Role of APAF-1, E-cadherin and peritumoural lymphocytic infiltration in tumour budding in colorectal cancer


  • No conflicts of interest were declared.


Tumour budding or dedifferentiation at the invasive margin of colorectal cancer (CRC) is an important prognostic marker and linked mechanistically to dysregulation of Wnt pathway signalling. Since budding is observed in only 40% of CRCs, we hypothesized that Wnt pathway dysregulation may be a necessary but insufficient explanation for budding and that buds may be destroyed selectively by tumour immune mechanisms. Twenty potential markers of tumour budding were evaluated in tissue microarrays (TMAs) obtained from the main tumour body of 1164 DNA mismatch repair-proficient CRCs and the findings were correlated with tumour budding, lymphocytic infiltration and survival. Loss of expression of E-cadherin and APAF-1 were independent predictors of budding (sensitivity 70.3% and specificity 48.2% when one or the other was lost). Peritumoural lymphocytes (PTLs) were observed more frequently in CRCs with loss of either E-cadherin or APAF-1 that were budding-negative. PTLs and tumour-infiltrating lymphocytes (TILs) were strongly correlated. The absence of TILs increased the adverse prognostic impact of E-cadherin and APAF-1 loss. Co-occurrence of E-cadherin loss, APAF-1 loss and low TIL counts in CRCs was an independent prognostic factor. The findings were verified in whole tissue sections from 88 CRCs with known KRAS mutation status (which was not associated with budding). Loss of E-cadherin and APAF-1 within the main body of CRCs are independent predictors of tumour budding. The prognostic benefit of lymphocytic infiltration may be explained by the immune destruction of budding cells. Copyright © 2007 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.


Morphological findings at the invasive tumour margin provide independent prognostic information in colorectal cancer (CRC) 1–3. A diffusely infiltrative margin is characterized by widespread dissection of normal tissue structures with loss of a clear boundary between tumour and host tissue. Diffuse infiltration is also associated with a feature described as tumour budding or dedifferentiation, in which there is a transition from glandular structures to single cells or clusters of up to four cells at the invasive margin 3. Tumour budding should be distinguished from a subclone showing poor differentiation by virtue of its linear distribution along the entire invasive margin. Budding cells have been credited with the properties of malignant stem cells, including the potential for redifferentiation both locally and at sites of metastasis 4. This suggests that the morphological and immunophenotypic features associated with budding cells are dynamic and reversible and likely to be under epigenetic control. Indeed, tumour budding has been compared with the process of gastrulation (differentiation of three germ layers, patterning and cell migration) that occurs during normal embryogenesis 5, 6.

A critical molecular event in both tumour budding and gastrulation is Wnt pathway activation, leading to nuclear translocation of β-catenin and activation of the transcription factor Tcf 5, 6. This underlies the differential gene expression pattern observed in budding cells, including the increased expression of laminin 5-γ2 7, matrix metalloproteinase-7 (MMP-7 or matrilysin) 8, membrane type-1 MMP 8, p16 9, cyclin D1 10, urokinase-like plasminogen activator receptor 11, CD44 11, COX-2 11 and tenascin-C 12, and reduced expression of E-cadherin 5, 13 and Cdx-2 14. Budding cells also show evidence of autonomous movement characterized by the presence of podia 15 that express P-glycoprotein at points of attachment to mesenchymal elements 16. Mesenchymal markers including vimentin and fibronectin are also expressed 5 and tumour budding resembles epithelial to mesenchymal transition in other cancer model systems 12.

While tumour budding occurs in CRC cells primed by APC inactivation, the process is largely restricted to the interface between malignant epithelium and stromal elements at the invasive margin. This suggests that the budding phenotype is triggered through an increased sensitivity to mesenchymally-derived growth signals 14. For example, myofibroblasts may be activated by PGE2 or TGFβ secreted by tumour cells and then complete a paracrine loop by secreting amphiregulin and neuregulin, which are ligands for EGFR (ErbB-1) and ErbB-3 respectively 17. EGFR down-regulates E-cadherin and drives Wnt pathway signalling through the transcriptional repressor Slug 18. Alternatively, FGF may down-regulate E-cadherin through the transcriptional repressor Snail 19.

