Colorectal cancer (CRC) is the third most common diagnosed malignancy among both sexes in the United States as well as in the European Union. While the incidence and mortality rates in western, high developed countries are declining, reflecting the success of screening programs and improved treatment regimen, a rise of the overall global CRC burden can be observed due to lifestyle changes paralleling an increasing human development index. Despite a growing insight into the biology of CRC and many therapeutic improvements in the recent decades, preclinical in vivo models are still indispensable for the development of new treatment approaches. Since the development of carcinogen-induced rodent models for CRC more than 80 years ago, a plethora of animal models has been established to study colon cancer biology. Despite tenuous invasiveness and metastatic behavior, these models are useful for chemoprevention studies and to evaluate colitis-related carcinogenesis. Genetically engineered mouse models (GEMM) mirror the pathogenesis of sporadic as well as inherited CRC depending on the specific molecular pathways activated or inhibited. Although the vast majority of CRC GEMM lack invasiveness, metastasis and tumor heterogeneity, they still have proven useful for examination of the tumor microenvironment as well as systemic immune responses; thus, supporting development of new therapeutic avenues. Induction of metastatic disease by orthotopic injection of CRC cell lines is possible, but the so generated models lack genetic diversity and the number of suited cell lines is very limited. Patient-derived xenografts, in contrast, maintain the pathological and molecular characteristics of the individual patient’s CRC after subcutaneous implantation into immunodeficient mice and are therefore most reliable for preclinical drug development – even in comparison to GEMM or cell line-based analyses. However, subcutaneous patient-derived xenograft models are less suitable for studying most aspects of the tumor microenvironment and anti-tumoral immune responses. The authors review the distinct mouse models of CRC with an emphasis on their clinical relevance and shed light on the latest developments in the field of preclinical CRC models.

Colorectal cancer (CRC) is the third most common diagnosed malignancy among both sexes in the United States as well as in the European Union[1,2]. A decrease in the incidence and overall mortality of CRC in countries with a very high development index over the last decades can be attributed to an interplay of screening programs with detection of precancerous lesions or early stage cancers on the one hand[3,4], and improved therapeutic concepts leading to an increased stage adjusted survival for all stages of CRC on the other hand[5]. This development is in sharp contrast to countries with a rapidly growing high development index. Together with an increased CRC incidence and mortality this reflects an adoption of the so-called “western lifestyle” including the risk factors for CRC. While obesity, smoking and red meat consumption are significantly associated with an elevated CRC risk[6], physical activity, high vegetable, fruit and fiber intake as well as metronomic aspirin therapy, have shown to decrease CRC risk[7,8]. Besides these modifiable risk factors, a variety of genetic factors influences CRC risk. About 5% of CRC cases are caused by hereditary, highly penetrant cancer syndromes, like familial adenomatous polyposis (FAP) and Lynch syndrome (LS); whereas up to 20%-30% of cases are considered as “familial” due to different germline mutations of varying penetrance[9]. Since the discovery of the link between APC germline mutations and FAP[10], followed by the genetic pathology of LS in the early 1990s[11,12], a myriad of mutations contributing to CRC genesis has been identified, constantly reshaping the genomic landscape of the disease[13]. The ideal model of CRC should recapitulate the progress from a precancerous adenoma to an invasive carcinoma with metastatic potential and at the same time it has to reflect the inter-individual molecular diversity of the disease. Consequently, a single (mouse) model of CRC simply cannot match all of these requirements. In this review, we discuss the different mouse models of CRC with their distinct advantages and disadvantages with a focus on their translational and clinical relevance.

Carcinogen-induced models (CIM) in rodents look back on a long tradition but maintained their usefulness for certain applications to the present day. They provide a platform for dietary studies and give insights into the pathways of food-borne carcinogens and colitis-associated carcinogenesis. Administration of the chemical compounds is possible via ad libitum feeding, oral gavage, intraperitoneal/ subcutaneous (s.c.) or intramuscular injection, or enema.

