Extent and Location of Tumor-Infiltrating Lymphocytes in Microsatellite-Stable Colon Cancer Predict Outcome to Adjuvant Active Specific Immunotherapy

This article is featured in Highlights of This Issue, p. 271

Colorectal cancer is one of the most common cancers worldwide for both males and females, with a 5-year survival rate of approximately 50%, depending on tumor stage. Mutations in oncogenes and tumor suppressor genes such as TP53, KRAS, BRAF, APC, B-Catenin, SMAD4, and AXIN contribute to the carcinogenesis in colorectal cancer (1).

Mutations can also arise in the DNA mismatch repair system in the tumor cells, which can lead to microsatellite instability (MSI). In colorectal cancer15% to 20% is MSI and 80% to 85% is microsatellite-stable (MSS) largely overlapping with chromosomal instability (2, 3). Frameshift mutations in protein-coding sequences may lead to the formation of frameshift peptides in MSI tumors. Frameshift peptide-derived T-cell epitopes can be recognized as foreign by the immune system and may therefore give rise to an increase of tumor-specific T cells (4).

Currently, the prognosis for patients with CRC is determined by using the UICC tumor–node–metastasis (TNM) classification system (5). The 5-year survival for patients with colorectal cancer ranges from almost 100% for patients with stage I disease to less than 5% for patients with stage IV disease (1). Besides the TNM criteria, tumor-intrinsic factors can play a role in determining the prognosis for patients with colorectal cancer. For example, KRAS mutations and specific TP53 mutations are correlated with a relatively poor prognosis (6, 7). Low expression of p21 and cyclin D1 and high expression of p53 and AURKA cell-cycle proteins have been correlated with disease recurrences (8). Conversely, patients with MSI tumors have a relatively good prognosis (9).

TNM classification, genetic factors, and microsatellite status of the tumor are important determinants of the prognosis for patients with colorectal cancer. In recent years, it has become clearer that the tumor microenvironment is also of importance for the prognosis for these patients (10–12). The presence of tumor-infiltrating lymphocytes (TIL) has been shown to play an important role in the survival of patients with colorectal cancer (11, 13, 14). Galon and colleagues have quantified TIL, in the center of the tumor (taking cancer cell nests and the stroma together) and in the invasive margin of the tumor, on the basis of cell surface markers CD3, CD8, and CD45RO and on intracellular granzyme B expression (15–18). For prognostic purposes, the markers CD3 and CD8 appeared to be most informative and currently under investigation for implementation of the Immunoscore (16, 19). Deschoolmeester and colleagues (20) and Naito and colleagues (21) published data on multivariate analysis of a number of prognostic factors. Both groups have shown that the presence of TIL in the invasive margin was of lesser importance than the presence of TIL in the tumor stroma or in the cancer cell nests. The markers CD3 and CD8 emerged as most informative in these and other studies (20–22). To date, no information about the predictive value of TIL in colon cancer in the clinical outcome after adjuvant (immuno)-therapy treatment is available.

The first line of treatment for patients with colorectal cancer is surgery (23, 24). In patients with advanced-stage disease, this is followed by adjuvant chemotherapy. Immunotherapeutic treatment options in the adjuvant setting have been explored as well (25). Harris and colleagues reported on an adjuvant active specific immunotherapy (ASI) versus control study performed in a patient cohort of 412 stage II/III patients with colon cancer (26). Significant differences neither in recurrence-free survival nor in overall survival between the two treatment groups were documented. Evaluation of treatment compliance however showed beneficial effects of adjuvant ASI in patients who showed measurable induration after the third vaccine (26). Harris and colleagues did not make a distinction between patients with stage II and III tumors but grouped them all together in the analyses. Vermorken and colleagues also conducted a multicenter clinical trial on adjuvant ASI for patients with colon cancer (27). A vaccine consisting of irradiated autologous tumor cells admixed with the adjuvant Bacillus Calmette-Guérin bacteria has been evaluated. In that study in patients with colon cancer, a comparison was made between surgery alone and surgery followed by adjuvant ASI treatment. The recurrence-free interval (RFI) for patients with stage II tumors was significantly extended for patients treated with surgery plus ASI compared with surgery alone but not for stage III patients (27). In a retrospective follow-up study, tumor samples from these patients were analyzed for their microsatellite status; 17% of the samples were MSI and 83% were MSS tumors (28). The group of patients with MSI tumors had an increased survival rate independent of ASI therapy. Patients with stage II MSS tumors showed a significantly increased recurrence-free survival when they received ASI (28).

