DNA Hypomethylating Agents Preserve T-cell Stemness and Potentiate the Efficacy of CD3-Bispecific Antibodies

The authors report that combination of CD3-bispecifics with epigenetic modifiers can preserve therapeutically relevant T-cell states and enhance immunotherapy responses.

Abstract

Bispecific antibodies targeting tumor-associated antigens and CD3 are promising therapeutic agents for both solid and hematologic cancers. CD3-bispecifics induce T-cell activation and cytotoxicity; however, prolonged T-cell receptor (TCR) stimulation can lead to chromatin rewiring and T-cell dysfunction, thereby limiting their full therapeutic potential. In this study, we investigate the combination of CD3-bispecifics with the DNA hypomethylating agent decitabine (DAC) and observe enhanced tumor growth inhibition in various preclinical models. Utilizing a prostate-specific membrane antigen (PSMA)×CD3 bispecific antibody for the treatment of prostate carcinoma and in vivo humanized mouse disease models, we catalog, at the single-cell level, the dynamics of T-cell epigenetic states during bispecific therapy and in combination with DAC. Importantly, this combination strategy preserves a T-cell factor 1 (TCF-1)–positive T-cell population and delays acquisition of a dysfunctional state at both chromatin and protein levels. At the DNA methylation level, TCR stimulation in the presence of DAC maintains a naïve-like pattern in gene loci associated with T-cell stemness. This study provides a resource for understanding the evolution of T-cell states during immunotherapy and mechanistic support for combining epigenetic modifiers with CD3-bispecifics in the clinic.

Introduction

Recent advances in T cell–based immunotherapies have demonstrated a curative potential for treating cancers that have been refractory to other therapeutic modalities. CD3-bispecific antibodies represent a targeted approach in immunotherapy in which they bind specifically to a tumor surface antigen and the CD3ε chain of the T-cell receptor (TCR), leading to T-cell activation, cytokine release, and target cell killing (1). To date, multiple CD3-bispecific antibodies have been approved for hematologic malignancies (2). However, success in solid tumors with this class of therapeutics has been limited and most patients typically do not achieve durable responses. Chronic exposure to tumor antigens in the cancer microenvironment results in a hypofunctional T-cell state, also known as dysfunction or exhaustion (3–5). Exhausted T (T_ex) cells are characterized by reduced T-cell effector function and proliferation and coexpression of multiple inhibitory markers (such as PD-1, CTLA-4, TIM3, and LAG3) and are associated with decreased efficacy of immunotherapy (6, 7). In addition to continuous stimulation by tumor antigens, prolonged bispecific antibody–mediated TCR stimulation (8, 9) or tonic chimeric antigen receptor (CAR) signaling (10) induces T-cell dysfunction contributing to acquired resistance to immunotherapy. The acquisition of a dysfunctional state is epigenetically encoded and associated with specific transcription factor (TF) and gene expression programs (11–16). Among various epigenetic modifications driving the establishment of T_ex state, de novo DNA methylation by DNA methyltransferase 3A (DNMT3A) results in silencing of genes associated with T-cell proliferative capacity and effector potential. Disruption of DNMT3A activity enhances T-cell reinvigoration after PD-1 blockade as well as CAR T-cell functionality (17, 18). Accumulating evidence suggests that a significant portion of tumor-infiltrating lymphocytes (TIL) may be in an epigenetically inflexible state that cannot be rewired by current immunotherapies, impeding durable responses (4, 13, 14).

The fixed dysfunctional chromatin state induced by chronic TCR stimulation by tumor antigens and/or therapeutics raises the question of which T-cell subpopulations mediate antitumor responses upon immunotherapy. Studies from multiple groups have demonstrated that a T-cell subset expressing T-cell factor 1 (TCF-1) can self-renew, undergo a proliferative burst after immune checkpoint blockade (ICB), and serve as a precursor for terminally exhausted CD8⁺ T cells in tumors (12, 16, 19, 20). A high frequency of this TCF-1⁺ stem cell–like or progenitor exhausted T-cell (T_pex) population is associated with favorable clinical responses (19, 21).

Here, we sought to determine whether disruption of epigenetic rewiring in combination with CD3-bispecific antibodies may improve immunotherapy by promoting favorable T-cell states. Using humanized murine models of prostate and colon cancers, we showed that the combination of decitabine (DAC), a DNA hypomethylating agent, with CD3-bispecific antibodies elicits superior antitumor immunity compared with the bispecific single agent. To gain mechanistic insights into trajectories of T-cell states upon combination of an epigenetic remodeler with a CD3-bispecific, we employed single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq; ref. 22) together with genome-wide methylation profiling. We observed that PSMA×CD3 in combination with DAC results in enrichment of a TCF-1⁺ T-cell population while reducing the frequency of dysfunctional T cells. Our data collectively provide a valuable resource for mapping the evolution of T-cell states during treatment with CD3-bispecific antibodies and serves as a foundation for the rational combination of immunotherapy with epigenetic modifiers to promote favorable immune cell states that boost the efficacy of immunotherapy.

Materials and Methods

Bispecifics

VelocImmune mice (Regeneron Pharmaceuticals, Inc.; refs. 23, 24) were immunized to generate the PSMA×CD3 (patent number: 10,179,819) and CD20×CD3 (patent number: 11,434,300) bispecific antibodies as previously described (2, 25).

Animal studies

All procedures were carried out in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Regeneron Pharmaceuticals Institutional Animal Care and Use Committee, and all animals were maintained under pathogen-free conditions.

