Biallelic mutation of the ADENOMATOUS POLYPOSIS COLI ( APC ) gene is a hallmark of sporadic colorectal cancer and colorectal, duodenal and desmoid tumours that develop in familial adenomatous polyposis (FAP) patients. The mutations affecting both APC alleles are interdependent, the position of the first APC mutation determining where the second hit will occur. This results in a complex pattern of mutation distribution in the APC sequence that translates into the stabilization of β-catenin that in turn feeds the affected cells with a permanent mitogenic signal. We describe here a new APC domain, the β-catenin inhibitory domain (CID) of APC located between the second and third 20 amino acid repeats and therefore present in many truncated APC products found in human tumours. In truncated APC, the CID is absolutely necessary to down-regulate the transcriptional activity and the level of β-catenin, even when an axin/conductin binding site is present. The activity of the CID is dramatically reduced in several colon cancer cell lines and can be inhibited by shorter truncated APC lacking the CID. The CID is a direct target of the selective pressure acting on APC during tumourigenesis. It explains the interdependence of both APC mutations, not only in colorectal but also in duodenal and desmoid tumours.
Familial adenomatous polyposis coli (FAP) is a dominantly inherited disease that manifests by the development of polyps in the colon and the upper gastrointestinal tract that ultimately evolve into fatal aggressive tumours when left untreated. FAP patients are also affected by desmoid tumours, which are abdominal neoplasms of fibroblastic origin ( 1 ). The underlying genetic cause of FAP is the mutation of the ADENOMATOUS POLYPOSIS COLI ( APC ) gene ( 2–4 ) located on 5q21. The APC protein is a component of the wnt signalling pathway that controls the proliferation of the epithelial cells of the gastrointestinal tract ( 5–9 ). The transcription factor β-catenin is the key effector of the pathway ( 10 ), stimulating proliferation through modulating the expression of specific target genes. In the absence of an extracellular proliferative signal, β-catenin enters a destruction complex where it is tagged for proteasomal degradation ( 11 ). APC displays binding sites for β-catenin and axin/conductin, the so-called 15 and 20 amino acid (15R and 20R) ( 12 ) and SAMP repeats ( 13 ), respectively. Axin/conductin recruit casein kinase I alpha (CK1α), that catalyses a priming phosphorylation event on Ser45 of β-catenin, and glycogen synthase kinase 3 beta (GSK3β) that subsequently promotes the sequential phosphorylation of Ser41, Ser37 and Ser33. The two latter phosphorylated residues constitute a recognition signal for the ubiquitination of β-catenin which is subsequently degraded in the proteasome. In FAP, β-catenin is stabilized constitutively ( 14 ), providing a permanent mitogenic signal to normally resting cells. This occurs when the second APC allele is inactivated somatically ( 15 , 16 ). Colorectal cancer develops also in a sporadic manner, involving somatic inactivation of both APC alleles in about two-third (if not more) of reported cases ( 17–20 ). It is thought that colon cancers in the context of FAP or sporadic settings follow the same genetic pathway, the former one occurring just ‘one hit quicker’ ( 21 ).
Germline mutations in APC are either nucleotide substitutions creating non-sense codons or small deletions/insertions leading to frame-shifts. Somatic mutations include, in addition, loss of heterozygosity (LOH) ( 17–19 ). Somatic recombination with the germline-mutated allele is the main mechanism of LOH ( 22 , 23 ). Importantly, somatic recombination is inherently much more frequent than any other genetic alteration ( 24 ). This implies that LOH happens when the first mutation is optimal and that a somatic point mutation reflects more likely the necessity of selecting an optimal product rather than a fortuitous event. As a consequence of these mutations, cells synthesize truncated APC products and this is thought to preclude the formation of an active destruction complex in addition to modifying the cytoskeleton dynamics ( 25–27 ). The two APC mutational events are interdependent, i.e. the position of the germline mutation in the open-reading frame determines where the somatic mutation will occur ( 28 , 29 ). As a first rule, the presence of the first, β-catenin binding, 20 amino acid repeat (20R1) is necessary for polyp development. Thus, when a germline mutation occurs before the 20R1, the second hit falls after it. Inversely, when the first hit occurs after the 20R1 but before the second 20 amino acid repeat (20R2), the second allele is removed by LOH in most cases, indicating that the presence of the 20R1 is optimal for tumour emergence. Interestingly, when the first mutation falls after the 20R2, the second hit is often a truncating mutation and the proportion of LOH decreases ( 20 , 28 , 30 , 31 ). As it was shown recently that the 20R2 cannot bind to β-catenin ( 32 ), this bias in the selection of APC truncations is apparently not related to the number of functional 20R. As a second rule, β-catenin binding to the 20R3 is prevented ( 17–19 ). Accordingly, there is a strong selection for both mutational hits to occur before or within the 20R3. When both rules are combined, one realizes that many mutations are located between the 20R1 and the 20R3, thereby defining the so-called mutation cluster region (MCR) of APC ( 17 ). Thus, the MCR reflects the selective pressure that leads to the retention of at least one truncated APC product containing a single functional 20R ( 32 ). The product of the second APC allele is either shorter or lost.
