The role of tumour heterogeneity and clonal cooperativity in metastasis, immune evasion and clinical outcome

It was recognised in the 1970s that heterogeneous subclones within tumours contain differing capacities for growth and metastatic ability [55‐57]. Early work also recognised that interactions between subpopulations resulted in differences in drug sensitivity when injected into opposite flanks of mice or co-cultured in collagen [58, 59]. The importance of these studies was not fully recognised until over two decades later [60].

Growing interest in tumour heterogeneity and subclonal interactions led to the development of a useful model for studying subclonal competition and cooperation in Drosophila melanogaster [60]. Eichenlaub et al. [61] used a D. melanogaster model of epithelial tumour formation to show that overexpression of both epidermal growth factor receptor (EGFR) and miR-8 in wing imaginal disc cells results in supercompetitive cells that engulf those surrounding them. These supercompetitive cells drive tumourigenesis and metastasis, whereas cells overexpressing either EGFR or miR-8 alone do not. Competition between normal and oncogenic tissues was also uncovered: in D. melanogaster, imaginal epithelial cells activate nonapoptotic JNK signalling in response to oncogenic mutant cells [62]. This leads to the activation of the ELMO/Mbc-mediated phagocytic pathway, which eliminates oncogenic cells [62]. Growing evidence suggests that subclones within tumours compete, but also cooperate. Several D. melanogaster studies demonstrated cooperation between cells with oncogenic Ras ^V12 mutations and cells lacking the scribbled gene (scrib-/-) to promote growth and invasion of the Ras ^V12 mutant cells [60, 63].

Xenotransplant models have been used to understand interactions between human tumour subclones. In a glioblastoma multiforme mouse xenotransplant model, a minor mutant EGFR subpopulation enhanced tumourigenicity of the entire tumour. This occurred via a paracrine mechanism that promoted growth in wtEGFR cells within the tumour, illustrating the advantages of heterogeneity [64]. In a zebrafish-melanoma xenograft model, inherently invasive (MITF^high) melanoma cells cooperated with poorly invasive (MITF^low) cells to invade away from the primary site through solid tissue [65]. The protease activity and extracellular matrix deposition of MITF^high cells around the primary tumour was critical for this co-invasion of MITF^high and MITF^low cells [65]. Heterogeneity may be critical in tumour progression and metastasis.

Polyak and colleagues [66] used a mouse xenograft model to investigate the effect of a polyclonal tumour population on tumour growth and metastasis. A minor IL11-overexpressing subclone changed the tumour microenvironment by increasing intratumoural vascularisation and reorganising extracellular matrix, in turn increasing tumour growth [66]. Importantly, this subclone was also outcompeted by a faster proliferating competitor, leading to tumour collapse [66].

These xenograft studies provide insights into subclonal interaction dynamics, demonstrating their complexity, and that they are involved in many aspects of tumour evolution including tumorigenesis, vascularisation, invasion and metastasis.

Advances in molecular biology have allowed a deeper study of subclonal cooperation in cancer mouse models. Autochthonous models permit long-term study of clonal dynamics in a specific genetic context [67].

In a mouse model of small cell lung cancer (SCLC), tumours were composed of distinct cell populations with either neuroendocrine or mesenchymal markers [68]. Both cell types shared specific genomic aberrations derived from a common ancestor [68]. Transition between the neuroendocrine and mesenchymal phenotype was achieved through ectopic expression of Ras ^V12 , and the mixed population endowed mesenchymal cells with metastatic ability [68].

Cleary and colleagues [69] used a mouse model of breast cancer to show that interclonal cooperation can be critical for tumour propagation. A portion of tumours harboured distinct Hras ^mut Wnt ^low and luminal Hras ^wt Wnt1 ^high subclones, which cooperated to maintain tumour propagation [69]. Although Wnt activation is rare in human breast cancer, this remains a valuable model for studying paracrine mechanisms of subclonal communication.

In a p53-null mouse model of basal-like breast cancer, two tumour cell populations were identified, one expressing mesenchymal markers (“mesenchymal-like” cells), and another defined as tumour-initiating cells (TICs) [70]. When both populations were co-transplanted using limited dilutions, mesenchymal-like cells promoted self-renewal and tumour initiation capacity of the TICs [67]. Surprisingly, the mesenchymal population was maintained as a minor subpopulation [70].

Mouse models have revealed how subpopulations within tumours cooperate to promote tumour growth, maintenance and metastasis. Although not fully representative of the clinical presentation of disease in patients, these autochthonous models show the diverse ways in which subclones can interact and affect tumour progression. These studies suggest that to more effectively treat heterogeneous tumours, we must first understand the dynamics between different populations within a tumour, and how targeted treatment changes these interactions.

