Clinical and molecular characterization of AXL in colorectal cancer, CALGB (Alliance)/SWOG 80405 and real-world data Authors: Karam Ashouri, Joshua Millstein, Yan Yang, Joanne Xiu, Shivani Soni, Pooja Mittal, Sandra Algaze, Lesly Torres-Gonzalez, Priya Jayachandran, Wu Zhang, Shannon Mumenthaler, Anthony F Shields, Richard Goldberg, Emil Lou, Benjamin A Weinberg, Thomas G Graeber, John L Marshall, Alan P Venook, Alexander Hoffmann, Stacey D Finley, Aaron S Meyer, Francesca Battaglin, Heinz-Josef Lenz Published: 2025 ABSTRACT AbstractBackgroundAXL dysregulation is associated with both intrinsic resistance in tumor cells and reprogramming of the immune microenvironment. However, it is still unclear which patients would benefit from AXL-targeting therapies. We conducted a clinical and molecular characterization of AXL in colorectal cancer (CRC).MethodsThis study integrated real-world molecular profiling (Caris cohort; n=24,257) and randomized clinical trial data (phase III Cancer and Leukemia Group B/Southwest Oncology Group (CALGB/SWOG) 80405; n=433) to assess AXL messenger RNA expression in patients with CRC. Tumor samples underwent RNA sequencing and analysis of the immune microenvironment. AXL expression was categorized into tertiles. We assessed associations between molecular features, immune biomarkers, and clinical outcomes, including overall survival (OS) and progression-free survival (PFS), using Kaplan-Meier and Cox proportional hazards models while adjusting for relevant covariates.ResultsElevated AXL expression correlated with increased programmed death-ligand 1 immunohistochemistry positivity (6.2% vs 2.5%, q<0.0001), immune checkpoint-related gene expression, and infiltration of immunosuppressive cell populations (T-regulatory cells, M2 macrophages, monocytes, and B cells). Pathway analyses demonstrated links between high AXL expression and epithelial–mesenchymal transition, inflammatory signaling, interferon-gamma response, and tumor necrosis factor alpha signaling. In the Caris cohort, high AXL predicted worse OS in patients treated with fluorouracil, leucovorin, and oxaliplatin/fluorouracil, leucovorin, and irinotecan (38.0 vs 34.7 months, p=0.027; HR 1.07), bevacizumab (36.8 vs 32.6 months, p=0.013; HR 1.21), and anti-epidermal growth factor receptor therapy (28.4 vs 22.2 months, p=0.005; HR 1.21), but profoundly improved OS in KRAS mutant patients treated with immunotherapy (11.6 vs 23.4 months, p=0.012; HR 0.65). CALGB/SWOG 80405 findings confirmed shorter PFS (9.2 vs 12.9 months, p=0.001; HR 1.56) and OS (24.2 vs 34.7 months, p<0.001; HR 1.68) with high AXL expression across treatment arms.ConclusionsElevated AXL expression in CRC correlated with an immunosuppressive microenvironment and worse outcomes across standard treatments. However, it identifies a distinct subgroup of KRAS-mutant patients who significantly benefit from immunotherapy, supporting AXL as a context-specific biomarker and therapeutic target. WHAT IS ALREADY KNOWN ON THIS TOPIC AXL is a receptor tyrosine kinase implicated in tumor progression, therapeutic resistance, and immune evasion in several malignancies, including colorectal cancer (CRC). Preclinical studies suggest that AXL contributes to epithelial–mesenchymal transition, immune modulation, and revascularization following antiangiogenic therapy. However, clinical data validating AXL’s prognostic or predictive role in metastatic CRC (mCRC), particularly in the context of KRAS mutational status and diverse treatment modalities, have been limited. AXL is a receptor tyrosine kinase implicated in tumor progression, therapeutic resistance, and immune evasion in several malignancies, including colorectal cancer (CRC). Preclinical studies suggest that AXL contributes to epithelial–mesenchymal transition, immune modulation, and revascularization following antiangiogenic therapy. However, clinical data validating AXL’s prognostic or predictive role in metastatic CRC (mCRC), particularly in the context of KRAS mutational status and diverse treatment modalities, have been limited. WHAT THIS STUDY ADDS This study is the largest to date evaluating AXL expression in CRC and reveals that high AXL expression is associated with poor prognosis, increased immune suppression, and resistance to standard therapies. Interestingly, high AXL expression is correlated with improved overall survival in patients with KRAS-mutant CRC treated with immunotherapy, suggesting a context-specific immune vulnerability. The findings support AXL as a robust prognostic biomarker and reveal its dual role in modulating both immunosuppressive and immune-activating pathways depending on molecular context. This study is the largest to date evaluating AXL expression in CRC and reveals that high AXL expression is associated with poor prognosis, increased immune suppression, and resistance to standard therapies. Interestingly, high AXL expression is correlated with improved overall survival in patients with KRAS-mutant CRC treated with immunotherapy, suggesting a context-specific immune vulnerability. The findings support AXL as a robust prognostic biomarker and reveal its dual role in modulating both immunosuppressive and immune-activating pathways depending on molecular context. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings provide a rationale for incorporating AXL expression and KRAS status into therapeutic decision-making for mCRC and support the prioritization of clinical trials targeting AXL, either alone or in combination with immunotherapy. Future research should explore AXL-targeted strategies in KRAS-mutant tumors and refine patient selection for immunotherapy based on AXL-driven immune profiles. These findings provide a rationale for incorporating AXL expression and KRAS status into therapeutic decision-making for mCRC and support the prioritization of clinical trials targeting AXL, either alone or in combination with immunotherapy. Future research should explore AXL-targeted strategies in KRAS-mutant tumors and refine patient selection for immunotherapy based on AXL-driven immune profiles. Background Colorectal cancer (CRC) is a leading cause of cancer-related deaths.1 Despite therapeutic advances, CRC (metastatic CRC (mCRC)) still carries a poor prognosis, with a 5-year survival rate of 13–18%.2 Targeted therapies against vascular endothelial growth factor (VEGF) and epidermal growth factor receptor (EGFR) have provided modest survival benefits in patients with unresectable or metastatic disease. Ongoing research aims to uncover new molecular markers and signaling pathways that could serve as prognostic indicators or therapeutic targets. AXL a member of the tumor-associated macrophage family of receptor tyrosine kinases (TYRO3, AXL, and MER), which regulates proliferation, survival, migration, and differentiation.3 4 While the AXL receptor tyrosine kinase is directly activated by its ligand growth arrest-specific protein 6 (Gas6), this activation typically requires a concurrent interaction with phosphatidylserine (PtdSer) through Gas6’s γ-carboxyglutamic acid domain.5 6 This Gas6-PtdSer-AXL interaction enables receptor clustering and autophosphorylation of the AXL intracellular domain.5 7 This activation drives proliferation through the PI3K/Akt and NF-κB pathways, increasing anti-apoptotic proteins and reducing proapoptotic proteins.810 AXL promotes epithelial–mesenchymal transition (EMT), suppressing epithelial markers while enhancing mesenchymal markers.11 Overexpression of AXL has also been found in several cancers, including hepatocellular carcinoma,12 clear cell renal cell carcinoma,13 and cholangiocarcinoma,14 conferring a poor prognosis in these malignancies. AXL overexpression contributes to EGFR inhibitor-resistant lung cancer,15 cisplatin-resistant ovarian cancer,16 and radiation-resistant head and neck squamous cell carcinoma.17 Similarly, AXL shows higher expression in CRC than in normal tissue by immunohistochemistry (IHC) and correlates with a worse prognosis in early stage CRC.18 In vitro experiments have uncovered the contributions of AXL to the revascularization of CRC tumors after the withdrawal of antiangiogenic drugs, indicating AXL suppression may improve outcomes in CRC.19 AXL also regulates the tumor immune microenvironment and may affect immunotherapy response.19 The novel AXL inhibitor, TP-0903, has completed a phase 1 trial in solid tumors—including CRC—demonstrating a favorable safety profile, though efficacy data are still pending. (NCT02729298). Although preclinical studies have established AXL’s role in CRC, clinical validation has been limited to two retrospective studies on metastatic and early-stage disease.18 20 Our dual-cohort analysis integrates real-world molecular profiling with phase III trial data to address three key gaps: (1) prognostic validation across treatment settings, (2) immune microenvironment characterization, and (3) KRAS-subgroup therapeutic implications. Our study comprehensively profiled AXL expression in CRC using both real-world and clinical trial cohorts, revealing that high AXL levels are associated with distinct molecular features, an immunosuppressive tumor microenvironment (TME), and the activation of EMT and inflammatory pathways. We found that elevated AXL expression independently predicted shorter progression-free survival (PFS) and overall survival (OS) in mCRC, particularly in patients with wild-type KRAS. These findings establish AXL as a clinically relevant prognostic biomarker and a potential therapeutic target in mCRC. Methods Patient samples The real-world cohort was derived from the Caris Life Sciences database (Phoenix, Arizona, USA), which contains 295,431 