3525 Background: Tumor microenvironment (TME) features of colorectal cancer (CRC) may influence detection of circulating tumor DNA (ctDNA) and further refine prognosis. We applied a deep learning histopathologic algorithm to quantify TME features in stage III colon cancers and evaluated their association with postoperative ctDNA status among participants in a phase III FOLFOX-based adjuvant trial (NCCTG N0147). Methods: QuantCRC (Aiforia Technologies) was applied to digitized hematoxylin and eosin-stained whole-slide images of stage III colon adenocarcinomas to extract quantitative histopathologic features. Among cases where data met quality control (N=1817), associations between 15 distinct pathologist-defined features and clinical outcomes—time to recurrence (TTR), disease-free survival (DFS), and overall survival (OS)—were evaluated, stratified by postoperative ctDNA status assessed by a tissue-free assay (Guardant Reveal). Relationships were analyzed using univariate and multivariable Cox proportional hazards models that were adjusted for clinicopathologic and molecular features ( KRAS , BRAF V600E , MMR status). Interaction testing was performed. Results: We found statistically significant quantitative differences for histopathological features by ctDNA status. Features associated with ctDNA positivity(+) included lymphovascular invasion (LVI), and increased %high grade, %tumor budding, %necrosis, and %stroma. ctDNA+ cases had lower %inflammatory stroma and fewer %signet ring cells (all p< 0.006). Multivariable analyses, irrespective of ctDNA status, identified %tumor budding, %mucin, %signet ring cells, %inflammatory tumor bed, and %inflammatory stroma as significantly associated with adverse outcomes (TTR, DFS, OS). Within ctDNA+ cases, higher %immature tumor bed was associated with adverse outcome for all 3 variables. Among ctDNA-negative cases, significantly poorer outcomes (TTR, DFS, OS) were observed for tumors with lower tumor infiltrating lymphocyte (TIL) density and reduced %necrosis. After adjustment, significant interactions with ctDNA status were observed for LVI (p< 0.006) and %signet ring cells (p< 0.035) across all outcome variables, for %necrosis with DFS, OS (p<0.040), and for TILs with TTR (p<0.015). Analysis is ongoing for ctDNA tumor fraction as well as tumor genotyping data (739 genes; Guardant360). Conclusions: QuantCRC identifies TME features associated with postoperative ctDNA status that enable risk-stratification within ctDNA groups. Among ctDNA-negative patients, reduced intratumoral TIL density and reduced tumor necrosis identify high-risk subgroups with inferior clinical outcomes, and may have clinical utility.

103 Background: Clinicians spend significant time extracting and synthesizing information to risk-stratify and make adjuvant chemotherapy (AC) recommendations for CRC. By automating parts of this process, LLMs could help reduce documentation burden and variability in decision-making. Methods: We selected 144 stage III, pMMR CRC patients treated at Mayo Clinic (2017–2021). HIPAA-compliant Gemini-2.0-Flash-001 and GPT-4o extracted clinical variables (age, neuropathy grade, pathological stage) from EHRs and generated AC recommendations. Two approaches were compared: (1) a rule-based algorithm with nested if/else statements translating variables into NCCN-based recommendations, and (2) a dynamic LLM agent generating recommendations using a structured prompt informed by NCCN guidelines. Concordance with treating physicians was assessed, with disagreements reviewed by a blinded GI oncologist. Results: The dataset included 144 patients with median age 67 (IQR 66–68), 48.6% female. The LLM pipeline extracted neuropathy grade with 100% accuracy and pathological T/N stages with 99.3% accuracy. Concordance with physician recommendations was 93.1% for the rule-based algorithm and 62.5% for the LLM agent. In discordant cases, blinded adjudication by a GI oncologist favored the algorithm and LLM agent over physicians in most cases (Table 1). Conclusions: This exploratory work shows LLM agents can reliably extract variables and generate evidence-based treatment recommendations for stage III CRC. The high rate of expert preference for LLM recommendations in discordant cases is intriguing but limited by small sample size and possible adjudicator bias. Discordant physician recommendations reflect the complexity of decision-making, including patient preferences, performance status, comorbidities, and subtle findings not fully captured in EHRs. Rather than replacing judgment, we envision LLMs as decision support tools—analogous to an advanced trainee reviewing charts and presenting evidence-based recommendations to the attending physician with ultimate authority. Larger multi-institutional studies with multiple adjudicators are needed to validate these findings and clarify AI’s role in oncology. Patient characteristics, LLM extraction accuracy, and concordance with physician recommendations. Metric Result Total patients 144 Median age (IQR) 67 (66-68) Female (%) 48.6% LLM extraction accuracy – Neuropathy grade 100% LLM extraction accuracy – Pathological T and N stages 99.3% Primary Concordance – Rule-based algorithm vs. physician 93.1% Adjudicator preference of rule-based algorithm over treating physician for discordant cases 100% Concordance – LLM agent vs. physician 62.5% Adjudicator preference of LLM agent over treating physician for discordant cases 82.0%

