A total of 96 patients with PDAC were examined; 46 underwent NAC (NAC group), whereas 50 did not undergo NAC (non-NAC group). The present study included a non-NAC group because a comparison between the NAC and non-NAC groups was necessary to understand the histological changes caused by NAC. Although it was possible to evaluate the histological therapeutic effects using specimens treated with NAC alone, evaluating the changes in cancer stroma and microvessels after NAC would be difficult if the treated specimens were not compared with non-treated specimens. The patients underwent surgical treatment at Hirosaki University Hospital (Hirosaki, Japan) between November 2011 and April 2021. Written informed consent for the use of clinical records and pathological specimens was obtained from each patient before commencement of the study. All of the patients underwent dynamic CECT; none of them had contrast media allergy or renal function problems that would prevent them from undergoing CECT. The NAC group underwent dynamic CECT before and after NAC. Pathological tumor, node and metastasis (TNM) categories and staging were conducted according to an up-to-date TNM classification from the Union for International Cancer Control (eighth edition) (17). Regarding histological differentiation, tumors were classified as well-differentiated, moderately differentiated or poorly differentiated according to the World Health Organization classification of tumors of the digestive system (fifth edition) (18). A total of 17 patients in the NAC group were administered 50 mg/m² S-1 on days 1–14 and 1,000 mg/m² gemcitabine on days 8 and 15 for two 21-day cycles (GS). A total of 29 patients in the NAC group were administered 75 mg/m² nab-paclitaxel, followed by 1,000 mg/m² gemcitabine on days 1, 8 and 15 for two 28-day cycles (GnP). For patients who received GS, the dosing period was 2–10 cycles. For patients who received GnP, the dosing period was 2–15 cycles (Tables SI and SII). The dosing periods were clinically determined based on preoperative stage and patient condition. At our institution, the usual dosing period for patients with resectable pancreatic cancer is two cycles of GS or GnP. In cases of locally advanced borderline pancreatic cancer, the dosing period was extended until surgical factors that may complicate radical resection, such as portal vein invasion or superior mesenteric nerve plexus invasion, were resolved. The clinicopathological characteristics of all patients, including age, sex, tumor location and dosing period of NAC, are summarized in Table I.

For the present study, the area of residual tumor (ART) grading system was adopted as the method of determination of histological therapeutic effect of NAC. The ART grading system was adopted because Matsuda et al (13) reported that this system was the most prognostic method for the determination of the histological therapeutic effects of NAC for PDAC, whereas other grading systems, such as the College of American Pathologists, Evans and MD Anderson Cancer Center grading systems, did not demonstrate significant associations with patient outcomes. In the present study, the largest cross-section of tumor tissue stained with hematoxylin and eosin (H&E) was evaluated in the NAC group according to the ART grading system using a BX53 light microscope (Olympus Corporation) with a 10× eyepiece and a 4× objective lens (UPlanSApo 4×; Olympus Corporation). High- and low-responders were defined as patients in whom the ART was less than or equal to three 40× microscopic fields, and greater than three 40× microscopic fields, respectively. Treatments for non-viable cancer cells and histological changes following NAC, including fibrosis, macrophage aggregates, vascular degeneration and acellular mucous pools, were adjusted in accordance with the criteria of the ART grading system. Notably, histological evaluation was performed by expert pathologists specializing in pancreatobiliary pathology (TY and HK). Furthermore, all histological evaluations were performed while these pathologists were blinded to the clinical information.

The surgical specimens of the NAC and non-NAC groups were examined. In our hospital, we routinely sample whole PDAC tissues for pathological diagnosis. In the NAC group, 27 patients underwent pancreatoduodenectomy and 19 patients underwent pancreatosplenectomy. In the non-NAC group, 26 patients underwent pancreatoduodenectomy and 24 patients underwent pancreatosplenectomy. No patient underwent total pancreatectomy. Surgical specimens from pancreatoduodenectomy were sliced axially, whereas surgical specimens from pancreatosplenectomy were sliced at right angles to the main pancreatic duct. All surgical specimens were fixed with 10% formalin at 24°C for 24 h. Tissues were embedded in paraffin and sliced to a thickness of 4 µm for H&E staining and immunohistochemistry. H&E staining was performed using Mayer's hematoxylin for 20 min and with eosin for 5 min.

