Establishment and genetically characterization of patient-derived xenograft models of cervical cancer

Patient-derived xenograft (PDX) models were established to reproduce the clinical situation of original cancers and have increasingly been applied to preclinical cancer research. Our study was designed to establish and genetically characterize cervical cancer PDX models. A total of 91 fresh fragments obtained from 22 surgically resected cervical cancer tissues were subcutaneously engrafted into female NOD-SCID mice. Hematoxylin and eosin (H&E) staining was performed to assess whether the established PDX models conserved the histological features of original patient cervical cancer tissues. Moreover, a Venn diagram was applied to display the overlap of all mutations detected in whole-genome sequencing (WGS) data from patient original cervical cancer (F0) and F2-, F3-PDX models. The whole exome sequencing (WES) and the “maftools” package were applied to determine the somatic mutations among primary cervical cancers and the established PDX models. Our study successfully developed a panel of cervical cancer PDX models and the latency time of cervical cancer PDX model establishment was variable with a progressive decrease as the passage number increased, with a mean time to initial growth of 94.71 days in F1 engraftment to 40.65 days in F3 engraftment. Moreover, the cervical cancer PDX models preserved the histological features of their original cervical cancer. WGS revealed that the genome of original cervical cancer was preserved with high fidelity in cervical cancer PDX models throughout the xenografting and passaging process. Furthermore, WES demonstrated that the cervical cancer PDX models maintained the majority somatic mutations of original cervical cancer, of which the KMT2D, LRP1B, NAV3, TP53, FAT1, MKI67 and PKHD1L1 genes were identified as the most frequently mutated genes. The cervical cancer PDX models preserved the histologic and genetic characteristics of their original cervical cancer, which helped to gain a deeper insight into the genetic alterations and lay a foundation for further investigation of the molecular targeted therapy of cervical cancer.

papillomavirus [2]. The therapeutic options for recurrent/metastatic cervical cancer are still limited, with the main treatment strategy of combining chemotherapy with targeted therapy [3]. Bevacizumab is the only agent approved for targeted therapy of cervical cancer, whose potential clinical benefits have not been sufficiently confirmed, primarily because the patients are not preselected for molecular alterations in the targets of interest [4,5].
Conventional wisdom holds that cancer initiation and progression is attributed to the evolutionary process of genomic rearrangements and somatic mutations [6]. Moreover, genes with high somatic mutation rates are generally considered potential driver genes of cancer [7]. Performing next generation sequencing (NGS) for somatic mutation analysis is prevalent for the identification of critical molecular events involved in tumorigenesis and cancer development [8,9]. Cervical cancer often harbors activating somatic mutations, including E74 Like ETS Transcription Factor 3 (ELF3) and Core-Binding Factor Subunit Beta (CBFB) genes, and inactivating somatic mutations in the F-Box And WD Repeat Domain Containing 7 (FBXW7), Major Histocompatibility Complex, Class I, B (HLA-B) and E1A Binding Protein P300 (EP300) genes, which may drive oncogenic transformation and chemotherapy resistance of cervical cancer [10]. Therefore, it is of great importance to determine the specific somatic mutations of cervical cancer, which is helpful for the development of precise therapies and targeted therapies for cervical cancer patients.
Patient-derived xenograft (PDX) models are generally constructed by implanting fresh surgical tumor tissues into immune-deficient mice [11], which can be performed orthotopically or heterotopically [12]. Moreover, PDX models maintain the principal molecular characterizations, histological architecture and genomic features of the original cancers [13]. Therefore, in the recent decade, PDX models have been established to reproduce the clinical situation of original cancers [14], and have been increasingly applied in preclinical cancer research [15].
The present study was designed to establish and histologically characterize patient-derived xenograft models of cervical cancer. Moreover, this study implemented Whole-Genome Sequencing (WGS) and Whole Exome Sequencing (WES) to evaluate the genomic features and somatic mutations between cervical cancer PDX models and patients' original cervical cancer tissues.