3626 Background: For patients (pts) with dMMR stage III colon cancer, the addition of atezolizumab (atezo) to adjuvant 5-fluorouracil, leucovorin plus oxaliplatin (mFOLFOX6) significantly improved disease-free survival in ATOMIC (NCT02912559). Here we present HRQOL data. Methods: Pts with surgically resected stage III dMMR colon cancer were randomized (1:1) to receive mFOLFOX6 + atezo (840 mg IV q2 weeks) for 12 cycles (6 months [mo]) followed by atezo monotherapy for 13 cycles (12 mo total) versus mFOLFOX6 for 12 cycles. HRQOL was an exploratory endpoint measured using the FACT-C (includes FACT-G), FACT/GOG-NTX, EQ5D-5L, and PRO-CTCAE. Optional participation was by paper surveys at trial registration, prior to treatment cycles 4 and 7, and at 6, 12, and 36 mo after registration. Mean changes from baseline (BL) were compared between treatment arms using general linear mixed models (negative mean changes indicate worsening). At 6 mo, a two-sided 95% confidence interval excluding a difference of 1.9 points (FACT/GOG-NTX) and 6 points (FACT-C Trial Outcome Index [TOI]) favoring the mFOLFOX6 arm would indicate non-inferiority of the addition of atezo. Rates of side effect bother (FACT-G item GP5) and pt-reported adverse events (AEs) by PRO-CTCAE (composite scores, baseline adjusted) were compared using Fisher’s exact tests. Results: Of 712 randomized pts, 581 (285 atezo+mFOLFOX6; 296 mFOLFOX6) consented to participate in the survey. Pts completed 2398/3021 (79%) of expected surveys across all time points (1218/1489 [82%] atezo+mFOLFOX6; 1180/1532 [77%] mFOLFOX6). The difference in mean changes from BL between arms indicated non-inferiority of the addition of atezo at 6 mo (table); differences remained small at 12 and 36 mo. No significant differences in mean changes from BL between arms were found at 6, 12, or 36 mo for any scale. At 6 mo, 16% vs 20% of pts in each arm reported at least “quite a bit” of side effect bother (p=0.34), favoring atezo. Across cycles, atezo was associated with higher rates of pt-reported itchy skin (62% vs 48%), cough (53% vs 43%), and shortness of breath (51% vs 41%) [all p<0.05]. Conclusions: Addition of atezo to mFOLFOX6 slightly increases pt-reported AEs, but its effect on HRQOL is minimal and not clinically meaningful. Results support tolerability of atezo+mFOLFOX6 as standard of care adjuvant treatment for pts with dMMR stage III colon cancer. Support: UG1CA189823, U10CA180821, U10CA180882; Genentech, a member of the Roche group; https://acknowledgments.alliancefound.org. Clinical trial information: NCT02912559 . Scale Mo Mean change from BL – atezo+mFOLFOX6 Mean change from BL – mFOLFOX6 Diff (95% CI) FACT/GOG-NTX 6 -10.0 -10.6 0.6 (-0.8, 2.1) 12 -8.0 -9.4 1.4 (-0.2, 2.9) 36 -7.2 -6.9 -0.4 (-2.1, 1.4) FACT-C TOI 6 -4.1 -2.5 -1.6 (-3.7, 0.6) 12 2.1 1.8 0.3 (-1.9, 2.6) 36 0.7 1.8 -1.1 (-3.7, 1.5)

