Podocalyxin-like protein as a predictive biomarker for benefit of neoadjuvant chemotherapy in resectable gastric and esophageal adenocarcinoma

The neoadjuvant cohort consists of 148 consecutive patients diagnosed with resectable gastric or esophageal adenocarcinoma who received neoadjuvant ± adjuvant chemotherapy at the Skåne University Hospital in Lund and Malmö between January 1, 2008 and December 31, 2014. This cohort has been described previously [30] but for the present study, patients treated with neoadjuvant chemoradiotherapy or with palliative chemotherapy followed by salvage surgery were excluded. Prior to start of neoadjuvant treatment, all patients were discussed at a multidisciplinary tumor board where clinical stage was determined based on findings on endoscopy and computerized tomography. Diagnostic laparoscopy was not part of routine work-up unless peritoneal carcinomatosis was suspected. Data on recurrence and vital status were updated until December 31, 2017. Patient and tumor characteristics in the neoadjuvant cohort are described in Table 1 and treatment data in Additional file 1. The vast majority (> 95%) of the patients were treated with fluoropyrimidine- and oxaliplatin-based chemotherapy (EOX, FOLFOX or FLOX).

The surgery up-front cohort was derived from a previously described [29, 31‐38] consecutive cohort of 174 patients treated with surgical resection between 2006 and 2010, without neoadjuvant therapy, but 13 patients who received adjuvant treatment were excluded. Follow-up was done until March 1, 2016. The neoadjuvant cohort and the surgery up-front cohort were merged into a pooled cohort. An overview of the cohorts is depicted in Fig. 1.

For both the neoadjuvant cohort and the surgery up-front cohort, clinical and pathological classification of tumor stage was done according to the 7th edition of the UICC/AJCC TNM classification, whereby tumors in the gastroesophageal junction Siewert type I–III were classified as esophageal cancer. Residual tumor status was classified as: R0 = no residual tumor (free resection margins according to the pathology report), R1 = possible microscopic residual tumor (narrow or compromised resection margins according to the pathology report), R2 = macroscopic residual tumor (according to the operative report).

From the neoadjuvant cohort, archival pre-neoadjuvant (diagnostic) biopsies were retrieved along with blocks from the post-neoadjuvant resected specimens. Tissue microarrays (TMAs) were constructed with duplicate cores from separate donor blocks from the resected primary tumor and (if available) from lymph node metastases as previously described [30]. For immunohistochemistry (IHC), 3 μm sections from the biopsies and 4 μm sections from the TMAs were prepared and stained using the rabbit polyclonal anti-PODXL antibody HPA002110 (Atlas Antibodies AB, Stockholm, Sweden) similarly as described in our previous study [29] on PODXL in the surgery up-front cohort. Staining of PODXL was scored as: negative (0), weak cytoplasmic positivity in any proportion of cells (1), moderate cytoplasmic positivity in any proportion of cells (2), distinct membranous positivity in ≤ 50% of cells (3) and distinct membranous positivity in > 50% of cells (4). For duplicate cores the highest score was used. Tumor-associated vascular endothelial cells and glomeruli in control renal tissue in the TMAs served as positive controls, and normal gastric mucosa and esophageal epithelium served as negative controls. Evaluation of the staining was carried out by two observers (AL and KJ) blinded to clinical and outcome data, and scoring discrepancies were discussed to reach consensus. Of note, for patients treated with surgery up-front, PODXL expression was assessed only in the resected primary tumor and not in lymph node metastasis which was done in our original report [29]. This resulted in an elevated proportion (from 17.0 to 21.2%) of cases denoted as PODXL negative in the surgery up-front cohort.

To validate the use of TMAs for PODXL scoring, whole tissue sections from 25 resected primary tumors were evaluated, blindly to the annotated scores of their corresponding TMA cores.

The extent of residual cancer cells in the post-neoadjuvant resected primary tumor site was histologically evaluated using the 4-tiered tumor regression grading system described by Chirieac [39], i.e. 0%, 1–10%, 11–50% or > 50% residual cancer cells.

Follow-up time was based on reverse Kaplan–Meier estimate. The scoring of PODXL expression was trichotomized into negative (0), low (1–2) or high (3–4) and then further dichotomized into negative (0) or positive (1–4). For assessment of the correlation of PODXL expression between tissue samples, Kendall’s tau-b (τ_b) was used. Associations of clinicopathological factors with PODXL expression were analyzed using Kruskal–Wallis or Mann–Whitney U test for continuous variables, and Chi square test (Fisher’s exact for 2 × 2 tables and linear-by-linear association for larger tables) for categorical variables. Differences in histopathologic response stratified by trichotomized PODXL expression was assessed using Chi square test (linear-by-linear). Since only resected patients were included in the surgery up-front cohort, we assessed TTR and OS only in those patients in the neoadjuvant cohort who had undergone surgical resection. To account for the fact that a recurrence, by definition, cannot occur until the cancer has been resected, and for the differences in timing of surgery between the neoadjuvant cohort and the surgery up-front cohort, we used the resection date as baseline for TTR and OS. TTR was thus defined as time from resection to the date of biopsy or radiology proven recurrent disease. OS was defined as time from resection to the date of death. Furthermore, patients who died within 90 days from resection were excluded since this was regarded as being related to postoperative complications. Differences in Kaplan–Meier curves were assessed using log-rank test. Hazard ratios (HR) for TTR and OS were derived from Cox proportional-hazards regression. The variables adjusted for in the multivariable analyses are described in each corresponding table. In case of a low number of events per variable, only established prognostic variables were included in the adjusted model, in order to avoid overfitting of the data. In the Cox regression analyses of the pooled cohort, the potential role of PODXL as a predictive biomarker was assessed using an interaction term between the binary variables PODXL expression (negative vs. positive) × treatment (surgery only vs. neoadjuvant chemotherapy). A p-value < 0.05 was regarded as statistically significant and all tests were 2-sided. IBM^® SPSS^® Statistics version 23.0.0.2 for Mac OS was used for the statistical analyses.

