Mutation-associated neoantigens (MANAs) are highly cancer-specific targets for immunotherapy where peptides derived from intracellular mutant proteins are presented on the cell surface via HLA molecules. T cell-engaging bispecific antibodies and CAR T cells can target MANAs to eliminate cancer cells via T cell activation. However, the low antigen density of MANAs on the cell surface can limit therapeutic efficacy. Here, we investigated whether increasing the affinity of the H2 single-chain variable fragment (scFv) targeting the p53 R175H MANA (HMTEVVRHC presented on HLA-A*02:01) improves its therapeutic effect. We identified higher-affinity H2 variants via phage biopanning and a thiocyanate elution method. Increasing bispecific antibody affinity to the low nanomolar range increased cancer cell killing and tumor control in mouse xenograft models without sacrificing antigen specificity. We next asked how increasing scFv affinity impacts CAR T cell function - a matter of debate. We appended each variant scFv to a CD28z CAR, CD3γ, or the T cell receptor. In striking contrast to the bispecific antibody results, increasing CAR affinity decreased function in each CAR format due to lower T cell activation upon interaction with target cancer cells. These results have important implications for the design of future immunotherapeutic approaches targeting low-density antigens.

Lethal toxins could become potent therapies against cancer, but their clinical utility is limited by adverse events upon systemic administration. These could be reduced if the toxins were delivered by effector cells that specifically infiltrate cancers, thereby releasing toxins locally into the tumor microenvironment. One of the challenges underlying this strategy is that cells delivering toxins would have to be resistant to them. We address this obstacle by showing that effectors derived from transformed human cell lines genetically engineered for resistance to bacterial adenosine diphos-phate ribosylating toxins (ADPRTs), including Pseudomonas aeruginosa exotoxin A (PE), can produce targeted immunotoxins that specifically kill cancer cells express-ing cognate tumor-associated antigens. Resistance to immunotoxins was achieved by knockout of genes in the diphthamide biosynthesis pathway (DPH1-4) required for the posttranslational modification of eukaryotic elongation factor 2 (EEF2) that is the target of ADPRTs or by mutation of EEF2 itself. We show that engineering resist-ance to ADPRTs, one of the most potent toxins acting on human cells, is essential to achieve robust function of armored effector cell lines. This work establishes a critical step on the path to equip effector cells with the ability to deliver powerful toxins to cancer cells and introduces a platform to investigate extension to primary autologous or allogeneic therapeutic cell types.

Allogeneic hematopoietic cell transplantation is the only curative option for many patients with acute myeloid leukemia (AML). In the current study, we designed and implemented a personalized assay, called v96, incorporating up to 96 mutations in 30 AML patients undergoing transplantation. The assay was performed on DNA derived from cells isolated from the bone marrow as well as in cell-free plasma. All 30 (100%) of patients harbored molecular evidence of residual leukemia during remission that was detectable by the v96 assay, while only 6 (20%) had evidence of disease as assessed by conventional clinical assays. Furthermore, cell-free DNA from plasma proved to be more sensitive than DNA from cells of the bone marrow for identifying residual leukemia. The median number of mutants was 352-fold higher in plasma taken prior to transplantation for patients who relapsed compared to those who did not relapse. At 2 mo posttransplantation, 27 of the 30 patients still harbored detectable leukemia as assessed by the v96 assay. Twenty-two of these patients had a subsequent decrease in leukemic burden assessed by the v96 assay. In the majority of them (20 of 22 patients), the decrease occurred only after immunosuppression was discontinued, supporting a graft-versus-leukemia effect. These results document the feasibility of using a relatively large panel of carefully chosen mutations and a highly specific assay as noninvasive markers of therapeutic response in AML patients, minimizing the need for multiple bone marrow biopsies.

