ONCOGENOMICS

APC and the three-hit hypothesis

S Segditsas1,4, AJ Rowan1,4, K Howarth1,4, A Jones1, S Leedham2, NA Wright2, P Gorman1,W Chambers1,3, E Domingo1, RR Roylance1, EJ Sawyer1, OM Sieber1,5 and IPM Tomlinson1

1Molecular and Population Genetics Laboratory, London Research Institute, Cancer Research UK, London, UK; 2HistopathologyLaboratory, London Research Institute, Cancer Research UK, London, UK and 3Department of Colorectal Surgery and GeneticsKnowledge Park, Oxford Radcliffe Hospitals, Oxford, UK

The seminal ‘two-hit hypothesis’ implicitly assumes thatbi-allelic tumour suppressor gene (TSG) mutations causeloss of protein function. All subsequent events in thattumour therefore take place on an essentially nullbackground for that TSG protein. We have shown thatthe two-hit model requires modification for the APC TSG,because mutant APC proteins probably retain somefunction and the two hits are co-selected to produce anoptimal level of Wnt activation. We wondered whether theoptimal Wnt level might change during tumour progres-sion, leading to selection for more than two hits at theAPC locus. Comprehensive screening of a panel ofcolorectal cancer (CRC) cell lines and primary CRCsshowed that some had indeed acquired third hits at APC.These third hits were mostly copy number gains ordeletions, but could be protein-truncating mutations.Third hits were significantly less common when the secondhit at APC had arisen by copy-neutral loss of hetero-zygosity. Both polyploid and near-diploid CRCs had thirdhits, and the third hits did not simply arise as a result ofacquiring a polyploid karyotype. The third hits affectedmRNA and protein levels, with potential functionalconsequences for Wnt signalling and tumour growth.Although some third hits were probably secondary togenomic instability, others did appear specifically totarget APC. Whilst it is generally believed that tumoursdevelop and progress through stepwise accumulation ofmutations in different functional pathways, it also seemsthat repeated targeting of the same pathway and/or geneis selected in some cancers.Oncogene (2009) 28, 146–155; doi:10.1038/onc.2008.361;published online 6 October 2008

Keywords: two hits; tumour suppressor; APC; Wnt;copy number change; ‘just right’

Introduction

The ‘two-hit’ hypothesis (Knudson, 2001) has domi-nated cancer genetics for over 20 years. In this model,one copy of a tumour suppressor gene (TSG) isinactivated by a nonsense or frameshift mutation, andthe other copy is inactivated by a similar mutation or byloss of heterozygosity (LOH). Although rarely statedexplicitly, the ‘two-hit’ model implicitly applies to adiploid cell. In contrast, the typical cancer cell hasbecome polyploid. Thus, two questions are unanswered.First, do TSGs require three, four or more hits if theyneed to be inactivated in a polyploid cancer cell, or isthis so unlikely that TSG inactivation is uncommon insuch a situation? The failure to find many commonlymutated TSGs in colorectal cancer (CRC) or breastcancer suggests that the latter might be correct (Woodet al., 2007). Second, for those TSGs that are inactivatedbefore the tumour cell becomes polyploid, does acquisi-tion of a near-triploid or near-tetraploid state simplyleads to random gain of one or both copies of themutant TSG with no functional consequences? Thisstudy addresses this latter question.Bi-allelic APC mutations are generally believed to

initiate colorectal tumorigenesis. Although there isdebate as to whether APC mutations might occur on abackground of genetic instability (Michor et al., 2005)or, indeed, cause instability (Fodde and Smits, 2001),most evidence indicates that APC is mutated when cellsare near-diploid rather than grossly aneuploid. We haveshown previously that the APC TSG does not acquiremutations that lead to simple protein inactivation. Thebi-allelic APC mutations within colorectal tumours aredistributed non-randomly within the gene (Lamlumet al., 1999), with the position and type of the second hitdepending on the random first hit (Figure 1). Theobserved APC mutations are probably selected becausethey lead to an optimal level of Wnt signalling (Lamlumet al., 1999; Albuquerque et al., 2002). Criticaldeterminants of the positions of APC mutations appearto be their locations relative to the protein’s seven20-amino acid repeats (20AARs) that take part inb-catenin binding and degradation.Given these selective constraints on APC mutations,

we reasoned that the process of tumour progressionmight take the colorectal tumour cell away from its

Received 20 February 2008; revised 7 July 2008; accepted 28 August2008; published online 6 October 2008

Correspondence: Dr IPM Tomlinson, Wellcome Trust Centre forHuman Genetics, University of Oxford, Roosevelt Drive, Headington,Oxford OX3 7BN, UK.E-mail: iant@well.ox.ac.uk4These authors contributed equally to this work.5Current address: Ludwig Colon Cancer Initiative Laboratory, LudwigInstitute for Cancer Research, PO Box 2008, Royal MelbourneHospital, VIC 3050, Australia

