br Most cases had diffuse strong

Most cases had diffuse strong complete expression of CDX2, as shown in Fig. 3. However, 11% had complete loss and 23% showed at least focal loss of CDX2 expression. Focal loss was defined as at least one high-power field meeting the criteria for loss of CDX2 expression. In many of the CDX2-negative cases, there were distinct areas of the tumor that were 
Table Case demographics
Patient characteristics
Figure 3 A, Representative CDX2 immunohistochemistry (high-power magnification ×200). Kaplan-Meier survival curves based on overall CDX2 score (B) and focal CDX2 score (C).
negative for CDX2 juxtaposed with areas of the tumor that maintained expression of CDX2. CDX2 expression was vari-able across the 3 slides examined, with 23 of 210 cases having variability in the overall score and 86 of 210 cases with vari-ability in the focal score across the 3 stained slides.
The cancer-specific survival for our study population was favorable, with 7% of cases dying of CRC, which is in keeping with a stage II colon cancer population with no adjuvant treat-ment. In contrast to the previous study [6], we did not identify a difference in the cancer-specific survival for either cases with
Figure 4 A, Representative CDX2 immunohistochemistry (original magnification ×200) using the DAK-CDX2 and CDX2-88 antibodies.
Kaplan-Meier survival curves based on overall CDX2-88 score (B) and focal CDX2-88 score (C).
CDX2 and Muc2 in stage II colon cancer 75
Figure 5 Representative immunohistochemistry (original magnification ×200) and Kaplan-Meier survival curves for CDX1 (A), Muc2 (B), GPX2 (C), and villin (D).
complete loss or focal loss of CDX2 (Fig. 3). We were adequately powered to detect a similar difference to the effect size seen in the 2 cohorts reported previously [6] with our 210 cases. All cases had at least 3 years of follow-up with an aver-age follow-up of 5.3 years to detect possible late recurrences.
We used a more sensitive antibody for the detection of CDX2 than the antibody used in the previous study that used CDX2–88 [6]. The CDX2–88 antibody has been shown to be the least sensitive for CDX2 expression in a comparison of 5 commercially available CDX2 771-97-1 [38]. To ensure
Figure 6 A, Representative Muc2 immunohistochemistry (original magnification ×200). Kaplan-Meier survival curves based on overall Muc2 score (B) and focal Muc2 score (C).
Figure 7 A, Prevalence of cases with low expression of CDX2 and Muc2 in the TCGA provisional database accessed August 2018. B-G, Kaplan-Meier curves for overall survival (B, D, and F) and progression-free survival (C, E, and G) for CDX2 (B and C), Muc2 (D and E), and combined CDX2 and Muc2 (F and G).
CDX2 and Muc2 in stage II colon cancer 77
that the difference did not relate to the choice of antibody, we selected a subset of 22 cases that had a recurrence with matched control cases that did not have a recurrence. We found similar results with the CDX2-88 antibody used in pre-vious studies with our DAK-CDX2 antibody, with slightly de-creased expression noted for the CDX2-88 antibody in comparison to the DAK-CDX2 antibody. As shown in Fig. 4B and C, no difference in outcome was identified for cases stratified based on CDX2 expression assessed by the CDX2- 88 antibody.
It has been shown that gene expression signatures can be
used to identify colon cancers with a so-called dedifferentiated or stem cell–like gene expression pattern to identify cases that might have a negative outcome [5,6]. We reviewed the gene expression data from these studies and found other markers that could be assessed using immunohistochemistry. We eval-uated the same 22 cases for expression of CDX1, Muc2, GPX2, and villin as potential markers to identify cases with a negative outcome. CDX1 was expressed in half of the cases and overall showed a weaker, patchier staining than CDX2. In a similar manner to CDX2, it was not associated with a differ-ence in outcome (Fig. 5A). Villin was expressed in 20 of 22 cases and did not predict outcome (Fig. 5D). GPX2 was dif-fusely expressed in all 22 cases (Fig. 5C). In contrast, we found that Muc2 expression was lost in 10 of 21 cases and the Muc2-negative cases had a significant decrease in can-cer-specific survival (Fig. 5B).
Given the observed difference in survival identified based on Muc2 status in our screening cohort of cases, we then expanded this to our larger cohort of cases to as-sess if Muc2 might be used to identify early-stage colon cancers with a poor outcome. As is shown in Fig. 6, Muc2 expression was completely lost in 42% of cases and fo-cally lost in 60% of cases using the same scoring criteria used for CDX2. We found that overall loss but not focal loss of Muc2 was associated with a significant decrease in cancer-spe-cific survival (hazard ratio, 3.32; 95% confidence interval, 1.20-9.20).