Chimeric antigen receptors (CARs) can induce T cells to kill cancer cells but also to kill normal cells that express the same antigens1. Designing CARs to recognize combinations of antigens, via Boolean logic, can simultaneously expand the scope of targetable antigens and make CAR T cells more specific to cancer2. For example, one antigen may be expressed on cancer cells and normal bone marrow cells, while a second antigen may be present on the same cancer cells but only in normal lungs. If recognition of both antigens is required for T cell activation, only the cancer cells will be killed. Creating such AND-gated CAR T cells has been challenging given the need to engineer non-natural signaling mechanisms that integrate two ligand binding events into a single T cell activation stimulus3-6. Here, we design a fundamentally new AND-gated receptor called M ulti- AN tigen T riggered I mmune S ynapse (MANTIS), which leverages differences in extracellular receptor dimensions to regulate CAR signaling. MANTIS initially prevents CAR activity by steric blocking with a bulky extracellular domain. Upon engagement of the first antigen, MANTIS sheds this blocking domain, releasing a free CAR that can bind a second antigen and activate the T cell in an AND-gated manner. This work demonstrates how differences in extracellular receptor size can be leveraged to spatially regulate intracellular signaling pathways in response to antigen patterns, paving the way for new applications in synthetic biology and cell engineering.

8517 Background: Acquired resistance to targeted therapy remains a major challenge in ALK-positive NSCLC and is often mediated by mutations in the ALK kinase domain. Prophylactic immune targeting of common ALK resistance mutations represents a novel strategy to delay or prevent the emergence of ALK inhibitor resistance. Methods: We conducted a first-in-human phase 1b clinical trial of a prophylactic peptide vaccine (ALK-Vac) in advanced ALK-positive NSCLC patients without progression on standard-of-care tyrosine kinase inhibitor (TKI) therapy. Patients continued their ALK TKI and received ALK-Vac, consisting of synthetic long peptides targeting seven common ALK resistance mutations (I1171T, I1171N, I1171S, L1196M, G1202R, D1203N, E1210K) plus poly-ICLC adjuvant. ALK-Vac was administered subcutaneously on days 1, 4, 8, 15, and 22 (priming) and weeks 12 and 20 (boost). Primary objectives were safety and vaccine-specific T cell responses assessed by IFN-γ ELISpot. Exploratory objectives included molecular and immune-phenotype dynamics assessed by ultrasensitive cell-free DNA (cfDNA) duplex sequencing and CyTOF mass cytometry. Results: Fifteen patients were enrolled and all completed the planned ALK-Vac regimen. Most patients (13/15, 87%) were receiving first-line TKI therapy. Concomitant TKIs included alectinib (7/15, 47%), lorlatinib (5/15, 33%), and brigatinib (3/15, 20%). At enrollment, median TKI duration was 43.7 months (range 4.6-74.2) and 67% of patients had no measurable disease. Treatment-related adverse events (TRAEs) were primarily grade 1 (93% of patients); most commonly injection site reactions (93%), fatigue (60%), and flu-like symptoms (40%). No grade ≥ 3 TRAEs were observed. T cell responses (≥2-fold increase in SFU) were detected in 71% (10/14) of evaluable patients, with a median 11.9-fold increase. Responses were observed to G1202R, L1196M, and D1203N (each 10/14, 71%), E1210K (9/14, 64%), and I1171N/S/T (each 7/14, 50%). With a median follow-up of 11.5 months, the disease control rate was 93% (14/15). One patient who achieved robust immune response against multiple resistance mutations developed oligoprogression on alectinib 8.5 months after starting ALK-Vac; molecular profiling of this lesion identified an emergent KRAS G12D mutation without detectable ALK resistance mutation. Conclusions: ALK-Vac was well-tolerated and induced vaccine-specific T cell responses in a minimal residual disease setting, demonstrating feasibility of prophylactic targeting of ALK resistance mutations as an adjunct to TKI therapy and supporting a broader immune-interception framework potentially applicable to other oncogene-driven NSCLC. Comprehensive cfDNA and immune-phenotyping analyses will be reported. Clinical trial information: NCT05950139 .

