Intratumoural heterogeneity is evident in human cancers and most likely contributes to differing chemotherapeutic responses. Hence, to improve cure rates, an understanding of the contribution by intratumoural heterogeneity to drug resistance is essential. The contributors to intratumoural heterogeneity are genetic variation, stochastic processes, microenvironment and cell and tissue plasticity (Fig 1). The evidence for the role of each of these is discussed below.
Genetic variation and heterogeneous intratumoural drug responses
The introduction of many new targeted therapies to clinical practice provides support for the role of intratumoural heterogeneity in the loss of drug sensitivity. For example, a number of studies have recently reported the use of comparative genomic hybridization analysis and next generation sequencing to analyse individual tumour cells isolated from primary breast cancers (Navin et al, 2011), pancreatic adenocarcinomas (Ruiz et al, 2011), acute myeloid leukaemia (Ding et al, 2012) and renal cell carcinoma (Gerlinger et al, 2012; Xu et al, 2012). There is now definitive evidence showing that primary human tumours contain genetically distinct subpopulations of tumour cells. In primary breast cancers, glioblastomas, melanomas and renal cell carcinomas, clonal variants not only exist within tumours but are also confined to different sub-anatomic sites within the tumours (Gerlinger et al, 2012; Navin et al, 2011; Snuderl et al, 2011; Takata et al, 2000). Moreover, exome sequencing of single cells isolated from a renal cell carcinoma showed that only 30% of the genetic lesions within a tumour are common to all the cancer cells (Gerlinger et al, 2012; Xu et al, 2012). Finally, several independent studies have demonstrated that multiple clonal variants exist within established head and neck cancer cell lines (Cameron et al, 2010; Erlich et al, 2012; Poth et al, 2010). Significantly, these variants differed with respect to their transcriptomic profile, their sensitivity to chemotherapy, their ability to initiate tumours, and their ability to interact with one another to initiate tumours (Cameron et al, 2010; Erlich et al, 2012; Poth et al, 2010). Combined, these studies unequivocally show that genetically distinct variants of tumour cells exist within individual tumours in multiple tumour types.
There is increasing evidence demonstrating a role for intratumoural heterogeneity in drug resistance. Many patients have an immediate response to conventional cytotoxic therapies, which can be followed by recurrence and resistance to rechallenge with the same chemotherapeutic agents (DeVita & Chu, 2008; Garraway & Janne, 2012). In some instances, a relapsed tumour may be sensitive to a different chemotherapy protocol and thus patients may undergo multiple cycles of differing chemotherapeutic cocktails in pursuit of a sustained response (DeVita & Chu, 2008; Garraway & Janne, 2012). Similar clinical scenarios have been observed with the newer targeted therapies. For example, the first generation BCR/ABL kinase inhibitor, imatinib, or the V600E mutant-specific BRAF inhibitor, Vemurafenib produce profound initial responses in patients followed in many instances by the development of resistance (Flaherty et al, 2010; Rosti et al, 2012; Villanueva et al, 2010). In chronic myleogenous leukaemia, imatinib resistance is frequently associated with tumour cells that no longer harbour imatinib-sensitive mutations in the BCR-ABL kinase (Garraway & Janne, 2012; Michor et al, 2005). Switching patients to second-generation drugs with broader specificity, such as dasatinib can overcome this resistance (Rosti et al, 2012). In the instance of BRAF V600E mutant-specific therapies, resistance arises in a sub-population of cells in which IGF1 receptor signalling has been deregulated (Villanueva et al, 2010). Similarly, the recent trial of a vaccine against an EGF receptor mutation in glioblastoma demonstrated a similar transient response (Sampson et al, 2010). These trials clearly show the clinical effectiveness of targeted therapies. However, they also show that a paradoxical weakness of targeted therapies may be the highly selective nature of their action. Thus, tumours act as a repository of genetically variant transformed cells that differ in their sensitivity to targeted therapies (Fig 2).
Figure 2. Model depicting the selective resistance of specific clonal variants in response to a chemotherapeutic. Clonal variants, of varying chemotherapeutic sensitivity are represented by different colours.