ASI treatment is expected to boost the patients' immune system to achieve antitumor activity. We explored whether the number and nature of pre-existing T cells in the tumor stroma and epithelium had any bearing on clinical outcome for these patients. As a start, we used the biomarkers CD3 and CD8 in combination with location in the primary tumor of patients with stage II or III MSS colon cancer.

In the current retrospective study, we used archival tumor material derived from patients with colon cancer who participated in the same randomized trial testing the effects of adjuvant ASI (27). Disease-specific survival (DSS) and RFI were evaluated in detail at the 5-year posttreatment timepoint. First, we investigated the prognostic value of stromal and intraepithelial T-cell infiltrates independent of treatment arm. Second, we investigated the predictive value of stromal T-cell infiltrates for clinical outcome after adjuvant ASI treatment. This was done by comparing the treatment effect of adjuvant ASI in patients with high TIL to the treatment effect in low TIL.

The patient population, inclusion criteria, and vaccination protocol have been described in detail previously (27). In summary, eligible patients with stage II or III resectable adenocarcinoma of the colon and a good performance status were randomly assigned postoperative adjuvant ASI or no adjuvant treatment. From 254 patients included in the original ASI trial (27) and the follow-up study (using tumor samples of 196 patients of the original 254 patients) on the effects of MSI/MSS (28), 106 tumor samples were available, and of sufficient quality, to be used in the current study. An overview of the relevant patient characteristics is shown in Fig. 1.

Formalin-fixed, paraffin-embedded (FFPE) tumor sample sections of 3 μm were deparaffinized in xylene and dehydrated in ethanol. Endogenous peroxidase was blocked in 0.3% H₂O₂/methanol. Antigen retrieval for CD3 and CD8 was performed by boiling the slides in TRIS/EDTA (pH 9) in a microwave for 10 minutes. Antibodies were applied against CD3 [CD3 polyclonal (Rabbit-anti-Human (code A0452), 1:200 dilution; Dako Cytomation] and CD8 (clone C8/144B, 1:100 dilution; Dako Cytomation)and incubated for 1 hour at room temperature. All stainings were performed using the Powervision Plus (Dako Cytomation) Method. Bound peroxidase was visualized with 3,3′-diaminobenzidine (DAB) and nuclei were counterstained with hematoxylin.

The stained tumor sections were digitized by the Mirax slide Scanner system (3DHISTECH) equipped with a 20× objective with a numerical aperture of 0.75 and a Sony DFW-X710 Fire Wire 1/3” type progressive SCAN IT CCD (pixel size 4.65 × 4.65 μm²). The actual scan resolution (effective pixel size in the same sample plane) at 20× is 0.23 μm. Pannoramic Viewer version 1.14.50 (3DHISTECH) software was used to view the virtual slides. Per slide 10 fields (20 times magnification, 1,082 × 786 pixels, 670 × 486 μm²) were randomly selected and explored in the TIFF image format to be quantified after excluding areas of insufficient quality (i.e., tissue folds or scanning errors) and areas not containing tumor fields. Various areas in the selected fields were demarcated to distinguish tumor fields from tumor stroma and necrosis or artifacts. Lumen was automatically recognized as white regions and excluded from further analysis.

The images were analyzed using an in-house developed macro for ImageJ (U. S. NIH, Bethesda, MD) to quantify the number of positively staining cells, as previously described (29). The densities were calculated by dividing the cell count by the corresponding surface area for each of the 10 fields per slide and then taking the average value. Slides with less than 10 interpretable fields were excluded from the analysis. Every slide received three scores: total cell density (in tumor nests and stroma together), cell density in stroma, and cell density in tumor nests, all in cells/mm².

Data analysis was performed with the Statistical Package for the Social Sciences (SPSS Inc.) version 20. Descriptive statistics were used to describe clinicopathologic and treatment-related factors in the cohort. The primary endpoints were DSS and RFI. The former was defined as the time in years after surgery until disease specific death in a follow-up period of 15 years. RFI was defined as the time in years after surgery until a recurrence of the disease was diagnosed in a follow-up period of 15 years. Survivors were censored on the date they were last known to be alive (DSS) or to be free of disease (RFI).

To assess comparability of the selected study cohort to previous publications, Kaplan–Meier curves were used to depict survival in groups defined by MSI status (MSI vs. MSS), stage (II or III), and treatment (ASI vs. control; ref. 28).

