Predicting Response to Neoadjuvant Therapy in Oesophageal Adenocarcinoma

Simple Summary Oesophageal adenocarcinomas are a distinct subtype of oesophageal cancer that has an increasing incidence in western countries. As these cancers are often late presenting, patients with locally advanced oesophageal adenocarcinomas are routinely treated with neoadjuvant chemotherapy with or without radiotherapy prior to surgery. Unfortunately, this neoadjuvant protocol demonstrates limited response, while exposing patients to the side effects of therapy. Biomarkers that can accurately predict neoadjuvant therapy response would save time, suffering, hospital resources and potentially improve survival. Abstract (1) Background: Oesophageal cancers are often late-presenting and have a poor 5-year survival rate. The standard treatment of oesophageal adenocarcinomas involves neoadjuvant chemotherapy with or without radiotherapy followed by surgery. However, less than one third of patients respond to neoadjuvant therapy, thereby unnecessarily exposing patients to toxicity and deconditioning. Hence, there is an urgent need for biomarkers to predict response to neoadjuvant therapy. This review explores the current biomarker landscape. (2) Methods: MEDLINE, EMBASE and ClinicalTrial databases were searched with key words relating to “predictive biomarker”, “neoadjuvant therapy” and “oesophageal adenocarcinoma” and screened as per the inclusion and exclusion criteria. All peer-reviewed full-text articles and conference abstracts were included. (3) Results: The search yielded 548 results of which 71 full-texts, conference abstracts and clinical trials were eligible for review. A total of 242 duplicates were removed, 191 articles were screened out, and 44 articles were excluded. (4) Discussion: Biomarkers were discussed in seven categories including imaging, epigenetic, genetic, protein, immunologic, blood and serum-based with remaining studies grouped in a miscellaneous category. (5) Conclusion: Although promising markers and novel methods have emerged, current biomarkers lack sufficient evidence to support clinical application. Novel approaches have been recommended to assess predictive potential more efficiently.


Introduction
Oesophageal cancer is the sixth most common cause of cancer-related death worldwide. It is an aggressive and often-late presenting cancer associated with a poor 5-year survival rate of 15-25% [1]. The two main histological subtypes of oesophageal cancer include oesophageal squamous cell carcinoma (OSCC) and oesophageal adenocarcinoma (OAC). Although commonly homogenised in studies, these entities are histologically, epidemiologically and genetically distinct [2]. Over the last four decades, the incidence of

Discussion
The literature for response prediction in NAT is characterised by two main approaches: the first involves using imaging techniques to compare derived variables at time points before, during and after NAT; the second approach relies on detecting molecular markers prior to NAT. Both approaches seek to influence management either through early cessation or avoidance of NAT with each group in recent years, incorporating novel techniques and modalities to identify markers of response.

Imaging Markers
In the past two decades, multiple systematic reviews have been published investigating the predictive potential of computed tomography (CT) and fluorodeoxyglucosepositron emission tomography (FDG-PET) [16,17]. Response prediction with CT predom-inantly involves comparing changes in tumour volume during NAT, whereas in FDG-PET, changes in standardized uptake value (SUV) of FDG reflect alterations in tumour metabolism. These modalities have been deemed insufficiently accurate in identifying complete pathological responders and are not recommended in clinical practice.

Computed Tomography
Although novel CT techniques, such as 3D-CT volumetry, appear more sensitive in response prediction, van Heijl notes in a 2011 study that tumour volume fluctuates paradoxically during the course of NAT and hence should not be relied upon to accurately reflect pathological response [18]. This insight may also be key to understanding the failure of CT findings overall as a predictor of response. CT radiomics-based risk factor models have previously been explored to predict survival outcomes; however, they have demonstrated limited prognostic power over standard clinical variables [19]. This review did not identify any studies assessing the utility of CT radiomics in predicting NAT response.

Fluorodeoxyglucose-Positron Emission Tomography
Although the OAC-specific FDG-PET studies included within this review demonstrate some efficacy in the NAT response prediction [20][21][22][23][24], it was neither accurate nor sensitive enough for clinical utility. This is consistent with conclusions drawn in recent systematic reviews that more broadly focus on oesophageal cancer [16]. Similar to CT radiomics, PET radiomic markers, such as intratumoural uptake heterogeneity, have also been explored. A study in 2013 by Tan et al. demonstrated the predictive potential of PET tumour heterogeneity markers in a low cohort study [25]. However, in a separate study by van Rossum et al. in 2016 with a larger cohort, PET heterogeneity only showed incremental predictive advantage over standard clinical prediction models [26]. Importantly, the past decade has observed a trend toward multimodal imaging techniques, such as FDG-PET-CT, and even more recently, the inclusion of magnetic resonance imaging (MRI).