tumor samples. After restricting to unique patients (n=287,415), 34,065 patients with CRC were identified. The final Caris analytic cohort comprised 24,257 patients with unique CRC with available next-generation sequencing data and formalin-fixed paraffin-embedded (FFPE) tumor samples. The clinical trial dataset was derived from the phase III trial CALGB/SWOG 80405: Cancer and Leukemia Group B (CALGB of the Alliance for Clinical Trials in Oncology)/Southwest Oncology Group (SWOG) 80405, in which 433 patients were randomized to receive bevacizumab (n=226) or cetuximab (n=207) with chemotherapy backbones of fluorouracil, leucovorin, and oxaliplatin (FOLFOX) or fluorouracil, leucovorin, and irinotecan (FOLFIRI).21 The trial enrollment was between 2005 and 2012. The median age of the participants was 60 years old and most (62.6%) were male. Patients maintained their study treatment until disease progression, unacceptable toxic effects, a gap of more than 28 days between treatments, or a decision to discontinue the treatment. The treating investigator conducted disease assessment and was not blinded to the treatment. The initial study protocol enrolled patients regardless of KRAS status; however, an interim amendment revised the inclusion/exclusion criteria to restrict enrollment to patients with known wild-type KRAS tumors due to emerging evidence demonstrating a lack of benefit of cetuximab in patients with KRAS mutation. Details of the patient population and power analysis for the CALGB 80405 trial have been described previously.21 The full clinical protocol used in this study is publicly available at ClinicalTrials.gov (NCT00265850). Tumor profiling In the real-world dataset, FFPE tumor specimens underwent a pathology review to determine the tumor content. Microdissection was performed in the regions with the highest cancer cell concentrations. For whole transcriptome sequencing, a minimum tumor content of 10% was required. RNA was extracted using the Qiagen RNA FFPE Tissue Extraction Kit and the quality was assessed using the Agilent TapeStation. Libraries were prepared and sequencing was conducted on the Illumina NovaSeq 6500 platform, targeting an average of 60 million reads. Data processing included demultiplexing, alignment to the hg19 human reference genome using the STAR aligner, and transcript quantification with salmon. Genomic DNA sequencing was performed using the Illumina platform. For NextSeq sequenced samples, a custom-designed SureSelect XT assay targeted 592 whole genes, whereas for NovaSeq sequenced samples, over 700 clinically relevant genes were enriched at high coverage, with an additional >20,000 genes sequenced at lower depths. Variants were interpreted by board-certified molecular geneticists in accordance with the American College of Medical Genetics and Genomics standards. Immune microenvironment characteristics were analyzed using the QuanTISeq RNA deconvolution program and interferon-gamma (IFN-γ) signatures associated with immune checkpoint inhibitor outcomes were calculated. In the clinical trial dataset, RNA was extracted from FFPE tumor samples using the High Pure FFPET RNA Isolation Kit (Roche). Libraries were prepared using 250 ng of input RNA, following the Illumina TruSeq RNA Access protocol. Sequencing was performed on an Illumina HiSeq 2500 platform, targeting 50 million paired-end reads with a read length of 2×100 bp. Data quality control and normalization were performed using MAPRSEQ V.3.1 and Conditional Quantile Normalization.22 23 Statistical analysis Both datasets evaluated AXL (Ensembl ID: ENSG00000167601) expression using tertile groups (low T1, mid T2, high T3 expression) for survival analyses. Sensitivity analyses were performed using alternative AXL expression categorizations (quartiles and median split) in addition to the primary tertile-based analysis, with similar results across categorization approaches. Missing data for molecular and clinical variables were handled through complete case analysis, with patients included in analyses only when AXL expression data and relevant covariates were available. Statistical analyses for the real-world dataset incorporated clinical data obtained from insurance claims, including prescribed medications, procedures, and diagnoses. OS was defined as the time from treatment initiation to death or last known follow-up record. Patients for whom the date of death was not recorded were censored if no insurance claims were observed for >100 days. Kaplan-Meier survival estimates and log-rank tests were used to compare survival outcomes between high tertile (T3) and low tertile (T1) subgroups of AXL expression, while Cox proportional hazards models were used to calculate the HRs. P values were adjusted for multiple comparisons using the Benjamini-Hochberg method (q<0.05). In the clinical trial