Keratin 17 is a prognostic and predictive biomarker in pancreatic ductal adenocarcinoma

Abstract

Objectives: To determine the role of keratin 17 (K17) as a predictive biomarker for response to chemotherapy by defining thresholds of K17 expression based on immunohistochemical tests that could be used to optimize therapeutic intervention for patients with pancreatic ductal adenocarcinoma (PDAC).

Methods: We profiled K17 expression, a hallmark of the basal molecular subtype of PDAC, by immunohistochemistry in 2 cohorts of formalin-fixed, paraffin-embedded PDACs (n = 305). We determined a K17 threshold of expression to optimize prognostic stratification according to the lowest Akaike information criterion and explored the potential relationship between K17 and chemoresistance by multivariate predictive analyses.

Results: Patients with advanced-stage, low K17 PDACs treated using 5-fluorouracil (5-FU)-based chemotherapeutic regimens had 3-fold longer survival than corresponding cases treated with gemcitabine-based chemotherapy. By contrast, PDACs with high K17 did not respond to either regimen. The predictive value of K17 was independent of tumor mutation status and other clinicopathologic variables.

Conclusions: The detection of K17 in 10% or greater of PDAC cells identified patients with shortest survival. Among patients with low K17 PDACs, 5-FU-based treatment was more likely than gemcitabine-based therapies to extend survival.

Keywords: chemotherapies; immunohistochemistry; pancreatic ductal adenocarcinoma; predictive biomarkers.

FIGURE 1

Flowchart, analysis of keratin 17 in 2 differential cohorts of PDAC. FFPE, formalin-fixed, paraffin-embedded; IHC, immunohistochemistry; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 2

Identification of optimal cutoff value for K17 on both discovery (A-C) and validation (D-F) cohorts, respectively. A, Akaike information criterion (AIC). P value (B) and HR values (C) plotted as a function of K17 expression PathSQ score in the discovery cohort and in the validation cohort (D-F). Gray shaded areas in A-F represent areas where P value was significant. G, Boxplot depicting K17 expression levels in the entire cohort (n = 379). Representative images of moderately differentiated PDACs with low K17 (H) and high K17 (I) expression. AIC, Akaike information criterion; HR, hazard ratio; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 2

Identification of optimal cutoff value for K17 on both discovery (A-C) and validation (D-F) cohorts, respectively. A, Akaike information criterion (AIC). P value (B) and HR values (C) plotted as a function of K17 expression PathSQ score in the discovery cohort and in the validation cohort (D-F). Gray shaded areas in A-F represent areas where P value was significant. G, Boxplot depicting K17 expression levels in the entire cohort (n = 379). Representative images of moderately differentiated PDACs with low K17 (H) and high K17 (I) expression. AIC, Akaike information criterion; HR, hazard ratio; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 3

Validation of K17 as an independent negative prognostic biomarker of PDAC. Kaplan-Meier curves for the overall survival analysis of K17 from PDAC cases in the discovery (A) and validation (B) cohorts. A, Low K17 (n = 15 [25%]; median = 19 months) and high K17 (n = 59 [75%]; median = 27 months). HR = 3.217 (95% CI, 1.604-6.450); P = .0177. B, Low K17 (n = 76 [25%]; median = 42 months); high K17 (n = 229 [75%]; median = 25 months). HR = 1.511 (95% CI, 1.065-2.124); P = .0338. Forest plots showing the multivariate analysis factors in the discovery (C) and validation (D) cohorts. Kaplan-Meier curves depict the overall survival, which integrate K17 status and the lymph node status for the combined discovery (E) and validation (F) cohorts. E, Lymph node–negative cases. Low K17 (n = 25 [26%]; median = 36 months); high K17 (n = 70 [74%]; median = 22 months). HR = 2.119 (95% CI, 1.174-3.823); P =.0258. F, Lymph node–positive cases. Low K17 (n = 76 [27%]; median = 33 months); high K17 (n = 208 [73%]; median = 23 months). HR = 1.272 (95% CI, 0.9163-1.765); P = .1659. Kaplan-Meier curves depict the overall survival, which integrate K17 status and pathological stage for the combined discovery (G) and validation (H) cohorts. G, Early-stage (I-IIB) cases. Low K17 (n = 77 [26%]; median = 36.6 months); high K17 (n = 225 [74%]; median = 25.1 months). HR = 1.441 (95% CI, 1.042-1.997); P =.0392. H, Advanced-stage (III-IV) cases. Low K17 (n = 23 [30%]; median = 33 months); high K17 (n = 54 [70%]; median = 18 months). HR = 1.772 (95% CI, 0.941-3.151); P =.081. P values were calculated with the log-rank test. Due to low number of advanced-stage patients, a link between K17 expression and pathologic stage could not be performed within the discovery cohort. The HRs and P values are shown for all panels. I, Graph showing expression of K17 immunohistochemistry for each case within the same tumor stage category. PathSQ score ranges from 0% to 100% in both categories. P value was calculated using the Mann-Whitney test. HR, hazard ratio; K17, keratin 17; ns, not significant; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 3

