Background
Cervical cancer (CC) is a severe public health concern, particularly in developing countries. It is the fourth most common cancer in women [
1] and accounts for more than 300,000 deaths annually [
2]. Although early-stage CC is mostly curable, r/m CC is always accompanied by a poor prognosis due to the lack of effective treatment modality [
3,
4]. To develop and evaluate novel therapeutic strategy for advanced CC, preclinical models that are capable of accurately resembling PTs of patients should be developed and used to facilitate the application of therapeutic approaches into clinical practice [
5].
Presently, cell lines and cell line-derived xenograft (CDX) models represent the most common tools in preclinical research for cancer [
6]. However, individual cell line cannot reflect the heterogeneity of patient’s tumor. In addition, upon multiple passages, cell lines may undergo genetic drifts or mutations. These intrinsic defects result in the poor performance of the above-mentioned models in assessing patient’s response to a certain treatment [
7]. Organoid model, originating from primary tumor and requiring limited culture, could largely preserve the genetic features of its corresponding PT and can be used for high throughput analysis [
8]. However, the in vitro nature of this model prevents it from assessing the systemic safety profiles of candidate therapeutic modalities. Patient-derived xenograft (PDX) tumor, an in vivo model derived from individual patients, is considered an ideal tool for tumor research and has been used to explore tumor mechanisms and therapeutic modalities [
9]. There have been multiple PDX models established for CC. However, these investigations have several caveats including the insufficient scale of established PDX, and limited parameters used for comparative studies to evaluate the consistency between primary tumor and PDX tumor, hence limiting the application of PDX model in cervical cancer treatment research [
5,
10‐
13]. Additionally, key factors beyond patient clinical information [
11,
12], such as the differentially expressed genes (DEGs) of tumor or stroma and the tumor immune microenvironment (TME), should be evaluated for their influence on the growth of PDX tumors for guidance in subsequent PDX establishment. Therefore, there is an urgent need, at least in CC, for larger and more comprehensive PDX biobanks establishment.
The prevalence of HER2 gene amplification or mutation in CC patients ranges from 4.8 to 17% [
14,
15], and HER2 positivity is associated with a more advanced disease stage and worse prognosis in CC [
16]. Fortunately, neratinib, an irreversible pan-HER inhibitor, has been shown effective in treating advanced-stage CC with HER2 mutation [
17,
18]. However, its efficacy remains to be improved. Some indirect but compelling evidences suggest that ACT therapy with tumor-infiltrating lymphocytes (TILs) may have a synergistic effect with neratinib. First, anti-HER2-targeted therapy increased the levels of tumor-infiltrating lymphocytes (TILs) in solid tumors [
19]. Second, several clinical trials have demonstrated that the higher the level of TILs infiltrated in tumors, the greater the clinical benefits of anti-HER2 agents were shown [
20‐
22]. Third, ACT itself is a promising option for advanced-stage CC, which has been proved to be a promising therapy for CC with objective response rates (ORRs) of 44% and 50% in two clinical trials [
23,
24]. Therefore, it is of great significance to explore whether TILs could be combined with neratinib to improve the therapeutic efficacy for CC patients with HER2-mutation.
In the present study, we established by far the largest panel of PDX models with a high success rate. We found that xenograft engraftment and grafting speed are influenced by the primary tumor size, the presence of follicular helper T cells and the expression of cell adhesion-related genes in primary tumor tissue. The established PDX tumors retained the genetic and histopathological characteristics of the corresponding patient biopsy samples, even after serial passage or cryopreservation. Furthermore, our PDX models faithfully recapitulated patients’ response to conventional chemotherapy. Finally, using the PDX model, we found that combinatorial therapy with ACT and neratinib could effectively inhibit the growth of PDX tumors derived from CC patients with HER2-mutation.
