Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Expression of CXC-chemokine receptor 4 (CXCR4) is associated with poor prognosis in multiple cancers. A small portion of patients with colorectal cancer (CRC) are eligible for immune checkpoint blockade (ICB) therapy.WHAT THIS STUDY ADDS High CXCR4 mRNA expression was associated with poor survival in CRC, but was associated with improved outcome after ICB therapy. Furthermore, high CXCR4 expression was positively associated with multiple ICB biomarkers.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Combined evaluation of CXCR4 expression with other immunotherapy biomarkers could help refine stratification of patient selection for ICB therapy in CRC.Introduction Colorectal cancer (CRC) is the third most common cancer globally, with liver metastasis being the leading cause of mortality among these patients. 1 2 For patients with liver-limited disease, intent-to-cure resection of hepatic metastasis provides the greatest survival benefit. Approximately 20% of these patients achieve a cure after liver resection; however, most patients will develop disease recurrence within the liver or at other distant sites, and novel systemic treatments are needed.3 While immunotherapy has revolutionised cancer treatment in many solid tumours, leveraging this strategy in microsatellite stable/mismatch repair proficient (pMMR) CRC remains a unique challenge. Although CRC patients with mismatch repair deficient (dMMR) tumours are often responsive to immune checkpoint blockade (ICB),4 less than 10% of metastatic CRC cases are dMMR.5 Therefore, it is critical to better understand the molecular landscape of CRC and the tumour microenvironment (TME) to uncover novel therapeutic targets and biomarkers of response.Chemokines are small, secreted proteins mostly known for their role in mediating immune cell (IC) trafficking and are considered to have a major impact on the cellular composition of the TME. CXC-chemokine receptor 4 (CXCR4) is a ubiquitous G protein-coupled cell surface receptor that is activated by C-X-C motif chemokine ligand 12 (CXCL12).6The activation of this axis plays an essential role in IC trafficking and physiological processes such as chemotaxis, migration, cell adhesion, proliferation and survival.7–9 Aberrant CXCR4 upregulation has been implicated in tumour proliferation, invasion and metastasis in various solid tumours including CRC.10 11 Clinical studies have shown that CXCR4 overexpression in early (stage I/II) CRC has been associated with higher risk of locoregional recurrence and/or liver metastasis. CXCR4 expression has been demonstrated to be upregulated in liver metastases compared with primary tumours.12 In patients with stage IV disease, CXCR4 overexpression has been shown to be an independent predictor of poor overall survival (OS).13 Furthermore, in patients undergoing liver resection, lack of CXCR4 expression has correlated with improved recurrence-free survival.14 However, associations between CXCR4 mRNA expression and molecular features such as genetic and immune landscape have not been studied in CRC.Within the TME, CXCR4 expression was observed in both human and murine models of solid tumours to influence tumour growth through regulation of IC migration.15 A recent study reported CXCR4 overexpression as a predictor of immunotherapy response in lung cancer;16 however, such a relationship has not been thoroughly explored in CRC. Notably, recent studies have provided evidence of enhanced immune response mediated by CXCR4 inhibition in CRC,17 which supports further investigation of CXCR4 as a potential immunotherapeutic target. Therefore, the goal of this study was to interrogate the role of CXCR4 mRNA expression on the TME and its potential prognostic and predictive value for immunotherapy strategies in patients with CRC. Using a large cohort of CRC tumours and clinical dataset, we compared genomic, TME and clinical characteristics between groups with high versus low CXCR4 expression. Although our results validate previous findings that high CXCR4 mRNA expression is a predictor for poor survival in CRC, we also describe its association with improved outcome after ICB treatment, indicating its potential predictive value in helping to inform future immunotherapeutic strategies in CRC.Methods Analysed cohort