Download figure to PowerPoint
Emergence of drug resistance in patients receiving targeted or non-targeted therapy is consistent with the presence of pre-existing variants of tumour cells with varying drug sensitivities. This is supported by molecular studies showing the presence of sequence-verified tumour cell variants within individual human tumours (Navin et al, 2011; Ruiz et al, 2011). Whilst intratumoural genetic heterogeneity clearly has a capacity to drive resistance it is ironic to note that drug treatment may contribute to intratumoural genetic heterogeneity. A recent study sequenced tumour cells from acute myeloid leukaemia (AML) patients prior to and following treatment and relapse (Ding et al, 2012). Relapse was accompanied by the emergence of drug resistant clones (Ding et al, 2012). Moreover, in one patient alone they found 330 tumour-specific mutations, 78 relapse-specific mutations and only 5 mutations that were shared between the primary and relapsed tumours (Ding et al, 2012). In total, eight patients were sequenced and in all instances they found that chemotherapy altered the mutational and variant composition of the tumour resulting in genetically distinct tumour cell variants in treated patients (Ding et al, 2012). These data indicate that the mutagenic properties of some of the therapies currently in use could contribute to heterogeneity and hence could contribute to resistance.
Stochastic processes contribute to heterogeneous intratumoural drug responses
The natural variation that occurs within any cell population is often overlooked as a source of variation in chemotherapy. A series of studies by Sorger and colleagues (Albeck et al, 2008; Spencer et al, 2009) showed the extent of variation that can occur within a genetically identical population of tumour cells in response to a cytotoxic stimulus. They demonstrated that the cytotoxic ligand TRAIL displayed considerable variability with respect to the time and extent of cell death. Using cells stably expressing proteins in the extrinsic apoptosis pathways, they showed that the time to apoptosis, within any culture of genetically identical cancer cells, varied and could be described by a normal distribution. Those cells at extreme ends of the distribution spectrum responded very differently to the same dose of TRAIL (Spencer et al, 2009). This is not an isolated observation. Gascoigne & Taylor (2008) reported a similar finding by measuring the response to antimitotic chemotherapeutics in a number of cancer cell-lines (Gascoigne & Taylor, 2008). Combined, these studies indicate that genetically identical cells under identical physical conditions differ in their response to a given chemotherapeutic to an extent that may impact on clinical response.
Microenvironmental factors contribute to heterogeneous intratumoural drug responses
There is evidence that the tumour stroma actively contributes to heterogeneous tumour behaviour and, in particular, chemosensitivity. Stromal components can constitute greater than 50% of tumour mass. Tumour stroma comprises cellular and non-cellular components such as fibroblasts, immunocytes and structural proteins/fibres through to cells and tissue associated with more complex structures such as blood vessels, muscles, bone marrow or nerves. Stromal elements directly control tumour cell behaviour and chemotherapeutic responses. For example, Muranen et al (2012) showed that treatment of breast and ovarian cancer cell lines with PI3K/mTOR inhibitors led to a rapid cytotoxic response. However, they also observed that a small population of cells consistently survived in 3D tissue culture systems. The surviving cancer cells were characterized by their close proximity and interaction with the matrigel in which the cultures were grown (Muranen et al, 2012). Stromal elements and stromal substitutes such as matrigel are known to engage cellular receptors such as integrins. In this instance, Muranen et al (2012) showed that PI3K/mTOR inhibitors induced IGF1 receptor and EGF receptor, on those cells which contacted the stroma. This led to the activation of antiapoptotic pathways (e.g.: BCl2) resulting in drug resistance (Muranen et al, 2012). Significantly, treatment of mice with EGF receptor or IGF1 receptor inhibitors resulted in improved drug responses to the PI3K/mTOR inhibitors in animal models of breast and ovarian cancer (Muranen et al, 2012). Two important concepts arise from this. Firstly, drug resistance may be attributable to a subpopulation of tumour cells that, through their contact with the basement membrane, have acquired drug resistance (Fig 3). Secondly, these data show that chemoresistance can be manipulated pharmacologically. Similarly, non-small cell lung carcinoma (NSCLC) and breast cancers are associated with significant stromal infiltration. In particular, expression of proteins such as the integrins and their basement membrane ligands, laminins, are overexpressed and disrupted in their expression pattern (Desgrosellier & Cheresh, 2010). Laminin/integrin ligation is known to activate intracellular pathways such as NFkB, MAPK/ERK, PTEN/PI3K/Akt, resulting in suppression of the cytotoxic response of breast, oral or NSCLC cells to anoikis (Weaver et al, 2002), etoposide (Sethi et al, 1999; Weaver et al, 2002), doxorubicin (Sethi et al, 1999) or cisplatin (Sansing et al, 2011). Thus, the interaction of tumour cell surface receptors with adjacent stromal elements can induce drug resistant behaviour in adjacent tumour cells (Fig 3). The importance of stroma-mediated chemosensitivity has been recognized and is the basis for the development, and clinical trial, of agents such as the RGD-based inhibitors of integrins (e.g.: cilengitide) in cancer patients (Vermorken et al, 2011).