In this communication, we describe a new APC domain, located between the 20R2 and the 20R3, required for the efficient degradation of β-catenin. The activity of this domain can be inhibited by short APC isoforms not containing it. It provides a third rule and a key explanation for the apparent complex selection of APC mutations, not only in colorectal, but also in desmoid and duodenal tumours from FAP patients.
To investigate the influence of truncated APC on the transcriptional activity of β-catenin, we first compared three YFP-APC fusion constructs (Fig. 1 A). yAPC contains the whole open-reading frame, whereas yAPC1641 extends up to amino acid 1641 and therefore contains the first SAMP repeat. These constructs are able to assemble the β-catenin destruction complex ( 33 ) and were used as references for the inhibition of β-catenin activity resulting from proteasomal degradation. As expected, transient expression of yAPC and yAPC1641 in SW480 cells containing high amounts of β-catenin led to its efficient disappearance, which was not observed upon expression of YFP, the negative control (Fig. 1 B). In addition, we also designed yAPC1519, extending up to approximately the last third of the 20R3 and therefore lacking the SAMP repeat (Fig. 1 A). Surprisingly, β-catenin was also no longer detectable in cells expressing this construct (Fig. 1 B). The effect was efficiently inhibited by the GSK3β inhibitors LiCl and BIO that simultaneously induced the recruitment of the yAPC1641 and yAPC1519, but not yAPC, in cytoplasmic inclusions. β-Catenin was also frequently present in these structures and absent from the nucleus, indicating that APC can retain β-catenin in the cytoplasm, in accordance with the literature ( 34 ). When transfected into HCT116 cells in which β-catenin cannot be targeted for destruction due to a point mutation at Ser45 precluding its phosphorylation, all three yAPC constructs failed to alter the intensity of the β-catenin signal (Fig. 1 B). The β-catenin staining of HCT116 cells was specific, because it disappeared upon transient transfection of a siRNA against β-catenin (data not shown). Thus, we concluded that APC truncated at position 1519 can target β-catenin for degradation in SW480 cells and this likely depends on the classical mechanism via phosphorylation of β-catenin, because it does not occur in HCT116 cells and is inhibited by LiCl or BIO in SW480 cells.
To delineate more precisely the domain within yAPC1519 responsible for β-catenin degradation in SW480 cells, we performed a progressive deletion analysis, starting with yAPC1641 (Fig. 2 A). The localization and the expression level of the different truncated yAPC constructs were monitored by fluorescence microscopy and western blotting, whereas the β-catenin level and its associated transcriptional activity were visualized by immunofluorescence staining and by using the β-catenin-dependent TOP/FOP reporter assay (see Materials and Methods), respectively. The capacity to target β-catenin for degradation and inhibit its transcriptional activity, the expression level and the intracellular localization of yAPC1641, yAPC1558, yAPC1519, yAPC1502, yAPC1494, yAPC1493, yAPC1485 and yAPC1466 were similar. In sharp contrast, yAPC1404 could not target β-catenin for degradation anymore, although it was expressed at a similar level as the other constructs. It retained some inhibitory activity on β-catenin-mediated transcription, probably due to the presence of the 20R1, because yAPC1247 lacking this repeat failed to inhibit transcription (Fig. 2 A). In addition, the inactive yAPC1404 and shorter constructs localized as intracellular inclusions instead of the predominantly cytoplasmic diffuse pattern displayed by the longer constructs.
