3D Imaging Resolves Human Pancreatic Duct–β-Cell Clusters During Cystic Change Free

The lobules of the human pancreas consist of both exocrine parenchyma (acini and ducts) and endocrine islets. During development, the ductal epithelium plays a central role by differentiating into acinar and islet cells, establishing exocrine and endocrine functions (1). In healthy adults, pancreatic endocrine function was thought to be maintained by a stable islet cell population, with both β- and α-cell masses preserved in individuals without apparent pancreatic diseases (2–4). However, recent data from cross-sectional imaging show that local and benign cystic changes in the human pancreas are common, particularly in the senior population (5–9). For instance, using magnetic resonance cholangiopancreatography for cyst detection (≥2 mm), Kromrey et al. (9) reported a weighted prevalence of 49.1% in 1,077 participants (mean age 55.8). Of these, 676 individuals were followed-up for 5 years, and no pancreatic cancer was observed. Additionally, a classic study of 300 Japanese autopsy cases (mean age 79.3) using hematoxylin and eosin (H&E) histology revealed that 24.3% of the analyzed pancreases harbored cysts (≥2 mm) (10). In the human pancreas, due to the direct exocrine-endocrine contacts in the lobule (note: rodent islets are primarily perilobular and sheathed by glia [11,12]), the impact of cystic change on the islet microenvironment is anticipated, but the details are unknown. From the perspective of endocrine pancreas, investigating the lobular microenvironment in the context of cystic change offers a unique opportunity to gain insight into associated exocrine-endocrine tissue remodeling, which otherwise cannot be fully evaluated under experimental conditions.

Previously, lineage tracing analysis of the mouse pancreas has shown that injuries induced by partial duct ligation and subsequent tissue repair can lead to islet β-cell regeneration, including β-cell neogenesis (or duct–to–β-cell differentiation) (13–16). However, conflicting results showing no β-cell neogenesis have also been reported (17–20), creating controversy in islet regeneration after injury (21,22). Because lineage tracing involves extensive cellular labeling and imaging processes, artifacts are prone to occur, necessitating careful positive and negative signal controls to avoid erroneous results and conclusions. This concern is also relevant for human pancreatic duct and β-cell analysis, in which high/superresolution imaging is essential for accurately specifying cell identity and cell-cell associations without ambiguity.

Despite a clear need to translate animal research to human pancreas analysis, several challenges currently limit the progress of human duct and islet imaging in health and disease. First, it is inherently difficult to visualize the three-dimensional (3D) ductal network and scattered islets using routine two-dimensional (2D) pathological sections. Second, pancreatic steatosis is prevalent in adult humans (23). The accumulated fats (peri- and/or intralobular fatty infiltration [24]) strongly scatter light, leading to false-negative results in in-depth imaging. Third, the resected human pancreas, such as the surgical biopsy, often contains residual blood, which can cause autofluorescence and result in false-positive signals in fluorescent imaging. Thus, to investigate β-cells in the context of cystic change, careful examination of the lobular components—including both the epithelium (duct cells) and islets (β-cells)—using multimodal and multidimensional signals is essential to compare and identify the associated tissue remodeling.

Given the aforementioned challenges, a fundamental improvement in human pancreatic tissue and cellular imaging is needed to advance our understanding of the lobular and islet remodeling. Here, using a new antifade 3D fluorescence imaging approach (25,26), we embed the cytokeratin 7 (CK7) (epithelial marker) and insulin-labeled human pancreas in an acrylamide and N-hydroxymethyl acrylamide (or A-ha)-based high-refractive-index (high-n) polymer for tissue clearing and 3D duct and β-cell imaging (25). By examining consecutive pancreatic sections, we demonstrate that this approach can integrate with clinical 2D histology (hematoxylin and eosin [H&E] stain, the gold standard) and Airyscan superresolution imaging (27) to characterize the lobular microenvironment with subcellular (vesicle)-level resolution. Our development of the multimodal and multidimensional human pancreatic duct and β-cell microscopy in the context of cystic change is discussed in this report.

Collection and use of human tissues were approved by the National Taiwan University Hospital Institutional Review Board (NTUH-REC no. 202309106RINA). Human pancreases were obtained from deceased organ donors whose pancreases were declined for transplantation but whose next of kin had given written consent for use in research.