Figure 3. Model depicting the impact of stromal interactions on the sensitivity of identical clonal variants to chemotherapeutics. Identical clonal variants are shown. Those cells that interact with the stroma are marked by a star.
Download figure to PowerPoint
Given the contribution of stromal elements to tumour cell behaviour it was quickly realized that fibroblasts associated with cancer tissue were different, phenotypically (Dicker et al, 2002; Elkabets et al, 2011; Place et al, 2011) and genomically (Eng et al, 2009; Hu et al, 2005; Qiu et al, 2008) from fibroblasts associated with normal tissues. The origin of this heterogeneity is unclear. A recent study shows that cancers can be infiltrated by stromal cells derived from the bone marrow (Elkabets et al, 2011). Thus, heterogeneity could arise in the resident tissue fibroblasts or result from infiltration with fibroblasts of a different origin. Despite the complexity of the origins of cancer associated stromal cells it is clear that tumour behaviour is dependent upon their presence and thus they offer an opportunity for therapeutic intervention. For example, it has been shown that cancer stroma could have a profound indirect effect on the chemosensitivity of pancreatic cancer cells (Olive et al, 2009; Provenzano et al, 2012). The insensitivity of pancreatic cancers to gemcitabine was due, in part or wholly, to the presence of a non-permeable stromal barrier that restricted the ability of gemcitabine to reach the tumour cells. It was shown that the use of Hedgehog antagonist, IPI-926, or the hyaluronic acid disrupter, PEGPH20, collapsed the stroma allowing gemcitabine to reach the tumour cells and induce tumour cell death (Olive et al, 2009; Provenzano et al, 2012). In this instance, the resistance to gemcitabine was directly attributable to the anatomic heterogeneity within the tumour.
Tumour stroma contributes indirectly to chemotherapeutic sensitivity by regulating tumour development/progression and by exerting selection pressure on the evolving tumour. In this way the stromal elements dictate the genetic/epigenetic/phenotypic composition of the tumour and thereby modulate chemotherapeutic sensitivity. Arguably the best example of the pro-tumourigenic activity of stromal elements is seen by the establishment and growth of tumour cells at metastatic sites. For example, it has been shown that primary tumour cells are able to contribute to the establishment of a premetastatic niche at distant sites, which in turn, serves to attract and nurture the growth of tumour cells that have left the site of the primary malignancy (Kaplan et al, 2006). The best example of this would be skeletal osteoclasts which are essential to the establishment and growth of breast cancer cells at distant sites within the skeleton (Guise et al, 2006; Mundy, 2002). The relationship between the breast cancer cells and the osteoclasts is often referred to as a ‘vicious cycle’ because primary breast cancer cells release growth factors such as RANKL which stimulate the growth and maturation of distant skeletal osteoclasts which in turn resorb bone releasing matrix-associated growth factors such as TGFβ1 that attract and promote the growth of breast cancer cells in the bone (Guise et al, 2006; Mundy, 2002). The establishment of skeletal metastases significantly reduces patient lifespan and ablation of osteoclasts, using bisphosphonates, significantly reduces patient morbidity and increases lifespan such that it is now standard of care for advanced metastatic breast cancer (Coleman, 2011). These data provide a strong line of biological and clinical evidence showing the importance of the tumour stroma to tumour cell growth and the enormous clinical value of targeting this process. It is noteworthy that recent studies have shown that metastatic foci of medulloblastoma are genetically divergent from tumour cells of the primary lesion (Wu et al, 2012) suggesting that the establishment of metastatic foci may be selective for specific variants of the primary tumour that have an inherent or acquired capacity to migrate to, or take up residence, in the premetastatic niche. Thus, the metastatic stroma and presence of genetically distinct metastatic variants will contribute to the differing chemosensitivities of metastatic lesions.
The innate immune system is an active participant in the development of tumours. M1 macrophages, for example, are tumour-suppressive and associated with good tumour responses to therapy whilst M2 macrophages are pro-tumourigenic and associated with tumour progression (Mosser & Edwards, 2008). The relationship between macrophages and chemotherapeutic response has now been demonstrated in breast cancer. Recent data have shown that a high tumour associated macrophage to T lymphocyte ratio in primary breast cancers was associated with a poor prognosis (DeNardo et al, 2011). DeNardo et al (2011) showed that high levels of colony stimulating factor-1 in breast cancer led to the recruitment of tumour associated macrophages which, in turn, suppressed the tumour-suppressive activity of T lymphocytes and inhibited taxane-mediated cytotoxicity. Pharmacological inhibition of tumour associated macrophage infiltration led to the sensitization of breast cancer cells to cytotoxic drugs confirming their causal association with drug resistance (DeNardo et al, 2011). Thus, there is clinical and experimental evidence to show that the local tumour immune system contributes to chemotherapeutic responses.
