It is well recognized that genomic copy number can change during the cell life cycle. For example, the amplification of oncogenes, the deletion of tumour suppressor genes, or some other rearrangements may result in alterations to gene transcription [15]. Chromosomal instability, which includes the amplification and deletion of chromosomal segments or entire chromosomes, is a common feature of most CRCs. In an exploratory study, Molinari et al. [16] analysed alterations of chromosomal copy number and found that non-responders had a specific alteration profile. These included more frequent changes at the 18q23, 17p13-12, 13q31-34, 13q12, 10p14-13, 16p13, 8q23-24, 7p22-21, 7q21, 7q36, 3q29 and 2q21 chromosomal regions. Insights into the profile of chromosomal alterations could thus provide helpful information in predicting the response to nCRT, leading to an optimized strategy for RC. Chen et al. [17] also reported that chromosomal copy number alterations (CNAs) were associated with the response to nCRT. In their prospective phase II study, they revealed that non-pCR patients showed more frequent loss of chromosomal region 15q11.1-q26.3 (p 0.00002). In contrast, pCR patients showed more frequent loss of chromosomal region 12p13.31 (p 0.0003). The same authors also found that some specific CNAs were significantly associated with persistent lymph node metastasis following chemoradiotherapy in RC, in particular the loss of chromosome 4 [18]. Subsequently, other researchers found that chromosome segregation errors affected the response of RC patients to nCRT. They noted that errors in chromosome segregation predicted an enhanced pathological response to nCRT (OR = 3.9; p = 0.02). Those errors caused downstream structural damage to chromosomes that in combination with defects in DNA repair mechanisms enhanced the effect of DNA-damaging therapies [19]. These findings identified a potential mechanistic predictive marker of treatment response to nCRT and suggest that targeting of chromosomal instability may be an effective therapeutic strategy.

Various molecular pathways are involved in the development of CRC, of which DNA methylation is an important pathway. Although the development of RC has been widely studied, methylation profiling is seldom explored for its ability to predict response to nCRT. Molinari et al. [20] first analysed methylation profiles in biopsy samples from normal individuals and from RC patients and evaluated whether these could be used to predict nCRT. Only TIMP3 methylation status showed a significant difference within the four TRG classes (p 0.05). Another study found that the methylation level was significantly reduced in post-treatment resection specimens (171.5 vs. 152.7, p = 0.01). T stage (p = 0.005) and tumour regression (p 0.001) were strongly associated with the pre-treatment methylation level, which was also closely linked to complete and partial pathological responses (p = 0.01) [21]. Kim et al. [22] identified 7 hypermethylated CpG sites (ARHGAP6 cg07828380, KLHL34 cg01828474, ZEB1 cg04430381, STL cg00991794, DKK3 cg041006961, DZIP1 cg26886381, and DZIP1 cg24107021) that were able to predict nCRT response. These authors also showed that the methylation status of KLHL34 cg14232291 could predict radiosensitivity in patients who received nCRT. Preclinical studies have shown that cellular retinol-binding protein 1 (CRBP1) is related to radiation sensitivity. Following comprehensive molecular analysis, Yokoi et al. [23] found that hypermethylation of the CRBP1 promoter DNA resulted in epigenetic silencing and that histological responses in RC patients treated with nCRT were significantly associated with the quantitative methylation value of CRBP1 (p = 0.031). Recently, Canto et al. [24] analysed differentially methylated (DM) CPGs in 32 RC tissue samples and 5 normal rectal tissues. They found that three DM CpGs linked to the INSIG1, GPR1 and OBSL1 genes could be used in a classifier to sensitively and specifically identify patients who had a pCR.

