A radiopharmaceutical enhances CAR T cells against radio-sensitive and radio-resistant neuroblastoma by tumor sensitization and TME remodeling Authors: Constanza Rodriguez, Robert S. Edinger, George Diehl, Jeyshka M. Reyes-González, Woonghee Lee, John Buckley, Jangsuk Oh, Divya Nambiar, Kwamena Baidoo, Reona Okada, Ira Phadke, Zoe Schmiechen, Hannah Khan, Yash Sakhalkar, Angel G. Cortez, Anders Josefsson, Jessie R. Nedrow, Ayla White, Deborah E. Citrin, Xiyuan Zhang, Carolyn J. Anderson, Freddy E. Escorcia, Ravi B. Patel, Rosa Nguyen Published: 2026 ABSTRACT SummaryChimeric antigen receptor (CAR) T cell therapy has limited efficacy against solid tumors such as neuroblastoma (NB). Key obstacles include extensive tumor burden and the presence of an immunosuppressive tumor microenvironment (TME). We employ targeted radiopharmaceutical therapy (RPT) using [67Cu]Cu-LLP2A and show that it potentiated the anti-tumor activity of CAR T cells in radio-sensitive and radio-resistant NB models via distinct mechanisms. In radio-sensitive NB, RPT is directly tumoricidal while also enhancing CAR T cell efficacy through pro-immune pathways, most notably via the TNF-α pathway, leading to paracrine activation of T cells. In radio-resistant NB, RPT improves CAR T cells by remodeling the myeloid compartment in the TME and increasing the formation of immunological niches of cytotoxic CD8+ GZMB+ and CD4+ GZMB+ CAR T cells. While neither treatment modality alone can effectively treat NB, the combination of VLA-4-targeted RPT and GD2 or B7-H3 CAR T cells augments anti-tumor efficacy, resulting in marked tumor regression in preclinical NB models. Introduction Chimeric antigen receptor (CAR) T cell therapy has shown unprecedented success in treating relapsed or refractory leukemias and lymphomas.1,2 However, its impact has been limited in most solid tumors, including neuroblastoma (NB), a rare but aggressive pediatric cancer.3,4,5,6,7 This reduced efficacy has been largely attributed to poor CAR T cell persistence, restricted infiltration and function in an immune-suppressive tumor microenvironment (TME), and their inability to overcome high tumor burden.8 To address these issues, pre-conditioning chemotherapy with agents such as fludarabine (FLU) and cyclophosphamide (CY) has been used to deplete host lymphocytes, lower tumor mass, and eliminate immune-suppressive cells, thereby enhancing adoptive T cell engraftment and function.9,10 Similarly, radiation therapy can have tumoricidal and immune-priming effects, promoting immune cell activation and infiltration into the TME.11 Traditional methods, such as total body irradiation (TBI) via external beam radiation (EBRT), can systemically reduce tumor load and induce lymphodepletion, supporting adoptive T cell engraftment and improving overall survival.12 However, by design, TBI delivers radiation to the whole body over a few minutes, which can cause unwanted off-target toxicity to healthy tissues, exacerbating short and long-term adverse effects, including those related to CAR T cells.13 In contrast, targeted radiopharmaceutical therapy (RPT) offers a systemically administered, precise delivery of radiation over an extended period of time, potentially minimizing off-target toxicity.14 131I-metaiodobenzylguanidine (MIBG), an RPT approved for patients with pheochromocytomas and paragangliomas,15 has been developed over decades for the therapy of NB.16,17 Thus far, 131I-MIBG has not been curative by itself, but in combination with myeloablative chemotherapy, it was shown to be tolerable and active against relapsed/refractory NB.16,18,19,20 More recently, it has been tested in newly diagnosed patients in the phase III Children’s Oncology Group trial ANBL1531, with pending results. In this study, we investigated the use of [67Cu]Cu-LLP2A as a treatment before CAR T cells in preclinical models of NB. Our RPT agent is different from MIBG because LLP2A is a peptidomimetic ligand for the very late antigen-4 (VLA-4, α4β1 integrin) integrin receptor expressed on neural crest cell-derived tumors like NB and human immune myeloid cells and lymphocytes.21,22 The short biological half-life of LLP2A and favorable decay characteristics of the β-emitter copper-67, with a mean energy of 141 keV and a half-life of 2.58 days, ensure minimal radiation exposure to adoptively transferred CAR T cells, limiting direct toxicity to these cells. We tested this approach in NB xenograft models using two clinically relevant CAR constructs: GD2-directed and B7-H3-directed CAR T cells, both of which are under clinical development for NB and other solid tumors (NCT04539366, NCT03721068, NCT04897321, and NCT06500819). We demonstrate that [67Cu]Cu-LLP2A enhances CAR T cell therapy to induce significant tumor regression in radio-sensitive and radio-resistant NB through different modes of action, offering a promising strategy for