Early functional reactivity to a CEACAM5 x CD3 T cell bispecific antibody in solid tumors associates with baseline immune contexture rather than target antigen density alone

Integrated translational framework for evaluating CEACAM5 x CD3 TCB pharmacodynamics

An integrated experimental framework was established to evaluate the biological determinants of CEACAM5 x CD3 TCB-induced PD activity. The mechanism of action of CEACAM5 x CD3 TCB, including tumor cell engagement, T cell activation, cytokine release and tumor cell killing, is schematized in Fig. 1A. By coupling humanized murine models with a patient-derived EVOC platform and Phase 1 clinical trial samples, we aim to resolve how the native human TME influences early drug-induced pharmacodynamics (Fig. 1B).

CEACAM5 x CD3 TCB induces tumor control and intratumoral immune activation in humanized mice

The activity of CEACAM5 x CD3 TCB was first evaluated in humanized NOD scid gamma (NSG) mice bearing established CEA-positive MKN-45 gastric adenocarcinoma tumors. Weekly intravenous (i.v.) dosing at 0.5 mg/kg resulted in significantly lower tumor volume compared to the vehicle (histidine buffer)-treated animals (p = 0.0093) with a tumor growth inhibition of 57% (Fig. 2A).

Intratumoral PD was assessed 24 h after the second dose to maximize detection of T cell infiltration. CEACAM5 x CD3 TCB treatment significantly increased intratumoral levels of multiple cytokines in treated mice, including IL-2, TNFα, IL-6, IL-8 and IL-10, with IFNγ showing a numerical increase as compared to the vehicle control (Fig. 2B). The distribution of cytokine data, assessed by the coefficient of variation (CV), was 52% for IFNγ with a range of 30–64% across cytokines (Fig. 2B and Supplementary Table S1). These intratumoral cytokine changes were accompanied by increased infiltration of intratumoral CD3⁺, CD4⁺ and CD8⁺ T cells relative to vehicle control-treated animals (Fig. 2C), confirming on-target immune activation in vivo.

Patient-derived tumor explants preserve native TME architecture, CEA expression and immune cell infiltration

To assess CEACAM5 x CD3 TCB activity within a preserved human TME morphology, tumor specimens from 14 patients with CRC, PDAC, and NSCLC were cultured ex vivo using an EVOC model previously developed by Gavert et al.¹⁷. Baseline clinicopathologic features of patient tumors, including tumor indication, histologic subtype, intratumoral CD3⁺ T cell density and CEA expression assessed by H-score, are summarized in Table 1. Histologic assessment confirmed that the patient's tumor native intratumoral stromal composition, architecture, and endogenous immune contexture were preserved in EVOC at baseline (Fig. 3A). Consistent with prior reports for CRC, PDAC and NSCLC^18,19, all tumors were CEA positive, with CRC demonstrating the strongest and most homogeneous expression, followed by PDAC and NSCLC. NSCLC tumors tended to have higher CD3 T cell infiltration as compared to PDAC or CRC, both of which exhibited lower CD3 T cell infiltration, although not statistically significant²⁰. Both CEA expression and baseline CD3⁺ T cell infiltration exhibited substantial inter-patient variability (Fig. 3B, C and Table 1; CEA H-score range 10 to 285; CD3% range 0.2–22.7). Short-term culture in EVOC up to 72 h maintained tissue viability, proliferative capacity and stable CEA expression (Supplementary Fig. S1).

CEACAM5 x CD3 TCB-mediated cytokine responses ex vivo tend to be higher in tumors with higher T cell infiltration and do not associate with baseline CEA expression

Exposure of patient-derived tumor tissue to CEACAM5 x CD3 TCB ex vivo for 48 h induced robust cytokine responses compared with an untargeted control (DP47 × CD3 TCB). Significant increases were observed for IFNγ, IL-2, IL-10 and TNFα, while Granzyme B showed a modest, non-significant increase (Fig. 4A). A substantial level of inter-patient heterogeneity was observed in cytokine response with CVs ranging from 78% to 135% across cytokines (Fig. 4A and Supplementary Table S1). Although not statistically significant, higher baseline CD3⁺ T cell infiltration trended with a greater magnitude of cytokine release in EVOC samples as compared to samples with lower CD3+ T cell infiltration, with a high degree of inter-patient heterogeneity (Fig. 4B). In contrast, baseline CEA expression did not associate with functional responses to CEACAM5 x CD3 TCB ex vivo (Supplementary Fig. S2).

