Stabilization of the classical phenotype upon integration of pancreatic cancer cells into the duodenal epithelium

Highlights • PDAC cells in the duodenal epithelium mimic intestinal cells and co-opt the basement membrane.• Intramucosal PDAC location is strongly coupled to the classical phenotype and to intestinal traits.• Intratumoral heterogeneity is linked to specific tissue compartments, which shape phenotype plasticity of PDAC cells.


Introduction
PDAC is one of the most lethal tumors with a five-year survival rate of less than 9% [1] . Although clinically perceived as uniformly aggressive, PDAC subtypes with varying clinical outcomes can be defined based on transcriptional profiling [2][3][4][5][6] . Classifications distinguish two major subtypes, "classical" and "basal-like", although intermediate states and less common subtypes exist [ 2 ]. Basal-like tumors are associated with a worse prognosis [ 2-4 , 6 ]. While subtyping based on bulk transcriptomic data is prognostically valuable, cells with both a basal-like and a classical phenotype co-exist in individual tumors, as revealed by single-cell RNA sequencing [ 2 ]. Genetic changes, such as allelic imbalances of mutant KRAS , contribute to this intratumor heterogeneity [ 2 , 7 ]; however, mouse models and organoid co-cultures have demonstrated a central role for the microenvironment in shaping the PDAC tumor cell phenotype [ 8 , 9 ], and there is strong support from experimental studies that the non-malignant microenvironment can reprogram malignant cells to a normal-like behavior [ 10 , 11 ]. In human PDAC, cancer cell states have recently been linked to adjacent fibroblast subtypes [ 12 ]. Nevertheless, the full extent to which microenvironmental cues shape tumor cell phenotypes remains unclear.
Routine pathological assessment regularly reveals morphological heterogeneity in pancreatic tumors [ 13 ]. In clinical cases where a small intestinal mass is biopsied endoscopically, PDAC can occasionally be misdiagnosed as an intestinal neoplasm, or even be mistaken for reactive small intestinal changes. This rare, but clinically important phenomenon -that poses diagnostic difficulties -has previously been termed "intestinal mimicry", and an immunohistochemical marker panel has been proposed to improve PDAC diagnosis based on intestinal biopsies [ 14 , 15 ].
The ability of PDAC cells to escape the pathologist's eye once settled in the duodenal epithelium suggests strong phenotypical changes in the tumor cells, while leaving the underlying stroma relatively intact. This is remarkable, given the destructive mode of growth and the strong desmoplastic stromal reaction that otherwise characterize PDAC [ 13 ].
Based on the emerging PDAC subtypes, we have studied in detail the changes relating to intestinal mimicry that occur in the tumor cell phenotype upon switching location from the pancreas to the small intestine. We have systematically mapped tumor cell phenotype dynamics in a unique cohort of PDAC patients with duodenal infiltration, collected over more than a decade of routine pathological diagnostics at a large tertiary care center specializing in pancreatic resections. We found that PDAC cells in the small intestinal epithelium revert to non-destructive growth, and switch to a purely classical phenotype upon epithelial integration. In the duodenal epithelium, protein expression of PDAC cells mimics that of small intestinal enterocytes, while the stroma retains its small intestinal identity devoid of desmoplasia. For the first time, our results link the small intestinal microenvironment to defined shifts in PDAC tumor cell subtypes. Together, they suggest that intestinal mimicry provides the remarkable -yet largely overlookedpossibility of studying cancer cell differentiation towards a less aggressive, near-normal phenotype in relation to microenvironmental cues and within spatially defined tissue compartments.

Patients
Patients with duodenal infiltration of primary PDAC were identified by a retrospective search in the pathology archive database of the Karolinska University Hospital, Huddinge, Sweden, and through routine diagnostic pathology between 2008 and 2020. In all cases, the tumor epicenter was located in the pancreas and all tumors were classified as PDAC histologically and by the local multidisciplinary tumor board.
Cases in which infiltration of PDAC cells into the small intestine was described in the pathology report, or for which the participating pathologists (CFM, Béla B, LS) noted this phenomenon, were selected and systematically reassessed based on available hematoxylin and eosin (H&E) staining and immunohistochemistry (IHC). Clinical data were obtained by retrospective chart review.

