FormalPara Key Points

This is the first study to show that expression of αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x, and sdi-Lea on primary pancreatic ductal adenocarcinoma (PDAC) tissues may be predicted using fine-needle biopsies (FNBs).

The predictive value of FNB expression for primary tumor expression was not negatively influenced by the application of neoadjuvant therapy.

The study facilitates the clinical translation of a personalized molecular imaging approach in PDAC, in which patients can be preoperatively screened for positive biomarker expression, potentially improving both tumor delineation and patient outcomes.

1 Introduction

Pancreatic ductal adenocarcinoma (PDAC) is the most common type of pancreatic cancer (± 90%), accounting for approximately 450,000 cases each year worldwide [1]. With a 5-year overall survival rate of merely 2–9%, the malignancy is highly lethal [1, 2]. To date, surgery offers the only curative treatment for PDAC [1,2,3]. Unfortunately, due to the asymptomatic course of the disease, PDAC is often diagnosed at an advanced stage and therefore only 15–25% of patients qualify for surgery [4]. In addition, positive resection margins are observed in up to 75% of patients who underwent surgery for PDAC, leading to early recurrence and shorter survival [5, 6]. Neoadjuvant therapy (NAT) has been shown to reduce the number of irradical resections and is therefore increasingly applied [7, 8].

Accurate guidance through computed tomography (CT), endoscopic ultrasound (EUS), magnetic resonance imaging (MRI) and/or positron emission tomography (PET) is essential to carefully select and stratify patients for surgery. However, due to the highly infiltrative and discontinuous growth pattern of PDAC, these modalities are not sufficiently able to make a clear distinction between vital tumor cells and benign lesions, such as tumor-associated pancreatitis [9, 10]. Moreover, the application of NAT may induce tissue fibrosis and necrosis, thereby further complicating pre- and intraoperative tumor detection. By providing enhanced contrast between malignant and healthy as well as benign tissues, molecular-targeted imaging agents could provide more reliable visualization of PDAC [11].

Accurate selection of molecular imaging biomarkers remains a key prerequisite for successful tumor delineation. A biomarker is considered suitable for molecular imaging when, among other criteria, it is expressed in a large group of patients within the same tumor type, while expression on surrounding healthy and benign tissues is absent, or at least significantly lower [12]. Previous research by, among others, our group has identified the proteins αvβ6, CEACAM5, epidermal growth factor receptor (EGFR), mesothelin, and tumor-associated glycans Lewisa/c/x (Lea/c/x) and sialyl-di-Lewisa (sdi-Lea) as promising tumor-specific biomarkers for molecular imaging in PDAC [13,14,15,16,17].

Due to the presence of intertumoral heterogeneity and application of NAT, none of these tumor markers is equally expressed in all patients, thereby preventing universal administration of molecular imaging tracers. To address these issues, patients could be preoperatively screened for expression of multiple tumor markers, for instance by using fine-needle biopsies (FNBs). During diagnostic work-up, FNBs are routinely obtained via EUS and may provide histological confirmation of PDAC [18,19,20]. Moreover, FNB material is increasingly used for molecular and genomic characterization of PDAC, as well as to generate patient-derived xenografts or organoids, thereby providing opportunities for patient-centered therapeutic selection [21, 22]. Similarly, preoperative evaluation of biomarker expression on FNBs could form a robust and efficient strategy to select the most suitable molecular imaging tracer in a patient-specific manner. However, it remains to be elucidated whether biomarker expression in FNBs corresponds to the expression on the primary tumor (PT). Moreover, very few studies have examined expression of tumor markers in PDAC after NAT [14, 15, 17].

Therefore, this study aimed to evaluate whether preoperative FNB specimens can be used to predict biomarker expression in the corresponding PT specimen. To accomplish this, expression of biomarkers αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea on FNB and PT specimens derived from the same patient was evaluated and compared. Moreover, since FNB specimens are acquired before NAT, the effect of pretreatment on this association between FNB and PT biomarker expression will additionally be studied.