Contribution of tumour cell plasticity to heterogeneous intratumoural drug responses
Tumour cells display considerable plasticity and this plasticity extends to sensitivity to chemotherapeutics (Fig 4). In cancer, plasticity refers to the ability of a cell to reversibly change lineage or to modify cell behaviour beyond the normal differentiation programme of that cell. Plasticity relating to lineage transition is generally silenced in adult tissues with the exception of some stem cell compartments (Tang, 2012). Thus, the reinstatement of plasticity in cancer cells reflects a pathological consequence of changes in the tumour cells or in the adjacent tumour environment. From a therapeutic point of view, plasticity is a confounding factor since cancer cells that respond to a particular cytotoxic therapy may be insensitive to chemotherapy if they have changed their phenotype. The best-described example of cancer cell plasticity is the continuum observed in the epithelial to mesenchymal transition (EMT) and the reverse of this process, the mesenchymal to epithelial transition (MET) (reviewed in Nieto, 2011). Studies of the EMT have revealed a causal link between the EMT and the acquisition of stem-like activities and chemoresistance. For example, the mesenchymal phenotype in lung, pancreatic, and head and neck cancer cells is associated with insensitivity to the clinically approved EGFR-targeted agent erlotinib/Tarceva (Thomson et al, 2005; Yauch et al, 2005). In particular, lung carcinoma cell lines, which have undergone an EMT, exhibit reduced sensitivity to erlotinib due to reduced dependence on the EGFR pathway (Thomson et al, 2008; Yao et al, 2010). Moreover, studies of drug sensitivity in various cancer cells, before or after the EMT, show that following a mesenchymal transition cancer cells are resistant to TRAIL (McConkey et al, 2009), radiation (Bao et al, 2006; Nieto, 2011), paclitaxel (Cheng et al, 2007), and cisplatin (Hsu et al, 2010; Latifi et al, 2011). Passage through the EMT is regulated at a transcriptional level by a suite of transcription factors such as Zeb1, Twist, Snail and Slug (Arumugam et al, 2009; Nieto, 2011) that are responsible for the phenotypic changes that accompany the EMT. In particular, the loss of E-Cadherin expression is the classic marker of the EMT and is controlled by Snail/Twist and Zeb1. Significantly, these same factors induce drug resistance (Arumugam et al, 2009; Nieto, 2011).
Figure 4. Model depicting impact of tumour cell plasticity on chemotherapeutic sensitivity. In this model a cell may give rise to individual tumour cells of different lineage that differ in their sensitivity to a chemotherapeutic agent.
Download figure to PowerPoint
Plasticity is not restricted to EMT (Thompson & Haviv, 2011). Recent work has shown that different pathological subtypes of breast cancer cells are able to give rise to one another (Chaffer et al, 2011; Gupta et al, 2011). Specifically, basal, ductal and stem-like cancer cell populations were isolated from two different breast cancer cell-lines indicating that established cell lines can stably retain intratumoural heterogeneity. Moreover, the authors showed that each of the different cell subpopulations could give rise to all three lineages in approximately the same proportions observed in the unsorted population. It can be concluded that the different subpopulations are not fixed in their phenotype. Of relevance to the present review, it was reported that the stem-like cells were chemo-resistant to paclitaxel and 5-fluorouracil and that expansion of the other subtypes of cells following chemotherapeutic exposure was due to the resistance of the stem cell fraction (Gupta et al, 2011). Interestingly, they found that all three populations of cells could initiate tumour formation in vivo. Moreover, Roesch et al (2010) showed that melanoma cells can be divided into slow and fast replicating populations. The slow-cycling population represented a small fraction of the melanoma cells and was characterized by high levels of expression of the demethylase enzyme Jarid1B (Roesch et al, 2010). Both Jarid1B+ve and Jarid1B−ve melanoma cells could initiate tumours in vivo and could give rise to mixed populations of Jarid1B+ve and Jarid1B−ve melanoma cells (Roesch et al, 2010). However, knockdown of Jarid1B reduced self-renewal suggesting that Jarid1B+ve cells had stem-like qualities. Unfortunately, the chemo-sensitivity of the Jarid1B+ve and Jarid1B−ve populations was not examined (Roesch et al, 2010). Although these studies are very recent and have not yet been validated in other cancer types, they provide important insight into how intratumoural heterogeneity evolves and how this may relates to drug responses (Fig 4).