Introduction
Gastric cancer is the third-leading cause of cancer-related deaths worldwide and was responsible for an estimated 769,000 deaths in 2020 [
1]. The overall 5-year survival rate for gastric cancer patients is only 32% in the United States, and novel therapies are urgently needed [
2]. However, efforts to develop effective anticancer therapies have been impeded by a lack of preclinical models that recapitulate the heterogeneity of gastric tumors. One strategy to overcome this limitation is the generation of patient-derived xenografts (PDX), in which human tumors are propagated in immunodeficient mice. PDXs are a completely in vivo preclinical platform that bypasses growing the tumor cells in in vitro environments under highly artificial conditions. As a result, PDXs retain a high degree of the histologic, genomic, and transcriptomic characteristics of the human tumors from which they are derived [
3‐
5]. PDXs also exhibit superior predictive values for clinical outcomes than traditional in vitro cell culture and cell line-derived xenografts [
6]. While gastric cancer PDX models have been established by East Asian and European institutions, few PDXs have been generated from Black/African American or Hispanic patients [
4,
7,
8]. This is a significant deficiency because gastric cancer biology, and response to novel therapies, has been shown to vary by patient race and ethnicity [
9‐
14].
The strain of recipient mice that should be used to establish gastric cancer PDX has not been determined. Commonly used strains include NOD.Cg-
Prkdcscid Il2rgtm1Wjl/SzJ mice (NSG) and
Foxn1nu (nude) mice [
5,
15]. NSG mice are highly immunodeficient and lack mature T cells, B cells, and natural killer cells and have defective macrophages and dendritic cells, whereas nude mice only lack mature T cells [
16]. In theory, highly immunosuppressed strains may allow for higher tumor engraftment rates. However, the use of immunodeficient mice to generate solid tumor PDXs is hampered by the growth of B cell lymphoma, the frequency of which may be affected by the degree of immunodeficiency of the recipient mice [
4,
5].
Lenvatinib is a multitargeted tyrosine kinase inhibitor and inhibits vascular endothelial growth factor receptors 1–3 (VEGFR 1–3), fibroblast growth factor receptors 1–4 (FGFR 1–4), platelet-derived growth factor receptors α and β (PDGFR α-β), RET (rearranged during transfection), and c-kit. However, it most potently affects the VEGFR family, suggesting a dominant anti-angiogenic mechanism [
17,
18]. Anti-VEGFR therapy does have a role in gastric cancer treatment; ramicirumab, which is an anti-VEGFR2 monoclonal antibody, is approved for second-line therapy in advanced gastric cancer [
19]. Recently, lenvatinib showed promising efficacy when given in combination with pembrolizumab, an immune checkpoint inhibitor, for the treatment of patients with advanced gastric cancer in a phase II clinical trial in Japan [
20]. However, lenvatinib has not been well studied in the gastric cancer context as a monotherapy [
21]. Thus, additional preclinical and clinical data are needed to understand the effects of lenvatinib monotherapy in a diverse gastric cancer patient population.
In this study, we generated PDXs from Western gastric cancer patients of various racial and ethnic backgrounds. We used NSG or nude mice as recipients to perform a head-to-head comparison of their rates of engraftment and lymphomatous growth. Finally, we used our PDX model to evaluate the efficacy of lenvatinib.
Materials and methods
Sample acquisition
This study was approved by the University of Texas Southwestern Medical Center institutional review board. All patients with gastric adenocarcinoma who were treated at Parkland Memorial Hospital (Dallas, TX) and William P. Clements Jr. University Hospital (Dallas, TX) were approached for the study. All enrolled patients provided written informed consent.
Tumor tissue was obtained from patients by endoscopic biopsy, gastric resection, and biopsy of metastatic sites. Samples for PDX generation were immediately placed into sterile phosphate buffered saline and transported on ice to the mouse procedure facility for implantation. Samples for transcriptomic analysis were immediately placed in RNAlater (Invitrogen) for at least 24 h at 4 °C and then transferred to − 80 °C.
Mice
All mice were handled in accordance with the guidelines of the institutional animal care and use committee at the University of Texas Southwestern Medical Center. NSG and nude mice were obtained from Jackson Laboratory.
PDX generation, passage, storage, and reanimation
Samples obtained by esophagogastroduodenoscopy (EGD) biopsy (each 1–3 mm3), biopsy of metastatic sites (1–3 mm3), and surgical resection (3–5 mm3) were coated in Matrigel Matrix (Corning Life Sciences) and implanted into the subcutaneous space in the flanks of recipient mice. Tumors were passaged when they reached a maximum diameter of 1.5 cm or skin necrosis overlying the tumor was observed. We considered a PDX line to be established if the tumor grew to a diameter of at least 1.5 cm and grew after passage to a second recipient mouse. At passage, the tumor was excised from the subcutaneous tissue, divided, and then: (1) passaged into the flanks of a new mouse, (2) fixed in 4% paraformaldehyde for histologic evaluation, and (3) stored in 10% dimethyl sulfoxide (DMSO) + 90% fetal bovine serum and submerged in liquid nitrogen for future use. To reanimate frozen PDX tissue, the tissue was warmed in a 37 °C water bath for 60 s and then implanted in an identical manner as a fresh sample. After reanimation, mice were maintained the same way as mice with primarily implanted xenografts.