In non-small cell lung cancer (NSCLC) a low frequency subclone with MET amplification was selected for during treatment with EGFR tyrosine kinase inhibitors (TKI), allowing for development of a resistant MET- amplified tumour [71]. The EGFR T790M mutation, which drives EGFR TKI resistance, was also present prior to treatment in NSCLC [72]. In colorectal cancer, EGFR antibody-resistant subclones with pre-existing KRAS mutations emerged in 38% of patients following 5–6 months of treatment with panitumumab [73]. In patients with NSCLC, a subset of EGFR TKI resistant cancers acquire alterations, such as RB and p53 loss contributing to small cell carcinoma transformation [74].

In patients with HER2-positive breast cancer, Janiszewska et al. [75] discovered a dramatic increase in a minor resistant population of PIK3CA mutant cells, and a slight decrease in the dominant population of HER2-amplified cells post neoadjuvant therapy with trastuzamab. Alterations in the phosphoinositide 3-kinase (PI3K)-AKT pathway commonly drive resistance in patients treated with the HER2-targeting antibody trastuzamab [75].

In patients with colorectal cancer, mutant KRAS clones emerging during treatment with EGFR-specific antibodies declined during treatment breaks [76]. Engelman and colleagues [77] demonstrated that resistance to EGFR TKIs can develop through pre-existing clones and newly developed drug-tolerant clones simultaneously, though through separate mechanisms. When treating the NSCLC cell line PC-9 with increasing concentrations of gefitinib, one resistant population derived from EGFR T790M pre-existing clones emerged early, while another with characteristics of a drug-tolerant state and the EGFR T790M mutation, emerged late. This late emergent EGFR T790M subpopulation arose from drug-tolerant cells that were, initially, partially resistant to gefitinib, and then gained the EGFR T790 mutation to become fully resistant [77].

These studies demonstrate that heterogeneous tumours develop resistance when specific clones are targeted, allowing pre-existing or newly evolved subclones to emerge.

Competition between subclones is frequently revealed during and following therapy exposure. Keats and collaborators [78] illustrated that, in a patient with multiple myeloma, clonal dominance alternated between two main subclones. Modelling these subclones in a mouse, treatment with bortezomib exerted selective pressure on all subclones, but one clone eventually emerged and outcompeted the others [78]. Conversely, minor therapy-resistant subpopulations can also support the survival of therapy-sensitive populations, preserving tumour heterogeneity. In patients with colorectal cancer, EGFR therapy-resistant KRAS mutant subclones support non-mutant therapy-sensitive cells by secreting increased levels of TGFalpha and amphiregulin [79], in turn sustaining EGFR/ERK signalling in sensitive cells [79]. These data demonstrate how resistance to targeted therapy can develop with both preservation and loss of heterogeneity.

Is intratumour heterogeneity associated with worse patient outcomes? In premalignant Barrett’s oesophagus, clonal diversity is an important predictor of tumour progression [80]. By adapting diversity measures from ecology to measure both the number and abundance of clones relative to others in the population, the upper quartile of the number of clones and the genetic divergence based on loss of heterozygosity (LOH) were strongly predictive of increased progression from Barrett’s oesophagus to oesophageal adenocarcinoma [80].

In breast, ovarian, gastric and NSCLC, there is a paradoxical relationship between CIN and prognosis. The worst outcomes are in tumours with intermediate CIN, while those with extreme CIN scores had an improved outcome [81, 82]. The tumour cell of origin and the order of somatic events may also influence tumour heterogeneity and patient outcome [83].

In a recent study of 12 different cancer types [84], mortality risk increased when more than two clones coexisted in the same tumour sample, but decreased with the coexistence of more than four clones, emphasising that heterogeneity levels within tumours can strongly influence outcome [84]. Tumours with high levels of somatic copy number alterations (SCNAs) have increased proliferation markers and a decreased immune signature [85]. Interestingly, both phenotypes are controlled by different types of aneuploidy, and may be important patient outcome markers, as patients with higher SCNA levels tend to have worse responses to immunotherapy [85]. Recently, our group demonstrated in the TRACERx prospective study of 100 patients with NSCLC, that heterogeneity of DNA copy number events rather than point mutations, was associated with poor outcome [86].

In breast cancer, enriched areas of tandem duplications previously thought to be unimportant passengers, were shown to be potentially important for tumourigenesis [87]. These tandem duplications were characterised into two different rearrangement signatures [87]: one enriched in areas that disrupted tumour suppressor genes such as PTEN and RB1, and the other enriched in oncogenes, including MYC, and in putative regulatory elements of genes such as ESR1, with direct effects on transcription [87]. These enrichments demonstrate that mutational processes not only stochastically mutate the genome and induce heterogeneity but are also likely to be critical in cancer evolution and patient outcome. Further work to understand how established clinical parameters of outcome such as tumour size, grade and stage reflect intratumour heterogeneity is required.