Oncogene (2009) 28, 146–155& 2009 Macmillan Publishers Limited All rights reserved 0950-9232/09 $32.00

www.nature.com/onc

ON

CO

GE

NO

MIC

S

optimal level of Wnt signalling. The optimal Wntsignalling of a late CRC might, for example, be verydifferent from that of an early adenoma. One reasonmight be that the tumour cell’s microenvironment haschanged, or that additional genetic changes—such as apolyploid karyotype—have been acquired. Thus,although a near-diploid colorectal tumour cell mighttypically harbour one APC mutation leaving no20AARs and another leaving two 20AARs, a triploidcarcinoma cell would end up with a total of either two20AARs or four 20AARs, depending on which copy ofchromosome 5 was gained when polyploidy wasacquired. It is entirely possible that neither of thesegenotypes is the optimum for that cell. We wondered,therefore, whether the ‘two-hit’ hypothesis requiredamendment in the typical CRC, in that the APC genemight sometimes acquire additional mutations duringtumour progression that returned the cell closer to itsoptimum Wnt level. We therefore tested the idea thatthird hits occur at APC in CRCs, by studying both celllines and primary tumours.

Results

Models of mutation, LOH and copy number changeat APCFor a TSG, the observed pattern of truncating muta-tions, LOH and copy number change usually allow theunderlying order and types of changes to be deduced,although this is easier in cell lines than in primarycancers. Figure 1 shows how multiple hits at the APClocus would be reflected in observed genetic data.Initially, two hits occur in a diploid cell. The cell thenbecomes polyploid, resulting in trisomy, through someunspecified mechanism. Third hits, if we discountpolyploidization as a hit, are then presumed to occur.Consider first the most common scenario, in which

two truncating mutations are the initial hits at APC.When near-triploidy occurs, one of the two chromosome5 homologues is duplicated. A third hit might then occurby the deletion of one copy of APC. If this affects thechromosome 5 that has not been duplicated in thetransition to polyploidy, there will be LOH of thatdeleted region (Figure 1(1A)). If the other (duplicated)chromosome 5 is deleted, there is no LOH (Figure1(1B)). If there is gain of APC, that change is readilydetectable, but there is no LOH (Figure 1(1C)).In the second scenario, there occurs at the near-

diploid stage a truncating mutation accompanied by asecond hit deletion, and hence LOH results. Whentriploidy occurs, duplication of one homologue leads toLOH with a two-copy deletion (Figure 1(2A)) or LOHwith a single-copy deletion (Figure 1(2B)), depending onthe homologue that is duplicated. The regions ofdeletion and LOH coincide (that is, their boundariesare identical). There is no true third hit in thesescenarios. However, a third hit—for example, in theform of a further copy number change—could occurand would be detectable by array comparative genomic

hybridization (aCGH) if it affected a region differentfrom that affected by the second-hit deletion (see belowfor an example in a primary CRC, no. 622).In scenario three, the second hit is copy-neutral LOH

(mitotic recombination, gene conversion or chromoso-mal non-disjunction and reduplication). Triploidy leadsto three essentially identical copies of a region ofvariable size around APC. A third hit in the form ofcopy number deletion (Figure 1(3A)) will usually bedetectable, and distinguishable from scenario 2 ofFigure 1 because the region is unlikely to matchperfectly the region of LOH. Third hit by copy numbergain is readily detectable (Figure 1(3B)). Also, third hitsmight take the form of additional truncating mutations,presumably 50 of the existing mutation (Figure 1(3C)).Such protein-truncating third hits might also occur inscenarios 1 and 2, but are omitted from Figure 1 for thesake of simplicity.Other genetic pathways involving multiple hits at

APC might occur. Those shown have been chosenbecause they tend to correspond to the observed data(see below). Some third hits, moreover, might be cryptic.For example, scenarios 1A and 2B of Figure 1 result inthe same observed data, whereas one has involved athird hit and the other has not. Other cryptic third hitsmight occur—for example, in scenario 3A of Figure 1—if the region of deletion is the same as that of the originalcopy-neutral LOH. In other cases, gains and deletions atAPC might overlap, causing complex patterns that aredifficult to decipher. In the analysis below, we score thepresence of third hits conservatively, but use Figure 1 asa framework for our interpretation.

Three hits at APC are frequent in CRC cell linesComprehensive screening for large- and small-scalemutations and for true LOH has rarely been undertakenin CRCs at the APC locus. We initially screened a set of47 CRC cell lines. Overall, 38 (81%) of the cell linesharboured one or more nonsense or frameshift APCmutations. Two protein-truncating mutations werefound in 14 (30%) of the cell lines. LOH was a commonevent, occurring in 18 (38%) cell lines. Most LOH wascopy neutral, resulting from whole-chromosome loss/regain in 4 cases and mitotic recombination, with abreak point between 67Mb and APC, in 10 cases. Anadditional four cell lines had LOH by deletion.In at least 13 of 47 (28%) of the CRC cell lines, the

genetic changes at APC could not readily be explainedby only two hits (Table 1, Figure 1). In the near-triploidlines LS123, SW948 and C75, two truncating APCmutations were accompanied by a deletion of one copyof the gene, but no LOH (Figure 1(1B)); a similarmechanism almost certainly applied to COLO205. InSW837, LS411, NCI-H747 and C106 cell lines, therewere two truncating APC mutations together with gainof one gene copy (Figure 1(1C)). The same sequence ofmutations had almost certainly occurred in PC/JW,SW1116 and C32; in these cell lines, there was only onedetected truncating mutation, but as the gains flankedAPC by at least several megabases and were extremely