Based on preclinical data, we hypothesized that adding FAP-4-1BBL to CEACAM5 x CD3 TCB would provide a potent, costimulatory signal (“Signal 2”) that enhances T cell activation and, consequently, functional reactivity of the TCB. However, the addition of a FAP-4-1BBL bispecific antibody did not enhance cytokine induction relative to CEACAM5 x CD3 TCB monotherapy ex vivo (Fig. 4C).

Baseline tumor characteristics and pharmacokinetics in Phase 1 patients

We benchmarked the in vivo and ex vivo findings with translational biomarker data from 18 CRC, PDAC, and NSCLC patients enrolled in a Phase 1 dose-escalation study of CEACAM5 x CD3 TCB (ClinicalTrials.gov identifier NCT03539484). Baseline clinical, tumor and circulating biomarker characteristics, including dose level, tumor immune phenotype, intratumoral CD8⁺ T-cell density, tumor CEA expression and baseline soluble CEA (sCEA) concentrations, are detailed in Table 2. Prior to treatment, patient tumors exhibited high CEA expression with H-scores >100 across nearly all patients evaluated (Fig. 5A), and diverse tumor immune phenotypes (Fig. 5B). Classification of tumors into immune-desert, excluded or inflamed phenotypes was based on intratumoral CD8⁺ T cell density (Supplementary Fig. S3). Similar to the patient tumors evaluated in EVOC, 50% of baseline tumors from patients in the Phase 1 study had high CEA expression (H-score >200 for 8 of 17 patient tumors from the Phase 1 study, 6 of 11 patient tumors evaluated in EVOC). Variable levels of baseline sCEA were observed across the patient cohort (Fig. 5C). Bulk RNA sequencing of patient tumor biopsies at baseline demonstrated that immune-inflamed and excluded tumors were characterized by higher baseline expression of IFNγ–responsive, effector T cell signatures and antigen presentation pathways as compared to immune-desert tumors (Fig. 5D).

Pharmacokinetic (PK) analyses demonstrated that patients with medium (71–100 ng/mL) or high (>100 ng/mL) baseline sCEA—that could sequester drug, acting as a decoy or “sink”—had reduced active drug exposure at lower dose levels, while adequate exposure was maintained in patients with low sCEA (<70 ng/mL) or at higher doses (Supplementary Fig. S4).

Systemic cytokine induction reflects baseline tumor immune contexture and drug exposure in Phase 1 patients

In the small cohort of 18 patients evaluated from the Phase 1 study, peripheral IFNγ concentrations increased rapidly following CEACAM5 x CD3 TCB administration and were most pronounced in patients with low baseline sCEA following the first dose (Supplementary Fig. S5). At 24 h post-dose, significant increases were observed in peripheral levels of IFNγ, TNFα, IL-2SR, IL-10, IL-6, and IL-8, compared with pre-treatment levels (Fig. 6A and Supplementary Fig. S5). Similar to ex vivo findings, a substantial level of inter-patient heterogeneity was observed in cytokine responses at 24 h post-dose (CVs ranging from 79 to 190% across cytokines; Supplementary Table 1).

Stratification by baseline tumor immune phenotype revealed significantly higher peripheral cytokine induction (IL2SR, IL-6, and TNFɑ) in patients with immune-inflamed or excluded tumors compared to patients with immune desert tumors (Fig. 6B). A similar trend was observed for IFNγ although not statistically significant. These clinical observations, including the high inter-patient heterogeneity in cytokine response, were consistent with the trends observed in EVOC. In contrast, baseline CEA expression or peripheral CD3+ T cell abundance did not correlate with systemic cytokine induction (Supplementary Figs. S6 and S7), consistent with ex vivo observations. It should be noted that no significant clinical responses were observed in any patients on trial (Fig. 5D).

No significant changes in intratumoral gene signatures and proteins on-treatment in Phase 1 patients

To evaluate the effects of CEACAM5 x CD3 TCB treatment in situ, paired tumor biopsies obtained prior to treatment and at Cycle 2 Day 8 (pre-dose) were evaluated. No significant changes were observed in TME immune or non-immune pathways based on transcriptional analysis or immunohistochemistry, including cytotoxic T cell signatures, IFNγ pathway activity or CEA expression (Supplementary Figs. S8, S9, S10 and Table S2).