Body and tail regions of grossly normal and perfused human donor pancreases (n = 10) were obtained from deceased organ donors without apparent pancreatic diseases or diabetes. The regions were first cut into strips ∼1.5 cm wide, fixed in 10% formaldehyde for 2 days, and then washed in PBS for 4 days at 4°C. The fixed specimens were sectioned to a thickness of 350 µm using a vibratome and stored in 0.25% formaldehyde at 4°C for preservation. Supplementary Table 1 summarizes the clinical information (age, sex, BMI, HbA_1c level, and cause of death) of the organ donors, along with the corresponding figures and videos.

Rabbit anti-CK7 (ab68459; Abcam, Cambridge, MA), Alexa Fluor (AF)-647–conjugated mouse anti-insulin (sc-8033_AF647; Santa Cruz Biotechnology, Dallas, TX), and mouse anti-glucagon (ab10988; Abcam) antibodies were used in immunohistochemistry. Before the antibodies were applied, tissue sections were rinsed in PBS. This was followed by a blocking step, incubating the tissue with a blocking buffer (2% Triton X-100, 10% normal goat serum, and 0.02% sodium azide in PBS). The primary antibody was diluted in a buffer containing 0.25% Triton X-100, 1% normal goat serum, and 0.02% sodium azide in PBS (1:100) and then incubated with the tissue for 2 days at 15°C. Negative staining controls were prepared by omitting the primary antibody from the buffer. AF-488– or AF-546–conjugated secondary antibodies (raised in goat, 1:200; Thermo Fisher Scientific) were used in combination to reveal the immunostained structures. The labeled specimens were immersed in 10% formaldehyde at 4°C overnight for postfixation before tissue clearing and imaging. Supplementary Table 2 summarizes the immunostaining reagents/dilutions used in illustrations.

The vibratome section of human pancreas was first immersed in one well of a six-well plate with 3 mL of A-ha monomer solution (note: A-ha solution was prepared by adding 5 g of water containing 0.015 g of photoinitiator Irgacure 2959 to a mixture of 13.48 g of acrylamide and 19.1 g of N-hydroxymethyl acrylamide; molar ratio 1:1) (25). The plate was placed on a platform rotator for 15 min, and then the tissue was transferred to a new well with 3 mL of A-ha monomer solution for another 45 minutes. Afterward, a 350-μm iSpacer (SunJin Lab, Hsinchu, Taiwan) was placed on a coverslip, fresh A-ha monomer solution (360 μL) was added to the spacer compartment, and the tissue was then transferred from the six-well plate to the spacer. After another coverslip was added on top, the sample was ready for photo-polymerization (Supplementary Fig. 1). Ultraviolet irradiation for 30 minutes was used to prepare transparent specimens for 3D confocal microscopy (Zeiss LSM 800, Carl Zeiss, Jena, Germany) and superresolution imaging (Zeiss Airyscan, a 32-detector array).

Prior to Airyscan, standard confocal imaging with 10× (C-Apochromat 10×/0.45), 25× (LD_LCI Plan-Apochromat 25×/0.8), and/or 40× (LD_C-Apochromat 40×/1.1) objectives were performed to evaluate the lobular environment and to identify the area of interest (lobule with cystic change). In Airyscan, superresolution imaging with 40× (Plan-Apochromat 40×/1.3 oil) or 63× (Plan-Apochromat 63×/1.4 oil) objective was applied. The multiple rounds of fluorescence imaging were made possible by the antifade properties of A-ha–based fluorescence imaging (25,26). Fluorescence signals in the Figures are pseudo-colored. Supplementary Table 3 summarizes the color codes for different markers in illustrations.

The human pancreas specimen was immersed in PBS and then imaged via stereomicroscopy (Carl Zeiss, SteREO Discovery.V12) to record the pixels occupied by the tissue (defined as 100) (Supplementary Fig. 2). Afterward, the tissue was transferred to the clearing condition and imaged again. The number of pixels recorded in the clearing condition (A-ha copolymer, methyl salicylate [organic solvent; 76631, Sigma-Aldrich], glycerol [88% in water; G5516, Sigma-Aldrich], Ce3D [aqueous solution; 427702, BioLegend, San Diego, CA] [28], or CytoVista [organic solvent; V11300, Thermo Fisher Scientific [29]) was divided by the pixel count in PBS to calculate the relative tissue size after optical clearing.

