Development and characterization of orthotopic patient-derived xenograft models of peritoneal metastatic mucinous appendiceal adenocarcinoma

Abstract

Background

Appendiceal adenocarcinomas (AAs) are a rare and heterogeneous group of tumors for which few preclinical models exist. The lack of preclinical models of AA has hindered drug development and is a major factor in why AA remains without a single Food and Drug Administration-approved systemic treatment.

Materials and methods

Tumors from 16 patients with appendiceal neoplasms (15 AAs and 1 high-grade appendiceal neoplasm) were implanted into the flank and the peritoneal cavity of immunodeficient mice leading to the successful establishment of three AAPDX models. Histological, immunohistochemical, genetic, and transcriptomic comparisons of patient and patient-derived xenograft (PDX) tumors were carried out.

Results

Higher tumor grade, peritoneal implantation, and RAS/RAF mutation were associated with successful tumor engraftment. Comparison of histological, immunohistochemical, and molecular analyses including both RNA and DNA sequencing revealed that the PDX models recreate many of the features of metastatic AAs, but also displayed several differences between paired PDX and human tumors, highlighting the intratumoral heterogeneity of AAs within each patient. Notably tumors from two patients with primarily low-grade mucinous adenocarcinoma converted to high-grade histology in PDX. Transcriptomic comparison of patient and PDX tumors identified increased Myc and E2F signaling, suggesting that activation of Myc may be a driver of the dedifferentiation of AAs. The established PDX models were able to undergo serial passaging and expansion and exhibited stable histological features during this process, allowing for drug testing.

Conclusions

These molecularly profiled, orthotopic PDX models of metastatic AAs represent a unique resource for future exploration to identify novel therapies for this orphan disease.

Highlights

• •AA is an orphan disease with limited preclinical models and no approved drugs.
• •We established three new orthotopic PDX models for drug discovery.
• •Activation of Myc may be a driver of the dedifferentiation of AAs.

Introduction

Appendiceal adenocarcinoma (AA) is a rare cancer with an age-adjusted incidence rate of 1.3 per 100 000 persons, although the incidence has been increasing recently, especially in younger patients.¹ AAs present significant clinical challenges due to limited therapeutic options, limited prospective data, difficulty in measuring disease burden in the peritoneal space, and marked heterogeneity between high- and low-grade tumors.^2,3 Appendix cancer has a unique natural history with metastatic spread limited to the peritoneal space in most patients. Currently, the standard-of-care treatment is cytoreductive surgery followed by hyperthermic intraperitoneal chemotherapy.⁴ However, roughly half of the patients with metastatic AAs have extensive peritoneal carcinomatosis and are not candidates for surgical resection. These patients have traditionally been offered chemotherapy designed for the treatment of colorectal cancer, although there are limited data to support this practice and a recent prospective trial has shown that 5-fluorouracil (5-FU)-based chemotherapy is ineffective in low-grade AAs.²

A lack of preclinical model systems appropriate for drug screening has limited drug development in AAs. AA tumors do not grow well in traditional two-dimensional cell culture and only a few cell line models of AAs exist.^5,6 The number of available patient-derived xenograft (PDX) models is also limited, including only one model from the Jackson Laboratory, and several models created at the University of Oslo.^7,8 PDX models previously developed by the University of Creighton are unfortunately no longer available.⁹ To facilitate drug discovery in AAs we sought to generate a larger panel of AA models by systematically implanting tumors from 16 patients with appendiceal neoplasia into immunodeficient mice, establishing three sustainable orthotopic metastatic PDX models.

Materials and methods

Patients and clinical data

Patients with AAs were approached and consented under an institutional review board-approved protocol (approved date: August 2023; LAB10-0982) at the University of Texas MD Anderson Cancer Center between 2020 and 2024. Clinical data, including patient demographics, clinicopathological, treatment, and outcome data, were abstracted from the Palantir Foundry software system to identify patients with a diagnosis of AAs.

NOD/SCID/IL2Rγ^null (NSG) mice models and maintenance

Female NSG mice, 5-7 weeks of age, were purchased from The Jackson Laboratory (Bar Harbor, ME) and used for tumor implantation. The animals were maintained under specific pathogen-free conditions. Housing and all procedures involving animals were carried out according to protocols approved by the Institutional Animal Care and Use Committee of MD Anderson Cancer Center and conducted in accordance with United States Common Rule.