Patient specimens
Surgically resected cervical cancer tissues and matched adjacent normal tissues were obtained from cervical cancer patients (n = 22) with written informed consents, who underwent radical hysterectomy at the Second Affiliated Hospital of Wenzhou Medical University. The size of each tissue sample measured approximately 1 × 1 × 1 cm. The tissue samples were preserved in ice-cold phosphatebuffered saline (PBS), and transported to the laboratory for processing within 2 h. Furthermore, after removal of necrotic tissues, the resected fresh cervical cancer tissues were sectioned into three sections: one section was prepared for cancer transplantation to establish the PDX models; one section was fixed in 4% paraformaldehyde, and then embedded in paraffin and sectioned for Hematoxylin-eosin (H&E) staining; and the remaining section was flash-frozen with liquid nitrogen for long-term storage at − 80 °C for the further study. The research protocol was ethically reviewed and subsequently approved by the ethics committee of the Second Affiliated Hospital of Wenzhou Medical University.

Animals
Female NOD-SCID mice, aged four to six weeks, weighed 18-20 g, were purchased from Beijing Vital River Laboratory Animal Technology (Beijing, China). The NOD-SCID mice were housed and maintained in isolator cages under a specific-pathogen-free (SPF) environment with a controlled humidity and temperature, on a standard 12 h light/dark cycle, at Wenzhou Medical University. All animal care and experimental procedures complied with the guidelines for ethical review of animal welfare and were approved by the Institutional Animal Care and Use Committee of Wenzhou Medical University.

Establishment of cervical cancer PDX models
The cervical cancer tissues from surgery patients (termed as F0) were washed with PBS and subsequently gently cut with sharp, sterile scissors into small pieces of approximately 3 mm 3 . Three to five fragments of minced cervical cancer tissues were aspirated into a 5 mm-trocar sheath individually, which was employed to perform the blunt separation of connective tissue following the cutaneous incision operated in the right buttock of NOD-SCID mice. By pushing the inner core needle of the trocar sheath, the cervical cancer tissues were subcutaneously engrafted into the right buttock of NOD-SCID mice, and the cutaneous incision was sutured with stitches. This generation of NOD-SCID mice harboring patient-derived cervical cancer tissues was termed as F1.
After tumor engraftment, the NOD-SCID mice were monitored for a maximum period of 6 months to confirm whether PDX models were successfully established. If the tumor nodules were not palpable within 6 months, they were considered as engraftment failure. The time interval from engraftment to the presence of palpable tumor was defined as the latency time [16]. By measuring the length and width of the subcutaneous tumor with a calliper weekly, its volume was calculated by the following formula: Volume = 0.5 × (width 2 × length) [17]. When the volume of the tumor was approximately 1000 mm 3 , the xenograft mice were euthanized by cervical dislocation, and the tumor was excised and sectioned into three sections as the procedure followed for the surgically resected human cervical cancer tissues. Parts of the xenograft tumor were continuously engrafted into new female NOD-SCID mice. Furthermore, the subsequent serially-transplanted xenografts were obtained in a manner similar to that described above, and the tumor engrafted NOD-SCID mice were sequentially termed as F2, F3…Fn respectively.

Histological analysis
The patient original cervical cancer tissues (F0), F2-and F3-PDX tissues were fixed in 4% paraformaldehyde for 24 h and embedded in paraffin. Tissue slices were sectioned at a thickness of 4 μm and subsequently were stained with hematoxylin and eosin (H&E), which were imaged on an optical microscope. Moreover, the histological comparison among patient original cervical cancer tissues (F0), F2-and F3-PDX tissues was conducted by a senior pathologist based on overall tumor cytoarchitectural features.

DNA extraction from tissues
The DNeasy Blood and Tissue kit (Qiagen, Hilden, Germany) was applied to extract the DNA from PDX tissues and patient original cervical cancer tissues, which was performed as described in the manufacturer's instructions.

Whole-genome sequencing (WGS)
A single patient original cervical cancer (F0, patient id = No.3) and one each of its corresponding cervical cancer PDX models (F2 and F3) were subjected to Whole-Genome Sequencing (WGS) to evaluate their genomic features. The library was constructed by randomly fragmenting the genomic DNA into 350 bp segments. The DNA libraries were high-throughput sequenced using the Illumina HiSeq 4000 platform for paired-end reads of 150 bp. WGS was applied in F0, F2 and F3 at mean coverages of 24.37X, 16.61X and 40.59X, respectively. The analysis of raw-read sequencing data was carried out by taking the human genome build hg19 as the reference genome. Pre-trimming of the DNA sequence was performed by Fastp 0.20.1, and the quality control assessment was carried out with FastQC v0.11.9. The sequence reads were aligned to the human genome GRCh37/hg19 with BWA (version 0.7.17). GATK 3.8-1-0 was conducted for variant calling, and annotation.