In 1915, Yamagiwa et al[14] proofed the carcinogenic properties of coal tar by its repetitive application on the ears of rabbits. At about the same time, first researchers worked on colon carcinogenesis by applying chemical or radioactive substances[15-17]. In the 1960s, cycasin and its metabolite, methylazoxymethanol, have shown to induce cancers in rodents[18-20]. In the following years, the more chemically stable substances, azoxymethane (AOM) and its precursor molecule, 1,2-dimethylhydrazine as well as methylazoxymethyl acetate, were extensively used to induce colon carcinogenesis in mice and rats. All three compounds are metabolized to methylazoxyformaldehyde, which is able to alkylate the DNA bases guanine and thymine[21]. After being processed by Phase-II-reaction, it is secreted to the bile and exceeds its carcinogenic effect to the intestinal mucosa[22]. Interestingly, these compounds show different carcinogenic potential depending on the mouse strain, housing conditions and the way of administration[23-25]. Although most authors claim a certain organotropism for AOM and dimethylhydrazine, tumor formation happens mostly in the small intestine and relevant amounts of alkylated DNA adducts can be observed in the liver and the kidneys[26]. Moreover, intestinal carcinogenesis can be achieved by the oral or rectal application of the direct alkylating topic agents N-methyl-N-nitrosourea (MNU), 3,2’-dimethyl-4-aminobiphenyl and N-Methyl-N’-nitro-N-nitrosoguanidine of which the latter two are traditionally used in rats[27-29].

Other carcinogens gained attention in connection with the association between meat consumption and increased CRC risk[30]. Heterocyclic aromatic amines (HAA) form from the reaction between free amino acids, sugars and creatine at high temperatures during the cooking process of meat and fish[31], whereby 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP) and 2-Amino-9H-pyrido[2,3-b]indole are the most abundant HAA in various foods[32]. PhIP is metabolized by the liver enzyme CYP1A2 to N2-Hydroxy-PhIP, which then, after sulfation or acetylation, forms activated esters capable of DNA adduct formation[33]. Detoxification of PhIP and its metabolites is driven by glutathione conjugation and glucuronidation[34,35]. Glucuronide conjugates are excreted through urine and bile[36]. In case of the latter, hydrolyzation by bacterial beta-glucuronidases in the intestines liberate reactive PhIP metabolites, which not only affect the intestinal mucosa, but undergo enterohepatic circulation[37]. Important to consider is, that the metabolism of PhIP in rodents results in less reactive metabolites than in humans, and its carcinogenic potential measured in animal studies might be even higher in humans[38]. Nakagama et al[39], by combining a high fat diet with PhIP intake, showed the tumor enhancing properties of this food borne agent simulating the carcinogenic effects of the s.c. “western diet”. Moreover, PhIP led to the formation of high-grade dysplasia and adenocarcinomas in a mouse model of chemical induced colitis[40]. Although other common foodborne HAAs have shown to induce dysplasia and carcinomas in rodents[41], they are rarely used for modelling colon carcinogenesis. Polycyclic aromatic hydrocarbons, as benzo[a] pyrene, may be used for chemoprevention studies but are insignificant for CRC modeling in general[42,43]. Dextran sodium sulfate (DSS) must also be mentioned when discussing chemical-induced CRC mouse models. Since the first report of an DSS-induced colitis model nearly 30 years ago[44], countless studies used DSS to simulate chronic inflammatory bowel diseases and we would recommend the reader to refer to excellent reviews discussing inflammatory bowel diseases and DSS[45]. As a sulfated polysaccharide, DSS directly damages the colonic epithelium resulting in an impairment of the mucosal barrier with consecutive entry of luminal bacteria and associated antigens into the mucosa, triggering inflammation[46]. Depending on the animal strain, DSS dosage and administration regimen, mice can develop acute and chronic colitis or even colitis-induced dysplastic lesions[47-50]. DSS in combination with carcinogenic compounds, primarily AOM, has been proven useful for the research of colitis-induced cancer[51]. By use of mice with germline Apc mutation (APC^Min mice) for DSS treatment, the rate of dysplasia and carcinoma formation was further enhanced[52]. Cooper et al[53] reported an increased CRC incidence of 40% in APC^Min mice after two cycles of 4% DSS treatment compared to untreated control animals and identified the loss of heterozygosity (LOH) of Apc as underlying cause.