Establishment of PDX models

Fresh peritoneal tissue specimens from 16 consenting patients with AAs were collected at the time of the cytoreductive surgery at the University of Texas MD Anderson Cancer Center. Pathologic diagnosis was determined by board-certified pathologists with fellowship training and expertise in gastrointestinal pathology following the American Joint Committee on Cancer, College of American Pathologists, and the World Health Organization guidelines.^10,11 The tumor core was divided into 0.5-cm³ blocks of tissue, and implanted in the peritoneal cavity or subcutaneous (flank) tissue of NSG mice (passage 0, P0). The well-being of the mice was carefully monitored, and mice were killed when signs of disease, mainly abdominal distension together with a rough coat and reduced mobility, were seen. The tumor was harvested and cut into ∼25-30 mm³ fragments, six of which were placed in the peritoneal cavity—distributed across the upper and lower abdominal quadrants, as well as both flanks—of NSG mice for subsequent passages. Orthotopic metastatic PDX models were established after subsequent intraperitoneal (IP) passages. Three PDX models were established from patient (Pt) #01, Pt #08, and Pt #11. The clinical histories and therapeutic strategies for these patients are shown in Supplementary Figure S1, available at https://doi.org/10.1016/j.esmogo.2025.100133.

Histological and immunohistochemistry analyses

Histological and immunohistochemical (IHC) analyses were carried out in the Department of Veterinary Medicine and Surgery at MD Anderson Cancer Center. To prepare the tissue microarray (TMA), a TMA Grandmaster machine (3DHistech/Epredia, Budapest, Hungary) was used. The TMA Grandmaster extracted tissue cores at a size of 1.5 mm, arranged in an 8 × 15 array configuration. Three or more tumor cores from at least three different tumor formalin-fixed paraffin-embedded blocks were used for TMA.

For multiplex IHC analysis, we simultaneously assessed multiple protein markers within the TMA sections. The staining procedure involved the application of multiple primary antibodies, including anti-MUC2 (EPR6145, catalog number ab134119; Abcam, Cambridge, UK), anti-CDX2 (EPR2764, catalog number ab76541; Abcam), anti-KU80 (C48E7, catalog number 2180; Cell Signaling, Danvers, MA), anti-Vimentin (D21H3, catalog number 5741; Cell Signaling), and anti-Ki-67 (SP6, catalog number 16667; Abcam). Subsequently, a series of fluorophore-conjugated secondary antibodies were utilized to enable the visualization and quantification of distinct protein markers. Multiple Pt #11 TMA cores displayed large regions of severe necrosis, so quantitative analysis for Pt #11 could not be carried out. The quantified data were obtained from three different tissue cores in each specimen; one tumor tissue from each patient tumor and three tumor tissues from each PDX tumor.

Drug treatment

Eribulin and nab-paclitaxel, obtained from residual unused doses from the Mays Clinic Pharmacy at the University of Texas MD Anderson Cancer Center, were freshly diluted with physiological saline. PDX mice were intraperitoneally injected with diluted eribulin (2 mg/kg) or nab-paclitaxel (250 mg/kg) weekly (three weekly treatments and 1 week off, for two cycles). The IP tumor was visualized using a 4.7T small animal MRI system (Bruker Biospin MRI, Billerica, MA), 30-40 sections of 0.75 mm thickness and 0.25 mm gap of each section, throughout almost the entire peritoneal cavity in each mouse. The tumor growth was evaluated by the modified peritoneal RECIST method, a quantitative measuring system designed for mucinous peritoneal disease, which is the sum of the longest diameter of up to five target lesions in the abdominal cavity.^2,12

Whole exome and RNA sequencing

Fresh tumor tissue samples of patients with appendiceal cancer were obtained from The UT MD Anderson Cancer Center’s Institute Tissue Bank. Germline genomic DNA was also extracted from peripheral blood mononuclear cells of the same patients. The PDX tumors were harvested to extract genomic DNA and total RNAs at The UT MD Anderson Cancer Center’s core biospecimen extraction facility. Genomic DNA from tumors and germline samples and RNA from tumor tissues were sent for whole exome sequencing (WES; average sequencing depth >100×) and mRNA sequencing at Novogene Corporation in Sacramento, California. To segregate mouse and human reads from the raw WES and RNA sequencing (RNA-seq) files of PDX models, Xenome (version 1.0.1-r) was used, with the mouse reference genome (GRCm39) and gene annotation file (GENCODE.vM26) being employed for mapping and annotating mouse reads. Reads that aligned with the human reference genome, including reads that could be aligned with both human and mice reference genomes and ambiguous reads, were considered for further analyses of somatic mutation calls and gene expression studies.

Somatic mutation calls of clinical and PDX tumor samples with matched germline samples were carried out in the Snakemake framework. Implementation of the Snakemake workflow for WES profiles of heterogeneous tumor tissues was described previously.¹³ Briefly, consensus mutation calls from three mutation callers, MuSe2,^13,14 Mutect2,¹⁵ and Strelka,¹⁶ were utilized in detecting somatic mutations of tumor tissues.

Simultaneously, after filtering mouse reads, the raw BAM/FASTQ RNA-seq files of human reads obtained from Xenome were aligned to the human reference genome (GRCh38) using the RNA-seq alignment tool STAR (version 2.7.2b). RNA sequencing and gene expression analysis were carried out as previously described.¹⁷ The RNA-seq data processing pipeline utilized the gene annotation file GENCODE (version 22) in conjunction with HTSeq (version 0.11.0) to extract raw read counts from the STAR-aligned RNA-seq files. Because of sample quality concerns, RNA-seq could not be carried out on Pt #11. Therefore, we only carried out these analyses for Pt #01 and Pt #08.