The histological characteristics of the NAC and non-NAC groups were investigated. The densities of cancer cells, CAFs, microvessels and stromal collagen fibers in the whole areas of the tumor were measured in a way similar to that described in our previous study (16). In the non-NAC group, the entire maximum cross-section of the tumor was examined. In the NAC group, ARTs, including viable cancer cells, were examined. Maximum cross-sections of tumor were divided into 40× microscopic fields using a BX53 light microscope with a 10× eyepiece and a 4× objective lens (UPlanSApo 4×) and a DP74 digital camera (Olympus Corporation). Whole images of the largest cross-sections of tumors with immunohistochemistry and Masson's trichrome staining were captured using CellSens software (version 2.3 64-bit; Olympus Corporation).

In the present study, ImageJ software [Java 1.6.0_24 (64-bit); National Institutes of Health] was used to analyze the histological images following immunohistochemistry and Masson's trichrome staining. The area of cytokeratin AE1/AE3, αSMA and CD31-positive components was measured in terms of the pixel number using thresholds with minimum and maximum values of 0 and 120, respectively. Finally, the average area ratios of positive components in the whole largest cross-sections of tumors were measured as the densities of cancer cells, CAFs, microvessels and stromal collagen fibers.

To measure the densities of cancer cells, CAFs, microvessels and stromal collagen fibers, immunohistochemistry (cytokeratin AE1/AE3, αSMA and CD31) and Masson's trichrome staining were performed on the slides of the maximum cross-section of the tumor. Cytokeratin AE1/AE3 is positive in cancer cells and nontumoral epithelial cells. αSMA is used as a general marker of CAFs in numerous types of human cancer (19–21). CD31 is positive in vascular endothelial cells of tumoral and nontumoral vessels (22). In the present study, Masson's trichrome staining was used to measure the density of stromal collagen fibers because it provides better visualization than immunohistochemistry (23).

Tissue slides were deparaffinized using the avidin-biotin-peroxidase complex method with Benchmark XT autoimmunostainer (Roche Tissue Diagnostics; Roche Diagnostics, Ltd.). Deparaffinized slides were treated with tris-EDTA buffer (pH 7.8) at 95°C for 44 min. The slides were then treated with 5% non-fat dry milk at 37°C for 15 min to block endogenous peroxides and proteins. Subsequently, the slides were incubated with primary antibodies for 60 min at 24°C. The clone numbers and dilution ratios of the primary antibodies were as follows: Cytokeratin AE1/AE3 (monoclonal mouse; clone AE1, AE3; 1:100; cat. no. 412811; Nichirei Bioscience Inc.), αSMA (monoclonal mouse; clone 1A4; cat. no. M0851; 1:100; Dako; Agilent Technologies, Inc.) and CD31 (monoclonal mouse; clone JC70A; cat. no. M0823; 1:40; Dako; Agilent Technologies, Inc.). All reaction products of primary antibodies were visualized using the iVIEW DAB detection kit with a biotin-conjugated goat anti-mouse immunoglobulin G secondary antibody (1:1,000; cat. no. 760-091; Roche Tissue Diagnostics; Roche Diagnostics, Ltd.). Slides were incubated with the secondary antibody at 37°C for 1 h. Subsequently, counterstaining with hematoxylin at 37°C for 8 min was performed and staining was observed under an optical microscope.