Chemical-induced mouse models are not homogenous and possess specific advantages and disadvantages. They are by far the oldest method inducing CRC in animals and a multitude of studies has been traditionally carried out in rats and other rodents diminishing the comparability of older data with more recent results. Most models reflect very well the progression from aberrant crypt foci to adenomas to carcinomas of the human adenoma-carcinoma sequence[54]. Therefore, they are still useful to evaluate the influence of diet[55], dietary supplements[56,57], chemopreventive interventions[58,59] and the gut microbiome[60]. Especially the combination of DSS with carcinogenic agents provided many insights in the link between CRC and inflammation[61]. However, chemical-induced carcinomas rarely show invasive properties and local or distant metastases are usually absent. Albeit Yang et al[62] reported lymph node metastases in an intrarectal MNU-model, the use of shrews (phylogenetic unrelated to rodents) as test animals interdicts the comparison to rodent animal models. A further exception is the work of Derry et al[63], who could observe a relevant number of lung metastases in an AOM-induced mouse model, which is to our knowledge the only report of metastatic spread in a chemical-induced mouse model. Besides the lack of invasiveness, a lot of CIM show a high latency from the first application to tumor development. Depending on the carcinogen, dosing protocol and mouse strain, latencies from 24 to 50 wk were reported[64-66]. By combining the carcinogen with DSS, the time to tumor development can be notably shortened to 10-18 wk[23,67]. Although the minority of sporadic CRC patients show synchronous adenomas[68,69], nearly all CIM show a “carpeting” of the colonic mucosa with polyps[70]. Tumor formation is not restricted to the colon, but can be commonly observed in the whole gastrointestinal tract[71,72]. Moreover, MNU additionally induces leukemia and lung adenomas[73] and PhIP leads to formation of mammary and prostate neoplasia[74]. Another complexing aspect of CIM concerns the genetic aberrations associated with the adenoma-carcinoma sequence, i.e., the accumulation of mutations, predominantly affecting APC, KRAS and P53[75]. While the AOM model shows frequent Kras and β-catenin mutations, Apc and P53 are rarely affected[76-79]. In contrast PhIP, IQ and MNU lead to Apc mutations, but show no P53 or Kras mutations[21]. In general, the vast quantity of different dosing protocols, application forms and animal strains, makes direct comparisons and the selection of the right CIM difficult[54]. Another aspect, not to be neglected, is the agenda of animal welfare. Quite a few protocols lead to significant weight loss and diarrhea, which is, in combination with often long study durations, detrimental for the animal wellbeing[80].

With the knowledge explosion concerning the genetic pathways altered in CRC at the end of the 20^th century, the scientific community demanded specific genetic mouse models to focus on certain molecular mechanisms of colorectal carcinogenesis.