APC and the three-hit hypothesisS Segditsas et al

147

Oncogene

unlikely to inactivate APC, the most plausible explana-tion was that these lines had a cryptic, protein-truncating second hit, with the gain as a third hit. Inall of these 11 CRC cell lines, the most parsimonioussequence of events (see detailed explanation in Figure 1)was that the truncating mutation(s) and/or LOH werethe initiating two hits at APC in a cell disomic forchromosome 5, and that the copy number changeoccurred subsequently during tumorigenesis.Four cell lines (C99, GP5D, C125 and SW1417) had

acquired one truncating mutation and an interstitialdeletion of one copy of APC, the position of the deletionmatching a region of LOH. The underlying order andnumber of mutations were uncertain in these lines. Themost likely scenarios were that a deletion was the secondhit in a diploid cell, causing LOH (Figure 1(2)), or that athird-hit deletion had occurred (Figure 1(1A)), aftereach chromosome had initially acquired two truncatingmutations.There was no simple tendency for cell lines to end up

with a particular total number of 20AARs in theirmutant proteins (details not shown). However, theposition/types of the APC mutations were associatedwith the presence of third hits. After excluding the C99,GP5D, C125 and SW1417 cell lines from the analysis(owing to the uncertainty regarding the order andnumber of hits in these cases), third hits were found inonly 2 of 14 (14%) cell lines with a truncating mutationplus copy-neutral LOH, the two exceptions (Table 1)being C70 (Figure 1(3A)) and LS1034 (Figure 1(3B)). Incomparison, third hits were significantly more frequent

(11/18, 61%; w21¼ 5.8, P¼ 0.016) in lines with truncatingAPC mutation(s) and no LOH. Importantly, polyploidy(P¼ 0.24) and chromosomal instability (measured bythe total genome-wide number of chromosome-scalegains or deletions of >10Mb, P¼ 0.38) were notindependent predictors of a third hit when thesevariables were incorporated into a logistic regressionmodel, whereas copy-neutral LOH remained a signifi-cant predictor (P¼ 0.025).

Demonstration of three hits at APC in primary CRCsTo exclude our observations of three hits at APC beingthe result of phenomena that were restricted to cancercell lines, we screened a panel of 70 primary sporadicCRCs for genetic changes at APC. In primary cancersthat have not been rigorously microdissected, allelicimbalance (AI) and aCGH analyses cannot alwaysdetect and distinguish between true LOH, deviation inallelic ratios without true reduction to homozygosity,copy number change and unbalanced polyploidy (Fig-ure 2). We therefore reasoned that the existence of threehits could not readily be assessed in every primarycancer and that accurate assessment of the frequency ofthird hits was not possible in our primary CRCs. Ouranalysis was therefore restricted to showing that thirdhits were not a cell line artefact.Using the criteria detailed in Figure 2, we found 12

CRCs, both polyploid and near-diploid, with evidenceof three hits at APC (Table 2). Half of these CRCs hada single truncating APC mutation, a presumed (but

Table 1 CRC cell lines with evidence of three hits at the APC locus

Cell line MSI Karyotype Truncatingmutation

aCGH SNP LOH FISHverified?

Suggestedmechanism

LS123 — 63 1450+1554 Del 51-128 (1 of 3 copies) None Yes 2 trunc.>delSW948 — 67 1114+1429 Del 51-ter (1 of 3 copies) None Yes 2 trunc.>delC75 — 70 811+1450 Del 51-ter (1 of 3 copies) None ND 2 trunc.>delCOLO205 — 78 1554 Del 50-113 (1 of 3 copies) None ND ?2 trunc.>del

SW837 — 40 213+1450 Gain 69-ter (1 of 2 copies) None Yes 2 trunc.>gainLS411 — 75 789+1554 Gain 99-ter (1 of 3 copies) None Yes 2 trunc.>gainC106 — 79 1238+1490 Gain all (1 of 3 copies) None ND 2 trunc.>gainNCI-H747 — 60 161+1429 Gain all (1 of 3 copies) None ND ?2 trunc.>gainPC/JW — 70 969 Gain 108-116 (2 of 3 copies) None ND ?2 trunc.>gain� 2SW1116 — 63 1430 Gain all (1 of 3 copies) None Yes ?2 trunc.>gainC32 — 74 776 Gain 75-ter (1 of 3 copies) None Yes ?2 trunc.>gain

C70 — 61 1309 Del 51-118 (1 of 3 copies) Loss 67-ter ND MR>delLS1034 — 77 1309 Gain 108-ter (1 of 3 copies) Loss 88-ter Yes MR>gain