Single-sample gene set analysis, differential gene expression analysis, and gene set enrichment analysis

Single-sample gene set analysis (ssGSEA) was conducted using the R Package GSVA (version 1.50.0; R Foundation, Vienna, Austria).¹⁸ First, count matrices were normalized using DESeq2’s variance stabilizing transformation. Then, the counts were run through the GSVA package using the Hallmark genesets (h.all.v2023.1.Hs.symbols) from the MSigDB Collections database.¹⁹ Note that ssGSEA settings were used and not the Gene Set Variance Analysis (GSVA) settings. The results were plotted using pheatmap.²⁰ Gene set clustering was conducted using the ward.D method with Euclidean distance measurements.

Differential gene expression analyses (DGEAs) were conducted using the R Package DESeq2 (version 1.42.0) as previously described.^17,21 All genes output from DGEA were ranked by their pi-value: –${\log }_{10}{p}_{\text{adj}}\times {\log }_2\text{FoldChange}$.²² The ranked list was then used for GSEA analysis using the Hallmark genesets (h.all.v2023.1.Hs.symbols) from the MSigDB Collections database.¹⁹ The R package fgsea (v1.28.0) was used to conduct GSEA with 5 000 000 permutations, a weighted enrichment statistic, and excluding gene sets with <15 or >500 genes.²³ Leading edge analysis was also carried out using fgsea.

Statistical analyses and software

Statistical analysis was carried out using GraphPad Prism 9 (GraphPad Software, San Diego, CA) or R v4.3.2 in RStudio v2023.09.01. One-way analysis of variance with Holm–Šidák post hoc test for multiple comparisons or Student’s t-test for two-group comparisons was used to determine the significance of differences. When appropriate, we estimated variation within each group of data and ensured that it was similar between groups that were being statistically compared. The log-rank test was used to determine statistically significant differences between two Kaplan–Meier survival curves.

Results

Clinical, histological, and molecular features of AA tumors implanted as PDX models

All implanted tumors had mucinous histology with one high-grade appendiceal neoplasm (HAMN), and 15 mucinous AAs ranging from well- to poorly differentiated (Table 1). All of the implanted tumors came from patients with metastatic disease limited to the peritoneal space undergoing either cytoreductive surgery or diagnostic laparoscopy, although in the case of the HAMN, no tumor cells were identified outside of the appendix. Peritoneal Carcinomatosis Index scores ranged from 8 to 36 (median 25) and there was also a broad range of cellularity (Table 1 and Supplementary Figure S2, available at https://doi.org/10.1016/j.esmogo.2025.100133); 8 of the 16 patients received neoadjuvant chemotherapy before the surgery during, which tumor tissue was harvested. Consistent with the prior report, the most commonly mutated genes were KRAS, GNAS, and TP53²⁴; one tumor was microsatellite instability-high (MSI-H) detected by the loss of MLH1 and PSM2 by IHC. The mutational status of six tumors was not available.

Table 1.

Histological and molecular features of appendiceal neoplasms from patients enrolled in this study