In the 1980s, the first genetically engineered mouse models (GEMM) of brain tumors, lymphoma, pancreatic cancer, breast cancer and osteosarcoma emerged[81-87]. Based on the groundbreaking work of Evans, Smithies and Capecchi[88] on gene targeting, the first tumor suppressor knock-out mouse models emerged in the early 1990s[89,90]. To circumvent the obstacle of frequent embryonic lethality caused by germline knock-outs of tumor suppressors, Cre-loxP mediated mouse models were designed to allow the tissue specific and conditional knock-out of tumor suppressor genes or activation of oncogenes, respectively[91-93]. Interestingly, the very first GEMM of CRC, the APC^Min mouse, was created without sophisticated methods. Moser et al[94] showed, that the application of N-ethyl-N-nitrosourea leads to nonsense mutations in codon 380 of the Apc gene and subsequent breeding of these animals established the first model for multiple intestinal neoplasia. APC^Min mice develop a large number of adenomas in the small intestine after 120-140 d due to LOH and show a high mortality with increasing age as a result of intestinal obstruction and anemia without progression to invasive carcinoma[95]. While these models contributed to the understanding of the early stages of FAP, they do not reflect the majority of spontaneous CRC[96]. Since a homozygous Apc mutation is lethal during embryonic development, breeding of homozygous APC^Min mice is impossible[97]. However, additional treatment of APC^Min mice with AOM or other carcinogenic compounds increases malignancy of the resulting tumors and simultaneously shortens the time to tumor development[98-100]. Till the present day, these models are in use for chemoprevention studies[101,102] and have enormously contributed to the understanding of the early tumor initiating events[103]. Interestingly, a change from C57BL/6 to a hybrid genetic background can extend the lifespan of APC^Min mice beyond one year, resulting in a high proportion of invasive adenocarcinoma[104]. Sødring et al[105] changed the APC^Min genetic background from C57BL/6 to A/J mice resulting in increased tumor formation in the colon, with a reasonable number of tumors progressing to carcinomas. Transgenic mice with alternative Apc mutations, like the APC^+/1638 mouse[106,107], the APC^Δ716 mouse[108] and the APC^Δ242/+ mouse[109] vary in tumor count and histopathology. Tumor formation predominantly in the small intestine instead of the colon is the major flaw of most Apc-based mouse models. Colnot et al[110] designed the APC^Δ14/+ mouse, which shows a more severe phenotype with invasion of the muscularis, an increased lethality and a higher colonic tumor burden compared to APC^Min mice but unfavorably still shows relevant tumor development in the small intestine. Early attempts of combining Apc mutations with homozygous P53 knockouts yielded conflicting results, with either no increase[111,112], or a small increase of gastrointestinal malignancy[113]. The most likely explanation is that in human cancers P53 missense mutations frequently act in a dominant negative fashion, overruling the tumor-suppressive function of the wildtype allele[114,115]. In contrast, targeting Kras without tissue specific promoters leads either to embryogenic lethality[116] or few viable animals succumbing to rapidly developing lung tumors[117]. To avoid abundant distribution of mutations in the whole organism, transgene expression controlled by a tissue specific gene promotor, most commonly by application of the Cre-loxP-system[91], has proven to be extraordinarily useful. Many workgroups used Villin-Cre transgenes to restrict recombination of floxP-flanked genes to the epithelial cells of the small and large intestines either with a constitutive expression (Vil-Cre) or with a tamoxifen-inducible expression (Vil-Cre-ER^T2)[118]. Another option of site-specific Cre expression is the fatty acid binding protein liver Cre transgene (Fabpl Cre), which can be combined with a tetracycline-inducible tet-on system[119]. Yet, both transgenes’ expression is not limited to the large intestine: While Villin-Cre is expressed in the epithelial cells of the whole intestines, Fabpl-Cre expression can be detected in the distal small intestine, cecum and colon[119,120]. Also, the AhCre strain, carrying Cre under control of the Cyp1A promotor, is commonly used for colon cancer models[121,122]. Here, Cre-expression is induced by β-naphthoflavone in the liver and intestines, but constitutive recombination can be observed in other tissues like the renal epithelium[123]. To achieve a more colon specific expression of Cre-recombinase, Hinoi et al[124] constructed a transgene of Cre and the promotor region of the CDX2 homeobox gene. By inserting a guanine repeat tract to this transgene (CDX2P9.5-G22Cre), stochastic activation of Cre-expression by means of spontaneous frameshift mutations leads to a mosaic-like recombination in the mucosa of the terminal ileum, cecum, and colon[125]. At last, carbonic anhydrase 1 promoter/enhancer-Cre recombinase transgene (CAC) facilitates recombination strictly limited to the large intestine[126]. Paralleling Cre-transgene implementation, others achieved spatiotemporal oncogene expression in the large intestine by delivering Cre by viral transfection via transanal injection, surgery or colonoscopy[127-129] leading to exquisite models of CRC with metastatic spread[130]. Supplied with this comprehensive genetic toolbox, a plethora of CRC mouse models were generated and used to evaluate the role of different mutations and their interplay. Among the non-hypermutated tumors, the most frequently mutated genes are APC, P53, KRAS, PIK3CA, FBXW7, SMAD4, and TCF7L2[13]. The role of Apc LOH as a driver mutation is highlighted in several mouse models. While generalized deletion of both Apc alleles (Apc^flox/flox) leads to rapid death by disorder of cell differentiation[131], mice with a mosaic–like deletion of both Apc genes die rapidly from florid polyposis[125]. Depending on the type and modality of Apc mutation, heterozygotes develop adenomas or invasive adenocarcinomas[120,132,133]. Whilst