KRT81 and HNF1A expression in pancreatic ductal adenocarcinoma: investigation of predictive and prognostic value of immunohistochemistry‐based subtyping

Abstract

Even after decades of research, pancreatic ductal adenocarcinoma (PDAC) remains a highly lethal disease and responses to conventional treatments remain mostly poor. Subclassification of PDAC into distinct biological subtypes has been proposed by various groups to further improve patient outcome and reduce unnecessary side effects. Recently, an immunohistochemistry (IHC)‐based subtyping method using cytokeratin‐81 (KRT81) and hepatocyte nuclear factor 1A (HNF1A) could recapitulate some of the previously established molecular subtyping methods, while providing significant prognostic and, to a limited degree, also predictive information. We refined the KRT81/HNF1A subtyping method to classify PDAC into three distinct biological subtypes. The prognostic value of the IHC‐based method was investigated in two primary resected cohorts, which include 269 and 286 patients, respectively. In the second cohort, we also assessed the predictive effect for response to erlotinib + gemcitabine. In both PDAC cohorts, the new HNF1A‐positive subtype was associated with the best survival, the KRT81‐positive subtype with the worst, and the double‐negative with an intermediate survival (p < 0.001 and p < 0.001, respectively) in univariate and multivariate analyses. In the second cohort (CONKO‐005), the IHC‐based subtype was additionally found to have a potential predictive value for the erlotinib‐based treatment effect. The revised IHC‐based subtyping using KRT81 and HNF1A has prognostic significance for PDAC patients and may be of value in predicting treatment response to specific therapeutic agents.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) represents 90% of pancreatic cancers [1]. It is a highly malignant disease with a 5‐year survival rate of ~10% [2] and is projected to become the second leading cause of cancer death before 2030 [3]. Because of its nonspecific symptoms and aggressive biological behaviors, the majority of PDAC patients are diagnosed at an advanced stage and have a median survival of only 10–12 months [4]. Frequent recurrence after surgery and a lack of effective treatment options strongly contribute to the poor outcome of affected patients.

Despite continuous efforts over the past years, novel therapeutic methods have turned out to have minimal effect in improving the prognosis of PDAC patients and predictive biomarkers are still lacking. Following the path of treatment in other cancer entities, one way to move forward is the development of molecularly based patient stratification strategies and to identify different types of PDAC with differing therapy responses. In this scenario, molecular alterations of pancreatic cancer need urgently to be further clarified to identify vulnerabilities for individualized molecularly guided patient treatments.

Over decades, PDAC had been perceived as a molecularly ‘boring’ disease, homogenously driven by KRAS and TP53 mutations. However, research conducted in the last decade indicates that pancreatic cancer has a more complex makeup than previously considered with broad intra‐ and intertumoral heterogeneity [5, 6, 7]. By RNA sequencing and DNA translocation profiling, the heterogeneity could be elucidated and different molecular subtypes of PDAC were defined, initially as the classical, exocrine‐like, and quasi‐mesenchymal types by Collisson et al [8]. Through further investigations, these three subtypes were summarized into two subtypes by other groups, differentiating only between a classical/pancreatic progenitor subtype, which is enriched in GNAS mutations and includes the former classical and exocrine‐like subtypes [9]; and the basal‐like/squamous subtype, with highly abundant TP53 mutations, consistent with the former quasi‐mesenchymal subtype [10]. Possible contributions to better selection of treatment for individual patients or refined definition of patient‐specific prognosis has been discussed for some of these new molecular classifiers [8, 10, 11, 12].

Although the novel subtyping methods could stratify tumors into prognosis‐related and potentially treatment relevant groups, the technical requirements as well as high per case costs continue to limit the wide clinical application of individualized RNA expression profiling for a clinically relevant subtype classification of PDAC patients. Therefore, a cheap, widely available immunohistochemical alternative, which can be used in clinical practice, is desperately needed.