PURPOSE:Detection of molecular residual disease using circulating tumor DNA (ctDNA) may enable postoperative risk stratification and guide adjuvant therapy. We evaluated the prognostic value of a tissue-free, epigenomic ctDNA assay in patients with stage III colon cancer (CC) enrolled in a phase III adjuvant chemotherapy trial. METHODS:Plasma samples were collected after surgery and before adjuvant infusional fluorouracil, leucovorin, and oxaliplatin alone or combined with cetuximab. ctDNA was analyzed using a tissue-free assay; in ctDNA-positive patients, tumor fraction (TF) was quantified and genotyping was performed with a 739-gene panel. Associations with disease-free survival (DFS), time to recurrence (TTR), and overall survival (OS) were assessed using multivariable Cox models adjusted for covariates. RESULTS:Among 2,260 evaluable patients, 461 (20.4%) were ctDNA-positive with significantly higher detection in advanced T-/N-stage, high-grade, obstruction/perforation, and BRAFV600E tumors. At a median follow-up of 6.1 years, ctDNA positivity was independently associated with shorter TTR (hazard ratio [HR], 5.96 [95% CI, 5.11 to 6.96]), DFS (HR, 5.03 [95% CI, 4.36 to 5.81]), and OS (HR, 4.45 [95% CI, 3.76 to 5.27]; all P < .0001). The 5-year DFS was 27.7% (95% CI, 23.8 to 32.2) v 77.1% (95% CI, 75.1 to 79.1) in ctDNA-positive versus ctDNA-negative patients, and adverse prognostic impact was greater in lower T/N stage, low-risk, and dMMR subsets (interaction P = .0012-.041). Among ctDNA-positive patients, TF was nearly double in those who recurred or died (P = .0002) and stratified patients for TTR, DFS, and OS (all adjusted P < .002). Genotyping identified mutations in FLT1 (OR, 8.99) and PREX2 (OR, 7.73) genes that were most strongly associated with recurrence (P < .03). CONCLUSION:Evaluation of ctDNA in resected stage III CC using a tissue-free assay provided robust and independent prognostic value. Higher ctDNA burden, dMMR, and specific mutations defined poor prognostic groups among ctDNA-positive patients.

Abstract Background: The BRAF(V600E) mutation in colorectal cancer (CRC) is associated with treatment resistance and a poor prognosis. CDK2/9 is frequently activated in human cancers, and we have previously shown that BRAFV600E can upregulate MCL-1, which can be suppressed by the selective CDK2/9 inhibitor fadraciclib. Herein, we determined whether targeting CDK2/9 by fadraciclib can be enhanced by concurrent Bcl-2/Bcl-xL inhibition using LP-118, which minimizes platelet toxicity in patients, compared to the Bcl-2 inhibitor venetoclax. Effects of the drugs on apoptosis, colony formation and tumor growth were studied in a zebrafish model. Methods: We utilized isogenic human RKO (BRAFV600E/V600E/WT) and T29 (BRAFWT/−/−) CRC cell lines in addition to HCT-116 (BRAFWT) and Bax knockout (HCT-116Bax-/-) CRC cells. Cells were treated for 24h with fadraciclib (500nM) plus encorafenib (500nM) in the presence or absence of LP-118 (1µM) or venetoclax (1µM). Apoptosis and cell growth inhibition were assessed using annexin V/PI staining and colony formation assays, respectively. Expression of MCL-1, Bcl-xL, Bcl-2, PARP, and cleaved caspase-3 (Asp175) proteins was analyzed. In a zebrafish tumor xenograft model, the effects of the drugs on caspase-3 cleavage and tumor growth were then evaluated. Results: Fadraciclib suppressed MCL-1 expression, induced apoptosis and significantly inhibited colony formation in all CRC cell lines. However, this effect was enhanced by encorafenib only in RKO (BRAFV600E/V600E/WT) cells. The addition of LP-118 to fadraciclib plus encorafenib potently enhanced apoptosis induction, shown by annexin V and cleavage of PARP, caspase-3, in RKO and HCT-116 cell lines, which was synergistic and Bax dependent. Treatment with LP-118 combined with fadraciclib and encorafenib suppressed both Bcl-xL and MCL-1 expression in RKO cells. The addition of LP-118 to fadraciclib plus encorafenib enhanced apoptosis (annexin V) to a greater extent than did venetoclax. In a zebrafish tumor xenograft model, the addition of LP-118 to fadraciclib plus encorafenib significantly enhanced caspase-3 activation and reduced the tumor xenograft size. Conclusion: LP-118 synergistically enhanced the antitumor activity of fadraciclib plus encorafenib in BRAFV600E human CRC cells, and to a greater extent than did venetoclax. Mechanistically, targeting Bcl-2/Bcl-xL concurrent with inhibition of MCL-1 and CDK2/9 by fadraciclib potently induced apoptosis and suppressed tumor cell growth. These findings suggest a novel and effective therapeutic strategy for BRAFV600E human CRCs. Citation Format: Md Mohiuddin, Vanda Póvoa, Beatriz Sebo, Rita Fior, Frank A. Sinicrope. Novel Bcl-2/Bcl-xL inhibitor, LP-118, enhances apoptosis induction by the CDK2/9 inhibitor, fadraciclib, in BRAF(V600E) human colorectal cancer cells [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4658.

