Non-malignant tissues are often overlooked as samples for establishing PDXs. However, they can be used to optimise xenografting techniques, study angiogenesis and the interactions between epithelium and stroma, investigate the normal physiological responses of tissues to treatment, compare the features of patient-matched benign and malignant tissue and identify cancer cells of origin [6, 10, 20–24]. Non-malignant tissue is often dissected from the same surgical specimen as the tumour by sampling regions that are distant from known tumour foci [10, 23, 25]. Other possible sources of non-malignant tissue include prophylactic surgeries, such as mastectomies and oophorectomies from women with pathogenic BRCA1 or BRCA2 mutations, or procedures for benign conditions, such as transurethral resection of the prostate (TURP) for treating benign prostatic hyperplasia [8, 24, 26].

Depending on the experiment and tumour type, thorough histopathology may be required to confirm that these specimens are truly non-malignant. Furthermore, some cancers are thought to exert a field effect on surrounding tissue, so non-malignant samples may be best defined as “benign” or “morphologically normal”, rather than “normal” [27, 28]. Notwithstanding this limitation, non-malignant samples are still useful because they often have high take rates as primary PDXs [10, 25]. Anecdotally, benign epithelial cells often persist in primary PDXs, even when cancer cells fail to grow. However, unlike tumours, non-malignant tissues do not produce serially transplantable PDXs. Collectively, this means that non-malignant specimens are a convenient source of tissue for short-term PDX experiments as long as their histopathology is carefully reviewed.

Localised tumours are a common source of tissue for PDXs, because surgery with curative intent is standard practice for treating many cancers. These specimens can provide large amounts of tumour tissue from each patient, and it is sometimes possible to obtain locally advanced cancer from surrounding lymph nodes. Regions of tumour tissue can be dissected from surgical specimens with a scalpel or biopsy needle, ideally avoiding benign or necrotic tissue [5, 29]. Whichever method is used, it is essential that the overall architecture of the specimen is preserved for routine pathology reporting of surgical margins, tumour differentiation and histology [5]. PDXs can also be established from primary tumours after patients have received neoadjuvant treatments [30, 31]. In cases where surgery is not performed, tissue can be obtained from biopsies, including fine needle aspirates [10, 13, 32, 33]. Thus, for many cancers localised tumours are a widely available source of tissue for xenografting.

PDXs of primary tumour have many applications. Large biobanks of PDXs have been established from primary tumours of numerous cancers [1–3, 34]. Genomic analyses of these large cohorts have shown that they approximate the inter-patient diversity of tumours in the clinic [1, 2]. Thus, they provide comprehensive preclinical platforms for drug screening. Serially transplantable PDXs of primary cancers are also useful for studying tumour biology. For example, some laboratories have grown PDXs under selective pressure to create models of therapy resistance [13], while others have established PDXs from matched localised and metastatic tumours from the same patient [1, 32]. Therefore, even though PDXs of primary tumours usually do not represent lethal disease, they are invaluable models for cancer research.

PDXs of metastatic tumours provide models of the most aggressive stages of cancer progression, including therapy resistance [14, 35]. This makes PDXs of metastatic tumours ideal for studying mechanisms of drug resistance and for testing the efficacy of novel therapeutics. For many cancers, however, there is limited access to metastatic tumours compared to localised disease. Patients with some cancers rarely undergo surgical resection of metastases and are instead treated with radiotherapy or systemic therapies like chemotherapy. Nevertheless, once the logistical and ethical challenges are overcome, it is still possible to obtain metastatic samples from patients during treatment or after death.

PDXs can be established from numerous sources of metastatic tumour cells from patients who are still undergoing treatment. For example, several laboratories have generated PDXs from surgically resected liver metastases of colon cancer [34, 36–38]. Surgery is less commonly performed on metastases of many other cancers; however, other types of samples are sometimes available. This includes malignant ascites or pleural effusions, from which cancer cells can be isolated and injected or grafted into mice [17, 39–42]. Metastatic tissue can also be obtained from palliative surgeries, which are sometimes used to alleviate pain or repair fractures due to bone or spinal metastases [43–45]. Biopsies are another common source of metastatic tumour tissue [46, 47]. They are usually performed during the course of patient treatment but are sometimes undertaken specifically to obtain tissue for research or clinical trials, which carries a small risk of complications to the patient [48–50]. This means that it is sometimes possible to plan the timing of biopsies and even use them to obtain serial samples from patients [49]. However, the limitations of biopsies include the small amounts of tissue they provide and the inability to sample some metastatic sites. Nevertheless, along with surgical resections, ascites and pleural effusions, biopsies are an essential source of metastatic tumour cells for establishing PDXs.