Although up to 90% of CRCs are characterized by dysregulation of the Wnt signalling pathway 20, only about 40% show tumour budding 3. We hypothesized that full expression of the budding immunophenotype might require synergizing alterations that could be identified in the main body of the tumour. Additionally, we hypothesized that the absence of budding might be explained by the immune destruction of budding cells. We have previously noted an inverse relation between tumour budding and peritumoural lymphocytic infiltration (PTL) 15 and have suggested that the marked lymphocyte reaction in microsatellite instability-high (MSI-H) CRCs might be related to the relative absence of budding in this subset 21. Tumour budding is uncommon in MSI-H CRCs 9, 22 and lacks the full budding immunophenotype 23. For example, budding cells in sporadic MSI-H CRCs do not show increased expression of β-catenin or laminin5-γ2 and lack the development of cytoplasmic podia 23. In this study we excluded CRCs that showed loss of expression of a DNA mismatch repair proteins. The study was based on 1164 tissue microarray (TMA) samples that were subdivided into two matched sets, one for the development of a predictive model and a second for its subsequent validation. Budding-associated markers are likely to be relatively subtly altered in TMA samples obtained from the main body of the tumour. Therefore, optimum scoring cut-offs were determined with the aid of receiver operating characteristic (ROC) curves. The scores were used to establish correlations with budding in both matched sets and the findings were validated in whole tissue sections.


The study was approved by the Institutional Review Board of McGill University.

TMA construction

A TMA of 1420 unselected, non-consecutive CRCs was constructed 24. Briefly, formalin-fixed, paraffin-embedded tissue blocks of CRC resections were obtained. One tissue cylinder with a diameter of 0.6 mm was punched from morphologically representative tissue areas of each donor tissue block and brought into one recipient paraffin block (3 × 2.5 cm), using a home-made semi-automated tissue arrayer.

Immunohistochemistry (IHC)

The 1420 CRCs were dewaxed and rehydrated in distilled water. Endogenous peroxidase activity was blocked using 0.5% H2O2. Following pressure cooker-mediated antigen retrieval in 0.001 M ethylenediaminetetraacetic acid (EDTA), pH 8.0, the sections were incubated with 10% normal goat serum (Dako Cytomation, Carpinteria, CA) for 20 min. In order to determine mismatch-repair (MMR) status, the 1420 CRCs were incubated with primary antibody for MLH1 (MLH1 clone MLH-1; BD Biosciences Pharmingen, San Jose, CA), MSH2 (clone MSH-2, BD Biosciences Pharmingen, San Jose, CA), and MSH6 (clone 44; Transduction Laboratories) at dilutions of 1/100 for 2 h at room temperature. Subsequently, the sections were incubated with HRP-conjugated secondary antibody (DakoCytomation) for 30 min at room temperature. For visualization of the antigen, the sections were immersed in 3-amino-9-ethylcarbazole + substrate-chromogen (DakoCytomation) for 30 min, and counterstained with Gill's haematoxylin. IHC was performed similarly for a panel of 20 markers, as outlined in Table 1. The markers were chosen because of their known association with budding or with other potentially relevant mechanisms (differentiation, proliferation, apoptosis and cell adhesion). In addition, CD8 was used as a marker for tumour-infiltrating lymphocytes (TILs).