Materials and methods
Patients and tissue samples
We enrolled 69 CC patients, including those with stage IB1 to IIIC2 disease, admitted in Tongji Hospital of Huazhong University of Science and Technology between June 2018 and January 2021. All patients provided written informed consent. Fresh tissue specimens, measuring 200–1000 mm3 in size, were collected from surgery or biopsy and transported immediately to the laboratory while submerged in tissue storage solution (MACS, 130-100-008) on ice. The tissue samples were selected for implantation in mice, ribonucleic acid (RNA) and deoxyribonucleic acid (DNA) extraction, and flow cytometry analysis based on specific criteria. The research protocol received ethical approval from the Ethical Committee of Tongji Medical College, Huazhong University of Science and Technology.
Animals
Female nonobese diabetic/severe combined immunodeficiency (NOD/SCID) NOD-SCID and NOD/SCID IL2Rγ−/− (NCG) mice, aged five to eight weeks and weighed 18–21 g, were purchased from Nanjing Biomedical Research Institute of Nanjing University (Nanjing, China). The experimental mice were housed in isolator cages maintained under specific-pathogen-free conditions, with precisely regulated temperature and humidity, and with a standardized 12 h light/dark cycle at Tongji Medical College. All animal care and experimental procedures strictly adhered to the guidelines for ethical review of animal welfare and were granted approval by the Institutional Animal Care and Use Committee at Huazhong University of Science and Technology.
Establishment of PDX models
The PT used for implantation measured 3–4 mm3 in size and was implanted subcutaneously into the flanks of NOD/SCID mice. When the tumors (P1 generation) reached 1500 mm3 in size, they were resected and introduced into the mice (P2 generation). The process was repeated until the P4 generation was generated. Of note, PT and PDX tumor are free of common human and animal pathogens as determined by PCR-based methodology (data not shown).
Establishment and culture of PDXO
PDXOs were established and cultured as previously described [
8]. Briefly, the cells were resuspended in basement membrane extracts (Corning, Corning, NY, USA) and cultured on a basal medium that containing growth factors such as Noggin, FGF7, B27, Y-27632, and EGF [
8].
Generation and rapid expansion of TILs
The surgical specimens were obtained and cut into 1–2 mm2-thick sections, which were then placed in a 24-well plate with 1 mL of culture medium consisting of 90% RPMI 1640 (Gibco) and 10% heat-inactivated human AB serum (HS). Recombinant human interleukin (IL)-2 (6000 IU/mL; GeneScript) was added, along with 1% penicillin and streptomycin. The TILs were expanded in a 75 cm2 cell culture flask using a standard rapid expansion protocol (REP) with irradiated (50 Gy) feeder cells (1 × 108); moreover, 40 mL culture medium containing 90% REP medium (X-Vivo, Lonza), 10% HS, 3000 IU/mL IL-2 (GenScript), and 25 mM β-mercaptoethanol (Sigma) were added as supplements, along with 30 ng/mL CD3 antibody (clone: OKT3, BioLegend) and TILs (1 × 106). On day 7 and every day thereafter, the cell densities were maintained at 1–2 × 106 cells/mL. On days 12–14, the cells were harvested and cryopreserved in liquid nitrogen. The TILs were then injected intravenously into the mice at a dose of 100 μl (10 × 106 TILs per mice).
Cell growth assay and organoid-TIL co-culture systems
Organoid growth was assessed utilizing the standard Cell Counting Kit-8 assay (Vazyme, Jiangsu, China) following the manufacturer’s protocol.
TILs were culture in X-Vivo medium, supplemented with 25 mM β-mercaptoethanol, 1:100 penicillin/streptomycin, and human AB serum (“TIL cell medium”). Prior to co-culture, cryopreserved REP TIL was thawed in pre-warmed (37 °C) TIL cell medium and incubate for 15 min at 37 °C with 60 IU/mL DNase I (Sangon Biotech). After that, resuspend cell at 2 × 10
6 per mL in TIL cell medium and add 150 IU/mL of IL2 and cultured overnight. Organoids are being cultured overnight with 200 ng/mL IFN-γ (PeproTech). 96-well U-bottom plates were coated with 5 μg/mL anti-CD28 (Biolegend) and kept overnight at 4 °C. On the coculture day, PDXO were dissociated to single cells with TrypLE Express and resuspended in TIL cell medium at 5 × 10
4 cell per mL. TIL was seeded at a density of 1 × 10
6 cells/well and stimulated with single cell organoids at a 20:1 TIL: tumor cell ratio, supplement with 150 IU/mL IL-2 and 20 μg/mL anti-PD-1. Plate 200 μL of dissociated organoid-TIL suspension per well and incubate at 4 °C for two days [
25].