Patient ID	Age, years	Sex	Histology	Signet ring features	Grade	Peritoneal Carcinomatosis Index score	Number of regions with tumor identified, n/N (%)	Neoadjuvant systemic chemotherapy	Microsatellite	Mutations					Flank, n/N (%)	Peritoneum, n/N (%)
										GNAS	KRAS/NRAS	BRAF	TP53	Others
Pt #01	61	F	Well-to-moderately differentiated mucinous adenocarcinoma	No	2	18	6/9 (67)	No	MSI-H	Not tested	Not tested	V600E	Wild type	MLH1 Methylation	0/6 (0)	8/9 (89)
Pt #02	64	F	Well-differentiated mucinous adenocarcinoma	No	1	33	15/17 (88)	No	Not done	Not done					0/3 (0)	0/3 (0)
Pt #03	61	F	Well-differentiated mucinous adenocarcinoma	No	1	20	2/18 (11)	No	Not done	Not done					0/3 (0)	0/3 (0)
Pt #04	68	F	Well-to-moderately differentiated mucinous adenocarcinoma	No	1.5	23	21/23 (91)	Yes	MSS	Wild type	Wild type	Wild type	D281Y, C176F		0/3 (0)	0/3 (0)
Pt #05	33	F	Well-differentiated mucinous adenocarcinoma (arising from HAMN)	No	1	29	2/21 (10)	No	MSS	Wild type	Wild type	Wild type	Wild type		0/3 (0)	0/3 (0)
Pt #06	64	M	Moderately differentiated mucinous adenocarcinoma	No	3	33	1/1 (100)	Yes	MSS	R201H	Wild type	Wild type	Wild type		0/3 (0)	2/5 (40)
Pt #07	43	F	Poorly differentiated mucinous adenocarcinoma	No	3	8	6/16 (38)	No	MSS	Wild type	Wild type	Wild type	Wild type	NF1 I679fs, PTPN11 F71C	0/3 (0)	0/3 (0)
Pt #08	31	M	Well-differentiated mucinous adenocarcinoma	No	1	35	16/16 (100)	Yes	MSS	Not done					0/3 (0)	6/8 (75)
Pt #09	35	F	High-grade appendiceal mucinous neoplasm	No	n/a	15	0/19 (0)	No	MSS	Wild type	Wild type	Wild type	Wild type		0/3 (0)	0/3 (0)
Pt #10	50	F	Well-to-moderately differentiated mucinous adenocarcinoma	No	2	23	4/7 (57)	No	Not done	Not done					0/3 (0)	1/6 (17)
Pt #11	40	M	Moderately to poorly differentiated mucinous adenocarcinoma	Yes	2.5	20	4/5 (80)	Yes	MSS	Wild type	KRAS G12V NRAS G10R	Wild type	R282W	MYC amplification, U2AF1 S34F	2/3 (67)	4/5 (80)
Pt #12	72	F	Moderately differentiated mucinous adenocarcinoma	No	2	9	3/13 (23)	Yes	MSS	Wild type	KRAS G12D	Wild type	Wild type		0/3 (0)	1/6 (17)
Pt #13	67	F	Moderately differentiated mucinous adenocarcinoma	No	2	31	17/20 (85)	Yes	MSS	R201C	KRAS G12R	Wild type	Wild type	VEGFA R339W	0/4 (0)	0/3 (0)
Pt #14	51	F	Well-differentiated mucinous adenocarcinoma	No	1	36	7/10 (70)	Yes	MSS	Wild type	Wild type	Wild type	Wild type		0/4 (0)	0/3 (0)
Pt #15	73	F	Well-differentiated mucinous adenocarcinoma	No	1	27	13/13 (100)	No	Not done	Not done					0/4 (0)	0/3 (0)
Pt #16	74	M	Moderately differentiated adenocarcinoma	No	2	27	10/18 (56)	Yes	MSS	Not done					1/4 (25)	0/3 (0)

HAMN, high-grade appendiceal neoplasm; MSI-H, microsatellite instability-high; MSS, microsatellite stable; Pt, patient.

Histologic and molecular features associated with successful growth of the PDX model

To generate PDX, tumor tissue from the same patient was implanted into both the flank and peritoneal cavity of multiple mice and monitored for tumor growth for up to 12 months. Peritoneal implantation, high grade, and possessing a RAS/RAF mutation were all predictors of tumor implantation success (Figure 1A and B). The HAMN tumor did not grow in mice, consistent with the low metastatic potential of the tumor and the fact that the peritoneal deposits in these patients contained only mucin and fibrotic tissue without tumor cells. In addition, while Pt #11’s tumor grew in both the flank and the peritoneal cavity, the orthotopic peritoneal PDX #11 exhibited higher cellularity with higher Ki-67 positivity (Figure 1C) and faster growth (Figure 1D), relative to the flank PDXs. Three tumors from the cohort of 16 patients were established as sustainable PDXs after serial passages in NSG mice (Pt #01, #08, and #11). Pt #01 and Pt #08 tumors could grow only in the peritoneal cavity, while Pt #11 tumors grew in both the flank and the peritoneal cavity (Table 1). Tumors from three other patients grew in the first mouse passage but failed to propagate after serial passages. To evaluate if tumor formation in PDX was associated with a more aggressive clinical course, it was first necessary to separate those patients who had complete cytoreductive surgery (all tumors removed) from those with incomplete surgery (tumors known to be left behind). In the four cases of incomplete cytoreduction, we found that progression-free survival was the shortest for the two patients where tumors established PDX. In the 11 cases of complete cytoreduction, we did not see a significant association between PDX establishment and disease-free survival (Supplementary Figure S3, available at https://doi.org/10.1016/j.esmogo.2025.100133).

Figure 1.

Tumor formation rates in AAPDX models. Percentages of tumor formation in AAPDX models. (A) The tumor formation rates in all flank (left) and peritoneal (right) PDX models were plotted. Patient ID numbers are shown in each plot (see all listed in Table 1). (B) The tumor formation rates of PDX models from wild-type (black) and RAS or RAF mutant tumors (red) were plotted. (C) Representative photos, hematoxylin and eosin (H&E)-stained, and Ki-67-stained sections of Pt #11 PDX flank and peritoneal tumors. Original magnification, ×2 or ×20; bar 2 mm (×2) or 50 μm (×20). (D) Pt #11 PDX tumor growth in the flank (n = 4 mice) and peritoneum (n = 4 mice) was monitored by magnetic resonance imaging and evaluated by the modified peritoneal RECIST (mpRECIST) method (see details in the ‘Materials and Methods’ section). Statistics were determined using a Student’s t-test (∗P < 0.05). AA, appendiceal adenocarcinoma; PDX, patient-derived xenograft.