We previously reported initial results from a clinical trial testing a strategy in which patients with muscle-invasive bladder cancer (MIBC) achieving a clinical complete response after cystoscopic resection of the bladder tumor plus systemic therapy could forgo removal of their entire bladder (cystectomy). While the results were highly promising, a subset of patients omitting initial cystectomy developed recurrence highlighting the need for biomarkers to refine selection of patients for this approach. We here report long-term follow-up of these patients and investigate whether tumor DNA in the plasma (ctDNA) or urine (utDNA) could inform prognosis and the need for cystectomy. Three-year bladder-intact survival among patients with a complete clinical response following four rounds of systemic therapy was 69%. Metastatic risk was significantly higher for patients with detectable versus undetectable ctDNA presystemic therapy (HR 4.68; 95% CI 1.10-43.35; log-rank P = 0.036). Only 4.5% (1 of 22) of patients with undetectable baseline ctDNA developed metastatic disease. Undetectable ctDNA before or after systemic therapy was associated with extremely low metastatic risk. Urine utDNA was more sensitive than plasma ctDNA at detecting residual disease within the bladder, and detectable urine utDNA in patients with a complete clinical response was associated with shorter bladder-intact survival (HR 6.47, 95% CI 1.34-31.31; log-rank P = 0.008). These findings establish the conceptual and experimental foundation for incorporating ctDNA and utDNA assays into the management of patients with MIBC, particularly with respect to the need for cystectomy.

Self-renewing normal tissues generate several somatic mutations at each division. Previous studies have reported that cancer cells have more mutations than their normal counterparts. It is not obvious why dramatic differences in mutation burdens between normal tissues and cancers should exist. To fully understand human tumorigenesis, the increase of mutation burden in cancers will have to be understood. Here, we provided a systematic comparison of mutational burdens in normal and cancer cells from five different organs, revealing a four-fold increase of mutation burdens in cancerous vs. non-cancerous cells. Three proposed hypotheses that could account for the increased mutation burdens in cancer are: the classical hypothesis, where driver gene mutations explain the higher mutational burden; the catastrophic hypothesis, where extreme mutational events lead to large-scale genomic alterations; and the tail hypothesis, where differences in baseline mutation rates among individuals account for the differences. Testing through orthogonal observations showed that the observed medians and distributions of mutation burdens in cancers could be explained by the hypotheses to various degrees of significance, and only the tail hypothesis could easily explain the increase in median mutation burdens in the normal tissues of cancer patients compared to the normal tissues of non-cancer patients. Overall, this study characterizes an increased mutation burden across multiple types of cancer compared to normal tissue and provides insights into the contributing factors. A tenable hypothesis proposed in this study involving fundamental differences in baseline mutation rates among individuals could have implications for cancer prevention strategies.

4101 Background: Gastric and gastroesophageal cancers (G/GEC) commonly spread to the peritoneum. Accurate peritoneal staging is critical for guiding intent and therapy, however staging PET, CT and peritoneal cytology have low sensitivity for peritoneal micro-metastases. Whilst circulating tumor DNA (ctDNA) poorly correlates with peritoneal disease, tumor DNA from the peritoneum (ptDNA) may identify early peritoneal metastases missed by conventional staging. We developed a tumor-informed whole-genome sequencing (WGS) platform to detect ptDNA and evaluated its accuracy in diagnosing peritoneal metastases and predicting survival in curative-intent G/GEC. Methods: This Australian multicenter prospective cohort study enrolled patients receiving curative-intent treatment for G/GEC (upfront surgery, perioperative FLOT or neoadjuvant CROSS/nivolumab). Tumor biopsies, blood, and peritoneal lavage fluid were collected at staging laparoscopy, and at surgery for those receiving neoadjuvant therapy. Tumor-informed WGS was used to identify ctDNA and ptDNA, which were correlated with disease free survival (DFS), peritoneal recurrence free survival (pRFS), non-peritoneal recurrence free survival (npRFS), and overall survival (OS) using multivariate Cox regression. The diagnostic utility of ptDNA for peritoneal metastases was evaluated by comparing true-negative patients ( > 2 years disease-free post-surgery alone) with an independently enrolled cohort of true-positive patients (macroscopic/cytology-positive peritoneal disease). Results: ptDNA was detected in 28 (30.4%) of 92 consecutive non-metastatic patients (59 with and 33 without neoadjuvant therapy). ptDNA-positivity significantly correlated with advanced cT and (y)pT stage. At a median follow-up of 17 months, there were 12 (13.0%) peritoneal and 10 (10.9%) non-peritoneal recurrences. ptDNA-positivity independently predicted poorer pRFS (HR 26.74, 95%CI 3.94-181.66), DFS (HR 4.51, 95%CI 1.77-11.47) and OS (HR 6.65, 95%CI 1.72-25.76), but not npRFS (HR 0.87, 95%CI 0.16-4.78). Importantly, 9 (60%) patients converted from ptDNA-positive to ptDNA-negative post-neoadjuvant treatment. Patients who remained ptDNA-positive had significantly worse pRFS (3/6 recurrences, HR 45.4, 95%CI 3.5-595.2) than those who were ptDNA-negative (0/25 recurrences). The diagnostic accuracy of ptDNA was validated in 10 true negative and 12 true positive patients. In this cohort, ptDNA achieved 100% sensitivity, specificity, positive and negative predictive values (95%CI 75.8-100%) for detecting peritoneal metastases. ctDNA data will be presented at the meeting. Conclusions: Tumor-informed WGS-based ptDNA accurately diagnoses peritoneal disease and predicts peritoneal recurrence post curative-intent treatment, informing its potential to personalize clinical management using intraperitoneal therapies.

