor of activation or inactivation by the translocation or insertion breakpoints.

Six cell lines showed losses close to the telomere, within 8p23, all including a minimum common region of deletion from 1.7 Mb to 1.9 Mb (Figure 4a). PMC42, HCC38 and HCC1569 had losses from the telomere to 2.2 Mb, 3.4 Mb and 3.7 Mb respectively; HCC1500 had a deletion between 1.3 Mb and 36.5 Mb; DU4475 had a heterozygous deletion between 1.1 Mb and 1.9 Mb and MIA PaCa-2 had a homozygous deletion between 1.7 Mb and 1.9 Mb, first reported by Bashyam et al. (4. In order to identify the genes affected by the minimum common region of loss, the homozygous deletion in MIA PaCa-2 was mapped by PCR. It included exon three, but not exon 2 of CLN8, and all of C8orf61, ARHGEF10 and micro RNA hsa-mir-596 (Figure 4a).

Since DU4475 had a heterozygous deletion, in order to fit the classical model of a tumour suppressor gene, we would expect the remaining copy of the target gene to be disrupted by some other mechanism, such as point mutation, in this line. Expression data for multiple cancer cell lines (not shown) suggested ARHGEF10 as a more likely target gene than CLN8, since ARHGEF10 expression was decreased by more than 50% in 83% (15/18) of lines tested. We therefore sequenced ARHGEF10 in DU4475 and found a C → T point mutation in exon 19 (Figure 4b). This causes a non-conservative histidine to tyrosine substitution in the protein product at position 733, a highly conserved region (Figure 4c). Although paired normal DNA is not available for this cell line this nucleotide change is not listed as a known SNP (single nucleotide polymorphism) in the Entrez SNP database (Build 129).

The breast cancer cell line DU4475 was unique in having five separate small deletions on 8p (Additional file 5), at least three of which coincided with candidate tumour suppressor genes: a deletion of the promoter of candidate tumour suppressor TUSC3 (Tumour Suppressor Candidate 3) (15.6 Mb) in 8p22, an approximately 1 Mb deletion between 21.9 Mb and 23.0 Mb in 8p21, containing candidate cancer genes RHOBTB2 (22.9 Mb) and DR5 (Death receptor 5) (23.0 Mb) 41 but not extending as far as DR4, and a deletion in 8p23.3, which emerged as a common region of loss in this study.

A novel amplicon was found in 8p21.3 in both cell lines and primary tumours. Amplification was a relatively rare event, seen in 6% (2/41) of cell lines and 1/98 primary tumours, perhaps due to the reported over-representation of tumours with amplicons in the cell line collection 19. However, the presence of the amplicon in primary tumour material confirms it as a genuine event in breast cancer. Furthermore, the minimum region of amplification contains candidate genes from families known to play a role in carcinogenesis.

FGF17 (fibroblast growth factor 17) is a member of the fibroblast growth factor family, a group strongly implicated in carcinogenesis. FGF pathway genes, including FGF3, FGF4 and FGF8, are common targets of MMTV integration 4243 and FGFR3 mutations have been reported in 41% of bladder cancers 44. FGF17 is upregulated in some prostate cancers and this correlates with a higher risk of metastases and lower survival 45. NPM2 (Nucleophosmin/nucleoplasmin member 2) can catalyse the assembly of DNA and histones into nucleosomes and is involved in decondensation and remodelling of sperm chromatin immediately after fertilisation. In the same family, NPM1 has the ability to regulate the function of the tumour suppressor protein ARF 46, and is mutated in several haematological malignancies including 35% of acute myeloid leukaemias 4748. GFRA2 mediates RET signalling in response to the ligand GDNF 49. RET is a well-known oncogene that plays a role in the development of thyroid carcinomas and the familial cancer syndrome multiple endocrine neoplasia (reviewed in 50).

Of these genes FGF17 and NPM2 showed over-expression in two cell lines, one with (MDA-MB-134), and one without (DU4475), amplification. However, neither was over-expressed in the second cell line with amplification in this region (BT-20). If, as the array CGH data suggests, the highest level of amplification in BT-20 lies outside the region amplified in MDA-MB-134, there may be multiple targets of amplification in this region. A similarly complex pattern of non-overlapping amplicon peaks is seen in 8p12 and 11q13 in breast cancer, and has led to the suggestion that they contain as many as four targets 1151.

SGCZ (14.6 Mb), although disrupted by two unbalanced changes and one balanced translocation breakpoint, is unlikely to be a cancer gene. It forms part of the sarcoglycan complex that links the intracellular cytoskeleton to the extracellular matrix in muscle 52, and our findings show that it is not expressed, either in normal breast or as a consequence of rearrangement. Although DLC1 (13.2 Mb) has a tumour suppressive effect in a variety of tumour types, including breast cancer (e.g. 40), it was not a likely target of these 8p22 rearrangements as it lay just outside the rearrangements in all except one case. This is consistent with suggestions that genomic rearrangement is not a mechanism of DLC1 inactivation 39. If a tumour suppressor gene is present in this region, then TUSC3 (15.6 Mb), previously suggested as a potential target in this region (45354, was the strongest candidate target of the rearrangements of 8p22 in this study. Two cell lines have deletions that solely affect TUSC3 and result in decreased expression and expression was lost or decreased in 31% of primary breast tumours. However, the presence of several breakpoints in this region that do not appear to have an effect at the gene level could alternatively suggest that this site is prone to breakage.