Certain limitations of collecting samples from living patients are overcome with rapid autopsy programmes. Rapid autopsy, also known as warm or immediate autopsy, involves the collection of tumour tissue within a few hours of a patient’s death [51]. The speed of this process is important for maintaining high-quality, viable tumour tissue before autolysis occurs. This creates logistic challenges, so rapid autopsy programmes typically involve team members from clinical care, palliative care, funeral services, forensic medicine, tissue banking and cancer research [14, 52]. Most rapid autopsy protocols use imaging and clinical notes to identify the locations of metastases. These sites are then reviewed macroscopically during dissection to avoid any necrotic tissue [14, 52]. Despite concerns about the viability of rapid autopsy samples, serially transplantable PDXs have been established from melanoma, rhabdomyosarcoma and breast, pancreatic, prostate and ovarian cancers [14, 35, 51–53]. The reported success rate for generating PDXs varies between tumour types, from 5% for prostate cancer to 100% for melanoma [14, 52].

There are benefits and limitations to rapid autopsy programmes and the samples they provide. One of the benefits of rapid autopsy is that it enables extensive sampling of multiple metastatic sites, including those that cannot be accessed before death [14]. This is particularly useful for studying intra-patient tumour heterogeneity [54–56]. Furthermore, compared to biopsies, rapid autopsies can provide greater amounts of tissue from more sites and without patient discomfort or the risk of complications. Yet, rapid autopsies are not routine. This makes them a low-throughput source of metastatic tissue for xenografting, especially compared to biopsies. Therefore, rapid autopsy programmes are often used for detailed studies of carefully selected patient cohorts, where samples are gradually accumulated over time. Overall, this means that rapid autopsies and other sources of metastatic tissue are complementary methods of collecting patient samples for xenografting.

Circulating tumour cells (CTCs) are an emerging source of cancer cells for PDXs. Successful PDXs have been established from breast, prostate and small-cell lung cancer CTCs directly implanted into immunocompromised host mice [57–59]. PDXs have also been generated from CTCs that were cultured as organoids before engraftment (see Section “Sorted or Cultured Cancer Cells”) [60–62]. CTCs have yielded tumours when directly injected into subcutaneous and bone sites, which is often cited as evidence that CTCs contain a subpopulation of metastasis-initiating cells [57, 58, 63]. Not surprisingly, samples with greater numbers of CTCs have higher take rates when xenografted [57, 58]. This could be due to the increased likelihood that some of the CTCs within a sample will be tumourigenic, as well as the association between high CTC counts and aggressive tumours. Since the process of collecting CTCs from blood or “liquid biopsies” is non-invasive, it might be possible to obtain serial samples to establish PDXs from different stages of disease progression, such as before and after therapeutic resistance. Therefore, CTCs are likely to become increasingly popular samples for establishing PDXs.

PDXs are usually established from samples of intact or digested patient tissue; however, they can also be grown from tumour cells that are preselected using sorting or primary culture. Cell sorting is used to enrich defined populations of cancer cells based on their expression of cell surface antigens or phenotype. Common techniques include flow cytometry, magnetic bead separation and differential attachment to coated plates. Xenografts of sorted cells have primarily been used to study cancer-repopulating cells [37, 64, 65]. The frequency of cancer-repopulating cells can be calculated by decreasing numbers of sorted cells to establish a tumour [66]. The other common method of preselecting cancer cells, primary cell culture, is also based on the premise that only a subpopulation of cancer cells may have the potential to form tumours. In this way, in vitro cell culture can be used to select patient specimens that are more likely to establish successful PDXs. These primary cultures are increasingly being established as organoids rather than adherent or suspension cultures [60–62, 67]. This method is particularly useful for samples with low take rates, because it is easier to monitor their growth in vitro and then subsequently graft the cells into host mice. This also provides matched in vitro and in vivo models to study tumour biology and drug responses in various contexts. An unresolved question is whether there are differences in PDXs established from pieces of tissue compared to cultured cells, other than the obvious lack of human stroma in early generations. Nevertheless, preselecting cancer cells through sorting or primary culture provides researchers with a way to control the success rate of PDXs.

It is essential to use immunocompromised host mice to avoid rejection of PDXs. Early methods of supressing the host immune system included X-ray irradiation and thymectomy [74, 75]. The subsequent discovery of athymic nude mice (nu/nu), which lack functional T cells, foreshadowed the use of increasingly immunocompromised mouse strains for xenografting [76]. Many laboratories then shifted to using severe combined immune-deficient (SCID) mice, which are deficient in both T and B cells [77]. Non-obese diabetic SCID (NOD-SCID) mice then became more popular for xenografting, because they avoid the leaky phenotype of SCID mice and also have impaired natural killer (NK) cell function [78, 79]. To further abrogate the host immune response in these strains, some laboratories pretreated mice with etoposide, an immunosuppressant, a few days prior to grafting [39, 80]. It was subsequently shown, however, that etoposide decreased the take rate and growth of breast cancer PDXs in thoracic fat pads [39].