CTL assay
The organoids were digested into single cell suspensions and stimulated for 24 h with 200 ng/mL of anti-interferon-α. After washing with phosphate-buffered saline (PBS), the cells were seeded at a density of 5 × 104 cells per well in 96-well round-bottom plates. The target cells were cultured in duplicate with effector Rep TIL cells in 200 μl TIL cell medium at 37 °C in a CO2 incubator for 4–6 h at the indicated an E/T ratio of 20:1, 10:1, and 5:1. Percent specific lysis was calculated using the following formula: 100 × [(experimental release − spontaneous release)/(maximum release − spontaneous release)], following the LDH Cytotoxicity Assay Kit instructions.
Mouse experiments
Under anesthesia, tumor fragments measuring 3–4 mm
3 in size were grafted into flanks of NCG mice. Treatment began after the tumor reached 70 mm
3 in size. The following drugs and dosage regimens were used. Neratinib (MedChem Express) 20 mg/kg was administered five times a week via oral gavage for 20 days. Autologous TILs were infused via tail vein injection with a dose of 10 × 10
6 cells. IL-2 was given at a dose of 45,000 IU for 3 days initially, followed by twice weekly dosing. Cisplatin 5 mg/kg diluted in water was administered intraperitoneally twice a week. The tumor size and mouse weight were assessed at the initiation of treatment, twice a week, and at the end of treatment. The volumes were determined using the following formula: 0.52 × length (L) × width (W)
2, where L represents the major tumor axis and W represents the minor tumor axis [
26].
DNA was extracted from frozen tumor and normal tissue samples of patients and matched PDX tumor samples using the DNeasy Blood and Tissue Kit (Qiagen, Germantown, ML, USA). The DNA quality was determined using Qubit 3.0, while the DNA samples were clustered using the Illumina PE Cluster Kit on the cBot Cluster Generation System (Illumina, San Diego, CA, USA). The whole exome sequencing (WES) libraries were prepared from 1 μg of genomic DNA using an Agilent liquid capture system (SureSelect Human All Exon V6; Agilent, Santa Clara, CA, USA) and sequenced with the Illumina HiSeq Instrument using 180–280-bp paired-end reads generated on the Novogene platform according to the manufacturer’s instructions.
Using the BMA program, the reference human + mouse (hg 19 + mm 10) genome was used to map the sequence reads from each model [
27]. The NovaSeq 6000 had a 20-fold mean read depth for PT samples that matched the PDX samples and a tenfold mean read depth for matching normal samples. Species disambiguation was performed using the software available at
https://github.com/disambiguate. VarScan v.2.3 was used to identify the single nucleotide variants (SNVs) [
28], and the results were annotated using ANNOVAR. The CNVkit (version 0.9.3) was utilized to detect the copy number variations (CNVs), while the segmented data were detected using Genomic Identification of Significant Targets in Cancer (GISTIC) 2.0 [
29]. The CNV discordance between samples was determined by assessing log
2 (CN ratio) CN gains and losses, and three samples were excluded from the analysis based on the criterion [
30].