Magnetic Resonance Imaging
Recently, MRI techniques, such as diffusion-weighted imaging (DWI) and dynamic contrast enhanced (DCE) MRI have been the focus of investigation. DWI-MRI exploits the random motion of water molecules to deduce tissue cellularity and assigns the degree of diffusion with an apparent diffusion coefficient (ADC) value. Tumours that have high cellularity appear hyperintense on imaging and can be monitored through changes in ADC. DCE-MRI uses a rapid series of contrast-enhanced T1 images to track changes in tumour microvasculature (measured by area-under-concentration-time (AUC) curve for inflowing contrast).
Two studies investigating the utility of DWI-MRI in predicting NAT response concluded that changes in ADC were highly predictive of histopathological response with high specificity [27,28]. An earlier study by Kwee et al. was unable to demonstrate this but may have been limited by its low sample size. The same study did, however, demonstrate robust interobserver reproducibility with DWI-MRI results at their centre, which is encouraging and necessary for widespread adoption [29]. Concomitantly, a unimodal DCE-MRI study also demonstrated predictive potential [30]. These successes have paved the way for multimodal imaging techniques that hope to further bolster predictive accuracy. In 2018, Heethuis et al. published the first study that demonstrated the complementary value of DWI-MRI and DCE-MRI in NAT response prediction [31]. DWI-MRI and FDG-PET-CT have also been investigated in two multimodal studies and similarly demonstrated stronger combined predictive values than what each individually was capable of [32,33].
Small sample size is a universal limitation of most of the discussed MRI studies and though the quoted c-statistics are promising, one must be wary of potential type 2 errors when interpreting these results. Higher-powered multicentre studies with histologically unpooled analyses are required to confirm these findings for OACs. Borggreve et al. also identifies the 10-15 day window after starting NAT as the optimal moment for an interim NAT MRI scan, as it is timely enough for changes to manifest but early enough to avoid distortion by potential radiation oesophagitis [28], as initially theorised by van Rossum et al. [27]. In contrast, within Heethuis's multimodality study of DWI and DCE-MRI [31], comparison between the preversus post-treatment MRI, as opposed to the pre-treatment versus during MRI, demonstrated higher predictive value. Interestingly, these results are inconsistent with other findings from the same institution [27,28]. This may be explained by the differences in sample size and the use of automatic contouring software to delineate the tumour rather than manual approaches used in the past [31]. Moreover, there is evidence that semiautomated delineation of the tumour on an ADC image was more reproducible than manual methods [29] and should be considered as the technique of choice for future studies.
Logistically, protocols to control for cardiopulmonary motion artefacts in DWI-MRI need to be published, as several studies noted this issue, which resulted in unusable ADC images and a further reduction in study sample size [27,30]. Furthermore, better predictive accuracy for DWI-MRI could be obtained in future studies if tumour heterogeneity were addressed by using voxel-wise approaches to map ADC and replace currently simplistic mean-based ADCs [34]. Multimodal imaging appears beneficial for better response prediction. The multicentre PRIDE study that is currently underway (ClinicalTrials.gov, accessed on 7 December 2021, Identifier NCT03474341) will investigate the utility of trimodal imaging for NAT response prediction and should secondarily validate many of the low powered MRI studies discussed in this review [35].

Epigenetic Biomarkers
Epigenetics refers to processes that manipulate chromatin structure to modify gene expression without fundamentally altering the DNA sequence. This is facilitated through three key processes: DNA methylation, histone modification and non-coding RNA gene interactions. MicroRNAs (miRNA) are a subset of non-coding RNA that regulate gene expression through transcriptional or post-transcriptional interactions [36]. MicroRNA and microRNA panels have been the more common avenue of inquiry in predictive epigenetic biomarkers, with only one study examining methylation and no studies on histone modification in OAC.
Although many novel miRNA have been found, the majority of these appear to be from the same discovery cohort used in Lynam-Lennon et al. and Bibby et al. and have not been clinically validated [38][39][40][41]. Moreover, Bibby et al. provides insight on the limitations of assessing miRNA's effects on chemo-or radiosensitivity in vitro, as the NAT response in vivo involves complex interplays between the tumour microenvironment [38].
This key implication further emphasizes the importance of validating novel miRNA's in larger cohorts to ensure generalisability before commencing preclinical studies to elucidate mechanisms. Given the similar discovery cohorts used between a large portion of miRNA studies and the lack of validation [38][39][40][41], there is inconclusive evidence to suggest any promising singular miRNA biomarkers for NAT response prediction in OAC.

MiRNA Panels
Despite the discovery of many individual miRNA, to date there are only two OACspecific miRNA panels that have been evaluated [44,45]. In 2015, Odenthal et al. generated a panel from a discovery cohort, which included previously investigated miRNA-192. Unfortunately, the panel failed to demonstrate any predictive impact [45]. The divergence in miRNA-192 s predictive potential between Odenthal's two studies [43,45] may be attributed to several reasons. Although miRNA-192 s significance as a marker was initially validated, it was not specific to OAC, as the validation cohort consisted of 52.5% OAC patients. Odenthal's subsequent study included miRNA-192 in a miRNA panel that failed to predict response in a cohort of only OAC patients. Furthermore, there was heterogeneity in chemotherapy regimens and serological miRNA samples were obtained in contrast to the biopsy tissue miRNA used in the first study. Despite being easier to obtain, it is unclear whether serological miRNA is reflective of tissue miRNA, as other studies have identified confounding sources of miRNA in blood [42]. Therefore, in future studies, it is important to minimise these sources of heterogeneity and focus on demonstrating predictive value of the panel before assessing whether the corresponding serological test is valid.
More successfully, Skinner et al. generated an internally validated novel miRNA expression profile (MEP) score from a panel of miRNA derived from a rigorous study design with a homogenous patient cohort to collectively determine the probability of pCR (MEP score ROC-AUC was 0.78 in model cohort) [44]. The MEP score demonstrated a novel approach to evaluating the predictive potential of combined miRNA profiles. These early successes justify further investigation into the utility of the MEP score as a predictor of NAT response [44]. Skinner et al. notes that only a partial complement of miRNA was investigated, leaving room for future studies to bolster the panel's predictive potential by adding new miRNA.

DNA Methylation
A reasonably powered study from Slotta-Huspenina et al. was the only study in this review to investigate DNA methylation related to NAT response in OAC and found that patients without histological response had significantly higher mean TFAP2E gene methylation [46]. Given the distinct mutational profile of OAC compared to OSCC [47], further studies should be done to investigate the difference in methylation that could explain the divergent responses to chemo-and radiotherapy between the two histological subtypes [48].

Genetic Biomarkers
Genetic biomarker studies have broadly been investigated in panels or as individual genes or alleles. Genes of interest are either derived from past literature or through applying a range of genetic detection techniques to discovery cohorts.

Genes
Two retrospective studies investigated key genes to predict NAT response in an OAC specific cohort [49,50]. In one study, overexpression of CCL28 (inflammatory chemokine) and underexpression of DKK3 (a tumour suppressor and prognostic marker [51]) were predictive of pCR [49]. CCL28 s predictive value was exclusively investigated in OAC for the first time in this study and requires further validation. A key limitation of manual gene selection is that it oversimplifies complex biological processes and prevents the observation of concomitant gene interactions that may improve predictive potential.
Ephrin B3 receptor gene was another candidate biomarker identified through gene expression analysis, which enables broader oncogenic trends to be identified in functionally similar genes [50]. However, these results are limited by a lack of external validation, and the use of a lower response discrimination threshold in this study may potentially lead to false positive genes being implicated.
It is also important to note for future studies that prognostic value does not necessarily correspond to predictive potential. It is difficult to differentiate between these outcomes without a surgery-only treatment arm in prospective studies, which is becoming less possible as NAT has shown clear benefits in increasing overall survival [5].

Gene Panel
A panel of 26 differentially expressed genes was used to create an artificial neuronal network (ANN) with predictive potential in a recent pilot study by Lloyd et al. [52]. ANN has previously been used in one study to investigate differential genes; however, these genes were manually selected, and the study did not focus on OAC [53]. Lloyd et al. demonstrates a novel approach to predicting NAT response in OAC with an ANN that preliminarily evaluated a 26-gene panel with an accuracy of 73%, sensitivity of 80% and specificity of 70% [52]. Further studies in larger cohorts are recommended to validate the ANN model and improve predictive performance.

Alleles
The alleles and single nucleotide polymorphisms (SNPs) investigated in the literature are largely heterogeneous and based on past evidence of predictive potential in other cancers [54][55][56]. Although sample sizes were large relative to other biomarker studies, most articles had mixed histology, albeit more OAC than OSCC [55][56][57].
Heterozygous ERCC1, which codes a component of a nucleotide excision repair complex that repairs platinum-induced DNA damage [54], demonstrated positive association in an OAC specific study but could not be confirmed in a recent prospective study due to incompletely reported results [57]. Hence its predictive performance remains statistically unquantified and should be confirmed in a future prospective study.
To add to the heterogeneity, one study found significantly reduced NAT response with vitamin D receptor polymorphism (Apal) [56], while another found no significance with the multidrug resistance protein gene, ABCB1 [55]. These findings were not corroborated by any other study identified within this review and have not been externally validated.
Furthermore, as 5FU and cisplatin act in a p53-dependent fashion [58,59], Kandioler explored the effect of mutant TP53 on NAT response and found significant difference in response between patients with normal and mutant TP53 [60]. Clinical validation of TP53 s status as a predictive biomarker is currently being explored in the multicentre Pancho trial (ClinicalTrials.gov, accessed on 7 December 2021, Identifier: NCT00525200) [61].

Copy Number Alterations
Qian et al. reported the first use of genomic gains as an independently validated biomarker [62]. Copy number increases in chromosome 14q11 and 19p13 were significantly associated with pCR [62]. Conceptually, copy number alterations present an interesting avenue of further genetic research, given these findings and the role of radiation-induced instability, which has been previously hypothesised as a mechanism of miRNA dysregulation [41].

Protein Expression
Protein biomarkers are a heterogeneous group of molecules that are categorised in this review by their potential role in oncogenesis.

Growth and Proliferation
Epidermal growth factor receptor (EGFR) dysregulation results in increased cell survival, proliferation and migration [63]. Aichler et al. supports the hypothesis that EGFR overexpression contributes to cisplatin-based resistance [63]. However, less than 40% of non-responders exhibited overexpression, suggesting that EGFR alone is not enough to accurately predict NAT response.
The sonic hedgehog (SHH) pathway is a key growth signal for epithelial-mesodermal interactions during embryonic gut development. Past studies suggest that in both OAC and OSCC, SHH is overexpressed due to possible genomic amplification [64]. A validated immunohistochemistry-based assay for SHH pathway proteins demonstrated high AUC-ROC values in a second validation study with consistent results to the first [65,66]. This assay also used labelling scores that minimise interobserver variability. A few minor limitations exist, including the use of older 5FU-based chemotherapy regimens rather than those with less toxicity, such as CROSS [5], although the authors do not anticipate this to significantly impact the test. Regardless, this remains to be clinically validated. A prospective phase II trial (ClinicalTrials.gov, accessed on 7 December 2021, Identifier NCT04018872) will further explore the correlation of hedgehog pathway markers with therapy response by using itraconazole as a signalling pathway inhibitor.

Metabolic Dysregulation
Changes in mitochondrial protein expression alter NAT sensitivity and affect response [67,68]. Reduced mitochondrial electron transport chain (ETC) protein expression was correlated to increased cisplatin response [68], while overexpression of ATP5B, a marker of oxidative phosphorylation, demonstrated poor pathological response and radioresistance [67].
Oxidative stress contributes to oncogenesis and cancer progression by activating a plethora of transcription factors, such as p53 and NF-kB that have been independently investigated as predictive biomarkers [66,69,70]. Alterations to ETC proteins and ATP5B may reflect the same treatment-resistant cellular response caused by oxidative stress. However, ETC underexpression has only been validated in chemotherapy-only patients, whereas ATP5B as a radioresistance biomarker has only been observed in vitro [67,68]. Contrastingly, the thioredoxin interacting protein, a redox buffer and surrogate marker of oxidative stress, yielded no significant difference between NAT response groups in Woolston et al. [71]. Although this appears to contradict the hypothesis that oxidative stress alters NAT sensitivity, their patient sample included a potentially confounding group of GOJ and gastric cancer patients [71]. Despite preliminary studies demonstrating promising predictive potential for oxidative stress markers, further validation is required in studies using newer chemoradiotherapy regimens, such as CROSS or FLOT before any conclusions can be drawn [5,10].