Most recently, NOD-SCID interleukin-2 receptor gamma chain null (NSG) mice have become a common strain used for xenografting. NSG mice are highly immunocompromised because they lack functional T, B and NK cells [81, 82]. It is often assumed that using highly immunocompromised host mice can improve the take rate of PDXs; however, it seems to depend on the patient specimens. Small numbers of melanoma cells have dramatically higher take rates in NSG versus NOD-SCID mice, whereas there is no difference in the engraftment rate of breast cancer tissue in SCID versus NSG mice or prostate cancer tissue in nude versus NOD-SCID mice [11, 42, 65]. Nevertheless, many laboratories now routinely use NOD-SCID or NSG mice for xenografting given that the take rate is greater, or at least equal, to that obtained with other strains. The trade-off for high engraftment rates with immunocompromised mice is the inability to study the interactions between tumour and immune cells, leading to increasing interest in humanised PDX models [83].

In addition to the immune system, steroid hormone levels are another systemic feature of host mice that affects the engraftment and growth of hormone-dependent cancers. Breast cancer PDXs are often established in female mice implanted with oestrogen pellets. Oestrogen supplementation increases the engraftment of both oestrogen receptor-positive and receptor-negative subtypes, presumably due to paracrine signalling from the stroma [17, 42, 84, 85]. Unfortunately, higher doses of exogenous oestrogen can cause side effects in host mice, such as urine retention and hydronephrosis, leading to the development of alternative protocols with lower oestrogen levels [26, 86, 87]. Similar to breast samples, prostate tissue is grafted into male mice with testosterone implants. Higher testosterone levels are required to maintain the differentiation of benign and malignant prostate glands in PDXs, because androgen levels in adult male mice are only equivalent to hypogonadal adult men [10, 24, 88, 89]. Therefore, supplementing steroid hormone levels in host mice is a simple way to maximise the success of PDXs of hormone-dependent cancers.

The graft site is an important consideration when establishing PDXs, because local features of the mouse host affect the take rate, fidelity and practicality of PDXs. Specimens are often engrafted heterotopically, that is, at a different site compared to their tissue of origin. Subcutaneous grafting into the shoulder or flank of host mice is particularly common. The advantages of subcutaneous grafting include the speed and technical simplicity of the method, the ability to graft large specimens and the ease of monitoring tumour growth [90]. Thus, it is ideal for tumours with high take rates or for serially transplantable PDXs that have already been established at other graft sites. Tumours with low take rates as subcutaneous grafts, such as oesophageal, prostate and low-grade ovarian cancer, often grow more successfully as subrenal capsule or intramuscular grafts [10, 11, 91, 92]. This is thought to be due to the greater blood supply at these sites. Subrenal grafting requires more expertise than subcutaneous grafting; however, once the grafts are implanted, they are held in place by the renal capsule, a thin membrane surrounding the kidney [5, 10, 89, 93]. Another limitation of subrenal capsule grafting is that it is more difficult to estimate tumour growth by palpating the grafts. Therefore, for heterotopic PDXs, the choice of graft site depends on the balance between optimal take rate and ease of grafting and monitoring of tumour growth.

PDXs can also be established orthotopically [10, 17, 28, 42, 90, 94–97]. It can be more difficult and time-consuming to establish orthotopic grafts, and they may need to be monitored using specialist imaging equipment [98]. Nevertheless, orthotopic grafting ensures that the host microenvironment mimics the patient tissue of origin as closely as possible. This is reflected in the high take rates of orthotopic grafts for many tumour types, compared to subcutaneous grafts [10, 17]. The graft site can also affect the phenotype of PDXs. For example, orthotopic PDXs may be more likely to metastasise [95, 99, 100] and more closely reflect patients’ responses to therapy than heterotopic PDXs [101]. Yet, most PDXs closely recapitulate the original patient tumour [2], so the degree to which orthotopic grafting improves fidelity is unclear and might depend on the tumour type. The precise anatomical site used for orthotopic grafting can also be important. For example, breast cancer PDXs exhibit better engraftment and growth in abdominal compared to thoracic fat pads [39]. Furthermore, injecting oestrogen receptor α-positive tumours into milk ducts instead of fad pads produces PDXs that more closely resemble the histopathological features of patient tumours [26]. Collectively, these studies emphasise that the local microenvironment can affect the take rate and phenotype of PDXs.