mutated Kras alone is insufficient to induce colorectal tumorigenesis, it increases the susceptibility of the intestinal mucosa to chemical carcinogenesis[134] and leads to accelerated tumor formation in combination with Apc loss[122,135]. Nras mutation, in contrast, does not alter the effect of Apc loss[136]. In humans, P53 mutations are often associated with vascular and lymphatic invasion and advanced cancer stages[137,138]. In fact, the combination of Apc mutation with a dominant-negative P53 mutation leads to increased invasiveness of intestinal tumors with signs of epithelial to mesenchymal transition[139,140]. Also, loss of P53 in a constitutively active Notch signaling background leads to intestinal tumor formation and metastasis[141], whereas Notch signaling does not cooperate with the Wnt-pathway[142]. Nevertheless, Apc deficiency seems to represent a key prerequisite of cancer progression, since Apc restoration leads to spontaneous tumor regression of Kras-mutated, P53-deficient adenocarcinomas[143]. FBXW7 codes for the F-box/WD repeat-containing protein 7, the substrate receptor of a ubiquitin ligase, which degrades several proto-oncogenes like MYC, CCNE1, NOTCH1 and JUN and acts synergistically with P53 as haploinsufficient tumor suppressor[144-146]. Intestinal Fbxw7 deletion enhances tumor development in an Apc^Min/+background[147] and a combined deletion of Fbxw7 and P53 results in highly aggressive intestinal cancers with metastatic spread to the lymph nodes and liver[148]. The PI3K/AKT- pathway is well known for its pro-oncogenic and anti-apoptotic signaling and PI3K mutations are common in CRC and many other human cancers[149]. As demonstrated by Leystra et al[150], the intestinal expression of a constitutively active Pi3k (PIK3ca) is a sufficient driver mutation to induce rapid tumorigenesis with invasion of adjacent organs and addition of PIK3ca to a homozygous loss of Apc, drives adenoma-to-carcinoma progression with metastatic spread[129]. Although loss of Pten, the counterpart of Pik3, does not affect intestinal cell proliferation; in the context of Apc deficiency or other mutations, it promotes cancer progression[151-153]. SMAD4 is considered as a tumor suppressor, similar to other constitutes of the TGFβ-pathway[154,155]. Since the genes Apc, Smad2 and Smad4 are all located on chromosome 18 in the mouse, they are suited to generate distinct cis- and trans compound heterozygotes by meiotic recombination. Compared to their single mutation littermates, mice with combined Apc and Smad4 mutations, show accelerated tumor development[156] and increased malignancy[157,158]. In contrast, compound heterozygotes of Apc and Smad2 mutations show no increased tumor development compared to littermates with a single Apc mutation[159]. Notably, homozygous Smad3 mutation leads to aggressive CRC with lymphatic spread and, upon Apc deficiency, drastically reduced life span[160,161]; but SMAD3 is rarely mutated in human CRC[162]. Findings from several CRC-GEMM highlight the role of TGFβ-signaling as a strong tumor suppressor, since Tgfβ[163-165], as well as Tgfβ-receptor 2[140,166,167] knockout, induce local invasion and metastatic spread. Regarding the role of TCF7L2 mutations in CRC, so far, no GEMM of CRC addressing this topic have been published. Besides the conventional adenoma-carcinoma sequence, the serrated pathway represents an alternative route of CRC development with distinct molecular and clinical features. The underlying BRAF^V600E mutation occurs in 15%-20% of sporadic CRC, causes a constitutive activation of the MAPK/ERK pathway and is strongly associated with the CpG Island methylator phenotype and microsatellite instability (MSI) due to MLH-1 methylation[168-170]. Braf^V600E causes crypt hyperplasia and combined with Apc or P53 mutations, as well as mutations affecting Ink^4A/Arf, gives rise to invasive carcinomas[171-174]. Although not common in human CRC, mutation of the GSK3-β phosphorylation site causes degradation-resistant β-catenin[175,176], and has been remodeled in the mouse. GSK3β-resistant β-catenin increases proliferation of the intestinal epithelium and causes adenoma formation, but does not mediate malignant progression[177-179]. Paralleling the research on canonical cancer pathways, there has been reasonable effort to reproduce MSI, a hypermutable phenotype caused by malfunction of DNA mismatch repair (MMR) enzymes[180,181]. MSI-CRC can occur in the context of hereditary MMR gene mutations (LS) or can be detected in up to 15% of spontaneous CRC, caused by hypermethylation of MMR genes (spMSI)[182,183]. The constitutes of the MMR machinery have been extensively studied by somatic knockout models. Since the knockout of the MMR genes MLH1[184], MSH2[185,186], MSH6[187] and PMS2[188] predominantly cause hematopoietic malignancies with consecutive reduced lifespan in the mouse and consequently only a minor fraction of homozygotes develop intestinal neoplasia[189,190], these knockouts were frequently put in an APC deficient setting, to increase intestinal carcinogenesis[184,191,192]. Kucherlapati et al[193,194] demonstrated, that mutations of Fen1 and/ or Exo1, both cooperation partners of MMR enzymes, lead to similar patterns of MSI tumor development. However, MSI-high tumors rarely show activation of the Wnt-pathway and are typically chromosomal stable, whereas APC mutations are typically associated with a chromosomal instable phenotype[195,196]. Therefore, these models do not adequately recapitulate Lynch-type or spMSI tumors. To overcome the aforementioned obstacles, a floxed Msh2 allele was combined with either a LS related missense mutation (Msh2^G674D) or an Msh2^Δ7 null mutation in mice carrying a Villin-Cre transgene, leading to intestinal carcinogenesis and chemoresistance typical for MSI-high tumors[197]. An overview of the above addressed mutations can be found in Figure 1 and are summarized in Table 1.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Overview of the frequently altered pathways in colorectal cancer. The numbers in square brackets label the corresponding model descriptions as given in Table 1.