A total of 15 026 CRC tumours, collected from primary tumour location and distant metastatic sites, were submitted to Caris Life Sciences (Phoenix, Arizona) for comprehensive molecular profiling including DNA, RNA and protein analysis. This was a retrospective analysis in which all consecutive CRC cases between 2008 and 2023 with available molecular profiling and survival data were included.Immunohistochemistry Immunohistochemistry (IHC) was performed on formalin-fixed paraffin-embedded (FFPE) sections of glass slides. The slides were stained using automated staining techniques, per the manufacturer’s instructions, and were optimised and validated per Clinical Laboratory Improvement Amendments/College of American Pathologists and International Organization for Standardization requirements. The staining was scored for intensity and staining percentage. Results were categorised as positive or negative by defined thresholds based on published clinical literature that associates biomarker status with patient responses to therapeutic agents. A board-certified pathologist evaluated all IHC results independently. The primary antibody used against PD-L1 was SP142 (Abcam Cat No ab228462, RRID:AB_2827816). The staining was regarded as positive if its intensity on the membrane of the tumour cells was 2+ (on a semiquantitative scale of 0–3: 0 for no staining, 1+ for weak staining, 2+ for moderate staining or 3+ for strong staining) and the percentage of positively stained cells was >5%.Next-generation sequencing Next-generation sequencing (NGS) was performed on genomic DNA isolated from FFPE tumour samples using the Illumina NextSeq 500 Sequencing System (RRID:SCR_014983). Prior to molecular testing, tumour enrichment was achieved using manual microdissection techniques. Matched normal tissue was not sequenced. A custom-designed SureSelect XT assay was used to enrich 592 whole-gene targets (Agilent Technologies, Santa Clara, California). For additional subjects included in survival analysis, whole-exome sequencing (WES) was performed using a hybrid pull-down panel of baits designed to enrich for more than 700 clinically relevant genes at high coverage (>500×), along with another panel designed to enrich for an additional >20 000 genes at lower depth (>250×). WES was performed on the Illumina NovaSeq 6000 Sequencing System (RRID:SCR_016387). All variants were detected with >99% confidence based on allele frequency and amplicon coverage, with an average sequencing depth >500× and analytical sensitivity of 5%. Identified genetic variants were interpreted by board-certified molecular geneticists and categorised as pathogenic, likely pathogenic or variant of unknown significance, according to the American College of Medical Genetics and Genomics standards. Only pathogenic and likely pathogenic variants were included in comparative analyses.Tumour mutational burden Tumour mutational burden (TMB) was measured by counting all non-synonymous mutations found per tumour that had not been previously described as germline alterations in dbSNP151, Genome Aggregation Database or benign variants identified by Caris geneticists. A threshold of ≥10 mutations per megabase (mt/MB) was used to define TMB-high status, following the KEYNOTE-158 pembrolizumab trial. 18 This study demonstrated that patients with a TMB of ≥10 mt/MB across various tumour types exhibited higher response rates compared with those with a TMB of <10 mt/MB. Caris Life Sciences is a participant in the Friends of Cancer Research TMB Harmonization Project.19Mismatch repair deficiency Multiple test platforms were to determine the MMR status of profiled tumours, including PCR, IHC and NGS. The three platforms generate highly concordant results as previously reported, 20 and in the rare cases of discordant results, the MMR status of the tumour is determined in the order of IHC, PCR and NGS. By IHC, the tumour was considered mismatch repair deficient (dMMR) if complete absence of any of four tested proteins was observed: MLH1 (M1 antibody, Roche Cat No 790-4535, RRID:AB_2336022), MSH2 (G2191129 antibody, Roche Cat No 790-5093, RRID:AB_2936886), MSH6 (44 antibody, Roche Cat No 790-4455, RRID:AB_2336020) or PMS2 (EPR3947 antibody, Roche Cat No 760-4531, RRID:AB_2336010). By PCR analysis, multiplex amplification of five mononucleotide repeat