Abbreviations: aCGH, array comparative genomic hybridization; CRC, colorectal cancer; Del, deletion; FISH, fluorescence in situ hybridization;karyotype, modal chromosome count; LOH, loss of heterozygosity; MR, mitotic recombination; MSI, microsatellite instability; ND, not determined;redup, reduplication; SNP, single nucleotide polymorphism; Ter, 5q telomere; trunc., protein-truncating mutation(s); >,‘followed by’.The table shows combined results of mutation screening, LOH analysis, copy number analysis, karyotyping and FISH using both an APC-specificprobe and a chromosome 5q paint. Codons of truncating APCmutations are shown. Boundaries of large-scale changes are shown in megabases (Mb).Estimated number of copies gained or deleted is shown, relative to the overall level of ploidy. Our suggested mechanism to explain the observedchanges is also shown. Note that (i) copy number is shown relative to the overall level of ploidy; third hits do not, therefore, result simply from theacquisition of the polyploid karyotype; (ii) no evidence of monoallelic expression without LOH—for example, as a result of methylation-relatedtranscriptional silencing—was detected in any line; (iii) the following cell lines without evidence of third hits were analysed, but data are not shown indetail: CX1, HCA46, HT29, HT55, LOVO, SW403 and VACO5 (all two trunc. only); HCA7 and SKCO1 (both one trunc. only); C10, HCT15/DLD1,LIM1863, SW620, VACO4S, SW1222, CACO2 and COLO678 (all trunc.+LOH by MR); HRA19, COLO320, VACO10MS and T84 (alltrunc.+LOH by whole chromosome loss and redup); C99, GP5D, C125PM and SW1417 (all trunc.+LOH by deletion); COLO741 (deletion, but noLOH); and H716, HCT116, HUTU80, LS174T, RKO, SNUC2B, SW1116 and SW48 (nil found).

APC and the three-hit hypothesisS Segditsas et al

148

Oncogene

x x

No LOH, no deletion xx x

No LOH, one-copy deletion

xx xx

xx xLOH, one-copy deletion

(coincide)

x

x

LOH, two-copy deletion(coincide)

LOH, one-copy deletion(coincide)

x

x

x x x x

LOH, no deletion

x x x

LOH, one-copy gain

1A

1B

1C

3A

3B

3C

1

Two mutations

x

Mutation + del

Mutation + copy-neutral LOH

LOH, one-copy deletion(not coincide)

x x x

x xx

No LOH, one-copy gain

2A

2B

LOH, no copy change,two mutations

xxxx

First and second hits TriploidizationNormal cell

x

Third hits

2

Figure 1 Third-hit genetic pathways involvingAPC. When analysing data on mutations, copy number changes and LOH at a locus such asAPC, the correspondence between the biological phenomena and their detection using technically imperfect molecular methods must initiallybe borne in mind. Some illustrations are as follows. First, deletion in a diploid cell leads to true LOH (in the sense of molecular reduction tohomozygosity), but deletion of one copy in a polyploid cell generally does not cause true LOH. Second, copy number gain does not lead totrue LOH (unless artefactually if many copies of the same homologue are gained). Third, aneuploidy/polyploidy per se does not cause trueLOH (except in the very rare haploid situation), although it often does cause allelic imbalance in primary cancers (see Figure 2). Fourth,microsatellites (and other polymorphisms) measure LOH through relative allelic dosage, aCGHmeasures copy number relative to the DNAcontent of the tumour as a whole and FISH measures total copy number. The figure summarizes the pathways that most plausibly underliemost of the observed combinations of genetic changes at APC in the CRC cell lines, and that can be detected using the above analyses. Thecancer cell’s complement of changes atAPC is shown, with a truncating mutation denoted by a cross, deletion by an open box and gain (of amutant allele) by a black box. Copy-neutral LOH is shown, for convenience, as occurring by mitotic recombination. It is assumed that the‘third hits’ occur after the cell has acquired a near-triploid karyotype, as most CRC cell lines are near-triploid; however, the principles of thethree-hit analysis are extendable to other karyotypes. Note that in some cases, such as cells with two truncatingAPCmutations and deletion(1B), deletion must have occurred at the polyploid stage (for otherwise, there would be a single truncating mutation plus a deletion thatcauses LOH (2A or 2B)), but in other cases, such as gains of APC (1C, 3B), that third hit could also have occurred in a near-diploid cell.

APC and the three-hit hypothesisS Segditsas et al

149

Oncogene

undetected) second truncating mutation and deletionwithout AI (although it must be borne in mind thatLOH analysis can occasionally produce false-negativeerrors and that some of these cancers may actually havehad LOH as a second hit by deletion). It is thereforeimportant to note that we identified other cancers withvery good evidence of three hits, specifically: (i) twoCRCs with two detectable truncating mutations and

deletion; (ii) two cancers with third hits by copy numbergain; (iii) a cancer that had acquired a truncatingmutation, LOH of the whole chromosome by deletion,and then a further smaller deletion involving APC(Figure 3a); and (iv) a CRC (no. 1015) that had acquiredtwo protein-truncating mutations (at codons 216 and1315), together with LOH at all three markers close toAPC, but without copy number change. In this last case,

Polyploid cancer:copy number deletion,two mutations found

Cancer Normal

xx

Polyploid cancer:copy number gain, one ortwo mutations found

x

Polyploid cancer: copynumber deletion but no AI,one mutation found

Near-diploid cancer:no copy number change, AI, two mutations found

x

xx

x

Figure 2 Third-hit genetic pathways involving APC in primary cancers. The figure shows scenarios where third hits can reliably bedetected in primary cancer specimens: (a) two truncating mutations and copy number deletion; or (b) one or two detected truncatingmutations and copy number gain; (c) one detected truncating mutation and copy number deletion in the absence of AI; or (d) twotruncating mutations and AI without copy number change in a near-diploid cancer.