Study design and rationale

An integrated translational framework combining humanized mouse studies, patient-derived ex vivo tumor explants, and clinical biomarker analyses was used to characterize early pharmacodynamic activity of CEACAM5-directed CD3 T cell bispecific antibodies (CEACAM5 x CD3 TCB) (Fig. 1B).

Humanized mouse studies

The humanized animal model has been previously described³². All animal experiments were conducted in accordance with institutional and governmental regulations and were approved by the Veterinary Office of the Canton of Zurich (license ZH193/2014), the institutional ethics and licensing authority overseeing this work. Mice were housed under specific-pathogen-free conditions with controlled temperature, humidity, and a 12-h light/dark cycle, and had ad libitum access to food and water. Mice were treated with drugs or controls intravenously in conscious, gently restrained manner without anesthesia, consistent with standard practice. Prior to injection, tails were prewarmed with a heat lamp to induce vasodilation and disinfected with 70% ethanol. Mice were humanely euthanized at the study endpoint, or earlier if signs of sickness or suffering were observed, prior to tumor harvest for flow cytometric and cytokine/chemokine analysis. Euthanasia was performed via isoflurane anesthesia followed by cervical dislocation and confirmatory exsanguination, in accordance with AVMA guidelines and institutional standard operating procedures. Mice were implanted subcutaneously with MKN45 human gastric carcinoma cells and randomized to treatment groups once tumors reached approximately 150–200 mm³. CEACAM5 x CD3 TCB or vehicle control (sterile histidine buffer) was administered intravenously once weekly for six consecutive weeks. Histidine buffer was selected as the control to allow for cross-comparability between multiple therapeutic modalities while adhering to animal conservation guidelines. Tumor growth was monitored by caliper measurements, and animals were assessed daily for clinical signs and adverse events.

For intratumoral PD assessment, a subset of mice (n = 4 per group) was sacrificed at 24 h post-second dose for flow cytometric analysis and for cytokine and chemokine analysis. The cohort size was selected based on the expected high consistency of PD data in the humanized mouse model. Results were compared against vehicle-treated controls to determine treatment-induced changes.

Tumor dissociation and single-cell preparation

Subcutaneous tumors of four mice per treatment condition were harvested from humanized mice and weighed immediately following extraction before immersion in sterile ice-cold phosphate-buffered saline (PBS). Tissue dissociation was performed using the gentleMACS™ system (Miltenyi Biotec) according to a modified manufacturer’s protocol. Briefly, tumors were transferred into gentleMACS C-tubes containing 5 ml RPMI-1640 medium supplemented with Collagenase D (1 mg/ml; Roche Diagnostics) and DNAse I (0.025 mg/ml; Sigma-Aldrich). Samples were processed using the gentleMACS program m_imptumor_01_01, followed by a 30-min incubation at 37 °C with mechanical mixing every 10 min. A second dissociation step was performed using the program m_imptumor_02_01. The resulting cell suspension was filtered through a 70 µm nylon cell strainer (BD Falcon), washed twice, and resuspended in FACS buffer (DPBS without Ca²⁺/Mg²⁺, supplemented with 2% FCS and 2 mM EDTA).

Flow cytometry

Single-cell suspensions were transferred to 96-well plates (approximately 100 µl per sample) and centrifuged at 1450 rpm for 5 min at 4 °C. The supernatant was discarded, and cells were resuspended in 100 µl of an antibody cocktail prepared in FACS buffer. To prevent non-specific binding, the master mix contained a purified rat anti-mouse CD16/CD32 Fc-block (clone 2.4G2; BD Biosciences, 553142) alongside the following human-specific fluorophore-conjugated antibodies: AlexaFluor 488-anti-CD45 (clone HI30), BV605-anti-CD3 (clone OKT3), APC-Cy7-anti-CD4 (clone OKT4), and AlexaFluor 700-anti-CD8 (clone SK1); all antibodies were sourced from BioLegend/Lucerna-Chem.