Table 1. Tumour markers and antibodies used for immunohistochemical analysis of 1164 CRCs
APAF-1Pro-apoptotic protein with tumour suppressing functionNCL APAF-11 : 40Novocastra, Newcastle, UK
APCTumour suppressor and promoter of β-catenin degradationC201 : 100Santa Cruz, CA, USA
β-CateninTumour promoter and mediator of WNT signallingB-catenin-11 : 100Dako Cytomation, Missassauga, Canada
Bcl-2Anti-apoptotic protein inhibiting release of cytochrome c from mitochondria1241 : 400Dako Cytomation, Carpinteria, USA
CD8Cytotoxic T cell markerC8\144B1 : 100Dako Cytomation, Carpinteria, USA
CD44sCell adhesion moleculeDF14851 : 50Dako Cytomation, Carpinteria, USA
E-cadherinIntercellular adhesion moleculeNCH-381 : 100Dako Cytomation, Grostrup, Denmark
EphB2Tyrosine kinase receptor involved in deregulation of cell–cell interactionsAF4671 : 200R&D Systems, Minneapolis, USA
Her2/neuTyrosine kinase receptor involved in cell proliferation and survivalPN2A1 : 100Dako Cytomation, Grostrup, Denmark
MST1Pro-apoptotic proteinPolyclonal1 : 200Cell Signaling, Danvers, USA
MUC1Involved in cell adhesion, signal transduction, maintenance of cellular polarity139H21 : 100Cedarlane Laboratories, Hornby, Canada
MUC2Producer of gel-forming mucin specific to goblet cellsCcp581 : 100Cedarlane Laboratories, Hornby, Canada
p21Cell-cycle arrest mediatorSX1181 : 20Novocastra, Newcastle, UK
p27Inhibitor of cyclin-dependent kinasesSX53G81 : 100Dako Cytomation, Carpinteria, USA
p53Tumour suppressor involved in cell cycle arrest, apoptosis, repair of DNA damage and angiogenesisDO-71 : 200Dako Cytomation, Gostrup, Denmark
pAKTInvolved in PI3-K signalling244F91 : 100Cell Signalling, Danvers, USA
pERKMAP kinase downstream of RAS20G111 : 100Cell Signalling, Danvers, USA
pSMAD2TGFβ signalling moleculePolyclonal1 : 500Biocare Medical, CA, USA
RHAMMMember of MAP kinase pathway involved in cell motility2D61 : 100Novocastra, Newcastle, UK
RKIPDown-regulator of MAP kinase signalling and marker of metastasisPolyclonal1 : 1000Upstate, New York, USA
TGF-βGrowth factor with both tumour suppressing and promoting functionsTB211 : 1000AbCam, Cambridge, MA, USA

Evaluation of IHC

Immunoreactivity was assessed for each protein in a semi-quantitative manner by scoring the proportion of positive tumour cells over the total number of tumour cells in the range 0–100%. Scores were given using 5% increments (0%, 5%, 10%, etc.). The staining intensity was not assessed. This scoring method has been evaluated previously for four of the markers used in the present study 25. Since observer reproducibility is always imperfect, the analysis of IHC scoring was performed on two matched datasets (see below). MLH1, MSH2 and MSH6 were scored as negative (0% staining) or positive (>0% staining).

DNA mismatch repair (MMR) status

The 1420 CRCs were stratified according to DNA MMR status and consisted of 1197 MMR-proficient tumours expressing MLH1, MSH2 and MSH6, 141 MLH1-negative tumours, and 82 presumed Lynch syndrome/HNPCC cases demonstrating loss of MSH2 and/or MSH6 at any age, or loss of MLH1 at < 55 years 26. Only MMR-proficient cases with information on tumour budding were included in this study (n = 1164).

Clinico-pathological data

The clinico-pathological data for the CRC patients included T stage, N stage, tumour grade, vascular invasion and survival. The distribution of these features is described elsewhere 27. PTL presence at the invasive front and tumour budding were evaluated in the original haematoxylin and eosin (H&E)-stained sections from the resection specimens corresponding with each TMA punch. Based on this independent assessment, CRCs were scored as positive for tumour budding if there was a poorly circumscribed invasive margin due to the presence of infiltrating single glands, small clusters of cells or single cells (Figure 1A). The number of cells in clusters and the number of clusters was not scored and the definition used in this study was therefore broad and unlikely to miss CRCs with the budding phenotype.

Figure 1.

(a) Tumour budding at the invasive margin of a colorectal cancer, characterized by small clusters of malignant cells that have separated from the glands above and are located within a loose stroma lacking peritumoural lymphocytes. (b) Marked peritumoural lymphocytic reaction at invasive margin with no tumour budding. Left malignant gland shows small numbers of intra-epithelial lymphocytes or TILs (H&E-stained sections)

Randomization of MMR-proficient CRCs

The 1164 MMR-proficient CRCs were randomly assigned into two groups, Study Group 1 (n = 582) and Study Group 2 (n = 582), matched on sex, age, tumour location, T stage, N stage, tumour grade, vascular invasion and survival time.