RNA sequencing and differential gene analysis
RNA was extracted from the patient’s tumors using the RNA Easy Mini Kit (Qiagen, CA), following the manufacturer’s instructions. Sequencing was performed on Illumina NovaSeq 6000 with paired 150-bp reads. The StringTie software was employed to extract read counts and fragments per kilobase of transcript sequence per million base pairs sequenced. The edgeR package was used to standardize the RNA sequencing data and identify differentially expressed genes (DEGs). In our dataset, the genes with fewer than 1 count in less than 5 samples were excluded, and genes with at least two-fold change with a p-value of < 0.05 were considered to be DEGs. Gene set enrichment analysis was performed using the GSEA software (V4.1.0). Meanwhile, the ClusterProfiler package was used to perform Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis.
Immunohistochemistry
PDX and fresh tumor specimens were fixed in 4% formalin, embedded in paraffin, and stained with hematoxylin and eosin (H&E) or immunohistochemistry (IHC). The tissue was then dehydrated, and antigen retrieval with AR6 (sodium citrate) or AR9 (EDTA) was optimized. The slides were then observed and photographed under an Olympus BX53 light microscope. The epithelial and stromal components of the tumor were demarcated using the inForm software v2.6 (Akoya)®, and the IHC markers were quantified.
Multiplex IHC staining
Briefly, the slides were dewaxed and rehydrated, and antigen retrieval was performed. The endogenous peroxidase and Fc receptor were blocked as described in the IHC protocols. The samples were stained with rabbit monoclonal anti-CDKN2A (1:1,000. Abcam), rabbit monoclonal anti-CD4 (1:5, SP8, MXB), and rabbit monoclonal anti-CD8 (1:5, SP16, MXB). The Opal 7-Color Manual IHC Kit (Akoya Biosciences, DE, USA) was used to generate immunofluorescent signals, using TSA dye 520 (1:100, anti-CDKN2A), dye 570 (1:100, anti-hCD8), and dye 620 (1:100, anti-hCD4). Spectral 4,6-diamidino-2-phenylindole (DAPI) (1:10, Akoya) was used to counterstain the samples. The Vectra® 3.0 Automated Quantitative Pathology Imaging System was used for multi-slide imaging. The image data were extracted using the inForm® Software v2.6 (Akoya).
Flow cytometry
The tumors were harvested and minced into small pieces, and then digested with collagenase type IV (Sigma, 1 mg/mL) and DNase I (Invitrogen, 20 ug/mL) for 30 min at 37 °C, followed by filtering using a 70 μm cell strainer. The cells were washed with PBS and then suspended in fluorescent-activated cell sorting (FACS) staining buffer (1 PBS with 1% BSA and 0.1% sodium azide) for staining. The BD LSRFortessa instrument (BD Bioscience) was used for data acquisition, while FlowJo V10 (BD Bioscience) was used for data processing.
Statistical analysis
The data were presented as mean ± standard error, with a p-value of < 0.05 denoting a significant difference. The differences between groups were analyzed using one-way ANOVA or t-test. All statistical analyses were performed using Prism 8 and SPSS 23.0. Data visualization was performed in R version 4.1.1 using the ggplot2, ComplexHeatmap, gtrellis, and maftools packages.
Discussion
Appropriate preclinical models that faithfully reflect the characteristics of the primary tumor can facilitate the development and the efficacy evaluation of novel treatment strategies for solid tumor [
9]. In this study, we established, to our knowledge, the largest PDX biobank to date with a 63.8% success rate from 44 patients with different stages of cervical cancer.
PDX models should maintain the genomic features of their parental tumors to serve as a reliable preclinical model for therapeutic drug testing [
9,
36]. We compared the difference in genomic alterations between PDX tumors and their matched PTs through WES analyses. Our results indicated that the tumor purity was increased after being engrafted into the mice, which was similar to the findings of other studies [
42]. Our analysis showed a 18.79% CNV discordance ratio between PDX-P1 and PT, which is also comparable to the 10–20% ratio in other studies [
30]. We obtained a 2.1% CNV discordance ratio between early-generation PDX tumor and late generation tumor, which is, unsurprisingly, similar to the 3%–9% range reported in other studies [
30]. In a single-cell genomic analysis of breast cancer, minor subclones dominated the xenografts in consecutive passages, leading to the changes in mutation clusters [
43]. We also observed some mutations that were not recorded consecutively in the SNV data, which suggested that clonal dynamics might play an essential role in the tumor evolution of PDX models. Nevertheless, the concordance ratio of SNVs between our PT and PDX models was 70.3%. Overall, our data indicated that the established PDX models retained the genomic mutations and molecular characteristics of the native tumors across serial passaging.