Vibratome sections of the human pancreas (350 µm in thickness) were examined with stereomicroscopy to detect likely areas with microcystic change or pancreatic intraepithelial neoplasia (PanIN). Afterward, the area of interest was examined by clinical 2D histology with H&E stain (gold standard; 3 µm in thickness) to confirm the tissue remodeling. Once confirmed, immunolabeling of adjacent vibratome sections was performed for 3D histology to reveal the epithelium and islets in the same microenvironment.

Avizo 6.2 (VSG, Burlington, MA), Zen (blue/black editions, Carl Zeiss), and LSM 510 (Carl Zeiss) software were used for noise reduction, projection, and image analysis. Image segmentation for quantification of the CK7⁺ and insulin-positive signal density was illustrated in Juang et al. (30). Briefly, feature extraction and image segmentation were performed by the Zen software to collect pixels of the area of interest (intralobular space, DAPI⁺ signals) and those with CK7⁺ or insulin-positive signals. Pixels with the signal were then divided by those of the area of interest × 100% to estimate the signal density. In Fig. 1G and N, the red, green, or blue signal intensity of a stereomicroscopic image was analyzed by the Zen software (0–255 intensity range). The Feret diameter was used to estimate the size of cystic change and low-grade PanIN in the pancreatic tissue map. The intensity graphs presented in Figs. 6F and 7C were acquired from the Profile function of Zen software to present the signal intensity.

The quantitative values are presented as means ± SD. Statistical differences were determined using the two-sided unpaired Student t test. Differences between groups were considered statistically significant when P < 0.05.

Source data for this study are provided in the Supplementary Materials. Owing to the large size of the 3D confocal and Airyscan superresolution image files, raw data sets are available from the corresponding author, S.-C.T., upon reasonable request.

The lobules of human pancreas strongly scatter light, thus appearing opaque under stereomicroscopy (Fig. 1A). In microcyst formation, however, the small sac-like cavities reduce opacity, creating contrast against a black background (Fig. 1B and C). We explored this unique feature by combining vibratome-based stereomicroscopy (350 µm tissue thickness) with microtome-based H&E histology (gold standard, 3 µm thickness) to detect and confirm microscopic cystic changes in the donor pancreas (Fig. 1D–F; additional examples are presented in Supplementary Fig. 3). In signal analysis, we quantified the changes by comparing the signal intensities of the three primary colors—red, green, and blue—of the vibratome section within the microcyst against those of the surrounding parenchyma on a black background (Fig. 1G). The results showed that the signal intensity decreased by 55% (red; P < 0.001), 52% (green; P < 0.001), and 47% (blue; P < 0.001) inside the microcyst compared with the normal parenchyma, demonstrating the sensitivity of stereomicroscopy in microcyst detection.

To validate our imaging approach, we used the low-grade PanIN (Fig. 1H–M and Supplementary Fig. 4), widely recognized and commonly found in adult pancreases (31–33), as a positive control for epithelial remodeling in the red, green, and blue signal analysis (Fig. 1N). The results showed that the signal intensity decreased by 27% (red; P < 0.001), 24% (green; P < 0.001), and 16% (blue; P < 0.001) in PanIN compared with the perilesional parenchyma, affirming the utility of stereomicroscopy for detecting lobular remodeling in the human pancreas.

We identified 36 microcysts and 54 low-grade PanINs in the body and tail regions of 10 human donor pancreases (survey of 5,592 vibratome sections; individual donor information is listed in Supplementary Table 1). The pancreatic head and neck regions were resected and excluded from the analysis due to logistical limitations.

At the cellular level, cystic changes and low-grade PanINs represent two distinct types of epithelial remodeling as defined by H&E histology: the former is lined by mucin-free, flattened or cuboidal duct cells and frequently exhibits intraluminal secretions (34,35), while the latter consists of columnar duct cells with abundant supranuclear mucin, accompanied by peri-PanIN fibrosis (36,37). Notably, despite their differences (especially in mucin content), microcysts and low-grade PanINs are not mutually exclusive. Figure 2A–F demonstrates that these two types of epithelial remodeling can be simultaneously detected and identified using stereomicroscopy and H&E histology. We identified 10 pairs of microcysts and PanINs within the same vibratome section among the 36 microcysts (10 of 36 [28%]) and 54 low-grade PanINs (10 of 54 [19%]) in the body and tail regions of the 10 donor pancreases. Additional examples are provided in Supplementary Fig. 5. Importantly, this unique finding enabled a side-by-side comparison and illustration of the microcyst and low-grade PanIN using paired CK7 and insulin immunohistochemistry.