2 Materials and Methods

2.1 Patient and Specimen Selection

Pathology reports of patients who underwent a pancreatic EUS-FNB procedure and resection at the Leiden University Medical Center (LUMC) between 2015 and 2020 were retrospectively reviewed. Representative formalin-fixed paraffin-embedded (FFPE) tissue blocks and hematoxylin and eosin (H&E)-stained slides of the PT and the FNB of each patient were selected, and obtained from the biobank at the LUMC. FNB samples were processed using the cell-block technique [23]. Selection and suitability of patient material included was determined by a hepatopancreaticobiliary pathologist based on H&E slides (ASLPC). One tissue block per FNB or PT tissue was used for each patient. Clinicopathological data were obtained from patients’ medical records. Pathological T (pT) and pathological N (pN) stages were defined according to the 8th edition of the American Joint Committee on Cancer/Union for International Cancer Control (AJCC/UICC) TNM staging system for pancreatic cancer. Partial and near-complete pathological response were defined as the presence of > 5% and ≤ 5% residual tumor tissue, respectively. The study protocol was approved by the Gastroenterology Biobank Review Committee (protocol reference: 2020-16) and the local medical Ethical Review Committee (protocol reference: B20.052). This study was conducted in agreement with the Dutch code of conduct for responsible use of human tissue in medical research. Tissue specimen and clinicopathological data were used anonymized and in compliance with the Declaration of Helsinki (1964).

2.2 Immunohistochemistry

Four-μm-thick tissue sections were cut using a microtome from FFPE tissues and placed on glass slides. Tissue sections were deparaffinized in xylene and rehydrated in degrading concentrations of ethanol (100%, 50% and 25%), until reaching demineralized water. Subsequently, endogenous peroxidases were blocked using a 0.3% hydrogen peroxide solution. Antigen retrieval was performed depending on the primary antibody as described in Table 1 of the electronic supplementary material (ESM). After antigen retrieval, slides were rinsed in phosphate-buffered saline (PBS; pH 7.4). Primary antibodies against αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea were applied on the tissue and incubated overnight at room temperature in a humid incubator. For information about primary antibodies, see ESM Table 1. After overnight incubation, primary antibodies were removed by rinsing the slides in PBS, and anti-mouse horseradish peroxidase (HRP) or anti-rabbit HRP secondary antibodies (Envision, Dako, Glostrup, Denmark) were applied for 30 min at room temperature in a humid incubator. Subsequently, secondary antibodies were removed by rinsing the slides in PBS. Antibody binding was visualized using a 3,3-diaminobenzidine tetrahydrochloride solution (DAB, K3468, Agilent Technologies, Inc., Santa Clara, CA, USA) for 10 min at room temperature. Lastly, the slides were counterstained with Mayer’s hematoxylin (Klinipath B.V., Olen, Belgium), dehydrated in a dry incubator for 2 h and mounted using pertex (Leica Microsystems, Wetzlar, Germany).

Table 1 Patient characteristics of the total cohort and subgroups of patients who did and did not receive NAT
Full size table

