Numerous academic institutions have begun renewing their interest in genetically engineered mouse models (GEMMs), cell line-derived xenografts, and PDX models.

GEMMs are classified into two categories: Transgenic and targeted. In transgenic GEMMs, exogenous oncogenes are expressed through pronuclear injection of embryonic stem cells. Targeted GEMMs involve homologous recombination in mouse embryonic stem cells, in which targeting vectors modify the homologous arms of a genomic locus to the accuracy of a single base. GEMMs have been used to elucidate multiple aspects of cancer development. Khaled et al (17) summarized data from multiple studies. For example, GEMMs are particularly suitable for deleting or overexpressing targeted genes in a tissue-specific manner, are easy to produce and use endogenous regulatory elements. However, establishing the model takes ~1 year. Another major disadvantage of GEMMs is the low success rate between two targeted sites. In addition, GEMMs may not be able to mimic the individualized therapy associated with a tumor-specific gene. Therefore, GEMMs may not represent the optimal preclinical trial model.

Cell lines are transplanted into immunodeficient mice to establish tumor models by numerous means, including subcutaneous implantation and orthotopic, venial or peritoneal injection. In 2007, Hajitou et al (18) established the first soft tissue sarcoma cell line-derived tumor in the hind limb of rats. Subsequently, the development of cell line-derived xenograft models has decreased the influence of irrelevant cells when studying a single factor (16), improved the use of experimental cellular operations in researching tumor-associated signaling pathways and molecular mechanisms (19,20), established acute myeloid leukemia animal models to mimic the progression of liquid tumors (21), and aided research into metastatic mechanisms via intravenous or intraperitoneal injection (22). In addition, cell line-derived xenograft mouse models are easily established, and generating tumors takes only 2–8 weeks (23).

However, since generating cancer cell lines may irreversibly alter the biological properties of the derived cells, cell line-derived xenografts have limited predictive value for cancer therapy. The major disadvantages of this model are as follows: Following the dissociation of tumor tissues into a single cell suspension in vitro, selection pressure tends to result in a decrease in the heterogeneous characteristics of the tumors; not all cell lines are suitable for patients with certain types of cancer (17) and cell lines may not accurately reflect the complexity of tumor heterogeneity (24), which impacts patient-specific responses to clinical therapy. Therefore, cell line-derived xenograft models may not satisfy the requirements of individualized medicine.

Compared with GEMMs and cell line-derived xenograft models, PDX models have more predictive value for clinical outcomes (25,26). Multiple studies have reported the high fidelity of PDXs regarding histology, the transcriptome and genome (12–16). Previous studies have obtained more detailed data on tumor cell population dynamics using deep-genome and single-cell sequencing techniques in PDXs for breast cancer (27,28). Therefore, the present study suggests that PDXs may be optimal models for studying tumor heterogeneity and currently represent the most powerful tool for assessing cancer-associated mechanisms. However, PDXs remain to be popularized in clinical application. Thus, the present study reviewed the progression of PDXs, particularly the challenges faced by these models, and future directions for individualized medicine (Table I).

PDXs may have originally failed to enter mainstream cancer research due to the limitations of using host animals that are not sufficiently immunodeficient and initiate xenograft rejection. However, using animal models is now common, since the number of immunodeficient host animal models has increased and the cost of immunodeficient mice has decreased (29). At present, nude, nonobese diabetic (NOD)/severe combined immunodeficiency (SCID) and NOD SCID γ (NSG) mice represent the three most commonly used types of immunodeficient mice. Nude mice are athymic, which results in a congenital deficiency of T and normal B lymphocytes and enhanced activity of natural killer (NK) cells (30). NOD/SCID mice are generated by crossing C.B.-17-SCID mice with NOD mice. SCID and NOD/SCID mice exhibit a congenital deficiency of T and B lymphocytes; the latter also exhibits decreased NK cell activity, which aids the use of the model since residual NK cells serve a key function in rejecting human tissues. NSG mice exhibit a deficiency of the interleukin 2 receptor γ-chain via the genetic engineering of NOD/SCID mice. NSG mice are the most severely immunodeficient since they exhibit T, B and NK cell deficiency. Therefore, the engrafting success rate in NSG mice is typically increased compared with that in nude or NOD/SCID mice (31). Therefore, NSG mice may be the most suitable of the three to generate tumors (32). NSG mice have been successfully utilized in multiple types of cancer and the present study has summarized the current state of PDXs (Table I).

Circulating tumor cells (CTCs) are released from primary tumors into peripheral blood vessels. Conditions permitting, CTCs may infiltrate distant tissues and thereby induce metastasis by providing ‘seeds’ to distant organs, in accordance with the ‘seed and soil’ theory (33). CTCs represent a readily accessible method for liquid biopsy. Numerous researchers over the past decades have focused on identifying and improving technically challenging methods of isolating CTCs (34). Currently, the CellSearch System is the only Food and Drug Administration-approved technique for CTC enumeration (35). However, the CellSearch System depends on epithelial cell adhesion molecules (EpCAMs) and therefore may only identify a small number of CTCs, potentially resulting in false negative or positive readings (36). Therefore, researchers have previously used the brain metastasis-selected markers (BMSMs) erb-b2 receptor tyrosine kinase 2 (ERBB2)⁺/epidermal growth factor receptor (EGFR)⁺/heparanase⁺/notch 1⁺ to identify EpCAM⁻ CTCs (37). The infiltration and metastasis of these CTCs were analyzed using BMSMs and these cells were then demonstrated to be invasive and capable of generating brain or lung metastasis in nude mice. Researching EpCAM⁻ CTCs may also serve to improve the understanding of metastasis. Enrichment steps are crucial to increase the isolation success rate. Although enriching CTCs may reach 98% of total CTCs by using the CTC-Chip, this chip has not become available on the market yet (38). Previously, the drug sensitivity of cultured CTCs was tested among multiple CTC lines (39). Furthermore, another study demonstrated that testing the drug sensitivity of CTC lines for clinical regimes is in accordance with the clinical treatment response (40). Drug sensitivity testing of CTCs may predict the donor patient's response and direct appropriate therapy for individualized medicine. In addition, CTC-derived xenografts have been used in the laboratory for studying certain types of cancer. One institution reported that the PDXs for small cell lung cancer (SCLC) may reflect the responses of patients to clinical regimes (34). Therefore, CTC-derived xenograft models (CDXs) may be used to detect drug resistance mechanisms.