APC and the three-hit hypothesisS Segditsas et al

150

Oncogene

the most parsimonious explanation was that an initialprotein-truncating mutation was followed by copy-neutral LOH and then by a further mutation on oneof the two copies of chromosome 5 (Figure 1(3C),Figure 2d); given that mutations near APC codon 1309are generally associated with LOH (Lamlum et al.,1999), it is likely that the codon 1315 mutation was thefirst event, and that the codon 216 mutation proximal tocodon 1315 provided an additional advantage byproducing a more severely truncated protein.

Loss of a mutant APC allele in progression from adenomato carcinomaIn general, copy number changes at APC are rare incolorectal adenomas and most of these lesions are near-diploid (Sieber et al., 2002; Jones et al., 2007). It seemsunlikely, therefore, that third hits are common inadenomas. During investigation of eight ‘malignantpolyps’ (CRC with contiguous adenoma), we discoveredan example of a third hit following laser capturemicrodissection of the two parts of the lesion(Figure 3b). The adenomatous component did not showAI and contained an APCmutation at codon 1406 and apresumed cryptic second hit. In the cancer, there wasnear-complete AI. Both parts of the lesion contained ap53 E171X mutation, demonstrating a common origin.Thus, this lesion had lost an APC-mutant allele duringprogression from adenoma to carcinoma.

Do some third hits not only involve APC but also target it?Some background copy number changes must inevitablyoccur in chromosomally unstable (CINþ ) cancers andsome of these may involve chromosome 5q. Alterna-tively, a locus other than APC on 5q might bespecifically targeted by these changes. However, thedetection of two protein-truncating APC mutations incancer no. 1015 with LOH shows that APC can be thethird-hit target. In addition, the following indirectevidence suggests that copy number third hits may alsosometimes target APC:

(i) We found several cancers with small regions ofthird-hit gain or deletion centred on APC, the

smallest being a gain of 8Mb in the cell line PC/JWand two deletions of just 1–2Mb in two primarycancers, one with two truncating mutations(Figure 3c) and the other with one identifiedtruncating mutation and no LOH.

(ii) The highest frequency of copy number change onchromosome 5 in CRCs was at APC (Figures 4aand b); this was not the site of maximum change incancers of the breast (Figure 4c) or ovary(Figure 4d) that we had studied previously.

(iii) Gain or deletion involving chromosome 5q wasone of the most frequent large-scale eventsgenomewide, occurring in about one-third ofCRCs, well above background rates.

(iv) Gain or deletion rarely (three cell lines and fourprimary cancers) involved the whole of chromo-some 5, showing that it did not simply occur as aresult of endoreduplication during the process ofpolyploidization.

(v) Where assessment was possible in the CRC celllines (details not shown), gain or loss of materialaround APC occurred by several different mechan-isms, including translocation, deletion and iso-chromosome formation, and involved differentregions of 5q, providing very little evidence forthe action of a specific form of genomic instabilityor chromosome fragility.

(vi) Some CIN cancers showed evidence of third hits(for example, gain around APC in the near-diploidline SW837 and in primary cancers nos. 621 and982).

Do third hits at APC affect mRNA and protein levels?Given that most truncated APC proteins found in vivoretain some b-catenin-binding activity, it is possible thatcopy number changes modulate Wnt signalling by thismeans. Direct in vitro evidence for the functionalconsequences of third hits at APC is, however, extremelydifficult to obtain, given that any effects are likely to bequantitative rather than qualitative, and there are severeproblems in precisely replicating the in vivo situation inany artificial or model system. However, it is known thatusing siRNA or the expression of exogenous mutant

Table 2 Summary of primary CRCs with evidence of third hits at APC

Cancer no. Mutation 1 Mutation 2 APC AI Copy no.change APC

Ploidy MSI Proposed mechanism

338 R283X C>T E1317X G>T No Deletion ? N 2 trunc>del575 1287FS insT No Deletion Polyploid N ?2 trunc.>del621 518FS del5bp 1383FS insA No Gain Near-diploid N 2 trunc>gain622 411FS insC Yes Deletion Polyploid N LOH by whole chromosome

del then further del close to APC667 Q247X C>T No Deletion Polyploid N ?2 trunc.>del681 1309FS delAAGA No Deletion Polyploid N ?2 trunc.>del951 1439FS del10bp No Deletion Polyploid N ?2 trunc.>del961 R1450X C>T No Deletion Polyploid N ?2 trunc.>del982 1516FS delCT Yes Gain Near-diploid N trunc.>LOH>gain1015 1315FS insA R216X C>T Yes None Near-diploid N trunc.>LOH>trunc.1344 R1114X C>T 1448FS delC No Deletion Polyploid N 2 trunc>del1409 1430FS delC No Deletion Polyploid N ?2 trunc.>del

Abbreviations and nomenclature are as for Table 1.