Mouse models

Syngeneic tumor studies were performed in mice genetically modified to express human CD3 and FOLH1 (PSMA) or human CD3 and CD20 using the VelociGene technology at Regeneron Pharmaceuticals, Inc. (26–28). The CD20 humanized mouse line was generated by replacing the mouse exons 2 (coding exon 1) to 7 with orthologous human sequences. The cassette containing the vector was electroporated into existing mouse embryonic stem cells of CD3g/3d/3e triple humanization on a B6 background and selected with hygromycin. The cassette was then removed with Cre recombinase in vitro. Cassette-deleted clones were reconfirmed by TaqMan screening, and microinjected Swiss Webster host embryos were surgically implanted into pseudopregnant female mice. Human CD3 (hCD3)/human PSMA (hPSMA) humanized male mice (8–12 weeks) were implanted with 2.5 × 10⁶ TrampC2/hPSMA tumor cells by subcutaneous injection in the right flank. hCD3/human CD20 (hCD20) humanized female mice (8–12 weeks old) were implanted with 0.5 × 10⁶ MC38/hCD20 tumor cells by subcutaneous injection in the right flank. NOD/SCID gamma (NSG; The Jackson Laboratory, strain #005557, RRID: IMSR_JAX:005557) male mice (8–12 weeks) were implanted with 2.5 × 10⁶ TrampC2/hPSMA tumor cells by subcutaneous injection in the right flank. PSMA×CD3 or CD20×CD3 bispecifics were administered by intraperitoneal injection of indicated doses and days after tumor implantation. For experiments with DAC (Meitheal Pharmaceuticals 71288-119-20), mice were first treated with DAC (0.2 mg/kg in PBS) by intraperitoneal injection on days 3 to 7, followed by a combination of DAC (0.2 mg/kg) and PSMA×CD3 (0.5 mg/kg) on days 10, 13, and 17 or combination with CD20×CD3 on days 10, 13, 17, 20, and 24. For all tumor experiments, tumor growth was measured twice per week using digital calipers (Mahr Metrology, MarCal 16 EWRi). Tumor volume was calculated as X*Y*(X/2), where Y is the longest dimension and X is the perpendicular measurement to Y. Mice with tumors >2,000 mm³ or with ulcerated tumors were euthanized. There was no attrition in the animal studies. The investigators were not blinded to allocation during experiments or outcome assessment. Data collection and analysis were not performed blind to the conditions of the experiments. No statistical method was used to predetermine sample sizes. No data were excluded from the analysis.

Cell lines

Generation of TrampC2/hPSMA cells was described in ref. (25). TrampC2/hPSMA cells were maintained in DMEM high glucose (Irvine Scientific) with 5% FBS (VWR), 5% Nu-Serum IV (VWR), 5 μg/mL of insulin (Gemini Bio-Products), 10 nmol/L dehydroandrosterone (Sigma-Aldrich), 2 mmol/L L-glutamine (Irvine Scientific), 500 μg/mL of penicillin, 500 μg/mL streptomycin (Thermo Fisher Scientific), and 500 μg/mL G418 (Thermo Fisher Scientific). The WSU-DLCL2 cell line was obtained from DSMZ (ACC 575, RRID: CVCL_1902). The WSU-DLCL2 cells were maintained in RPMI-1640 (Thermo Fisher Scientific) with 10% FBS (VWR) supplemented with 500 μg/mL of penicillin, 500 μg/mL of streptomycin (Thermo Fisher Scientific), 2 mmol/L glutamine (Irvine Scientific), and 1 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES; Gibco). TrampC2/hPSMA and WSU-DLCL2 cell lines were received in 2020 and authenticated via short tandem repeat profiling (IDEXX BioAnalytics). Cell lines were not reauthenticated within the past year. Cell lines were tested regularly (once per year) for Mycoplasma via real-time PCR assay. All experiments were conducted with low-passage cell cultures (<passage 5).

Cytokine analysis

Serum and tumor samples were collected for cytokine analysis on day 10, 4 hours after receiving indicated treatments. First, blood was collected by retro-orbital bleeds into Microtainer serum tubes (BD 365967). Tumors were harvested and then homogenized using BeadBug 2-mL stainless steel bead tubes (Millipore Z763829-50EA) in ProcartaPlex Cell Lysis Buffer (Invitrogen EXP-99999-000) with additional protease inhibitors (Thermo Fisher Scientific 78429) at 500 μL buffer per 100 mg of tumor per the manufacturer’s instructions. Protein content was measured by Bradford assay (Bio-Rad 5000006) against BSA standards (Thermo Fisher Scientific 23208) and samples diluted to 10 mg/mL in PBS. Cytokine (IL2, IFNγ, TNF, and IL6) concentrations in serum and tumor lysate were analyzed using the V-PLEX Proinflammatory Panel 1 Mouse Kit (Meso Scale Discovery K15048D) and SQ120 (MESO QuickPlex, RRID: SCR_020304) instrument per the manufacturer’s instructions.

Tumor processing

At early (11 days after tumor inoculation, 1 day after first dose) and late (>25 days after tumor inoculation, after the establishment of exhaustion) time points, mice were euthanized and tumors excised. Tumors were weighed and processed using the mouse tumor dissociation kit (Miltenyi Biotec 130-096-730) per the manufacturer’s protocol with the following modifications: Enzyme R was used at 40% of the recommended volume. Tumors were minced using razors (VWR 55411-050), transferred to 5-mL tubes containing enzyme mix, and incubated in a water bath at 37°C for 45 minutes, vortexing every 5 minutes. Dissociated tumors were forced through a 70-µm mesh filter (Miltenyi Biotec 130-110-916) to generate single-cell suspensions for flow cytometry.

In vitro long-term T-cell stimulation assay

T cells were isolated from peripheral blood mononuclear cells (PBMC; ReachBio) of three healthy donors. Data included in this study are derived from one representative donor. The investigators received only deidentified specimens and had no access to donor-identifying information. No ethical approval was required for the study, as informed consent was obtained prior to sample acquisition by the vendors, as stated by their policies. PBMCs were received cryopreserved in 2024 and were stored in liquid nitrogen until used. Naïve T cells were isolated using the EasySep Human Naïve Pan T Cell Isolation Kit (StemCell 17961). Following isolation, T cells were primed with 300 nmol/L DAC at 37°C for 24 hours. Next, WSU-DLCL2 cells were incubated with DAC-primed T cells in a 1:5 cancer:T-cell ratio in the presence of 300 nmol/L DAC and 100 pmol/L CD20×CD3 bispecific. Cancer cells and bispecific were replenished every 2 days.

In vitro cytotoxicity assay

TrampC2/hPSMA cells were plated at 2 × 10⁵/mL in 96-well plates with 1× CellTox Green (Promega G873B) and incubated at 37°C. DAC (Meitheal Pharmaceuticals 71288-119-20) was administered in a twofold serial dilution from 1.752 µmol/L to 1.71 nmol/L. To generate a cell viability standard, TrampC2/hPSMA were heat-killed for 5 minutes at 90°C and combined at varying proportions (from 0% to 100%) with viable cells. Fluorescence was measured after 48 and 72 hours at 485 nm excitation and 535 nm emission wavelengths.

