Promising preclinical patient-derived organoid (PDO) and xenograft (PDX) models in upper gastrointestinal cancers: progress and challenges

Based on the few published studies, organoids could be cultured from both the mucosal epithelium and tumour tissues of EC patients using suitable media contianing a variety of growth factors, inhibitors and hormones, using matrigel or basement membrane extract as ECM substitutes [23, 24]. Zheng et al. developed a novel oesophageal minimum essential organoid culture medium (E-MEOM) for culturing murine oesophageal organoids that were immunohistologically and transcriptomically similar to the normal oesophageal epithelium [25]. Vega et al. developed a 3D organotypic culture system and demonstrated that inhibition of Notch signalling promoted the transdifferentiation of normal oesophageal squamous epithelium to a Barrett’s oesophagus (BE)-like metaplasia partially through KLF4 upregulation [26]. Karakasheva et al. developed a protocol that could successfully generate primitive organoids from oesophageal squamous cell samples, allowing the study of the molecular mechanisms underlying oesophageal cancer cell propagation in PDO culture [27]. Kijima et al. established PDOs from both tumour and adjacent normal mucosa biopsies of ESCC patients using medium with exogenous stem cell factors, and the success rate of PDO culture was 68.75% (11/16) for ESCC tumour tissues [28]. They evaluated the growth and structural characteristics of ESCC PDOs treated with 5-FU ex vivo and found that treatment resistance was more conducive to tumour-like organ formation, and the potential treatment-resistant cell population was characterized by high CD44 expression and high autophagy capacity [28]. These organoid culture systems provided a comprehensive experimental platform for studying the molecular mechanisms of oesophageal cancer cell development and drug response. However, at present, the successful rate of esophageal cancer organoids is generally low, and it’s sincerely expected that further technical optimization can overcome this issue.

Organoids have been reported to maintain high accuracy in predicting therapy response in cancer patients (Table 2). Vlachogiannis et al. reported the potential of PDOs to predict therapy response using a living biobank, which included 110 fresh biopsies from 71 patients, with 100% sensitivity and 93% specificity to forecast patients’ responses to chemotherapy or targeted therapy [29]. Li et al. established long-term expansion PDOs from resection tissues of oesophageal adenocarcinoma cancer (EAC) patients using a glandular-preferred protocol which added additional Wnt3A to the growth factors combination of culture medium and tested 24 anticancer compounds in these PDOs [30]. For target agents, the drug sensitivity of PDOs was mostly in accordance with the molecular status, such as TP53 and EGFR mutations. For chemotherapeutic drugs, including 5-fluorouracil, epirubicin, and cisplatin, the resistance of PDOs was consistent with the response in patients [30]. Another study reported that a total of 16 EAC PDOs were successfully grown and characterized using short tandem repeat (STR) analysis, whole-exome sequencing (WES), histology, and immunohistochemistry, indicating recapitulation of the tumour histology and genomic characteristics [31]. Additionally, drug testing using clinically appropriate chemotherapeutics and targeted therapeutics showed an overlap between the patient tumour response and the corresponding organoid response. Together with genomic features, an EAC PDO carrying ERBB2 amplification responded to the targeted HER2 agent mubritinib, while wild-type organoids showed no response, providing insight into personalized treatment using PDO-based drug sensitivity tests. Taken together, due to the capacity of PDOs to recapitulate patient responses, these results verified the feasibility of using organoids for testing drug sensitivity.

[i]ESCC: oesophageal squamous cell carcinoma; EAC: oesophageal adenocarcinoma cancer; CRC: colorectal cancer; GI: gastrointestinal; H&E: haematoxylin-eosin staining.

Among reported studies establishing preclinical models of EC, PDX models derived from ESCC patients were more frequently observed than those derived from EAC patients (Table 3). A study established 18 EAC PDX models and confirmed their clinicopathological features [32]. A previous study established 4 ESCC and 13 EAC PDX models from a tumour tissue bank [33]. Zhu et al. reported a 61-PDX sequence, one of the largest ESCC PDX cohorts ever reported, and revealed that EGFR may function as a predictive biomarker for cetuximab response [34]. Dodbiba et al. established 21 PDX models from oesophageal/gastroesophageal junction cancers with a success rate of 38%, and among 7 xenografts with drug tests, only the chemosensitivity of 2 xenografts correlated with the patients’ clinical responses [35]. Zou et al. completed the modelling of 25 PDXs derived from 188 fresh endoscopic biopsy tissues of ESCC patients and established PDX models that retained the histologic and genomic characterizations. Tumour growth inhibition from 5 xenografts exposed to paclitaxel and platinum correlated well with the clinical response of patients [36]. Zhang et al. established 37 ESCC PDX models for preclinical drug discovery, in which models carrying HER2 expression had no response to 5-FU/cisplatin [37]. By applying a humanized ESCC PDX model, Liu et al. reported that indomethacin exerted antitumour activity and enhanced cancer immune responses [38]. Ma et al. analysed the heterogeneity of ESCC across 10 cell lines, 80 TCGA tissues and 2 PDX models at the DNA, RNA and protein levels and characterized various novel TP53 mutations, ECM-receptor interactions, focal adhesion, and olfactory transduction pathways (CNGB1) as indicators for accurate research and precision therapeutic development [39]. These observations highlight that EC PDXs has been established as a reliable preclinical model system with histology, genomic variation, and gene expression patterns consistent to the primary tumor, and have proved to be of great application value in oesophageal cancer drug screening.