Lenvatinib testing
Four PDX lines were selected for drug testing. When tumors reached a volume of 300–500 mm
3, mice were randomized to receive either the drug delivery vehicle (DMSO mixed with corn oil) or lenvatinib (MedChemExpress; cat #HY-10981) 10 mg/kg daily by oral gavage. Tumor dimensions were measured at baseline and once weekly thereafter by a blinded observer. After 21 days, all mice were euthanized. Tumor volume was calculated using the formula (π/6) × length × width
2.[
22] The percent tumor volume change (%Δv
tumor) from baseline of all tumors was calculated. Mice were weighed weekly.
Histology, immunohistochemistry, and immunofluorescence
Tissue samples were fixed in 4% paraformaldehyde for 24 h and then embedded in paraffin. Sections were cut to 4 μm. All samples were stained with hematoxylin and eosin (H&E) for histologic examination. Tumor and PDX histology were evaluated in a blinded fashion by a board-certified pathologist with expertise in gastrointestinal malignancies (S.T.G.H.).
Antibodies used for immunohistochemistry (IHC) were mouse anti-human pan-cytokeratin AE1/AE3 monoclonal antibody (1:500 dilution; Santa Cruz Biotechnology; SC-81714), mouse anti-human CD20 monoclonal antibody (Agilent; IR60461-2), rabbit anti-human CD31 monoclonal antibody (1:50 dilution; Abcam; ab28364), rabbit anti-human Ki-67 antibody (1:1000 dilution; Abcam, ab15580). Terminal deoxynucleotidyl transfer dUTP nick end label (TUNEL) immunofluorescence (IF) was performed using the In Situ Cell Death Detection Kit, AP (Roche). Hoechst 33342, Trihydrochloride, Trihydrate (Invitrogen) was used for nuclear counterstaining.
Quantification of intratumoral vascular density, tumor cell proliferation, and apoptosis
Blood vessel density was quantified using the hot spot method [
23]. In brief, each slide was scanned under light microscopy at 40× magnification to identify the three areas of highest vascular density. Those three areas were then evaluated at 200× magnification and the number of vessels per 200× magnification field were counted. Any stained ring of endothelial cells surrounding a lumen or discrete cluster of endothelial cells was counted as a single vessel. Two tumors from each PDX line were evaluated. Ki-67-positive cells were manually counted in three 400× fields per tumor. The Ki-67 of two tumors per treatment arm was calculated, and their average value was reported. To quantify TUNEL-positive cells, each slide was scanned under fluorescent microscopy at 40× magnification to identify the three areas of highest TUNEL-positivity. Those three areas were then evaluated at 200× magnification and the number of apoptotic cells per 200× magnification field were manually counted. The TUNEL-positivity rate of two tumors per treatment arm was calculated, and their average value was reported. All quantification was performed in a blinded fashion.
RNA sequencing and analysis
RNA was extracted using the RNeasy Mini Kit (Qiagen). Total RNA was quantified using the Ribogreen method using Victor X2 fluorometry (Life Technologies). RNA integrity was assessed using an Agilent Technologies 2100 TapeStation. Libraries were made with TruSeq Stranded Total RNA kits (Illumina) and sequenced on a NovaSeq 6000 with 2 × 150 base pair paired-end reads.
RNA-sequencing data was refined by Fast p v0.20.1 (default settings) such that Illumina universal adapters were trimmed and reads with lower quality bases, or noninformative reads, were filtered out. A total of 51.3 million paired-end reads were retained. All samples engrafted into mice were aligned into both human (hg19) and mouse reference assembly (m38) using STAR (v2.7.9, —twopassMode Basic). Each aligned read was examined to determine the organism of origin. Reads uniquely aligned to the mouse genome (i.e., with a higher matching score than the human genome) were removed. Then, the refined human reads were converted to transcriptome abundant matrix in the format of transcripts per kilobase million (TPM) via Salmon v1.5.2 with input parameters: ‘-I IU—validateMappings’). The Salmon reference genome sequence index was built on human reference transcript sequences (GRCm38.v29) with the gene model Gencode v29 (Ensembl 94).
Statistical analysis
Categorical variables were presented as counts and compared with chi-square or Fisher’s exact tests. The %Δvtumor between vehicle and lenvatinib-treated tumors was compared with Mann–Whitney U tests, and the same tests were used for comparisons between vehicle and lenvatinib-treated vessel counts. A p value of < 0.05 was considered statistically significant. Statistical computation was performed using GraphPad Prism (Version 9.1.0).