To define more precisely the minimal APC construct that would target β-catenin for degradation, we created the constructs yAPC1417 and yAC1430 (Fig. 2 B). When transiently expressed in SW480 cells, yAPC1417 could not target β-catenin for degradation significantly, although it retained some inhibitory activity on the transcriptional activity of β-catenin, similar to yAPC1404. This was again attributed to its β-catenin binding activity, because a construct spanning amino acids 1–1519 where all 15R and 20R had been mutated to destroy any possibility of β-catenin binding (yAPC1519-15Rµ-20R1µ-20R3µ) ( 32 ) could not inhibit the transcriptional activity of β-catenin. In contrast to yAPC1417, yAPC1430 had retained the capacity to target β-catenin for degradation, since 34% of transfected SW480 cells stained negative for β-catenin in immunofluorescence. yAPC1430 was also significantly more active than yAPC1417 in repressing β-catenin-dependent transcription. Both yAPC1430 and yAPC1417 were recruited into cytoplasmic inclusions. Thus, the C-terminus of the minimal APC construct competent for β-catenin degradation lies between amino acid residues 1417 and 1430, shortly after the 20R2.
Next, we compared the inhibitory activity of yAPC1466, yAPC1493 and yAPC1494 with that of yAPC1641 under limiting conditions, using the TOP/FOP reporter assay (Fig. 2 C). We observed that the inhibitory activity exerted by yAPC1641 on the transcriptional activity of β-catenin was progressively lost upon dilution of the transfected expression vector. Importantly, the loss of activity of yAPC1466, yAPC1493 and yAPC1494 paralleled that of yAPC1641, indicating that the truncated APC lacking the SAMP repeat are as efficient in inhibiting the transcriptional activity of β-catenin as a molecule containing it. To confirm this further, we deleted amino acids 1404–1466 from the yAPC1641 construct (Fig. 3 A). We found that this internal deletion abolished β-catenin degradation despite the presence of the SAMP repeat. The construct retained some inhibitory activity on β-catenin-dependent transcription probably due to the presence of the 20R1 and 20R3 which provide a high β-catenin binding activity. The same observations were made when the fragment 1404–1466 was deleted from the yAPC1519 construct (Fig. 3 B). Interestingly, the removal of amino acids 1404–1466 from yAPC1641 and yAPC1519 drastically affected the intracellular localization of the resulting constructs, which were concentrating now in cytoplasmic inclusions (Fig. 3 A and B), similarly to yAPC1404, yAPC1289 and yAPC1247 (Fig. 2 A). This opened the possibility that truncated APC shorter than 1466 might have kept some intrinsic β-catenin degradation activity, masked by the induction of the inclusion phenotype. To answer this, we compared constructs containing and lacking the 1404–1466 fragment in a background where the whole N-terminal part of APC from amino acids 1–957 had been deleted (Fig. 3 C). In the absence of the N-terminal part, β-catenin degradation was somewhat less efficient. Again, deletion of the fragment 1404–1466 abolished β-catenin down-regulation. Importantly, none of the constructs displayed a dot-like phenotype, demonstrating that their loss of activity is independent of the induction of inclusions (Fig. 3 C). Therefore, the ability of truncated APC to target β-catenin for degradation requires the APC fragment spanning amino acids 1404–1466, to which we refer as the β-catenin inhibitory domain (CID) of APC.