Selecting cut-off scores for protein expression

The selection of cut-off scores for protein expression was based on ROC curve analysis and performed on Study Group 1 28. At each percentage score for a given protein, the sensitivity and specificity for discrimination of budding versus non-budding was plotted, thus generating a ROC curve. The score located closest to the point with both maximum sensitivity and specificity, i.e. the point (0.0, 1.0) on the curve, was selected as the cut-off score leading to the greatest number of tumours correctly classified as having budding or not. In addition, 100 bootstrapped replications were performed to re-sample the data and determine the reliability of the cut-off score for each protein. With bootstrapping, 100 re-samples of equal size were created and ROC curve analysis was performed for each subgroup. The most frequently obtained cut-off score (mode) over the 100 re-samples and the area under the ROC curve (AUC) and 95% CI were obtained for each analysis. AUCs summarize the discriminatory power of the protein for budding over the entire range of scores, with values of 0.5 indicating low power and those closer to 1.0 higher power.

Whole tissue sections

Surgical resections from 88 MMR-proficient CRCs, including 44 cases with tumour budding and 44 cases with no tumour budding, were selected. The clinico-pathological data for these cases is described elsewhere 15. IHC was performed on these cases for markers identified as independent predictors of budding in multivariate analysis in TMA Study Groups 1 and 2. Protein expression was scored as positive or negative by consensus and according to the previously determined cut-off scores by two independent pathologists. H&E slides were used to evaluate the presence of PTLs at the invasive front (present or absent) (Figure 1B) and the number of TILs in 10 × 40 fields of H&E sections. A tumour was considered negative for TILs when the median number within 10 × 40 fields was 0.

Since the expression of some of the markers assessed in this study (notably pERK) could be explained by oncogenic activation of KRAS, KRAS mutation in codons 12 and 13 was assessed in the CRCs from which the whole tissue sections were derived, as described previously 29. Briefly, fragments containing codons 12 and 13 in exon 1 were amplified using the following primers: KRAS forward 5′-GCCTGCTGAAAATGACTGAA-3′ and KRAS reverse 5′-AGAATGGTCCTGCACCAGTAA-3′, generating a 167 bp fragment. The PCR reaction was carried out in a total volume of 25 µl in a Biometra TGradient Cycler. The PCR reaction mixture contained 10× PCR Gold Buffer, 25 mM MgCl2 solution, 0.2 mM of each dNTP, 20 µ M each of forward and reverse primer, 2.5 U AmpliTaq Gold DNA Polymerase (Applied Biosystems, Branchburg, USA) and 2 µl DNA template. Amplification was carried out with 5 min of initial denaturation at 95 °C, followed by 35 cycles of denaturation at 95 °C for 40 s, primer annealing at 57 °C for 30 s and extension at 72 °C for 30 s. The final extension was performed at 72 °C for 6 min. To enhance the concentration of PCR product, the same PCR was carried out again using 2 µl of the previous PCR product. Samples were analysed on an 8% polyacrylamide gel and submitted to Genome Quebec (Montreal, Canada) for automated sequencing.

Association of protein expression with tumour budding and survival analysis

The expression of each protein was dichotomized around its cut-off score and its association with tumour budding was analysed on both Study Groups 1 and 2 by means of logistic regression. The p values, odds ratio (OR) and 95% CI for each analysis were obtained. All variables significant (p < 0.05) in univariate analysis were entered into a multivariate logistic regression model. A selection procedure was used to identify the independent predictors of tumour budding. The reliability of the model was established by 1000 bootstrapped replications of the data. The most frequently selected model from the 1000 re-samples was chosen as the final predictive model. Testing of the sensitivity and specificity of the model was carried out using 100-fold cross-validation. The association between TILs, PTLs and budding was evaluated using the χ2 test. The log-rank test (univariate analysis) and Cox proportional hazards regression (multivariate analysis) were used to analyse survival time for different combinations of markers. All analyses were carried out using SAS (Version 9.1, The SAS Institute, Cary, NC, USA).


