Systematic Review of Patient-Derived Xenograft Models for Preclinical Studies of Anti-Cancer Drugs in Solid Tumors

Patient-derived xenograft (PDX) models are used as powerful tools for understanding cancer biology in PDX clinical trials and co-clinical trials. In this systematic review, we focus on PDX clinical trials or co-clinical trials for drug development in solid tumors and summarize the utility of PDX models in the development of anti-cancer drugs, as well as the challenges involved in this approach, following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines. Recently, the assessment of drug efficacy by PDX clinical and co-clinical trials has become an important method. PDX clinical trials can be used for the development of anti-cancer drugs before clinical trials, with their efficacy assessed by the modified response evaluation criteria in solid tumors (mRECIST). A few dozen cases of PDX models have completed enrollment, and the efficacy of the drugs is assessed by 1 × 1 × 1 or 3 × 1 × 1 approaches in the PDX clinical trials. Furthermore, co-clinical trials can be used for personalized care or precision medicine with the evaluation of a new drug or a novel combination. Several PDX models from patients in clinical trials have been used to assess the efficacy of individual drugs or drug combinations in co-clinical trials.


Introduction
Cancer is the leading cause of death in economically developed countries, and many physicians and scientists are using their efforts to develop new treatment approaches. At present, surgery, radiation therapy, and chemotherapy are the three pillars for curing and prolonging survival of patients with cancer. Among these treatments, cancer chemotherapy plays a central role for patients with advanced or recurrent cancer. Conventional anti-cancer drugs can directly kill rapidly proliferating cells, while small molecule inhibitors and therapeutic antibodies can inhibit the intracellular growth signal cascade and lead to cancer cell-specific death [1,2]. For the development of anti-cancer drugs, in vitro cell killing assays using commercially available patient-derived cell lines or in vivo tumor growth inhibition assays using cell-line-derived xenograft (CDX) models are commonly employed to measure the efficacy of drugs and to make a "go or no-go" decision for further clinical study. Unfortunately, few drugs are approved even if the drugs demonstrate a good response in preclinical studies. Indeed, only 5% of the anti-cancer drugs that have anti-cancer activity in preclinical studies are approved for clinical application by the United States Food and Drug Administration (FDA) [3].
To develop anti-cancer drugs for solid tumors, knowledge of the hallmarks of cancer and the cancer microenvironments is important [4]. The cancer microenvironments consist of cancer cells and

Mouse Strain and Implantation Site for PDX Model
The mouse strains used for PDX models are shown in Figure 4. A PDX model can be developed by implantation of cancer tissue into an immunodeficient mouse. Although a variety of mouse strains were used, the most common strains were nude (47%, 47/100), scid (14%, 14/100), NOD-scid (11%, 11/100), and NOG/NSG (the same as NOD/scid/IL2γ-receptor null; 21%, 21/100). Nude mice lack CNS