PDOs of GC could be cultured from both tissue specimens and malignant ascites (Table 4) [40‐43]. Gao et al. developed 15 GC PDOs from 5 patients and identified similar KRAS alterations and drug sensitivity in primary tumours and paired organoids [41]. Malignant ascites from advanced GC with peritoneal metastasis could be collected to generate organoids, showing divergence between individuals but comparability between ascites and PDOs in histological and genomic landscapes. Additionally, ascites-derived PDOs could be used to evaluate the response of chemotherapy regimens and showed similar drug sensitivity to that of patients [40]. Steele et al. reported that GC PDOs from individuals exhibited divergent morphological features and therapeutic regimen efficiency [42]. One patient exhibited a complete response clinically in accordance with the high sensitivity of their corresponding organoids, while another patient did not show a response to therapy agents even though the derived organoids partially responded. Overall, PDO culture techniques allow the generation of preclinical models from metastatic ascites or primary cancer sites, which representing the molecular characteristics and corresponding medical responses similar to parental tumour.

GC PDOs could be expanded long term and subjected to whole-genome sequencing, which can reveal the characteristic mutation style in specific subtypes of GC, such as TP53 mutation in the CIN (chromosomal instability) group, PIK3CA alteration in EB virus, MSI (microsatellite instability), and GS (genomically stable) subtypes. Different types of ERBB2 alterations including amplification and Ser310Phe showed similar regulatory patterns involving the c-MYC-mediated genes CCND2, CDKN1A and THBS1 [44]. On the basis of biomarker detection, ex vivo targeted therapy tests were set up, showing that PDOs harbouring HER2 mutations responded to trastuzumab alone or with 5-fluorouracil. The divergent response to classic chemotherapeutics was investigated, and the IC₅₀ was compared with several cell lines showing a resistance tendency [45]. Therefore, PDO may be a powerful tool for the investigation of molecular pathogenesis and the discovery of biomarkers and targeted medicine.

GC organoid biobanks have been established and comprise an assortment of histological and molecular subtypes, and these normal and cancerous organoid lines recapitulate the morphological, histological, genetic, and transcriptomic characterization of corresponding tumour tissues [46]. Whole-exome sequencing in the GC PDOs revealed the well-documented driver mutations previously reported in GC, such as frequent alterations of CDH1 in diffuse type, TP53 in intestinal type and some other mutations involving RHOA, ERBB2, FGFR2, and MYC, and the similarities in the CIN and GS status to those previously reported in GC were also demonstrated. Organoid-based drug sensitivity ex vivo correlated well with clinical response. For instance, two patients who benefited from 5-fluorouracil and cisplatin after gastrectomy with this combined treatment had sensitive organoids, and another organoid derived from a patient with progressive disease showed no response to capecitabine. High-throughput drug screening was performed in PDOs from 7 patients, and the heterogeneity of agent response was assessed under the conditions of the same patient being given an array of drugs, the same drug being given to various individual patients or spatially different tumour regions from same patient being assessed [46]. Five characteristic organoids derived from gastroesophageal cancer patients in a gastrointestinal cancer cohort were established that captured the histological and genomic features of the parent tissues, such as the intestinal type, diffuse type, ERBB2 amplification type, and temporal intratumor heterogeneity from the same patient (from baseline and posttreatment) [29]. The drug sensitivity of organoids correlated well with clinical treatment response, and the transformation of PDXs from sensitive to resistant to paclitaxel was sequentially generated before and after treatment [29].