Anti-Apoptosis
Two studies highlighted anti-apoptotic proteins that modulated chemo-or radio therapy response in preclinical settings [72,73]. Piro et al. demonstrated in OAC cell lines that downregulating BIRC3, a direct caspase inhibitor in apoptosis, with a transforming-growthfactor-β activated kinase 1 (TAK1) inhibitor increased chemo-and radiosensitivity [72]. The autophagy markers, LC3B and p62, taken together as an 'autophagic index', demonstrated non-response to chemotherapy if p62 was high, both alone or together with low LC3B [73]. Unfortunately, the study was limited by incomplete validation due to compromised tissue samples. Given these findings, the studies suggested TAK1 and p62 inhibitors to be investigated as potential sensitising agents in chemoradiotherapy [72,73]. Unfortunately, both studies lack external validation for BIRC3 and autophagy markers to be serious candidates for NAT response prediction. Higher BIRC3 expression in OAC compared to OSCC, however, could suggest that BIRC3 contributes to the divergence in NAT response between the two histological subtypes [47] and thus presents an avenue to elucidate molecular differences between the two cancers.
Moreover, two different studies found that SCCA1 overexpression increased chemoresistance [74,75]. Specifically, high SCCA1 levels were clustered in patients with higher TRGs [74]. SCCA1 is believed to have an anti-apoptotic affect against the lysosomal membrane permeabilisation response to cell stress [76]. In a recent externally validated preclinical study, SCCA1 has also been directly and inversely associated with peritumoural leptin and immune activation markers, respectively, as well as PD-L1 expression [75]. These findings suggest a role for SCCA1 in tumour, microenvironmental and immune cell crosstalk and may be a key marker for future studies in understanding these interactions.
Despite OAC's genetic differences to OSCC and gastric cancer, there is mounting evidence that increased nicotinamide N-methyltransferase (NNMT) enzyme expression may play a pivotal role in cancer growth, metastasis and chemoresistance [77]. NNMT is involved with the catabolism of structurally related compounds, such as nicotinamide and pyridine, allowing urinary excretion of these products [78]. Nicotinamide has been described to have an inhibitory effect on histone deacetylases and poly(ADP-ribose) polymerases (PARPs), which promote genomic stability in response to acquired DNA damage through genotoxic events, such as radiotherapy [79]. Therefore, it has been hypothesised that NNMT overexpression and subsequent reduction of intracellular nicotinamide in cancer stem cells removes PARP inhibition, conferring improved survival characteristics of cancer stem cells against DNA damage and programmed cell death [77].
In a preclinical study on cancer stem cells, Pozzi et al. identified NNMT to be associated with cancer stem cell enrichment in a variety of epithelial and mesenchymal cancer cell lines, including bladder, lung, colorectal and osteosarcoma [77]. Moreover, further work in melanoma cell lines found that NNMT silencing through enzyme knockdown conferred chemosensitivity, highlighting it as a potential molecular target for future chemotherapeutic agents [80]. These findings were also supported in OSCC cell lines by Cui et al. who demonstrated that NNMT upregulation may additionally confer chemoresistance though promoting the Warburg effect, though the underlying mechanisms that facilitate this interaction remain unclear [81]. Lim et al. also noted NNMT overexpression in gastric cancer tissue and hypothesised that post-translational modifications could explain this differential expression [82]. Although underlying molecular mechanisms are unclear, broader trends of NNMT overexpression and its effects on treatment resistance in a variety of cancer cell lines, including gastric cancer and OSCC, warrants further investigation into NNMT's role in facilitating treatment resistance in OAC and its potential as a biomarker to predict NAT response [82,83].

Loss of DNA Repair and Cell Cycle Regulation
Genes previously discussed, such as ERCC1 and TP53, have also had corresponding proteins examined [69,84]. Of note, p53 expression has demonstrated no correlation with NAT response in contrast to gene studies [60,84], while borderline significance was demonstrated in van Olphen et al. [69]. A possible reason for this divergent result in Fareed et al. may be due to the study's homogeneous treatment of p53 [84]. Recent studies suggest that significant differences exist for NAT response between groups that have normal p53 compared to a mutated or absent p53 [60,61,69]. Therefore, future biomarker studies must be wary of mutant p53. Moreover, further heterogeneity existed within chemotherapy regimens and patient histology, which could also have contributed to the discordance in results [84].

Molecular Panels
Langer et al. manually selected a panel of proteins based on prior studies to determine predictive value. From this selection, thymidylate synthetase, the target of 5FU, and MRP-1, a multidrug-resistance protein, demonstrated significant correlation with non-response to chemotherapy. ERCC1 was also studied but contrary to Fareed et al. did not show any associations with response [84,85]. Bronson et al. similarly selected a distinct panel of prognostic markers to predict pCR but discovered that marker expression was highly heterogeneous between patients and shared no correlation with NAT [86].  [85,86]. Future studies should determine their thresholds judiciously, as there is potential for false positives in Langer et al. and, inversely, a possibility for false negative in Bronson et al. This review did not find any 'true' molecular panel with collective predictive value.

Immunologic Biomarkers
Interactions between immune cells and the tumour microenvironment are a topic of recent research interest [87]. Although specific mechanisms are not yet understood, studies on predictive immunologic biomarkers are already underway.