markers (BAT-25, BAT26, NR-21, NR24 and MONO-27) was performed using the MSI Analysis System (Promega, Cat No MD1641), then amplified fragments were analysed on the Applied Biosystems 3500xL Genetic Analyzer System (RRID:SCR_021901). A tumour sample was considered dMMR if two or more mononucleotide repeats were abnormal. By NGS, dMMR was examined using over 7317 target microsatellite loci and compared with the reference genome hg19. The number of microsatellite loci altered by somatic insertion or deletion was counted for each sample. Only insertions or deletions that increased or decreased the number of repeats were considered. dMMR status was assigned for tumours with >45 altered loci, and pMMR status was assigned for tumours with <43 altered loci.Whole-transcriptome sequencing FFPE specimens underwent pathology review to assess tumour content, with a minimum threshold of 10% required to ensure enrichment of tumour-specific RNA. Biotinylated RNA baits (Agilent SureSelect Human All Exon V7 panel) were hybridised to synthesised cDNA targets, and bait–target complexes were amplified in a post-capture PCR reaction. The resultant libraries were pooled and sequenced using the NovaSeq platform.The reference genome used was GRCh37/hg19, and analytical validation of this test demonstrated ≥97% positive per cent agreement, ≥99% negative per cent agreement and ≥99% overall per cent agreement with a validated comparator method. For transcript counting, transcripts per million (TPM) molecules were generated using the Salmon expression pipeline. CXCR4high (Q4) and CXCR4low (Q1) groups were defined based on the top and bottom quartile expression in each cohort. Gene fusions were detected from these transcriptomic data as previously described.21 CRC samples were classified into previously described consensus molecular subtypes (CMS).22 IC fractions were calculated from deconvolution of bulk RNA-seq data using the quanTIseq computational pipeline (RRID:SCR_022993),23 and mRNA expression data were used to perform gene set enrichment analysis (GSEA, RRID:SCR_003199) (for which the significance was determined by a p value <0.05 and a false discovery rate (FDR) q value <0.25), as previously described.24Statistical analysis Molecular alterations among CXCR4 mRNA expression quartiles were analysed using χ2 or Fisher’s exact tests. TME cell fractions were analysed using non-parametric Kruskal-Wallis testing. A value of <0.05 was considered a trending difference; p values were further corrected for multiple comparison using the Benjamini-Hochberg method to avoid type I error, and an adjusted p value (q value) of <0.05 was considered a significant difference.Survival analysis For a subset of the molecular profiling cohort described above, information for real-world survival and pembrolizumab treatment was obtained from the CODEai database using insurance claims data and reported as a surrogate for OS. OS was calculated from date of either first specimen collection or treatment to last of contact with HR and p value calculated using the Cox proportional hazards model and log-rank test, respectively. Significance was determined as p values <0.05. As each Kaplan-Meier analysis represents a distinct clinical or molecular subgroup, p values were not adjusted for multiple testing.Results Heterogeneity of CXCR4 expression in CRC tumour sitesIn this retrospective study, we identified 15 026 patients with confirmed diagnosis of CRC for whom data were available from molecular tumour profiling (WES, whole-transcriptome sequencing, IHC) performed at Caris Life Sciences. Based on RNA sequencing data, samples were analysed to quantify CXCR4 expression quartiles, and tumours in the lowest expression quartile (Q1, n=3857) were compared with those in the highest quartile (Q4, n=3856) for numerous molecular characteristics. Comparisons of Q1 versus Q4 distributional extremes were used for exploratory purposes, not biologically defined cut-offs. Patient demographics are summarised in table 1. CXCR4 mRNA expression for all samples ranged from 0.56 to 489 TPM with a median of 20 TPM. Notably, median CXCR4 mRNA expression in metastatic tissue was significantly higher than in primary tumours (22.7 TPM vs 18.6 TPM, p<0.001) (figure 1A). Further stratification of tumours by location revealed significantly lower CXCR4 