**4. A Platform for Treatment Response Studies; Moving towards Personalised Treatment?**

The concept of precision treatments has been of growing interest in many fields of research in recent years, particularly oncology, as there is a wide variability of patient responses to standard "one size fits all" treatment regimens. In some cases, genetic factors which can be specifically targeted in a "personalised" manner are already known, for example non-small cell lung cancer patients with an activating mutation in tyrosine kinase are particularly sensitive to treatment with tyrosine kinases inhibitors such as gefitinib [87]. However, for other cancers, such as oesophageal cancers and locally advanced rectal cancers, there are currently no accurate predictors of patient responses to treatment. The standard of care for oesophageal cancer consists of neo-adjuvant chemoradiotherapy followed by surgery, with a complete pathological response observed in approximately a quarter at the time of surgery but no response in approximately one fifth of patients [88,89]. Similarly, for neoadjuvant chemoradiotherapy treatment of colorectal cancer while approximately one fifth of patients show a complete pathological response, almost 40% of patients show no benefit to the treatment [90]. In both cancers, patients would clearly benefit from more robust pre-treatment predictive models.

Therefore, there has been a concentrated effort in the field of organoids to establish reliable predictors of colorectal cancer treatment response to both chemotherapy alone [91,92] and neoadjuvant chemoradiotherapy [93,94]. Van de Wetering et al. [91] established colorectal cancer organoids, alongside paired healthy tissue, and demonstrated that the organoids recapitulated the genetic profiles and mutational spectra of the tumours of origin. Furthermore, by performing screening of 83 compounds, including both clinically used drugs and experimental compounds, the authors showed that the organoids facilitated the high-content drug screening [91], which could facilitate precision treatments in the future. Interestingly, a later study by Ooft et al. [92] investigating treatment response of metastatic colorectal cancer using organoids, was able to predict accuracy of irinotecan monotherapy and 5-flurouracil/irinotecan dual therapy, with 80% and 83.3% respectively. While greater accuracy is required to implement predictive models in a clinical setting, these studies show the developing potential of organoids in precision medicine. Furthermore, in recent years, there has been an increasing number of studies aimed at identifying and repurposing already available drugs as radiosensitisers [95–98]. Drugs which can be repurposed offer cheaper and quicker alternatives to developing new drugs from scratch, while many of the adverse side effects are already known [99]. The possibilities to quickly and accurately screen drugs, as shown in the studies of van de Wetering et al. [91] and Ooft et al. [92], in cancer organoids will greatly increase the possibilities in precision medicine and further benefit the search for potentiators of radiation therapy.

Indeed, recent studies by Ganesh et al. [93] and Yao et al. [94] have focussed on rectal cancer organoids for predicting patient responses to neoadjuvant chemoradiotherapy (Table 1 summarises the different cancer organoids that have been used in studies of radiation responses). Both studies further consolidated other evidence that rectal cancer organoids faithfully recapitulate the tumours of origin, performing histopathological and mutational comparisons between the two [93,94]. Moreover, Ganesh et al. showed that upon xenotransplantation of the organoids into mice they were found to metastasise to the same locations as the original tumours. Importantly, upon treating the organoids with chemotherapeutic drugs (such as 5-Flurouracil and oxaliplatin) heterogeneous treatment responses correlated with the clinical progression-free survival of patients. Interestingly, organoids which displayed resistance to radiation were derived from patients who either were resistant to therapy or showed disease recurrence following treatment [93]. Yao et al. [94] also correlated the therapeutic clinical outcomes to the standard neoadjuvant chemoradiotherapy with the organoid outcomes following treatment 5-Flurouracil, irinotecan or radiation. In sixty-eight out of the 80 patient-derived organoid lines generated, at least one of the three treatment courses was found to be predictive of the patient's tumour regression score after surgery [94]. Furthermore, in a recent study, Pasch et al. established patient-derived cancer organoids and were prospectively able to predict the treatment response of a patient with metastatic colon cancer [100]. These studies combined with the works of

van de Wetering et al. and Ooft et al. provide a significant step towards a model for patient-specific response prediction.