Whole-exome sequencing (WES)
A single patient original cervical cancer (F0, patient id = No. 19) and one each of its corresponding cervical cancer PDX models (F2 and F3), and the adjacent noncancerous tissue were subjected to Whole-Exome Sequencing (WES) to identify the somatic mutations among them. For library preparation, the qualified genomic DNA was randomly broken into 180-280 bp fragments. Exonic DNA capture was conducted with Agilent SureSelect Human All ExonV6 (Agilent Technologies) following the manufacturer's instructions and was subsequently sequenced using the Illumina NovaSeq 6000 platform for paired-end reads of 150 bp. WES was performed on F0, F2, F3 and adjacent noncancerous tissues at mean coverages of 74.41X, 63.90X, 69.87X and 90.16X, respectively. The analysis of rawread sequencing data was carried out by taking the human genome build hg19 as reference genome. Pretrimming of DNA sequence was performed by Fastp 0.20.1, and the quality control assessment was carried out with FastQC v0.11.9. The sequence reads were aligned to the human genome GRCh37/hg19 with BWA (version 0.7.17). GATK 3.8-1-0 was conducted for variant calling, and annotation.

Somatic mutation analysis
The somatic mutation data were annotated and converted to the Mutation Annotation Format (MAF), and the "maftools" R package was applied to screen the differentially mutated genes in F0, F2 and F3 -PDX tissues [18]. And we have removed outliers by setting rmOutlier = T option when using plotmafSummary function. Moreover, TCGA somatic exomic mutations of 289 cervical cancer samples were acquired from The Cancer Genome Atlas (TCGA) portal (https:// portal. gdc. cancer. gov/) [19].

Data availability
The raw data of WGS and WES were submitted to the Baidu Netdisk (https:// pan. baidu. com/s/ 1X7KK nt8he 2GXiM 1X0FA Dhg), with the dataset code: h69i.

Statistical analysis
SPSS Statistics Version 25 (IBM Corp., Armonk, NY, USA) was applied for the statistical analyses. The normality of variables was assessed by the Shapiro-Wilk normality test. Normally and non-normally distributed variables were described as mean ± SD and median (P25, P75), respectively. Qualitative variables were described as numbers and percentages. Non-normally distributed data were analyzed with the nonparametric Mann-Whitney U-test. The rates were compared using the Chi-square test. P < 0.05 was considered statistically significant.

Establishment and validation of cervical cancer PDX models
Surgically resected primary cervical cancer tissues were collected from twenty-two patients to establish cervical cancer PDX models, and the tissues were histopathologically confirmed as cervical squamous cancer. In more detail, fragments of twenty-two primary cervical cancer tissues were engrafted into ninety-one female NOD-SCID mice aged six to eight weeks (Additional file 1: Table S1), among which twenty-eight F1 mice formed tumors. Therefore, the F1-PDX engraftment success rate was 30.77%, with a median latency time being 87.00 (77.25,104.00) days. Six of twenty-eight F1-PDX died in the process of tumor growth, with a mortality of 21.43%. Moreover, the specific information of F2-and F3-PDX were shown in the Table 1. The differences among latency time to establish F1-PDX, F2-PDX and F3-PDX were statistically significant. Furthermore, statistically significant differences were found among success rates of developing F1-PDX, F2-PDX and F3-PDX. The mortality between F1-and F2-PDX, F1-and F3-PDX, F2-and F3-PDX showed no statistical difference. Unfortunately, cryopreserved F0 cervical cancer tissues were implanted into NOD-SCID mice, showing a 0% engraftment success rate.
Histological analysis was performed to assess whether the established PDX conserved the histological features of patient original cervical cancer tissues. H&E staining of cancer tissues from F2-and F3-PDX illustrated that the cytoarchitectural characteristics were similar to those of the patient original cervical cancer tissues morphologically, with the representative micrographs showing in Fig. 1.

Genomic features retained during PDX tumor passages
The whole-genome sequencing (WGS) was performed to evaluate the genome-wide mutations between a single patient original cervical cancer (F0, patient id = No.3) and one each of its corresponding cervical cancer PDX models (F2 and F3). A total of 9222 overlapping mutations were found among patient original cervical cancer (F0) and F2-, F3-PDX, of which including the Zinc-Finger Protein family genes, such as ZNF275 and ZNF280A. Moreover, 9226 and 9225 mutations in original cervical cancer were also found in the corresponding F2-PDX and F3-PDX respectively, which accounted for 99.79% and  Therefore, the cervical cancer PDX models maintained the genetic characteristics and retained most of the mutations of original cervical cancer throughout the xenografting and passaging process, and the detailed results were presented in Fig. 2.