expression in liver (n=2988) compared with non-liver (n=3392) metastases (21.5 TPM vs 24.8 TPM, q<0.001), but was significantly higher compared with primary tumours (n=8335) (21.5 TPM vs 18.6 TPM, p<0.001) (online supplemental table 1). Because patients with right-sided CRC have been reported to demonstrate differential responses to therapies such as ICB,25 we compared CXCR4 expression according to sidedness of the primary tumour. Right-sided tumours demonstrated higher median expression compared with left-sided tumours (21.6 TPM vs 19.1 TPM, p<0.001),25 which was also observed in the pMMR subgroup (20.3 TPM vs 18.8 TPM, p<0.001). We also compared CXCR4 expression in CMS groups, which are classified according to distinct molecular properties and clinical characteristics, including immune activation.22 26 27 Comparison of CXCR4 expression in pMMR samples stratified by CMS revealed a significant difference in median CXCR4 RNA expression between all four CMS (figure 1B). CMS 4 was associated with the highest CXCR4 mRNA expression (33.3 TPM), while CMS 3 was associated with the lowest CXCR4 mRNA expression (13.0 TPM).SP110.1136/bmjonc-2025-001003.supp1Supplementary dataTable 1Patient demographics and characteristics in CXCR4low (Q1) and CXCR4high (Q4) quartiles of CXCR4 mRNA expression in patients with CRCCXCR4 Q1CXCR4 Q4Count (n)38573856Average age (range)61.9 (20–89)61.6 (15–89)Sex Male55.4% (2135/3857)53.1% (2048/3856) Female44.6% (1722/3857)46.9% (1808/3856)MMR status dMMR3.6% (136/3823)10.6% (404/3821) pMMR96.4% (3687/3823)89.4% (3417/3821)CRC, colorectal cancer; CXCR4, CXC-chemokine receptor 4; dMMR, mismatch repair deficient; MMR, mismatch repair; pMMR, mismatch repair proficient.Figure 1CXCR4 mRNA expression in CRC tumours. Median CXCR4 mRNA expression (TPM). Comparison between primary, metastatic tissue or unclear (primary and/or metastatic) tissue (A, **p<0.05; ****p<0.001). Comparison in pMMR tumours across CMS (B, all comparisons marked with red bar: q<0.0001). CMS, consensus molecular subtypes; CRC, colorectal cancer; CXCR4, CXC-chemokine receptor 4; pMMR, mismatch repair proficient; TPM, transcripts per million.Molecular landscape variations associated with CXCR4 expressionGenomic alterations were compared between the CXCR4low and CXCR4high cohorts (online supplemental table 2). TP53 and APC mutations were more frequently observed in the CXCR4low compared with the CXCR4high cohort (78.6% vs 65.6%, p<0.001; 82% vs 62.7%, p<0.001, respectively), whereas BRAF, NF1 and SMAD2 mutations were more frequent in CXCR4high compared with CXCR4low (12.9% vs 6.9%, p<0.001; 6.0% vs 2.5%, p<0.001; 3.9% vs 2.3%, p<0.001, respectively). Genetic analyses focused on the pMMR subgroup showed similar results. In pMMR CRC, TP53 and APC mutations were also more frequently observed in CXCR4low compared with CXCR4high tumours (80.4% vs 70.3%, p<0.001; 83.8% vs 66.9%, p<0.001, respectively). SMAD2, BRAF and NF1 were more frequently mutated in CXCR4high compared with the CXCR4low cohort (3.9% vs 2.2%, p<0.001; 8.3% vs 5.6%, p<0.001; 3.3% vs 1.7%, p<0.001, respectively) (online supplemental figure 1). GSEA evaluation of the pMMR cohort showed significant enrichment of gene expression in CXCR4high compared with CXCR4low tumours for transforming growth factor-beta, interleukin-2-signal transducer and activator of transcription 5 (STAT5) and interferon (IFN)-γ signalling pathways (all p<0.05 and FDR<0.25) (online supplemental table 3). In all CRC tumours, RET and MET fusions were detected at higher frequency in the CXCR4high compared with CXCR4low cohort (0.3% vs 0.1%, p=0.004; 0.2% vs 0%, p<0.015, respectively). Within the pMMR subgroup, these trends were not statistically significant, but RSPO2 and RSPO3 fusion events were more frequent in the CXCR4high compared with CXCR4low cohorts (0.2% vs 0%, p<0.039; 1.7% vs 1.1%, p<0.035, respectively) (online supplemental figure 2).CXCR4 expression is associated with immunotherapy biomarkers and infiltration of several ICsWithin the entire cohort, TMB-high, dMMR and PD-L1 positive status were more frequently observed in the CXCR4high compared with the CXCR4low cohort (10.6% vs 3.6%, p<0.0001; 6.4% vs 2.2%, p<0.0001; 13.8% vs 6.4%, p<0.0001, respectively) (figure 2A). In the pMMR cohort, only PD-L1 expression was associated with high CXCR4 expression (4.3% vs 