For Masson's trichrome staining, tissue slides were deparaffinized and rehydrated in 100, 95 and 70% alcohol. After washing in distilled water, the slides were stained in Weigert's iron hematoxylin working solution for 10 min. After further washing in distilled water, slides were stained in 1% Biebrich scarlet-acid fuchsin solution for 15 min. Slides were then differentiated in 5% phosphomolybdic-phosphotungstic acid solution after washing in distilled water. Subsequently, slides were directly stained in aniline blue solution for 10 min without rinsing. After washing in distilled water, slides were differentiated in 1% acetic acid solution for 5 min. After further washing in distilled water, slides were dehydrated quickly in 95% ethanol and cleared in xylene. Masson's trichrome staining was performed at 24°C and staining was observed under an optical microscope.

All of the patients in the NAC group underwent dynamic CECT before and after NAC using the fixed protocol at Hirosaki University Hospital. The protocol was the same as that described in our previous study (16). Dynamic CECT was conducted using a 64-detector row CT scanner (Discovery CT750 HD; GE Healthcare) with the following parameters: Detector configuration of 64×0.625, tube voltage of 120 kV, automatic tube current modulation, collimation of 40 mm, tube rotation time of 0.5 sec, pitch of 0.8, field of view of 35×35 cm, image matrix of 512×512, and slice thickness of 5 mm. After obtaining unenhanced images, a non-ionic contrast medium dose of 600 mgI/kg body weight with an iodine content of 300 mgI/ml (Iopamiron 300/370, Bayer Yakuhin Ltd.; Omnipaque 300, Daiichi-Sankyo Co. Ltd.; Iopromide 300/370, Fujifilm Toyama Chemical Co., Ltd.; Iomelon 350, Eisai Co. Ltd.; Optiray 320, Fuji Pharma Co., Ltd.) was injected intravenously within 30 sec, and scanning of the arterial, portal venous and equilibrium phases began 35–40, 60–70 and 180 sec after the start of contrast medium injection.

TDCs were drawn using Digital Imaging and Communications in Medicine data [EV insite R version 3.4.0.0 (PSP Corporation); or ShadeQuest/ViewR V1.24 (Yokogawa Medical Solutions)]. The method used to draw regions of interest (ROIs) in our previous study was used in the present study (16). The ROIs were drawn on the central areas of tumors, avoiding tumor margins, contrasted vessels and artificial materials, such as stents. Tumor margins were avoided in ROIs because histological analysis excluded tumor margins since they included a number of nontumoral components. After measuring the number of ROIs on CT for each time phase, the digital imaging software created TDCs before and after NAC. At our institution, all patients with PDAC underwent dynamic CECT via a fixed protocol to limit the influence of patient-related factors, such as body weight, and cardiac and renal function, which can affect TDCs. Additionally, concentrations of contrast agents and their administration rates were adjusted according to patient body weight, and cardiac and renal function. However, all patient-related factors could not be completely controlled. Therefore, in the present study, the slopes of the TDC were adopted as parameters of radiological images to minimize error factors because we considered that the changes in the rates of contrast enhancement were more appropriate than individual CT values. δ1 was defined as the first slope between the non-contrast and arterial phases before NAC, δ2 was defined as the second slope between the arterial and portal phases before NAC, and δ3 was defined as the third slope between the portal and equilibrium phases before NAC. δ1′ was defined as the first slope after NAC, δ2′ was defined as the second slope after NAC, and δ3′ was defined as the third slope after NAC. δδ1 was defined as δ1′ minus δ1, δδ2 was defined as δ2′ minus δ2, and δδ3 was defined as δ3′ minus δ3. CT number [in Hounsfield unit (HU)] was plotted on the vertical axis of the TDC, and time phase was plotted on the horizontal axis of the TDC.