Adjuvant chemotherapy in stage III colon cancer provides uncertain benefit at the individual level. Circulating tumor DNA (ctDNA) may help refine risk-adjusted treatment selection. In this multicenter, randomized, phase 2/3 trial, patients with stage III colon cancer underwent ctDNA testing 5-6 weeks after surgery and were assigned (1:1) to ctDNA-guided or standard management. In the ctDNA-guided arm, patients negative for ctDNA received de-escalated therapy, whereas ctDNA-positive patients received escalated therapy. Clinicians prespecified the standard regimen. Primary endpoints were 3-year recurrence-free survival (RFS) for ctDNA-negative patients and 2-year RFS for ctDNA-positive patients. Secondary endpoints included treatment-related hospitalization and ctDNA clearance. Among 968 evaluable patients, 702 (72.5%) were ctDNA negative. With a median follow-up of 47 months, ctDNA-negative patients experienced significantly fewer recurrences than ctDNA-positive patients (3-year RFS 87% versus 49%; P < 0.001). In ctDNA-negative patients, de-escalation reduced oxaliplatin use (34.8% versus 88.6%) and hospitalizations (8.5% versus 13.2%) but yielded slightly lower RFS than standard management (85.3% versus 88.1%), not meeting the non-inferiority margin. In ctDNA-positive patients, higher ctDNA burden correlated with recurrence risk (3-year RFS 77% to 23% across quartiles; P < 0.001). Escalated therapy did not improve outcomes over standard management (2-year RFS 51% versus 61%). There was no unexpected toxicity. Persistent ctDNA after treatment predicted markedly worse prognosis (3-year RFS 14% versus 79%). ctDNA is validated as a strong prognostic classifier. ctDNA-guided de-escalation reduced oxaliplatin exposure and adverse events with outcomes approaching standard of care, whereas exploratory chemotherapy intensification conferred no RFS benefit, suggesting a need for novel strategies in ctDNA-positive disease.Australian New Zealand Clinical Trials Registry Identifier: ACTRN12617001566325 .