1.7%, p<0.0001) (figure 2B). From RNA-seq data, quanTIseq was used to quantify IC infiltration patterns in the TME. In the CXCR4high cohort, we observed an association with higher infiltration of B cells, M1 macrophages, M2 macrophages, natural killer (NK) cells, CD8+ T cells and T regulatory cells (Tregs). Conversely, the CXCR4low cohort demonstrated an association with increased presence of neutrophils, myeloid dendritic cells and CD4+ T cells (all p<0.001) (figure 2C). Similar associations were observed in the pMMR cohort.Figure 2Comparison of immune-related characteristics. Prevalence of immunotherapy biomarkers in CXCR4low (Q1) and CXCR4high (Q4) expression groups in all CRC tumours (A) and pMMR CRC tumours (B). Heatmap showing immune cell comparisons between CXCR4low (Q1) and CXCR4high (Q4) expression groups by quanTIseq analysis (C). Median % infiltration shown (non-zero % shown where medians are zero). **P<0.001. CRC, colorectal cancer; CXCR4, CXC-chemokine receptor 4; dMMR, mismatch repair deficient; IHC, immunohistochemistry; MSI-H, microsatellite instability-high; NK, natural killer; pMMR, mismatch repair proficient; TMB-H, tumour mutational burden-high.High CXCR4 RNA expression associates with decreased OS in CRCFrom the larger cohort of patients with CRC with available molecular profiling data (n=12 128), we compared the survivorship between CXCR4 expression cohorts from a smaller subset of patients with CRC with available insurance claims data. In the entire study population (n=12 128), the CXCR4high cohort exhibited decreased OS (28.7 months, 95% CI 27.6 to 29.9) compared with the CXCR4low cohort (30.5 months, 95% CI 29.3 to 31.7) (HR=0.95, 95% CI 0.90 to 0.99, p<0.03) (figure 3A). A multivariate analysis incorporating established CRC prognostic factors (age, sex, MMR status, TMB, PD-L1, tumour site and sidedness) further demonstrated that low CXCR4 expression is associated with significantly better survival compared with high expression (HR=0.94, 95% CI 0.89 to 0.99, p=0.013) (figure 3A; online supplemental table 4). Similarly, in the pMMR cohort (n=11 018), the CXCR4high cohort demonstrated significantly lower OS (27.8 months, 95% CI 26.7 to 29.1) compared with CXCR4low (30.0 months, 95% CI 28.9 to 31.4) (HR=0.92, 95% CI 0.87 to 0.97, p=0.001) (figure 3B). In the dMMR subset (n=860), no significant differences in OS were observed (figure 3C).Figure 3Overall survival and progression-free survival for all patients. KM analysis of CRC comparing CXCR4low (Q1) and CXCR4high (Q4) expression groups in all patients with CRC (6063 CXCR4low vs 6065 CXCR4high) (A), patients with pMMR CRC (5508 CXCR4low vs 5510 CXCR4high) (B) and patients with dMMR CRC (430 CXCR4low vs 430 CXCR4high) (C). Event-free proportion was calculated from date of collection to last day of contact. P values were derived from log-rank tests to compare survival distributions between groups. These p values were not adjusted for multiple testing, as each KM curve represents a distinct clinical or molecular subgroup. CRC, colorectal cancer; CXCR4, CXC-chemokine receptor 4; dMMR, mismatch repair deficient; KM, Kaplan-Meier; pMMR, mismatch repair proficient.We next compared OS between CXCR4 expression quartiles in primary and metastatic tumours. In primary tumours, the CXCR4high cohort exhibited decreased OS compared with CXCR4low (31.8 months vs 34.8 months, HR=0.92, 95% CI 0.86 to 0.99, p=0.02) (online supplemental figure 3A). This observation held within the pMMR subgroup (30.5 months vs 34.3 months, HR=0.89, 95% CI 0.83 to 0.96, p=0.001) (online supplemental figure 3B). In contrast, in metastatic tissue, the CXCR4low cohort exhibited decreased OS compared with CXCR4high (23.4 months vs 25.1 months, HR=1.12, 95% CI 1.04 to 1.20, p=0.003) (online supplemental figure 3C). This association also held within the pMMR subgroup (23.4 months vs 24.9 months, HR=1.12, 95% CI 1.04 to 1.21, p=0.004) (online supplemental figure 3D). Stratification by metastatic site revealed no significant association between OS and high or low CXCR4 expression in liver, lungs or other metastatic sites (online supplemental figure 4).High CXCR4 RNA expression associates with improved OS in ICB-treated CRCWithin the entire cohort, we evaluated OS across CXCR4 expression quartiles in a subgroup of patients with CRC treated with