In our previous study, we suggested hepatocyte nuclear factor 1A (HNF1A) and cytokeratin‐81 (KRT81) as two surrogate protein markers to categorize the subtypes with innate molecular features [13]. By evaluating the expression levels of these proteins by immunohistochemistry (IHC), we proposed three subtypes described as KRT81‐positive, HNF1A‐positive, and double‐negative; and concordance between our IHC‐based subtypes and those defined by Collisson et al [8] was shown. The KRT81‐positive group recaptured the main features of the quasi‐mesenchymal subtype, while the HNF1A‐positive group showed an association with the exocrine‐like subtype and the double‐negative group with the classical subtype [13]. The biological and prognostic relevance of the IHC‐based subtypes has been further confirmed in our subsequent studies [14]. The subtyping method was applied in two independent primary resected PDAC cohorts, and we could identify robustly different prognostic PDAC subtypes: the KRT81‐positive subtype with the worst survival, the HNF1A‐positive subtype with the best survival, and the double‐negative subtype with an intermediate survival. However, so far, a certain number of double‐positive cases, which we always encountered, had to be excluded as not definitely classifiable, as they represented the smallest subtype group and patients showed variable biological behavior.

In the study presented here, we expand our previously investigated cohort from 130 to 269 cases, reinvestigate the prognostic value of our IHC‐based subtyping and revise our subtyping method to avoid the exclusion of double‐positive cases. Furthermore, we validate the revised subtyping model in the tissue cohort of a controlled, prospectively documented, and randomized phase III trial (CONKO‐005), which investigated the efficacy of erlotinib in the adjuvant treatment of resectable PDAC.

Materials and methods

Patient cohorts

Two cohorts, one retrospectively assembled from our archives, the second prospectively recruited in the context of the prospective CONKO‐005 trial, were investigated in this study. The primary resected cohort (cohort 1, Munich), consisting of 269 individuals who received an elective pancreatic resection for PDAC, between 2007 and 2016 at the Department of Surgery, Klinikum rechts der Isar, TU Munich, Germany. Grading and staging followed the World Health Organization (WHO) recommendations at the time of cohort generation (TNM classification of the seventh edition). Clinical data and follow‐up were obtained from a patient database, by reviewing the medical charts and directly contacting the patients and/or their physicians. The observation period for each patient started with the surgical resection. The study was approved by the Institutional Review Board (ethics committee) of the TU Munich, Germany (documents no. 1926/2007 and 126/2016 S).

Second, a sub‐cohort of the CONKO‐005 was investigated (cohort 2, Berlin) consisting of 286 patients with resectable PDAC after R0 resection between 2008 and 2013. The tissue cohort was prospectively generated accompanying this previous published controlled, open‐label, multicenter, randomized phase III trial. Patients received six cycles adjuvant gemcitabine (1,000 mg/m² intravenously; days 1, 8, 15; every 4 weeks) with or without erlotinib (100 mg orally once per day; days 1–28; every 4 weeks). The primary endpoint was to improve disease‐free survival (DFS) by adding erlotinib to gemcitabine, which achieved a statistical power of 80% from 14 to 18 months. The protocol required patients to have R0 resection with histologically tumor‐free surgical margins as defined by International Union Against Cancer criteria (sixth edition, 2002). Overall, 436 patients were randomly assigned at 57 study centers between April 2008 and July 2013. With a median follow‐up of 54 months, no difference in median DFS (GemErlo 11.4 months; Gem 11.4 months) or median overall survival (GemErlo 24.5 months; Gem 26.5 months) was detected. There was a trend toward long‐term survival in favor of GemErlo (estimated survival after 1, 2, and 5 years for GemErlo was 77%, 53%, and 25% versus 79%, 54%, and 20% for Gem, respectively) [15]. Histological examinations and assessments were conducted in the pathology department of the recruiting centers under the WHO recommendations (TNM classification of the seventh edition). Follow‐up examinations were performed every 3 months for 2 years or until disease recurrence, and then every 6 months for up to 5 years or until death. The start point of the observation was the surgical resection. Use of patient data was approved by the Central Ethics Committee of Berlin (EudraCT2007‐003813‐15). The study has been registered in the German Clinical Trials Register (DRKS) with the registration number DRKS00000247. Detailed study outcome data have been published elsewhere [15].

Patient and public involvement

All patient data were collected after informed consent and retrospectively analyzed in accordance with the ethics committee vote. Patients had the right to withdraw their permission of data and material usage at all stages of the study. Because of the retrospective nature of the study, patients were not directly involved in the study design or plans to disseminate study results. Nevertheless, patient interests were considered at all points of the study, as the avoidance of unnecessary side effects due to better tumor stratification and therapy selection is of self‐evident value to future patients.