Supplemental Figure 8. The relative contribution of cfDNA from various tissues before and ~24 hours after surgery for pancreatic cancer. * p < 0.05, ** p < 0.01, *** p < 0.001

AI is now a cornerstone of modern dataset analysis. In many real world applications, practitioners are concerned with controlling specific kinds of errors, rather than minimizing the overall number of errors. For example, biomedical screening assays may primarily be concerned with mitigating the number of false positives rather than false negatives. Quantifying uncertainty in AI-based predictions, and in particular those controlling specific kinds of errors, remains theoretically and practically challenging. We develop a strategy called multidimensional informed generalized hypothesis testing (MIGHT) which we prove accurately quantifies uncertainty and confidence given sufficient data, and concomitantly controls for particular error types. Our key insight was that it is possible to integrate canonical cross-validation and parametric calibration procedures within a nonparametric ensemble method. Simulations demonstrate that while typical AI based-approaches cannot be trusted to obtain the truth, MIGHT can be. We apply MIGHT to answer an open question in liquid biopsies using circulating cell-free DNA (ccfDNA) in individuals with or without cancer: Which biomarkers, or combinations thereof, can we trust? Performance estimates produced by MIGHT on ccfDNA data have coefficients of variation that are often orders of magnitude lower than other state of the art algorithms such as support vector machines, random forests, and Transformers, while often also achieving higher sensitivity. We find that combinations of variable sets often decrease rather than increase sensitivity over the optimal single variable set because some variable sets add more noise than signal. This work demonstrates the importance of quantifying uncertainty and confidence-with theoretical guarantees-for the interpretation of real-world data.

Supplemental Figure 1. Overview of the patient samples included in the present study.

Supplemental Figure 10. In silico mixing experiments (N = 10) of buffy coat bisulfite sequencing data with liver (A), lung (B), colon epithelial cell (C), and left atrium (D) bisulfite sequencing data deconvoluted using the Moss et al. (43) reference matrix and quadratic programming shows excellent agreement between predicted and actual fractional contribution.

Supplemental Figure 4. Methylation profiles using quadratic programming vs. non-negative least- squares regression using the reference matrix described in Sun et al. (3). Pearson’s correlation coefficient and p values are presented at the bottom of this figure, showing the derived contributions from each of the 12 tissue types that could be assessed.

Patients with radiographically detectable lesions in their brain or other symptoms compatible with brain tumors pose challenges for diagnosis. The only definitive way to diagnose such patients is through brain biopsy, an invasive and dangerous procedure. In this study, we present a new workflow termed "CSF-BAM" that simultaneously identifies B-cell or T-cell receptor sequences, aneuploidy, and mutations using amplification of both strands of the DNA from cerebrospinal fluid (CSF) samples. We applied CSF-BAM to a validation set of 209 samples from patients with brain cancers. Among the 129 samples from patients with the most common aggressive cancer types, the sensitivity of detection was 81%. None of 30 CSF-BAM assays were positive in CSF samples from patients without brain cancers (100% specificity). CSF-BAM provides an integrated approach to identify neoplasia in the central nervous system, provides information about the genetics and immune environment, and has the potential to inform patient management. SIGNIFICANCE:There is a paucity of technologies beyond surgical biopsy that can accurately diagnose central nervous system neoplasms. We developed a novel, sensitive, and highly specific assay that can detect brain cancers by comprehensively identifying somatic mutations, chromosomal copy-number changes, and adaptive immunoreceptor repertoires from samples of CSF. See related commentary by Weiss, p. 1976.

Supplementary Table S1: Cohorts for CSF-BAM components. Supplementary Table S2A: SafeBSeqS primer sequences. Supplementary Table S2B: SafeTseqS primer sequences. Supplementary Table S2C: Aneuploidy analysis primer sequences. Supplementary Table S2D: Mutation analysis primer sequences. Genomic coordinates refer to hg19. Supplementary Table S3: Summary of SafeBSeqS analysis in WBC control samples. Supplementary Table S4: Summary of CSF-BAM SafeBSeqS analysis in CSF. Supplementary Table S5: Summary of SafeTSeqS analysis in WBC control samples. Supplementary Table S6: Summary of CSF-BAM SafeTSeqS analysis in CSF. Supplementary Table S7: SafeBSeqS and SafeTSeqS analysis for CSF validation set 2. Supplementary Table S8: Summary of aneuploidy analysis in control samples. Supplementary Table S9: Summary of CSF-BAM aneuploidy analysis in CSF samples. Supplementary Table S10: Reproducibility of aneuploidy analysis. Supplementary Table S11: Comparison of aneuploidy analysis with CSF-BAM and Real-CSF. Supplementary Table S12: CSF-BAM mutation training set in non-cancer CSF samples. Supplementary Table S13: CSF-BAM mutation analysis in CSF samples. Supplementary Table S14: Demographic characteristics by sample. Supplementary Table S15: Diagnostic categories by sample. Supplementary Table S16: CSF-BAM results summary. Supplementary Table S17: CSF-BAM detection based on CSF reservoir abutment by sample. All samples with available clinical data are included. P-values were calculated by Fisher’s exact test. Supplementary Table S18A: TCR Clusters obtained via CSF-BAM. Supplementary Table S18B: TCR specificity annotation. Supplementary Table S19: CSF samples from patients with multiple sclerosis analyzed with SafeTSeqS and SafeBSeqS.