The engraftment rate of PDX model varied widely according to cancer type and individual patient characteristics [
44]. In general, CC showed a higher tumor engraftment rate (> 60%) in immunodeficient mice compared with other solid tumor types such as breast cancer (12.5–31.3%) [
45,
46]. The success rate of CC PDX establishment is affected by various factors, such as the transplantation site and the time interval between surgery and xenograft implantation [
5]. To eliminate batch processing interference, the procedure was standardized to 6 h in order to eliminate the influence of the time interval between surgery and xenograft implantation. The transplant sites for CC PDX model were mainly subcutaneous, orthotopic, and sub-renal capsule. Although the sub-renal capsule and cervical orthotopic engraftment rates were higher (75 and 71.4%, respectively) than the subcutaneous engraftment rate (48–70%) [
31,
33,
47], we used subcutaneous implantation in the current study, which is preferred for easier monitoring of the tumor size, and the engraftment rate of 63.8% is sufficient for preclinical investigation.
Our study also investigated several clinical factors that may affect the success rate of PDX engraftment in CC, including tumor size, perineural invasion (PNI), lymphatic vascular space invasion (LVSI), LN metastases, parametrial invasion (PI), and tumor stage. Among them, only tumor size of > 4 cm was associated with the PDX engraftment rate (
p < 0.05), which was consistent with the report of previous studies on various cancer types [
11,
48‐
50]. Tumor size is an indicator of tumor burden according to the National Cancer Institute and a significant prognostic indicator of advanced-stage CC [
51]. Although other pathologic features such as PNI, LVSI, LN metastases, PI, and tumor stage seem to be related to a higher engraftment rate in our study, and these features have been previously associated with PDX formation and poor clinical outcome in other cancer types [
50,
52], we did not observe a significant difference in our study (
p > 0.05). Hence, further studies with a larger sample size and more comprehensive analysis are warranted to confirm these findings.
We also compared the differences in the TME and genomic expression between engrafters and non-engrafters. To compare immune cell infiltration, we utilized flow cytometry to identify immune cell subsets that may correlate with engraftment; however, no significant difference was identified. Using the CIBERSORT algorithm, we observed an increase in the subset of follicular helper T cells, which correlated with engraftment. This finding is in agreement with a previous study, which demonstrated the association of Tfh-like cells with unfavorable outcomes, such as lymphatic metastasis or distant metastasis [
53‐
56]. Furthermore, we demonstrated that activated NK cell and resting CD4
+ memory cells were unregulated in the non-engrafters group. Consistent with our results, previous studies also revealed that memory T cell and NK cell correlated with better outcomes in other cancer types [
57,
58]. With regard to the correlation between tumor molecular features and tumorigenicity, the engrafters group exhibited an enrichment in molecular adhesion-related pathways, which is consistent with the report of previous studies [
59]. Moreover, engrafters group was also enriched in the DNA replication and cell cycle pathways in other study [
59]. Together, our results suggest that intrinsic factors of primary tumors, such as immune cell subsets and differentially expressed genes, may be useful in predicting the establishment of PDX models.