PURPOSE:Routine testing for mismatch repair deficiency (dMMR)/microsatellite instability (MSI) in colorectal cancers is recommended for Lynch syndrome screening, prognosis, and treatment guidance. In the metastatic setting, RAS/BRAF mutations guide treatment decisions. The impact of these mutations on the tumor immune microenvironment (TiME) in MSI/dMMR colorectal cancer is not known. EXPERIMENTAL DESIGN:The study involved retrospective analysis of 448 patients with stage I to IV MSI/dMMR colorectal cancer profiled using next-generation sequencing (Tempus xT DNA sequencing of 595-648 genes at 500x coverage and Tempus xR whole-exome capture RNA sequencing). MSI status was determined by assessing 44 or 239 loci using next-generation sequencing. dMMR was determined using IHC. Tumor mutational burden, tumor neoantigen burden (NTB), PD-L1, immune infiltration, and canonical immune pathways (76 gene set signatures) were analyzed. RESULTS:The median age at diagnosis was 67 years (range, 21-86); 59% of patients were female, and 70% were stage III to IV. Among the 448 patients, 100 (22%) harbored RAS mutations (RASmut), 119 (27%) a BRAFV600E mutation, and 229 (51%) were double wild type. RASmut exhibited lower NTB (median, 12 vs. 15 vs. 16; P = 0.003) and PD-L1 (3.6% vs. 13% vs. 24%; P < 0.001) than BRAFV600E and wild-type tumors, respectively. The RASmut TiME had lower overall inflammation and fewer infiltrating CD8+ T cells than wild-type or BRAFV600E tumors. In contrast, BRAFV600E tumors exhibited hyperproliferative characteristics associated with broad metabolic reprogramming but a similarly inflamed TiME compared with wild-type tumors. CONCLUSIONS:Our data suggest that MSI/dMMR colorectal cancers harboring RASmut are less immunogenic, and the TiME contains a lower inflammatory profile than wild-type or BRAFV600E tumors. Further analysis and validation are required to confirm these findings.

HLA loss of heterogeneity (HLA LOH) and tumor neoantigen burden in relationship to clinical outcome variables in dMMR CRC.

Background Standard adjuvant chemotherapy for stage III colon cancer consists of a fluoropyrimidine-plus-oxaliplatin regimen. Whether the addition of atezolizumab (an anti-programmed death ligand 1 agent) to a modified FOLFOX6 regimen (fluorouracil, oxaliplatin, and leucovorin; called mFOLFOX6) would improve outcomes in patients with stage III colon cancer with mismatch repair-deficient (dMMR) status is unclear. Methods In a phase 3 trial, we randomly assigned, in a 1:1 ratio, patients with resected stage III dMMR tumors to receive either adjuvant atezolizumab plus mFOLFOX6 for 6 months, with atezolizumab continued as monotherapy (for a total of 12 months of therapy), or mFOLFOX6 alone for 6 months. The primary end point was disease-free survival. Secondary end points were overall survival and the adverse-event profile. Results A total of 355 patients were assigned to receive atezolizumab plus mFOLFOX6 and 357 to receive mFOLFOX6 alone. The median age of the patients was 64 years, 55.1% were women, and 53.9% had tumors that were T4, N2, or both (indicating high risk). At a median follow-up of 40.9 months, the 3-year disease-free survival was 86.3% (95% confidence interval [CI], 81.8 to 89.8) in the atezolizumab-mFOLFOX6 group, as compared with 76.2% (95% CI, 70.9 to 80.6) in the mFOLFOX6 group (hazard ratio for disease recurrence or death, 0.50; 95% CI, 0.35 to 0.73; P<0.001). Adverse events of grade 3 or 4 occurred in 84.1% of the patients who received atezolizumab plus mFOLFOX6 and in 71.9% of those who received mFOLFOX6 alone. Conclusions The addition of atezolizumab to mFOLFOX6 significantly improved disease-free survival among patients with stage III dMMR colon cancer. (Funded by the National Cancer Institute of the National Institutes of Health and Genentech; ATOMIC ClinicalTrials.gov number, NCT02912559.)