PDXs are maintained by passaging tumor tissues directly from mouse to mouse. Heterotopic or orthotopic PDXs involve implanting tissues into the subcutaneous flank or the organs of mice. PDXs are considered to be superior to traditional cell line xenografts since the former may be more similar to parental tumors compared with the latter, particularly in terms of tumor, intratumor and intrametastasis heterogeneity (26). Traditional tumor models exhibit poor predictive values due to the associated heterogeneity. The development of PDXs satisfies the requirements of an effective preclinical tool, and PDXs are a predictive model of carcinogenesis physiology and clinical therapy. As genomic research develops, further subtypes of cancer are predicted. For example, five molecular subtypes of breast cancer have been characterized, studied and categorized using the Prosigna Breast Cancer Prognostic Gene Signature assay test: Luminal A, luminal B, ERBB2+, basal-like and normal-like (41). Subtypes are further divided into more detailed types for each individual in clinical practice (42,43).

In addition, studies on targeted drug treatments confirm the accurate predictive value, fidelity and stability of PDXs and PDX-associated clinical prognosis (Fig. 1A, B and C).

In 2014, Stebbing et al (44) established PDXs using puncture tissues derived from 29 patients with advanced sarcoma, and tissues derived from 16 of the 29 patients successfully established PDXs. According to the results of the study, 6 of the patients benefited from PDX-guided therapy. The remaining patients exhibited an association between clinical outcomes and their PDXs as demonstrated by retrospective analysis. In 2014, Elena Garralda et al (45) combined next-generation sequencing with PDXs to guide individualized treatments for 25 patients with advanced solid tumors. The most effective individualized drugs were selected according to the sequencing results of the PDXs. Subsequently, 14 of the 25 patients received personalized treatments, and 11 achieved durable partial remission. Subsequently, PDX models have increased in popularity.

The use of PDXs is advantageous in drug-screening and resistance mechanism research. PDXs possess an effective predictive value in targeted drug-screening for clinical treatments. The development of optimal regimes, based on drug-screening in PDXs, may improve the survival rate of patients with cancer. Furthermore, gene mutations cause tumorigenesis, particularly those in p53 and phosphatase and tensin homolog (24,46). Although gene mutations may be identified using whole-genome sequencing, identifying the significant mutations with which specific diseases are associated remains a challenge. Therefore, Berg et al (47) aimed to construct a comprehensive dataset, including driver mutations, tumorigenicity variants and clinical responses, but this was discovered to be time-consuming. In addition, drug resistance mechanisms may consist of primary or acquired resistance (48). PDXs may represent intratumor and intrametastasis heterogeneity, and more accurately predict resistance mechanisms to clinical treatments. Therefore, PDXs may potentially represent a platform for evaluating personalized resistance mechanisms. Personalized medicinal strategy may become a future direction for personalized treatment and translational medicine.

The poor predictive value of the preclinical models used to select novel drugs is partly responsible for the low success rate of novel agents in clinical application. PDXs are more predictive of clinical outcomes and possess a vast development foreground in the preclinical screening of novel anticancer drugs. Chiron et al (49) compared the anticancer effect of aflibercept with that of bevacizumab by using PDXs in multiple genetic backgrounds, demonstrating that aflibercept had increased anticancer functions compared with bevacizumab. Monsma et al (50) revealed that vemurafenib was effective in PDXs of melanoma with B-Raf proto-oncogene, serine/threonine kinase (BRAF)^V600E or BRAF^V600V. Furthermore, the combination of mitogen-activated protein kinase inhibitors and vemurafenib improved the effectiveness of anticancer therapies. Consequently, screening of the susceptibility of drugs may be effectively integrated into clinical translational medicine (Fig. 1C and D).

Coclinical trials are characterized by parallel studies between mouse models and patients, which may help determine treatment strategies for patients and to identify underlying cancer-associated mechanisms with the aid of PDXs (51,52). As cancer progresses, the drug becomes less effective and novel resistance mechanisms appear. When and how such mechanisms develop is unpredictable during clinical treatment. Drug resistance may be observed earlier with the aid of PDXs, which may assist in determining subsequent treatment regimes. GEMMs have been used in coclinical trials and exhibited positive results in parallel clinical trails (53), including those involving leukemia, melanoma, prostate cancer and non (N)SCLC (53–56). Furthermore, coclinical trials may also verify relevant hypotheses in a clinical setting and thereby affect the design of future clinical studies (57). However, this may be associated with increased costs (51). PDXs have not been used in large scale coclinical trials (38). Bertotti et al (38) reported that 8 patients with metastatic colorectal cancer were successfully treated using targeted PDX-based therapies combining anti-ERBB2 with anti-EGFR to predict resistance to anti-EGFR targeted therapy. Coclinical trials monitor the responses of individuals and parallel mouse models simultaneously and provide an in vivo model to research suspicious resistance mechanisms and test combination strategies for overcoming novel spontaneous resistance mechanisms (Fig. 1E).