The CID creates an apparent paradox. It is encoded within the MCR and thus present in many truncated APC fragments found in human tumours. We reasoned that the activity of the CID might be cell-specific and therefore we measured it in different colon cancer cell lines expressing different endogenous versions of truncated APC by comparing the efficiencies of yAPC1641, yAPC1519, yAPC1466, yAPC1404 and yAPC1247 in down-regulating β-catenin-dependent transcription. The experiment allowed assigning the different cell lines to three groups, depending on the position of the endogenous APC mutations and the activity of the CID in the exogenous constructs. The first group contains the DLD1, VACO4A and SW480 cell lines in which the endogenous APC mutations are located before the CID. In these cells, the CID is the main contributor to the down-regulation of β-catenin activity, since the APC activity is lost only upon its deletion (compare yAPC1466 and yAPC1404 in Fig. 4 ). This indicates that SW480, VACO4A and DLD1 cells cannot target β-catenin for degradation because the truncated products they express are lacking the CID. The second group contains SW837, SW1417, SKCO1, T84 and GP2D cell lines, in which the selection has led to the retention of at least one truncated APC product harbouring the CID. In these cells, yAPC1466 was poorly active, indicating that they are lacking or expressing at a low level a transacting factor conferring its activity to the CID. This explains why the APC mutations found in these cell lines have been selected despite their location downstream of the CID. Of note, in these cells, yAPC1519 containing the 20R3 displays a good β-catenin down-regulating activity. The situation in the third group is more complex. LoVo and SW948 cells express endogenous APC isoforms extending, respectively up to positions 1430 and 1429 that are intrinsically able to target β-catenin for degradation because they have retained part of the CID (Fig. 2 B and data not shown). It was expected that these cells would lack the transacting factor mediating the CID activity, but yAPC1466 was surprisingly still active (Fig. 4 ). Thus, LoVo and SW948 cells express an APC isoform that has been selected in the original tumour despite its ability to target β-catenin for degradation. Interestingly, LoVo and SW948 cells express also a shorter APC fragment extending up to position 1114. Therefore, we investigated whether short APC constructs lacking the CID would influence the activity of those containing it. yAPC1485 (containing the CID) was co-expressed together with either yAPC777 or yAPC1147 in SW480 cells (Fig. 5 A). yAPC1485 targeted β-catenin for degradation and inhibited its transcriptional activity efficiently, while yAPC777 and yAPC1147 did not. Remarkably, the co-expression of yAPC1147, but not yAPC777 interfered with the β-catenin degrading activity of yAPC1485, without decreasing its expression level. Similarly, larger constructs including yAPC1247, yAPC1289 and yAPC1404 (Fig. 5 B) interfered also with the activity of yAPC1466. There is some discrepancy between the TOP/FOP values and the percentage of β-catenin negative cells, but the two assays match in a qualitative way. Thus, truncated APC lacking the CID inhibit β-catenin degradation mediated by the CID. To define further which domains are required for the inhibition, we performed progressive deletions from the N-terminus of yAPC1404. This revealed that amino acids 1–205 were contributing significantly to the inhibition of yAPC1485 activity (Fig. 5 C). We concluded that an APC fragment spanning amino acid 1 to a position located between amino acids 777 and 1148 was necessary and sufficient to inhibit efficiently β-catenin degradation mediated by the CID.
Next, we determined which of the APC constructs targeting β-catenin for destruction could be inhibited by a shorter isoform lacking the CID. In this experiment, yAPC1247 was used as a test inhibitor of the β-catenin degrading activities of yAPC1466, yAPC1485, yAPC1493 and yAPC1641 (Fig. 6 A). Remarkably, with the exception of yAPC1641 whose activity was poorly affected, all other constructs were severely compromised in their ability to target β-catenin for degradation and to repress its transcriptional activity (Fig. 6 A). There is some discrepancy between the TOP/FOP values and the percentage of β-catenin negative cells, but the two assays match in a qualitative way. Further analysis revealed that yAPC1502 containing part of the 20R3 and yAPC1558 (data not shown) displaying the entire 20R3 were inhibited by yAPC1247 to a lesser extent than constructs lacking the 20R3 (Fig. 6 B). Thus, truncated APC containing the 20R3 or additionally the first SAMP repeat, which are rarely selected in human tumours, cannot be efficiently inhibited by an APC construct lacking the CID. Furthermore, an APC fragment as short as yAPC958-1466, our shortest construct targeting β-catenin for degradation, could also be efficiently inhibited by yAPC1289 or yAPC1147 (Fig. 6 C).