Global genomics research into single-nucleotide polymorphism (SNP) variations and the initial stages of human genomic haplotype mapping both contributed to the discovery of cancer-promoting genes. This led to the identification of specific molecular characteristics as potential biomarkers to predict therapy response and prognosis [25]. Several studies have reported that polymorphisms in DNA repair genes are associated with the sensitivity of cancer patients to nCRT. For example, several groups investigated the potential association between thymidylate synthase (TS) polymorphisms and the response of RC to nCRT, but no consensus has been reached [26, 27, 28]. A meta-analysis was conducted by Yang et al. [27] to determine whether TS polymorphisms could predict the response to nCRT in RC. They found that patients with the TS 2R/3R genotype showed a positive response and that patients with the 2R/3R or 2R/2R genotype could benefit more from nCRT than patients with other genotypes. However, both the 1494del6 and the 5’-untranslated region expression allele polymorphisms showed little predictive value. Lamas et al. [28] collected blood samples from 93 stage II-III RC patients and determined their genotypes for TS, excision repair cross-complementing group 1 (ERCC1) and X-ray cross-complementing group 1 (XRCC1). An overall tumour response rate of 47.3% was observed, and the authors found that XRCC1 G/G carriers were more likely to show a better response than G/A carriers (odds ratio (OR) 4.18; 95% confidence interval (CI): 1.62–10.74, p = 0.003). In addition, higher expression of TS linked to the 3G/3G, 3C/3G and 2R/3G genotypes was associated with a better treatment response rate than lower expression genotypes (OR= 2.65; 95% CI: 1.10–6.39, p = 0.02) [28]. Kim et al. [29] found 9 SNPs that were associated with nCRT response. In particular, the reference allele (C) of the SNP CORO2A rs1985859 was more likely to be associated with a positive response than the substitution allele (T) (p = 0.01). In clinical analysis, the SNP FAM101A rs7955740 showed no relation to radiosensitivity, but dysfunction of FAM101A in RC cells in vitro was closely associated with early phase apoptosis and colony formation. Sebio et al. [30] analysed polymorphisms in epidermal growth factor receptor (EGFR) and its ligands, TS, and DNA repair genes in 84 patients with stages II-III RC who underwent nCRT. They found the rs11615 C $>$ T polymorphism in the ERCC1 gene and the rs11942466 polymorphism in the amphiregulin gene region (OR = 0.26; 95% CI: 0.06–0.79; p = 0.014) were significantly linked to pCR (p = 0.023). Moreover, the C/C genotype was a ssociated with resistance to nCRT. Researchers have evaluated SNPs in RC patients who participated in the phase III trial ACCORD-12, with the Dworak score used to assess therapeutic response. Sixty-six germline SNPs were found in 10 candidate DNA repair genes. Boige et al. [31] concluded that 5 SNPs located in ERCC1, MTHFR, excision repair cross-complementing group 2 (ERCC2) and XPA were closely related to the Dworak score. XPA rs3176683 was valuable for predicting the response to nCRT, while the ERCC2 rs1799787 and ERCC1 rs10412761 variants were reported to be promising prognostic markers. Of note, the likelihood of response was decreased by 60% in the T/G haplotype of rs10412761 and rs1799787 (p 0.001). Sclafani et al. [32] reported that a SNP (rs61764370, T $>$ G base substitution) in the let-7 complementary site 6 of KRAS messenger RNA (mRNA) was a potential biomarker for predicting response to nCRT. The TG genotype was more likely to result in complete response (CR) (p = 0.02). Rampazzo et al. [33] analysed 8 SNPs (rs11742908, rs2736108, rs2736098, rs2736100, rs2736122, rs2735940, rs2853690 and rs35241335) located in the regulatory and coding regions of the telomerase reverse transcriptase (TERT) gene. Their results showed the rs2853690AA/GG and rs2736108CC genotypes were associated with less telomere erosion and lower levels of circulating TERT following nCRT. Both genotypes were also potential biomarkers for better response to nCRT [OR= 4.6 (1.1–19.1) and 3.0 (1.3–6.9), respectively]. Low levels of circulating TERT ($\le$median) after nCRT were also good predictors of better treatment response. Chiang et al. [34] reported that E346A/rs867228 homozygosity of the formyl peptide receptor 1 (FPR1) was associated with poor 5-year OS (p = 0.014). They also found this FPR1 genotype could influence nCRT-elicited anticancer immunity in their animal model by decreasing T lymphocyte infiltration and migration. Moreover, the CC genotype of FPR1-E346A could independently predict the response to nCRT.

Studies have suggested that RC patients with or without p53 mutations respond differently to neoadjuvant radiation-based therapy. Whether p53 mutation is a reliable biomarker to predict therapeutic response to nCRT remains controversial [35, 36], with early studies reporting that it did not predict response to radiotherapy [35]. A meta-analysis published in 2012 showed that patients with wild-type p53 or with low expression levels of p53 protein were more likely to show pCR when treated with nCRT (poor response: RR = 0.85; 95% CI: 0.75–0.96; p = 0.007; good response: RR = 1.30; 95% CI: 1.14–1.49; p = 0.001; complete response RR = 1.65; 95% CI: 1.19–2.30; p = 0.003) [36]. Duldulao et al. [37] screened for TP53 and KRAS mutations in pre-treatment tumour biopsies and in paired normal surgical tissue from 148 stage II–III RC patients treated with nCRT. They concluded that mutations in TP53 and in different KRAS codons could influence the response to nCRT. For example, wild-type KRAS patients were more likely to show pCR than those with any KRAS mutation (p = 0.006), while KRAS codon 13 mutations were negatively associated with pCR (p = 0.03). Nevertheless, the role of KRAS mutation in predicting the response to nCRT is still controversial [38]. TP53 mutation was reported to be associated with radioresistance, while patients with both TP53 and KRAS mutations were more likely to show less response to nCRT and to suffer lymph node metastasis [39, 40]. Jiang et al. [41] analysed the genes of RC patients who underwent nCRT and found that BRAF and SMAD4 mutations were associated with positive response to chemoradiotherapy and better prognosis. Some researchers have attempted to construct models of predictive genotype signatures (PGS) to predict nCRT responses in RC. Xiao et al. [25] recently built a PGS model, based on target sequencing of 15 genes, whose predictive value was proved better than that of any clinical factor. Predictive models that contain multiple gene characteristics could therefore assist in the more accurate selection of patients who might benefit from nCRT, thus facilitating the personalization of treatment strategies.