3503 Background: Despite adjuvant chemotherapy (ACT) a proportion of patients (pts) with stage III colon cancer (CC) will recur. Most at risk are those with detectable ctDNA, whereas those with undetectable ctDNA have a reduced recurrence risk. The DYNAMIC-III study explored the impact of ACT de-escalation or escalation as informed by post-surgery ctDNA results. Here, we report the primary analysis on the impact of treatment escalation in ctDNA-positive pts. Outcome data for treatment de-escalation in ctDNA-negative pts is immature. Methods: DYNAMIC-III is a multi-center, randomized, phase II/III trial. Eligible pts had resected stage III CC and were fit for ACT. Pts were randomly assigned 1:1 to ctDNA-informed or standard of care (SOC) management. Clinicians nominated the selected SOC ACT regimen prior to randomization. For ctDNA-informed management, a ctDNA-positive result at 5-6 weeks after surgery with a tumor-informed assay prompted an escalation ACT strategy (from single agent fluoropyrimidine [FP] to oxaliplatin-based doublet, from 3 months doublet to 6 months doublet or FOLFOXIRI [clinician choice], or from 6 months doublet to FOLFOXIRI). The primary efficacy endpoint for the ctDNA-positive cohort was 2-year RFS. The target sample size of 250 provided 80% power with 90% confidence to confirm superiority of ctDNA-informed treatment escalation compared to SOC with a HR of 0.746. Results: Of 961 eligible pts randomized between Oct 2017 and Apr 2023, 259 (27%) were ctDNA-positive. Of these, 113 (44%) had clinical low risk disease (non-N2 + non-T4). Median follow-up was 42.2 months (range 0.78 – 63.0). 115 (89%) of 129 ctDNA-informed pts received ACT escalation, with 65 (56%) receiving FOLFOXIRI. Of 130 SOC pts, 14 (11%) and 112 (86%) received single agent FP and oxaliplatin doublet, respectively. 2-year RFS for ctDNA-informed treatment escalation was 52% (90% CI: 44 - 59%) vs 61% (90% CI: 54 - 68%) for SOC (HR 1.11, 90% CI: 0.83 - 1.48; P = 0.6). The 3-year RFS for ctDNA-positive pts receiving FOLFOXIRI and FOLFOX/CAPOX was similar (47% vs 51%, HR 1.09, 90% CI 0.78 to 1.53; P = 0.7). In a pre-specified correlative analysis of all ctDNA positive pts, recurrence risk increased with ctDNA burden, with 3-year RFS of 78%, 63%, 36% and 22% for tumor-derived mutant molecules/mL quartiles < 0.06, 0.06 – 0.17, 0.18 – 1.31, and > 1.31, respectively (P < 0.01). Treatment-related hospitalisation was similar for escalated and SOC pts (OR 1.21, P = 0.58). Analysis of post-ACT ctDNA is underway. Conclusions: In this first randomised study of ctDNA-informed management in stage III CC, we confirm the prognostic significance of detectable ctDNA, with the novel finding of recurrence risk increasing markedly with ctDNA burden. Treatment escalation, including to FOLFOXIRI, did not improve RFS. Future studies in ctDNA positive pts should explore other escalation strategies. Clinical trial information: ACTRN12617001566325 .