dataset, OS and PFS were defined as the time from randomization to death or disease progression, respectively. Kaplan-Meier curves were used to assess survival differences between AXL expression tertiles, while Cox proportional hazards models were used to evaluate AXL as a continuous variable and to test for interactions between AXL expression, treatment, and patient characteristics. Covariates included age, sex, ethnicity, ECOG performance status, tumor location, number of metastases, KRAS/BRAF mutation status, microsatellite instability (MSI) status, randomized treatment, and chemotherapy regimen. Results AXL expression levels in KRAS subgroups and tumor sites The median age of the Caris cohort was 62 years (IQR 44–80) with most of the patients being male (54.9%). The median AXL expression in all CRCs was 5.72 transcripts per million (TPM) (IQR 3.42–9.18), with KRAS wild-type having higher AXL expression relative to KRAS mutant tumors (5.83 vs 5.42, q<0.0001) (online supplemental figure S1). AXL expression was higher in tissue samples obtained from primary tumors than in those obtained from metastatic sites (6.11 vs 4.89, q<0.0001). When evaluating the most common metastatic sites, AXL expression was highest in the peritoneal tissue (median 8.89, IQR 5.03–14.78), followed by the lymph node (median 5.95, IQR 2.98–9.87), lung (median 5.11, IQR 3.06–8.19), liver (median 4.62, IQR 2.77–7.46), bone (median 4.32, IQR 2.70–8.28), and brain (median 2.64, IQR 1.67–3.84) (online supplemental figure S1). The clinical and demographic characteristics were highlighted for the CALGB/SWOG 80405 cohort using AXL tertiles (online supplemental table S2). When evaluating CRC by AXL tertile expression, the cohort of patients with tumors that had high AXL expression demonstrated increased programmed death-ligand 1 (PD-L1) positivity (T1: 2.5% vs T3: 6.2%, q<0.0001) and decreased tumor mutation burden–high (TMB-H) (14.3% vs 12.1% q=0.0001), but no difference in MSI-high/deficient mismatch repair (MSI-H/dMMR) status (7.9% vs 7.2% q=0.1116) (table 1). In the KRAS mutant and wild-type subgroups, increased AXL was correlated with increased PD-L1 positivity and decreased TMB-H. Samples with low AXL expression in the KRAS mutant subgroup revealed increased MSI-H/dMMR status (T1: 11.1% vs T3: 10.0%, q=0.0177) (online supplemental table S1). Higher AXL TPM correlated with tumor inflammation score (T1: −134 vs T3: 53, q<0.0001) and IFN-γ score (-0.43 vs −0.36, q<0.0001) and persisted in both KRAS mutant and wild-type subgroups (table 1) (online supplemental table S1). PD-L1 IHC was not available in CALGB/SWOG 80405; however, analysis using NGS expression of the CD274 (PD-L1) gene demonstrated a statistically significant, although modest, positive correlation between AXL expression and CD274 (PD-L1) expression (R=0.19, p=7.3e-5) (online supplemental figure S2). Table 1Immune-related markers by AXL tertilesT1 (N=8005)T3 (N=8005)q-value*MSI-H/dMMR534 (7.9%)466 (7.2%)0.1116TMB-H (≥10/Mb)1104 (14.3%)921 (12.1%)0.0001PD-L1 ≥5%194 (2.5%)477 (6.2%)<0.0001TIS score−13453<0.0001IFN score−0.4293−0.3582<0.0001*Benjamini-Hochberg method.IFN, interferon; Mb, megabase; MSI-H/dMMR, microsatellite instability-high/mismatch repair deficient; PD-L1, programmed death-ligand 1; TIS, Tumor Inflammation Score; TMB-H, tumor mutational burden-high. Immune-related markers by AXL tertiles Benjamini-Hochberg method. IFN, interferon; Mb, megabase; MSI-H/dMMR, microsatellite instability-high/mismatch repair deficient; PD-L1, programmed death-ligand 1; TIS, Tumor Inflammation Score; TMB-H, tumor mutational burden-high. Association with tumor molecular characteristics, immune-related biomarkers, and pathway enrichment analysis AXL expression was positively associated with SMAD4 (T1: 13.8% vs T3: 15.4%; q=0.0086), BRAF (8.9% vs 13.5%; q<0.0001), and RNF43 (6.5% vs 8.8%; q<0.0001) mutations. Additionally, AXL TPM was negatively associated with APC (T1: 82.9% vs T3: 68.2%, q<0.0001), KRAS (49.4% vs 44.0%, q<0.0001), PIK3CA (19.1% vs 15.8%, q<0.0001), FBXW7 (11.2% vs 10.0%, q=0.0318), SOX9 (10.5% vs 8.9%, q=0.006), AMER1 (7.3% vs 5.5%, q<0.0001), MGA (6.7% vs 4.0%, q<0.0001), PTEN (5.7% vs 4.6%, q=0.0047), BCL9 (4.9% vs 4.1%, q=0.0305), NRAS (4.5% vs 2.9%, q<0.0001), and PRKDC (3.2% vs 2.4%, q=0.0081) (figure 1). Molecular alterations stratified by AXL tertiles. The pathways with the top five normalized enrichment scores (NES) during gene set enrichment analysis were EMT (NES: 3.595, q=0.002), inflammatory response (NES: 2.906, q=0.002), allograft rejection (NES: 2.819, q=0.002), IFN-γ response (NES: 2.793, q=0.002), and tumor necrosis factor alpha (TNF-α) signaling (NES: 2.790, q=0.002) (figure 2A, online supplemental table S3). The tumor expression of immune regulatory genes (CD80, IDO1, CD86, LAG3, HAVCR2, IFNG, PDCD1LG2, PDCD1, CTLA4) was