Immunofluorescence

Tumors were fixed for 72 hours in 10% neutral buffered formalin (Azer Scientific PFNBF-1000) and then transferred to 70% EtOH. All staining procedures were performed at HistoWiz, Inc. using the Leica BOND RX automated stainer (Leica Microsystems, RRID: SCR_025548). Samples were processed, embedded in paraffin, and sectioned at 4 μm. Slides were dewaxed using xylene and alcohol-based dewaxing solutions. Heat-induced epitope retrieval was performed using an ethylenediaminetetraacetic acid (EDTA)-based (pH 9) retrieval buffer (Leica Biosystems AR9640) for 20 minutes at 100°C. Samples were blocked with 1× Antibody Diluent/Block (Quanterix ARD1001EA) for 10 minutes at room temperature. Tissues were incubated with anti-CD8 (Cell Signaling Technology 98941, RRID: AB_2756376) at 1:600 for 30 minutes, followed by poly–horseradish peroxidase rabbit IgG secondary antibody (Leica Biosystems PV6119) and either Opal 570 (Quanterix FP1488001KT) or Opal 690 (Quanterix FP1497001KT). Primary antibody was stripped using citrate-based pH 6 epitope retrieval buffer (Leica Biosystems AR9961). Tissues were incubated with either anti–PD-1 (Abcam ab214421, RRID: AB_2941806) at 1:500 or anti–TCF-1 (Cell Signaling Technology 2203, RRID: AB_2199302) at 1:1,000 for 30 minutes, followed by secondary antibody and either Opal 690 (for PD-1) or Opal 620 (for TCF-1, Quanterix FP1495001KT). Secondary antibody was then stripped in the same manner, followed by incubation with 4′,6-diamidino-2-phenylindole (DAPI; Quanterix FP1490) for 5 minutes. Slides were washed, coverslipped (Leica Biosystems CV5030), and visualized using a Vectra Polaris slide scanner (Akoya, RRID: SCR_025508) at 20×.

Image analysis was performed at HistoWiz, Inc. using Visiopharm software (Visiopharm, v2026.02.1 x64, RRID: SCR_021711). Viable tumor area was identified, and large necrotic issue areas were excluded using manual annotation. Cell nuclei were detected and identified using a predeveloped artificial intelligence (AI)–based DAPI detection algorithm; cytoplasm was defined by a 7-pixel margin around each nucleus.

Flow cytometry and cell sorting

For immunophenotyping experiments, tumors were harvested and processed into single-cell suspension as described above. Human TruStain FcX (BioLegend 422302, RRID: AB_2818986) or TruStain FcX anti–mouse CD16/32 (BioLegend 101320, RRID: AB_1574975) was used to block nonspecific staining. For intracellular epitopes, intracellular staining was performed using the eBioscience Foxp3/Transcription Factor Buffer Set (Invitrogen 00-5523-00) per the manufacturer’s protocol. Samples were acquired on an Aurora (Cytek Biosciences, RRID: SCR_019826) or FACSymphony (BD Biosciences, RRID: SCR_022538) cytometer. Data analysis was performed using OMIQ (Dotmatics, RRID: SCR_027879). For single-cell sequencing experiments, cells were sorted using FACSymphony S6 (BD Biosciences, RRID: SCR_028281) or MA900 (Sony, RRID: SCR_026300). For the full list of flow cytometry antibodies used, see Antibodies Used (Supplementary Table S1).

scATAC-seq

scATAC-seq library preparation and sequencing were performed by Singulomics Corporation (https://singulomics.com/). Freshly sorted live mouse T cells were washed with PBS containing 0.04% BSA, and cell count was determined. Nuclei were isolated using the 10x Genomics Low Cell Input Nuclei Isolation protocol. Briefly, cells were centrifuged at 300 × g for 5 minutes at 4°C and resuspended in 50 µL PBS + 0.04% BSA. After centrifugation, 45 µL of the supernatant was removed, and 45 µL of chilled lysis buffer was added. The suspension was gently mixed and incubated on ice for 3 to 5 minutes. Chilled wash buffer was added without mixing, followed by centrifugation at 500 × g for 5 minutes. The supernatant was removed, and 45 µL of chilled diluted nuclei buffer was added. After a second centrifugation, the pellet was resuspended in 7 µL of chilled diluted nuclei buffer. The nuclei concentration was counted and used immediately for transposition. scATAC-seq libraries were prepared using the Chromium Next GEM Single Cell ATAC Reagent Kits v2 (10x Genomics) per the manufacturer’s instructions. Transposed nuclei were loaded with gel beads and reagents onto a Chromium Next GEM Chip H and processed on the Chromium X instrument (10x Genomics, RRID: SCR_024537) to generate barcoded gel beads in emulsion (GEM). GEMs underwent thermal cycling for barcoding, followed by recovery, Dynabead and SPRIselect cleanup, sample indexing PCR, and dual-sided solid-phase reversible immobilization beads (SPRI) size selection. Final libraries were quantified with the Qubit dsDNA High-Sensitivity Assay Kit (Thermo Fisher Scientific Q32854) and assessed using the TapeStation (Agilent Technologies, RRID: SCR_018435). Libraries were sequenced on NovaSeq (Illumina, RRID: SCR_024569), producing approximately 200 million paired-end 150-bp reads per sample.

Reduced representation bisulfite sequencing

Reduced representation bisulfite sequencing (RRBS) library preparation and sequencing were performed by Active Motif (www.activemotif.com). Genomic DNA (gDNA) was extracted using the KingFisher Apex instrument per the manufacturer’s instructions. gDNA (100 ng) was digested with TaqI and MspI at 37°C. Following enzymatic digestion, samples were used for library generation using an RRBS Methyl-Seq System. In brief, digested DNA was randomly ligated and, following fragment end repair, bisulfite-converted using a DNA bisulfite conversion. After conversion and cleanup, samples were amplified using the RRBS Methyl-Seq System protocol for library amplification and purification. Libraries were quantified and sequenced on the Illumina platform at SE75.

RRBS analysis

Raw single-end reads were adapter-trimmed and aligned to the genome using Bismark (29) and Bowtie 2 (30) allowing for no mismatches (−N 0) and a seed substring length of 20 (−L 20). PCR duplicates were then removed, and methylation at CpG sites was extracted using Bismark. The resulted CpG reports were processed with the methylKit R package (RRID: SCR_005177; ref. 31), and only CpG sites covered with at least 10 reads were retained for downstream analyses. For differential methylation analysis, the genome was tiled into 1-kb regions, and significantly differentially methylated tiles were identified. The methylKit package (RRID: SCR_005177)–provided Fisher exact and χ² tests were used to identify differentially methylated sites. Visualization of the methylation levels in selected genomic regions were performed using the Gviz R package (32).