Selection of cut-off scores

Tumour budding was present in 734 (63.0%) cases. ROC curve analysis was performed with tumour budding as the endpoint. The relevant cut-off scores for all proteins are listed in Table 2. The scores and tissue distributions were found to reflect the known biological function of each protein. A tumour with ⩽ four CD8+ TILs per TMA punch was considered negative for this feature.

Table 2. Cut-off scores, AUC (95% CI), and univariate analysis of tumour markers with respect to their association with budding in Study Groups 1 and 2
ProteinCut-off (%)AUC (95% CI)Study Group 1Study Group 2
p ValueOR (95% CI)p ValueOR (95% CI)
  • *

    Tumours entered into multivariate analysis.

  • **

    Only 8% of TMA cancer samples showed complete loss of expression of APC suggesting that the antibody used may give non-specific reactivity in formalin-fixed tissues.

    Decreased or increased expression is denoted by an odds ratio < 1 or > 1, respectively.

APAF-1900.58 (0.53–0.63)0.002*0.55 (0.38–0.79)< 0.001*0.49 (0.33–0.74)
APC**900.5 (0.45–0.56)0.029*0.65 (0.44–0.96)0.9351.02 (0.7–1.49)
β-Catenin50.55 (0.5–0.59)0.4841.14 (0.79–1.65)0.036*1.5 (1.03–2.19)
Bcl-2200.55 (0.5–0.6)0.002*0.57 (0.39–0.81)0.041*0.69 (0.48–0.99)
CD44s100.58 (0.53–0.63)0.021*0.67 (0.45–0.94)0.001*0.51 (0.35–0.75)
E-cadherin950.57 (0.51–0.62)0.009*0.57 (0.37–0.87)0.002*0.5 (0.32–0.77)
EphB2400.56 (0.51–0.62)0.009*0.59 (0.49–0.87)0.016*0.62 (0.42–0.92)
Her2/neu50.5 (0.45–0.55)0.4381.17 (0.78–1.75)0.9211.02 (0.69–1.51)
MST1800.55 (0.49–0.60)0.014*0.57 (0.37–0.89)0.047*0.66 (0.44–0.99)
MUC1100.53 (0.47–0.58)0.3580.84 (0.58–1.22)0.2441.24 (0.86–1.79)
MUC200.53 (0.47–0.58)0.1520.77 (0.54–1.1)0.2940.82 (0.57–1.19)
p2150.51 (0.46–0.56)< 0.001*0.44 (0.3–0.63)0.8251.04 (0.73–1.47)
p27850.51 (0.46–0.56)0.2310.8 (0.56–1.15)0.7120.93 (0.65–1.34)
p53300.52 (0.47–0.57)0.5271.12 (0.78–1.62)0.4051.17 (0.81–1.69)
pAKT100.51 (0.46–0.56)0.3921.17 (0.82–1.69)0.6390.92 (0.64–1.32)
pERK50.53 (0.48–0.58)0.045*1.49 (1.01–2.21)0.0891.43 (0.95–2.18)
pSMAD2600.55 (0.51–0.61)0.1190.74 (0.51–1.08)0.016*0.63 (0.43–0.92)
RHAMM1000.53 (0.47–0.58)0.271.24 (0.85–1.8)0.2891.23 (0.84–1.8)
RKIP900.57 (0.51–0.63)0.370.83 (0.55–1.25)0.004*0.54 (0.36–0.82)
TGFβ150.54 (0.49–0.6)0.032*0.67 (0.46–0.97)0.060.7 (0.49–1.02)

Univariate analysis (Table 2)

Of the 20 markers evaluated, 10 proteins were found to be significantly associated with tumour budding in Study Group 1, namely APAF-1 (p = 0.002), APC (p = 0.029), Bcl-2 (p = 0.002), CD44 (p = 0.021), E-cadherin (p = 0.009), EphB2 (p = 0.009), MST1 (p = 0.014), p21 (p < 0.001), pERK (p = 0.045) and TGFβ (p = 0.032)). Analysis of Study Group 2 identified important associations between tumour budding and APAF-1 (p < 0.001), β-catenin (p = 0.036), Bcl-2 (p = 0.041), CD44 (p = 0.001), E-cadherin (p = 0.002), EphB2 (p = 0.016), MST1 (p = 0.047), pSMAD2 (p = 0.016) and RKIP (p = 0.004). Decreased or increased expression of markers shown in Table 2 is denoted by odds ratios < 1 or > 1, respectively.