integrating RPT with adoptive T cell therapy in solid tumors. Results [67Cu]Cu-LLP2A potentiates CAR T cell efficacy in radio-resistant and radio-sensitive NB in vivo To test the combination of [67Cu]Cu-LLP2A with CAR T cell therapy in vivo, we used GD2-CAR T cells against GD2+ radio-sensitive IMR-5 cells (MYCN-amplified, NRAS-WT, and TP53-WT) and B7-H3-CAR T cells against B7-H3+ radio-resistant SK-N-AS cells (MYCN-WT, NRAS-mutant, and TP53-mutant). IMR-5 and SK-N-AS both have intermediate to low VLA-4 expression and thus show low but distinct uptake of radiolabeled LLP2A by biodistribution and dosimetry in vivo (Figures S1A–S1E).22 Mice bearing subcutaneous IMR-5 tumors were treated with either non-radioactive (cold) mass equivalent LLP2A or radiolabeled [67Cu]Cu-LLP2A at 0.5 (18.5 MBq) or 1.0 mCi (37 MBq) 4 days before the intravenous administration of untransduced (UT) mock T cells or CAR T cells targeting GD2 (Figure 1A). Cold LLP2A did not affect CAR T cell performance (Figure S2). This delayed injection schedule was applied to allow sufficient time for radiation to decay by an effective half-life (T1/2 = 61.8 h).14 Tumor progression was monitored via caliper measurements at regular intervals. Complete regression occurred in 5 of 7 mice (71%) treated with cold LLP2A and CAR T cells, whereas 6 of 8 (75%) receiving 0.5 mCi [67Cu]Cu-LLP2A with CAR T cells achieved complete tumor clearance (Figures 1B and 1C). Interestingly, increasing the dose to 1.0 mCi did not yield additional benefit (4 of 7 mice were tumor-free [57%]). In an orthotopic IMR-5 model, which is less responsive to GD2 CAR T cells alone, bioluminescence imaging demonstrated that the combination of GD2-CAR T cells with 0.5 mCi [67Cu]Cu-LLP2A was the only condition to induce tumor regression and achieve the highest survival by day 23 (Figure 1D). We then evaluated SK-N-AS tumors established orthotopically via para-adrenal implantation (Figure 1E); bioluminescence imaging showed that all mice treated with cold LLP2A combined with B7-H3 CAR T cells had disease progression. In contrast, treatment with 0.5 mCi and B7-H3 CAR T cells resulted in complete responses in all mice, with only one showing a small metastasis on the opposite kidney. Interestingly, the higher RPT dose of 1.0 mCi with B7-H3 CAR T cells induced tumor clearance in 6 of 10 (60%), potentially also affecting adoptive T cells, as noted in the heterotopic IMR-5 model (Figures 1F and 1G; Figure S3). These findings correlated with significantly reduced primary and metastatic tumor burden at endpoint in both RPT-treated groups (Figure 1H), showing that more animals had complete responses (CRs) upon dissection on day 23 post-CAR compared to the cold LLP2A + CAR group (Figure 1I). The decreased efficacy at the higher RPT dose level was associated with a decline in body weight during the course of therapy and clinical anemia, suggesting that while there is an enhanced effect with the combination, exceeding a certain radiation threshold becomes detrimental, likely harming the effector CAR T cells and animals (Figures 1J and 1K).Figure 1CAR T cells and RPT cooperate in preclinical models of NB(A) Schema of the heterotopic IMR-5 model. Caliper measurements of subcutaneous IMR-5 tumors.(B and C) (B) Average tumor volume with multiple comparisons table and (C) individual tumor curves are shown where each line represents an individual mouse. Plots are divided by RPT regimen: cold LLP2A + GD2 CAR (blue), 0.5 mCi [67Cu]Cu-LLP2A + GD2 CAR (purple), and 1.0 mCi [67Cu]Cu-LLP2A + GD2 CAR (red), while untransduced (UT) mock T cells are shown in gray. The number of mice that achieved complete tumor regression is indicated on each plot. Two-way ANOVA with multiple comparisons test; ∗p < 0.05; ∗∗p < 0.01.(D) Average IVIS (in vivo imaging system) bioluminescence imaging signals in mice with orthotopic IMR-5 treated with UT mock or GD2 CAR T cells with and without 0.5 mCi [67Cu]Cu-LLP2A. Paired t test; ∗p < 0.05.(E) Schema of orthotopic SK-N-AS model.(F) Average IVIS bioluminescence signals of mice, subdivided by RPT group as in (B). Two-way ANOVA with multiple comparisons test; ∗∗p < 0.01.(G) Individual IVIS bioluminescence signal curves are shown where the light gray area in each plot marks background luminescence levels. Mice whose curves reached background levels were marked as complete remission.(H) Tumor weights at the end of the study on day 23. One-way ANOVA with multiple comparisons test; ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01.(I) Stacked bar graphs with the frequency of complete responders (CR) and non-responders (no CR) per group. Fisher’s exact test; ∗∗∗∗p < 0.0001.