generally increased in the AXL high CRC cohorts (all q<0.0001), except for CD274 which was increased in the low AXL CRC cohorts (q=0.0301) (figure 2B). Increased AXL expression was correlated with B cells, M1/M2 macrophages, monocytes, T-regulatory cells (T-regs), and natural killer (NK) cells in the TME, while increased neutrophils, dendritic cells, and CD4+ T cells were associated with low AXL expressing CRC (all q<0.05) (figure 3). (A) GSEA results and (B) immune checkpoint‐related marker expression comparing AXL T3 to AXL T1 (q<0.05). GSEA, gene set enrichment analysis. Tumor microenvironment cell infiltration according to AXL expression. NK, natural killer; Treg, T regulatory cell. *indicates p≤0.05, while **** indicates p≤0.0001 Impact of AXL gene expression on clinical outcomes Caris cohort. CRC samples with upper tertile AXL expression demonstrated worse OS in the FOLFOX/FOLFIRI-treated CRC group (T1: 38.0 vs T3: 34.7 months, p=0.027; HR 1.07, 95% CI 1.01 to 1.14, figure 4A). The effect persisted within the KRAS wild-type CRC subgroup (44.8 vs 39.8 months, p=0.043; HR 1.11 (95% CI 1.01 to 1.23)), but not in the KRAS mutant CRC subgroup (33.1 vs 34.7 months, p=0.576) (figure 4E,F). Similarly, increased AXL expression was correlated with decreased OS in the CRC cohorts treated with bevacizumab (36.8 vs 32.6 months, p=0.013; HR 1.21 (95% CI 1.02 to 1.23)) and anti-EGFR therapy with cetuximab/panitumumab (28.4 vs 22.2 months, p=0.005; HR 1.21 (95% CI 1.06 to 1.38)) (figure 4C,D). Interestingly, AXL expression did not significantly predict OS in the pembrolizumab/nivolumab-treated CRC cohort (19.2 vs 20.6 months, p=0.931), but increased AXL expression conferred improved OS in the KRAS mutant subgroup (11.6 vs 23.4 months, p=0.012; HR 0.65 (95% CI 0.46 to 0.91)) (figure 4B and G,H). When using a median cut-off or quartile analysis, increased AXL expression continued to confer worse OS in the same subgroups above (FOLFOX/FOLFIRI, bevacizumab, anti-EGFR, FOLFOX/FOLFIRI—KRAS wild-type) and improved OS with immunotherapy-treated KRAS mutant CRC, although FOLFOX/FOLFIRI—KRAS wild-type lost significance in quartile analysis (p=0.059) (online supplemental figures S3-S4). OS of the Caris cohort by AXL tertiles in CRC (A) treated with FOLFOX/FOLFIRI, (B) immunotherapy, (C) anti-EGFR therapy, (D) bevacizumab, (E) FOLFOX/FOLFIRI KRAS wild-type, (F) FOLFOX/FOLFIRI KRAS mutant, (G) immunotherapy KRAS wild-type, and (H) immunotherapy KRAS mutant. Patient data was obtained from the Caris CODEai. T1 and T3 are defined in global CRC cohorts to ensure consistent cut-off levels. CRC, colorectal cancer; EGFR, epidermal growth factor receptor; FOLFIRI, fluorouracil, leucovorin, and irinotecan; FOLFOX, fluorouracil, leucovorin, and oxaliplatin; OS, overall survival. Given the strong correlation between AXL expression and PD-L1 status, we examined the prognostic value of their combined expression by comparing patients with high AXL expression (above median) and PD-L1 positivity to those with low AXL expression (below median) and PD-L1 negativity. In the conventional treatment subgroups (FOLFOX/FOLFIRI and cetuximab/panitumumab), patients with high AXL/PD-L1 positivity had significantly worse OS compared with those with low AXL/PD-L1 negativity. In contrast, among patients treated with pembrolizumab or nivolumab, high AXL/PD-L1 positivity was associated with improved OS (online supplemental figure S5). This analysis is limited by the substantial asymmetry between the two groups, as PD-L1 positivity was only 5% in the Caris cohort, resulting in markedly unequal sample sizes. In CALGB/SWOG 80405, univariate analysis demonstrated that patients with CRC with high AXL expression had significantly shorter PFS (T1: 12.9 vs T2: 11.7 vs T3: 9.2 months, p=0.001; HR: 1.56 (95% CI 1.23 to 1.99)) and OS (34.7 vs 34.0 vs 24.2 months, p<0.001; HR: 1.68 (95% CI 1.30 to 2.16)) (figure 5). When stratifying by treatment groups, the CRC cohort with higher AXL expression had significantly shorter PFS in FOLFOX (T1: 12.6 vs T2: 10.9 vs T3: 9.2 months, p=0.012; HR 1.52 (95% CI 1.15 to 2.01)), cetuximab (11.1 vs 12.9 vs 8.7 months, p=0.006; HR 1.58 (95% CI 1.11 to 2.26)), and bevacizumab (13.5 vs 10.1 vs 10.1 months, p=0.027; HR 1.54 (95% CI 1.11 to 2.14)) treated patients (online supplemental figure S6). Additionally, all treatment subgroups demonstrated a similar trend with decreased OS correlating with increased AXL expression (FOLFOX: T1: 31.8 vs T2: 32.8 vs T3: 24.2 months, p=0.002, HR 1.62 (95% CI 1.21 to 2.16); FOLFIRI: 39.6 vs 39.1 vs 24.7 months, p=0.018, HR 1.95 (95% CI 1.15 to 3.31); bevacizumab: 34.4 vs 31.3 vs 25.9 months, p=0.04, HR 1.55 (95% CI 1.10 to 2.18); cetuximab: 34.9 vs 39.1 vs 20.2 months, p<0.001, HR 1.84 (95% CI 1.26 to 2.70)) (online supplemental figure S6). When using a median cut-off for