Multivariate analysis (Table 3)

In Study Group 1, E-cadherin (p = 0.015), APAF-1 (p = 0.016), p21 (p = 0.003) and Bcl-2 (p = 0.007) were found to independently predict tumour budding, while E-cadherin (p = 0.006) and APAF-1 (p < 0.001) had independent predictive value in Study Group 2. The independent markers common to both groups were analysed in all 1164 CRCs. The sensitivity and specificity of E-cadherin and APAF-1 for budding were 0.32, 0.80 and 0.65, 0.51, respectively. The resulting final predictive model for budding included decreased E-cadherin (p < 0.001; OR = 0.57 (95% CI = 0.41–0.78) and decreased APAF-1 (p < 0.001; OR = 0.55 (95% CI = 0.42–0.73). The cross-validated sensitivity and specificity of this model, combining loss of either marker, were 70.3% and 48.2%, respectively.

Table 3. Multivariate analysis of tumour budding Study Groups 1 and 2
Study Group p ValueOR (95% CI)
1E-cadherin0.0150.51 (0.3–0.88)
 APAF-10.0160.51 (0.3–0.88)
 p210.0030.48 (0.29–0.78)
 Bcl-20.0070.51 (0.31–0.83)
2E-cadherin0.0060.53 (0.33–0.84)
 APAF-1< 0.0010.48 (0.31–0.73)

TILs and PTLs (Table 4)

Negative TILs and PTLs in Study Groups 1 and 2 combined were both significantly associated with the presence of tumour budding (p = 0.049 and p < 0.001, respectively) in univariate analysis. A strong correlation between TILs and PTLs was also observed (p = 0.03). The sensitivity and specificity of the absence of TILs for budding were 0.59 and 0.48, respectively, while for the absence of PTLs these were higher, at 0.85 and 0.7. PTL positivity in tumours with loss of E-cadherin was significantly more frequent in non-budding (26.7%) versus budding CRCs (15.8%) (p = 0.04). This effect was more pronounced in tumours with APAF-1 loss, where PTL positivity was found in 32.4% of budding-negative versus 14.2% of budding-positive CRCs (p < 0.001). When TILs were included in the predictive model for budding, along with loss of expression of E-cadherin and APAF-1, the sensitivity increased to 90% but the specificity was low (18%).

Table 4. Frequency of PTL positivity in budding-negative and budding-positive CRCs with loss of E-cadherin and APAF-1 expression
 FeaturesPTL positivity 
Budding-negative (%)Budding-positive (%)p Value
TMA CRCsLoss of E-cadherin26.715.80.04
 Loss of APAF-132.414.2< 0.001
Whole tissue sectionsLoss of E-cadherin66.722.20.032
 Loss of APAF-166.710.00.041

Survival analysis (Table 5)

In univariate analysis, budding-positive tumours had a significantly (p < 0.001) shorter survival time [mean 38 (range 34.0–47.0) months] than budding-negative tumours [mean 107 (range 104.0–114.0) months]. In multivariate analysis, along with T stage, N stage and tumour grade, tumour budding was an independent predictor of worse survival (p < 0.001), HR = 1.74 (95% CI = 1.39–2.17).

Table 5. Univariate survival analysis of tumour markers and clinico-pathological features in TMA CRCs
 p ValueHR (95% CI)
Loss of E-cadherin0.4291.09 (0.89–1.32)
Loss of APAF-10.0121.27 (1.05–1.54)
Negative TILs< 0.0011.61 (1.35–1.96)
Negative PTLs< 0.0011.49 (1.22–1.89)
Loss of E-cadherin, APAF-1 and negative TILs< 0.0011.75 (1.33–2.3)
Negative TILs and PTLs< 0.0011.63 (1.36–1.94)
Loss of E-cadherin and negative TILs0.0051.39 (1.1–1.74)
Loss of E-cadherin and negative PTLs0.3431.1 (0.9–1.34)
Loss of E-cadherin, negative TILs and PTLs0.0071.41 (1.09–1.8)
Loss of APAF-1 and negative TILs< 0.0011.65 (1.37–1.98)
Loss of APAF-1 and negative PTLs< 0.0011.37 (1.14–1.64)
Loss of APAF-1, negative TILs and PTLs< 0.0011.67 (1.38–2.02)
T stage< 0.0013.53 (2.67–4.69)
N stage< 0.0013.11 (2.6–3.7)
Tumour grade< 0.0011.92 (1.49–2.47)
Tumour type0.3271.1 (0.91–1.34)
Budding< 0.0012.51 (2.05–3.08)