Messenger RNA (mRNA), transcribed from a strand of template DNA, is a class of single-stranded RNA that carries genetic information to guide protein synthesis, thus gene alterations caused by nCRT can change the expression profiles of mRNA. These may in turn also be potential biomarkers for predicting therapeutic outcomes in RC. Some newly reported mRNA-associated biomarkers are listed in Table 2 (Ref. [42, 43, 44, 45, 46, 47, 48, 49, 50]).

Huh et al. [42] analysed pre-nCRT biopsies of RC patients to determine the predictive value of 13 tissue biomarkers: nuclear factor-kappa B, survivin, proliferating cell nuclear antigen, TS, p53, vascular endothelial growth factor (VEGF), matrix metalloproteinase-9, matrix metalloproteinase-2, cluster of differentiation 44 (CD44), CD133, thymidine phosphorylase, cyclooxygenase-2 and BCL2-associated X protein. These authors concluded that only CD44 mRNA expression was significantly predictive of therapeutic response (OR = 4.69, p = 0.030), even though it also correlated with expression of the other 12 markers (all p 0.05) [42]. Overexpression of ERCC1 mRNA was later reported to be associated with poor response to FOLFOX (5-fluorouracil, leucovorin, and oxaliplatin)-based nCRT in RC patients [43]. Patients with a higher level of C-C motif chemokine receptor 6 (CCR6) mRNA were prone to a poor response (p = 0.004), suggesting that CCR6 may be a biomarker for radiosensitivity and may also be a promising target for radiosensitization [44]. Yan et al. [45] found that an E3 ubiquitin-linked enzyme named RAD18 was a promising predictive biomarker for the efficacy of nCRT in RC. Overexpression of RAD18 mRNA helped to identify nCRT-resistant patients with an accuracy of 65%. Wang et al. [46] reported that a high level of chromodomain helicase DNA-binding protein 4 (CHD4) mRNA showed 60% accuracy for predicting nCRT resistance in patients with RC, while in vitro studies demonstrated that knockdown of CHD4 mRNA increased the sensitivity to nCRT in microsatellite stable (MSS) CRCs. Flores et al. [47] analysed blood samples of RC patients and evaluated TS mRNA levels in circulating tumour cells (CTCs). These authors found that 100% of non-responders expressed TS mRNA in all samples (p = 0.001). Cho et al. [48] developed a multigene mRNA-based biomarker model to predict nCRT response in RC patients. They found that the model performed well in predicting the response to nCRT, with an AUC of 0.84. Moreover, to evaluate the prediction stability of the model, internal cross-validation among three cohorts resulted in AUC values of 0.808–0.909. This suggests that the multigene, mRNA-based biomarker model may be valuable for identifying patients who could benefit from nCRT. Hur et al. [49] measured the expression levels of 7 mRNAs and used these results to develop a prediction model to predict response to nCRT. They found that expression of p21, p53, CD133 and Ki67 at the mRNA level was closely related to pCR. Furthermore, the prediction model could discriminate pCR with a preferable accuracy [50]. Recently, Ferrandon et al. [50] reported that a higher level of coenzyme A synthase (COASY) mRNA expression might be associated with radioresistance in RC. This was confirmed both in CRC cell lines and in independent patient cohorts. The authors also suggested that by regulating DNA repair and the status of phosphoinositide 3-kinase (PI3K), the level of COASY could serve as a predictive marker of radiation response. Besides, Li et al. [51] identified that 5 hub genes (YES1, PPP2R5C, PPP2R1B, PDPK1 and KRAS) were associated with the response of RC patients to nCRT.

Proteins are the most sophisticated and crucial biomacromolecules in the human body and are the ultimate product of gene expression. Previous studies have reported that epigenetic and/or genomic mutations, as well as alterations in the transcription process can result in changes to protein expression. The level and spectrum of certain proteins might therefore indicate vital signs of the internal environment and of some diseases. Moreover, alterations in protein expression could be used as potential markers of the response to cancer treatment. Researchers have therefore explored the ability of proteins to predict the response to specific treatments. By using prior knowledge of genes and transcriptomics, it has been revealed that some proteins are linked to the radiosensitivity of RC and to clinical outcomes. Usually, researchers focus on a single protein or protein panel in which the expression level can be assessed by immunohistochemistry (IHC) or western blot (WB) analysis of biopsy tissues or serum samples, respectively.

Certain proteins with predictive potential are involved in the process of DNA repair. These include meiotic recombination 11 homolog A, ataxia telangiectasia mutated (ATM), proliferating cell nuclear antigen-associated factor 15 (Paf15), ERCC1, X-ray repair cross-complementing protein 2 (XRCC2), and cell cycle proteins such as cyclin D, vaccinia-related kinase-1 and -2, and polo-like kinase 1 (Plk1). Other proteins contribute to cell proliferation, including nuclear factor-$\kappa$B (NF-$\kappa$B), Golgi phosphoprotein 3 (GOLPH3), proliferating cell nuclear antigen, EGFR, fibroblast growth factor receptor 4 (FGFR4), c-MYC, VEGF, Ki67 and focal adhesion kinase (FAK). Yet other proteins participate in apoptosis, including B-cell CLL/lymphoma 2 (Bcl2), cyclooxygenase-2 (COX-2), BCL2-associated X protein (Bax), p53, p21, survivin and apoptotic protease-activating factor 1. Still other proteins serve as tumour biomarkers (carbohydrate antigen 19-9, carcinoembryonic antigen) or contribute to metabolism, such as 3-hydroxy-3-methylglutaryl coenzyme A synthase, transketolase, hydroxyacyl-CoA dehydrogenase, 17-$\beta$-hydroxysteroid dehydrogenase type 2 and vascular non-inflammatory molecule 1. All of these proteins have been widely studied and reviewed with regard to their predictive significance in RC. However, the results are controversial and no consensus has been reached [13, 14, 72]. Herein, we focus mainly on newly identified proteins from biopsies and blood which were mainly listed in Table 4 (Ref. [73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84]).

Pucci et al. [73] reported that aberrant Ku70, Ku80 and sClusterin (a partner of Ku70) expression were significantly associated with radioresistance and may form a potential “cluster” of predictive factors for nCRT response in patients with RC. Another study examined differential gene expression between RC and normal tissue and identified 8 genes with a $>$16-fold difference (TAP2, SLC39A7, PSMB8, PPP1R18, PPBP, KRBOX1, HSPA1B and B3GALT4). The expression of these proteins was closely associated with the therapeutic outcome of nCRT (p 0.0005) [74]. Chen et al. [75] investigated the association between chloride channel accessory 1 (CLCA1) expression and nCRT response and found that high levels of CLCA1 expression predicted vascular invasion (p = 0.028), pre-treatment lymphatic metastasis (p = 0.032), and poor therapeutic response (p = 0.042). The transcription factor DEK is expressed in RC and is thought to be a reliable biomarker for achieving pCR following nCRT. DEK may be related to the pro-apoptotic factor P38 [76]. Expression of the transforming acidic coiled-coil protein-3 in RC patients was found to negatively correlate with sensitivity to nCRT (RR = 2.23, 95% CI: 1.44–3.45; p = 0.001), resulting in less pCR (p = 0.001) [77]. Yan et al. [45] reported that low expression of RAD18 (an E3 ubiquitin-linked enzyme) in pre-nCRT biopsies of RC patients correlated to better clinical outcomes after treatment. This may be linked to stimulation of the caspase-3- and caspase-9-mediated apoptotic pathway, leading to increased cell apoptosis. Using an animal model, these authors further demonstrated that dysfunction of RAD18 prevented the tumour from growing when exposed to 5-FU and/or irradiation in vivo. Besides, high expression of CHD4 was reported to be significantly associated with tumour regression grade (p = 0.001). In vitro studies revealed that CHD4 contributes to radio-resistance in CRC patients with microsatellite instability-high (MSI-H) tumours, while a lack of CHD4 expression enhances radio-sensitivity in microsatellite stable (MSS) CRC patients [46].

The forkhead box (FOX) family members FOXK1 and FOXK2 participate in cell proliferation and carcinogenesis. Recently, Zhang et al. [78] evaluated the expression of these proteins in RC and their ability to predict nCRT responses. They found that lower levels of FOXK1 and FOXK2 expression were detected in patients with pCR (p 0.05). In another study, Hsu et al. [79] found that overexpression of B-cell-specific Moloney murine leukemia virus insertion site 1 (BMI 1) was significantly associated with post-treatment tumour stage (T1–T2; p = 0.015), advanced pre-treatment nodal status (N1–N2; p 0.001), poor therapeutic response (p = 0.001) and prognosis. In addition, the lack of BMI 1 expression could modulate the expression of Kruppel-like factor 4 to enhance the radio-sensitivity of MSS CRCs. The ATF3-driven overexpression of serum- and glucocorticoid-regulated kinase 1 (SGK1) leads to radio-resistance in non-pCR RC patients [80]. Park et al. [81] evaluated the gene expression of tissue samples in RC patients and found that a signature of 9 genes (Table 4) could identify responders with relatively good accuracy, specificity and sensitivity. This signature was also pertinent for clinical and pathological features and could serve as a useful biomarker for predicting the response to nCRT.