This study was performed in accordance with the Declaration of Helsinki. Written consent of subjects was obtained.

Immunohistochemistry

Tissue microarrays of primary tumors were constructed and used for immunohistochemical staining. In both cohorts, two to three tissue cores with a diameter of 1‐mm tumor area were taken from each FFPE block and inserted into the blank tissue microarray block. The tumor areas were previously marked by a board‐certified pathologist (AM). TMAs were made using a tissue microarrayer (Beecher Instruments, Tartu, Estonia).

For cohort 1 (Munich), IHC staining of cases from 2007 to 2011 was performed by hand; more details of the staining method were shown in our previous published paper [14]. Staining of cases from 2012 to 2016 was performed on a Ventana Benchmark XT (Ventana Medical Systems, Tucson, AZ, USA). For cohort 2 (Berlin), IHC staining of all the cases was performed on a Ventana Benchmark XT (Ventana Medical Systems). Epitopes were unmasked by boiling slides in citrate buffered distilled water (pH 6) for 15 min in a pressure cooker and allowing a 30‐min cool‐down period. The Dako REAL Peroxidase Detection System kit (Dako, Jena, Germany) was used according to the manufacturers' specifications, including the ready‐to‐use anti‐rabbit/mouse secondary antibody (catalog no. K5003). Primary antibodies used were rabbit polyclonal anti‐HNF‐1A antibody (catalog no. sc‐8986) at a dilution of 1:100, and mouse monoclonal anti‐keratin 81 antibody (catalog no. sc‐100929) at a dilution of 1:500, both from Santa Cruz Biotechnology Inc (Dallas, TX, USA). Primary antibodies were incubated for 2 h at room temperature. Slides were digitalized using a Leica Aperio AT2 slide scanner (Leica, Wetzlar, Germany) and examined with an Olympus CX31 light microscope (Olympus Corporation, Tokyo, Japan). The images were obtained using Leica Aperio ImageScope software (version 12.4.6.5003, Leica).

IHC assessment

The IHC staining of all slides was reviewed and evaluated by an experienced, board‐certified pathologist (AM). The staining intensity scores of HNF1A were graded as follows: 0, no staining; 1, ‘weak’ indicated by barely discernable light brown nuclear staining only visible at high magnifications (at least ×100); 2, ‘medium’ identified by heterogeneous nuclear staining of varying shades of medium to dark brown; and 3, ‘strong’ where there was homogeneous dark brown staining. Tumor cells with medium to strong nuclear staining were classified as HNF1A‐positive, regardless of the percentage of stained tumor cells. Normal small intestine was used as positive control, due to its moderate to strong positivity for HNF1A in epithelial cells.

KRT81 staining was evaluated using a threshold of ≤30% positive tumor cells as KRT81‐negative, and >30% positive tumor cells as KRT81‐positive, regardless of staining intensity. The proportion of KRT81positive tumor cells was visually estimated, as it is hard to recognize the cellular borders and manually count the exact numbers of single tumor cells under an interferential background of strong cytoplasmic staining of KRT81. Thirty percentage was selected as a cutoff value because it was high enough to avoid possible overinterpretation caused by KRT81‐positive budding tumor cells while maintaining the necessary detection sensitivity. Most cases in our cohorts could be clearly classified as above or below the threshold resulting in a minimal number of borderline cases. Samples of normal skin served as positive control as KRT81 is strongly expressed in hair follicles.

In the previously used four‐tier classification system, tumors negative for both HNF1A and KRT81 were classified as double‐negative, and tumors with expression of both markers above the thresholds double‐positive. In the three‐tier classification system proposed in this study, tumors with expression of both markers were included in the new HNF1A‐positive subtype due to their comparable clinical behavior.

Statistical analysis

All statistical analyses were carried out using SPSS software (IBM Corp., Armonk, NY, USA, version 27.0). The Kaplan–Meier analysis was performed to estimate and compare survival curves between different subgroups followed by log‐rank test. Associations of clinicopathological features with IHC‐based subtypes were examined using chi‐square test (or Fisher's exact test when appropriate) for categorical variables, Spearman correlation coefficient for two interval variables, and one‐way ANOVA for interval variable with nominal/ordinal variables. The independent prognostic factors in PDAC were determined by univariate Cox regression and multivariate Cox regression. p values lower than 0.05 were considered as statistically significant.