Gene set enrichment analysis

Gene set enrichment analyses (GSEA) were performed using the fgsea R package (bioRxiv 2021:060012) using previously published signatures for naïve, effector, exhausted (33), and memory (34) CD8 T cells.

scATAC-seq quality control and filtering

Cell Ranger ATAC v2.1.0 (10x Genomics) was used to perform demultiplexing and read alignment of scATAC-seq raw data from all treatments and time points. Single-cell data were filtered based on the number of unique fragments per cell and enrichment of ATAC-seq accessibility at transcription start site (TSS), as described in ref. (22). TSS positions were acquired from the TxDb.Mmusculus.UCSC.mm10.knownGene Bioconductor package (RRID: SCR_006442).

Genome accessibility–based cell clustering

Cell clustering based on cell genome accessibility was performed as described in ref. (13), with the minimum cluster size set to 150 for the first round of clustering and a final resolution set to 0.8. For subclustering of T cells, the minimum cluster size was set to 100 and a final resolution set to 0.8.

Gene activity score, TF motif enrichment calculation, and pseudotime analyses

All analyses were performed as described in ref. (13).

Results

PSMA×CD3 induces potent antitumor efficacy at the expense of T-cell fitness over time

To investigate T-cell states induced by a CD3-bispecific in solid tumors, we generated humanized mice for CD3 (Cd3e, Cd3d, and Cd3g) and PSMA and inoculated mice with TrampC2, a well-established murine prostate adenocarcinoma cell line, which was engineered to express hPSMA (TrampC2/hPSMA; refs. 25, 28). Treatment with a PSMA×CD3 bispecific antibody (REGN4336) 10 days after tumor inoculation (Fig. 1A) induced dose-dependent antitumor control (Fig. 1B) and prolonged animal survival at all three concentrations tested (Fig. 1C). The bispecific-mediated tumor-targeted T-cell activation resulted in increased intratumoral proinflammatory cytokines including IL2, IFNγ, and TNF (Fig. 1D), consistent with antitumor activity. In addition, we observed an increase in circulating cytokines, including IL6 (Fig. 1E). Hallmarks of chronic TCR stimulation, which leads to dysfunction, include decreased expression of memory/stem-like markers, suppression of proliferation programs, and upregulation of inhibitory receptors. Flow cytometry analysis of tumor-infiltrating CD8⁺ T cells revealed decreased expression of TCF-1 (Fig. 1F) and a lower frequency of proliferating cells (Ki67⁺; Fig. 1G). CD8⁺ T cells from PSMA×CD3 bispecific–treated animals also demonstrated increased coexpression of inhibitory receptors (TIM3⁺CD39⁺PD-1⁺CTLA-4⁺; Fig. 1H), a hallmark of T_ex cells, compared with cells from isotype-treated mice. These data demonstrate that PSMA×CD3 boosts the effector function of CD8⁺ T cells and inhibits tumor growth while accelerating progression toward terminally differentiated/dysfunctional states.

Figure 1.

PSMA×CD3 drives potent antitumor efficacy and alters T-cell states. A, Experimental schematic detailing treatment regimen with PSMA×CD3. B, Tumor growth and (C) survival curves of mice inoculated with TrampC2/hPSMA cells and treated with PSMA×CD3 at indicated concentrations. Representative data from three independent experiments; n = 7–8 mice per group. B, Two-way ANOVA with the Tukey multiple comparisons test. Mean tumor volumes as SEM. C, Log-rank (Mantel–Cox) test. D and E, Analysis of (D) intratumoral and (E) serum cytokines at 4 hours after the initial dose of PSMA×CD3. Representative data from two independent experiments; n = 7–8 mice per group. One-way ANOVA with Tukey Honestly Significant Difference. F and G, Representative flow cytometry data (left) and summary graph (right) for (F) TCF-1 and (G) Ki67 in CD8⁺ TILs from mice treated with isotype or PSMA×CD3 (0.5 mg/kg) at a late time point (>25 days after tumor inoculation). H, Frequency of CD8⁺ TILs coexpressing TIM3, CD39, PD-1, and CTLA-4 from mice treated with isotype or PSMA×CD3 (0.5 mg/kg) at a late time point (>25 days after tumor inoculation). F–H, Gated on CD44^highCD8⁺CD3⁺ live singlets. Representative data from three independent experiments; n = 5 mice per group. Two-tailed unpaired t test. *, P < 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001. MFI, mean fluorescence intensity. [A, Created in BioRender. Wu, J. (2026) https://BioRender.com/y1wuqmw]

DAC potentiates efficacy of CD3-bispecific antibodies

Given that DNA methylation is a critical epigenetic mechanism for establishing T-cell dysfunction and restricts responses to checkpoint blockade in chronic viral infection and cancer (17), we examined whether pretreatment with DNA demethylating agents prior to CD3-bispecific treatment could enhance the bispecific’s antitumor effect. Humanized mice bearing TrampC2/hPSMA tumors were treated with a low dose of the DNA demethylating agent 5-aza-2′-deoxycytidine (decitabine or “DAC”), mimicking the 5-day dosing schedule for DAC in patients with myelodysplastic syndrome (MDS), followed by the addition of suboptimal dose of PSMA×CD3 (Fig. 2A). The PSMA×CD3+DAC combination significantly suppressed tumor growth (Fig. 2B; Supplementary Fig. S1A) and prolonged survival (Fig. 2C) compared with either DAC or bispecific monotherapy. Intratumoral levels of most proinflammatory cytokines were comparable between PSMA×CD3 monotherapy and PSMA×CD3+DAC combination groups (Fig. 2D). Serum cytokine levels were similar between PSMA×CD3 monotherapy and PSMA×CD3+DAC combination groups (Fig. 2E). In addition, we did not observe any treatment-related effects on animal body weight across any of the treatment groups (Supplementary Fig. S1B). Notably, we observed robust antitumor control mediated by combination of PSMA×CD3 and DAC, even when DAC dose was reduced to 0.01 mg/kg (Supplementary Fig. S1C).

Figure 2.