pembrolizumab (n=500). The CXCR4high cohort showed significantly improved survival in pembrolizumab-treated patients (26.8 months vs 19.6 months, HR=1.39, 95% CI 1.07 to 1.80, p=0.012) (figure 4). Time on treatment was also higher in the CXCR4high cohort compared with the CXCR4low cohort (5.1 months vs 3.5 months, HR=1.21, 95% CI 1.002 to 1.465, p=0.051) (online supplemental figure 5).Figure 4Overall survival and progression-free survival in immunotherapy-treated patients. KM analysis of pembrolizumab-treated patients with CRC in CXCR4 expression groups, comparing 198 CXCR4low (Q1) to 302 CXCR4high (Q4) patients. Event-free proportion was calculated from date of first treatment to last day of contact. P values were derived from log-rank tests to compare survival distributions between groups. These p values were not adjusted for multiple testing, as each KM curve represents a distinct clinical or molecular subgroup. CRC, colorectal cancer; CXCR4, CXC-chemokine receptor 4; KM, Kaplan-Meier.Discussion CXCR4 is associated with poor differentiation, metastasis and prognosis in CRC. Blockade of the CXCR4-CXCL12 axis has been shown to reduce the pleiotropic role in metastatic progression and tumour-induced immunosuppression. In preclinical murine models, a small molecule (MSX-122) acting as a partial CXCR4 antagonist has exhibited long-term blockade of metastasis and inflammatory changes. 28 AMD3100 (plerixafor) has been characterised as a CXCR4 antagonist with activity in leukaemia and multiple solid tumours.29 Disruption of the CXCR4-CXCL12 axis is being evaluated in preclinical and clinical studies (online supplemental table 5), although these trials have not incorporated CXCR4 gene expression as a criterion for patient selection for therapeutic efficacy and its role within the TME and response to ICB remains unclear. To address this gap, we leveraged a large real-world dataset to conduct a comprehensive molecular, genetic and immunological analysis of CRC based on CXCR4 mRNA expression levels. In the present study, CXCR4 mRNA expression was positively associated with TMB-high, dMMR, positive PD-L1 status and higher IC infiltration. Based on the evaluation of OS from the last healthcare contact in insurance claims data, high CXCR4 mRNA expression was associated with poor prognosis in the primary tumour but was associated with improved survival in all ICB-treated patients with CRC, irrespective of MMR status. To date, there is a general belief in the field that high CXCR4 expression in patients with CRC is unfavourable, as it correlates with advanced tumour stage, increased risk of recurrence and distant metastasis.13 In congruence with previous studies,30 31 our study confirms that high CXCR4 expression is associated with worse median OS in the pMMR cohort. We also observed that CXCR4 expression differs across anatomical sites, with higher levels in metastatic lesions compared with primary tumours, consistent with prior reports.12 30 32A recent study further implicated CXCL12-CXCR4 signalling as a driver of immune exclusion in hepatic metastases.33 Notably, the lower CXCR4 mRNA levels seen specifically in liver metastases may reflect the unique biology of the hepatic microenvironment rather than true loss of CXCR4 signalling. The liver is enriched in CXCL12-producing sinusoidal endothelial and stellate cells, and sustained ligand abundance can promote CXCR4 internalisation and downregulation despite preserved pathway activity.34 35 This provides a hypothetical mechanistic explanation for why CXCR4 expression appears reduced in liver lesions, even though CXCR4 remains associated with inflamed/IFN-rich phenotypes in non-liver metastases. The poorer OS observed in CXCR4low liver metastases is consistent with the hypothesis that these tumours reside within an immune-excluded, stromally remodelled hepatic microenvironment. Our transcriptomic analysis captures CXCR4 expression across primary and metastatic contexts, whereas microenvironment-driven receptor downregulation in liver does not contradict the broader genomic enrichment patterns seen in large CRC datasets.Although immunotherapy responsiveness is generally limited in patients with liver metastases, including those with pMMR CRC,36–41 several studies have shown that combinatorial approaches, including kinase inhibitors, can