High expression levels for the V-set and transmembrane domain-containing 2 like protein in RC have been associated with the therapeutic results of nCRT (p = 0.030). These proteins are thought to contribute to radio-resistance in an IL-4-mediated signalling pathway by downregulating $\gamma$-H2AX expression and modifying the progression of cell apoptosis and proliferation [82]. Fratte et al. [83] investigated the expression of 11 tumour-related proteins (RAD51, CD44, HIF1, CXCR4, COX2, Ki67, GLUT1, CA-IX, VEGF, CXCL12, and MLH1) in pre-nCRT tissue biopsies of RC patients. These authors report that pre-treatment expression levels of RAD51, HIF1, CXCR4, COX2 and Ki67 were predictors of the response to nCRT. Moreover, Ki67 expression together with that of CXCR4 could increase the predictive ability. SMAD7 is a member of the SMAD family and a key mediator of the intracellular signalling cascade, especially the TGF-$\beta$ pathway. SMAD7 expression was significantly elevated in the primary tumour tissue of responders to nCRT (p = 0.014) [84]. In addition, Koyama et al. [85] identified 350 differentially expressed genes, of which 199 were downregulated and 123 were upregulated in poor responders. These authors also found that inhibition of Akt activation improved the therapeutic response to CRT. Another genome-wide RNAi screening study identified FICD, NHP2, LDLRAD2, SYNE3, NCAPH, and replication factor C subunit 4 (RFC4) as potential biomarkers of radio-resistance in RC. Further analysis revealed that RFC4 expression in tissue samples was associated with a poor response to nCRT in RC patients and with worse prognosis. The authors speculate this may be due to the promotion of DNA repair by non-homologous end joining (NHEJ), including RFC4 and Ku70/Ku80 [86]. However, a recent gene expression-based study showed opposite results. Momma et al. [72] screened for potential markers of the response to nCRT by analysis of pre-CRT tissue samples from 53 responders and 61 non-responders. The researchers also employed a further six independent datasets that included 99 responders and 176 non-responders in order to validate the predictive values of the signatures. Surprisingly, they found that current signatures based on gene expression and identified through microarray platforms were not sufficiently robust to predict nCRT response and thus to influence decision-making in the treatment of RC.

Metabolites are generated by various types of biochemical reactions in the body and also participate in the pathophysiological process of tumourigenesis and progression. Recent work in laboratory medicine has applied metabolomics for the classification of tumours and for the evaluation of treatment response, recurrence and prognosis. This has allowed metabolomics to produce biomarkers in various fields [87]. Early studies showed that serum concentrations of phosphoenolpyruvic acid, hypoxanthine, myo-inositol, creatine, and glycine were potential metabolic biomarkers for predicting response to CRT and prognosis in RC [88, 89]. A prospective cohort study (NCT03149978) by Jia et al. [87] found that RC patients who underwent nCRT could be discriminated between responders and non-responders by conducting metabolomic analyses of their serum. Potential responders could be identified using 15 differentially expressed metabolites. These results suggest that complete resection for all patients is not required in order to achieve the desired clinical and therapeutic outcomes. Using proper evaluation, potential non-responders could avoid unnecessary nCRT and oncologists could directly advise patients for whom surgery is the preferred option. Tomàs et al. [90] recently conducted a study of RC patients to identify biomarkers related to energy metabolism and to circulating levels of paraoxonase-1 that are able to predict nCRT responses. A low pre-nCRT level of plasma valine was associated with pCR (AUC = 0.826), while low concentrations of succinate were associated with relapse (AUC = 0.833). The content of serum exosomes has the potential to predict responses to neoadjuvant radiotherapy in RC. Recent work detected 129 metabolites in the plasma and exosomes, of which 23 differentially accumulated metabolites (DAMs) were present at significantly different levels between responders and non-responders. The exosome levels of pentadecanoic acid and of sucrose were higher in poor responders. Proteome components of serum-derived exosomes are also promising biomarkers, and the combination of proteomic and metabolomic biomarkers is likely to be a fruitful area of investigation [91].

The regulation of immunity is complex, with many researchers investigating the connection between circulating bio-macromolecules and the immune system. Once validated, serum biomarkers could be more applicable clinically than biopsy biomarkers due to their non-invasive characteristics.

A prospective study analysed 9 biomarkers (25-OH-vitamin D, osteopontin, CA IX, IL-6, IL-8, CRP, LDH, CA19-9 and CEA) of RC patients who underwent CRT. The authors found that IL-8 (OR 0.94, p = 0.036) and CEA (OR= 0.97, p = 0.029) were significant predictors of response [101]. The levels of C-C motif chemokine ligand-5, TNF-$\alpha$ and soluble CD40-ligand were reported to be excellent markers of a good response to nCRT [102]. The inflammatory response was also found to be associated with the genesis and prognosis of numerous cancers [115]. The neutrophil-to-lymphocyte ratio (NLR) has been widely investigated for the prediction of nCRT response in RC. However, a consensus has yet to be reached on the time points to measure NLR and on the NLR cut-off values (used to determine whether a test is positive or negative) to be used. The time points for evaluation of pre-CRT and post-CRT outcomes were also limited [13, 14]. In a retrospective study of 1052 RC patients, both a lower LMR (p = 0.0001) and a higher NLR (p = 0.0001) were associated with the response to nCRT. An NLR value of $\ge$3.11 was associated with a lower likelihood of achieving complete total mesorectal excision or of preserving the sphincter [116]. Elevated pre-nCRT NLR was also associated with poor therapeutic response to nCRT [116]. Dreyer et al. [117] evaluated the modified Glasgow prognostic score and the albumin, C-reactive protein and hemoglobin levels of patients with RC. These workers found that the body’s inflammatory status prior to nCRT was associated with a poor response to nCRT [118]. More recently, the neutrophil $\times$ monocyte (N $\times$ M) ratio and the lymphocyte-to-C-reactive protein ratio (LCR) were found to be markers of a good therapeutic response to CRT (p = 0.005 and p = 0.016, respectively) [103]. These biomarkers are readily accessible and economically feasible, making them applicable to clinical practice. In combination with other validated biomarkers or clinicopathological characteristics, these biomarkers could be used to design more personalized treatment strategies for RC patients.