2.3 Semi-Automatic/Digital Image Analysis

Whole slide images of the stained tissue slides were captured using the PANNORAMIC® 250 Flash III DX scanner (3DHISTECH Ltd, Budapest, Hungary). For quantification of DAB staining, QuPath version 0.2.3 open-access software was used [24]. A detailed description of the QuPath workflow used in this study is provided in the ESM. A graphical representation of the used workflow is depicted in ESM Fig. 1. Briefly, random forest object classifiers for PDAC FNB and PT tissues were trained and built for each investigated biomarker separately, yielding a total of 12 object classifiers [25]. Image type settings, color deconvolution stain settings and positive cell detection (DAB staining) parameters were optimized. Object classifiers were trained to specifically recognize and discriminate between tumor cells, stroma cells and red blood cells (RBC), until providing a sensitivity, specificity, negative-predictive value, and positive-predictive value of ≥ 85%. Subsequently, the described workflow was scripted, allowing for semi-automated batch processing of the whole cohort (n = 45). Tumor regions were annotated on FNB and PT tissues by a pathologist (ASLPCC), whereafter the corresponding script was run. The H-score scoring system was used to quantify per-cell biomarker expression [H-score formula: 1 × (% cells 1+) + 2 × (% cells 2+) + 3 × (% cells 3+)], in which 1+, 2+ and 3+ were defined as low, medium, and high immunohistochemical staining intensity, respectively. Overall tissue staining with an H-score of ≤ 50 was considered negative, 51–100 as low, 101–200 as moderate, and 201–300 as strong expression of the biomarker. Concordance between fine needle aspiration biopsy (FNAB) and PT biomarker expression was defined as an FNAB versus PT H-score difference of ≤ 50 (∆H-score ≤ 50).

2.4 Statistical Analysis

IBM SPSS statistics version 25 (IBM Corporation, Armonk, NY, USA) was used for all statistical analyses, and graphs were created using GraphPad Prism version 8 (GraphPad Software, La Jolla, CA, USA). For categorial data, groups were compared using a Chi-square test. An independent samples t-test was used to compare continuous variables of patient characteristics. Correlation analyses were performed using Spearman correlation. Differences or correlations with a p-value < 0.05 were considered significant.

3 Results

3.1 Patient Characteristics

Forty-five patients diagnosed with PDAC were included in this study, of whom 23 (51%) received NAT and 22 (49%) did not. Ten (43%) NAT patients received gemcitabine/radiotherapy, while 13 (57%) patients received FOLFIRINOX. Characteristics of patients included in the full cohort and within two subgroups (patients who received NAT and patients who did not receive NAT) are shown in Table 1. No significant differences were observed between the two subgroups for all characteristics.

3.2 Object Classifier Training and Validation

Twelve object classifiers were trained and validated. Each of the object classifiers allowed detection of its cell type of interest with sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of ≥ 85%, as shown in ESM Table 2. Examples of QuPath detection images after batch processing analysis in FNB and PT tissues are shown in ESM Fig. 2.

3.3 Biomarker Expression on Fine Needle Biopsy and Primary Tumor Tissues of Pancreatic Ductal Adenocarcinoma

In total, 45 FNB and 45 PT tissues were included in the study and immunohistochemically stained for αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea expression, followed by staining quantification using the described semi-automated imaging analysis workflow. Sequential images of staining of FNB and PT tissues derived from one representative patient are shown in Fig. 1.

Fig. 1
figure 1

Images of HE and immunohistochemical staining of αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea on FNB and PT tissues derived from one representative patient. Overview images and inserts are taken at 5× and 40× magnification, respectively. Scale bars represent 200 µM. FNB fine-needle biopsy, HE hematoxylin-eosin, PT primary tumor

Full size image

All biomarkers showed a membranous staining pattern on tumor cells in both FNB and PT tissues, with CEACAM5 showing generally more staining on the luminal side of cells. αvβ6 was homogeneously expressed throughout the positive tumors, whereas Lea/c/x and sdi-Lea showed a slightly more heterogenous staining pattern. Membranous staining of CEACAM5, EGFR and mesothelin on PDAC cells was considerably more heterogeneous compared with the other biomarkers (Fig. 1).