Emerging evidence suggests that intratumour heterogeneity may also influence the anti-tumour immune response. Rapid advances in cancer genome sequencing have enabled scientists to decipher the impact of somatic coding alterations upon immune recognition and surveillance. Non-silent mutations generate neoantigens that can be recognised by the immune system [88, 89]. Both CD4+ and CD8+ T lymphocytes were shown to recognise tumour neoantigens [89‐93]. Using whole genome sequencing, scientists have established that mutation burden is highly variable among different tumour types [94]. Mutation levels in melanomas and lung cancers are among the highest because of exposure to ultraviolet light and tobacco carcinogens respectively [94, 95]. This makes these cancer types ideal candidates for immunotherapy; however, others such as clear cell renal cancer also respond to checkpoint therapy despite a much lower mutational burden [96].

In patients with melanoma, whole exomes from pre-treatment melanoma tumour biopsies and matching germline tissue samples were examined to determine if neoantigen burden affects responses to antibodies directed against cytotoxic T lymphocyte-associated antigen-4 (CTLA4) [97]. Overall mutational load and expression of cytolytic markers in the immune microenvironment were significantly associated with an anti-CTLA4 response and clinical benefit [97].

Rizvi and colleagues [98] explored how mutational burden affects sensitivity to PD-1 blockade in NSCLC. Expressed by activated T cells, PD-1 is a key immune checkpoint receptor that mediates immunosuppression [88]. They concluded that a higher nonsynonymous mutation burden was associated with improved objective response, durable clinical benefit and progression-free survival [98]. However, some patients with both high and low mutational burdens failed to respond to therapy [98], suggesting that other factors are involved in immunotherapy responses.

We investigated whether melanoma and NSCLC sensitivity to PD-1 and CTLA-4 treatment was enhanced in tumours where tumours had a high clonal neoantigen burden. Considering a combination of neoantigen clonality and neoantigen burden allowed us to better discriminate responder from non-responder patients than either metric alone [99]. These findings suggest that the degree of intratumour heterogeneity may be associated with a differential response to checkpoint blockade, and that multiple factors are involved in tumour immune responses. These studies raise the possibility that technologies now exist to exploit clonal neoantigens, present in every tumour cell for therapeutic benefit, either through vaccination or cell therapy approaches.

Multiple laboratories are now focusing on a deeper understanding of immune evasion and resistance to immunotherapy. One study [100] uncovered two novel mechanisms of evasion: first, the elimination of neoantigen-containing tumour cells within a subset of the tumour population, and second, the acquisition of one or more genetic events that resulted in neoantigen loss. Both clonal and subclonal alterations were lost following immunotherapy, but all clonal neoantigens were eliminated by chromosomal deletions and LOH, whereas subclonal neoantigens were lost through LOH and elimination of tumour cells [100]. We uncovered that CIN likely contributes to mutational heterogeneity within tumours through SCNAs [86]. More than 14% of subclonal mutations (range: 0–56%) are subclonal because of SCNA loss events of clonal alterations [86]. This suggests that CIN contributes to neoantigen heterogeneity, and may play a role in immune evasion and immunotherapy resistance. These findings raise questions about whether immunotherapy will be effective as tumours evolve and become heterogeneous, and to what extent underlying CIN, resulting in rapid karyotype evolution, drives resistance to such approaches. Understanding neoantigen heterogeneity within tumours will aid the development of more effective, patient-specific immunotherapies, which may not prevent eventual relapse but could significantly extend patient’s lives.

T cell exhaustion is another mechanism of immune evasion. This state of T cell dysfunction arises during chronic antigen exposure, which is defined by poor effector function, sustained expression of inhibitory receptors, and an abnormal transcriptional state [101‐103]. In this state, PD-1 is one of many cell surface inhibitory receptors to co-regulate T cell exhaustion [103]. Anti-PD-1 immunotherapy partially restores T cell function from an exhausted state [102, 104, 105]. It is also clear that the relative abundance of partially exhausted tumour-infiltrating CD8+ T cells predicts response to anti-PD-1 therapy [106]. However, resistance to PD-1 therapy develops in many patients.

Alterations in the genes encoding interferon receptor-associated Janus kinase 1 (JAK1) or 2 (JAK2), along with deletion of the wild-type allele, lead to defects in pathways involved in interferon receptor signalling and in antigen presentation. This results in resistance to PD-1 blockade in melanoma [107]. Loss of PTEN also inhibits T cell-mediated tumour-killing by decreasing T cell infiltration in tumours, and is associated with worse outcomes with anti PD-1 therapy [108]. These studies suggest that tumour heterogeneity is beneficial to immune evasion and the development of immunotherapy resistance. Approaches that attenuate tumour heterogeneity by inhibiting the molecular mechanisms driving cell-to-cell variation, or that target multiple truncal mutations occurring consistently in all regions of the tumour while stimulating the immune response using checkpoint blockade, might help limit diversity and resistance acquisition [99].