Histological comparison of human and PDX tumors

Grossly, both PDX #01 and PDX #08 were soft solids while PDX #11 was jellylike and mucinous (Figure 2A). Histologically, Pt #01 and Pt #08 were well- to moderately differentiated and well-differentiated (low-grade) mucinous AAs, respectively, while PDX #01 and PDX #08 were poorly differentiated (high-grade). Both Pt #11 and PDX #11 were poorly differentiated mucinous adenocarcinoma (Figure 2B, and summarized in Supplementary Table S1, available at https://doi.org/10.1016/j.esmogo.2025.100133). All three patient tumors displayed mucinous features; however, both PDX #01 and PDX #08 no longer displayed mucinous features and were far more cellular than their respective patient tumors. Only PDX #11 maintained mucinous features following implantation and passaging. These features were further confirmed using a TMA with staining for CDX2 (intestinal epithelial cells), KU80 (human cells), Vimentin (stromal cells), mucin short variant S1 (MUC1, associated with tumor progression), and Ki-67 (proliferative cells). In Pt #01 and Pt #08 tumors, limited numbers of Ki-67⁺CDX2⁺MUC1^low cells were found while such Ki-67⁺CDX2⁺MUC1^low cells were highly enriched in PDX #01 and #08 tumors (Figure 2C, Supplementary Figure S4A-D, available at https://doi.org/10.1016/j.esmogo.2025.100133). By contrast, Pt #11 and PDX #11 overexpressed MUC1 (Supplementary Figure S4E and F, available at https://doi.org/10.1016/j.esmogo.2025.100133). Ki-67⁺CDX2⁺MUC1^high cells enriched in Pt #11 as well as in PDX #11. These IHC results suggest similarities in proliferative cells between Pt #11 and PDX #11.

Figure 2.

Histological comparison of human and patient-derived xenograft (PDX) tumors. PDX models from patient (Pt) #01, #08, and #11 were successfully established. (A) Representative images of the tumors formed in the peritoneum of NSG mice. The tumors from PDX #01 and #08 were solid tissues, while those from PDX #11 formed jellylike soft tissue. (B) Histological comparison of patient and PDX tumors. Original magnification, 10×; bar 200 μm. (C) Multiplex immunohistochemical staining for 4′,6-diamidino-2-phenylindole (DAPI; blue), CDX2 (green), Vimentin (Vim; red), and Ki-67 (white) in Pt and PDX tumors on a tissue microarray (TMA). Original magnification, 10×; bar 200 μm. (D) Tumor cellularity (top left), stromal cellularity (top right), Ki67 positivity in tumor cells (bottom left), and Muc1 positivity in tumor cells (bottom right) on TMA cores were computationally quantified. Tumor cells were characterized as cells staining positive for Ku80 and CDX2. Stromal cells were characterized as Vim positive. Red bars represent comparisons between PDX samples from different tumors and black bars represent comparisons between Pt (n = 3 each) and PDX samples (n = 9 each) in each corresponding tumor. (E) Transcriptomic expression of oligomeric mucins and FCGBP in Pt tumors (blue, n = 1 or 2) and PDX tumors (red, n = 2 or 3). Statistics were determined using a Student’s t-test (ns, P > 0.05; ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001).

Finally, in the computational quantification of TMA and multiplex IHC, PDX #01 and PDX #08 had a higher number of cells (tumor, CDX2⁺ and KU80⁺, and stromal, Vimentin⁺) compared with Pt #01 and Pt #08 while PDX #11 had a similar number of cells to the two Pt samples (Figure 2D, top left and top right). A higher percent of PDX tumor cells were Ki-67⁺ relative to the Pt samples (Figure 2D, bottom left). Similar to our gross and histological findings, a lower percent of PDX tumor cells were MUC1⁺ relative to the Pt samples; PDX #11 was the only sample with a higher percentage of MUC1⁺ cells (Figure 2D, bottom right). In summary, three PDX models were established, with the low-grade AAPDX models PDX #01 and PDX #11 exhibiting different histological features compared with their patient tumors.

Molecular comparison of human and PDX tumors

DNA sequencing of the PDX tumors and comparison to the patient tumors highlight intratumor heterogeneity in AAs. Clinical molecular testing identified Pt #01 as MSI-H with loss of MLH1 and PSM2 by IHC, along with hMLH1 methylation and BRAF^V600E. Additional clinical molecular testing studying other genes was not carried out on this tumor. WES of a tumor taken from the same surgery that generated PDX #01 confirmed a very high TMB consistent with MSI-H (Supplementary Table S2, available at https://doi.org/10.1016/j.esmogo.2025.100133). Interestingly, after two passages, the WES of PDX #01 showed a TMB of only 1.2 mut/MB, indicating that an microsatellite stable clone was selected during growth as a PDX. For both Pt #08 and PDX #08, WES discovered no mutations in KRAS, GNAS, BRAF, or TP53. However, this may be a false-negative result given low cellularity. Interestingly, a mutation in the tumor suppressor gene TSC1 was found in PDX #08 but not in Pt #08 (Supplementary Table S2, available at https://doi.org/10.1016/j.esmogo.2025.100133). On clinical sequencing, which is carried out at a significantly higher depth than research WES, Pt #11 was found to have KRAS^G12V and TP53^R282W; however, these were not seen in research WES. PDX #11 showed retention of the KRAS^G12V, but not the TP53 mutation, again consistent with clonal intratumor heterogeneity in AAs (Supplementary Table S2, available at https://doi.org/10.1016/j.esmogo.2025.100133). PDX #11 also had an NRAS^G10R mutation, which was observed in Pt #11 in clinical sequencing but not in research sequencing.