Validation of K17 as an independent negative prognostic biomarker of PDAC. Kaplan-Meier curves for the overall survival analysis of K17 from PDAC cases in the discovery (A) and validation (B) cohorts. A, Low K17 (n = 15 [25%]; median = 19 months) and high K17 (n = 59 [75%]; median = 27 months). HR = 3.217 (95% CI, 1.604-6.450); P = .0177. B, Low K17 (n = 76 [25%]; median = 42 months); high K17 (n = 229 [75%]; median = 25 months). HR = 1.511 (95% CI, 1.065-2.124); P = .0338. Forest plots showing the multivariate analysis factors in the discovery (C) and validation (D) cohorts. Kaplan-Meier curves depict the overall survival, which integrate K17 status and the lymph node status for the combined discovery (E) and validation (F) cohorts. E, Lymph node–negative cases. Low K17 (n = 25 [26%]; median = 36 months); high K17 (n = 70 [74%]; median = 22 months). HR = 2.119 (95% CI, 1.174-3.823); P =.0258. F, Lymph node–positive cases. Low K17 (n = 76 [27%]; median = 33 months); high K17 (n = 208 [73%]; median = 23 months). HR = 1.272 (95% CI, 0.9163-1.765); P = .1659. Kaplan-Meier curves depict the overall survival, which integrate K17 status and pathological stage for the combined discovery (G) and validation (H) cohorts. G, Early-stage (I-IIB) cases. Low K17 (n = 77 [26%]; median = 36.6 months); high K17 (n = 225 [74%]; median = 25.1 months). HR = 1.441 (95% CI, 1.042-1.997); P =.0392. H, Advanced-stage (III-IV) cases. Low K17 (n = 23 [30%]; median = 33 months); high K17 (n = 54 [70%]; median = 18 months). HR = 1.772 (95% CI, 0.941-3.151); P =.081. P values were calculated with the log-rank test. Due to low number of advanced-stage patients, a link between K17 expression and pathologic stage could not be performed within the discovery cohort. The HRs and P values are shown for all panels. I, Graph showing expression of K17 immunohistochemistry for each case within the same tumor stage category. PathSQ score ranges from 0% to 100% in both categories. P value was calculated using the Mann-Whitney test. HR, hazard ratio; K17, keratin 17; ns, not significant; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 4

High K17 expression is correlated with shorter survival in patients treated with gemcitabine-based, 5-FU–based therapies or radiotherapy. Kaplan-Meier curves for the overall survival analysis of K17 from PDAC cases of all stages in both cohorts combined. A, Overall survival in patients with PDAC treated with 5-FU–based therapies. Low K17 (n = 29 [32%]; median = 69.5 months); high K17 (n = 61 [68%]; median = 31 months). HR = 2.256 (95% CI, 1.220-4.172); P = .0118. B, Overall survival in patients with PDAC treated with gemcitabine-based therapies. Low K17 (n = 54 [30%]; median = 38.4 months); high K17 (n = 126 [70%]; median = 27.2 months). HR = 1.549 (95% CI, 1.050-2.284); P = .0309. C, Overall survival in patients with PDAC treated with radiotherapy. Low K17 (n = 8 [24%]; median = 38 months); high K17 (n = 25 [76%]; median = 28 months). HR = 3.678 (95% CI, 1.031-13.12); P = .0214. P values were calculated using the log-rank test. Gemcitabine-based therapies include the following: gemcitabine, gemcitabine and nab-paclitaxel, or any other chemotherapy combination that includes gemcitabine; 5-FU–based therapies include the following: FOLFIRINOX, FOLFIRI, FOLFOX, or any other chemotherapy combination that includes 5-fluorouracil. Radiotherapy includes all forms of radiation in combination with or without other chemotherapeutic agents. HR, hazard ratio; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 5

K17 immunohistochemistry prediction of overall survival in lymph node–positive and advanced PDAC according to receipt of chemotherapy. Kaplan-Meier curve for overall survival in patients with lymph node–positive PDAC receiving adjuvant FOLFIRINOX (A) and patients receiving gemcitabine (B). A, Low K17 (n = 18 [31%]; median = 69.5 months); high K17 (n = 40 [69%]; median = 31 months). HR = 2.141 (95% CI, 1.006-4.558); P = .0431. B, Low K17 (n = 18 [35%]; median = 34 months); high K17 (n = 34 [65%]; median = 31 months). HR = 1.078 (95% CI, 0.5346-2.172); P = .8295. Kaplan-Meier curve for overall survival in patients with advanced-stage PDAC receiving adjuvant 5-FU–based chemotherapies (C) and in patients receiving adjuvant gemcitabine-based chemotherapies (D). C, Low K17 (n = 15 [29%]; median = 29 months); high K17 (n = 37 [71%]; median = 25 months). HR = 5.016 (95% CI, 2.258-11.14); P = .003. D, Low K17 (n = 10 [24%]; median = 20.3 months); high K17 (n = 31 [76%]; median = 18.4 months). HR = 1.117 (95% CI, 0.5201-2.401); P = .7684. P values were calculated using the log-rank test. Gemcitabine-based therapies include the following: gemcitabine, gemcitabine and nab-paclitaxel, or any other chemotherapy combination that includes gemcitabine. The 5-FU–based therapies include the following: FOLFIRINOX, FOLFIRI, FOLFOX, or any other chemotherapy combination that includes 5-fluorouracil. HR, hazard ratio; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma.