enhance ICB efficacy by modifying the otherwise ‘cold’ hepatic microenvironment.42 43 Furthermore, recent work in hepatocellular carcinoma demonstrates that CXCR4 inhibition can improve responses to ICB.44 Together, these findings underscore the clinical relevance of CXCL12-CXCR4 signalling in CRC and support further investigation into targeting this pathway to overcome therapeutic resistance, particularly in patients with liver metastases.Investigation of genomic alterations revealed several significant differences between CXCR4high and CXCR4low expression cohorts. Our study shows a link between p53 mutations and CXCR4 expression, which has also been shown in other cancers.45 The significance of higher TP53 mutations associated with low CXCR4 expression in our study is unknown. However, in line with our findings, prior studies have also reported a similar inhibitory pattern of p53 on CXCR4 in pancreatic cancer.46 Moreover, a p53-mediated pathway suppresses the CXCR4-CXCL12 axis in distant metastasis via inhibition of CXCL12 production by stromal fibroblasts.47 Wild-type TP53 has also been shown to repress CXCR4 expression in breast cancer. It was also recently suggested that TP53 mutation in cancer stem cells leads to CXCR4 upregulation.48 49 Further studies are warranted to elucidate the underlying mechanisms driving these associations. BRAF mutations were also more frequent in the CXCR4high cohort. BRAF inhibition in patients with CRC has not demonstrated clinical benefit in the monotherapy setting due to upregulation via the epidermal growth factor receptor (EGFR) pathway, which suggests the need for combinatorial strategies for patients with BRAF-mutated CRC.50 Gala et al observed a comparable positive association between CXCR4 expression and pMMR CRC harbouring BRAFV600E mutational status,51 while other studies have described CXCR4 activation of EGFR and extracellular signal-regulated kinase (ERK) pathways involved in resistance to BRAF inhibitors.52–54 Thus, the availability of CXCR4 inhibitors may present new therapeutic opportunities in specific CRC genotypes.Although some studies have explored CXCR4 as a predictor of chemotherapy response, its role in forecasting immunotherapy outcomes in CRC remains unknown. The basis of immunotherapy is to overcome the mechanisms within the TME that block the immunosuppressive response.Our analysis showed that high CXCR4 mRNA expression is associated with increased infiltration of multiple IC populations—including B cells, M1 and M2 macrophages, NK cells, CD8+ T cells and Tregs—independent of MMR status. Because CXCR4 is a central regulator of leucocyte trafficking, high expression may reflect enhanced IC recruitment into the TME rather than a uniformly inflamed or immunosuppressed state. This mixed infiltration pattern suggests that CXCR4high tumours may exist in a biologically active TME with concurrent effector and suppressive signals. In this context, we observed that high CXCR4 expression is associated with poorer OS in the broader CRC cohort, yet paradoxically corresponds to improved responsiveness to ICB, regardless of MMR status. Similar findings have been reported in non-small cell lung cancer, where CXCR4high tumours demonstrate worse prognosis but higher immune infiltration and better responses to ICB.16 Together, these observations indicate that CXCR4 expression may mark a TME with substantial immune engagement that remains functionally constrained, but which can be effectively reactivated by immunotherapy. This highlights the need for further mechanistic studies to define how CXCR4-driven immune architecture influences ICB sensitivity in CRC.High CXCR4 mRNA expression was also positively associated with TMB-high, dMMR and positive PD-L1 status, which are positively associated with ICB response in many cancer types, but have shown limited predictive value in CRC. Similar observations were reported in a pancreatic ductal adenocarcinoma study with high CXCR4 mRNA expression associated with enhanced infiltration of CD4+ T cells, T regs, CD8+ T cells and macrophages.46 CXCR4high tumours often align with an ‘inflamed/IFN-rich’ transcriptional state also enriched for high TMB, dMMR and PD-L1 expression. This is biologically consistent with the idea that TMB-high and dMMR/microsatellite instability (MSI)-high tumours