APC and the three-hit hypothesisS Segditsas et al

151

Oncogene

APC mRNA can affect Wnt signalling and cell function(for example Schneikert et al., 2007). We deter-mined whether the level of APC mRNA co-varied withtotal APC copy number using 24 of our CRC cell lines.APC expression was a median of 0.014 (inter-quartile

range ¼ 0.004–0.039) of that of the per copy expressionof the control GAPDH mRNA. Higher genomic APCcopy number was significantly associated with increasedAPC mRNA expression (Spearman’s tb¼ 0.53,P¼ 0.008). We then showed, using subclones of theSW837 cell line that differ in APC copy number, thatthis mRNA difference was proportionately reflected inprotein levels (details not shown).

Discussion

The two hits at APC that initiate colorectal tumorigen-esis are usually bi-allelic truncating mutations that occurat non-random positions with respect to each other. Theresulting residual protein function produces a level ofWnt signalling that is close to the optimum (Lamlumet al., 1999; Albuquerque et al., 2002). When true LOH(reduction to homozygosity) occurs at APC, it usuallyinvolves no copy number change (Sieber et al., 2002;Gaasenbeek et al., 2006). This observation has producedan apparent paradox, in that in CRCs, copy numberchanges at APC are common. These copy numberchanges can take the form of gain or deletion in bothnear-diploid and aneuploid cancers. We have shownthat the copy number changes are effectively third hits.Very occasionally, moreover, CRCs can acquire threehits without copy number change, for example by twosomatic mutations plus copy-neutral LOH. Our datasuggest that at least some of these third hits specificallytarget APC and can affect APC mRNA and proteinlevels. Our findings support a model whereby someCRCs respond to a changing environment by modulat-ing Wnt signalling through copy number changes orother types of third hit at APC.We recognize that the accepted genetic model of

colorectal tumorigenesis might be incorrect, in that APCinactivation is not the initiating event in the pathogen-esis of some colorectal tumours, but might occur afterthe carcinoma has become polyploid, such that three ormore hits are needed to inactivate the protein. This iscontrary to the bulk of evidence from colorectaladenomas (Sieber et al., 2005; Jones et al., 2007) andcannot apply to third hits that occur by gain, bytruncating mutation or in near-diploid tumours. It is notclear as to why third hits at APC only occur in someCRCs, or why some tumours have gains and othersdeletions. Perhaps different cellular environments—including accumulated mutations—confer different se-lective constraints, leading to different selection pres-sures and hence different requirements for third hits.Cancers without third hits may have acquired alter-native molecular changes that allow them to acquire anear-optimal level of Wnt signalling. An intriguing, butunexplained, observation is the paucity of third hitswhere the cancer’s second hit is by copy-neutral LOH.We found no evidence to suggest that this associationresulted from some secondary effect of chromosomalinstability. One explanation might lie in the fact thatcopy-neutral LOH causes all APC mutations in the cellto be identical. Polyploidization does not, therefore,

Figure 3 Examples of ‘third hits’ at APC in primary cancers. (a)The plot shows aCGH-derived copy number along chromosome 5relative to the overall ploidy level in cancer no. 622. There isdeletion of the whole of chromosome 5, probably involving onecopy of three present in this polyploid cancer, plus a smaller regionof deletion (red line) around APC, probably involving anothercopy, leading to a double-copy deletion around APC. (b) The left-hand panel shows an hematoxylin and eosin-stained sectiondemonstrating normal mucosa (1), adenoma (2) and carcinoma(3) present in this section of a ‘malignant polyp’. The right-handpanel shows AI following laser capture microdissection atmicrosatellite D5S346 in the carcinoma (lower panel), but not theadenoma (upper panel), an observation also reflected in thesequence electropherogram (data not shown). The allelic ratios innormal tissue and the adenomatous portion of the tumour werealmost identical, suggesting that contamination by normal cells didnot obscure AI in the adenoma. (c) The plot shows deletion of fourclones (circled) around APC in cancer no. 667 that did not have AI(Table 2).

APC and the three-hit hypothesisS Segditsas et al

152

Oncogene

‘unbalance’ the original two hits and third hits are notselected.Three hits at APC have previously been shown to occur

in tumours from patients with specific germline mutationsthat cause attenuated familial adenomatous polyposis(AFAP). In AFAP, these third hits often take the form ofloss of a germline mutant allele or a mutation at codon1554 in cis with a more proximal germline mutation(Spirio et al., 1998; Su et al., 2000; Sieber et al., 2006). It isintriguing that some of our sporadic CRCs had changescompatible with the loss of a mutant allele and others hadmutations at codon 1554, a change that is found quiterarely in tumours from patients with classical FAP. It isconceivable, therefore, that some codon 1554 (or similar)mutations in sporadic CRCs were unrecognized third hits.In their influential review of carcinogenesis, Hanahan

and Weinberg (2000) stated that tumours develop andprogress through stepwise accumulation of mutations indifferent functional pathways. Our data suggest thatrepeated targeting of the same pathway and/or geneoccurs in some cancers, perhaps to allow adaptation tochanging genetic, cellular and tissue environments. Ourdata provide further evidence to suggest that APC doesnot follow the conventional TSG model in which twocopies are simply inactivated. In the conventional

model, all subsequent events in that tumour cell takeplace on a background of the loss of TSG function, andthere is no realistic possibility of the TSG acquiringfurther mutations to alter its function. We tentativelysuggest that, for APC, this ‘multiple hit’ model besubstituted for the ‘two-hit’ model. Evidently, we do notcurrently know whether this notion applies to genesother than APC. There is evidence, for example, thatgains on chromosome 12p preferentially occur in CRCsthat acquired KRAS2 mutations when they were adeno-mas (Leslie et al., 2006), and this might be a case of amutant oncogene genotype tracking the selective land-scape. Where third hits occur at TSGs, either the cancermight be dependent on the sequential genetic changes forits growth or it might merely be ‘fine-tuning’. If the formerwere the case, there would be considerable therapeuticpotential for disrupting carcinogenesis by targeting genesthat rely on ‘multiple hits because small environmentalchanges might cause the genotype to become sub-optimal.