Serial multiplex quantitative immunohistochemistry
Serial multiplex quantitative immunohistochemistry (smq-IHC) was performed as described previously [ 16 ]. Briefly, formalin-fixed paraffinembedded (FFPE) samples were cut to a thickness of 4 μm and stained on an automated stainer (BOND-MAX, Leica Biosystems, Germany) as part of the diagnostic routine in a clinically accredited histology lab. Staining procedures have been described previously [ 16 ]. Antibodies and staining protocols are presented in Supplementary Table 1 .
For quantification, H&E stains as well as IHC for Mothers against decapentaplegic homolog 4 (SMAD4) and Tumor protein 53 (p53) guided the identification of tumor regions and individual cells, which were then manually matched to corresponding regions in serial sections, while SMAD4/p53 stains were used to navigate through the sections ( Supplementary Fig. 1 ). All identifiable tumor cells in the mucosa and submucosa of one representative paraffin block were included in the quantification process. The average areas on which the calculations are based were: 15.3 mm 2 (range 3.7-36.6 mm 2 ) for the mucosa and 37.2 mm 2 (range 7.0-153.9 mm 2 ) for the submucosa, respectively.
Quantification was achieved by calculating the percentage of positive cells with respect to all tumor cells in the two separate regions, i.e. mucosa and submucosa.
For all antibody stains, at least n = 10 cases were included for the final assessment, based on staining quality and availability. For all quantified antibody stains, statistics are based on the evaluation of at least n = 15 matched mucosa/submucosa pairs. This study was approved by the responsible Ethical Review Board (no. 2020-06115, Etikprövningsmyndigheten and 2015/259-31/2, Etikprövningsnämnd , Sweden).

FFPE
tissue sections from a cohort of Kras LSL −G12D/ + ;Trp53 LSL −R172H; + ;Pdx1-Cre (KPC) mice [ 17 ] described previously [ 18 ] were analyzed for the presence of duodenal invasion based on available H&E sections. One animal was identified where PDAC cells had infiltrated the duodenum; FFPE sections from this mouse were assessed for expression of the high-motility group AT-hook 2 (HMGA2) protein by immunohistochemistry, as described previously [ 18 ]. The antibody used is listed in Supplementary

Consistent morphological changes in PDAC cells in the duodenum
A total of n = 20 patients in whom PDAC cells had infiltrated the entire thickness of the duodenal wall were identified. All patients ( n = 8 females, n = 12 males, aged 64-82 years, median age 71.8 years) underwent resection according to Whipple at Karolinska University Hospital, Stockholm, Sweden. A total of n = 19 patients passed away during the observation time, median overall survival was n = 559 days after operation (range n = 3 to n = 2460 days). One patient was alive when data collection was completed in January 2021.
In patients where PDAC cells had infiltrated the duodenal epithelium, we observed consistent morphological changes that accompanied tumor cell integration into the epithelial layer ( Fig. 1 A ). In all cases, the mucosal architecture was strikingly preserved, and the epithelial lining exhibited either normal morphology or mild reactive atypia, interspersed with areas of columnar epithelium with a dysplastic/neoplastic appearance, in line with the previously recognized diagnostic challenge [ 15 ]. To confirm unequivocally that the neoplastic cells in the duodenal mucosa were of pancreatic origin, rather than reactive intestinal cells secondary to PDAC infiltration into the submucosa, we used two independent immunohistochemical markers, SMAD4 and p53; SMAD4 is lost in approximately half of all pancreatic cancer cases [ 19 ]; while TP53 mutations, which lead to accumulation of mutant p53 protein, are detectable in 50% of patients, independently of alterations in SMAD4 19 . Both markers confirmed the seamless integration of PDAC cells into the SMAD4 + ve /p53 −ve/low duodenal epithelium, without destruction of the mucosal architecture ( Fig. 1 B ).
Intramucosal PDAC cells were well differentiated and polarized, in contrast to the submucosa, which harbored pleomorphic cancer cells and irregular glands, consistent with primary PDAC histomorphology. A hallmark of PDAC is its desmoplastic stroma, in which tumor glands are embedded [ 12 ]. Notably, IHC for the desmoplasia marker, podoplanin (D2-40) [ 20 ], revealed no desmoplasia of the subepithelial stroma adjacent to mucosal cancer cell integration, while desmoplasia was present in the submucosa ( Fig. 1 C ). In contrast, protein expression of CD146 and WT1 ( Fig. 1 D ), markers of the intestinal lamina propria [ 21 , 22 ], was preserved in regions adjacent to intraepithelial tumor cells.

Stabilization of the classical phenotype upon epithelial integration
The diverging phenotypes of PDAC cells in mucosal vs . submucosal locations implied a high degree of location-dependency, suggesting that the local microenvironment is a major contributor to the tumor cell phenotype. To quantify these phenotypic differences, we used smq-IHC to assess a panel of 12 protein markers, selected to identify tumor cell characteristics, and to approximate the transcriptional subtypes [ 2 ] (see also Supplementary  Table 2 ), as previous studies had shown that condensed marker panels can differentiate between basal-like and classical subtypes with high accuracy [ 23 ]. The panel comprised intestinal as well as pancreatobiliary differentiation markers (CK20, CDX2, MUC2, MUC1, MUC5AC), all of which are included in the classical transcriptional profile [ 2 ], a pancreatobiliary marker specific to the basal-like profile (CK17), general markers for PDAC cells (CK7, MUC6) and tumor-specific glycoproteins (CA19-9, CA125, and CEA), together with Ki67 to assess proliferation.
The smq-IHC results revealed significantly reduced expression of the basal-like marker, CK17, along with CA125, in the mucosa vs. submucosa. In contrast, the expression of the classical/intestinal markers, MUC5AC, CK20, and MUC2 was significantly increased in the mucosa compared to the submucosa ( Fig. 2 A & B, Supplementary Table 2 ), supporting a strong phenotypic switch away from basal-like towards classical differentiation following intramucosal integration, which was accompanied by enhanced Ki67 positivity ( Fig. 2 A ).