Haemoglobin is a kind protein responsible for carrying oxygen in higher organisms. The role of hemoglobin in predicting response of certain therapeutic strategies has been widely investigated. McGrane et al. [120] analysed blood samples from 273 patients with RC and treated with nCRT. They found that a hemoglobin level of 120 g/L (anemic) at presentation correlated with a higher regression grade and thus an inferior response to nCRT (p = 0.006). These workers also reported that patients with anemia had higher mortality rates than non-anemic patients (HR 1.73; 95% CI: 1.05–2.86). Multiple lines of evidence show that platelets (PLTs) have a strong influence on the progression and metastasis of tumours, with the effects mediated by a variety of mechanisms. PLTs promote epithelial-mesenchymal transformation, thus allowing cancer cells to escape immune surveillance, obstruct the microvasculature, and stimulate angiogenesis [113, 121]. A retrospective study investigated the role of PLT count in predicting the clinical outcomes of nCRT. Patients with an elevated PLT count showed a lower pCR rate compared to those with lower counts (12.8% vs. 22.1%, respectively; p = 0.001) [122].

Lymphocytes are extremely sensitive to radiation. Radiation therapy therefore leads to exhaustion of lymphocytes in the hematopoietic system, with radiation doses of 1 Gy being sufficient to directly destroy lymphocytes in the circulatory system [123]. Radiation-induced lymphocytopenia (RIL) may counteract the anti-tumour effects of radiotherapy and has recently become a promising area of research. In solid tumours, RIL was shown to be significantly associated with the therapeutic response, PFS, and even OS [123, 124]. Heo et al. [124] investigated whether the level of circulating lymphocytes during nCRT could predict therapy response in patients with RC. They found a sustained blood lymphocyte count (lymphocyte count at 4 weeks/baseline lymphocyte count $>$0.35; OR = 8.33, p = 0.02) during CRT was significantly associated with pCR. Liu et al. [125] evaluated the absolute lymphocyte count (ALC) of RC patients during nCRT. They found that a high ALC nadir was associated with pathologic response (OR= 4.32, 95% CI: 1.22–15.26, p = 0.023) and suggested that ALC may serve as a stratification marker for RC patients scheduled to receive nCRT. Recent studies have suggested that micronuclei frequency (MNf) in peripheral blood lymphocytes may be a prognostic biomarker for monitoring the response of RC patients who undergo induction chemotherapy and then operation. However, it is not known whether MNf could also serve as a predictive biomarker of the response to chemoradiotherapy [126]. Other indexes based on changes to circulating lymphocytes have also been suggested as potential biomarkers of the response to nCRT. These include the platelet-to-lymphocyte ratio, neutrophil-to-lymphocyte ratio, lymphocyte-to-monocyte ratio, and lymphocyte-C-reactive protein ratio [127, 128].

Circulating tumour cells (CTCs) are released into the bloodstream by primary tumours which then could metastasize to distant organ sites. Moreover, CTCs are a minimally invasive and preferred alternative to tumour biopsy [14, 119]. The clinical utility of CTCs in RC has been widely investigated. Most studies have confirmed that a high level of CTC counts correlates with poor prognosis [119]. When integrated with tumour characteristics, CTC counts could be used as a substitute for primary tumour cells. Previous research demonstrated that cytokeratin 20-positive CTCs were predictors of the response to nCRT in RC patients [129]. Troncarelli et al. [47] reported that TS expression was completely absent in CTCs from patients with a pCR (p = 0.001). In contrast, CTCs from 83% of non-responders expressed TS (p 0.001). Furthermore, RAD23 homolog B (RAD23B), an excision repair protein, was expressed in the CTCs of non-responders. These authors concluded that TS/RAD23B and/or TS mRNA expression in CTCs were possible biomarkers for predicting the response to nCRT.

It has been hypothesized that cell-free DNA (cfDNA) results from cell lysis caused by apoptosis or necrosis of cells [14]. In a study of 34 RC patients, Sun et al. [130] analysed cfDNA for the methylation status of the O6-methylguanine-DNA methyltransferase (MGMT) promoter region and to screen for KRAS mutations. They concluded that MGMT promoter methylation status, the 400-/100-bp DNA ratio, and the 400-base pair DNA concentration in the baseline cfDNA were helpful for predicting the response to nCRT. Schou et al. [131] analysed cfDNA in plasma samples of RC patients treated with nCRT and surgery. These authors concluded that cfDNA had the potential to improve pre- and post-nCRT risk assessment and thus promote the development of personalized therapy in RC.

Circulating tumour DNA (ctDNA) represents 0.01%–10% of the total cfDNA and has become a promising biomarker in various types of cancers [118]. Numerous studies have shown the value of ctDNA genomic alterations or ctDNA concentrations in tumour diagnosis, monitoring of treatment response and of resistance, selection of targeted therapy, and the detection of residual disease [119, 132, 133].