generate abundant neoantigens, exhibit strong immune infiltration and upregulate checkpoints, while CXCL12-CXCR4 signalling intersects with Janus kinase/STAT3 and nuclear factor-kappa B pathways that further induce PD-L1. However, within the dMMR/MSI-high subgroup, OS is characterised by their profound sensitivity to PD-1 blockade,55 56 which produces durable treatment effects that overshadow more subtle TME distinctions. Although CXCR4 can support recruitment and retention of suppressive myeloid and Treg populations, and blockade of this axis enhances anti-PD-1 activity in preclinical models,17 57 these disadvantages become secondary once tumours are strongly ICI responsive. As a result, CXCR4 expression provides limited additional OS stratification within the dMMR/MSI-high stratum.These findings seem to corroborate previous reports showing CXCR4 overexpression as a potential biomarker for response to immunotherapy in lung cancer, as indicated by other markers including low tumour immune dysfunction and exclusion score, low T cell exclusion score and high T cell dysfunction score.16 Biasci et al also showed that inhibition of the CXCR4-CXCL12 axis via AMD3100 led to increased IC infiltration in both pMMR CRC and pancreatic cancer.17 Of note, the phase II COMBAT/KEYNOTE-202 trial assessed the efficacy of the CXCR4 antagonist BL-8040 (motixafortide) in combination with the anti-PD-1 antibody pembrolizumab±chemotherapy in patients with pretreated metastatic pancreatic ductal adenocarcinoma.58 This combination therapy led to changes in TME composition as depicted by an increase in intratumour CD8+ T cells and decrease in immunosuppressive cells. Considering the current evidence combined with the recent publication of neoadjuvant immunotherapy outcomes in patients with early-stage CRC, these findings justify exploratory analyses of completed clinical trials to enhance our understanding of the role of CXCR4 expression in CRC.This study has several limitations. Survival estimates were derived from retrospective claims-based data and reflect last healthcare contact, which may not accurately represent true OS. As with all retrospective analyses, the dataset is subject to selection and information biases and lacked key variables—including demographics, staging and treatment details—that could influence molecular patterns. These constraints limit our ability to refine CRC subgroups and fully interpret associations between CXCR4 expression, advanced disease and ICB response; thus, conclusions regarding prognostic or predictive value should be viewed as exploratory. Additionally, bulk sequencing restricts resolution of immune infiltration, as deconvolution tools such as quanTIseq cannot distinguish intratumoral from peritumoral localisation. Finally, continuous or dose–response modelling of CXCR4 expression was not performed, and employing such analyses may add resolution in future work. Nonetheless, the observed association between CXCR4 expression and IC infiltration supports further investigation of CXCR4 as a potential biomarker in defined clinical contexts.To the best of our knowledge, this study represents the largest and most comprehensive genetic and transcriptomic analysis of CXCR4 mRNA expression in CRC. These findings underscore potential interplay of molecular pathways with CXCR4 and the TME and its contribution to the evolution of metastasis in CRC, highlighting possible actionable targets. In addition to describing a potential prognostic association in CRC, our findings suggest further exploration of CXCR4 expression as a potential predictive biomarker for ICB response in CRC. Moreover, these findings add to the growing body of evidence for targeting chemokines and chemokine receptors, or combinations with current therapies as potential novel treatment strategies for CRC. In this regard, an ongoing clinical trial is assessing the safety of combining the CXCR4 peptide antagonist LY2510924 with the anti-PD-L1 antibody (durvalumab) in treating advanced solid tumours.59 Despite the limitation of the study due to its retrospective nature and small sample size of ICB-treated patients, our results provide a comprehensive molecular and immunological characterisation in CRC that could potentially establish a framework for selecting appropriate participants for future clinical trials.