A recent study also established an organoid model for metastatic gastrointestinal cancers which were histologically, genetically and molecularly similar to the tumour of origin [104]. Following drug treatment of the organoids, the outcomes were compared with the clinical outcomes of the patients enrolled in Phase I/II clinical trials and were found to closely mimic the clinical outcomes of the patients [104]. Moreover, the study successfully identified differential inter- and intra-patient responses to common chemotherapeutic agents for gastrointestinal cancer treatment [104]. This study represents an important advance for organoids in the field of personalised precision medicine.

As mentioned above, currently the ability to predict patient responses to chemoradiotherapy for oesophageal cancer is also extremely limited. Great strides are being made towards the optimisation of imaging techniques for predicting treatment outcomes for oesophageal cancer treatment [89,105–107]; however, there is still no means to accurately predict patient outcomes. Recent advances in the culturing of oesophageal adenocarcinoma organoids have established new models to study the development and heterogeneity of the disease [108]. The established patient-derived oesophageal adenocarcinoma organoids shared histopathological features with patient-matched tumour samples and genetic mutations were conserved at a patient-specific level [108]. They further showed a loss of cellular polarity, which is often considered a hallmark of cancer. Drug screening in the organoids revealed a highly diverse range of responses, which tallies with the difficulties in predicting patient responses. However, the diversity of the responses remained throughout passaging, indicating the stability of the model through time [108]. Unfortunately, the findings of this study were somewhat limited due to a low success rate of establishing organoids (organoids were established from only 10 out of 32 patients). Reasons for a low success yield included failure to initiate culture, infection, fibroblast overgrowth, and arrested growth [108], while others also working on developing oesophageal adenocarcinoma organoids have recently identified the presence of Barrett's epithelium as another

potential contamination source in culture [109]. These new models will be essential to opening new avenues for testing new drugs and treatment regimens for oesophageal adenocarcinomas. Furthermore, as mentioned above, radiotherapy is an also important arm of treatment for other cancers in the head and neck region. Recently established protocols for generating organoids from oral mucosa and head and neck squamous cell carcinomas may facilitate a more personalised treatment planning for more tumours in this region [101]. Comparisons of the responses of tumour organoids with matched normal tissue organoids may even allow for studies of the therapeutic window on a personalised scale.

Glioblastoma is a highly aggressive brain tumour with an extremely poor prognosis for patients for whom radiotherapy is an integral arm of treatment [110]. This remains the case even with significant advances in the understanding of glioblastoma development, cellular heterogeneity within the tumour, and the role of cancer stem cells play in this [111–113]. Many of the models used for studying glioblastoma utilise adherent monolayers which, although they have been highly revealing of the mechanisms of glioma stem cell resistance [114], have thus far not been representative of the tumour microenvironment or levels of therapeutic resistance of glioblastoma seen in vivo. However, recently new organoid models have been established that could shed light on the initiation, development, tumour invasion, and treatment of glioblastoma. In two independent studies, Bian et al. [115] and Ogawa et al. [116] utilised CRISPR/Cas9 genome editing technology to manipulate cerebral organoids towards tumorigenesis. In both studies, cells derived from the generated tumour organoids exhibited epithelial-mesenchymal properties, indicative that they are representative of the invasive mesenchymal subtype of glioblastoma. Indeed, the cells were invasive when seeded with normal cerebral organoids [115,116] and were capable of forming tumours when xeno-transplanted into mouse recipients [116]. While neither group determined radiation responses of the glioblastoma organoids, Bian et al. demonstrated that CRISPR/Cas9 generated glioblastoma organoid models are appropriate for preclinical in vitro drug screening [115].

Indeed, studies which have investigated the radiosensitivity of glioblastoma organoids have demonstrated that they more closely resemble in vivo tumour sensitivity than monolayer cultures [102,103]. Furthermore, importantly, particularly from a radiobiology point-of-view, Hubert et al. showed that although the non-stem cells of the organoids were radiosensitive, the tumour-initiating cancer stem cells were indeed resistant [102], recapitulating important in vivo findings from previous studies [114]. Although these glioblastoma models offer excellent platforms to study glioblastoma development and biology, and to test new treatments, the duration of culturing generally does not facilitate rapid screening for a more personalised approach to treatment. However, recently a robust and rapid (within 1-2 weeks) protocol for establishing glioblastoma organoids capable of facilitating moderate to high throughput screening for a potentially more personalised response prediction [117].