Samples were incubated in the dark for 30 min at 4 °C. Following incubation, cells were washed with 200 µl FACS buffer and stained with DAPI (4’,6-diamidino-2-phenylindole; Roche, Cat. no. 10236276001) for dead cell exclusion according to the manufacturer’s instructions. Data were acquired using a BD LSRFortessa™ cell analyzer (BD Biosciences). Manual gating and analysis were performed using FlowJo software v.10.4 (FlowJo, LLC). Tumor-infiltrating immune cells were identified as living (DAPI^-) CD45⁺ cells, with further gating to define CD8⁺ and CD4⁺ T cell subpopulations. For the quantification of intratumoral T cells, following staining and flow cytometric acquisition, raw flow cytometry counts were adjusted by the dilution factors applied. These counts were normalized to the initial tissue weight to express infiltration density as cells per milligram (mg) of tumor tissue to account for variations in tumor size at the time of harvest, enabling standardized comparison across samples.

Cytokine and chemokine analysis

Small tumor fragments were snap-frozen in liquid nitrogen and stored at −80 °C. Whole protein was isolated by homogenizing tissues in Bio-Plex® Cell Lysis Buffer supplemented with EASYpack Protease Inhibitor Cocktail (Roche) using a Precellys®24 Homogenizer. Protein concentrations were determined using the BCA™ Protein Assay Kit (Thermo Scientific). Quantitative analysis of 40 human cytokines and chemokines was performed using the Bio-Plex Pro™ Human Chemokine Panel (Bio-Rad Laboratories) on a Bio-Plex system. Final cytokine concentrations were normalized to the total protein content and are expressed as pg/mg of protein.

Patient samples and ex vivo tumor explant culture

Fresh tumor tissue was obtained from surgical resections of consenting patients with CRC, PDAC or NSCLC (N = 14; Table 1). Tumor specimens were transported in preservation medium and processed within 2 to 6 h of resection.

Tumor samples were sectioned into approximately 300 µm thick slices using a vibratome and cultured as organotypic tumor explants under conditions that preserve tissue architecture and immune cell viability, as previously described. Two to three explant slices randomly assigned per condition were cultured for up to 72 h.

Explants were treated with a CEACAM5 x CD3 TCB either alone or in combination with a FAP 4-1BBL costimulatory agent. Control conditions included an untargeted TCB (DP47 × CD3 TCB) or medium alone. Culture supernatants were collected at 48 h and stored at −80 °C for cytokine analysis. Corresponding tissue slices were fixed in formalin for histological and immunohistochemical evaluation.

Biomarker analysis of ex vivo explants

Formalin-fixed paraffin-embedded (FFPE) tumor explants were serially sectioned and stained with hematoxylin and eosin (H&E) to assess tissue morphology. Immunohistochemistry (IHC) was performed using antibodies directed against CD3 to quantify T cell infiltration and CEA to assess antigen expression. Whole-slide digital scanning and quantitative image analysis were used to determine CD3 T cell quantities using QuPath software. In short, tissue sections were analyzed using the positive cell detection function. Cells were first identified based on nuclear detection parameters, after which CD3 staining intensity was assessed, and cells were classified as positive or negative according to predefined thresholds. The percentage of CD3-positive cells was calculated as the number of CD3-positive cells divided by the total number of detected cells within the analyzed region of interest. CEA expression was evaluated using a semiquantitative H-score. All analyses were performed using identical detection and classification parameters across samples to ensure consistency and comparability. Cytokine concentrations in explant culture supernatants were quantified using the ProcartaPlex assays (Thermo Fisher Scientific) and read on a Luminex instrument according to manufacturer instructions. Analytes included interferon gamma, interleukin-2 receptor alpha, tumor necrosis factor alpha, interleukin-10, interleukin-6, interleukin-8, and granzyme B. Due to limitations in tissue and sample availability, not all 14 tumor samples were subject to the full set of IHC and cytokine-based biomarker analysis.

Clinical trial patient samples

Clinical biomarker analyses were performed using samples obtained from patients enrolled in a first-in-human, open-label, multicenter, dose-escalation Phase 1 clinical study of CEACAM5 x CD3 TCB (ClinicalTrials.gov identifier NCT03539484, registration date 29/05/2018). The study was approved by Videnskabsetiske Komitéer for Region Hovedstade, Hillerød, Denmark; CEIC Hospital Vall D’Hebron, Barcelona, Spain; and Independent Belgian Ethics Committee, Leuven, Belgium, as well as local ethics and governance committees. The study was conducted in accordance with the Declaration of Helsinki, the International Council for Harmonization Good Clinical Practice guidelines, and all applicable local and national regulations. Institutional review boards at each study site approved the study, and all patients provided written informed consent.