Mouse Strain and Implantation Site for PDX Model
The mouse strains used for PDX models are shown in Figure 4. A PDX model can be developed by implantation of cancer tissue into an immunodeficient mouse. Although a variety of mouse strains were used, the most common strains were nude (47%, 47/100), scid (14%, 14/100), NOD-scid (11%, 11/100), and NOG/NSG (the same as NOD/scid/IL2γ-receptor null; 21%, 21/100). Nude mice lack mature T cells, scid mice lack mature B and T cells, NOD-scid mice lack mature B and T cells and have impaired NK cells, and NOG/NSG mice lack mature B and T cells, as well as NK cells [114]. Two other immunodeficient mouse strains, scid-beige [70,101] and RAG2 null/γC null [30], were also used. The use of NOG/NSG mice is increasing because these are most severely immunodeficient and the tumor engraftment rate (take-rate) is the best among the mouse strains. However, nude mice are still the most commonly used strain, owing to their low cost and because hairless mice are easy to assess for tumor growth. Recently, several cancer immunotherapies have been assessed in preclinical and clinical studies. Conventional PDX models are difficult to use to analyze the efficacy of immuno-oncology drugs, as the immunodeficient mice have no cytotoxic T cells. Therefore, syngeneic mouse models are commonly used for the development of these drugs. To resolve this issue, humanized PDX models have been developed. Vargas et al. reported on humanized PDX models used to evaluate treatment with the immune checkpoint inhibitor nivolumab using NSG mice with an injection of tumor tissue as well as the patient's peripheral blood lymphocytes (PBLs) [15].
The use of NOG/NSG mice is increasing because these are most severely immunodeficient and the tumor engraftment rate (take-rate) is the best among the mouse strains. However, nude mice are still the most commonly used strain, owing to their low cost and because hairless mice are easy to assess for tumor growth. Recently, several cancer immunotherapies have been assessed in preclinical and clinical studies. Conventional PDX models are difficult to use to analyze the efficacy of immunooncology drugs, as the immunodeficient mice have no cytotoxic T cells. Therefore, syngeneic mouse models are commonly used for the development of these drugs. To resolve this issue, humanized PDX models have been developed. Vargas et al. reported on humanized PDX models used to evaluate treatment with the immune checkpoint inhibitor nivolumab using NSG mice with an injection of tumor tissue as well as the patient's peripheral blood lymphocytes (PBLs) [15].
The tumor implantation sites are shown in Figure 5. Although subcutaneous implantation was most common (80%, 79/99), orthotopic implantation was sometimes selected for several types of cancer, including the implantation of primary or metastatic brain tumor into the brain [57,61,63,83,101], breast cancer into mammary gland [26,46,91], and renal cancer into renal capsule [30]. Although the patient-derived orthotopic xenograft (PDOX) models mimic metastasis of cancer and are important models for basic and applied research compared to PDX models [115], the subcutaneous PDX models are commonly used in the PDX clinical trials or co-clinical trials because the PDX models are easy to assess the drug efficacy compared to the PDOX models.  The tumor implantation sites are shown in Figure 5. Although subcutaneous implantation was most common (80%, 79/99), orthotopic implantation was sometimes selected for several types of cancer, including the implantation of primary or metastatic brain tumor into the brain [57,61,63,83,101], breast cancer into mammary gland [26,46,91], and renal cancer into renal capsule [30]. Although the patient-derived orthotopic xenograft (PDOX) models mimic metastasis of cancer and are important models for basic and applied research compared to PDX models [115], the subcutaneous PDX models are commonly used in the PDX clinical trials or co-clinical trials because the PDX models are easy to assess the drug efficacy compared to the PDOX models.

Applications of PDX Models in Cancer Research
Applications of PDX models in cancer research are shown in Figure 6. The efficacy of potential anti-cancer drugs can be investigated using cell lines in vitro and CDX models in vivo. In addition, PDX models can then be used to conduct a drug efficacy study. The majority of articles were categorized into this traditional method (54%, 53/98). An article representative of this method was published by Cho et al. in 2018 [92]. In the study, the efficacy of anti-prostate-specific membrane antigen (PSMA) antibody drug conjugate (ADC) was assessed using prostate cancer models. First, the IC50 of the ADC was analyzed using six cell lines with several PSMA expression levels. Next, tumor growth inhibition by ADC administration was assessed in three CDX models with several PSMA expression levels. Finally, the efficacy of the ADC was assessed in eight PDX models with several PSMA expression levels. The PSMA-ADC demonstrated potent and specific cytotoxicity in PSMA-positive cell lines, and robust and durable antitumor activity in PSMA-positive CDX and PDX models. Drug efficacy was also assessed by PDX clinical trials and co-clinical trials as summarized in the following two sections.
Tumor biology is also an important subfield of cancer research. Bonanno et al. clarified the relationship of liver kinase B1 (LKB1) expression level with sensitivity to bevacizumab using PDX models [68]. Bertolini et al. showed that CD133 + /CXCR4 + cancer initiating cells isolated from PDX models had a superior ability to seed and initiate metastasis at distant organs [40]. Comparing the 2010-2016 period to 2017 onwards shows that the percentage of studies investigating mechanisms or biomarkers for cancer growth or metastasis using PDX models has doubled (first period:second period, 8%:19%).