Previous retrospective studies have suggested that PDX models can be used to predict drug response and outcomes of patients in various cancer types [
9,
59,
60]. Our study demonstrated that the PDX models of CC are a promising tool for conventional drug screening and could provide insight into the development of novel therapy. HER2 mutation is a poor prognostic factor of advanced-stage CC [
18]. The therapeutic activity of neratinib, a HER2 inhibitor, has been shown in patients with HER2 mutations who were unresponsive to platinum-based chemotherapy (ORR: 25%; 95% confidence interval: 5.5–57.2%) [
18]. A previous clinical trial demonstrated that tyrosine kinase inhibitors (TKIs) increased the infiltration of T cells in tumor tissue [
19]. In the NeoALTTO trial, higher levels of TILs were associated with improved outcomes in HER2-positive patients treated with lapatinib and/or trastuzumab [
20]. Despite the synergistic effect of TILs and HER2 inhibitors suggested by aforementioned investigations, combinatorial strategy using these two treatments has never been explored in CC. Taking advantage of our PDX model, we were able to show that the novel TIL/neratinib combinatorial strategy, featured by an impressive elevation in the infiltration of GrzB
+ T cells in PDX tumor, may be a promising therapeutic alternative for patients with HER2 mutation. Our results demonstrated that CD4
+ T cells were dominant in both transfused TIL cell products and PDX tumors that were effectively controlled by combinative therapy. These results suggest that CD4
+ T cells may be an important anti-tumor cell population in cervical cancer immunotherapy, and it is likely to play a synergistic effect with neratinib against cervical cancer. Previous preclinical studies have supported that CD4
+ CAR-T cells in maintaining antitumor responses [
61]. Although the exact mechanism of action of CD4
+ T cells remains unclear, their significant antitumor potential can provide new insights into the development of effective tumor immunotherapy [
62,
63].
Based on our studies, the PDX tumor for CC could potentially be used to evaluate the effects of various treatment strategies, such as the efficacy of a variety of chemotherapy drugs, the efficacy of chemotherapy drug combination strategies, the screening of second-line drugs, the efficacy of immunotherapy, etc., which will assist in the selection and evaluation of clinical treatment strategies. More importantly, after classifying cervical cancer patients with different molecular subtypes, the PDX tumor model can be used to explore the best treatment strategy for patients with a specific subtype of cervical cancer, thereby providing more rapid and reliable treatment for cervical cancer patients with the same subtype characteristics. Furthermore, we are using the cervical cancer PDX models to further explore the tumor growth characteristics of cervical cancer patients with different molecular subtypes, and the possible intervention strategies and their efficacy.
However, there are several caveats in PDX models, including ours. First, the stromal content is decreased in PDX tumors compared to their corresponding PTs (Fig.
2c). Previous investigations also indicated that human stromal cells are replaced by mouse counterpart over serial passages. Thus, this model is not fit for studying the crosstalk between human cancer cells and stromal cells, and is incapable of recapitulating the contribution of stromal cells to drug sensitivity/resistance. Second, upon implantation into immunocompromised mice and over consecutive passages, human immune cells are gradually lost due to the lack of human cytokines to support their survival (data not shown). Therefore, PDX model is intrinsically unable to assess the impact of patients’ immune system when used for preclinical evaluation. Nevertheless, humanized mice that harbor reconstituted human immune system could overcome this drawback to some extent. In addition, this model is particularly useful for assessing the therapeutic efficacy of adoptively transferred immune cells including CAR-T, TCR-T and TILs. Indeed, our PDX model is a good fit for evaluating the efficacy of TIL-based monotherapy and combinatorial therapy, as demonstrated by our current study. TIL-based immunotherapy is a highly personalized treatment modality, for which the expansion of autologous TIL requires 4–8 weeks and is therefore time-consuming. However, this time interval can be fully utilized to establish PDX tumors, which could be ready at the time of TIL product collection. In fact, we are designing a co-clinical trial for TIL/neratinib combinatorial therapy in r/m CC, which is similar to our current strategy but in a prospective setting. In this scenario, our PDX model will act as an “avatar” that is able to timely guide clinical decision-making in selecting appropriate therapeutic strategies, i.e., monotherapy with either TIL or neratinib, or TIL/neratinib combinatorial therapy. Based on our current study, this co-clinical trial is promising and will provide insight into future development of novel therapeutic strategies for r/m CC.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.