In further analyses on the relation between TIL expression and primary study outcomes, only the MSS cohort was used. Untransformed and log-transformed TIL expression data were analyzed for normality using descriptive statistics and Normality plots. To satisfy normality requirements in further analyses, the log-transformed data of epithelial CD3 count, stromal CD8 count, and epithelial CD8 count were used, whereas original stromal CD3 count was used. T-testing was used to compare levels of the 4 TIL variables between events and no-events. Both a 5-year and a 10-year cutoff was used to define events, with the group of no-events consisting of patients without any event at 15 years posttreatment or an event after 5 and 10 years, respectively. Data are shown for the 5-year evaluation point. T-testing of difference in TIL was repeated after stratification for stage and ASI treatment, respectively.

From 106 of the 254 patients originally included in the ASI trial reported by Vermorken and colleagues (27), FFPE tumor samples were still available and of sufficient quality to perform immunohistochemical (IHC) staining and quantification. Of these 106 tumor samples, 99 were available for CD3 staining and 104 for CD8 staining. Figure 1 shows a diagram connecting the different studies, a panel showing the number of patients in each subgroup according to microsatellite status, tumor stage, and treatment, and examples of IHC staining. The currently investigated group contained most (24 of 34) of the patients with MSI tumors and about half of the patients with MSS tumors (82 of 162) identified in our previous study on the effects of microsatellite status and adjuvant ASI treatment (28). The distribution of patient characteristics between the various subgroups was similar to that in the original study reported by Vermorken and colleagues (27).

Figure 1 also shows examples of tumor sections stained for CD3 and CD8 (left and right, respectively, top shows high numbers of TIL, bottom shows low numbers of TIL). Tumor stroma and tumor nests can readily be identified and were demarcated to facilitate computer-assisted analysis as described in Materials and Methods. Intraepithelial T cells (also referred to as “tumor nest”) and T cells in the tumor surrounding stroma (also referred to as “stromal”) were quantified for both CD3 and CD8. In the present study, we focused on the parameters tumor stage, microsatellite status of the tumor, T-cell infiltration, and (adjuvant ASI) treatment and their respective bearing on clinical outcome measured as DSS and RFI over a follow-up period of 15 years. To assess comparability of the selected study cohort to our previous study (28), Kaplan–Meier curves were used to depict survival in groups defined by microsatellite status (MSI vs. MSS), stage (II or III), and treatment (ASI vs. control). The DSS and RFI data on the 106 patients in the current study are presented in Fig. 2 and are at large in agreement with the data published by us previously on the larger group of 196 patients (28).

Computer-assisted T-cell counting revealed a lack of significant differences in the average number of TIL in either of the two categories (tumor nest or stroma) for either of the two markers (CD3 or CD8) between stage II and III tumors (data not shown). On the other hand, MSI tumors compared with MSS tumors contained significantly higher numbers of both CD3- and CD8-positive TIL in the tumor nests (all P < 0.01) but not in the stroma (data not shown). As previously reported by de Weger and colleagues (28), the effects of tumor stage and treatment in the group of patients with MSI tumors were undetectable because of the small size of some of the subgroups. Therefore, we excluded patients with MSI tumors in subsequent analysis with respect to the potential influence of pre-existing TIL.

Next, we analyzed the prognostic and predictive values of pre-existing tumor-infiltrated T cells on the clinical outcome measured as DSS or RFI in patients with MSS tumors. Figure 3 shows scatterplots for the numbers of pre-existing stromal CD3 T cells in two groups of patients; those that experienced an event later on in life and those that did not (Fig. 3A for DSS and 3B for RFI). A wide range of T-cell counts was noted. As expected, there were no significant differences in pre-existing T-cell counts between the event and the no-event groups displayed (data not shown). A normal distribution of the continuous T-cell counts was noted for stromal CD3 numbers, but not for epithelial CD3, stromal CD8, and epithelial CD8 T-cell numbers. Log transformation of the numbers of the latter three showed a normal distribution (data on file). T-testing of the differences in T-cell counts in patients with an event (death or a recurrence within 5 years) versus those without was performed on the original stromal CD3 counts and on log-transformed epithelial CD3, stromal CD8, and epithelial CD8 counts.