Loss of 8p23.3, with a minimum common region of between 1.7 Mb and 1.9 Mb, was found in both breast and pancreatic cell lines, consistent with evidence from LOH studies of a cancer gene situated between 1.0 Mb and 2.1 Mb. 1 Mb array CGH data 55 from primary breast tumours suggests that 8p23.3 is a genuine target in primary tumours. 8% (6/73) of tumours may have specific loss distal to 10 Mb. However, analysis of that dataset is difficult due to the low resolution and the effect of contaminating normal DNA on log₂ ratio shifts in primary material. In combination, these results suggest the region between 1.0 Mb and 2.1 Mb as the location of a tumour suppressor gene. The genes affected by the minimum common region of deletion, and therefore candidates, are CLN8, C8orf61, has-mir-596 and ARHGEF10. ARHGEF10 has emerged as a candidate colorectal cancer gene in the sequencing screen carried out by Wood et al. 56, and, in conjunction with our discovery of a heterozygous deletion of ARHGEF10 in DU4475 with mutation of the remaining copy, this makes it a strong candidate for the 8p23.3 tumour suppressor. Functional data about ARHGEF10 supports a role for it in carcinogenesis. It activates RhoB 57, which is down-regulated in multiple tumour types 5859 and is necessary for apoptosis in response to DNA damage in transformed cells 60.

A large number of cancers lose all, or nearly all of 8p, potentially suggesting the presence of tumour suppressor loci, while the occurrence of specific rearrangements on distal 8p supports the theory that there are multiple cancer genes located on 8p. Many previous LOH and copy number studies, as well as our data, especially the specificity of the deletions in DU4475, can narrow down the regions of loss to 8p22 and 8p23; we suggest TUSC3 and ARHGEF10 as possible targets of genomic rearrangement. There may well be other tumour suppressor genes, such as DLC1, located on 8p but these appear to be inactivated by alternative mechanisms, including methylation. These results lead us to suggest a parsimonious model to explain the varied rearrangements seen on 8p in cancer (Figure 5)

The most frequent rearrangements are loss of distal 8p, from 8p12 to the telomere, sometimes with proximal amplification of 8p11-12. As previously suggested, unbalanced translocations through 8p11-12 can be explained by the presence of a major tumour suppressor gene at the most distal of these breaks, around 30 Mb 10. However, we argue that if this were the only tumour suppressor gene on 8p, we might expect to see a mirror-image pattern of breaks distal to this gene with the telomeric sequences being retained, as well as interstitial deletions of the tumour suppressor gene. A tendency towards loss of the whole of distal 8p suggests that there is at least one further tumour suppressor gene, and that it is located close to the telomere (Figure 5). This is consistent with our data suggesting a tumour suppressor gene is located in 8p23.3. There can therefore also be any number of additional tumour suppressor genes in between these two which would give the same pattern of rearrangement, and our data suggests that there may be at least one further candidate gene located in 8p22. In line with the application of Occam's razor loss of both/all of these tumour suppressor genes is most often seen as loss of the whole of distal 8p, but can also be achieved by several more specific events, therefore explaining the complex and inconsistent pattern of changes on distal 8p (Figure 5) – tumours may have deletions of varying sizes encompassing two or more tumour suppressor genes as well as losing them all by loss of the whole of 8p or a proximal translocation.

Cell lines with multiple small deletions are invaluable in allowing us to narrow down the number and position of tumour suppressor genes on 8p. The unique pattern of deletions in DU4475, three of which target candidate tumour suppressor genes or regions previously reported as frequently deleted, supports 8p23.3, 8p22 and 8p21 as candidate regions. Potential targets in 8p23.3 and 8p22 have been discussed above while 8p21, although only affected by a single small deletion in this study, is the location of RHOBTB2, a candidate breast cancer gene based on the discovery of two somatic mutations 41, and the TRAIL receptors DR4 and DR5, which have been suggested as candidate cancer genes owing to their pro-apoptotic function 61. MIA PaCa-2 has two deletions on 8p, also supporting the existence of tumour suppressor genes at 8p22 and 8p23.3.

Although the premise of our Discussion is that the driving force behind 8p rearrangements is selection for the alteration of cancer genes, the same pattern of rearrangements might be observed if the driving force was the mechanism of breakage e.g the presence of fragile sites on 8p. For example, the breaks at 8p12 with proximal amplification could be due to the presence of a fragile site that initiates breakage-fusion-bridge cycles 62 although no fragile site has been shown to exist on 8p 63 and 8p12 breaks frequently occur without accompanying amplification.