Next, we collected data from the literature where the sequence of APC mutational events occurring in tumours from FAP patients had been determined. These data were re-evaluated, taking into account the CID and the fact that it can be inhibited by APC fragments lacking it (Fig. 7 ). The first condition for the successful selection of an APC mutation in a human colorectal tumour is the presence of the 20R1 ( 20 , 28 , 30 , 31 ). Accordingly, when the first mutational hit falls before amino acid 1270 (before the 20R1), the second hit occurs almost systematically after position 1284, indicating that the 20R1 is necessary for the outgrowth of a tumour (Fig. 7 A). Inversely, when the first hit leaves an intact 20R1 but removes the CID at the same time (between positions 1284 and 1417), the second allele is lost in most cases by LOH, indicating that the presence of the 20R1 is also sufficient for tumour outgrowth. Interestingly, when the first hit falls before position 777 and therefore generates a fragment lacking the interfering activity, there is a clear tendency for the second hit to occur preferentially between position 1284 and 1417, that is, a tendency to avoid the presence of the CID while satisfying the necessary condition of keeping the 20R1. However, when the first hit falls between positions 777 and 1270 and therefore may allow the synthesis of an inhibitory APC fragment, the second hit falls again after the 20R1 (after position 1284), but mutations leading in addition to the retention of the CID or part of it (after position 1429) are more represented in proportion, relative to the former situation with a first hit occurring before position 777. On the other hand, when the first mutation keeps the CID (after position 1429), most second hits occur between position 778 and 1417, leading possibly to the synthesis of inhibitory fragments. A significant proportion of LOH is also tolerated in this case, suggesting that the loss of activity of the CID as seen in the second group of cell lines (Fig. 4 ) might be also a frequent event during tumoural progression. This may also apply to tumours selecting the presence of the CID even when the first mutation falls before position 777. It should be noted that tumours with a first hit falling after position 1429 are rare (31/319). In comparison to a first mutation occurring between position 1284 and 1417 and whose selection requires only LOH as a second hit, the selection of a first hit falling after position 1429 would require either a truncating mutation in the second allele which is less frequent than LOH or, LOH accompanied in addition by the inactivation of the protein mediating the effects of the CID, the two situations being obviously less frequent than just LOH. We conclude that the CID and the possibility to inhibit its activity provide a satisfying explanation for the interdependent selection of both APC mutations in colorectal tumours from FAP patients.
We extended our observations to duodenal and desmoid tumours that may also develop in FAP patients as a consequence of APC mutations (Fig. 7 B and C). In these tumours, when the germline APC mutation occurs before the CID (before position 1417), the somatic hit leads almost systematically to the retention of the CID. Inversely, when the CID is already present in the truncated APC resulting from the first mutation (between positions 1429 and 1564), the second allele is lost in most cases. Thus, in tumours of the upper gastrointestinal tract and desmoid tumours, the presence of the CID appears necessary and sufficient for tumour development.
We also compared the spectrum of APC mutations in FAP patients to a panel of mutations found in sporadic colorectal tumours (Fig. 8 ). Although the sequence of APC mutations in sporadic tumours is not known, we found that both APC mutational hits were combining in a similar manner as in FAP tumours. When a hit in one allele occurs before position 777, most hits in the other allele have the tendency to fall between the 20R1 and the CID, rather than after position 1429. In contrast, when one hit occurs between positions 777 and 1270, most second hits allow the presence of the CID. Thus, colorectal FAP and sporadic tumours respect the same rule of selection related to the presence or the absence of the CID. However, an important difference must be noted. Sporadic tumours tolerate a higher proportion of LOH, independently of the location of the first hit. This suggests that the first APC mutation in FAP patients restricts the number of potential genetic pathways leading to tumour development, because they need to be affected in the correct sequence. It implies that unknown tumourigenic events may well occur before the first APC mutation.
The present study describes the identification of a new APC domain, the inhibitory domain of APC (CID) that contributes to a large extent to APC-mediated degradation of β-catenin. The CID is retained in many truncated APCs from tumours of FAP patients and can be inhibited by shorter APC fragments lacking it. It is an important criterion for the interdependent selection of the two APC mutational hits occurring in colorectal, desmoid and upper gastrointestinal tumours from FAP patients. The CID also suggests that desmoid and upper gastrointestinal tumours keep some residual control on β-catenin transcriptional activity through a degradation-dependent mechanism. This type of control is counter-selected in colorectal tumours that rather rely only on β-catenin binding to the 20R1 of APC to modulate the transcriptional activity of β-catenin.