A Japanese study investigated the clinical utility of ctDNA to predict nCRT responses and post-operative recurrence in RC patients. ctDNA levels at baseline and after nCRT were found to be significantly different (p = 0.0003). The authors concluded that the change in ctDNA was independently associated with pCR (p = 0.0276) [132]. A prospective multicentre Chinese trial further explored the value of ctDNA for monitoring tumour burden, predicting nCRT response, and predicting survival. Zhou et al. [134] found that baseline ctDNA levels were strongly associated with positive nCRT responses and that MFS was closely related to the median variant allele frequency (VAF) of mutations in the baseline ctDNA (HR 1.27, p 0.001). They also demonstrated that ctDNA closely mirrored the tumour burden and served as a real-time monitoring indicator. A recent meta-analysis of 21 publications concluded that higher baseline levels of the longer fragments of cfDNA, certain methylated genes and tumour-specific mutations, and the integrity index were potential predictors of poor response to treatment. Furthermore, undetectable baseline ctDNA levels and a decrease in common RC mutations during nCRT (dynamic monitoring) may be robust indicators of pCR [133].

Computed tomography (CT) is extremely important for diagnosing and staging cancers and for evaluating the efficacy of certain therapeutic strategies. Over the past few years, researchers have been exploring the role of CT imaging in determining the prognosis of RC and for predicting the response to nCRT. A retrospective study (n = 95) analysed the texture features of pre-nCRT CT images, including standard deviation, skewness, kurtosis, uniformity and entropy. Chee et al. [135] concluded that features of homogeneous textures were associated with better nCRT responses. Other studies have also suggested that kurtosis and fractal dimension (FD) are potential CT-derived biomarkers for predicting nCRT responses [136, 137]. Perfusion CT imaging was also investigated in a prospective study of RC treated with nCRT. The results showed that hot-spot blood volume and a decline in hot-spot permeability were significant predictors of pCR outcome (p 0.0001) [138]. Texture analysis contributes to the assessment of heterogeneity of medical images by analysing grey-level intensities on a pixel-by-pixel basis. Some researchers have constructed prognostic models of RC by analysing pre-treatment, contrast-enhanced CT textures to identify patients with poorer down-staging prospects. These could then be offered more intensive treatment via a higher radiation dose or by using other strategies [139]. Another retrospective study evaluated the total subcutaneous, visceral, mesorectal and abdominal fatty tissue components based on the findings of CT images. Dilek et al. [140] demonstrated that a cut-off value of $\ge$69.4 for mesorectal fat tissue volume (MRV) was associated with positive nCRT responses. In ROC analysis, the specificity was 58.5% and the sensitivity was 82.9% (AUC = 0.75 (0.65–0.84), p 0.001).

F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography (18F-FDG-PET/CT) is a helpful imaging tool used extensively in clinical oncology for disease diagnosis, staging, evaluation of therapeutic efficacy, diagnosis of relapse, identification of the underlying lesion, and long-term follow-up. Researchers have explored FDG-PET/CT-derived imaging markers for predicting the response of RC patients to nCRT by studying pre- and post-CRT radiological characteristics and correlating them with the response to nCRT. The most challenging aspect was to choose the best parameter for semi-quantitative analysis to assess response so that it could be used as a predictive biomarker. Previous studies in patients with RC have shown that changes in the standard uptake value ($△$SUV), SUVmax, or %$△$SUV may predict positive nCRT responses [55]. Lovinfosse et al. [141] analysed PET/CT images obtained before nCRT in RC patients and evaluated the histogram-intensity features, total lesion glycolysis (TLG), metabolic tumoural volume (MTV), maximum and mean standard uptake values, and 11 regional and local textural features. These authors concluded that TLG could be a robust marker of good therapeutic responses to nCRT (TRG 3-4) [141]. Early total lesion glycolysis (TLG-early) and its percent change compared to baseline ($\mathrm{\Delta }$TLG-early) were also reported as potential biomarkers to discriminate responders from non-responders. Furthermore, $\mathrm{\Delta }$TLG-early exhibited the highest accuracy for response prediction, especially in high-risk RC patients treated with nCRT using bevacizumab [142].

Using baseline FDG-PET/CT and metric learning (ML), Wu et al. [143] built a novel artificial intelligence (AI) model to predict the response to nCRT in 236 newly diagnosed RC patients. ML determines the dissimilarity or similarity between objects in an AI-mediated manner based on a distance metric. The authors found that the model had a preferable accuracy, specificity and sensitivity in predicting therapeutic response of nCRT. Using the AI-mediated ML model, baseline FDG-PET/CT images were concluded to be robust biomarkers for predicting nCRT responses in RC. The AI-based ML model is non-invasive and may allow significant progress in personalized, precision therapy.

Magnetic resonance imaging (MRI) is also a helpful imaging tool used extensively in clinical oncology which has several desirable characteristics, particularly its exceptional resolution and contrast definition between the lesions and surrounding tissues. These have contributed to wide application of MRI in the disease setting, including the evaluation of response to nCRT in RC [55, 144]. Although the alterations in tumour morphology occur later than other changes at the molecular and biological levels, MRI can still accurately assess tumour characteristics. The potential role of MRI in predicting response to nCRT in RC has been thoroughly investigated.