Table 1 shows the mean number of TIL and the mean differences correlated with DSS for patients with MSS tumors. Table 1A shows the data for DSS on all patients (so stage II plus stage III) with MSS tumors. Subanalyses are shown for stage II patients in Table 1B and for stage III patients in Table 1C. The data show there is a significant difference in the number of stromal CD3 T cells between patients with stage II tumors who died from the disease versus those who did not within a period of 5 years after treatment. The patients who experienced an event (death) appeared to have lower stromal CD3 T cells compared with patients who were alive at the 5-year evaluation time point (Event: mean, 594; SD, 235; No-Event: mean, 899; SD, 248; difference: P = 0.01). Significance was not reached for the other three T-cell categories in patients with stage II MSS tumors. No significant differences were noted in patients with stage III MSS tumors in any of the four T-cell categories, nor if we group stage II and III together (Table 1C and A, respectively). Data were also analyzed at the 10- and 15-year evaluation point. This revealed little differences compared with the original 5-year time point (data not shown).

Similar analyses were performed for RFI. The data are shown in Table 2. Significant differences were found in the number of pre-existing CD8-positive T cells found in the tumor nest in the group of patients with MSS tumors in stage II as well as in the whole group (stage II plus stage III together), but not in the separate group of patients with stage III tumors Table 2B, A, and C, respectively). Patients who had a recurrence within 5 years had significantly lower CD8-positive T cells in the tumor nest than those patients who did not have a recurrence (stage II separately and stage II plus III). Data were also analyzed at the 10- and 15-year evaluation point. This revealed little differences compared with the original 5-year time point (data not shown).

On the basis of these data, we concluded that (i) high numbers of stromal CD3 T cells have positive prognostic value measured as DSS for patients with stage II MSS tumors and (ii) high numbers of epithelial CD8-positive T cells have positive prognostic value measured as RFI for the group of patients with stage II MSS tumors as well as for the whole group (so stage II plus stage III together).

Previously, we have documented beneficial effects of adjuvant ASI in patients with MSS colon cancers (28). Here, we investigated whether these beneficial effects of adjuvant ASI correlated with the extent and location of the T-cell infiltrates in tumor samples derived from patients with MSS tumors. Data from patients with stage II or stage III MSS tumors were grouped together for the separate analyses per treatment arm (control vs. adjuvant ASI). The data are shown for DSS in Table 1D (ASI) and E (control) and for RFI in Table 2D (ASI) and E (control). No statistically significant differences in T-cell counts between patients with an event or without an event were noted in the control groups for DSS and for RFI. For DSS in the ASI-treated patient group, there was a highly significant difference in the number of pre-existing stromal CD3-positive T cells between events versus no-event (event: mean, 523; SD, 160; no-event: mean, 895; SD, 228; difference: P = 0.00). For stromal CD8 and for epithelial CD8 T-cell numbers, the differences were not significant (P = 0.08 and P = 0.06, respectively). For RFI in the ASI-treated patient group, there was a highly significant difference between event and no-event in the number of pre-existing stromal CD3-positive T cells (event: mean, 635; SD, 194; no-event: mean, 904; SD, 236; difference: P < 0.01) and in the number of pre-existing epithelial CD8-positive T cells (event: mean, 168; SD, 228; no-event: mean, 232; SD, 295; difference: P = 0.01).

On the basis of these data, we concluded that high numbers of pre-existing stromal CD3-positive T cells are of positive predictive value in adjuvant ASI treatment measured as DSS as well as RFI. In addition to this, epithelial CD8-positive T cells also have positive predictive value in adjuvant ASI treatment measured as RFI.

In the previous paragraphs on prognostic and predictive values of TIL, we used continuous data. We next dichotomized the continuous data on TIL counts in high and low on the basis of the mean T-cell count for the most relevant category; stromal CD3. High is defined as above the mean value and low is defined as equal to or below the mean value. In Tables 1 and 2, we reported on the differences in T-cell count between patients who had experienced an event and who that did not. Here, we show survival as Kaplan–Meier plots for DSS and RFI for the marker CD3 stroma. We compared ASI/High with Control/High and ASI/Low with Control/Low and show the data in Fig. 4 for DSS as well as RFI. The data clearly show that patients with high stromal CD3 T cells benefit from ASI treatment, both in DSS and in RFI [Fig. 4A (P = 0.01) and 4C (P = 0.01), respectively]. On the other hand, patients with low stromal CD3 T cells did not benefit from ASI treatment in DSS nor RFI (Fig. 4B and D, respectively). To see what effect dichotomization of the continuous data has on the outcome, we also performed the analysis with a 25th and a 75th percentile cutoff. Figure 4E shows the pre-existing stromal CD3 T-cell counts for the controls and the ASI treatment groups. As expected, there is no significant difference in the counts between these two groups (data not shown). The figure also shows the cutoff at the 25th, 50th, and 75th percentiles. Resulting P values are shown in a matrix identical to the way the panels of Fig. 4A–D are shown. From the data, it is clear that the beneficial effect of high stromal CD3 T cells in combination with ASI treatment is seen over a wide range of cutoff values. In conclusion, (i) prognostic as well as predictive values may be attributed to stromal CD3 T cells and (ii) prognostic value may be attributed to epithelial CD8 T cells.