Functional characterization of the CID
From the modular structure of the central domain of APC consisting of β-catenin binding sites intermingled between axin/conductin binding sites, it is thought that APC requires association with both β-catenin and axin/conductin to fulfill its function in degradation of β-catenin. It therefore came as a surprise to us that APC fragments lacking all SAMP repeats such as yAPC1519 could target β-catenin for destruction as efficiently as a construct containing a SAMP repeat, i.e. yAPC1641 (Fig. 2 C). We cannot formally exclude a potential influence of endogenous truncated APC in our experiments, but our data are in line with previous observations highlighting the APC sequences involved in β-catenin inhibition. Accordingly, a first study described an internal APC fragment spanning amino acids 1034–1554 that could reduce the amount of endogenous β-catenin when over-expressed in SW480 cells ( 35 ). Another study demonstrated that over-expression of APC truncated at position 1556 was dramatically reducing the β-catenin level and associated transcriptional activity in SW480 cells ( 36 ). A third study showed that APC truncated at position 1465 was inhibiting efficiently the transcriptional activity of β-catenin in SW948 cells, but the authors did not report about the level of β-catenin ( 37 ). An internal deletion analysis of APC showed that amino acids 1404–1466 are essential for degradation of β-catenin by both SAMP-lacking (yAPC1519) and SAMP-containing (yAPC1641) APC fragments. Progressive deletion from the C-terminus allowed us to set the N-terminal border for this activity between residues 1417 and 1429. Thus, we define the CID as the APC sequence located between the 20R2 and the 20R3. The CID is operating in a cell line-specific manner, indicating that a cell-specific cofactor is conferring its activity to the CID. It is possible that this cofactor binds directly to the CID, inducing phosphorylation and degradation of β-catenin. Alternatively, deletion of the CID may modify the APC conformation, altering indirectly the binding of the cofactor somewhere else in the APC sequence.
Deletion of the CID provokes a drastic change of the intracellular localization of truncated APC, inducing its recruitment into cytoplasmic inclusions at the expense of a diffuse cytoplasmic pattern. These inclusions have already been described previously in cells ectopically expressing truncated APCs ( 38 , 39 ) and may also occur at low APC concentrations, because a dot-like pattern is also seen when endogenous APC from SW480 cells is detected with an anti-APC antibody ( 40 ). Inhibition of GSK3β with either LiCl or BIO also led to the formation of inclusions by constructs containing the CID, indicating that they are not simply aggregates resulting from overexpression, and might be regulated by phosphorylation. The formation of inclusions requires the N-terminus and sequences between positions 777 and 1147. Therefore, the minimal aggregating APC fragment contains the N-terminal dimerization domain of APC ( 41 ) as well as the recently described N3 domain (residues 782–1018) that can also homodimerize ( 42 ). We suggest that the formation of inclusions may involve polymerization, similarly to what has been shown recently for Dishevelled ( 43 ), and that it is prevented by the CID. It was possible that formation of intracellular inclusions upon deletion of the CID might have inactivated the APC fragments, thus explaining their loss of activity on β-catenin degradation. However, APC constructs lacking amino acids 1–957, such as yAPC958-1466 and yAPC958-1404, cannot be incorporated into inclusions. Yet the former can target β-catenin for degradation, whereas the latter that has lost the CID cannot, which indicates that the CID itself provides a β-catenin degrading function to APC.
Inhibition of the CID
The β-catenin degrading activity of truncated APC constructs containing the CID is counteracted by shorter versions of APC lacking it. The inhibition occurs as soon as the CID is deleted, as exemplified by yAPC1404 (Fig. 5 B) and yAPC1417 (data not shown). The inhibitory activity is not present in yAPC777 and also lost when the N-terminus is deleted (as in yAPC205-1404), suggesting that the sequences required for an efficient inhibitory activity on β-catenin degradation are contained within the N-terminal domain and between amino acids 777 and 1147. The inhibiting constructs can recruit all N-terminus-containing, β-catenin-degrading APC fragments into the inclusions (data not shown), which could represent a potential mechanism for the inhibition. However yAPC958-1466 cannot be recruited into the inclusions but its β-catenin degrading activity is strongly inhibited by constructs building inclusions (Fig. 6 C), which is speaking against recruitment of an active APC construct into the dots as the mechanism of inhibition. However, the inclusions may well contribute to the inhibition of β-catenin degradation in an indirect manner.