DAC enhances the efficacy of PSMA×CD3 in vivo. A, Experimental schematic detailing treatment regimen with PSMA×CD3 (0.5 mg/kg) ± DAC (0.2 mg/kg). B, Tumor growth and (C) survival curves of each treatment group. Concatenated data from three independent experiments; n = 7–10 mice per group. B, Two-way ANOVA with the Tukey multiple comparisons test; ****, P ≤ 0.0001. Mean tumor volumes as SEM. C, Log-rank (Mantel–Cox) test; ****, P ≤ 0.0001. D and E, Analysis of (D) intratumoral and (E) serum cytokines at 4 hours after the initial dose with the indicated treatment. Representative data from two independent experiments; n = 7–8 mice per group. D and E, Two-tailed unpaired t test between PSMA×CD3 and PSMA×CD3+DAC; *, P < 0.05. [A, Created in BioRender. Wu, J. (2026) https://BioRender.com/b3xmeu1]

Next, we tested whether our observation is specific to PSMA×CD3 or if addition of a hypomethylating agent can also improve antitumor efficacy of other CD3-bispecifics. To answer this question, we utilized mice humanized for CD3 (Cd3e, Cd3d, and Cd3g) and CD20 and inoculated them with MC38, a murine colorectal cancer cell line, which was engineered to overexpress human CD20 (MC38/hCD20). Humanized mice bearing MC38/hCD20 tumors were primed with DAC followed by the addition of a suboptimal dose of CD20×CD3 [odronextamab (2), approved in Europe for relapsed/refractory follicular lymphoma and diffuse large B-cell lymphoma (DLBCL); Supplementary Fig. S1D]. The combination of CD20×CD3+DAC significantly suppressed tumor growth (Supplementary Fig. S1E) and prolonged survival (Supplementary Fig. S1F) compared with either DAC or bispecific monotherapy. These results collectively demonstrate that disruption of epigenetic reprogramming via a hypomethylating agent enhances the antitumor efficacy of bispecific T-cell engagers.

T cells acquire a spectrum of epigenetic states during bispecific treatment

To catalog the spectrum of T-cell epigenetic landscapes promoted by PSMA×CD3, the epigenetic modifier DAC, and combination treatment, we generated scATAC-seq profiles (10x Genomics platform; ref. 22) of CD2⁺CD90.2⁺ TILs sorted from mice bearing TrampC2/hPSMA tumors (Fig. 3A). To remove low-quality cells, we filtered the dataset using cutoffs of at least 1,000 unique nuclear fragments per cell and a minimum TSS enrichment score of 8 (Supplementary Fig. S2A; “Materials and Methods”). In total, we generated epigenetic profiles of 60,380 TILs. Chromatin accessibility–based clustering of TILs from three time points (early, middle, and late) and four treatments (isotype control, PSMA×CD3 and DAC monotherapies, and PSMA×CD3+DAC combination) resulted in a total of 16 major clusters (Supplementary Fig. S2B). To annotate these clusters, we assessed the general accessibility of lineage-defining marker genes such as Cd3e, Cd4, Cd8, and Ncr1 by computing gene activity scores (Supplementary Fig. S2C). Moreover, examination of the most distinctively accessible genes for each cluster led to the identification of a complex composition of T-cell states. We identified clusters with accessibility at loci of known genes associated with naïve/memory (C12_CD8 and C0_CD4), effector (C3, 6, and 9_CD8 and C2_CD4), dysfunctional (C4 and 7_CD8), and regulatory (C1_CD4) T-cell states, such as Lef1, Prf1, Tox, and Foxp3, respectively (Supplementary Fig. S2D). Although most clusters were populated by immune cells from every treatment at various ratios, we also observed clusters dominated by specific treatments (Supplementary Fig. S2E). Thus, the single-cell epigenetic approach revealed dynamic chromatin accessibility patterns of TILs during cancer progression and upon therapy.

Figure 3.

PSMA×CD3+DAC combination therapy preserves T-cell epigenetic states associated with stemness. A, Experimental schematic detailing treatment regimen with PSMA×CD3 (0.5 mg/kg) and DAC (0.2 mg/kg) and sample collection for scATAC-seq. B, Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) of 35,599 scATAC-seq profiles of CD2⁺CD90.2⁺ TILs from PSMA×CD3- and PSMA×CD3+DAC-treated mice, colored according to different T-cell clusters. C, UMAP of TILs colored by gene scores, reflecting general chromatin accessibility of selected lineage marker genes. D, UMAP of TILs colored by chromVAR TF motif bias–corrected deviations, inferring activity of TFs associated with T-cell stemness. E, Cell alignment to the pseudotime developmental trajectory within CD8⁺ T-cell populations. Smoothened arrow represents a visualization of the interpreted trajectory in UMAP embedding. F, Pseudotime heatmap ordering of gene activity scores of the top 10% most variable genes across the inferred CD8⁺ T-cell trajectory. Cluster identities of ordered cells along the pseudotime trajectory is marked at the top. G, Percent contribution of TILs from PSMA×CD3- and PSMA×CD3+DAC-treated mice in each T-cell cluster included in trajectory analysis from (E). [A, Created in BioRender. Wu, J. (2026) https://BioRender.com/b3xmeu1]

Combination of PSMA×CD3+DAC preserves a pool of TCF-1⁺ T cells and decreases T-cell dysfunction

The addition of DAC to the bispecific treatment regimen demonstrated a significant combinatorial antitumor effect (Fig. 2B and C). To profile chromatin landscapes induced by combination of an epigenetic modifier and CD3-bispecific with higher resolution, we reclustered a subset of 35,595 CD2⁺CD90.2⁺ TILs isolated from animals treated with PSMA×CD3 monotherapy or PSMA×CD3+DAC combination at three different time points. We identified 16 T-cell subclusters, suggesting a diversity of cell states (Fig. 3B). Examination of cluster-specific top gene activity scores revealed CD4 (C0, 1, 2, 4, 8, 12, and 14), CD8 (C3, 5, 6, 9, 10, and 13), and NK (C7) cell clusters (Fig. 3C). A small number of cells with high chromatin accessibility in genes encoding for myeloid or stromal markers (C15 and C11, respectively) that escaped our cell sorting strategy were also detected (Supplementary Fig. S2F).