In summary, GEMM have contributed enormously to the understanding of the molecular processes of CRC initiation, progression and crosstalk of common cancer-associated pathways. Besides the models for spontaneous CRC and common cancer syndromes, like Familiar Adenomatius Polyposis and LS, several models recapitulate metastatic disease, either as “classic” GEMM[141,148,166,198] or upon viral Cre delivery[129,130]. Yet, these models have several limitations. First of all, cancer development is a stepwise process with an initial driver mutation and subsequent acquisition of further mutations[199], and thus can only be partly reflected in tumor mouse models by the combination of a constitutively active with an inducible mutation[172]. Roper et al[128] recently demonstrated in vivo genome editing of Apc and P53 by viral delivery of the correspondent sgRNA in mice expressing CRISPR-Cas9 under the control of a Villin-Cre transgene. Second, the number of combined mutations is limited in the mouse, since the resulting phenotype shows often a drastically reduced lifespan[152]. Triple mutant (Car1^CreER/+; Apc^fl/fl; Kras^LSL–G12D/+) and quadruple mutant mice (Car1^CreER/+; Apc^fl/fl; Kras^LSL–G12D/+; P53KO; Smad4^fl/fl) showed a reduced lifespan of merely one month. Moreover, GEMM are time consuming and expensive, since breeding of the transgenic mice often takes multiple generations and requires careful interbreeding to yield the desired alterations. In terms of animal welfare, it should be noted that the breeding process yields many “reject” mice, which are neither used for further breeding nor for research. Also, the construct of the transgene, or the viral vector, respectively, is complicated. While GEMM represent a valuable tool for basic research, their use for preclinical studies is limited due to a lack of genetic heterogeneity on the one hand, and discrepancies to the human tumor development on the other hand. On a final note, it should be added, that GEMM of CRC can be applied the other way round: Mice harboring mutagenic SB transposons were crossed to mice expressing SB transposase under control of a Villin-Cre-transgene, to generate mice, that develop intestinal lesions due to random insertional mutagenesis and can be screened for unknown CRC driver mutations[200]. Moreover, by combining this approach with well-known driver mutations, new pathway-associated mutations could be identified[201,202].