PURPOSE:Immune checkpoint blockade (ICB) induces frequent and durable responses in metastatic deficient DNA mismatch repair (dMMR) colorectal cancer, yet substantial molecular heterogeneity and resistance remain. We sought to identify candidate tumor- and immune-related biomarkers associated with clinical outcomes following anti-PD-1 therapy. EXPERIMENTAL DESIGN:Consecutive patients with metastatic dMMR colorectal cancer (N = 39) treated with anti-PD-1 therapy underwent tumor profiling using a validated immune-enhanced exome and transcriptome platform. Microsatellite instability (MSI) burden was quantified as the percentage of unstable microsatellite loci using MSIsensor-pro. Associations with objective response were evaluated, and progression-free survival (PFS) and overall survival (OS) were analyzed using Cox proportional hazards models. RESULTS:Higher MSI burden was associated with improved objective response (P = 0.018) and survival. The dichotomized MSI level (Q2-4 vs. Q1) was associated with longer PFS [hazard ratio (HR), 0.18; 95% confidence interval (CI), 0.06-0.56; P = 0.003] and OS (HR, 0.20; 95% CI, 0.07-0.58; P = 0.003), with similar results when modeled continuously. MSI burden correlated with neoantigen clonality but not burden (R = 0.53, P = 0.01). Responders exhibited significantly greater T-cell receptor repertoire diversity; both T- and B-cell receptor diversities were associated with survival. Although human leukocyte antigen (HLA) A, HLA-B, and HLA-C expression was not prognostic, the HLA-B*07:02 allele was associated with the best overall response. In contrast, overexpression of immune exhaustion-related genes and cytotoxic T-cell and natural killer cell exhaustion phenotypes were associated with ICB resistance and poorer prognosis. CONCLUSIONS:Integrated exome/transcriptome profiling with MSI quantification identified MSI burden and adaptive immune repertoire diversity as potential correlates of response and survival following ICB. These findings suggest a mechanistic link among genomic instability, antigen recognition diversity, and immunotherapy benefit.

The treatment paradigm for deficient mismatch repair (dMMR) colorectal cancer (CRC) has evolved substantially in recent years. Universal mismatch repair testing and broad use of next-generation sequencing have increased the detection of dMMR tumours, which exhibit microsatellite instability-high (MSI-H), hypermutation and abundant neoantigens. The presence of dMMR/MSI-H serves as a robust predictive biomarker for immune checkpoint inhibitor (ICI) therapy, which has demonstrated superior efficacy to cytotoxic chemotherapy in the metastatic setting and led to the first tumour-agnostic Food and Drug Administration approval in 2017 for metastatic dMMR/MSI-H solid tumours. Recent evidence also supports the benefit of ICIs in non-metastatic dMMR CRC. Neoadjuvant immunotherapy has produced high rates of pathological response in both colon and rectal cancers. In locally advanced dMMR rectal cancer, ICI therapy has enabled omission of chemoradiation and surgery in most patients. In resected node-positive dMMR colon cancer, the addition of ICI therapy to chemotherapy has substantially improved disease-free survival, expanding its role earlier in the disease course. Despite these advances, the optimal treatment strategy for non-metastatic dMMR CRC remains undefined due to the lack of direct comparative studies. Importantly, a subset of patients derives limited or no benefit from ICIs despite dMMR/MSI-H status, underscoring the need to further elucidate resistance mechanisms and to develop strategies to overcome them.