Currently, immunotherapy has become a major therapeutic option in the clinic for most cancers, including gastroesophageal cancers. PDOs are being developed to explore the potential mechanisms of immune therapy resistance/response in gastroesophageal cancers by using coculture or air-liquid interface (ALI) systems [47‐50]. A study presented a system for the coculture of mouse-derived gastric cancer organoids with immune cells, allowing the identification of a subgroup of gastric cancer patients who would potentially benefit from immunotherapy [49]. Chakrabarti et al. cocultured human gastric cancer organoids (huTGOs) generated from biopsied or resected tissues with cytotoxic T lymphocytes (CTLs) and myeloid-derived suppressor cells (MDSCs) and suggested that HER2-targeted therapy could inhibit CTL effector functions and PD-L1 expression [51]. In ALI system, tumor tissue containing stromal cells and immune cells are separated physically or enzymatically, following by seeded in the collagen gel in a upper surface which is exposed to air-conditions with a porous membrane underneath for nutrient diffusion occurring, so that oxygen can be transported in a more efficient manner [52]. By using ALI technology, Neal and colleagues successfully established co-culture PDOs containing immune cells or fibroblasts from 100 patients representing 28 different tumour types with the success rate of 73% after culture for one-month. These co-culture models maintained the diversity of T cell clones in patients for several weeks [50]. With the rapid development of organoid coculture technology, it provides an valuable platform for further research of personalized immunotherapy. However, the coculture PDOs still need to be more validated.

GC PDX models have been established so that the correlations compared to parent tumours, characterized by histology, genetics and clinical responses, can be evaluated (Table 5) [54‐57]. Wang et al. constructed 9 PDX models from 32 GC patients (28.1% success rate) harbouring molecular heterogeneity, including HER2 positivity, c-Met overexpression, and FGFR2 amplification, that responded to molecular targeted therapeutic agents [54]. Gastroscopic biopsies of GC patients were obtained to establish the PDX model, and the overall success rate was 34.1%, in which samples obtained before chemotherapy showed a higher transplantation rate. In addition to the concordance of histopathology and HER2 expression, chemosensitivity between parent tumour tissues and xenografts was investigated, revealing comparable therapeutic responses of corresponding regimens used in clinical treatment [55]. Within these cases, histological transformation from intestinal to diffuse type occurred in case 144, displaying no correlation between PDX-based drug sensitivity and clinically stable disease status [55]. Wang et al. developed mini PDX models for 4 GC patients to achieve personalized screening of chemotherapy or targeted therapy agents [58].

Notably, PDXs had become a successful tool for drug discovery in GC cancer. Ryan et al. constructed a comprehensive PDX collection of gastroesophageal cancer, including 46 (47%) GC adenocarcinomas, 25 gastroesophageal junction adenocarcinomas (26%), 21 oesophageal adenocarcinomas (32%), and three squamous cell carcinomas (3%), and then evaluated the antitumour activity of rational combination strategies [59]. Song et al. established patient-derived cell lines with peritoneal carcinomatosis, transformed them into orthotopic mouse models, identified major expression and activation traits, and then recapitulated the molecular and phenotypical features of donors [60]. Kuwata et al. successfully established 35 gastric cancer PDX models from 232 engrafted tissues and compared the clinicopathological factors associated with the establishment of PDX and CDX models [61]. Yagishita et al. built a large-scale Japanese patient-derived xenograft library (J-PDX) composed of 298 cross-cancer PDXs, in which 9 PDXs were gastric cancer, with a success rate of 16.7% (9/54) for engraftment [62]. Corso et al. established a comprehensive collection of gastric cancer preclinical models composed of 100 PDX and derivative cell lines or organoids, which included all the major gastric cancer histologic and molecular types identified by The Cancer Genome Atlas [63]. Chen et al. provided a 50-case PDX cohort of gastric cancer, characterized each of their individual histopathological and molecular features, and then evaluated anticancer agents targeting MET, EGFR, HER2 and CDKs in these models [64]. The broad application of these PDX accelerate the development of individualized combination therapies and guide the design of future clinical trials.

Precision medicine therapies generally require genomics based on targeted therapies and immunotherapies instead of “one-size-fits-all” chemotherapies [65]. Organoids recapture key characteristics of their corresponding normal or diseased organs and are amenable to current experimental technology. Inspired by cutting edge organoid technology, a deeper understanding of developmental biology and cancer biology could be achieved, as well as the filling of the gap between bench and bedside [19, 66‐69]. Great progress has been accomplished in the establishment of PDO models and their application as a predictive preclinical model to evaluate therapeutic responses in vitro, involving lung cancer [70‐72], colorectal cancer [29, 73‐75], breast cancer [76, 77], ovarian cancer [78, 79], pancreatic cancer [80, 81], liver cancer [82, 83], neuroendocrine neoplasms [53], oesophageal cancer and gastric cancer. However, the following practical challenges still need to be addressed to completely maximize the potential value of upper gastrointestinal PDOs in predicting clinical response:

As we all known, the PDX is widely used in basic research, drug development and clinical practice because it shows significant heterogeneity of primary tumour and is highly consistent with patient response to treatment. Nevertheless, there are still some limitations of PDX models in cancer research: