Keratin 17 Is a Prognostic Biomarker in Endocervical Glandular Neoplasia

Daniel Mockler, MD; Luisa F. Escobar-Hoyos, PhD; Ali Akalin, MD, PhD; Jamie Romeiser, MPH; A. Laurie Shroyer, PhD; Kenneth R. Shroyer, MD, PhD


Am J Clin Pathol. 2017;148(3):264-273. 

In This Article

Materials and Methods

Institutional Review Board Approval

The study was coordinated as part of the Stony Brook University Institutional Review Board Committee on Research in Human Subjects (CORIHS protocol 94651). This study's biomarker assessments used formalin-fixed, paraffin-embedded (FFPE) surgical tissue specimens that had been linked to deidentified patient records that included information on patient risk characteristics, pathology/histological tumor assessments, radiation and chemotherapy treatment details, and clinical outcomes (ie, survival status, disease recurrence, and cause of death assessments).

Case Selection

This study included a total of 178 FFPE surgical tissue blocks that were retrospectively selected from the archival collections of both the Stony Brook BioBank and the Department of Pathology at the University of Massachusetts, in compliance with institutional review board–approved protocols. For cases from the Stony Brook BioBank, a search for each diagnostic category was performed encompassing the years 1980 to 2010, and all of the endocervical adenocarcinoma cases were selected while the most recent cases were selected for the other diagnostic categories as described below. For cases from the Department of Pathology at the University of Massachusetts, two separate searches encompassing the years 2000 to 2013 were performed for the diagnosis of adenocarcinoma in situ (AIS) and invasive endocervical adenocarcinoma. Thirty-six cases of invasive adenocarcinoma (between 2000 and 2013) and 19 of the most recent cases of AIS (between 2009 and 2013) were selected. These study cases comprised the following diagnostic categories: adenocarcinoma (n = 90), AIS (n = 32), microglandular hyperplasia (n = 16), tubal metaplasia (n = 13), endometriosis (n = 5), tunnel clusters (n = 2), and normal (n = 5). In all cases, a tissue block was selected following histologic review by a pathologist (K.R.S., D.M., and A.A.) of H&E-stained sections from hysterectomy, cervical cone, loop electrosurgical excision procedure, and/or cervical biopsy to confirm that diagnostic tissue as originally reported was adequately represented in the remaining tissue blocks. Cases that had insufficient residual tissue were not included in the study.


Formalin-fixed, paraffin-embedded tissue blocks were sectioned at 5 μm, and the whole tissue sections were mounted on charged glass slides (Superfrost Plus; Fisher Scientific, Pittsburgh, PA). After incubation at 60°C for 1 hour, tissue sections were deparaffinized in xylene and rehydrated using graded alcohols. Antigen retrieval was performed using a decloaking chamber at 120°C for 10 minutes in citrate buffer (20 mmol, pH 6.0; Vector Laboratories, Burlingame, CA) for K17 and p16, as well as a high pH antigen unmasking solution (Vector Laboratories) for Ki-67. Endogenous peroxidase activity was blocked by applying 3% hydrogen peroxide for 5 minutes. Nonspecific antibody binding was blocked by preincubation with 5% horse serum. Primary antibodies used were mouse monoclonal anti–human K17 antibody (clone E3; Abcam, Cambridge, MA; 4°C overnight), p16INK4a (CINtec, 60 minutes at room temperature; Ventana Medical Systems, Tucson, AZ), and Ki-67 (clone MIB-1, 1:100 dilution, 60 minutes at room temperature; DAKO, Carpentaria, CA). Negative controls were performed on all cases using an equivalent concentration of a subclass-matched mouse immunoglobulin, generated against unrelated antigens (BD, Franklin Lakes, NJ), in place of primary antibody. Following incubation with the primary antibody, slides were processed by an indirect avidin-biotin–based immunoperoxidase method using biotinylated horse secondary antibodies (R.T.U. Vectastain Universal Elite ABC kit; Vector Laboratories), developed in 3,3'-diaminobenzidine (DAKO) and counterstained with hematoxylin. K17 staining was scored using PathSQ,[14] a manual semiquantitative scoring system that is based on the proportion of tumor cells with strong (2+) staining. Ki-67 staining was scored as a percentage of tumor cells with positive nuclear staining, and p16INK4a-stained sections were scored positive when strong (2+) nuclear and cytoplasmic staining of more than 90% of lesional cells was present.


Briefly, sections were cut at 4 μm and heat immobilized at 60°C for 60 minutes. After deparaffinization with xylene and rehydration through graded alcohols, antigen retrieval was performed in a High PH buffer (Vector Laboratories) at 120°C for 10 minutes in a Decloaking Chamber (Biocare Medical, Walnut Creek, CA). Sections were rinsed in phosphate-buffered saline (PBS) and blocked for 1 hour at room temperature using 5% nonfat milk. Primary antibodies were added and incubated overnight at 4°C using the following: monoclonal CK17 (E3 clone) prediluted (Abcam) and rabbit polyclonal CK7, diluted 1:1,000 (Abcam). Sections were rinsed in PBS and incubated with Alexa secondary 488 goat anti-mouse and 594 goat anti-rabbit, both diluted to 1:250, incubated for 2 hours at room temperature in a light-proof chamber. The slides were rinsed in PBS and cover slipped using Vectashield mounting media with 4',6-diamidino-2-phenylindole (Vector Laboratories). Negative controls were performed on all cases.

Statistical Analysis

Diagnostic Value of K17. To determine an optimal diagnostic K17 staining cutoff value to differentiate between normal vs abnormal tissue, biomarker findings were dichotomized into two groups: normal tissue vs abnormal (AIS + cancer) tissue. As a visual inspection of the K17 scoring distribution identified potential natural data-driven cut-points, a negative likelihood ratio (NLR) analysis was used to identify the lowest K17 value (as the new "threshold") that would differentiate a normal vs an abnormal diagnosis; the NLR represents the probability of a negative test result given the presence of the disease (ie, false-negative probability), divided by the probability of a negative test result given the absence of disease (ie, true-negative probability). A low NLR would indicate a low chance of misclassifying tissue as "normal" when actually the tissue is abnormal. NLRs are known for ruling conditions out and thus are an appropriate measure to use when defining normal tissue using K17 thresholds.

Prognostic Value of 17. Cox proportional hazards models for survival time from date of index surgical biopsy were generated for K17 as both a continuous variable and through an exploratory categorical cutoff process using increments of 10, from 10% or more to 90% or more. Once the cutoffs were determined, a second reviewer (D.M.) rescored the K17 score for all cancer tissue slides. This second reviewer was blinded to the first reviewer's (K.R.S.) scores. Interrater reliability was examined using a weighted κ score. All disagreements were then reviewed by both raters, and a consensus value was achieved. Kaplan-Meier survival curves were then generated for the categorical K17 variable. Finally, the relationship between K17, p16, stage, and survival was explored as univariate predictors of survival, testing for interactions with K17, and then as subgroup analyses. Finally, a multivariable Cox regression was performed using any predictors with a univariate significance of P < .1.

As this was an exploratory, pilot study analysis of K17 biomarker findings, all calculations were performed at a 95% confidence level using SAS 9.4 software (SAS Institute, Cary, NC).