Loss of APAF-1, negative TIL count, and absence of PTLs were associated with shorter survival time, as were T stage, N stage and tumour grade. E-cadherin was not found to be predictive of a shorter survival time on its own (p = 0.429). Tumours simultaneously displaying loss of E-cadherin, APAF-1 and negative TILs had a significantly shorter survival time (p < 0.001) than tumours without this set of characteristics. This combination of adverse features was independently predictive of survival (p = 0.027), HR = 1.36 (95% CI = 1.04–1.78) in a multivariate analysis when adjusting for the effect of tumour budding.

Whole tissue sections

Loss of E-cadherin (p < 0.001) and loss of APAF-1 (p = 0.037) were significantly associated with tumour budding in univariate analysis. This loss was observed within tumour buds as well as within the main tumour body. In multivariate analysis, however, only E-cadherin was found to be an independent predictive factor (p < 0.001). The sensitivity of E-cadherin for tumour budding was 80.5% and the specificity 81.1%. Negative TILs (p = 0.02) and PTLs (p = 0.008) were linked to the presence of budding. There was an association between TILs and PTLs, as in the TMA samples (p = 0.003). In multivariate analysis with E-cadherin, TILs were not significantly associated with budding (p = 0.25), while PTLs had independent predictive value (p = 0.049). In tumours with loss of E-cadherin, PTL positivity occurred in 66.7% of budding-negative versus 22.2% budding-positive CRCs (p = 0.032). Similarly, PTL positivity in tumours with loss of APAF-1 was more frequently observed in budding-negative (66.7%) than in budding-positive CRCs (10%; p = 0.041) (Table 4). There was no association between mutation of KRAS and tumour budding (p = 0.88), E-cadherin loss (p = 0.36), APAF-1 loss (p = 0.61), TILs (p = 0.66), or PTLs (p = 0.41). KRAS mutation occurred in 42% of CRCs (42.9% and 41.2% of budding-positive and budding-negative CRCs, respectively).


‘Tumour budding’ refers to a process of de- differentiation that occurs at the invasive margin of colorectal cancer. Budding has been shown to be an independent risk factor with respect to local spread 30, lymph node metastasis 31–36, hepatic metastasis 37, 38, lung metastasis 38, recurrence following curative surgery 39 and survival following curative surgery 2, 3, 40–43. Indeed, budding has been shown to be the most important prognostic feature in the pathological examination of CRC specimens after lymph node spread 3.

The budding immunophenotype is not restricted to budding cells but is generally expressed to a higher degree in budding cells as compared with the main tumour body 43. This means that thresholds for identifying a pro-budding immunophenotype within the main tumour body might need to be set at appropriately sensitive levels (with the risk of compromising specificity). TMAs provide an efficient and cost-effective means for testing a comprehensive panel of potential budding markers on a large number of tumour specimens. By univariate analysis, reduced expression of six markers (APAF-1, E-cadherin, EphB2, Bcl-2, MST1 and CD44s) was significantly associated with tumour budding in two well-matched TMA datasets. Increased expression of β-catenin occurred in one dataset but not the other. While this may represent a limitation of TMAs, others have also failed to show a correlation between budding and expression of β-catenin in the main tumour body 43. This suggests that Wnt pathway activation is a necessary but insufficient explanation for tumour budding. Using multivariate analysis, only reduced expression of E-cadherin and APAF-1 were independent budding markers in both TMA datasets. Loss of expression of E-cadherin in the main body of the tumour could be explained by methylation of the promoter region. Additionally, reduced expression of E-cadherin may occur through the activation of transcriptional repressors, such as Slug and Snail, which may be up-regulated through growth signals derived from stromal cells (see Introduction). Loss of E-cadherin will itself drive the Wnt signalling pathway and increase the expression of laminin5-γ2 18.