A model for the selection of APC mutations
The characteristics of the CID explain the interdependent distribution of the two APC mutational hits observed in tumours from FAP patients (Fig. 7 ). Thus, a first APC mutation occurring after the CID is followed in most cases by a second truncating mutation leading to the loss of the CID. According to our data, the short fragments would block the β-catenin degradation activity of the longer ones, which would relieve the selective pressure to delete the CID. In line, more CID-containing fragments are allowed as a second hit when the first hit generates potentially interfering fragments rather than shorter pieces lacking the interfering activity. For the remaining cases in which the CID domain is present but no interfering fragments are formed from the second allele, we have to assume that most of the CID activity is lost, for instance through down-regulation of the cofactor required for the CID activity. Such a scenario is exemplified by the second group of colorectal carcinoma cell lines (Fig. 4 ), where the CID is poorly active in inhibiting β-catenin-dependent transcription. A similar combination of mutations at the APC alleles is also seen in sporadic colorectal carcinomas. Thus, the CID is a direct target of the selective pressure imposed to the cells. The three different possibilities of inactivating it (i.e. deletion of CID, loss of CID activity and generation of an interfering fragment) explain to a large extent the complex distribution of APC mutations in colorectal tumours.
In desmoid and duodenal tumours, the selection leads to the retention of the CID in truncated APC. Accordingly, a first hit falling before the CID is almost always followed by a second hit after it. There is apparently no selective pressure for the synthesis of an inhibitory fragment in these tumours because a first hit occurring after the CID is almost systematically followed by loss of the second allele. Thus, the pattern of APC mutations distribution indicates that the CID is also a direct target of the selective pressure in desmoid and duodenal tumours, but in contrast to colorectal tumours it is required for tumour outgrowth. The two different kinds of selective pressure for APC mutations in colorectal as compared with desmoid and duodenal tumours reflect two different molecular mechanisms by which the activity of β-catenin may be partially kept under control despite the primary alteration of the degradation complex. Colorectal tumours rely on β-catenin binding ( 38 , 39 ), and degradation is counter selected, while desmoid and duodenal tumours may retain some ability to target β-catenin for degradation.
In colorectal tumours, APC truncations are almost systematically selected for the absence of the 20R3, i.e. up to position 1493 ( 32 ). FAP tumours expressing a 20R3-containing APC isoform, i.e. from position 1494 (actually 10/319 cases), are less frequent than tumours in which at least one APC hit is located between positions 1429 to 1493 (79/319). APC constructs containing the 20R3 are not as efficiently inhibited by a fragment lacking the CID as constructs lacking the 20R3 (Fig. 6 B and data not shown). In addition, constructs containing the 20R3 still display a considerable β-catenin down-regulating activity in the second group of cell lines, where the CID is poorly active (Fig. 4 ). These observations suggest that the selection excludes the presence of the 20R3 because its very high affinity towards β-catenin ( 44 ) would render inefficient the inhibition of β-catenin degradation achieved by either a fragment lacking the CID or a reduction of the activity associated with the CID.
Interestingly, desmoid and duodenal tumours tolerate a high proportion of 20R3-containing isoforms as the result of a somatic mutation (4/9 and 10/18 cases for desmoid and duodenal tumours, respectively), and this occurs only when a first hit has led to the synthesis of an inhibitory fragment (from positions 778 to 1417). This suggests that the too strong β-catenin down-regulating activity provided by the presence of the 20R3 might be tolerated because it would be counteracted by the shorter fragments. As a result, this would be equivalent to the situation where a first hit falling between 1429 and 1493 is followed by LOH. The confirmation of this hypothesis requires however a larger sampling of desmoid and duodenal tumours.