Patients received intravenous infusions of the drug every three weeks, at dose levels ranging from 400 µg to 1800 µg (Table 2). Peripheral blood samples included in this study were collected prior to treatment and longitudinally for pharmacokinetic and pharmacodynamic biomarker analyses at 2, 24, 48 and 168 h following the first dose. Fresh paired tumor biopsies were obtained prior to treatment and on treatment at Cycle 2 Day 8 (pre-dose) for characterization of the TME and were formalin fixed and paraffin embedded.

Clinical biomarker analyses

FFPE tissue blocks were obtained and sectioned consecutively for H&E and IHC, which were performed at Discovery Life Sciences (Kassel, Germany) and CellCarta (Antwerp, Belgium). Samples passing quality control for sufficient tumor content without necrotic areas were included in the analysis. Tumor biopsies were assessed by immunohistochemistry for CD8-positive T cells and CEA expression. CD8⁺ T-cells were scored visually using a validated multiplex PanCK-CD8 assay (Clone SP239 and Clone AE1/AE3/PCK26). This dual staining approach facilitated the classification of tumors based on the spatial location and density of CD8⁺ T cells within tumor intraepithelial (IE) and intratumoral stroma (ITS) compartments. Using this approach, samples were categorized into three phenotypes: Inflamed (IE2 + IE3 ≥ 20%), Excluded (ITS2 + ITS3 > 20% and IE0 + IE1 ≥ 80%), or Desert (IE0 + IE1 ≥ 80% and ITS0 + ITS1 ≥ 80%)²⁰ (Fig. S4). Scoring was performed by CellCarta pathologists. This approach has been extensively validated for use in biopsy samples. CEA expression was evaluated using Cell Marque CEA31 monoclonal antibody. Staining was evaluated by a pathologist to determine the percentage of positive tumor cells at varying intensities (1+, 2+, 3+), which were integrated into a weighted H-score (range 0–300) calculated as (% cells at 1+) × 1 + (% cells at 2+) × 2 + (% cells at 3+) × 3. Tumor tissue sections with at least 1+ tumor cells with a level of 1 + CEA staining were considered CEA positive.

Bulk RNA sequencing was performed on paired baseline and on-treatment tumor biopsies to assess immune-related transcriptional programs. FFPE tissue was macrodissected using H&E as a guide to enrich tumor content. RNA was extracted using the High Pure FFPE RNA isolation kit (Roche), assessed by Qubit and Agilent Bioanalyzer for quantity and quality. The prepared libraries were sequenced using the Illumina platforms. RNA-seq data were analyzed as previously described²⁴. Downstream analyses focused on pathway-level immune signatures (refer to Table S1 for gene signature list). Signature scores were estimated via rank-biserial values (BioQC V.3.15) and compared using the Wilcoxon test.

Serum and plasma samples were collected and processed centrally for measurement of soluble CEA and circulating cytokines using ELISA and multiplex ProteinSimple immunoassays on an Ella platform (R&D Systems).

Statistical analysis

All statistical analyses were performed using R software (version 4.2.0). For the humanized mouse study, differences in tumor volume, cytokine concentrations and immune cell densities between treatment groups were evaluated using the Wilcoxon rank-sum test. For the ex vivo explant study, cytokine concentrations were compared between treatment groups using the Wilcoxon signed-rank test; additionally, cytokine responses were assessed by calculating log-transformed treatment-to-control ratios and testing their deviation from zero using the Wilcoxon signed-rank test within CD3 low and CD3 high subgroups. For the phase I study, cytokine concentrations were compared across timepoints, and fold changes at 24h were compared between desert and excluded/inflamed tumors, both using the Wilcoxon signed-rank test. Variability in cytokine response was quantified using the CV, calculated on the original concentration scale. Associations between baseline intratumoral T cell density and cytokine induction were evaluated using Spearman’s rank correlation. All statistical tests were two-sided and a p-value of less than 0.05 was considered statistically significant.