Applications of PDX Models in Cancer Research
Applications of PDX models in cancer research are shown in Figure 6. The efficacy of potential anti-cancer drugs can be investigated using cell lines in vitro and CDX models in vivo. In addition, PDX models can then be used to conduct a drug efficacy study. The majority of articles were categorized into this traditional method (54%, 53/98). An article representative of this method was published by Cho et al. in 2018 [92]. In the study, the efficacy of anti-prostate-specific membrane antigen (PSMA) antibody drug conjugate (ADC) was assessed using prostate cancer models. First, the IC 50 of the ADC was analyzed using six cell lines with several PSMA expression levels. Next, tumor growth inhibition by ADC administration was assessed in three CDX models with several PSMA expression levels. Finally, the efficacy of the ADC was assessed in eight PDX models with several PSMA expression levels. The PSMA-ADC demonstrated potent and specific cytotoxicity in PSMA-positive cell lines, and robust and durable antitumor activity in PSMA-positive CDX and PDX models. Drug efficacy was also assessed by PDX clinical trials and co-clinical trials as summarized in the following two sections.
Tumor biology is also an important subfield of cancer research. Bonanno et al. clarified the relationship of liver kinase B1 (LKB1) expression level with sensitivity to bevacizumab using PDX models [68]. Bertolini et al. showed that CD133 + /CXCR4 + cancer initiating cells isolated from PDX models had a superior ability to seed and initiate metastasis at distant organs [40]. Comparing the 2010-2016 period to 2017 onwards shows that the percentage of studies investigating mechanisms or biomarkers for cancer growth or metastasis using PDX models has doubled (first period:second period, 8%:19%).

PDX Clinical Trials
Fourteen articles reporting on PDX clinical trials are summarized in Table 1. PDX clinical trials are important for the development of anti-cancer drugs and clinical decision making before human clinical trials. This is reflected in the fact that 20 to 30 PDX models were enrolled in most of the PDX clinical trials. As PDX clinical trials are also referred to as "phase II type clinical trial like models", several solid tumors were enrolled following an "all-comer model" [10]. The tumor growth in PDX models in such PDX clinical trials have commonly been assessed by the modified response evaluation criteria in solid tumors (mRECIST) guidelines and shown in a waterfall plot as is done in human clinical phase II trials. During the first study period, five to 10 animals per treatment group were commonly used. In 2015, Gao et al. reported a high-throughput in vivo drug screening method using a large number of PDX models [10]. The study design was "1 × 1 × 1", which meant one animal per model per treatment. After the publication of this article, 1 × 1 × 1 approaches were mainly used in PDX clinical trials. However, as one animal per PDX model might not be representative of generalizable drug response, two or three animals per model per treatment (2 × 1 × 1 or 3 × 1 × 1) PDX clinical trials have recently become more common.

PDX Clinical Trials
Fourteen articles reporting on PDX clinical trials are summarized in Table 1. PDX clinical trials are important for the development of anti-cancer drugs and clinical decision making before human clinical trials. This is reflected in the fact that 20 to 30 PDX models were enrolled in most of the PDX clinical trials. As PDX clinical trials are also referred to as "phase II type clinical trial like models", several solid tumors were enrolled following an "all-comer model" [10]. The tumor growth in PDX models in such PDX clinical trials have commonly been assessed by the modified response evaluation criteria in solid tumors (mRECIST) guidelines and shown in a waterfall plot as is done in human clinical phase II trials. During the first study period, five to 10 animals per treatment group were commonly used. In 2015, Gao et al. reported a high-throughput in vivo drug screening method using a large number of PDX models [10]. The study design was "1 × 1 × 1", which meant one animal per model per treatment. After the publication of this article, 1 × 1 × 1 approaches were mainly used in PDX clinical trials. However, as one animal per PDX model might not be representative of generalizable drug response, two or three animals per model per treatment (2 × 1 × 1 or 3 × 1 × 1) PDX clinical trials have recently become more common.  4 Mouse strain. 5 Implantation site. 6 Number of animals per each treatment group. 7 Non-small cell lung cancer. 8 Colorectal cancer. 9 Glioblastoma. 10 Renal cell carcinoma. 11 Head and neck squamous cell carcinoma. 12 Subcutaneously implantation. 13 Orthotopic implantation.

Co-clinical Trials
Nine articles described co-clinical trials. These are summarized in Table 2. Co-clinical trials are recognized as "avatar" models for personalized care or precision medicine [11,12]. Thus, the number of PDX models enrolled in most co-clinical trials was less than 10. Since the biopsy specimens collected from metastatic lesion were small, NOG/NSG mice were used in half of the co-clinical trials to increase the engraftment rate. Unlike PDX clinical trials, five to 10 animals per treatment group were used in most of these co-clinical trials. PDX models using co-clinical trials were established from patients enrolled in clinical trials, and most of the cancer tissues were collected prior to treatment. Stebbing et al. reported a correlation between PDX model results and clinical outcomes (81%, 13/16) [37]. Frankel et al. reported that responses to anti-cancer drugs in a PDX model reflected the responses in patients [76]. The PDX model reported by Campbell et al. was generated from a patient treated with a mitogen-activated protein kinase kinase (MEK, also known as MAPKK) inhibitor for two weeks in a neoadjuvant setting [90]. Although the patient had a clinical response with down-staging of the tumor and underwent surgical resection, the patient later developed metastasis. In the co-clinical trial, the PDX model was treated with a longer course of the MEK inhibitor to determine whether this regimen had a clinical benefit. In this case, the MEK inhibitor suppressed tumor growth of the PDX model over the first 50 days of treatment; however, the tumor re-grew after 50 days. Thus, the PDX model displayed responses consistent with the clinical findings in the patient. In some co-clinical trials, not only were the same anti-cancer drugs used in clinical trials evaluated, but novel drugs or combinations were also assessed [81]. Interestingly, Vargas et al. reported a case study of real-time personalized care in a co-clinical trial [15]. In the case study, the response to first-and second-line therapies in PDX models could predict the response to first-line therapy, development of resistance and response to second-line therapy before these events were observed in the patient.