Somatic mutational status of cervical cancer and PDX models
The whole-exome sequencing (WES) was performed to identify somatic mutations in one single patient original cervical cancer (F0, patient id = No. 19) and one each of its corresponding cervical cancer PDX models (F2 and F3). When compared with the somatic mutations in the adjacent noncancerous tissues, the original cervical cancer (F0), F2-and F3-PDX samples exhibited 196 overlapping somatic mutations, with 272 somatic mutations in F0, 286 somatic mutations in F2-PDX, and 315 somatic mutations in F3-PDX (Fig. 3).
Altogether, these somatic mutations were further characterized and represented in different categories by conducting "maftools" analysis (Fig. 4). In particular, the "maftools" analysis displayed the top seven variant classifications in original cervical cancer (F0), F2-and F3-PDX samples, of which the missense mutation was the most frequently observed variant classification (Fig. 4A). Moreover, the somatic variant counts of missense mutation, frame shift del, in frame del, nonsense mutation, splice site, frame shift ins and in frame ins in separately original cervical cancer (F0), F2-and F3-PDX samples were shown in Fig. 4B. Figure 4C exhibited the mean values with standard deviations of the counts of these seven variants in the original cervical cancer (F0), F2-and F3-PDX samples. The single nucleotide polymorphism (SNP) was the most abundant mutation type (Fig. 4D). Meanwhile, the C > T variant was the most common type that observed in the single nucleotide variant (SNV) classification (Fig. 4E). In addition, the horizontal histogram exhibited the top ten mutated genes with 100% mutation rates, which implied that these ten genes were mutated in all the F0, F2-and F3-PDX samples ( Fig. 4F and Additional file 2: Table S2). These top ten mutated genes included phospholipase C beta 1 (PLCB1), lysine methyltransferase 2D (KMT2D), low-density lipoprotein receptor-related protein 1B (LRP1B), neuron navigator 3(NAV3), tumor protein P53 (TP53), marker of proliferation Ki-67 (MKI67), FAT atypical cadherin 1 (FAT1), PKHD1 like 1 (PKHD1L1), sperm associated antigen 17 (SPAG17) and KIAA1109, which had different counts of somatic mutations (Fig. 4F).
The distribution of mutations of these ten frequently mutated genes was exhibited by a Lollipop plot (Fig. 5). D659E and P813R were the most common missense mutations in PLCB1. Q1377*, G266* and E255* were the most frequent nonsense mutations in KMT2D, TP53 and KIAA1109, respectively. D2230H, R3158W, E2592Q, E1996Q and E438D were the most common missense mutations in LRP1B, FAT1, MKI67, PKHD1L1 and SPAG17, respectively. N2045Tfs*7 was the most frequent frameshift deletions in NAV3. In addition, the Variant Allele Frequency (VAF) information of these ten representative genes identified by WES in patient original cervical cancer (F0) and cervical cancer PDX models (F2 and F3) was displayed in Table 2.
Furthermore, the somatic mutated genes recognized using WES in original cervical cancer, F2-PDX and F3-PDX were next confirmed in the TCGA database. The seven somatic mutated genes (KMT2D, LRP1B, NAV3, TP53, MKI67, FAT1, PKHD1L1) identified by WES were part of the top 100 mutated genes in cervical   cancer samples, which were highlighted in the waterfall plot obtained from the TCGA database (Fig. 6).