Materials and methods

A panel of 47 CRC cell lines was derived from public sourcesand collaborators. DNA and mRNA were extracted from

0

5

10

15

20

25

0

5

10

15

20

25

0

2

4

6

8

10

12

14

16

50 55 60 65 70 75 80 85 90 95100105110115120125130135140145150155160165170175180 50 55 60 65 70 75 80 85 90 95100105110115120125130135140145150155160165170175180

50 55 60 65 70 75 80 85 90 95100105110115120125130135140145150155160165170175180

50 55 60 65 70 75 80 85 90 95100105110115120125130135140145150155160165170175180

0

2

4

6

8

10

12

Figure 4 Frequencies of copy number change along the long arm of chromosome 5. Only cancers with copy number change on 5q areshown. X-axis, distance in Mb. Y-axis, number of cancers showing copy number change. (a) Colorectal cancer cell lines (N¼ 23); (b)primary colorectal cancers (N¼ 34); (c) primary grade III ductal breast carcinomas (N¼ 35); and (d) ovarian cancer cell lines (N¼ 22).The breast and ovarian cancer data are derived from our unpublished work. Note the peak frequency of change at APC (112Mb) inthe two types of colorectal cancer, but no evidence of such a peak in the other cancer types despite chromosomal instability in many ofthese cancers.

APC and the three-hit hypothesisS Segditsas et al

153

Oncogene

these samples using standard methods. A panel of 70 fresh-frozen primary CRCs and paired normal tissue was alsostudied. These samples had been snap-frozen at surgery andeach cancer was shown to contain >65% malignant cells byhistological review. Formalin-fixed, paraffin-embedded(FFPE) material was also available from the same cancers.DNA was extracted from each sample and from paired normaltissue.For copy number assessment, aCGH analysis was under-

taken on 1-Mb density bacterial artificial chromosome(BAC) arrays as described (Fiegler et al., 2003). We usedthe terms ‘gain’ and ‘deletion’ for aCGH (copy number)changes. For consistency, we also referred to the absence of awhole chromosome as ‘deletion’. The determination ofsignificant copy number changes was performed using theprogram DNAcopy (http://www.bioconductor.org/repository/release1.5/package/html/DNAcopy.html).For assessing ‘true LOH’—that is, reduction to homozygo-

sity—in cancer cell lines, DNAs were studied using theAffymetrix (Santa Clara, CA, USA) GeneChip Human Map-ping 10K Xba131 single nucleotide polymorphism arraysaccording to the manufacturer’s instructions. Median call ratesof 96% (IQR¼ 90–97%) were achieved. The MGCOS andGDAS (Affymetrix) software were used to process the outputdata according to the manufacturer’s protocols and thresholdsfor scoring homozygotes and heterozygotes. Copy numberchange was not analysed using the Affymetrix system, sincefound aCGH to be more reliable and to provide betterquantitation.True LOH at APC could not readily be assessed in the

primary cancers, owing to inevitable contaminating ‘normal’cells that prevented differentiation between copy numberchanges that caused relative shifts in allelic dosage and truereduction to homozygosity. AI, a measure of nominallysignificant deviation from a 1:1 allelic ratio, was thereforeassessed instead. We used microsatellite markers just distal tothe APC locus (D5S346, D5S421 and D5S656) on the AppliedBiosystems (Foster City, CA, USA) 3100 sequencer. Consti-tutionally, homozygous markers were scored as non-informa-tive. At each marker, AI was considered present if the areaunder one allelic peak in the adenoma was less than 0.5 timesor greater than 2 times that of the other allele, after correctingfor the relative allelic areas using constitutional DNA.Although these somewhat arbitrary thresholds are commonly

used to score AI/LOH, we had previously confirmed that theyshowed good performance (specificity of 86% and sensitivityof 68%) by finding (i) a bimodal distribution separating fresh-frozen samples with LOH from those without, with a divisionat ratioo0.58 or >1.72; (ii) measurement of allelic ratios in apanel of normal DNA samples and cancer cell lines withvarious known copy numbers at APC; and (iii) use of the CRCcell lines to verify concordance between LOH as assessed bymicrosatellites with that assessed by SNP microarrays (OMSieber et al., in preparation). The ratios of 0.5 and 2.0 wereused to err on the side of sensitivity rather than specificity andto provide consistency with previous studies. All tests wereperformed in duplicate, and in the infrequent event ofdiscordance between markers, precedence was given to themarker closest to the APC gene.The APC gene was screened for mutations proximal to the