Immune Cells
Neutrophil-to-lymphocyte ratio (NLR) is a novel predictive biomarker that was investigated in three recent studies [88][89][90]. The studies were unanimous in demonstrating that NLR was correlated with pCR. Two studies concentrated on pre-treatment NLR but were dichotomous with defining a threshold for significance, although they also differed in tumour subtypes, chemotherapy regimens and histopathological response thresholds [88,89]. Moreover, Al Lawati et al. only investigated NLR as a secondary outcome and was not able to provide an AUC-ROC [88]. Contrastingly, Sherry et al. measured concurrent NLR changes during chemotherapy and specifically found that a high NLR in the second week of NAT makes pCR less likely [90]. Future studies to elucidate the most optimal timings to measure NLR to maximise predictive accuracy may be indicated. Overall, NLR has thus far shown consistency in response prediction despite differences between studies, although a validation of its efficacy within an OAC-only prospective cohort treated by newer trimodal NAT regimes is indicated.

Immune Markers
Over the last few years, several novel immunologic biomarkers have been investigated in vitro to demonstrate radioresistance [91][92][93]. Most prominently, ADAM12 is a biomarker for cancer-associated fibroblasts (CAF) found in the tumour stromal microenvironment. Ebbing et al. showed that CAF-induced IL-6 conferred chemoradioresistance to in vitro cells treated with CROSS regimen [91]. High levels of ADAM12, as well as the complement C3 and leukaemia-inducible factor (LIF) presented in other studies are all associated with poor NAT response [92,93]. Of the three biomarkers, only LIF has been internally validated but with a small cohort [92], although ADAM12 may soon be externally validated in an upcoming clinical trial (ClinicalTrial.gov, accessed on 7 December 2021, Identifier NCT04554771). However as of now, each of these preliminary studies require external validation to confirm their effects. Furthermore, these immune markers can all be derived from blood samples, making them advantageously easier to obtain and monitor compared to tissue biomarkers.

Immune Panels
This review identified one immunogenetic signature investigated by Ghatak et al., which when combined with five differentially expressed genes in a 31 patient cohort, predicted NAT response with AUC-ROC of 0.96 [94]. Although these results are promising, they were presented in a conference abstract and therefore should be interpreted cautiously and should be verified in larger cohorts.

Blood and Serum Markers
Recent studies have shown interest in novel blood markers, such as circulating tumour DNA (ctDNA) and circulating tumour cells (CTCs) as well as adapting older biomarkers, such as CEA and CA19-9 to predictive uses in OAC [95][96][97]. In past studies, non-specific markers, such as plasma lipoproteins, albumin, fibrinogen and platelets have also been explored [98][99][100].

Circulating Tumour Markers
Circulating tumour DNA is a subset of total circulating cell-free DNA and has been touted, along with CTCs to have "liquid biopsy" potential [95,101]. Although there is some evidence to support the predictive value of CTCs and ctDNA in other cancers, there are few studies that investigate their efficacy in OAC [101,102]. Egyud et al.'s results were limited by a small dataset and a low ctDNA detection rate hampered further by the unavailability of DNA from patients for verification [95]. Despite demonstrating some evidence of ctDNA correlation with NAT response, study limitations prevent any robust conclusions from being drawn. As noted with other studies where blood draws are required, future studies require protocols to ensure homogeneity for when blood is collected [90,95]. This factor further contributed to the limitations of this study.
The predictive potential of CTCs has been more extensively studied in gastric and colorectal cancers, less so in OAC. Several barriers currently exist for widespread adoption of CTC detection, specifically, faster, cheaper and more accurate CTC detection devices are needed along with the establishment of CTC detection protocols [101]. This review identified a conference abstract by Seyedin et al., who used a nanotechnology-based CTC capture system to detect a reduction of CTC levels during NAT to predict response but again, limits with sample size and a heterogeneity in tumour types prevented meaningful conclusions from being drawn [97]. An upcoming case-control study (ClinicalTrials.gov, accessed on 7 December 2021, Identifier NCT02812680) aims to further assess CTCs' feasibility as a predictive marker. A contrastingly larger cohort in van der Kaaij et al. demonstrated that concurrently high CEA and CA19-9 levels were predictive of NAT non-response with high specificity [96]. If successfully validated in prospective studies, high CEA and CA19-9 could prompt a reconsideration of NAT in the context of other clinical findings.