Tumor expression of oligomeric mucin genes MUC2, MUC5AC, and MUC5B

To see if the earlier findings were also found in the transcriptome, we compared patient tumor and PDX tumor RNA expression of several oligomeric mucins²⁵ as well as FCGBP, which is associated with extracellular mucin structure and function²⁶ (Figure 2E). Both Pt #01 and #08 expressed large amounts of FCGBP, MUC2, MUC5AC, and MUC5B relative to their respective PDX tumors. PDX #11, the only mucinous PDX, expressed high levels of these MUCs as well. PDX #11’s KRAS^G12V mutation might be stabilizing the mucinous histological features.^27,28

High expression of MYC signaling drives dedifferentiation in AAPDX

To further compare the molecular features of the patient and PDX tumors, we analyzed the samples’ transcriptomes of Pt #01 and Pt #08 using RNA-seq. ssGSEA using Hallmark genesets was carried out, which generated four clusters of gene sets with either similar or contrasting expression (Figure 3A).^18,19 Gene sets with high activity in human samples and very little activity in PDX samples included angiogenesis and epithelial to mesenchymal transition; these differences were expected given that murine reads, which in a PDX model is generally from stromal cells, were computationally removed from the PDX transcriptome but remain in the patient transcriptome.¹⁷ The human samples also had greater activity of transforming growth factor (TGF) beta signaling and ultraviolet response down. Gene sets with greater activity in PDX included proliferative gene sets such as E2F TARGETS, G2M CHECKPOINT, MYC TARGETS V1, and MYC TARGETS V2.

Figure 3.

Transcriptomic comparison of #01 and #08 patient and patient-derived xenograft (PDX) tumors. (A) Single-sample gene set enrichment analysis (ssGSEA) on patient (Pt) #01 (left, n =1) and Pt #08 (right, n = 2 each) and PDX tumors (n = 2 each) using the Hallmark genesets. Gene sets were clustered into four clusters, labeled C1-C4. (B) GSEA using the Hallmark genesets to compare transcriptomes from the Pt #01 and Pt #08 PDX tumors (n = 4 total) against patient tumors (n = 3 total). (C) Analysis of leading edge genes from gene sets enriched in PDX tumors, limited to leading edge genes with a Log2 fold change >3.32 (fold change >10). FC, fold change.

Next, to explore the differences between the transcriptomes of the low-grade patient tumors and their paired high-grade PDX tumors, we ran a DGEA and then conducted a gene set enrichment analysis (GSEA) using the ranked list of genes. Using GSEA instead of ssGSEA highlights differences between the patient tumors and the PDX tumors. Many of the findings in the ssGSEA were recapitulated in the GSEA (Figure 3B, Supplementary Figure S5, available at https://doi.org/10.1016/j.esmogo.2025.100133). The gene sets that appeared to be relatively enriched in the patient tumors were similarly relatively enriched in the ssGSEA [epithelial to mesenchymal transition, Normalized Enrichment Score (NES) = −1.26; Coagulation, NES = −1.21]. Meanwhile, the PDX tumors were enriched in the two MYC TARGETS gene sets (MYC TARGETS V1: NES = 2.21; MYC TARGETS V2: NES = 1.72). While these signatures showed higher activity in PDX tumors compared with patient tumors (Figure 3B), they were also strongly expressed in patient tumors (Figure 3A). These data suggest that the well-known oncogene Myc, which has not been previously implicated in the pathogenesis of AAs, may be an important driver of transformation to higher-grade appendiceal tumors.