For multivariable survival analysis, tumor grade and UICC (Union for International Cancer Control) stage were considered as prognostically relevant factors. Additionally, in cohort 1 the resection status after surgery by histopathological examination was considered. In cohort 2, the CA19.9 value in blood was tested after surgery as an indicator of residual tumor. The cases were divided into three groups according to their postoperative CA19.9 values (low: ≤100 U/ml; medium: >100 U/ml and ≤500 U/ml; high: >500 U/ml).

Results

Frequency distribution and clinicopathologic patient characteristics

By using the revised algorithm, we could assign every case in both cohorts to one of the defined groups. For the staining of a double‐positive case, see Figure 1. For example images of HNF1A and KRT81 staining, see supplementary material, Figure S1. In cohort 1, the most prevalent subtype was double‐negative with 113 cases (42.0%), followed by the HNF1A‐positive subtype with 84 cases (31.2%), the KRT81‐positive subtype with 51 cases (19.0%), and the double‐positive subtype 21 cases (7.8%). In cohort 2, the double‐negative subtype remained the most numerous with 143 cases (50.0%) as in the first cohort, followed by the KRT81‐positive subtype (77cases, 27.0%) and the HNF1A‐positive subtype (47 cases, 16.4%).The double‐positive subtype was still the least common with 19 cases (6.6%) (supplementary material, Figure S2). An overview table of the clinicopathologic characteristics of the investigated patient cohorts is provided as supplementary material, Table S1. An overview table of the clinicopathologic characteristics according to IHC‐based subtypes is provided as supplementary material, Table S2.

Figure 1.

Representative images of (A) HNF1A and (B) KRT81 staining of a double‐positive case (scale bar: 300 μm).

In this study, we further analyzed the association between the revised IHC subtypes and clinicopathologic characteristics. Sex was found to have a significant correlation with subtypes in a previous cohort [14]; however, the distribution did not differ in both cohorts investigated in this study (cohort 1: p = 0.227, cohort 2: p = 0.346). Additionally, no associations were observed between subtype and age or other prognostic variables, namely grade, UICC stage, nodal status, and surgical resection status (cohort 1) or postoperative CA19.9 levels (cohort 2) (supplementary material, Table S3).

Prognostic value of IHC‐based subtype in PDAC cohorts

In a direct comparison between the previously used four‐tier and the three‐tier classification system proposed in this study, we found that the previously unclassifiable double‐positive cases showed relatively good survival, similar to HNF1A‐positive cases, allowing us to integrate cases of both subtypes into a combined new HNF1A‐positive subtype proposed here. Kaplan–Meier survival analysis for each subtype in cohort 1 indicated that the new ‘combined’ HNF1A‐positive subtype was associated with the best survival (median 619 days), followed by the double‐negative subtype with an intermediate survival (median 600 days). The KRT81‐positive subtype was associated with the worst survival (median 413 days; p < 0.001). For the clinical study cohort 2, similar survival curves for the different subtypes were observed; the new ‘combined’ HNF1A‐positive group was associated with the best outcome with a median survival time of 1,067 days, followed by the double‐negative subtype (median 821 days) and the KRT81‐positive subtype (median 531 days; p < 0.001) (Figure 2A–D). Multivariate analysis in both cohorts showed that IHC‐based subtypes were an independent prognostic factor (p = 0.001 in cohort 1; p = 0.003 in cohort 2). As expected, other independent prognostic factors were tumor stage (p < 0.001 in cohort 1; p = 0.002 in cohort 2), incomplete resection status in cohort 1 (p = 0.002) and postoperative CA19.9 level in cohort 2 (p < 0.001). Tumor grade in cohort 2 was found to significantly relate to survival (cohort 2 p = 0.008; cohort 1 p = 0.049). For an overview of multivariate analysis, see Table 1.

Figure 2.

Kaplan–Meier curves of IHC‐derived subtypes of PDAC. (A, C) Kaplan–Meier curves of four‐tier subtypes and three‐tier subtypes in cohort 1. (B, D) Kaplan–Meier curves of four‐tier subtypes and three‐tier subtypes in cohort 2. p values calculated by log‐rank test. Ticks, censored cases.

Table 1.

Multivariate analysis of cohort 1 and cohort 2 calculated by Cox regression modeling

Cohort 1 – Munich
Variables	Level	HR (Hazard Ratio, 95% CI)	p
Subtype			0.001
	Double‐neg
	New HNF1A‐pos	0.774 (0.552–1.085)
	KRT81‐pos	1.667 (1.121–2.479)
UICC stage			<0.001
	Stage I
	Stage IIA	0.500 (0.174–1.437)
	Stage IIB	0.899 (0.325–2.487)
	Stage III	0.877 (0.273–2.815)
	Stage IV	3.232 (0.994–10.513)
R1 status		1.642 (1.200–2.246)	0.002
Grade			0.049
	G1
	G2	1.350 (1.001–1.820)

Cohort 2 – Berlin
Variables	Level	HR (Hazard Ratio, 95% CI)	p
Subtype			0.003
	Double‐neg
	New HNF1A‐pos	0.650 (0.434–0.975)
	KRT81‐pos	1.402 (1.004–1.960)
UICC stage			0.002
	Stage I
	Stage IIA	0.503 (0.267–0.947)
	Stage IIB	1.018 (0.571–1.814)
	Stage III	1.409 (0.387–5.131)
	Stage IV
CA19.9			<0.001
	G1
	G2	2.152 (1.402–3.304)
	G3	3.292 (1.681–6.447)
Grade			0.008
	G1
	G2	1.506 (1.114–2.306)

Predictive value of IHC subtyping in cohort 2

Multivariate analysis in cohort 2 showed that patients with the new HNF1A‐positive subtype tended to benefit more from a cotreatment of gemcitabine with erlotinib than the double‐negative subtype or the KRT81‐positive subtype despite failing to reach the threshold of statistical significance (p = 0.064, supplementary material, Figure S3A–C). In the cotreatment group, the new HNF1A‐positive subtype achieved the best overall survival and the KRT81‐positive subtype the worst (p = 0.007, Figure 3B). However, in the gemcitabine‐only treatment group there was no significant difference in overall survival between the three IHC‐based subtypes (p = 0.134, Figure 3A).

Figure 3.

Multivariate analysis of survival effects of chemotherapeutic regimens administered in cohort 2. (A) Survival differences between three‐tier PDAC subtypes in the gemcitabine‐only treatment group. (B) Survival differences between three‐tier PDAC subtypes in the gemcitabine–erlotinib treatment group.

Discussion

In a wide range of solid tumors, a clinically relevant correlation between molecular subtyping and prognosis has been observed, and the understanding of molecular processes underlying different patient outcomes has led to more precise treatment and improved survival [16, 17, 18, 19, 20]. Studies in resectable PDAC patients proposed RNA expression‐based molecular subclassification systems based on the characterization of activated/suppressed pathways regulated by epigenic modifications and posttranscriptional mechanisms [8, 10, 11]. Although the proposed subtyping methods were somewhat different, certain kinds of distinct subtypes could be identified by most research groups, showed a certain degree of overlap and were confirmed to have a significant relevance for treatment responses and overall survival [21]. It was found that currently used postoperative chemotherapy may produce very different results for PDAC patients even within the same histopathological type of tumor [22].

However, there are inherent difficulties for subtyping methods based on gene expression analysis hindering its implementation in routine diagnostics. In our previous research [13], we attempted to discover potential IHC markers based on the established RNA classification systems, which could help in building a clinically implementable novel classification system. The IHC classifier was proposed to recapitulate the major functions of the molecular subtyping method used by Collisson et al [8]. while being easy to use, cheap, and widely available. We identified two surrogate IHC markers, KRT81 and HNF1A, which we used to define three subtypes. KRT81 was initially characterized in hair follicle formation [23] but was later found to have associations with several kinds of cancer such as non‐small cell lung cancer or breast cancer [24, 25, 26]. It has been reported that KRT81 plays a predominant role in cancer progression by activating genes related to tumor invasion and migration, such as matrix metallopeptidase 9 and lipocalin 2 [26]. KRT81 may also function as a regulator of inflammatory cytokine interleukin‐8 and be involved in malignant transformation in melanoma [27]. The transcription factor HNF1A was originally identified as an important factor in the regulation of glucose metabolism and plays a role in the development of diabetes [28, 29, 30], and also in B‐cell differentiation [31]. HNF1A has been reported to have conflicting functions in tumor progression and drug resistance among different cancer types. According to Fujino et al and Wang et al, HNF1A inhibition led to a significant reduction of proliferation and anticancer drug resistance of non‐small cell lung cancer and colorectal cancer via glucose metabolism [32, 33]. However, another recent study suggested that upregulation of HNF1A induces remarkable inhibition of tumor growth and platinum‐based chemotherapy resistance in PDAC through the activation of p53‐binding protein 1 [34].

In a cohort including 217 patients, the HNF1A‐positive subgroup (exocrine‐like) ranked best for overall survival, while the KRT81‐positive subgroup (basal‐like) had the worst survival, and double‐negative (classical) cases showed an intermediate survival. This was in line with survival data reported for the RNA‐based classifications [8, 10, 11].

To validate the feasibility of the IHC‐based subtyping method, a subsequent study investigating three independent PDAC cohorts was conducted by us [14]. In two primarily resected cohorts, we found a significant association between the identified subtypes and patient outcome. Again, the KRT81‐positive subtype had the worst prognosis, while the HNF1A‐positive subtype had the best, and the double‐negative subtype showed intermediate survival. In a cohort of advanced stage, nonresectable patients receiving primary chemotherapy, the KRT81‐positive subtype also showed dismal survival compared to the other two subtypes, while no significant difference in survival between the HNF1A‐positive and double‐negative subgroup could be found. Additionally, the KRT81‐positive patients did not seem to profit from intensified chemotherapy with FOLFIRINOX, whereas patients with HNF1A‐positive tumors seemed to have the biggest benefit. In the adjuvant treatment of PDAC patients, gemcitabine‐based chemotherapy or mFOLFIRINOX are standards of care resulting in 5‐year survival rates from 20% to almost 50% [35, 36, 37]. So far, the choice of these different regimens is only done with regard to the general condition. Especially in patients with good performance status, predictive biomarkers are urgently needed to identify the best treatment option with the lowest side effects. This is even more relevant in the ongoing discussion about neoadjuvant and perioperative treatment strategies [38, 39].

In our previous research, a few double‐positive cases (3/262, 18/130, 5/125) were defined as ‘unclassifiable’ as the survival curves of this small patient subgroup seemed to cross with the curves of the other three subtypes due to small sample size, providing limited prognostic value. In the study presented here, we expanded one of the previously investigated cohorts from 130 to 269 patients and introduced a new, independent primary resected cohort. Now, in our enlarged cohort, these double‐positive cases showed relatively good survival, similar to the HNF1A‐positive subtype. Therefore, a revised stratifying method was generated in which we aggregated the double‐positive and HNF1A‐positive cases into a combined new HNF1A‐positive group. Compared with the classifiers used in our previous research, the modified classification system remains essentially the same but allows us now to avoid exclusion of any individuals of the cohort, thus enabling the IHC‐based subtyping of all PDAC patients. In both cohorts, the double‐negative subtype remained the largest subgroup and double‐positive subtype the smallest. The number of HNF1A‐positive cases was more than that of KRT81‐positive cases in cohort 1 but the opposite was observed in cohort 2. The reason for the variance of subtype frequencies might be due to the fact that the two cohorts investigated in our research adopted different criteria to select patients. Compared to the retrospective cohort 1, cohort 2 was generated in the context of a phase‐3 clinical trial, which demand significantly stricter inclusion criteria which might have induced a selection bias. We could validate again that there was a marked survival difference between the IHC‐based subtypes. In short, the new HNF1A‐positive subtype was associated with the best overall survival at a median survival of 619 days in cohort 1 and 1,067 days in cohort 2, the KRT81‐positive the worst (413 days in cohort 1, 531 days in cohort 2) with the double‐negative subtype in between (600 days in cohort 1 and 821 days in cohort 2). The revised three‐tier classifier including the double‐positive cases in the HNF1A‐positive subgroup reliably retained the statistically significant difference in overall survival across the three subtypes.