Dynamic contrast-enhanced MRI (DCE-MRI) has been employed extensively to obtain functional imaging. This can in turn be used to provide useful information on vascular and tissue permeability. Kim et al. [145] reported that a decrease in the tumour perfusion parameter (Ktrans) was significantly associated with good response to nCRT in RC (p = 0.0007). In contrast, none of the other parameters examined were effective at predicting the efficacy of nCRT [145]. Ciolina et al. [146] confirmed that pre-Ktrans predicted response to therapy, while wash-out and Kep measured before nCRT correlated with RC grading.

Diffusion-weighted MRI (DW-MRI) detects water proton mobility in tissues to provide information on microscopic structures. This technique has also been applied to monitor treatment and to predict nCRT response in RC. The apparent diffusion coefficient (ADC) in DW-MRI is a useful instrument for quantitative analysis owing to its various characteristics, including cellularity, tumour proliferation, tumour necrosis, tumour grade, extracellular space tortuosity and tissue organization [55]. However, the use of ADC as a predictive biomarker of nCRT response has been controversial, as discussed elsewhere [55, 144]. A prospective study showed that texture features of RC on T2-weighted (T2w) magnetic resonance images were potential imaging biomarkers for the clinical outcome of nCRT. The authors found that pre-treatment medium texture-scale quantified as kurtosis, mid-treatment kurtosis without filtration, and changes of kurtosis in the pre-treatment and mid-treatment images differed significantly between the PR+NR and pCR patient groups (p = 0.045, 0.038, and 0.01, respectively). In particular, the ROC value in pre-treatment kurtosis was significantly higher than all other parameters (0.907, p 0.001) [147]. Another prospective observational cohort study found that a lower baseline tumour blood flow from dynamic susceptibility contrast MRI was associated with better response to nCRT (p = 0.01) [148].

Recently, a multivariable model incorporating mrT stage and quantitative parameters from baseline MRI, including T2w volume and T2w signal entropy, was used to distinguish responders from non-responders prior to nCRT in RC [149]. The combination of T2-weighted MRI volumetry, diffusion-weighted imaging (DWI), and 18F-FDG PET/CT obtained before CRT and before surgery was useful for predicting therapeutic results in RC [150]. Furthermore, quantitative imaging outperformed molecular markers for the prediction of response to nCRT. These findings may help physicians in selecting the most appropriate patients for organ preservation in RC cases.

Radiomics, which converts digital images into quantitative data, is based on the concept that radiological images are comprised of data that reflects underlying pathophysiology. Advances in science and technology have allowed the extraction of large amounts of quantitative data from tomographic images (CT, MR or PET images). The end goal of radiomics is to generate imaging biomarkers as decision support tools for clinical practice [151]. The implementation of radiomics has progressed rapidly as our understanding of tumour biology has improved. This has in turn contributed to the introduction of precision medicine. Recently, several studies evaluated the potential role of radiomics as a biomarker for predicting tumour responses in RC.

One study explored the clinical and pre-nCRT multi-parameter MRI features of 186 RC patients in order to develop and validate a radiomics model for predicting therapeutic response. Cui et al. [152] concluded that their pre-treatment radiomics-based model was of great value for predicting pCR and could potentially guide the selection of patients for a “watch-and-wait” policy. Another study showed that MRI and FDG-PET radiomics features could serve as potential biomarkers. A logistic regression model comprised of six second-order texture features (one from T2w MRI: T2w correlation and five from PET: metabolic volume, glycolytic volume, PET 10th percentile, PET homogeneity, PET contrast) gave the best results for predicting nCRT responses in RC patients (AUC = 0.86) [153]. By comprehensively analysing pre- and post-nCRT MRI data, Liu et al. [154] developed a radiomics model that showed excellent performance and could be employed as a non-invasive and individualized tool to predict cases of pCR in patients treated with nCRT. An international multicentre study of RC patients who underwent nCRT and total mesorectal excision showed that pre-treatment, MRI-based radiomics of RC and/or the characteristics of the surrounding mesorectal compartment were able to predict TRG, the neoadjuvant rectal (NAR) score, and pCR. Both the mesorectal compartment and tumour provided useful and authentic information on therapeutic response and prognosis [155]. More recently, elevated heterogeneity in skewness maps of baseline tumours in T2w-based radiomics has been shown to correlate with nCRT responses [156].

Variation in the collection and analysis of imaging data has impeded the clinical use of imaging biomarkers. Hence, standard and independent software is urgently needed to acquire original data and to further analyse the post-processing data. The primary obstacles with regard to radiomics are the optimal collection and combination of diverse multimodal data sources in a quantitative fashion, the variety of radiomic features, unbalanced datasets, and having few observations. Moreover, the study populations are usually small and limited to single institutions, and the reproducibility of techniques has seldom been investigated. Consequently, the usefulness of the above imaging biomarkers requires further validation and larger prospective studies are warranted before applying these biomarkers as predictive factors in clinical trials.