Leading edge analysis on the GSEA results showed sizable overlap between the MYC TARGETS, E2F TARGETS, and G2M CHECKPOINT leading edges (Supplementary Figure S6, available at https://doi.org/10.1016/j.esmogo.2025.100133). Notably, two of the genes driving the significance of G2-M CHECKPOINT are E2F1 and E2F2, pointing to a direct connection between E2F signaling and cell division in the PDXs (Figure 3C). Other notable leading edge genes involved in promoting cell division are PLK1 (polo-like kinase 1),²⁹ AURKB (aurora kinase B),³⁰ and CDC20.³¹

Histological comparison over the serial passage of Pt #11 tumor

To better characterize the stability of PDX #11, we compared PDX tumors over successive passages. Initial passages (P0) were ∼250 days long, but subsequent passages were shorter and eventually stabilized (Figure 4A). The PDXs maintained the mucinous histological features originally seen in Pt #11 (Figure 4B, Supplementary Figure S7, available at https://doi.org/10.1016/j.esmogo.2025.100133). TMA with multiplex IHC from these serially passaged PDX tumors found that tumor cellularity and the proportion of Ki-67 cells tended to increase with successive cycles (Figure 4B, Supplementary Figures S8 and S9, available at https://doi.org/10.1016/j.esmogo.2025.100133). However, the proportion of Ki-67-positive cells was initially high in P0 before dropping in P1. By P3, Ki-67 had recovered and slightly increased in P4. The decreased passage time, increased cellularity, and increased positivity for Ki-67 point to the PDX tumor adapting to its murine tumor microenvironment with successive passages while maintaining its mucinous histology.

Figure 4.

Multiplex immunohistochemical (IHC) analysis on patient and serially passaged patient-derived xenograft (PDX) tumors. Tumors from patient (Pt) #11 were repeatedly passaged in NSG mice. (A) Schematic representation of terms of each passage (passages 0-4). (B) Multiplex IHC staining for CDX2 (green), Vimentin (red), Ki-67 (white), and MUC1 (magenta), and 4′,6-diamidino-2-phenylindole (DAPI) in a tissue microarray (TMA) of the serial passaged PDX tumors. Original magnification, ×4 (left) and ×20 (right).

Comparison of chemotherapeutic response in patient and matched PDX

Before surgery, Pt #11’s cancer progressed despite treatment with the quadruplet drug regimen FOLFOXIRI (folinic acid, 5-FU, oxaliplatin, and irinotecan; Supplementary Figure S1, available at https://doi.org/10.1016/j.esmogo.2025.100133). We tested the efficacy of FOLFOXIRI on organoids derived from PDX #11 (Organoid #11). Compared with colorectal cancer organoids (LS174T), which are known to be sensitive to FOLFOXIRI,^32,33 organoid #11 was highly resistant with a concentration that causes 50% inhibition of growth (half-maximal inhibitory concentration) two orders of magnitude greater [0.0036 clinically used dose (CUD, see Supplementary Methods, available at https://doi.org/10.1016/j.esmogo.2025.100133) versus 0.45 CUD, sum-of-squares F test P < 0.0001; Figure 5A]. This finding that PDX-derived organoids replicate the clinical resistance observed in Pt #11 suggests that these AAPDX models provide a useful platform for testing drug efficacy. Additionally, the individual drugs folinic acid, 5-FU, oxaliplatin, and SN-38 were tested in organoid #11, with 5-FU showing the greatest efficacy (half-maximal inhibitory concentration = 0.34 CUD, Supplementary Figure S10, available at https://doi.org/10.1016/j.esmogo.2025.100133) which was similar to the quadruplet combination. These data suggest that in contrast to low-grade mucinous AAs, where 5-FU has been shown to be ineffective,² 5-FU is active in high-grade tumors.

Figure 5.

Chemotherapeutic responses of patient-derived xenograft (PDX) #11 tumor. (A) The resistance of PDX-derived organoids to FOLFOXIRI. The relative total areas of organoid #11 in the presence of various doses of FOLFOXIRI for 96 h measured in Incucyte, in comparison with colorectal cancer (CRC) organoids (LS174T), were plotted as % of dimethyl sulfoxide (DMSO; left). Inserted images of organoids are represented. Bar, 100 μm. (B) Mice were injected with saline, eribulin (2 mg/kg), or nab-paclitaxel (250 mg/kg) intraperitoneally (weekly for 3 weeks and 1-week rest, and two cycles); n = 4 or 5 mice/group. (C) Representative magnetic resonance images (MRIs) at 56 days are shown. (D) Tumor size in MRI was evaluated by modified peritoneal RECIST (mpRECIST). (E) Changes in body weight after 1 week (left) or over 49 days (right) are plotted. (A, B, D, and E) Error bars, standard error of the mean. Statistics were determined using a one-way analysis of variance with Dunnett’s post hoc test (ns, P > 0.05; ∗P < 0.05; ∗∗P < 0.01). (F) Kaplan–Meier survival curves showing the percentage of mice treated with eribulin or nab-paclitaxel. The statistical significance of each treatment versus saline was assessed by a log-rank (Mantel–Cox) test. CUD, clinically used dose.