CD8⁺ T cells play a critical role in cancer cell elimination and are crucial targets of immunotherapy. TCF-1⁺CD8⁺ T cells, also known as T_pex cells, can self-renew and give rise to intermediate T_ex cells, which regain some cytotoxic and effector function, and terminal T_ex cells, marked by the highest inhibitory receptor expression. The presence of TCF-1⁺ TILs has been shown to correlate with improved responses to immunotherapy and longer progression-free survival in patients with cancer (19, 21). Given the critical role of this cell state in cancer immunotherapy, we sought to identify those CD8⁺ T cells in our chromatin accessibility dataset. scATAC-seq profiling allows for inference of TF activity (22, 35, 36). We observed that cluster C10 was characterized by high activity of TFs associated with T_pex, including TCF-1 and LEF1 (Fig. 3D). We reconstructed a pseudotime trajectory to represent the developmental hierarchy of CD8⁺ T cells within tumor using a nearest-neighbor approach as previously described (22). All possible starting clusters were tested followed by sequentially selecting clusters with the highest epigenetic similarity, and candidate trajectories were further ranked by statistical significance (22). On this basis, the developmental sequence was most consistent with a trajectory that begins with the TCF-1–high progenitor-like cluster C10, passes through intermediate clusters C5 and C3, and terminates in cluster C6 (Fig. 3E). Next, we sought to identify genes with dynamic accessibility patterns across the trajectory. Genes that were highly accessible early in the trajectory included genes encoding T_pex markers, such as Ccr7 and Bach2 (Fig. 3F), followed by genes encoding for effector molecules, including Gzmb, Il2, and Ifng (corresponding to C5). In contrast, genes encoding for terminal dysfunction-related markers (e.g., Nr4a3, Tigit, Entpd1, Tox, and Nr4a2) became accessible late in the trajectory (corresponding to C3 and C6; Fig. 3F). These findings indicate that tumor-infiltrating CD8⁺ T cells can be partitioned into T_pex, intermediate T_ex, and terminal T_ex clusters. Notably, T_pex cluster C10 was mainly populated with T cells from tumors treated with PSMA×CD3+DAC combination. The intermediate T_ex cluster C5 was equally represented by cells from bispecific monotherapy and bispecific+DAC combination. Finally, the terminal T_ex cluster C6 was predominantly composed of CD8⁺ T cells from tumors treated with PSMA×CD3 monotherapy (Fig. 3G). Thus, our data indicate that combination of a CD3-bispecific with a hypomethylating agent preserves an epigenetic state associated with T_pex in tumor-infiltrating CD8⁺ T cells and delays acquisition of a chromatin accessibility signature associated with terminal T_ex.

We next tested whether changes in chromatin accessibility promoted by PSMA×CD3 and DAC were reflected at the protein level. To address this question, we used flow cytometry to validate the expression of markers associated with T_pex (TCF-1) and terminal T_ex (including TIM3, CD39, PD-1, and CTLA-4) in tumor-infiltrating CD8⁺ T cells isolated at a late time point (>25 days after tumor inoculation). CD8⁺ T cells from mice treated with PSMA×CD3 alone were enriched for terminal T_ex (marked as TCF-1⁻TIM3⁺) at the expense of T_pex (marked as TCF-1⁺TIM3⁻). Conversely, CD8⁺ T cells from mice treated with PSMA×CD3+DAC maintained significantly higher proportions of TCF-1⁺TIM3⁻ T_pex and lower proportions of TCF-1⁻TIM3⁺ terminal T_ex (Fig. 4A and B). Also, combination treatment with PSMA×CD3+DAC significantly reduced the proportion of CD8⁺ TIL coexpressing TIM3, CD39, PD-1, and CTLA-4 (Fig. 4C). In agreement with flow cytometry results, immunofluorescence analysis showed a similar trend toward increased CD8⁺TCF-1⁺ T cells and reduced CD8⁺PD-1⁺ T_ex in PSMA×CD3+DAC-treated tumors relative to PSMA×CD3 alone (Supplementary Fig. S3A and S3B). Preservation of T_pex and lower frequency of TILs coexpressing multiple inhibitory receptors upon PSMA×CD3+DAC combination was also observed at an earlier time point (11 days after tumor inoculation; 1 day after the initial combination treatment; Supplementary Fig. S3C–S3E). Moreover, PSMA×CD3+DAC combination did not affect the infiltration of major immune cell populations and did not induce upregulation of suppressive myeloid markers compared with bispecific monotherapy (Supplementary Fig. S4A and S4B). These results align with our epigenomic profiling data and suggest that the addition of DAC maintains a stem-like, progenitor T-cell state at both epigenetic and phenotypic levels. Finally, we examined the effect of DAC on cancer cells and observed that DAC did not induce cytotoxicity of TrampC2 cells in in vitro assays (Supplementary Fig. S4C). Moreover, low-dose DAC (0.2 mg/kg) alone had limited impact on survival in NSG mice bearing TrampC2/hPSMA tumors, indicating minimal intrinsic antitumor activity at this dose (Supplementary Fig. S4D and S4E).

Figure 4.

PSMA×CD3+DAC combination therapy reduces the frequency of terminal T_ex in TrampC2/hPSMA tumors. A, Representative flow cytometry data of TIM3 and TCF-1 expression in CD8⁺ TILs from mice treated with PSMA×CD3 (0.5 mg/kg), DAC (0.2 mg/kg), or combination. B, Percentage of TCF-1⁺TIM3⁻ (left) and TCF-1⁻TIM3⁺ (right) TILs at a late time point (>25 days after tumor inoculation) of mice treated with PSMA×CD3, DAC, or combination. C, Percentage of CD8⁺ TILs coexpressing TIM3, CD39, PD-1, and CTLA-4 at a late time point (>25 days after tumor inoculation) of mice treated with PSMA×CD3, DAC, or combination. A–C, Gated on CD44^highCD8⁺CD3⁺ live singlets. Representative data from two independent experiments; n = 5–8 mice per group. B and C, One-way ANOVA with Tukey Honestly Significant Difference. *, P < 0.05; ***, P ≤ 0.001; ****, P ≤ 0.0001.

DAC treatment alters de novo DNA methylation programs associated with chronic stimulation of human T cells

We next sought to examine the impact of DAC on bispecific-mediated activation of human T cells. We first designed an in vitro model system to induce long-term T-cell stimulation with the CD20×CD3 (odronextamab) bispecific molecule over 14 days. The DLBCL cell line WSU-DLCL2 was utilized as a CD20⁺ target, and experimental conditions were optimized to provide continued stimulation via target antigen. T cells from healthy donors were pretreated with DAC, and the hypomethylating agent was replenished every 2 days (Fig. 5A). In agreement with epigenetic profiling data from our humanized model, immunophenotyping analysis showed that human T cells stimulated by CD20×CD3 for 14 days in the presence of DAC demonstrated higher expression of TCF-1 compared with T cells stimulated with bispecific alone (Supplementary Fig. S5A). Next, to further investigate the impact of the epigenetic modifier on the epigenetic landscape of chronically stimulated T cells, we performed RRBS analysis of chronically stimulated T cells with or without DAC. Principal component analysis revealed that naïve and bispecific-stimulated T cells with or without DAC harbor distinct methylomes (Supplementary Fig. S5B). As expected, treatment with DAC decreased methylation across the genome (Supplementary Fig. S5C), primarily at promoter regions (Supplementary Fig. S5D). To resolve these differences with more granularity, we compared methylation profiles at individual loci associated with known T-cell states. Compared with the increased promoter methylation of TCF7 (encodes for TCF-1) observed upon chronic stimulation with CD20×CD3, the addition of DAC resulted in a methylation pattern similar to that of naïve T cells (Fig. 5B). Moreover, we observed demethylation of the effector-associated locus PRF1 during the naïve-to-effector transition driven by bispecific stimulation. The addition of DAC further reduced PRF1 methylation both at the promoter and in the gene body (Fig. 5C).

Figure 5.

DAC treatment preserves a naïve-like methylation chromatin landscape during bispecific-mediated T-cell activation. A, Experimental schematic detailing in vitro chronic stimulation of human T cells with CD20×CD3 ± DAC in the presence of cancer cells. B and C, Methylation profiling of TCF7 (B) and PRF1 (C) loci in naïve and chronically stimulated cells with CD20×CD3 ± DAC. Vertical lines represent individual CpG sites within the genomic region, and line height indicates the level of methylation; gray boxes highlight DMRs at promoters. D, GSEA of gene signatures corresponding to different CD8⁺ T-cell functional states (33, 38) in ranked DMRs from T cells after chronic bispecific stimulation in vitro ± DAC. Vertical lines represent differential methylation levels of individual genes (“CD20xCD3+DAC” minus “no DAC”); height of vertical lines represents the magnitude of differential methylation. Positions of genes on the rank-ordered list with normalized enrichment score (NES) and false discovery rate (FDR) q value of the set. E, Model of the impact of DAC on Tcf7 promoter methylation during CD3 bispecific–mediated chronic stimulation. TAA, tumor-associated antigen. [A, Created in BioRender. Wu, J. (2026) https://BioRender.com/d015z49; E, Created in BioRender. Wu, J. (2026) https://BioRender.com/105abbr]

We next sought to examine the impact of DAC on gene signatures representing various T-cell differentiation states. We first cataloged differentially methylated regions (DMR) in CD20×CD3 chronically stimulated T cells versus those stimulated in the presence of DAC. We then performed GSEA (37) using previously published signatures of naïve, effector, exhausted (33), and memory (38) CD8⁺ T cells. This analysis revealed that genes specifically upregulated in effector and naïve CD8⁺ T cells were significantly enriched in DMRs from DAC-treated cells (Fig. 5D; Supplementary Fig. S5E); genes upregulated in memory CD8⁺ T cells also trended toward enrichment (Fig. 5D). In contrast, no specific pattern was observed for genes specifically upregulated in exhausted CD8⁺ T cells (Fig. 5D). Collectively, our data indicate that CD3 bispecific–forced methylation rewiring is mitigated by DAC treatment, resulting in prolonged accessibility of stemness/memory-related loci like TCF7 (Fig. 5E).

Therapeutic drugs targeting epigenetic modifiers have been efficacious in treating hematologic malignancies, and several epigenetic therapies have been approved for MDS, cutaneous T-cell lymphoma, systemic peripheral T-cell lymphoma, and acute myeloid leukemia. Although various clinical trials are investigating the combination of immunotherapy and epigenetic drugs in solid tumors (39), there is a lack of epigenomic or transcriptomic profiling data available in the context of these treatments. Therefore, to infer the effect of DAC addition on immunotherapy, we interrogated a database compiling a large number of solid tumor samples from patients with cancer treated with immunotherapy (40) for clinical association between expression of genes encoding targets of DAC, specifically DNMT1, DNMT3A, and DNMT3B, and survival of patients treated with PD-1 blockade. In the absence of ICB therapy, DNMT1 and DNMT3A expression levels did not reveal overt differences in survival, whereas patients with low DNMT3B expression had increased survival compared with those with high DNMT3B (Fig. 6A). Analysis of overall survival (OS) in patients treated with PD-1 blockade (nivolumab) revealed that patients with low DNMT3A or DNMT3B expression had significantly increased survival compared with those with high expression (Fig. 6A). DNMT1 expression levels did not show significant differences in OS upon immunotherapy (Fig. 6A). Notably, utilizing a tool to link gene expression and therapy response based on transcriptomic data in diverse solid tumor types (https://rocplot.com; ref. 41), we observed that the expression of DNMT1 and DNMT3A was significantly increased in nonresponder compared with responder patients treated with nivolumab (Fig. 6B). These findings from patients support an association between expression levels of genes encoding DAC targets and survival and prognosis.

Figure 6.

Expression of genes encoding for DAC targets is associated with patient survival and response to PD-1 blockade. A, OS of patients with cancer with no ICB treatment (top graphs) and nivolumab treatment (bottom graphs) based on high (≥median) versus low gene expression of DNMT1, DNMT3A, and DNMT3B. The log-rank test was used to compare survival between groups. B, Gene expression of DNMT1, DNMT3A, and DNMT3B in responders versus nonresponders following nivolumab treatment. The Mann–Whitney test was used to compare response between groups. αPD-1, anti–PD-1.

Discussion

Chronic TCR stimulation by tumor antigens and/or therapeutic modalities such as CD3-bispecifics results in extensive rewiring of the epigenome and a tumor microenvironment composed of heterogeneous T-cell states. Refined understanding of the impact of therapy on evolution of immune cell states at the single-cell level will provide guidance for the design of efficient therapeutic strategies that amplify productive T-cell states and delay progression to counterproductive states. Here, we tested the hypothesis that perturbation of epigenetic rewiring via a hypomethylating agent may improve the efficacy of CD3-bispecific antibodies by utilizing humanized mouse models, CD3-bispecifics with different specificities, a hypomethylating agent approved for hematologic malignancies, and deep single-cell epigenetic profiling, along with DNA methylation and flow cytometry analysis.