AXL, higher AXL levels continued to confer worse PFS and OS in the general cohort, though statistical significance was lost for some treatment subgroups (online supplemental figure S7). Association between AXL expression and patient outcomes in the CALGB/SWOG 80405 trial for (A) PFS and (B) OS. CALGB/SWOG, Cancer and Leukemia Group B/Southwest Oncology Group; OS, overall survival; PFS, progression-free survival. When analyzed as a continuous variable, higher AXL expression was associated with worse PFS and OS (Plinear=2E-04 and 2.2E-05, respectively). On multivariate analysis, higher AXL expression continued to demonstrate significantly shorter PFS (T3 vs T1 (reference) adjusted HR 1.53 (95% CI 1.19 to 1.97)) and OS (adjusted HR 1.73 (95% CI 1.33 to 2.25)). While the high AXL-expressing cetuximab group had a numerically lower OS relative to bevacizumab-treated patients with high AXL expression (cetuximab T3: 20.2 months vs bevacizumab T3: 25.9 months) in CALGB/SWOG 80405 (online supplemental figure S6), no significant interaction between AXL expression and treatment was observed for PFS or OS (likelihood ratio test). Discussion The AXL signaling pathway has emerged as a regulator of tumor progression through increased EMT, proliferation, and migration.24 Preclinical studies exploring AXL inhibitors have shown potential in overcoming therapeutic resistance, especially when combined with immune checkpoint inhibitors and chemotherapy.25 Our study expands on these findings by correlating the molecular profile of AXL with clinical outcomes in both a randomized clinical trial cohort and a real-world dataset, providing new insights into its impact on PFS, OS, and immune characteristics within the CRC TME. We emphasize the potential translational implications of AXL-targeted therapies by bridging the controlled clinical trial data with real-world evidence. Our study highlighted the increased AXL messenger RNA (mRNA) expression in primary CRC samples relative to metastatic sites, indicating that alterations in AXL expression may play a more prominent role in the early stages of CRC growth and local invasion (online supplemental figure S1). Notably, AXL has previously been shown to be upregulated (mRNA) in primary CRC tumor samples relative to normal colonic tissues.18 Studies of matched tumor samples in breast cancer showed increased AXL IHC expression in metastatic sites relative to primary tumors, highlighting key differences in tumor biology between primary and metastatic tumor sites.26 When evaluating AXL expression in individual metastatic sites in our cohort, samples obtained from peritoneal metastasis demonstrated the highest relative AXL TPM, exceeding the levels of AXL mRNA expression observed in primary tumors. These findings correlate with a previous study that found a single peritoneal metastatic sample to have 10× the RNA expression of its matched primary CRC sample.27 The development and growth of metastasis in the poorly vascularized lining of the peritoneum often requires tumor cells to undergo EMT to implant and grow within the peritoneal cavity, which correlates with the upregulation of AXL.28 A hypoxic environment is also typical in peritoneal metastasis, leading to increased levels of hypoxia-inducible transcription factors, which promote AXL/GAS6 signaling.29 The worse prognosis of peritoneal metastasis relative to liver/lung spread CRC is partly due to decreased treatment efficacy, but targeting AXL either directly or through antibody-drug conjugates may offer improved outcomes for this subgroup.30 31 Elevated AXL expression levels were correlated with increased PD-L1 positivity (table 1), consistent with preclinical studies in melanoma and lung cancer, demonstrating a similar AXL/PD-L1 association.32 Although high AXL expression was associated with increased PD-L1 positivity, it correlated with decreased TMB-H and showed no significant difference in MSI-H/dMMR status. This finding diverges from the classic “immune hot” phenotype, which is characterized by concurrent elevation of PD-L1, TMB, and MSI-H.33 One plausible explanation is that AXL promotes EMT, a process strongly associated with PD-L1 upregulation independent of tumor mutational burden.34 Mechanistically, AXL signaling can directly enhance PD-L1 expression via activation of the PI3K/Akt pathway and transcriptional regulation, fostering adaptive immune resistance even in the absence of increased neoantigen load.35 Thus, the immunosuppressive phenotype observed in high AXL tumors may reflect alternative, non-mutational routes of immune escape. The correlation between high AXL expression and elevated levels of immune cell types, such as B cells, M1/M2 macrophages, T-regs, and NK cells, suggests that higher levels of AXL expression might contribute to an immunosuppressive yet complex immune environment (figure 