There is considerable experimental evidence linking laminin5-γ2 21, 44–46 and E-cadherin 13, 47, 48 with tumour budding and malignant invasion through their specific effects on cell motility, adhesion, matrix degradation and proliferation. However, the additional synergistic role for APAF-1 is an entirely novel finding. APAF-1 plays a pivotal role in the activation of caspases implicated in mitochondria-mediated apoptosis 49 and is critical for normal mammalian development 50. Based mainly on studies of malignant melanoma 49, 51 but also in colorectal cancer 52, APAF-1 has also been highlighted as a tumour suppressor gene. Budding cells show loss of adhesion with each other and with basement membrane proteins. The resulting loss of pro-survival signals would be expected to trigger apoptosis 53, 54. Therefore, inactivation of a key pro-apoptotic gene might confer an important survival advantage and explain the observed correlation between tumour budding and loss of expression of APAF-1.

Epithelial–stromal reactions may drive tumour budding through the secretion of growth factors by both activated myofibroblasts (see Introduction) and tumour-associated macrophages 55. However, macrophages may also have an important anti- tumoural role and they have been linked with tumour cell apoptosis at the invasive front of colorectal cancer 56, 57 as well as with a favourable prognosis 58. A marked peritumoural lymphocytic reaction has also been shown to be an important prognostic marker 1, 59. We were interested in the lack of specificity of changes in E-cadherin and APAF-1 for tumour budding, and hypothesized that an intense peritumoural inflammatory reaction at the invasive margin of CRC might be specifically targeting budding cells. In cancers predicted to show budding on the basis of loss of expression of E-cadherin or APAF-1 but with no apparent budding, there was a greater likelihood of observing a marked peritumoural inflammatory reaction at the invasive margin. This finding was confirmed in the whole tissue sections. TILs were correlated with the presence of PTLs in both TMAs and whole sections, and the absence of TILs and PTLs was associated with budding and a poor prognosis. The possibility that an immune response to CRC may account for an improved prognosis by specifically targeting budding cells offers a novel rationale for immune-based therapies. Such an approach may be conceived as reversing a highly invasive phenotype through ‘nipping CRC in the bud’.

Despite being an independent predictor of tumour budding in the TMA series, loss of E-cadherin was not a significant prognostic marker on its own (Table 5). One explanation of this finding is the low sensitivity of E-cadherin loss for budding (32%). The minute tissue sample available for analysis within TMAs is clearly a limiting factor when evaluating markers with heterogeneous patterns of loss. E-cadherin loss showed a higher sensitivity for budding (80.5%) in whole tumour sections, in which APAF-1 loss was not an independent predictor of budding. Application of these findings to the prediction of budding in biopsy specimens will require further testing. The evaluation of loss of expression of immunohistochemical markers is problematic in diagnostic practice, since different factors may affect the intensity of expression. In this study, all immunohistochemical grading was based on the proportion of positive tumour cells, regardless of expression intensity, and excellent levels of agreement were achieved between observers in the whole tissue sections and in TMAs using this approach (data not shown).

In summary, we have shown that reduced immunohistochemical expression of E-cadherin and APAF-1 identifies most CRCs with the aggressive phenotype of tumour budding. The lack of specificity for the immunohistochemical test appears to be explained by the presence of a peritumoural inflammatory response that probably leads to the apoptotic destruction of budding cells. Tumour immunity may thereby temper the adverse prognostic significance of a pro-budding immunophenotype.


We thank Hassmig Minassian for technical support. This study was supported by grants from the Canadian Institutes of Health Research (MOP 67206) and the Swiss National Foundation (PBBSB-110417) and the Novartis Foundation, formerly the Ciba–Geigy–Jubilee Foundation. We thank Privatdozent Dr Hanspeter Spichtin, Institute of Clinical Pathology, Basel, Switzerland, and Professor Dr Robert Maurer, Institute of Pathology, Stadtspital Triemli, Zürich, Switzerland, for providing the cases.