In the FNB specimen, positive biomarker expression (H-score > 51) was observed in 41/42 (98%) for αvβ6, 29/41 (71%) for CEACAM5, 39/39 (100%) for EGFR, 16/42 (38%) for mesothelin, 37/41 (90%) for Lea/c/x and 39/43 (91%) for sdi-Lea. In the PT specimen, positive biomarker expression was observed in 41/42 (98%) for αvβ6, 32/41 (78%) for CEACAM5, 35/39 (90%) for EGFR, 20/42 (48%) for mesothelin, 35/41 (85%) for Lea/c/x and 41/43 (95%) for sdi-Lea. Categorized biomarker staining levels for FNB and PT tisssues separately are shown in Table 2. Moderate or strong αvβ6, Lea/c/x and sdi-Lea staining was observed in most FNB and PT tissues, whereas EGFR was moderately stained in the majority of cases. CEACAM5 and mesothelin staining was generally weaker compared with the other biomarkers.

Table 2 Categorized staining levels of αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea investigated targets in PDAC FNB and PT specimens, separated into cases showing negative (H-score 0–50), low (H-score 51–100), moderate (H-score 101–200) and strong expression (H-score 201–300)
Full size table

3.4 Concordance Between Biomarker Expression on PDAC FNB and PT Tissues

To assess the concordance between FNB and PT biomarker expression, differences between H-scores on FNB and PT tissuse were calculated. Concordance was considered to be present when the difference between FNB and PT H-score was ≤ 50. Concordance rates of all biomarkers are shown in Table 3. For the full cohort, concordance rates were 62% for αvβ6, 61% for CEACAM5, 85% for EGFR, 69% for mesothelin, 76% for Lea/c/x and 79% for sdi-Lea, indicating high concordance between biomarker expression on FNB and PT tissues. Next, to study the effect of NAT on concordance between FNB and PT biomarker expression, patients who did and did not receive NAT were divided into two subgroups. In subgroup analyses, medium to high concordance rates between dichotomized IHC staining were observed for patients who received NAT (range 64–83%) and patients who did not (range 60–90%), suggesting that NAT does not substantially affect concordance between positive or negative FNB and PT biomarker expression (Table 3).

Table 3 Concordance rates biomarker expression on FNB and PT tissues for the full cohort as well as the NAT and no NAT subgroups, along with the number of FNB and PT pairs
Full size table

3.5 Correlation between Biomarker Expression on PDAC FNB and PT Tissues

To study the predictive value of biomarker expression on FNB tissues for PT expression, the relationship of biomarker expression levels between FNB and PT specimens was investigated. For the full cohort, H-score scatterplots showing biomarker expression on FNB and PT tissues are shown in Fig. 2a–f. Moreover, Spearman’s rank correlation coefficients indicating the linear association between the FNB and PT biomarker expression are shown in Table 4 as well as in Fig. 2a–f. H-scores on FNB tissues were positively correlated with H-scores on PT tissues for CEACAM5, EGFR, mesothelin, sdi-Lea and Lea/c/x, suggesting that biomarker expression on FNB tissues is predictive for expression level in PT tissues. Correlation coefficients indicated a moderate–strong linear relationship between FNB and PT expression. However, for αvβ6, no correlation between H-score on FNB and PT tissues was found (rs = 0.041, p = 0.795).

Fig. 2
figure 2

H-score scatterplots showing the relationship between immunohistochemical staining of a αvβ6, b CEACAM5, c EGFR, d mesothelin, e Lea/c/x and f sdi-Lea on FNB (x-axis) and PT (y-axis) tissues. Blue dots represent an H-score difference between FNB and PT staining of ≤ 50, whereas the red dots represent a difference of > 51, thereby indicating an outlier. Spearman’s rank correlation coefficients (rs) with their corresponding p-values are shown in each graph. FNB fine-needle biopsy, PT primary tumor

Full size image
Table 4 Correlation scores between biomarker expression on FNB and PT PDAC tissues
Full size table

Next, subgroup analyses were performed to study the effect of NAT on the predictive value of biomarker expression on FNB tissues for PT biomarker expression. As shown in Fig. 3a–l and Table 5, H-scores of CEACAM5, EGFR, sdi-Lea and Lea/c/x expression on FNB were positively correlated with H-scores on PT tissues in both NAT and no NAT patients, suggesting that biomarker expression is predictive for PT expression regardless of application of NAT. However, for αvβ6 and mesothelin, biomarker expression on FNB tissues was not predictive for PT expression regardless of application of NAT or in the no NAT subgroup, respectively.