Previously, in the human PanIN microenvironment, we observed a localized increase in islet density accompanied by peri-PanIN/lobular islet aggregation (33,38). Here, using PanIN as a positive control, we detected and identified the microcyst-islet association by integrating stereomicroscopy (Fig. 2A), H&E histology (Fig. 2B–D), and fluorescence imaging (CK7 and insulin immunoreactive signals) (Fig. 2E and F). This multimodal approach shows 1) the primary structure of the microcyst is lined with CK7⁺ flattened/cuboidal duct cells, 2) the CK7⁺ epithelium contacts β-cells in the intralobular domain (Fig. 2E), and 3) the mean size of microcysts is significantly smaller than that of low-grade PanINs (1.22 ± 0.56 mm vs. 2.42 ± 1.05 mm; P < 0.001) (Fig. 2G). Both are markedly smaller than the detection limit of standard cross-sectional imaging (computed tomography or MRI, ∼1 cm). The size comparison of the 10 pairs of microcysts and PanINs on the same tissue slide confirms the result (microcysts at 1.20 ± 0.77 mm vs. PanINs at 1.97 ± 0.80 mm; P < 0.05) (Fig. 2H).

We next conducted a side-by-side analysis of microcysts and PanINs (positive control) on the same tissue slide to quantify the increases in CK7 and insulin signals associated with epithelial remodeling (Fig. 3). Compared with the normal lobule, the microcyst and PanIN exhibited 4.50 ± 1.65- and 3.03 ± 1.41-fold increases in CK7 signal density, respectively. In both the microcyst and PanIN analyses, the convoluted epithelium appears to contribute to the marked increases in CK7 density (Fig. 3D–F). In the same microcyst and PanIN microenvironment, we identified 2.10 ± 1.45- and 2.68 ± 1.48-fold increases in insulin signal density, respectively, compared with the normal lobule (Fig. 3C and G–I).

Similar increases in CK7 and insulin signals were also observed in individually detected/formed microcysts. Close contacts between CK7⁺ and insulin-positive pixels were observed, with noticeable overlaps in the low-magnification micrographs. Figure 3J–N presents five examples of the close epithelial and β-cell contacts in the microcystic environment. Overall, the average insulin signal density in the 36 microcystic environments was 83 ± 27% higher (P < 0.001) compared with the adjacent normal lobule (Fig. 3O). In a control experiment, we used paired glucagon and insulin labeling to reveal that α- and β-cells are both present in the same microenvironment (Fig. 3P–R), confirming the duct and islet cell association in the microcystic changes.

Inside the microcysts, luminal contents, such as fluids, necrotic materials, and cellular debris, were frequently observed, similar to those seen in the clinical specimens of pancreatic cysts. In particular, these luminal contents were visible in four-channel fluorescence imaging of the duct–α-/β-cell clusters with fluorescence signals overlaying on the transmitted light micrograph (Fig. 4). Using CK7⁺ duct epithelium as a boundary landmark, we identified three distinct signal types associated with the luminal contents.

First, luminal fluids appeared as a fluorescent smear distributed across the luminal space (Fig. 4A–C), closely resembling the pink/eosinophilic pattern seen in H&E-stained sections of microcysts (e.g., Fig. 1D’). Second, luminal debris was visible under transmitted light microscopy, appearing as dark or translucent areas (Fig. 4D–H; also seen in Fig. 3L, M, and Q). Third, autofluorescence was associated with the luminal debris. This artifact became especially prominent with anti-glucagon AF-488 labeling intended to identify periluminal α-cells, as comparable signal intensity was observed within the microcyst (Fig. 4I). This high background signal is due to the increased autofluorescence at lower wavelengths, where endogenous fluorophores in the luminal secretions/debris absorb and emit efficiently, mimicking specific labeling signals (39). In contrast, at the higher-wavelength AF-647–nm channel, autofluorescence was significantly lower, allowing for a clearer distinction of AF-647–labeled β-cells from the background (Fig. 4J).

These findings highlight that 1) luminal contents within microcysts are heterogeneous; 2) these contents are sources of autofluorescence that must be accounted for to avoid false-positive signals; and 3) high-wavelength fluorescence is preferable for identifying the duct–β-cell cluster, while low-wavelength signals, such as those from AF-488–labeled α-cells, require careful interpretation to differentiate specific labeling from autofluorescence artifacts.