3). Specifically, T-regs and M2 macrophages are known for their immunosuppressive effects that can promote immune evasion. Cell line studies of oral squamous cell carcinomas have implicated AXL activation in M2 macrophage polarization.36 However, a study evaluating AXL expression within bone marrow-derived macrophages noted increased expression following the activation of M1 (IFN-γ and TNF-α) and M2 (interleukin-4) phenotypes, indicating the context-dependent role of AXL within the TME.37 In the Caris cohort, lower AXL expression was correlated with greater infiltration of dendritic cells and neutrophils, which are often involved in proinflammatory and antigen-presenting functions (figure 3). Similarly, AXL knockout preclinical models have demonstrated an increase in the infiltration of bone marrow-derived and splenic dendritic cells relative to wild-type AXL.38 39 These findings suggest that AXL may play a role in suppressing dendritic cell recruitment and activation, potentially limiting antigen presentation, and reducing the antitumor immune response.39 40 In both our clinical trial cohort and real-world data set, increased AXL expression was associated with poorer OS across multiple treatment regimens in CRC, suggesting that AXL may serve as a prognostic marker for aggressive disease (figures 4 and 5). Our multivariate analysis of CALGB/SWOG 80405 confirmed that AXL expression independently predicted shorter PFS and OS, further establishing its prognostic relevance in mCRC. When analyzed as a continuous variable, high AXL expression was consistently correlated with worse survival outcomes, indicating that higher levels of AXL expression correlated with tumor progression. These findings complement previous research where AXL positivity was noted to confer worse OS in mCRC (AXL-positive, 23.7 months; AXL-negative, 30.8 months; HR, 1.455, p=0.032) treated with chemotherapy+targeted therapy (tailored to RAS alterations) by Cardone.20 When stratified based on RAS status, only the RAS wild-type subgroup showed improved OS and PFS. It is important to note that most patients enrolled in CALGB/SWOG 80405 had wild-type RAS. This likely contributes to the large prognostic effect of increased AXL expression (table 1). Initially, the impact of KRAS mutations on resistance to anti-EGFR therapy was not yet known when CALGB/SWOG 80405 had begun. Three years after the trial began, it was discovered that KRAS-mutant tumors conferred resistance to this treatment. Consequently, the study protocol was amended to require KRAS wild-type status (codons 12 and 13) as the inclusion criterion. When conducting subgroup analysis based on KRAS status in FOLFOX/FOLFIRI-treated patients in the Caris dataset, increased AXL expression continued to confer worse OS in the KRAS wild-type subgroup (p=0.043), but not in the KRAS mutant (p=0.576) subgroup (figure 4E,F). These findings highlight an important interaction between AXL expression and the RAS/KRAS status in CRC. In KRAS mutant tumors, the constitutively active KRAS protein drives tumor progression through MAPK/ERK and PI3K/AKT cascades, which may make these tumors less dependent on upstream RTKs, such as AXL, for survival signals. AXL expression may have a reduced impact on prognosis in RAS-mutant tumors because mutant RAS activates tumor-driving signaling pathways. Additionally, the high AXL-expressing cetuximab group had a numerically lower OS relative to bevacizumab-treated patients with high AXL expression (20.2 months vs 25.9 months) in CALGB/SWOG 80405 (online supplemental figure S6). While we did not observe a significant treatment interaction effect, this numerical difference may be explained by downstream MAPK expression secondary to high AXL expression. However, it is important to note that this observation is exploratory and hypothesis-generating and should be interpreted cautiously due to the lack of statistical significance and the potential for confounding factors. Further studies are warranted to clarify the clinical implications of differential OS in these subgroups. Interestingly, when evaluating the CRC cohort that was treated with immunotherapy in the Caris database, AXL was not prognostic in the general cohort, but higher AXL expression was correlated with improved OS in the KRAS mutant subgroup (figure 4H). While KRAS mutations drive immune evasion through lactic acid-mediated suppression of CD8+ T cell activity and NF-κB inactivation, high AXL expression in these tumors could partially counteract this immunosuppression by fostering an IFN-γ-enriched microenvironment (figure 2B) or recruiting proinflammatory macrophages and NK cells (figure 3).41 42 Notably, AXL has been shown to regulate both immunosuppressive (M2 macrophages and T-regs) and immunostimulatory (M1 macrophages and activated dendritic cells) populations depending