Discussion
Constructing an ideal animal model for studying cervical cancer is essential for translational cancer research. Previous studies have displayed different protocols to establish cervical cancer PDX models, of which the engraftment success rates varied from 0 to 75% and the latency time fluctuated between 10 days and 12 months [20][21][22][23][24]. In our current study, we successfully developed a panel of cervical cancer PDX models by subcutaneously inoculating fresh cervical cancer tissues into NOD-SCID mice, which preserved the histologic characteristics of original cervical cancer. The latency time of cervical cancer PDX model establishment was variable with a progressive decrease as the passage number increased, with a mean time to initial growth of 94.71 days in F1 engraftment to 40.65 days in F3 engraftment. Additionally, the increased engraftment success rates of cervical cancer PDX models were revealed in our study during serial transplantation, with the success rate of F3 engraftment reaching 87.93%.   Furthermore, the next generation sequencing (NGS) provided a comprehensive approach to identify genomic biomarkers with high fidelity between the cervical cancer PDX models and the patients' original cervical cancer tissues, which potentially offers a molecular basis for clinical targeted therapy. Our study revealed that most genomic features of original cervical cancer were retained in cervical cancer PDX models throughout the xenografting and passaging process by performing WGS. And the cervical cancer PDX models maintained the majority of somatic mutations of their original cervical cancer, which was confirmed by WES. Full concordance in representative mutated genes, including KMT2D, LRP1B, NAV3, TP53, FAT1, MKI67 and PKHD1L1, was observed in original cervical cancer, F2-PDX models and F3-PDX models, which were also identified as frequently mutated genes in cervical cancer samples in the TCGA database. Largely compatible with previous researches [25][26][27][28], our study also identified genetic mutations of TP53 and MKI67 in the cervical cancer. In addition, our study firstly revealed the somatic mutations of KMT2D, NAV3 and PKHD1L1 in cervical cancer, to our knowledge. There are limited researches that reported the mutations of LRP1B and FAT1 genes in cervical cancer previously.
Low-density lipoprotein receptor-related protein 1B (LRP1B), which is localized on chromosome 2q, encodes an endocytic LDL-family receptor [29]. The LRP1B gene is considered as a putative tumor suppressor, and mutations in LRP1B are frequently observed in many cancers [30,31]. Based on the whole-genome sequencing and high-throughput viral integration detection of HPV integration in cervical cancer, Hu et al. [32] pointed out that LRP1B was identified as an HPV integration hotspot in cervical cancer, and the downregulated expression of LRP1B protein was found when HPV integrated in LRP1B introns, which potentially become a novel biomarker for early screening, diagnosis and personalized therapy of HPV integration-driven cervical cancer. Consistently, Cao et al. [33] stated that LRP1B mutation was correlated with HPV 16 integration status, and HPV-LRP1B integration decreased the expression of LRP1B, which was obtained from mining TCGA database. Moreover, this study also indicated that LRP1B mutation was in a relationship with unfavorable outcomes of cervical cancer patients [33]. In our present study, the missense mutation of LRP1B was detected in original cervical cancer, F2-PDX models and F3-PDX models, which may potentially become a new therapeutic target for cervical cancer patients, and further study will be needed to discover a novel and efficient targeted therapy.
FAT atypical cadherin 1 (FAT1) is a member of the family of fat-like atypical cadherins that encodes 4588 amino acid residues [34,35]. The mutations in FAT1 gene were detected frequently in diverse cancers, particularly in squamous cell carcinomas, potentially serving as an oncogene or a tumor suppressor [36,37]. Reportedly, inactivating mutations in the FAT1 gene potentially activated the Wnt signaling pathway and promoted the carcinogenesis, such as in head and neck squamous cell carcinomas [38]. Moreover, Chung et al. [39] performed the WES of 15 cervical adenocarcinoma-normal pairs and indicated that FAT1 was the most significantly mutated gene in cervical adenocarcinoma, with missense and nonsense mutations. Huang et al. [40] conducted the WGS or WES of 102 cervical cancer-normal pairs, and the mutational analysis discovered that the FAT1 was frequently mutated in cervical cancer, which accounting for 8.8% of all cases (9/102). In our current research, the missense mutation of FAT1 was identified in original cervical cancer, F2-PDX models and F3-PDX models. Therefore, the cervical cancer PDX models conserved the somatic mutations of FAT1 in original cervical cancer, and the mutated FAT1 gene could be utilized as the molecular target of new therapeutic regimens for cervical cancer patients.

Conclusion
In summary, we succeeded in generating serially transplantable cervical cancer PDX models by subcutaneously inoculating fresh cervical cancer tissues into NOD-SCID mice, which exhibited a high concordance of histopathology and genome of the primary cervical cancer patient tissues. Performing WES in original cervical cancer tissues and cervical cancer PDX models helped to gain a deeper insight into the genetic alterations that were essential for the carcinogenesis of cervical cancer. Our established cervical cancer PDX models combined with genomic annotations can be employed to explore the novel molecular targeted therapies, validate novel therapeutic regimens, and predict the sensitivity of chemotherapeutic drug in cervical cancer patients in the future. Therefore, the present study lays a foundation for further investigation of the personalized treatments of cervical cancer patients and has important implications for translational cancer research.