Because artifacts (both false-positive and false-negative results) are a major concern in investigating duct–β-cell associations in a novel environment, in this research the human pancreases with microcysts were embedded in the high-refractive-index (high-n) A-ha copolymer (25) (Supplementary Fig. 6A) for tissue clearing and antifade 3D/Airyscan superresolution imaging (25,26).

Macroscopically, when a specimen is embedded in the A-ha copolymer (similar to resin embedding in electron microscopy), the pancreatic lobules become transparent with a substantial increase in transmittance across the specimen (n = 1.53) (Supplementary Fig. 6B). This tissue-clearing phenomenon is similar to immersing the pancreas in high-n liquids such as glycerol (n = 1.46), methyl salicylate (organic solvent; n = 1.54), clearing-enhanced 3D (Ce3D; aqueous solution; n = 1.50), and CytoVista (organic solvent; n = 1.50) (Supplementary Fig. 6C–F, positive controls). In modern 3D histology, tissue clearing is essential to prevent false-negative results due to light scattering during in-depth fluorescence imaging (40).

It is important to note that A-ha–based clearing via polymerization (Supplementary Fig. 1) results in a mild expansion of the formalin-fixed pancreas by 7.3 ± 1.2%, whereas liquid-based clearing with glycerol, methyl salicylate (organic solvent), Ce3D (aqueous solution), and CytoVista (organic solvent) causes 12.3 ± 1.8%, 30.8 ± 3.1%, 9.8 ± 2.9%, and 15.0 ± 1.7% shrinkage, respectively (Supplementary Fig. 6G). In clinical sample fixation, formalin-induced protein cross-linking is known to cause tissue shrinkage, with ∼30% shrinkage observed during the processing of surgical human colorectal specimens (41). The A-ha–based tissue clearing gently counteracts this artifact rather than aggravating the situation.

Next, we used consecutive pancreas sections—one for microcyst detection and confirmation (Fig. 5A–C) and the other for panoramic-to-3D/Airyscan superresolution imaging (Fig. 5D–N)—to perform a detailed characterization of the exocrine-endocrine association in microcystic changes. This cm-to-µm magnification of the duct and β-cells reveals 1) β-cells in the epithelium (Fig. 5F–J) and 2) the duct–β-cell cluster (Fig. 5K–N). Notably, in the epithelium, CK7⁺ cytokeratins in the duct cells accumulate in the apical domain, while insulin-positive vesicles in the β-cells are located near the basal domain (Fig. 5H–J), highlighting polarized tissue and cellular organizations. In the cell-cell association, the periluminal β-cells were juxtaposed with duct cells, showing alignment and direct contacts between the two distinct cell types. Supplementary Videos 1–6 provide six examples from six donors of in-depth Airyscan superresolution imaging, validating the duct–β-cell association in a 3D space continuum.

In a control experiment, using the CK7⁺ cytokeratins as the luminal boundary to avoid false-positive results (Fig. 4), we applied the paired glucagon and insulin labeling to confirm the duct–α-/β-cell clusters in the microcystic environment at the vesicle level (Supplementary Fig. 7 and Supplementary Videos 7 and 8).

Technically, we would like to stress that because panoramic-to-3D/Airyscan β-cell imaging requires multiple rounds of 3D fluorescence microscopy (Fig. 5E–N), the photostability of the labeled β-cell vesicles is crucial to prevent photobleaching (false-negative result) (Supplementary Fig. 8) from prolonged illumination light exposure under high-power lenses. To confirm the stability, we used a second superresolution imaging method—stimulated emission depletion microscopy, known for its tendency to induce photobleaching in fluorophores (42,43)—to identify cytokeratins and vesicles within the same duct–β-cell cluster (Supplementary Fig. 9). In the A-ha copolymer, because the solid condition limits diffusional contact between indicator dyes and reactive species (e.g., oxygen [44,45]), photochemical alterations that are difficult to avoid in liquid clearing conditions are minimized (25).