Discussion
In this study, we established 23 gastric cancer PDX lines from patients of diverse racial and ethnic backgrounds. PDXs are unique among preclinical models in that they recapitulate the native tumor biology of the patients from whom they are derived. However, to date, gastric cancer PDXs have primarily been generated from European and East Asian patients [
4,
7,
8]. The lack of diversity in previously generated gastric cancer PDXs is a major shortcoming for the field as there is significant biological heterogeneity associated with patient race and ethnicity [
9‐
14]. For example, Strong et al. [
9] found that patients in Korea had improved gastric cancer-specific survival when compared to patients in the US, even after controlling for other prognostic variables. Additionally, differing races and ethnicities have exhibited divergent responses to therapy; for example, in a phase II study of regorafenib, Asian patients were found to have a more favorable response than patients of other races and ethnicities [
11]. To our knowledge, this is the first description of PDXs generated from a large number of Black/African American and Hispanic patients.
The preferred recipient mouse strain for PDX generation has not been established [
4,
5]. Theoretically, a highly immunodeficient mouse strain, such as the NSG line that lacks mature T cells, B cells, and natural killer cells, may allow for a higher tumor engraftment rate, but at a cost of increased lymphomatous growth as compared to a modestly immunodeficient strain, such as nude mice, which lack only mature T cells [
16]. To answer this question, we performed a head-to-head comparison of NSG and nude mice as recipients. We found that while engraftment rates were similar between the two strains, there was a lower rate of lymphoma formation in nude mice. The unregulated proliferation of xenografted patient B cells leads to lymphoma formation in immunodeficient mice. Our findings demonstrate that the presence of some functional elements of the immune system in nude mice is sufficient to suppress lymphoma overgrowth without affecting the rate of gastric adenocarcinoma engraftment. Moreover, we found that PDXs engrafted into NSG and nude mice similarly retained the histologic and transcriptomic characteristics of the human primary tumor from which they were derived. While administration of the anti-CD20 antibody rituximab has been suggested as a method to reduce lymphomatous transformation in gastric cancer PDXs, we have not found this to be necessary due to the success of our nude mouse protocol [
15,
24]. Thus, our results demonstrate that nude mice are a preferred platform over NSG mice for gastric cancer PDX generation. Our findings are consistent with the report by Choi et al. who also concluded that nude mice are the preferred recipient for gastric cancer PDX generation [
5]. This has significant logistical implications as nude mice are less expensive than NSG mice and may have fewer housing and handling restrictions.
The efficacy of novel therapies in the preclinical setting is infrequently replicated in clinical trials [
25]. This may be explained, in part, by the degree to which preclinical research relies on cancer cell lines that are established as a monolayer growing in a nutrient-rich environment. In contrast, PDXs are unmodified by in vitro adaptation and have been shown to maintain a high degree of the histologic, genomic, and transcriptomic features of the parent tumors from which they were derived [
5,
26]. As a result, PDXs are a demonstrably better platform for preclinical drug testing [
27].
We found that lenvatinib inhibited the growth of all four PDX lines tested, regardless of patient race/ethnicity, Lauren subtype, histologic differentiation, anatomic tumor location, and history of prior chemotherapy treatment. This suggests that lenvatinib monotherapy may be effective against a broad range of gastric cancers, although there is a paucity of data in gastric cancer patients. Lenvatinib has only been tested as a monotherapy in 6 advanced gastric cancer patients in a phase I study in which 3 of the 6 patients showed stable disease [
21]. In the KEYNOTE-061 trial, the objective response rate of gastric cancer patients treated with pembrolizumab was approximately 15% in patients with a combined positive score (CPS) of ≥ 1 [
28]. In contrast, Kawazoe et al.’s findings from a phase II trial that tested a combination of lenvatinib plus pembrolizumab showed an objective response in 69% of patients, despite only 66% of the patient cohort having a CPS of ≥ 1 [
20]. As such, we reasoned that in the Kawazoe et al. cohort, either: (1) lenvatinib was the primary driver of the increased response rates observed, or (2) lenvatinib enhanced the efficacy of pembrolizumab. Because we found that lenvatinib was effective in the absence of concurrent immune checkpoint inhibition, our data suggest that lenvatinib may, at least in part, be an independent driver of the results observed in the Kawazoe et al. study. Another notable finding was the efficacy of lenvatinib in all three treatment-naïve PDX lines. This is significant because ramucirumab, an anti-angiogenic monoclonal antibody that blocks VEGFR-2, has demonstrated efficacy only in a second-line setting [
29]. The broader range of lenvatinib’s anti-angiogenic inhibition via blockade of VEGFR1-3 may be more effective for the treatment of gastric cancer than selective VEGFR-2 inhibition that ramucirumab provides.
There are several limitations to our study. There is interest in using PDXs as a “patient avatar” to test therapies and identify the treatment options on an individualized basis. However, we were only able to generate PDXs from approximately 1 in 6 patients and our median time to engraftment was 13 weeks. Therefore, while this personalized approach is attractive in principle, it may not be feasible at most centers. Additionally, PDX-bearing mice are immunocompromised; as such, tumor-immune interactions and immune-oncologic therapies cannot be easily evaluated in PDXs. Finally, the influence of the tumor microenvironment in the subcutaneous space of the mouse flank is different than influences from gastric tissue and the unique environment of the stomach lumen. Despite these limitations, we believe that PDXs are a robust model for generating high-quality preclinical data for novel gastric cancer therapies.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.