From a pathogenetic point of view, the underlying mechanism of different survival between IHC‐based as well as molecular subgroups is not yet fully understood. In the previous studies, the molecular subtypes proposed by different research groups and used as references in our research are almost entirely based on transcriptomic data, which only reflects substantial epigenetic changes in PDAC [8, 10, 11]. As described in those studies, the differences in the genomic landscape between these subtypes appeared to be minor. Except the four predominate driver genes (KRAS, TP53, SMAD4, and CDKN2A), just a handful of genes mutated at 5–10% prevalence in PDAC cohorts were detected, which could not be well linked with the histological characteristics, transcriptomic data, or prognosis of the tumors [40]. This finding is also in line with the unpublished results of the present study. It was suggested by Collison et al based on the data of Singh et al that the classical subtype, which is recapitulated to a certain extent by our double‐negative subtype might be more dependent on the common KRAS mutations than the other subtypes [8, 41]. However, none of the authors suggested possible differences in carcinogenesis between the subtypes as it still remains unclear whether they represent differences in tumor generation and progression or different equilibriums of tumor–host interaction. Therefore, the interpretation of biological relevance and oncogenesis between subtypes at a genomic level needs further investigation and more convincing evidence.

Apart from its prognostic value, the potential predictive value of the IHC‐based subtype was also investigated. In cohort 2, patients with the new HNF1A‐positive subtype showed a propensity to profit from cotreatment with gemcitabine and erlotinib in contrast to those with double‐negative and KRT81‐positive subtypes (p = 0.064). Besides, in the multivariate analysis of cotreatment group, patients with the new HNF1A‐positive subtype achieved the best survival (p = 0.007), while in the gemcitabine‐only treatment group no significant difference in survival was found among the three subtypes. What merits proper attention is that the revised IHC‐based subtype is not a simple recapitulation of the expression‐based subtype identified by Collisson et al [8], but incorporates its own subtyping characteristics, especially when the double‐positive cases are taken into account. This subtype, which was expected to relate to worse survival in previous research [26, 42], appears instead to have a more favorable prognosis. This seems to indicate that biological shift caused by HNF1A expression dominates if KRT81 is coexpressed.

In summary, our study again confirms that, by evaluating the IHC expression level of KRT81 and HNF1A as partial but not full surrogates for RNA‐defined PDAC subtypes, PDAC patients could be classified into three distinct subgroups with prognostic implications, which is highly consistent with previous findings. Through an improvement of the classifier, we could include all cases therefore providing more reliable information without patient dropouts, potentially contributing to better clinical decision‐making in the future. The straightforward, easy to use evaluation criteria, and the wide availability of these immunohistochemical assays make the subtyping method practicable and easy to implement into pathological routine diagnostics. The classifier based on our data could not only be investigated in retrospective cohorts, but also included in future clinical trials with an instructive perspective. We believe that research into biologically relevant subtypes of PDAC patients may bring new opportunities for better patient management by individualizing patient treatments based on prognosis and expected treatment response. Overall, our results warrant further research to elucidate the role of KRT81 and HNF1A in PDAC and other tumor entities.

Author contributions statement

JR carried out experiments, performed statistical analysis and visualization and wrote the original draft of the paper. MS conceived and designed the study, reviewed and revised the paper and provided material support. UP, HR, HO, IED, HF and CJ provided material support, curated data, and reviewed and revised the paper. KS provided material support, project administration, and reviewed and revised the paper. AM conceived and designed the study, developed methodology, validated data, and reviewed and revised the paper. All authors were involved in writing the paper and had final approval of the submitted and published versions.

Supporting information

Figure S1. Example images of HNF1A and KRT81 staining of different intensities and percentages

Figure S2. Frequency distributions of four‐tier and three‐tier subtypes in cohort 1 and cohort 2

Figure S3. Multivariate analysis of survival effects of chemotherapeutic regimens in patient groups with different subtypes

Table S1. Overview table of the clinicopathologic characteristics of investigated patient cohorts

Table S2. Overview table of the clinicopathologic characteristics of IHC‐based subtypes

Table S3. Correlation analysis between IHC‐based subtypes and other clinicopathologic characteristics

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available due to privacy and ethical restrictions.