Fig. 3
figure 3

H-score scatterplots showing the relationship between immunohistochemical staining of a, b αvβ6, c, d CEACAM5, e, f EGFR, g, h mesothelin, i, j Lea/c/x and k, l sdi-Lea on FNB (x-axis) and PT (y-axis) tissues for NAT and no NAT patients separately. Blue dots represent an H-score difference between FNB and PT staining of ≤ 50, whereas the red dots represent a difference of > 51, thereby indicating an outlier. Spearman’s rank correlation coefficients (rs) with their corresponding p-values are shown in each graph. FNB fine-needle biopsy, NAT neoadjuvant therapy, PT primary tumor

Full size image
Table 5 Correlation scores between biomarker expression on FNB and PT PDAC tissues, subdivided into patients who received NAT and patients who did not
Full size table

4 Discussion

Tumor-targeted molecular imaging could play a crucial role in solving current challenges during diagnosis, monitoring and resection of PDAC by providing high-contrast delineation of malignant tissues [11, 26]. Patient-specific screening and selection of molecular imaging biomarkers using FNB tissues may improve PDAC delineation and prevent unnecessary tracer administration.

Our study shows that αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea are highly expressed in FNB and PT tissues of PDAC. The results suggest that there is a high level of association in expression between FNB and PT tissues of PDAC for the biomarkers evaluated herein. Although no correlation was observed for αvβ6, concordance rate was the highest among the investigated biomarkers. This suggests that αvβ6 expression is similar between FNB and PT, and the biomarker was expressed in an homogenous manner, i.e. strongly in most tissues. Moreover, when discriminating between patients who received NAT and patients who received no NAT, no substantial differences between concordance rates and correlations, except for mesothelin, were observed, suggesting that NAT does not affect the expression level of the biomarkers investiged and therefore also not the predictive value of biomarker expression on FNB versus PT tissues. Our study therefore demonstrated the feasibility of a molecular imaging biomarker screening workflow for PDAC in a clinically relevant setting.

The current findings are largely in line with previous studies that showed high expression of αvβ6, CEACAM5, EGFR, mesothelin, Lea/c/x and sdi-Lea in PDAC tissues [13,14,15, 17]. Interestingly, our study found positive mesothelin expression in 48% of PT tissues, while most studies report positive mesothelin expression in 80–85% of PDAC tissues [15, 27, 28]. This discrepancy could have been caused by, among others, the use of different antibody clones directed against mesothelin, a lack of staining reproducibility between laboratories, or the use of manual scoring methods in other studies, which may not be sufficiently reproducible [29,30,31]. Biopsy specimens have the potential to aid in proper treatment selection, as was previously shown for rectum, lung, breast and gastric cancer [32,33,34,35,36]. In each of these studies, high levels of concordance between biomarker expression on biopsy and PT tissues was observed. Although PDAC FNB specimens are already used in research into PDAC diagnosis [37,38,39], the association between FNB and PT biomarker expression has, to the best of our knowledge, not yet been investigated for PDAC.

The lack of reproducibility of manual scoring was compensated in our study through the use of QuPath, which could provide superior cell detection efficacy, leading to more accurate and reproducible H-scores as opposed to manual scoring [24, 31]. However, the QuPath workflow we used herein was more time-consuming compared with manual scoring. We therefore believe that for this study, the advantages outweighed the disadvantages as the workflow allowed us to obtain highly accurate and reproducible results.