PODXL expression could be assessed in pre-neoadjuvant biopsies from 99/148 (67%) patients, in post-neoadjuvant resected primary tumors from 102/118 (86%) patients, and in post-neoadjuvant resected lymph node metastases from 47/66 (71%) patients. Sample IHC images are shown in Fig. 2. The distribution of PODXL expression in the pre-neoadjuvant biopsies was: negative 33.3%, low 53.6% and high 13.1%. There was no correlation between PODXL expression in paired pre-neoadjuvant biopsies and resected primary tumors. There was a moderately strong correlation of PODXL expression between paired post-neoadjuvant resected primary tumors and lymph node metastases (trichotomized: τ_b = 0.48, p < 0.001, dichotomized: τ_b = 0.56, p < 0.001). The correlations and conversions of PODXL expression are further detailed in Additional file 2.

As shown in Additional file 3, there was a strong correlation between PODXL scoring in whole tissue sections and corresponding TMA cores (τ_b > 0.91, p < 0.001). In one of the 10 tested cases assessed as negative in the TMA, weak positive expression (score 1) was denoted on the whole tissue section.

The only factor significantly associated with PODXL expression was post-neoadjuvant pathological T-stage (ypT), which was lower for patients with high PODXL expression in their pre-neoadjuvant biopsies (p = 0.029). For details see Additional file 4.

As shown in Fig. 3, patients with high PODXL expression in their pre-neoadjuvant biopsies had a superior histopathologic response (notably 36% with no residual cancer cells) compared to those with negative (p = 0.010) or low (p = 0.013) PODXL expression. However, there were no significant differences in histopathological response between tumors with negative and low PODXL expression.

Kaplan–Meier plots and log-rank analyses (Fig. 4) demonstrate that patients with high PODXL expression in their pre-neoadjuvant biopsies were all recurrence-free at last follow-up, and had a superior TTR compared to PODXL negative cases (p = 0.026). Of note, one of the patients with a PODXL high tumor had an R2-resection, thus, TTR was not applicable in this case. Furthermore, for patients with PODXL high tumors treated with neoadjuvant fluoropyrimidine and oxaliplatin for a minimum of 8 weeks, no deaths had occurred at last follow-up, and OS was superior compared to PODXL negative cases (p = 0.038). There were no statistical differences between PODXL negative and PODXL low cases for neither TTR nor OS. As a comparison, survival plots stratified by PODXL expression in resected primary tumors from the surgery up-front cohort are also shown in Fig. 4.

Cox regression analyses on the neoadjuvant cohort for TTR and OS could not be performed with trichotomized PODXL expression due to few events (mostly none) for PODXL high tumors, thus dichotomized PODXL expression was used (see tables in Additional file 5). PODXL was not prognostic neither for TTR nor OS, except in patients treated with neoadjuvant fluoropyrimidine and oxaliplatin for a minimum of 8 weeks, where PODXL was prognostic for TTR in univariable analysis (HR 0.40, 95% CI 0.17–0.96, p = 0.041), but not when adjusting for known prognostic factors.

The pooled cohort, stratified by PODXL expression, was well balanced for all clinicopathological factors (Table 2), except for a larger proportion of gastric cancer in PODXL negative compared to positive cases (58.2 vs. 40.0%, p = 0.020), and a larger proportion treated with neoadjuvant chemotherapy in PODXL negative compared to positive cases (45.5 vs. 30.0%, p = 0.048).

Kaplan–Meier analyses (Fig. 5) demonstrate that for patients with PODXL negative tumors in the pre-neoadjuvant biopsies, neoadjuvant ± adjuvant chemotherapy had no effect on TTR or OS compared to surgery alone. In contrast, for PODXL positive cases, a superior TTR (estimated recurrence-free rate at 5 years: 69% vs. 41%) as well as OS (estimated surviving rate at 5 years: 61% vs. 31%) was shown with neoadjuvant ± adjuvant chemotherapy compared to surgery alone. Cox regression analyses of TTR and OS (Table 3) stratified by PODXL expression were consistent with these findings in both univariable and multivariable analyses. Furthermore, in Cox regression analyses of the entire (unstratified) pooled cohort, a beneficial effect of neoadjuvant ± adjuvant chemotherapy on TTR and OS was seen only in univariable analyses but not when adjusting for other factors including an interaction term (PODXL expression x treatment), thus indicating a possible predictive role for PODXL. The interaction term was statistically significant for TTR in unadjusted analysis for neoadjuvant chemotherapy NOS (not otherwise specified) ± adjuvant chemotherapy NOS vs. surgery alone (p = 0.018) but not in the adjusted analysis (p = 0.069). Limiting the neoadjuvant chemotherapy group to those treated with neoadjuvant fluoropyrimidine and oxaliplatin for a minimum of 8 weeks, the interaction term was significant in both unadjusted (p = 0.006) and adjusted (p = 0.024) analyses, further supporting a potential predictive role for PODXL. For OS, however, the interaction term was not significant, neither in unadjusted nor in adjusted analyses.