The identification of the CID allows complementing the panel of rules governing the selection of APC mutations in colorectal tumours. The CID must be removed or inactivated, the 20R1 must be kept and the 20R3 excluded. Most tumours respect these rules (303/325, 93%). The few exceptions include the tumours where the SAMP repeat is retained (6/325, 2%) and others where the 20R1 is lacking (16/325, 5%). Altogether, our data provide an appealing, because unifying, model of APC mutations selection in tumours from FAP patients. Two important issues follow: identifying the protein mediating the CID-dependent degradation of β-catenin and deciphering the mechanism leading to its inhibition.
MATERIALS AND METHODS
BIO and LiCl were from Sigma (Taufkirchen, Germany) and Merck (Darmstadt, Germany), respectively.
HCT116, SW480, T84, LoVo, DLD1, SKCO1, SW837, SW948, SW1417, VACO4A and GP2D colorectal cancer cell lines were all maintained in DMEM medium (PAA Laboratories, Cölbe, Germany) supplemented with 10% fetal calf serum (Perbio Laboratories, Frankfurt am Main, Germany) and 1% penicillin and 1% streptomycin (PAA Laboratories). APC mutations specific to each cell line are described in ref. ( 29 ).
The rabbit antibody H102 against β-catenin was purchased from Santa Cruz (Heidelberg, Germany). Secondary antibodies coupled to either horseradish peroxidase or Cy3 were from Dianova (Hamburg, Germany), the anti-β-actin from Sigma and the anti-GFP from Roche (Mannheim, Germany).
The plasmids expressing the YFP-APC fusion proteins were constructed by standard molecular biology methods, using pCMV-APC and YFP-APC ( 34 ) as template for PCR reactions and recipient vector, respectively. The sequence of any of these plasmids is available upon request. The constructs yAPC1417 and yAPC1430 reproduce the mutations seen in the DLD1 and LoVo cell lines, respectively ( 29 ). pcDNAflag ( 13 ) was used as a control vector.
Plasmids were transfected into cells overnight using 5 µl polyethylenimine (1 mg/ml) per microgram of DNA. For transient transfection of plasmids, 2 µg total DNA/200 000 cells/35 mm dish were used. When two yAPC constructs were cotransfected, the ratios of the β-catenin down-regulating construct to the inhibitory construct were 3:1 for subsequent western blotting and 1:5 for the reporter assays. Expression was allowed to proceed for 36 h for subsequent western blotting or 24 h for immunofluorescence.
TOP/FOP reporter assays
TOP/FOP reporter assays were conducted as described elsewhere ( 45 ). The TOPglow reporter consists of a tandem repeat of four TCF/LEF1 (T-cell factor/lymphoid enhancer factor 1) binding sites inserted in front of a TATA box ( 46 ), driving the expression of luciferase in a β-catenin-dependent manner. In the FOPglow reporter, the four binding sites are mutated to abolish the binding of TCF/LEF1. The internal control pUHD16.1 encoding the β-galactosidase was transiently transfected together with either FOPglow or TOPglow plasmids at an equimolar ratio (500 ng each). The total amount of other co-transfected constructs was 1 µg unless otherwise specified. The transcriptional activity measured 20 h post-transfection is defined as the ratio of TOPglow and FOPglow luciferase values normalized to the β-galactosidase values. FOPglow values were not significantly affected by the expression of any YFP-APC construct.
Western blotting was performed according to Behrens et al . ( 13 ). The blots were developed using the chemiluminescence reagents Western Lightning™ (Perkin Elmer Life Sciences, Boston, MA) and the signals were detected under a LAS-3000-Fuji camera from Raytest (Straubenhardt, Germany).
Immunofluorescence staining was conducted as described ( 13 ).
Several tumours, in which the precise site of APC mutation was unknown or unusual, including translocations and triple mutations, were not taken into account. Therefore, patients 623.iii.3 [3 tumours ( 31 )], Ds2, Ds7 and Ds11 [3 tumours ( 47 )], PLK214 and PLK253 [6 tumours ( 20 )], were excluded, as well as Patient 4 and all underlined tumours reported in ( 30 ).
This study was supported by a grant from the Wilhelm-Sander-Stiftung to J.B.
We thank I. Tomlinson and A. Rowan for providing cell lines, G. Daum for technical, and A. Doebler for secretarial assistances.
Conflict of Interest statement. None declared.