Murine PDAC recapitulates phenotypic plasticity
The KPC mouse model [ 17 ] closely recapitulates PDAC morphology, driven by mutations in Kras and Tp53 under the control of a pancreas-specific promoter ( Pdx1 ). To assess whether small intestinal infiltration of PDAC leads to morphological changes similar to those in humans, we analyzed tissues from n = 6 KPC mice with PDAC. We identified one animal, in which PDAC cells infiltrated the small intestine to the level of the epithelium. Morphological changes were similar to those observed in humans, such that the tumor cells in the epithelial layer were polarized and morphologically mimicked enterocytes ( Fig. 3 A ). Next, we assessed the expression of HMGA2, a transcriptional marker for basal-like tumor cells in human PDAC [ 2 ], that is also expressed in murine KPC tumors and for which staining of murine tissue has been established previously [ 18 ]. In concordance with human PDAC, HMGA2 was downregulated in murine PDAC cells located in the epithelial layer ( Fig. 3 B ) compared to PDAC cells in the submucosa, consistent with the attenuation of the basal-like phenotype upon intestinal integration.

Discussion
The tumor microenvironment influences cancer development and progression, exerting both cancer-promoting and restraining effects. The extent to which tumor location -and hence the spatial relationship of tumor cells to their specific microenvironment -shapes the PDAC cell phenotype is unclear. Here, we show that the integration of PDAC cells into the duodenal mucosa is associated with a quantifiable phenotypic shift towards intestinal differentiation, identifying the duodenal epithelium as a specific PDAC microniche ( Fig. 4 ). Location-dependent morphological changes are accompanied by a loss of basal-like subtype markers in favor of classical subtype markers, corresponding to a switch towards a less aggressive molecular phenotype, and strong intestinal cell-like differentiation of the integrated tumor cells [ 2-4 , 6 ]. Interestingly, mucosal PDAC cells were cycling (Ki67 + ve ) at higher levels than submucosal tumor cells, suggesting the uncoupling of differentiation from proliferation. While the increase in Ki67 positivity in conjunction with a less aggressive phenotype may appear counterintuitive, results from studies on the prognostic value of Ki67 expression in PDAC have been mixed [ 24 , 25 ]. The intestinal mucosa is a highly proliferative tissue that renews every five days [ 26 ] and hence, the increase in proliferative activity may be interpreted as part of the alignment of PDAC cells with their epithelial location.
Our results establish that PDAC cells integrate into the epithelial compartment of the duodenum, where they progress in an in situ -like manner. Importantly, for the first time, we provide evidence that this phenomenon is connected to the distinct tumor subtypes. Non-destructive growth in the epithelium requires PDAC cells to respect the native basement membrane, and it does not induce stromal desmoplasia; hence, it differs significantly from tumor growth in the pancreas. In the process of epithelial co-option, PDAC cells establish direct contact with adjacent non-neoplastic duodenal cells, and we speculate that the intercellular crosstalk between PDAC and enterocytes is key for tumor growth. Recent studies have begun to shed light on the molecular pathways that allow tumor cells to replace their non-malignant neighbors. These pathways involve tumor-host cell competition driven by Hippo or JNK signaling [ 27 , 28 ], and the secretion of Wnt antagonists by tumor cells to gain a competitive advantage [ 29 ]. It is remarkable that PDAC might be capable of adopting similar mechanisms outside of its host organ, and that the change in the mode of growth (destructive vs . replacement) is tightly connected to the cellular phenotype. Further studies are warranted to disentangle the underlying molecular pathways. However, a particular challenge is the fact that intestinal mimicry is infrequently observed in both    humans and mice, limiting the number of specimens available to study the underlying mechanisms. It is unclear to what extent sampling bias might contribute to the scarcity of cases we found, given that assessing the duodenal mucosa for tumor cell integration is only feasible for a limited region of the mucosa. We hope that our smq-IHC data together with, for example, spatial transcriptomic analysis of FFPE tissue at single-cell resolution will help to identify key pathways of non-destructive PDAC growth in the future.
Together, our data define a real-life endpoint of the phenotypic plasticity of PDAC cells in humans. They strongly support a model in which basallike vs . classical tumor subtypes are highly influenced by microenvironmental cues. The consistency of this phenomenon suggests that this and similar cohorts displaying duodenal invasion can be invaluable for deciphering the molecular underpinnings of PDAC subtype emergence orchestrated by the microenvironment.