Plasma and Serum Markers
Non-specific markers, such as plasma lipoproteins, albumin, fibrinogen and platelet counts have also been investigated for predictive potential in three studies [98][99][100]. The issue with non-specific markers is that probable confounders may influence their accuracy. Kelly et al. partially accounts for this by suggesting the use of a triple plasma protein panel [98]; however, none of these studies have externally validated their findings yet.

Adipocytes and Leptin
Tumour leptin as a biomarker was identified initially through gene expression analysis by Bain et al., who then found that higher leptin expression demonstrated significant association with NAT non-response as well as cisplatin resistance [103]. In vitro investigations involving leptin receptor antagonists subsequently demonstrated increased cisplatin sensitivity, which suggests that they could be useful in combination with cisplatin-based chemotherapy. Additionally, visceral obesity has been reported to increase radioresistance by stimulating spindle formation during anaphase [104]. Paradoxically, obesity does not appear to be predictive of poorer disease outcomes and, in fact, may even be associated with reduced risk of acute NAT toxicities [105]. Further studies should aim to clarify these findings.

Cancer Stem Cells
ALDH1 is a marker of cancer stem cells (CSCs), which are a chemoresistant cell population present in many cancers [106]. Ajani et al. scored each histological specimen based on ALDH1 presence to establish a labelling index which, when scored highly, suggested non-pCR [107]. This was congruent with expectations, as high ALDH-1 levels suggest high densities of chemoresistant CSCs. Given the strong predictive potential presented in this study, external validation should be pursued.

Tumour Proportions
The relative proportion of tumour cells to intratumoural stroma demonstrated efficacy as a predictive biomarker in Hale et al. [108]. Interestingly, high proportions of tumour to stroma were associated with a poorer NAT response [108]. Unfortunately, Hale et al. included both OAC and OSCC within the study, making it difficult to ascertain differences in tumour proportion between them. Future prospective studies would be useful to confirm these findings and clarify this difference between tumour subtypes.

Organoid Cultures
Organoid cultures are a novel in vitro technique that may facilitate preclinical studies that are better representations of NAT interactions within an in vivo tumour environment compared to conventional cell lines. The OPPOSITE study seeks to correlate in vivo to in vitro responses to NAT and elucidate potential biomarkers in the process (ClinicalTrials. gov, accessed on 7 December 2021, Identifier NCT03429816).

Limitations
This review was limited to articles published from 2010 onwards, as this allowed more focus to be given to studies using clinically relevant chemotherapy regimens. This may have caused a biased perspective on biomarkers that had more evidence published prior to 2010. However, there are several reviews that have a better focus on biomarkers published prior to 2010 and it is recommended that the evidence provided for older biomarkers be reviewed in conjunction with past reviews.
Secondly, although studies were critically appraised, a validated tool was not used. Instead, appraisal was guided by a 10-question appraisal process outlined by Young et al. to uphold rigor [109]. Nevertheless, objective quality comparisons between studies could not validly be made.

Conclusions
This review presents a heterogeneous landscape of predictive biomarkers that have been studied in the last decade. Although recent developments in imaging and immunologic markers have been promising, no biomarker currently has sufficient evidence to support clinical application. The shift from single biomarker studies to investigating biomarker panels or signatures will likely be driven by increasingly sensitive detection methods that will contribute dozens of differentially expressed biomarkers. Although concurrent validation of existing biomarkers needs to occur, tools, such as artificial neuronal networks, are also required to efficiently assess the predictive potential of a growing crowd of heterogeneous molecular biomarkers. To provide better clarity within the field, this review presents the following recommendations for future studies.

Minimise Heterogeneity Where Possible
• OAC and OSCC should be treated as different entities; future studies should either focus on a specific histological subtype or provide unpooled subtype analyses.

•
Tumour locations, such as oesophageal, gastro-oesophageal and gastric should similarly be differentiated. • TRG classification systems should aim for standardisation across the literature along with a definition of a 'good' and 'bad' response to allow for more valid comparisons between studies. To reduce the effect of interobserver variability with TRG between institutions, consider comparing Mandard's TRG1-2 to 4-5 for detecting biomarkers. • Consistency with NAT regimen between patients.

Clinical Validation
• All preclinical studies should aim to include an internal validation cohort. • All studies are recommended to pursue external validation.

Future Research Directions
• Biomarker panels may provide better-combined predictive potential than singular biomarkers, just as multimodality imaging has improved predictive value of imaging. • Development of artificial neuronal networks for multi-molecular biomarker panel assessment.