Transplant mouse models can be classified in various ways: Syngeneic tumor transplantation is characterized by tumor tissue or cancer cell line engraftment within the same mouse strain; whereas xenogeneic grafts are derived from a different mouse strain or human donors. Additionally, it can be distinguished between heterotopic and orthotopic models. As cancer grafts, tumor cells, organoids and tumor tissue pieces can be employed. In 1876, Novinsky successfully transferred tumors (likely canine venereal sarcomas) from one dog to another in two independent trials[203]. Shortly thereafter, Hanau[204] and Morau (reviewed in[205]) reported independently the successful passaging of epithelial rodent tumors. Around the turn of the century, several transplantable rodent tumors, mostly of a sarcomatoid phenotype, were established[206]. Later, the discovery of the human leukocyte antigens and oncogenic viruses elucidated the results of the earlier transplantation studies[207,208]. Toolan pioneered in the xenograft field by attenuating graft rejection through X-radiation and cortisone treatment[209,210].

PDX models

PDX differ greatly from the aforementioned heterotopic tumor models. They are established by the s.c. implantation of a tumor piece from a surgical resection or biopsy into the flanks of immunocompromised mice; with lower tumor take rates for biopsy samples[237]. Tissue can either be implanted directly after resection or cryopreserved in fetal calf serum containing 10% DMSO for implantation at a later time[229]. Incubation of the tissue in Matrigel^® prior to tumor implantation, significantly increases engraftment rates[228]. In order to obtain sufficient tumor tissue for larger scale studies, the resulting tumor can be fragmented and re-grafted subsequently. In recent years, our workgroup focused on the build-up of a large CRC biobank consisting of more than 140 PDX-models (general procedure is presented in Figure 2). This PDX panel reflects adequately the clinical and molecular heterogeneity of the patient population undergoing surgical resection of primary or metastatic CRC[252]. It is well accepted that PDX closely recapitulate the histology of the original “donor” tumor over several passages[253-255] and are also genetically stable[256,257]. However, Ben-David et al[258] demonstrated recently that changes in copy number alterations occurred in early passages of PDX tumors compared with both the P0 PDX and the donor tumor. A common criticism in heterotopic mouse models is related to the absence of tumor-stroma interaction or the “tumor microenvironment”[259]. While this is true for immune cells, the stromal component remains intact in PDX[255]. Although the human stromal compounds (fibroblasts, blood vessels etc.) are quickly and steadily replaced by their murine counterparts, the overall architecture of the tumor remains unaffected in the majority of cases[260,261]. Moreover, these murine stroma cells adopt and maintain a human-like metabolic phenotype[262]. These features indicate that PDX are indeed good and valuable models for preclinical testing of conventional and novel anticancer agents in the era of personalized cancer therapy[263,264]. A proof of concept study with advanced refractory cancers demonstrated that drug responses measured in a PDX model, can be used for successful clinical decision making[265]. A PDX model of CRC metastases closely resembled the efficacy of cetuximab and identified druggable targets in resistant tumors[266]. Moreover, a PDX clinical trial approach (one animal per model per treatment) reflects well the heterogeneity of the patient population and allows testing of new drugs and combinatorial regimen[267]. A large study with over 1000 PDX models confirmed the consistency between clinical and PDX clinical trial drug response[268]. The ultimate goal of precision medicine would comprise of the profound genetic and functional characterization of a given tumor to identify relevant drug targets and subsequent validation of potential therapies with aid of a PDX bearing “avatar” mouse to provide the most efficient treatment for the individual patient[264,269]. Currently, several clinical trials following this approach for colorectal[270,271] and pancreatic cancer are recruiting[272]. Yet, a median duration of 12.2 mo until PDX model establishment[237] remains an unsolved issue for patients urgently in need for treatment. When it comes to drug testing, it should be considered, that immunotherapy approaches can only be restrictedly evaluated in immunodeficient host mice. A further disadvantage of PDX models is a potential selection for more aggressive tumors. The data concerning the association between successful PDX engraftment and clinical or molecular features is in part conflicting. While we and others could not find significant associations between tumor grading and PDX engraftment[237,252], a Korean study observed significant correlations with tumor staging and grading[273]. A smaller study with a high PDX establishment rate found a significant correlation of PDX success with positive nodal status and grading[274], while Julien et al[275] only found significances for the combination of a positive nodal status with an elevated carcinoembryonic antigen level. Moreover, we observed a significant correlation between PDX engraftment and molecular features like KRAS and BRAF mutations as well as MSI[252]. Apart from the tumor biology, choice of the host mouse strain, repeated attempts of engraftment, quantity and quality of the resected tissue as well as previous treatment of the patient are additional factors influencing the success of model establishment. Collins et al[276] recently reported PDX engraftment rates varying from 14 to 100% with a median PDX establishment rate of 68% reviewing 14 CRC-PDX studies. Compared to establishment rates for primary CRC cell lines of about 10%, the PDX is clearly superior[252,277]. A selection bias relating to cancer biology can be diminished by increasing the number of enrolled patients and molecular characterization of the individual PDX. In fact, a very recent PDX study found an underrepresentation of the consensus molecular subtype number 2, due to worse engraftment[278]. Another crucial pitfall of the PDX model is the development of EBV-associated lymphomas at the implantation site, which can mimic successful engraftment[279]. Depending on the mouse strain and cancer entity, between 2.3% (colorectal) and 75% (prostate cancer) of primary engrafted bona fide xenografts turn out to be human de-novo lymphomas[280-284]. Since this condition is more frequently reported in NSG and comparable strains, development of de-novo lymphoma in NMRI^nu/nu mice might be hindered by high NK cell activity[285]. Thus, after successful engraftment of a PDX in NSG mice, we conduct the subsequent passaging in NMRI^nu/nu mice and xenograft histology is frequently evaluated by an experienced pathologist. Interestingly, Butler et al[286] significantly reduced the frequency of lymphomas in an ovaria cancer PDX model by a unique dosage of rituximab during implantation. Additionally, PDX serve not only in the field of therapy development; they are also a vital tool for the maintenance of a healthy biobank, allowing the establishment of secondary cell lines and supply sufficient tumor samples for the exchange with other work groups[252].

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Establishment of PDX. After surgery, a small sample of the tumor, which is not needed for pathological diagnosis, is obtained and cut into pieces of 27 mm³. These can be either implanted immediately in recipient mice or vitally cryopreserved in liquid nitrogen. The resulting patient-derived xenograft (b) closely reflects the histology of the donor tumor (a) (Previously published in[229]). Patient-derived xenograft can be further processed for subsequent implantation or cryopreservation.

In summary, we consider the PDX model as keystone of cancer research, holding great potential in the developing field of precision medicine for CRC.

The increasing field of precision medicine has a growing need for highly translational cancer models. Conversely, the increasingly negative public perception of animal studies constrains the scientific community to further stress the 3R-principle (replacement, reduction and refinement) in cancer research[335]. Aside from the improvement of in vivo models, this implies the refinement of in vitro methods as well.