The presence of TIL in colorectal cancer has been documented in a number of studies (21, 30–37). In recent years, Galon and colleagues have introduced the concept of the so-called Immunoscore, which is aimed at providing additional prognostic value to the universally implemented TNM classification for patients with colorectal cancer, based on the presence of infiltrating T cells in the invasive margin and in the center of the tumor (16, 17, 38, 39). With a few exceptions (20, 40), most of the studies on infiltrating T cells have not taken the microsatellite status of colorectal cancer into account as one of the important parameters influencing clinical outcome. The microsatellite status of colorectal cancer tumors has a strong prognostic value (3, 9, 41). Recently, Melo and colleagues identified three different molecularly distinct subtypes of colon cancer (42). One of which is the MSI subgroup (CCS2-MSI). Sadanandam and colleagues identified five different subgroups, two of which were named the inflammatory subset and the goblet-like subtype (43). A close comparison between these two datasets has shown that the CCS2-MSI subgroup encompasses the inflammatory and goblet subtypes (44). According to the authors, these associations make sense also in light of previous studies (45), as MSI tumors are often associated with an inflammatory immune infiltrate and a mucinous phenotype (44).

Adjuvant ASI treatment has been shown to have beneficial effects in patients with stage II colorectal cancer (27). Recently, we have documented that these beneficial effects of adjuvant ASI treatment were most prominent in patients with MSS tumors (28). Here, we report on the presence of infiltrating T cells in the tumor nests and the stroma of tumors obtained from patients with MSI (n = 24) or MSS (n = 82) tumors and correlate the presence of these T cells with prognostic value for the patients and predictive value with respect to the outcome after adjuvant ASI treatment. Larger numbers could not be included because of a lack of sufficient tumor material, as parts had already been used previously (28) or because of poor quality and staining, as the material was obtained in the 1990s (27).

In the patient group with MSI tumors, the T-cell numbers are significantly higher than the group with MSS tumors (data not shown). This is in agreement with data reported by other investigators using IHC and mRNA expression levels (46). Furthermore, we found significant differences between MSI and MSS tumors in tumor nest infiltrating CD3⁺ and tumor nest CD8⁺ cells, but not in stromal CD3⁺ and stromal CD8⁺ cells. This is in good agreement with what has been reported by others using IHC (4, 47).

In general, high T-cell infiltrates are indicative for a more favorable prognosis (20). We have documented previously that patients with MSI tumors fair better than patients with MSS tumors (28). It is tempting to attribute this difference to the number of TIL. High TIL counts would be beneficial for clinical outcome. However, this raises the issue of determining a cutoff, for which ROC methodology has been applied frequently (17, 48). However, by ROC dichotomizing TIL data, the probability of type I errors is inflated in subsequent analyses. Here, we have avoided this problem by using continuous data on T-cell counts in the analyses reported in Tables 1 and 2. A normal distribution was noted for stromal CD3 numbers, but not for epithelial CD3, stromal CD8, and epithelial CD8 T-cell numbers. Log transformation of the numbers of the latter three showed a normal distribution. For the Kaplan–Meier survival plots shown in Fig. 4, we have arbitrarily used the mean as cutoff value. Although there was an anticipated effect on the P values, the data in Fig. 4F indicate that a wide range of cutoffs may be used to dichotomize continuous data.

