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


Immunohistochemical analysis for K17 in normal endocervical tissue showed sporadic staining of subcolumnar reserve cells, predominately located near the squamocolumnar junction. Dual immunofluorescence staining for K7 and K17 showed these reserve cells to be K17+/K7– and were a discrete population of cells present subjacent to K7-positive endocervical columnar epithelium Image 1. K17 staining was not detected in endocervical columnar epithelium with the exception of occasional staining of immature squamous metaplasia in five (12%) of 41 benign cases. K17 was not detected in columnar epithelium in the other benign categories with the exception of intense and diffuse staining of reserve cells only in 16 (100%) of 16 microglandular hyperplasia cases Image 2.

Image 1.

Keratin 17 (K17) expression in normal endocervical mucosa. Keratin immunohistochemistry: (A, ×200), (B, ×600), (C, ×600), and (D) dual immunofluorescence staining, K17 (green), K7 (red) (×600). K17 immunohistochemistry highlights a population of subcolumnar cells that are discrete from the surface K7-positive columnar cells. These cells are predominately located near the squamocolumnar junction.

Image 2.

Keratin 17 (K17) immunohistochemical expression in benign endocervical patterns. A, Tunnel clusters (×100). B, Endometriosis (×400). C, Tubal metaplasia (×200). D, Microglandular hyperplasia (×200). Intense K17 staining in subcolumnar reserve cells in microglandular hyperplasia but only faint staining in other benign endocervical patterns.

For the 90 adenocarcinoma tissue specimens analyzed, the mean patient age was 49 years. The mean time a patient was followed was 105 months (or 8.75 years). Around 16% of study patients had a high-stage (stage III or IV) tumor; 3.3% had positive lymph nodes and 10% had distant metastasis. p16 was present in 79% of cases. The mean K17 score was 34% Table 1 and Table 2.

K17 was detected in most AIS and adenocarcinoma cases (88% and 83%, respectively), but there was no K17 expression seen in the columnar epithelium of any normal/benign tissue samples Image 3. To distinguish AIS/adenocarcinoma from normal mucosa/benign endocervical lesions, a potential diagnostic cutoff value of K17 was determined using a two-step process: a histogram, first generated for all 178 tissue samples, revealed a natural cutoff point for K17 at or below 10%. Then, NLR tests confirmed the K17 cutoff value of 0% or more to have the lowest NLR (NLR = .16), indicating a moderate refinement in the diagnostic process. Thus, this threshold may likely be useful for clinical decision-making purposes when combined with other relevant and/or expert clinician judgment. At this K17 diagnostic level of 0% or more, sensitivity was maximized (84.4%), as was specificity (100%).

Image 3.

Keratin 17 immunohistochemical expression in adenocarcinoma in situ (AIS) and invasive adenocarcinoma. A, AIS (×100). B, AIS (×400). C, Invasive adenocarcinoma (×200). D, Invasive adenocarcinoma (×100).

Within the cancer-specific tissues, there was no difference in continuous K17 score between low and high stage, presence or absence of p16, or mortality. There was a significant difference in rank score between tissues that were originally from Massachusetts compared with Stony Brook (table not shown).

Distribution of K17 was examined and two values were identified as natural break points in the distribution. A polychotomous K17 variable was created to indicate those with K17 values from 0% to 39%, 40% to 89%, and 90% to 100% Image 4. Using these categories, a second reviewer (blinded to the first reviewer's scores) reviewed and scored each slide. Interrater reliability was very strong (raw agreement = 96.7%; weighted κ = 0.95; 95% confidence interval, 0.90–1.0). After the consensus process, Kaplan-Meier survival curves were plotted Figure 1, and the Cox proportional hazards model found K17 to be a significant predictor of survival (Wald P = .03) Table 3 and Table 4. While the 40% to 89% category did not significantly differ from the less than 40% category (hazard ratio [HR], 2.12; P = .9), those with 90% to 100% K17 scores had 3.47 times higher risk of death compared with the risk of death for those with a less than 40% K17 score (P = .01). Univariate analyses also revealed that the absence of p16 and high stage were both significantly associated with an increase in survival HR (p16 HR, 3.2 [P = .003]; stage HR, 5.62 [P < .01]) (Table 3 and Table 4). While the relationship seemed to slightly attenuate after controlling for both stage and p16 in the multivariable analysis, the 90%+ K17 group remained a significant predictor of hazard of death (HR, 2.76; P = .0479). There were no significant interactions between K17 and p16 or K17 and stage; thus, no interaction variables were included in the final multivariable analysis. Finally, a subgroup analysis revealed that for those with p16 present, survival HRs increased for both K17 medium and high groups (40%-89% vs <40%: HR, 3.71 [P = .03]; 90%-100% vs <40%: HR, 5.19 [P = .01]). Within the p16 absent subgroup, the relationship between K17 and survival became insignificant. However, this may be likely due to the limited sample size (n = 18). Additional subgroup analyses revealed that within the low-stage subgroup, the 90%+ K17 group showed a marginally significant higher hazard of death compared with the less than 40% K17 group (HR, 3.09; P = .097).

Figure 1.

Kaplan-Meier curves for keratin 17 as a polychotomous variable.

Image 4.

Keratin 17 immunohistochemical expression in invasive adenocarcinoma. A, Low K17 expression (×200). B, Intermediate K17 expression (×200). C, High K17 expression (×200).