Chemotherapeutic response to eribulin and nab-paclitaxel

Previously, we tested the IP paclitaxel on AA tumor growth in PDX models and confirmed its efficacy with controllable toxicity.¹² Here, we tested IP delivery of two additional agents that also target microtubules. Eribulin, a chemotherapy drug that exhibits microtubule inhibitory activity, is used to treat heavily pretreated metastatic breast cancer.^34,35 Nab-paclitaxel is an albumin-bound paclitaxel. To test these drugs on the established PDX #11 model, we implanted fresh PDX #11 tumors into the peritoneal cavity and treated them with IP-delivered eribulin (2 mg/kg) or nab-paclitaxel (250 mg/kg, corresponding to 25 mg/kg of paclitaxel). Two cycles were administered, composed of weekly doses for 3 weeks followed by 1 week of rest (Figure 5B). Saline was used as a negative control. The PDX #11 tumor grew and was readily detectable by MRI in the peritoneal cavity (Figure 5C). While treatment with nab-paclitaxel more effectively reduced tumor burden in the peritoneal cavity (65.2% reduction versus IP saline at 56 days, P = 0.0032), eribulin also significantly reduced tumor burden (56.7% reduction versus IP saline at 56 days, P = 0.028; Figure 5D). Both eribulin and nab-paclitaxel did not significantly reduce the body weight of mice after the treatment (Figure 5E). Measured at 200 days after starting IP treatments, the survival rate of mice treated with nab-paclitaxel was 25% (P = 0.027 and hazard ratio 4.19, versus saline control); for mice treated with eribulin, the survival rate was 20% (P = 0.494 and hazard ratio 1.58, versus saline control; Figure 5F). Taken together with the known safety record of these agents from multiple prior trials,34, 35, 36, 37 we propose that IP administration of microtubule targeting agents such as eribulin and nab-paclitaxel is a promising therapeutic strategy that should be tested prospectively in patients with metastatic mucinous AAs.

Discussion

This study addresses the scarcity of existing preclinical models by introducing and characterizing three new orthotopic PDX models of peritoneal metastatic AAs. Tumors implanted into the peritoneal cavity, usually the only metastatic site for AAs, demonstrated significantly enhanced growth compared with those implanted in the flank. This observation aligns with previous studies that highlighted accelerated tumor growth and the upregulation of signaling pathways associated with proliferation, such as E2F Targets, G2-M Checkpoint, and Myc Target pathways, in the peritoneal environment.¹⁷ Higher tumor formation rates were also associated with higher grade and mutations in the RAS/RAF pathway. In addition, with the caveat that only four patients in our study had incomplete cytoreductions, we found that human tumors that generated PDX had shorter progression-free survival than those that did not.

One of the three models (PDX #11) maintained its mucinous histology through subsequent passages, although it did display some adaptation to the selective pressures of the murine peritoneal environment. The other two models (PDX #01 and PDX #08) displayed substantial differences between patient tumors and PDX models. PDX #01 lost the MSI-H classification seen in the human tumor, which indicates that there was at least one microsatellite stable subclone in the human tumor. Additionally, both relatively differentiated mucinous adenocarcinomas in the patients became poorly differentiated nonmucinous neoplasms in PDX mice, mimicking the transition from low to high grade that is sometimes seen clinically. Utilizing our single-institution cohort of >3000 patients with appendix cancer,³⁸ we were also able to identify 755 patients with more than one tumor biopsy separated by more than 6 months. Using the Foundry platform³⁹ to extract each patient’s pathological tumor grade, we found that 16% (121/755) of the patients have their appendix cancer tumor progress to a higher grade in the subsequent biopsy (Supplementary Figure S11, available at https://doi.org/10.1016/j.esmogo.2025.100133), consistent with a previous report of 17% dedifferentiation rate.⁴⁰ Although some of the differences in pathologic grade may be the result of subjective differences in grade assignments and/or intratumor heterogeneity,⁴¹ the majority of patients with low-grade tumors do not have a repeat biopsy at the time of progression, so it is difficult to know the true incidence of transformation to higher grade among patients with AAs. Thus while PDX #01 and PDX #08 may not be representative specifically of the low-grade human tumors from which they were derived, they appear to be representative of high-grade AAs.

Notably, these high-grade PDX tumors had relatively increased expression of MYC TARGETS V1 and V2 as well as E2F targets. MYC amplification and oncogenic mutations in MYC have been associated with high-grade AAs.⁴² In general, MYC signaling and E2F signaling interact to promote cell proliferation and block differentiation.43, 44, 45 Our finding that relative to paired low-grade patient tumors, higher-grade PDX tumors exhibit higher levels of MYC/E2F signaling points to increased MYC/E2F signaling as a key driver of high-grade AAs. However, it is unknown if what occurred was due to low-MYC/E2F-expressing cells replicating to form high-MYC/E2F-expressing daughter cells or simply selection for an existing high MYC/E2F-target-expressing clonal population.

The data and methods from this study can be used for establishing and characterizing PDXs of AAs and other appendiceal histologies such as goblet cell carcinoma, where no PDX models exist. Given the lack of Food and Drug Administration-approved systemic chemotherapies for AAs, future resources must go into finding and testing drugs for AAs. Importantly, the three models we established will serve as a viable preclinical platform for AA drug discovery, paving the way for future clinical trials. In summary, our study significantly contributes to the evolving understanding of AAs and establishes a foundation for future advancements in treatment strategies.