Recent single cell–based studies have revealed extensive heterogeneity of tumor-infiltrating immune cells (13, 21). Importantly, the abundance of specific states may influence therapeutic outcomes. However, our understanding of the long-term impact of therapeutics on shaping T-cell epigenetic landscapes and consequently T-cell states is limited. We provide the first comprehensive map of dynamic epigenomic changes in tumor-infiltrating T cells activated in the tumor microenvironment by a CD3-bispecific alone and in combination with an epigenetic modifier. Our findings, corroborated by protein readouts, reveal treatment-specific and cancer progression stage–specific chromatin landscapes.

Progressive de novo DNA methylation programs are critical for the establishment of T-cell exhaustion. Perturbation of these programs rejuvenates T cells and improves responses to ICB (17). DNMT inhibitors directly enhance activation and cytolytic potential in primary human CD8⁺ T cells (42), whereas deletion of DNMT3A from CAR T cells prevents dysfunction and improves efficacy (18). In addition, recent findings suggest that DAC enhances the antitumor potential of PD-1 blockade in tumor models (43). Consistent with these observations, our findings indicate that strategies combining DNA demethylating agents with immunotherapies that unleash and amplify the effector potential of T cells such as CD3-bispecific antibodies may delay acquisition of a fixed epigenetic landscape and preserve immunotherapy-responsive states, thereby improving therapeutic outcomes. Intriguingly, a recent report suggests that inhibition of EZH2, another epigenetic modifier, prevents T-cell exhaustion and enhances T-cell immunotherapies (44). The T-cell epigenetic landscapes driven by modulation of different epigenetic modifiers and their association with response to immunotherapy require further investigation.

Our results provide mechanistic insights into improved antitumor control of PSMA×CD3 or CD20×CD3 in combination with DAC. Employing an in vitro long-term bispecific-mediated stimulation assay and analysis of methylome changes, our results reveal that TCR stimulation in the presence of DAC maintains the promoter of stem-like markers such as TCF7 in a hypomethylated state. Although DAC is not expected to confer locus-specific activity, selective enrichment of naïve/memory signatures among hypomethylated genomic regions in bispecific+DAC-stimulated cells may reflect availability of stemness-related loci for methylation during the progenitor-to-intermediate-to-terminal transition. In this context, DAC inhibition of methylation could preferentially preserve epigenetic features associated with progenitor-like states.

In line with previous reports (42), our single-cell profiling of TILs and in vitro data demonstrate a direct impact of DNA hypomethylating agents on T-cell molecular programs. Although the low dose of DAC used in our in vivo studies (0.2 mg/kg) had minimal impact on the survival of tumor-bearing NSG animals, we cannot exclude potential effects of DAC on cancer cells, such as induction of viral mimicry (45, 46). Further studies are also needed to profile the impact of hypomethylating agents on chromatin landscapes of the broader tumor microenvironment. Another limitation of our study is the use of expression data from human studies based on total RNA isolated from tumors that cannot be directly assigned to T cells. Data from multiple ongoing clinical trials that evaluate the efficacy of epigenetic modifiers in combination with immunotherapies in various solid tumors (47) and adoption of single-cell analysis of clinical samples will provide valuable insights into how epigenetic modifiers may amplify immunotherapy-compatible cell states. Moreover, tertiary lymphoid structures play a role in maintenance of progenitor T-cell states and have been associated with responses to immunotherapy (48, 49). The spatial arrangement and distribution of TCF-1⁺ T_pex upon PSMA×CD3+DAC combination needs further investigation.

In conclusion, our study provides a unique resource of single-cell epigenomic data, along with protein-level validation, of T-cell states upon treatment with bispecific antibodies and outlines the rationale for the combination of epigenetic modifiers with CD3-bispecific antibodies in the clinic. The combination of DAC and CD3-bispecifics delays progression of T cells along the progenitor-to-intermediate-to-terminal T_ex trajectory and significantly improved efficacy of the bispecific. This approach may preserve therapeutically relevant T-cell states and establish durable T-cell immunotherapeutic responses.

Supplementary Material

Figure S1

Decitabine enhances efficacy of CD3-bispecifics in vivo.

Figure S2

Single-cell chromatin accessibility of T cells in TrampC2/hPSMA tumors during disease progression and therapy.

Figure S3

PSMAxCD3+DAC combination therapy reduces frequency of terminal Tex in TrampC2/hPSMA tumors.

Figure S4

PSMAxCD3+DAC combination therapy has limited impact on diverse cell populations within the microenvironment of TrampC2/hPSMA tumors.

Figure S5

Decitabine potentiates global demethylation.

Supplementary Table 1

Antibodies used.

Data availability

The scATAC-seq and RRBS data generated for this study will be publicly available in Gene Expression Omnibus (RRID: SCR_005012) at GSE310545 and GSE310428.

Authors’ Disclosures

Q. Wang reports other support from Regeneron Pharmaceuticals, Inc. during the conduct of the study, as well as a patent for provisional patent issued. C. Guo reports other support from Regeneron Pharmaceuticals, Inc. during the conduct of the study, as well as a patent for provisional patent issued. L.E. Macdonald reports other support from Regeneron Pharmaceuticals, Inc. during the conduct of the study, as well as a patent for provisional patent issued. N. Kourtis reports a patent for 63/898,323 pending to Regeneron Pharmaceuticals, Inc. D. Skokos reports a patent for 63/898,323 pending. No disclosures were reported by the other authors.

Authors’ Contributions

M. Aleynick: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Q. Wang: Data curation, software, formal analysis, investigation, visualization, methodology, writing–review and editing. J.E. Wu: Formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Lemus: Data curation, software, formal analysis, investigation, visualization, methodology, writing–review and editing. M. Chaimowitz: Formal analysis, validation, investigation, visualization, methodology, writing–original draft. H. Mir: Investigation, methodology. J.J. Warsaw: Investigation, methodology. C. Guo: Investigation, methodology, project administration. L.E. Macdonald: Methodology, project administration. J.C. Waite: Project administration, writing–review and editing. N. Kourtis: Conceptualization, supervision, writing–original draft, project administration. D. Skokos: Conceptualization, supervision, writing–review and editing.