Humanized mice are severely immunocompromised mice, which can be reconstituted with various types of human bone marrow-derived cells or CD34⁺ hematopoietic stem cells[336,337]. Since human stem and progenitor cells can be attained from umbilical cord blood or from peripheral drawn blood samples after GM-CSF treatment and cultured in vitro[338], these cells can be transferred to sub-lethally irradiated NSG mice. Morton et al[339] observed that PDX of head and neck cancer engrafted into these, so called “Xact mice”, are infiltrated with human B and T lymphocytes. Many transgenic mice further support engraftment with CD34⁺ stem cells by overexpression of human interleukins and signaling molecules. Thus NBSGW[340], hIL2-NOG[341], NSG-SGM3[342] and SRG-15 mice[343] have been introduced recently and antitumor effects against PDX of different cancers could be observed[341,343]. These models hold great promise for the research of CRC immunotherapy, especially for highly immunogenic hypermutated CRC. Capasso et al[344] showed very recently that check point inhibition with nivolumab leads to growth inhibition of human MSI-H PDX thereby accurately reflecting the clinical response of this CRC subtype. In contrast, no sustainable growth inhibition was observed in MSS tumors or MSI-H tumors in “standard” NSG mice[344,345].

Although patient-derived cell cultures are a valuable tool for high-throughput drug screenings, they exhibit considerable shortcomings[346]. First, the establishment rate of primary patient-derived CRC cells with conventional 2D culturing methods approximates some 10%[252,275], although higher success rates of 40% and more can be found in the literature[347,348]. Second, conventional 2D cultures change the biological properties of cells, possibly altering drug response in vitro[349]. Cell polarization, lack of stroma and abundance of growth factors, nutrients and oxygen are factors that might change the behavior of tumor cells[350]. In recent years, more complex cell culturing methods have emerged. CRC cells, cultured in an extracellular matrix form three dimensional spheres, so called spheroids, that differ in their biological properties from 2D cultured cells[351]. In contrast, organoids are three dimensional structures, derived from intact tumor pieces or tumor stem cells cultured in an extracellular matrix scaffold[352]. Patient-derived organoids (PDO), quite similar to PDX, recapitulate closely the histological and genetic properties of their parental tumors. Moreover, high rates of successful establishment have been reported[353] and reliable drug response prediction seems possible[354-356]. PDOs can be implanted s.c. or orthotopically into mice, resulting in PDO xenografts (PDOX)[264,354,357]. Furthermore, the Clevers group pioneered in the inauguration of non-malignant intestinal organoids exploiting the stem cell niche in vitro[358]. These organoids can be modified in vitro to exhibit malignant properties and used to enlighten the role of cancer-driving pathways by in vivo engraftment[359,360]. The available data strongly suggests, that PDOs reflect more faithfully the biological virtue and drug response of the parental tumor compared to conventional 2D cell cultures[352]. The circumstance, that organoids can be derived from CTC, could render them an excellent tool for the preclinical testing of patients with advanced stage cancer that do not undergo surgery[361]. Additionally, CTC reflect genetic changes associated with acquired drug resistance during chemotherapy[362]. Further steps to a reduction of animal experiments imply the faithful remodeling of the host organism in vitro. Several research groups created a “cancer on a chip” model that combines the advances of 3D cell culture connected with artificial organs that resemble the most common organs of metastatic spread[363]. Miller and Shuler introduced a “body on a chip model” with 14 artificial organs, which could be modified for cancer research[364]. At last, the widely acknowledged work of Guinney et al[365] regarding the consensus molecular subtypes of CRC draw great attention to the value of computed models in cancer science and the capabilities of bioinformatic research. Retrospective analysis of clinical trial samples partly demonstrated the association of drug response with molecular subtypes[366]. The constantly growing knowledge of cancer pathways and their crosstalk on the one hand, and the increased inter-individual complexity of tumors on the other hand, call for a method to integrate and interpret the overwhelming amount of data[367]. In silico methods, like data mining, pattern recognition, machine learning and network approaches, are able to predict the behavior of “virtual” HCT116 cells[368], can reveal genetic patterns associated with survival[369], can be used to detect new biomarkers[370], allow the identification of unknown driver mutations[371] and potential preclinical compounds[372]. Yet, in silico models often lack explanatory power and need careful interpretation by bioinformaticians. Their ability to correctly predict treatment response for an individual patient to a new compound is still a long way off[373].

We kindly thank Jenny Burmeister, graphical assistant, for the supply with excellent pictures.