•
Use of organoid cultures to identify biomarkers of interest. • Investigation into the predictive potential of the biomarker NNMT in OAC. • Implementation of robust predictive models that integrate data from different diagnostic streams (e.g. radiological and pathological response indicators).     Borggreve [35] Biomarker studies with asterisk (*) are conference abstracts. Preclinical (with discovery cohort).
There are 67 differentially altered miRNA identified. Low Mir-187 was found in patients with poor NAT response. In vitro, miR-187 modulates radiation and cisplatin sensitivity and alters variety of pathways, including C3 serum levels (increased in poor responders). Supports C3 increase as predictive marker.
In vitro, miR-17-5p significantly sensitises radioresistant cells to radiation and promotes repression of genes with miR-17-5p binding sites. In vivo, miR-17-5p is significantly decreased with poor NAT responders. Subpopulation of cells had low miR-17-5p, high ALDH activity and increased radioresistance.
Lynam-Lennon [40]  Biomarker studies with asterisk (*) are conference abstracts.  Genomic gains in chromosome 14q11 and 19p13 were significantly associated with pCR. First biomarker study with independent validation with a focus on genomic gains.
Qian [62] Biomarker studies with asterisk (*) are conference abstracts.  High p62 cytoplasmic expression alone, or in combination with low LC3B was associated with NAT non-response. LC3B or p62 demonstrated no independent prognostic value post-NAT. Issues with biopsy specimens prevented tissue response prediction from being conducted.
Adams [73]  CKD9 inhibitor increases radiosensitivity of cells to prolonged DNA damage in vitro by enhancing G2/M arrest and apoptosis. Axl found as candidate biomarker for CDK-9 inhibition-Axl mRNA, and protein significantly reduced (52%) with CDK-9 use with radiation (p < 0.006).

3-4) Preclinical
CAF induces Il-6 secretion and drives epidermal-to-mesothelial transition in vitro which confers chemoradioresistance and increased migratory capacity. Il-6 inhibition resensitised cells to therapy. Since Il-6 is non-specific, ADAM12 was found in an 86 patient cohort as a more specific marker of stromal CAFs. High serum ADAM12 was correlated with poor response to NAT (CROSS).

3-5) Retrospective
Significantly higher PFR (plasma fibrinogen), CRP and PBPC (peripheral blood platelet count) levels were observed in patients with good TRG. Only PFR was an independent factor influencing tumour regression.
Ilhan-Mutlu [99] Biomarker studies with asterisk (*) are conference abstracts. Preclinical (with discovery cohort) Gene enrichment analysis was done to identify overrepresented pathways within a cohort of 520 differentially expressed genes in radiological non-responders vs. responders. Higher leptin protein expression was associated with lack of histological response to neoadjuvant chemotherapy (p = 0.007). Higher leptin protein expression was associated with resistance to cisplatin (p = 0.008) Bain [103]

4-5) RCT
Proportion of tumour cells per tumour area (PoT) was measured to predict chemotherapy response.
PoT between 40% and 70% received survival benefit from NAT. High pre-treatment PoT related to lack of primary tumour regression (TRG4-5).
Hale [108] 2021 Molecular markers 40 OAC/GOJ (estimate) CROSS -Prospective ClinicalTrials.gov Identifier: NCT03429816. Organoid cultures will be used to correlate in vivo to in vitro response to neoadjuvant chemotherapy. Molecular subtypes with histological response will be correlated to identify biomarkers.
Biomarker studies with asterisk (*) are conference abstracts.

ClinicalTrials.gov Search Strategy
Search terms: neoadjuvant therapy, oesophageal adenocarcinoma, adenocarcinoma of the esophagus, neoadjuvant, esophageal, etc. Applied filters: recruiting, not yet recruiting, active not recruiting, completed, enrolling by invitation and suspended.

of 50 included
Records Identified through Database Searching 531 Table A9. Inclusion and exclusion criteria.

Inclusion Criteria Exclusion Criteria
• All peer-reviewed full-text articles and conference abstracts that were published from 2010 onwards; • Articles where >50% of the patient sample had oesophageal adenocarcinoma requiring neoadjuvant therapy; • Neoadjuvant chemotherapy with or without radiotherapy is a regimen that is clinically accepted at the date of authorship: FLOT, MAGIC, CROSS or CALBG 9781; Variations within the same drug class were accepted (e.g., docetaxel and paclitaxel); • Biomarker to predict neoadjuvant therapy outcome was investigated.
• Articles where <50% of the patient sample had oesophageal adenocarcinoma were excluded if: Oesophageal adenocarcinoma and squamous cell carcinoma analysis was pooled; Gastroesophageal and gastric cancer analysis was pooled; • Non-English articles; • Review Articles.  Table A11. NTumour regression grade classification systems.