Discussion
Since the articles were searched using the terms "PDX clinical trial" or "patient-derived xenograft clinical trial", some articles using PDX models were not identified and this review is done at the study level. Cell lines or CDX models used for drug screening fail to accurately predict the clinical efficacy of anti-cancer drugs. Thus, PDX models are powerful preclinical models for cancer research, especially drug development, in several cancer types [116]. Although major cancers were often used for PDX models, the need for rare cancer PDX models for use in PDX clinical trials or co-clinical trials is increasing [15]. Regarding pediatric cancer, US Pediatric Preclinical Testing Consortium (http://www.ncipptc.org/) and Children's Oncology Group cell culture and xenograft repository (https://www.cccells.org/index.php) have a lot of PDX models of rhabdomyosarcoma, Ewing sarcoma, neuroblastoma, and osteosarcoma, and are contributed to drug development. The PDOX models are commonly used for neuroblastoma because the tumors spontaneously metastasize to the liver, lung and bone marrow and are promising models for studying and targeting human neuroblastoma metastases [117,118]. The other topic of PDX model is humanized PDX model for the development of immuno-oncology (IO) drugs [119]. The conventional PDX models are lack of mouse immune systems and human immune cells, thus, IO drug, such as immune checkpoint inhibitor, is difficult to be assessed by using these models. To resolve this issue, PBLs or tumor-infiltrating lymphocytes (TILs) are injected into NOG/NSG mice [15]. However, these procedures cause severe graft-versus-host disease within two to five weeks after injection of PBL or TIL, thus, these mice are used only for short-term assessing of IO drugs. Compared to PBL or TIL mice, the models of transplantation of CD34-positive human hematopoietic stem cells (HSCs) isolated from umbilical cord blood into NOG/NSG mice have complete hematopoietic reconstitution. These next-generation PDX models are used for the development of IO drugs.
The average engraftment rate of PDX models was recently estimated at 49% [8], allowing the PDX model to be commonly used not only for basic cancer research but also for assessing drug efficacy in preclinical research. However, the resources and processes for creating and characterizing PDX models differ greatly among academic institutions. To address this, Meehan et al. published the PDX models minimal information standard (PDX-MI) for reporting on the generation, quality assurance, and use of PDX models. This standard was developed by four independent PDX model resources (the EurOPDX consortium, the IMODI consortium, the Patient-Derived Models Repository at NCI-Federick, and the Jackson Laboratory PDX Resource) [120]. For consistency, PDX models should be reported and generated according to this standard.
The traditional PDX models have commonly been used for the development of anti-cancer drugs. PDX clinical trials and co-clinical trials using PDX models are becoming an increasingly important method for drug development. PDX clinical trials, such as "phase II type clinical trial like mouse models", can be used for the development of anti-cancer drugs before human clinical trials, and the results assessed by the mRECIST guidelines can be shown in a waterfall plot. In these studies, a few dozen PDX models from a PDX repository are enrolled and the efficacy of the drugs is assessed by 1 × 1 × 1, 2 × 1 × 1 or 3 × 1 × 1 approaches. On the other hand, co-clinical trials, such as "avatar" models, can be used for personalized care or precision medicine for the evaluation of a new drug or a novel combination. In these studies, several cases of PDX models from patients in clinical trials are used to determine the efficacy of the drugs or drug combinations.

Conclusions
PDX models are useful preclinical models for the development of anti-cancer drugs employed in PDX clinical trials and co-clinical trials. Information regarding the generation, quality assurance, and use of PDX models should be recorded according to the PDX-MI.