FIGURE 6

High K17 expression predicts poor therapeutic response to 5-FU–based and gemcitabine (Gem)–based therapies in advanced-stage PDAC. A-D, Kaplan-Meier curve for progression-free survival in patients with lymph node–positive PDAC treated with gemcitabine-based or 5-FU–based therapies with low K17 expression (A) and high K17 expression (B). A, Gemcitabine (n = 41 [51%]; median = 23.5 months); 5-FU (n = 29 [36%]; median = 67.2 months); not treated (n = 10 [13%]; median = 28.7 months). B, Gemcitabine (n = 100 [51%]; median = 18.9 months); 5-FU (n = 61 [31%]; median = 21.0 months); not treated (n = 34 [17%]; median = 13.3 months). P values were calculated using the log-rank test. Kaplan-Meier curve for progression-free survival in patients with advanced-stage (III-IV) PDAC treated with gemcitabine-based or 5-FU–based therapies with low K17 expression (C) and high K17 expression (D). C, Gemcitabine (n = 10 [34%]; median = 7.5 months); 5-FU (n = 15 [52%]; median = 21.3 months); not treated (n = 4 [14%]; median = 19.1 months). D, Gemcitabine (n = 31 [40%]; median = 9.9 months); 5-FU (n = 35 [45%]; median = 15.2 months); not treated (n = 12 [15%]; median = 11.7 months). E, Comparison of K17 expression levels in responders and nonresponders with lymph node–positive PDACs treated with gemcitabine- or 5-FU–based therapies using RECIST v1.1 criteria. P value was calculated using the Mann-Whitney test. F, Percentage of patients’ response based on RECIST v1.1 in low and high K17-expressing lymph node–positive PDACs. Gemcitabine-based therapies include the following: gemcitabine, gemcitabine and nab-paclitaxel, or any other chemotherapy combination that includes gemcitabine; 5-FU–based therapies include the following: FOLFIRINOX, FOLFIRI, FOLFOX, or any other chemotherapy combination that includes 5-fluorouracil. P value was calculated with the Welch test. HR, hazard ratio; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma; RECIST, Response Evaluation Criteria in Solid Tumors.

FIGURE 6

High K17 expression predicts poor therapeutic response to 5-FU–based and gemcitabine (Gem)–based therapies in advanced-stage PDAC. A-D, Kaplan-Meier curve for progression-free survival in patients with lymph node–positive PDAC treated with gemcitabine-based or 5-FU–based therapies with low K17 expression (A) and high K17 expression (B). A, Gemcitabine (n = 41 [51%]; median = 23.5 months); 5-FU (n = 29 [36%]; median = 67.2 months); not treated (n = 10 [13%]; median = 28.7 months). B, Gemcitabine (n = 100 [51%]; median = 18.9 months); 5-FU (n = 61 [31%]; median = 21.0 months); not treated (n = 34 [17%]; median = 13.3 months). P values were calculated using the log-rank test. Kaplan-Meier curve for progression-free survival in patients with advanced-stage (III-IV) PDAC treated with gemcitabine-based or 5-FU–based therapies with low K17 expression (C) and high K17 expression (D). C, Gemcitabine (n = 10 [34%]; median = 7.5 months); 5-FU (n = 15 [52%]; median = 21.3 months); not treated (n = 4 [14%]; median = 19.1 months). D, Gemcitabine (n = 31 [40%]; median = 9.9 months); 5-FU (n = 35 [45%]; median = 15.2 months); not treated (n = 12 [15%]; median = 11.7 months). E, Comparison of K17 expression levels in responders and nonresponders with lymph node–positive PDACs treated with gemcitabine- or 5-FU–based therapies using RECIST v1.1 criteria. P value was calculated using the Mann-Whitney test. F, Percentage of patients’ response based on RECIST v1.1 in low and high K17-expressing lymph node–positive PDACs. Gemcitabine-based therapies include the following: gemcitabine, gemcitabine and nab-paclitaxel, or any other chemotherapy combination that includes gemcitabine; 5-FU–based therapies include the following: FOLFIRINOX, FOLFIRI, FOLFOX, or any other chemotherapy combination that includes 5-fluorouracil. P value was calculated with the Welch test. HR, hazard ratio; K17, keratin 17; PDAC, pancreatic ductal adenocarcinoma; RECIST, Response Evaluation Criteria in Solid Tumors.