BACKGROUND:Deficient DNA mismatch repair/microsatellite instability-high (dMMR/MSI-H) cancers are very sensitive to immune checkpoint inhibitors (ICIs), yet their use is frequently complicated by immune-related adverse events (irAEs). Our study analyzed the impact of grade ≥3 irAEs (immune-related severe adverse events (irSAEs)) on survival in patients with dMMR/MSI-H digestive cancers treated with ICIs. METHODS:We conducted an international, multicenter, ambispective study involving 1,175 patients from 34 centers. The primary endpoint was the correlation between the occurrence of irSAEs and progression-free survival (PFS). Secondary endpoints included factors associated with irSAEs, correlation between irSAEs and overall response rate (ORR) and overall survival (OS). Prespecified landmark and time-dependent survival models were used to account for the time to irSAE occurrence. RESULTS:Among 1,175 patients treated with ICIs for digestive cancers, 49.1% were female, median age was 66.9 years (IQR 53.7-77.3) and 82.9% had colorectal cancer. Overall, 382 patients (32.5%) had an irAE, including 117 (10.0%) irSAEs. Median time to irSAEs occurrence was 3.78 months (IQR 1.97-7.62). The most frequent irSAEs were gastrointestinal (3.4%), hepatic (1.8%) and dermatologic (1.0%). The only factor associated with irSAEs was the use of an ICI combination versus monotherapy (20.0% vs 11.4%, p=0.03). Among patients with irSAEs, 71.5% received oral and 29.8% intravenous corticosteroids, and 15.7% immunosuppressive agents. Overall, in 26.9% of patients, ICIs were resumed following irSAEs, with an irAE recurrence rate of 57.1%. irSAE was associated with a better ORR (48.6% vs 34.6%, p<0.0001). The multivariable Cox model revealed that irSAE as a time-dependent variable was not associated with better PFS (HR=1.280 (95% CI 0.898 to 1.824), p=0.172) or OS (HR=1.043 (95% CI 0.712 to 1.528), p=0.828). Likewise, landmark analysis at 3 months and 6 months also showed that irSAEs were not associated with better PFS or OS. CONCLUSIONS:The occurrence of irSAE was not independently associated with better survival in patients with dMMR/MSI-H digestive cancers treated with ICIs.

Genes that have mutations co-occurring with KRAS G12C in the Mayo cohorts. CRC = colorectal cancer, PDAC = pancreatic ductal adenocarcinoma

251 Background: The development of colorectal liver metastases (CRLM) is associated with poor prognosis, and recent data suggest that metastasis to the liver is associated with resistance to immunotherapy. We characterized the microenvironment of primary colorectal carcinomas (CRCs) relative to their synchronous CRLM using a validated segmentation algorithm that quantifies 15 distinct morphologic tumor features. Methods: Adult CRC patients with synchronous CRLM (N=57) at Mayo Clinic were identified from the electronic health record using Epic Slicer Dicer and from an internal database. Tumor H&E sections were digitized and reviewed for quality control (RKP). QuantCRC (Aiforia, Inc) was applied to digitized images to extract 15 GI pathologist pre-defined morphological features. Tumor features were compared between primaries and CRLM using the Kruskal–Wallis test. The project was approved by the Mayo Clinic Institutional Review Board. Results: The study included 57 patients (median age 59 years [IQR 50, 73], 51% female) with CRC primaries and synchronous CRLM. Among primaries, 20 (35%) were right-sided and 37 (65%) were left-sided. QuantCRC identified 6 of 15 morphological features that differed significantly between primaries and their CRLM, including reduced stroma, more high-grade and necrosis, and a higher tumor: stromal ratio (TSR) in CRLM (Table). The increase in TIL density in CRLM vs the primary tumor was of borderline significance (p =0.053). Among patients with left-sided primary tumors, their CRLM had significantly higher TSR, percent high-grade, percent necrosis, and TIL density compared to the primary (all p values ≤ 0.02). Among right sided primaries, CRLM had a significantly reduced percent mature stroma (p=0.034) whereas percent necrosis was increased (p =0.01). Conclusions: Using deep learning, we identified tumor morphological features that differed significantly between primary CRC and their synchronous CRLM. This included higher TSR in CRLMs compared to primaries that is associated with epithelial-mesenchymal transition and has been shown to contribute to treatment resistance. Analysis of tumor morphological features with patient prognosis is ongoing. Deep learning-derived morphological features of primary CRC and CRLM. Morphological feature (Median) Primary tumor Liver metastasis P-value Tumor-Stroma Ratio 0.78 1.50 0.008 % High-grade 23.08 31.72 0.005 TIL Count 44.36 60.20 0.053 % Necrosis 6.35 19.32 <0.001 %Signet Ring Cells 0.25 0.13 0.088 TB/PDC 0.89 1.13 0.484 % Stroma 54.89 40.98 <0.001 % Immature Stroma 46.76 34.13 <0.001 % Mature Stroma 6.27 5.08 0.030