Although moderate to strong correlation coefficients between FNB and PT tissues were observed for expression levels of most biomarkers, along with high concordance, several H-score outliers were present. A possible explanation may be that, considering the average time of 100 days between the FNB and surgical procedure (data not shown), biomarker expression changed over time. PDAC is a highly progressive disease and distinct (epi)genetic events or remodeling could lead to a certain extent of polyclonality [17, 18]. Different subclones of tumor cells could (over)express distinct proteins and, over time, this could result in intra-tumoral heterogeneity and, subsequently, varying expression patterns. Second, a biopsy is merely a small fragment of the entire PT specimen. Considering intra-tumoral heterogeneity, it could very well be possible that a biopsy sample contains an area of the PT that does not express the biomarker of interest, while other parts of the tumor do express the biomarker [17]. Of note, and in line with previous research, five of six biomarkers evaluated herein showed considerable heterogeneity, further substantiating this hypothesis [13,14,15,16,17, 27]. Lastly, FNB cell blocks and FFPE tisues are fixed and processed differently, which can influence the level of IHC staining [40]. However, our study shows that in most cases, biomarker expression on FNB tissues shows considerable predictivity for expression on PT tissues, indicating that biopsy specimens provide a representative sample of PT tissues.

Due to their low abundance in surrounding tissues, and also high expression in lymph node metastases, the biomarkers evaluated herein are promising targets for both diagnostic and intraoperative imaging of PDAC. Regarding the former, several studies recently reported the phase I/II evaluation of targeted PET tracers directed against mesothelin and αvβ6 and showed their safety and feasibility for PDAC visualization [41,42,43]. With regard to the latter, (early-phase) clinical trials have particularly focused on NIRF tracers targeting EGFR and CEA [44,45,46,47,48]. NIRF imaging using the anti-EGFR tracers panitumumab-IRDye800CW and cetuximab-IRDye800CW, and anti-CEA tracer SGM-101, allowed clear delineation of primary PDAC lesions, as well as lymph node and distant metastases intraoperatively [44,45,46,47,48]. Within the preclinical arena, the anti-αvβ6 tracers R01-MG-IRDye800CW and anti-Lea/c/x CH88.2-IRDye800CW have enabled clear localization and delineation of human PDAC xenografts in mice, thereby strengthening their clinical potential for real-time intraoperative NIRF imaging of PDAC [16, 49]. Considering the (pre)clinical availability of molecular tracers directed against these biomarkers, the workflow we evaluate and discuss in this study may be instantaneously translated towards a clinical setting, where patients with positive biomarker expression in FNB can be selected for patient-specific tracer administration.

This study has some limitations. First, the present study cohort was relatively small for exact estimation of biomarker expression in PT tissues. Therefore, evaluation of the relationship between FNB and PT expression on a larger cohort could strengthen the findings of this study. Second, due to the limited number of patients available, the NAT subgroup was small, which allowed no separate evaluation of the effect of gemcitabine/radiotherapy or FOLFIRINOX. Third, due to staining artefacts, the number of pairs available for analysis differed and did not overlap precisely for all biomarkers.

Future research could focus on the expansion of the workflow to other tumor types for which biopsies are obtained, such as breast, (colo)rectal and gastric cancer, and additionally to other tumor biomarkers of interest. The workflow evaluated herein can conveniently be integrated within standard pathological work-up, as IHC stainings on FNB tissues for markers such as Ki-67, p53 and SMAD4 are regularly performed to guide histological diagnosis [50,51,52]. Taken further, this method of biomarker screening can also be applied in the therapeutic arena, where PDAC patients eligible for targeted therapy can be screened for biomarker expression (53). Overall, this study provides the foundation for clinical translation of a patient-specific, FNB-based molecular imaging biomarker selection workflow that may improve PDAC diagnosis, delineation, and, ultimately, patient outcomes.

5 Conclusion

Our findings show that biomarker expression on FNB is considerably predictive for biomarker expression on corresponding PT tissue, irrespective of the application of NAT. This study thereby provides the foundation for the clinical application of an FNB-based biomarker-screening workflow, eventually facilitating a personalized approach of molecular imaging tracer administration in PDAC.