on the cellular context, which may explain its divergent effects across molecular subgroups. These findings suggest that AXL overexpression in KRAS mutant CRC may overcome the intrinsic resistance mechanisms to immune checkpoint blockade. Further studies are warranted to evaluate the interaction between AXL and KRAS in immunotherapy. The poor prognostic effect of increased AXL expression coupled with its integral role in cancer progression and metastasis highlights its potential for future targeted therapies. The first-in-human trial of an AXL kinase inhibitor (TP-0903) in solid tumors, including CRC, demonstrated a favorable safety profile and suggested that higher AXL expression may correlate with an improved therapeutic response (NCT02729298).43 Beyond selective AXL inhibitors, multitarget tyrosine kinase inhibitors such as cabozantinib and zanzalintinib, which simultaneously target AXL, MET, and VEGFR2, have shown promising activity in combination with immune checkpoint inhibitors in CRC. In the CAMILLA (multi-cohort, multi-center clinical trial of cabozantinib and durvalumab in advanced gastrointestinal malignancies) basket trial, cabozantinib plus durvalumab demonstrated an objective response rate of 27.6% in patients with refractory mCRC, with particularly encouraging results in RAS wild-type patients (50% response rate) (NCT03539822).44 These findings have led to Phase III trials evaluating zanzalintinib in combination with atezolizumab in microsatellite-stable CRC (NCT05425940), demonstrating the potential for multitarget approaches to overcome immunotherapy resistance. The immunosuppressive effects of AXL have also inspired a phase 1 trial to evaluate an AXL/MER inhibitor (PF-07265807) combined with programmed cell death protein 1 inhibition (sasanlimab) in MSS CRC (NCT04458259). In addition to conventional approaches, the relative specificity of AXL expression in CRC compared with that in normal colonic cells has also prompted the investigation of AXL-targeted CAR-T cells in a phase 1 study (NCT04842812). The limitations of our study include its retrospective design, heterogeneity of the population, and absence of detailed information on treatment sequences in the Caris dataset. Additionally, we did not have AXL IHC data to validate our gene expression findings or to confirm the functional activation of AXL. Furthermore, gene expression thresholds were derived internally from this dataset, which may have affected generalizability. Given the large numbers within the Caris dataset, differences in OS for some treatment regimens were statistically significant but may not be clinically significant due to relatively small differences between the groups. It is also important to note that the CALGB/SWOG 80405 trial was conducted between 2005 and 2012, prior to the widespread adoption of immunotherapy in CRC treatment. As a result, differences in the timing and therapeutic landscape between the CALGB/SWOG 80405 trial and the Caris real-world cohort may limit the direct applicability of our findings to the current treatment era. Despite these limitations, our study focused on biomarker discovery, underscoring the need for future prospective studies to validate the prognostic value of AXL expression in CRC. Such validation will be especially critical in the context of contemporary trials that incorporate immunotherapy-based regimens, to determine whether our findings remain relevant in the current standard of care. In summary, our study provides new insights into the role of AXL in CRC, highlighting its association with aggressive tumor behavior, immune modulation, and poor survival outcomes across conventional treatment regimens. By examining AXL expression in both a clinical trial cohort and real-world data, we demonstrated its potential as a prognostic biomarker, particularly in KRAS wild-type tumors. While immunotherapy, along with many other therapies, is less effective in KRAS-mutant CRC, our findings suggest that the immunosuppressive microenvironment driven by elevated AXL expression may paradoxically prime these cancers for a stronger response to checkpoint inhibitors. TME analysis shows that AXL-high tumors exhibit dual immunosuppressive (M2 macrophages, T-regs) and inflammatory (NK cells, IFN-γ) features, suggesting distinct therapeutic vulnerabilities. For KRAS-wildtype AXL-high CRC, combination strategies targeting both AXL and chemotherapy may overcome intrinsic resistance (eg, AXL inhibitors+FOLFOX), whereas KRAS-mutant AXL-high tumors could be more responsive to therapies that enhance antigen presentation or proinflammatory immune responses. These findings reinforce the potential importance of AXL as a therapeutic target and support further exploration of tailored treatment strategies, especially in combination with checkpoint inhibitors. Supplementary material