First, comparisons of the densities of cancer cells, CAFs, microvessels and stromal collagen fibers between the non-NAC group, low-responders and high-responders were examined using the Kruskal-Wallis test, followed by the Steel-Dwass post hoc test. Second, the differences between δ1 and δ1′, δ2 and δ2′, and δ3 and δ3′ in low-responders and high-responders were analyzed using Wilcoxon test. Third, differences in δδ1, δδ2 and δδ3 between low-responders and high-responders were analyzed using Mann-Whitney U-test, which was also used for comparison of CAFs and microvessels, A receiver operating characteristic curve was drawn to calculate the cutoff value of radiological parameters between low- and high-responders. Furthermore, Kaplan-Meier curves and log-rank test were used to analyze recurrence-free survival according to the ART grading system and radiological parameters. Contingency tables were analyzed using Fisher's exact test.

All statistical analyses were performed using EZR version 1.54 (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is a modified version of R commander designed to add statistical functions frequently used in biostatistics (24).

The ART grading system classified 29 patients as low-responders and 17 patients as high-responders. The clinicopathological characteristics of the NAC group are summarized in Table II using Pearson's chi-square test.

The ART grading system classified the NAC group into high-responders and low-responders. High-responders had ARTs less than or equal to three 40× microscopic fields. Low-responders had ARTs greater than three 40× microscopic fields (Fig. 1).

At the time of analyzing histological images, cytokeratin AE1/AE3 was positive in cancer and nontumoral epithelial cells on tumor margins. Nontumoral epithelial cells were excluded following H&E staining. Even though αSMA was positive in CAFs and vascular smooth muscles, Mann-Whitney U-test showed that the number of CAFs was significantly greater than the number of vascular smooth muscles in the NAC and non-NAC groups (data not shown). Therefore, the effect of smooth muscles on the measurement were ignored because the number of CAFs was predominantly greater than the number of vascular smooth muscles.

Immunohistochemical analysis revealed a marked decrease in cancer cells (Fig. 2B and G) and CAFs (Fig. 2C and H) in the high-responders compared with those in the low-responders and non-NAC group (Fig. 2). Kruskal-Wallis test revealed that the density of cancer cells was significantly decreased in order from the non-NAC group to low-responders to high-responders (Fig. 3A). Similarly, the density of CAFs was significantly decreased in order from the non-NAC group to low-responders to high-responders (Fig. 3B). There were no significant differences in the density of microvessels between the non-NAC group, low-responders and high-responders (Fig. 3C). Furthermore, collagen fiber density was significantly increased in the NAC groups compared with that in the non-NAC group; however, there was no significant difference in collagen fiber density between low-responders and high-responders (Fig. 3D).

The curve shape of the TDC was markedly altered before and after NAC (Fig. 4). In low-responders, δ1′ was significantly higher than δ1 (Fig. 5A), whereas there were no significant differences between δ2 and δ2′ (Fig. 5B). δ3′ was significantly lower than δ3 (Fig. 5C). In high-responders, there was no significant difference between δ1 and δ1′ (Fig. 5D), δ2′ was significantly lower than δ2 (Fig. 5E), and there was no significant difference between δ3 and δ3′ (Fig. 5F). No significant difference was noted in δδ1 between low-responders and high-responders (Fig. 5G). Furthermore, δδ2 was significantly lower and δδ3 was significantly higher in high-responders than in low-responders (Fig. 5H and I). Receiver operating characteristic curve showed that the cutoff values between low-responders and high-responders were δδ2=−3.448 HU/sec [sensitivity=82.4%, specificity=69.0%, area under the curve (AUC)=0.775] and δδ3=4.983 HU/sec (sensitivity=58.8%, specificity=89.7%, AUC=0.718) (Fig. 6). Regarding the ART grading system, there was no prognostic significant difference between high- and low-responders (P=0.166). In addition, there was no prognostic significant difference between high and low δδ2; the cutoff value of δδ2 was −3.448 HU/sec. Furthermore, there was no prognostic significant difference between high and low δδ3; the cutoff value of δδ3 was 4.983 HU/sec (Fig. 7).