In consideration of the fairly large number of variables in the current dataset, we have embarked on multivariate analysis (MVA) of the data. The variables were: stage (stage II or III), treatment (ASI treatment or control), CD3 TIL in the stromal compartment (CD3 high vs. low), CD3 TIL in the epithelium (CD3 high vs. low), CD8 TIL in the stromal compartment (CD8 high vs. low), CD8 TIL in the epithelium (CD8 high vs. low), events and no-events for DSS and for RFI separately. In a first attempt, we have used ROC dichotomized data, revealing, among other findings, a clear interaction between high stromal CD3 TIL and ASI treatment. In a second attempt, we used continuous TIL data and again found this interaction. However, because of the large number of variables and possible interactions between these variables, the main effects (e.g., tumor stage or treatment) sometimes appeared in the wrong direction. This made the analysis of the data after MVA extremely complex and the outcome almost uninterpretable. Hence we have excluded these results from presentation here. More in general studies like the one we have performed here are hampered by the large number of variables and possible interactions and the relatively low numbers of patients per (sub)-group.

The prognostic value of TIL in colorectal cancer has been documented by a number of groups (4, 14, 17, 20, 49, 50). And although an association between TIL and response to chemotherapy has been found in several studies (36, 51), the predictive value of TIL is still under investigation and under debate (52–55). With the currently available data presented here and analysis thereof, we concluded that (i) prognostic as well as predictive values may be attributed to stromal CD3 T cells and (ii) prognostic value may be attributed to epithelial CD8 T cells. We suggest that the value of enumerating T-cell infiltration of tumor nests and tumor stroma could be further improved by distinguishing patients with MSI tumors from those with MSS tumors in large cohorts of patients. Future multicolor staining and automated counting procedures will further enhance data extraction from small tumor samples.

A comprehensive review of the scientific literature on the treatment of colon cancer and the possible role for immunotherapeutic strategies is beyond the scope of this article. However, we would like to offer a number of suggestions here. Adjuvant ASI based on autologous tumor cells will not be implemented on a large-scale anymore, as it is too laborious, labor-intensive, and expensive. However, on the basis of our analyses on the role of TIL in the clinical outcome after adjuvant ASI, we do see a window of opportunity for immunotherapeutic intervention, especially, but not exclusively, for patients with stage II disease. Vaccination strategies based on frameshift peptides might very well be applicable to patients with MSI tumors (56). For patients with MSS tumors, application of antibodies directed against T cell expressed markers like CTLA-4 or PD-1 might be more appropriate to unleash antitumor immune responses (57). For adjuvant treatment of patients with colon cancer, we foresee that immunotherapeutic approaches will regain scientific interest and hopefully clinical application in the near future. The biomarkers CD3 and CD8 in combination with TIL numbers and location may aid oncologists in determining which patient is likely to benefit most from these immunotherapeutic approaches.

G.A. Meijer reports receiving other commercial research support from the Center for Translational Molecular Medicine. C.J.L.M. Meijer reports receiving speakers bureau honoraria from GlaxoSmithKline, Qiagen, and Roche and is a consultant/advisory board member for Qiagen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.W. Turksma, M.C. Shamier, V.A. de Weger, C.J.L.M. Meijer, E. Hooijberg

Development of methodology: A.W. Turksma, M.C. Shamier, V.A. de Weger, J.A.M. Belien, C.J.L.M. Meijer, E. Hooijberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.C. Shamier, K.L.H. Lam, V.A. de Weger, J.A.M. Belien, C.J.L.M. Meijer, E. Hooijberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.W. Turksma, V.M.H. Coupé, M.C. Shamier, K.L.H. Lam, J.A.M. Belien, A.J. van den Eertwegh, C.J.L.M. Meijer, E. Hooijberg

Writing, review, and/or revision of the manuscript: A.W. Turksma, V.M.H. Coupé, M.C. Shamier, J.A.M. Belien, A.J. van den Eertwegh, G.A. Meijer, C.J.L.M. Meijer, E. Hooijberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.W. Turksma, M.C. Shamier, K.L.H. Lam, V.A. de Weger, J.A.M. Belien, E. Hooijberg

Study supervision: C.J.L.M. Meijer, E. Hooijberg

The authors thank the following pathologists from the participating hospitals in retrieving the FFPE material: Drs. M. Flens, Zaans Medisch Centrum, Zaandam; M. Brinkhuis, Stichting Laboratorium Pathologie Oost, Enschede; H. Doornewaard, Gelre Ziekenhuis, Apeldoorn; A. Uyterlinde, Medisch Centrum Alkmaar, Alkmaar; and E. Barbe, St Lucas, Amsterdam. The authors also thank the reviewers for their helpful suggestions for improvement of the article.

This work was supported by grant VU2007-3814 from the Dutch Cancer Society.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.