In Fig. 5D–N, the integration of stereomicroscopy with paired CK7 and insulin immunohistochemistry enables 3D/Airyscan superresolution imaging of the duct–β-cell cluster. Using this sensitive approach, we identified 17 small (<800 µm; arbitrary cutoff) areas exhibiting distinct CK7 localization and duct–β-cell clusters in the same microenvironment (Figs. 6 and 7). By using H&E histology (performed on adjacent microtome sections), 9 of the 17 areas were confirmed as microcysts, showing flattened and/or cuboidal duct cells lining the sac-like pockets (Fig. 6 and Supplementary Fig. 10; lower quartile of microcysts in Fig. 2G). Examination of these microcysts shows 1) clustered and scattered β-cells in the cystic environment (Fig. 6J and K) and 2) small-diameter ducts (10–20 µm) surrounded by β-cells (Fig. 6L–N), resembling the intraislet ducts previously reported (33,38,46,47).

In the remaining eight areas, epithelial remodeling could not be determined by gold-standard H&E histology, as the areas of interest were too small to be visualized or processed via microtome sectioning—a limitation inherent to H&E histology. However, as shown in Fig. 7, panoramic-to-3D/Airyscan imaging of these areas successfully detected and identified the duct–β-cell cluster (Fig. 7D–K) and β-cells in the epithelium (Fig. 7L–N). Supplementary Fig. 11 provides a gallery display of the duct–β-cell clusters visualized in this condition. The average size of the 17 microenvironments (<800 µm) with duct–β-cell clusters is 0.61 ± 0.16 mm. At this scale, we demonstrate that the local CK7⁺ epithelium remodeling is detectable and that duct–β-cell clusters can be characterized at the cytokeratin and vesicle levels, confirming the remodeled duct and associated β-cells.

The lobules of the human pancreas are predominantly normal; however, local tissue remodeling due to aging, injury, and/or fatty infiltration is inevitable (48,49). In this study, to detect lobular changes, we examined cadaveric donor pancreases using 3D/2D multimodal imaging—a combination of stereomicroscopy, H&E histology, transmitted light imaging, and 3D/Airyscan superresolution imaging. This approach successfully detected and identified microcysts and duct–β-cell clusters in the same microenvironment. Upon magnification, the cytokeratins of duct cells and insulin-positive vesicles of β-cells were simultaneously revealed in the apical and basal domains of the cystic epithelium, respectively. The results highlight the polarized tissue and cellular organizations as well as the high resolving power of our imaging approach. Technically, embedding the human pancreas in the solid, high-n A-ha copolymer (Supplementary Fig. 1) allowed for repeated and panoramic-to-3D/superresolution imaging, facilitating a robust and clinically relevant investigation of the duct–β-cell association.

The precise tissue and cellular events leading to the formation of duct–β-cell clusters remain unknown. In the formation of microcysts, one potential mechanism involves the blockage of a terminal duct, causing fluid accumulation and subsequent epithelial dilation. As this process progresses, the expanding epithelium likely compresses the surrounding tissues, including the islets. However, duct expansion cannot explain 1) the cystic epithelium consisting of juxtaposed duct cells and β-cells (Fig. 5G–N), and 2) the presence of small ducts, 10–20 µm in diameter, surrounded by β-cells (Fig. 6L–N). The latter resembles the intraislet ducts observed in rodent models, children, and the PanIN microenvironment (33,38,46,47). Moreover, because the duct–β-cell cluster was identified in microcysts <800 µm in diameter (<10³ in volume vs. a 1-cm cyst observed in clinical conditions), we speculate that at this scale, the cluster formation is unlikely due to cell-cell compression. Instead, it may represent a regenerative response to local insults, such as intralobular duct obstruction, inflammation, or injury. This concept suggests a dynamic and adaptive process, potentially involving the emergence of newly formed or transformed β-cells. However, given that the image data represent only a snapshot in time, this interpretation remains speculative and requires cautious consideration.

Morphologically, the formation of microcysts represents a distinct phenomenon in which the cystic structure—composed of flattened and/or cuboidal epithelium—appears to form de novo, with no preexisting sac-like epithelial structures in the lobules. This contrasts with the hyperplastic ducts seen in human PanINs, where neoplasia arises from the overgrowth of existing interlobular columnar epithelium. Because of this fundamental difference, the presence of β-cells in the flattened cystic epithelium cannot be explained by the intercalation mechanism proposed by Blaine et al. (50). Blaine et al. (50), through lineage tracing experiments, investigated the origin of β-cells in epithelial structures, linking hyperplastic ductal epithelium in mice (induced by transforming growth factor-α overexpression) to analogous human conditions, including the epithelia of chronic pancreatitis and PanIN associated with pancreatic ductal adenocarcinoma, two conditions that markedly differ from cystic changes in risks and malignancy. Still, despite the differences in species (mouse vs. human) and epithelium (columnar vs. flattened/cuboidal) in duct and β-cell examinations, it is crucial to recognize that distinct tissue remodeling mechanisms might independently drive the formation of duct–β-cell clusters under varying physiological or pathological contexts, as evidenced by the coexistence of PanINs and microcysts (Fig. 2), each associated with duct–β-cell clusters in the human pancreas.