site of the first serine-alanine-methionine-proline (SAMP)repeat at codon 1580 using direct sequencing in forward andreverse orientations (details available on request). Any possiblemutations were re-sequenced from a new PCR product. The b-catenin gene (exon 3) was sequenced similarly. Microsatelliteinstability was assessed using the markers BAT25 and BAT26.Bandshift at either of these mononucleotide repeats was takenas indicative of microsatellite instabilityþ status.From selected CRC cell lines, we undertook full chromo-

some counts and karyotype analysis using standard methods.Fluorescence in situ hybridization was performed on selectedcell lines by standard methods using an APC-specific BACprobe and a chromosome 5q paint was used in a similarmanner (details available on request). For assignment ofploidy in the primary CRCs, we used flow cytometric analysis.All tumours with evidence of an aneuploid and/or polyploidflow cytometry peak, distinct from the diploid peak andcorresponding to a DNA index of greater than 1.2, wereclassed as polyploid.

Acknowledgements

We are grateful to colleagues at St Mark’s Hospital for tissuecollection, to several kind providers of cell lines and to theMutation Detection Facility, Cancer Research UK LondonResearch Institute.

References

Albuquerque C, Breukel C, van der Luijt R, Fidalgo P, Lage P, Slors FJet al. (2002). The ‘just-right’ signaling model: APC somatic mutationsare selected based on a specific level of activation of the beta-cateninsignaling cascade. Hum Mol Genet 11: 1549–1560.

Fiegler H, Carr P, Douglas EJ, Burford DC, Hunt S, Scott CE et al.(2003). DNA microarrays for comparative genomic hybridizationbased on DOP-PCR amplification of BAC and PAC clones. Genes

Chromosomes Cancer 36: 361–374.Fodde R, Smits R. (2001). Disease model: familial adenomatouspolyposis. Trends Mol Med 7: 369–373.

Gaasenbeek M, Howarth K, Rowan AJ, Gorman PA, Jones A,Chaplin T et al. (2006). Combined array-comparative genomichybridization and single-nucleotide polymorphism-loss of hetero-zygosity analysis reveals complex changes and multiple forms of chromo-somal instability in colorectal cancers. Cancer Res 66: 3471–3479.

Hanahan D, Weinberg RA. (2000). The hallmarks of cancer. Cell 100:57–70.

Jones AM, Thirlwell C, Howarth KM, Graham T, Chambers W,Segditsas S et al. (2007). Analysis of copy number changes suggests

chromosomal instability in a minority of large colorectal adenomas.J Pathol 213: 249–256.

Knudson AG. (2001). Two genetic hits (more or less) to cancer.Nat Rev Cancer 1: 157–162.

Lamlum H, Ilyas M, Rowan A, Clark S, Johnson V, Bell J et al.(1999). The type of somatic mutation at APC in familial adenomatouspolyposis is determined by the site of the germline mutation:a new facet to Knudson’s ‘two-hit’ hypothesis. Nat Med 5:1071–1075.

Leslie A, Stewart A, Baty DU, Mechan D, McGreavey L,Smith G et al. (2006). Chromosomal changes in colorectaladenomas: relationship to gene mutations and potential for clinicalutility. Genes Chromosomes Cancer 45: 126–135.

Michor F, Iwasa Y, Vogelstein B, Lengauer C, Nowak MA. (2005).Can chromosomal instability initiate tumorigenesis? Semin Cancer

Biol 15: 43–49.Schneikert J, Grohmann A, Behrens J. (2007). Truncated APCregulates the transcriptional activity of beta-catenin in a cell cycledependent manner. Hum Mol Genet 16: 199–209.

APC and the three-hit hypothesisS Segditsas et al

154

Oncogene

Sieber OM, Heinimann K, Gorman P, Lamlum H, Crabtree M,Simpson CA et al. (2002). Analysis of chromosomal instability inhuman colorectal adenomas with two mutational hits at APC. Proc

Natl Acad Sci USA 99: 16910–16915.Sieber OM, Segditsas S, Knudsen AL, Zhang J, Luz J,Rowan AJ et al. (2006). Disease severity and genetic pathwaysin attenuated familial adenomatous polyposis vary greatlybut depend on the site of the germline mutation. Gut 55:1440–1448.

Sieber OM, Tomlinson SR, Tomlinson IPM. (2005). Tissue, cell and stagespecificity of (epi)mutations in cancers. Nat Rev Cancer 5: 649–655.

Spirio LN, Samowitz W, Robertson J, Robertson M, Burt RW,Leppert M et al. (1998). Alleles of APC modulate the frequencyand classes of mutations that lead to colon polyps. Nat Genet 20:385–388.

Su LK, Barnes CJ, Yao W, Qi Y, Lynch PM, Steinbach G. (2000).Inactivation of germline mutant APC alleles by attenuated somaticmutations: a molecular genetic mechanism for attenuated familialadenomatous polyposis. Am J Hum Genet 67: 582–590.

Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, Leary RJ et al.(2007). The genomic landscapes of human breast and colorectalcancers. Science 318: 1108–1113.

APC and the three-hit hypothesisS Segditsas et al

155

Oncogene