Given the potential for artifacts in analyzing duct–β-cell associations, we embedded the human pancreas in the solid high-n A-ha copolymer for tissue clearing and fluorescence signal preservation (Supplementary Figs. 6, 8, and 9) (25,26). This chemical environment is specifically designed to mimic paraffin and resin embedding—methods commonly used in clinical laboratories to solidify tissues, such as formalin-fixed surgical or needle biopsy specimens—for long-term preservation and to facilitate micro- and ultrastructural analyses. As shown in Figs. 5–7, upon embedding, the fluorescently labeled human pancreas was stably preserved, allowing for high-power magnification of CK7⁺ cytokeratins and insulin-positive vesicles to accurately identify the duct–β-cell clusters while minimizing photobleaching during imaging (25).

Despite its effectiveness in superresolution imaging, this polymer-based 3D analysis in the solid state has three limitations. First, similar to resin embedding in electron microscopy, once the human pancreas is solidified, the process is irreversible, precluding additional tissue or cellular analyses, such as gene sequencing. Second, because the polymerization process is irreversible, the gold-standard H&E histology must be performed on adjacent tissue sections rather than on the same slide to confirm epithelial remodeling (as seen in Figs. 5A–G and 6A–I). Third, in superresolution imaging, the working distance of high-power lenses is limited. Therefore, special care is required to protect the objective from scratching during deep-tissue imaging. Despite these limitations, the A-ha embedded human pancreas (Supplementary Fig. 1) resembles a standard 2D pathological section: both are optically clear and chemically stable, enabling repeated imaging, preservation, transfer, and sharing among laboratories. These qualities are crucial for building consensus on the β-cell microenvironment in regeneration and for resolving conflicting results in the literature.

In summary, before this research, microscopic cystic changes had been reported in the human pancreas, but their impact on the local β-cell population was unclear. By using low-grade PanINs as a positive control for epithelial remodeling, we used panoramic-to-3D/superresolution imaging to qualitatively and quantitatively characterize the microcystic changes, revealing β-cells in the cystic epithelium and the duct-β-cell cluster. Although this study is observational and does not elucidate the mechanisms underlying the presence of clustered β-cells in the microcystic environment, the chemical and imaging tools we developed to characterize (and preserve) the duct–β-cell association offer valuable insights and advancements for continued investigation into the human duct and islet cell remodeling in health and disease.

Acknowledgments. The authors are grateful for the support from the Confocal Imaging Core at National Tsing Hua University, Taiwan, which is sponsored by the National Science and Technology Council (NSTC 112-2740-M-007-001).

Funding. This work was supported in part by grants from Taiwan National Health Research Institutes (NHRI-EX112-11225EI and NHRI-EX113-11225EI) to S.-C.T., Y.-W.T., and Y.-M.J., Taiwan National Science and Technology Council (NSTC 113-2314-B-007-004-MY2) to S.-C.T. and (NSTC 113-2740-B-001-005) to C-.N.S. and S.-C.T., and National Tsing Hua University, Taiwan (intramural 110F7MAKE1 and 112QI034E1) to S.-C.T.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. C.-Y.L., T.-C.K., Y.-H.C., H.-P.C., C.-C.C., Y.-M.J., and Y.-W.T. contributed to human pancreas acquisition, preparation, and clinical analysis; Y.-H.C., S.-J.P., F.-T.H., M.-H.C., L.-W.L., C.-N.S., H.-J.C., and S.-C.T. contributed to 3D histology and image presentation. C.-N.S., Y.-M.J., Y.-W.T., and S.-C.T. obtained funding. S.-C.T. contributed to drafting of the manuscript. All authors contributed to the study concept and design. All authors contributed to data analysis and interpretation of data, revised the manuscript critically for intellectual content, and approved the final version of the manuscript. S.-C.T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.