(J and K) Longitudinal body weight of treated mice, suggesting toxicity at the higher RPT dose level in the (J) IMR-5 and (K) SK-N-AS model. Two-way ANOVA with multiple comparisons test; ∗∗p < 0.01; ∗∗∗p < 0.001.Error bars in subpanels are standard deviations. CAR T cells and RPT cooperate in preclinical models of NB (A) Schema of the heterotopic IMR-5 model. Caliper measurements of subcutaneous IMR-5 tumors. (B and C) (B) Average tumor volume with multiple comparisons table and (C) individual tumor curves are shown where each line represents an individual mouse. Plots are divided by RPT regimen: cold LLP2A + GD2 CAR (blue), 0.5 mCi [67Cu]Cu-LLP2A + GD2 CAR (purple), and 1.0 mCi [67Cu]Cu-LLP2A + GD2 CAR (red), while untransduced (UT) mock T cells are shown in gray. The number of mice that achieved complete tumor regression is indicated on each plot. Two-way ANOVA with multiple comparisons test; ∗p < 0.05; ∗∗p < 0.01. (D) Average IVIS (in vivo imaging system) bioluminescence imaging signals in mice with orthotopic IMR-5 treated with UT mock or GD2 CAR T cells with and without 0.5 mCi [67Cu]Cu-LLP2A. Paired t test; ∗p < 0.05. (E) Schema of orthotopic SK-N-AS model. (F) Average IVIS bioluminescence signals of mice, subdivided by RPT group as in (B). Two-way ANOVA with multiple comparisons test; ∗∗p < 0.01. (G) Individual IVIS bioluminescence signal curves are shown where the light gray area in each plot marks background luminescence levels. Mice whose curves reached background levels were marked as complete remission. (H) Tumor weights at the end of the study on day 23. One-way ANOVA with multiple comparisons test; ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01. (I) Stacked bar graphs with the frequency of complete responders (CR) and non-responders (no CR) per group. Fisher’s exact test; ∗∗∗∗p < 0.0001. (J and K) Longitudinal body weight of treated mice, suggesting toxicity at the higher RPT dose level in the (J) IMR-5 and (K) SK-N-AS model. Two-way ANOVA with multiple comparisons test; ∗∗p < 0.01; ∗∗∗p < 0.001. Error bars in subpanels are standard deviations. Next, we profiled CAR T cells by multi-color flow cytometry in both models to better understand the effect of RPT after pre-treatment with cold RPT and 0.5 mCi [67Cu]Cu-LLP2A. When evaluating the engraftment from the spleens of subcutaneously injected IMR-5-bearing mice on day 37, we observed similar CD3+ T cell proportions among groups but higher proportions of CD8+, but not CD4+ T cells, among splenocytes in the 0.5 mCi [67Cu]Cu-LLP2A + CAR T cell group compared to UT mock T cell groups (Figures 2A and 2B; Figure S4A). To test for differences in T cell phenotypes that might explain enhanced efficacy or T cell persistence, we compared CAR expression, T cell memory, and exhaustion markers but found no significant differences between cold and RPT-treated groups in the IMR-5 model (Figures 2C–2E). In contrast, in SK-N-AS-bearing mice, total CD3+ T cells were enriched at end point in animals treated with CAR T cells and RPT, but neither CD8+ nor CD4+ T cells were differentially enriched in the CAR/RPT compared to the CAR alone group (Figures 2F and 2G; Figure S4B). Similar to GD2-CAR expression, B7-H3-CARs were highly expressed across all CAR groups (Figure 2H). The different CD4+/CD8+ splenic enrichments between models (IMR-5 vs. SK-N-AS) suggest that RPT can differentially impact T cell subsets. While tumor-infiltrating T cells were very scarce in the subcutaneous IMR-5 tumors at the study endpoint, precluding robust analysis, SK-N-AS tumors had abundant T cells at the study endpoint but very few samples due to complete tumor regression in the combination groups. Thus, we collected tumor tissue at 8 days after adoptive T cell transfer. We found that at the early time point, SK-N-AS tumors also exhibited considerable T cell infiltration and produced distinct clusters of CAR+ tumor-infiltrating CD44+ CD45RO+ CD45RA− CD4+ and CD8+ antigen-experienced T cells that were enriched 8 days after adoptive transfer (Figure 2I). While no major differences in T cell phenotypes were found, higher frequencies of T cells and CAR T cells among all live cells were evident in tumor tissues. Importantly, the myeloid compartment in these tumors was also remodeled, exhibiting higher fractions of Ly6G+ neutrophils and lower proportions of F4/80+ Ly6C+ macrophages in [67Cu]Cu-LLP2A- and B7-H3 CAR-treated animals. Additionally, tumor-infiltrating and splenic T cells from the SK-N-AS group with 0.5 mCi [67Cu]Cu-LLP2A had lower expression of inhibitory receptors and a large expansion of central memory T cells at the study endpoint (Figures 2J–2L). Specifically, evaluating CD39+ CD69+ T cell subsets, which are regarded as terminally differentiated effectors, and CD39− CD69− T cells with a memory-progenitor stem-like phenotype,23 we noted lower expression of PD-1 and TIM-3 in RPT-treated samples, indicating desirable T cell differentiation under low radiation exposure (Figures S4C–S4F).24,25Figure 2RPT increases CD8 T cell expansion in treated mice(A–C) Frequency plots of (A) CD4+, (B) CD8+, and (C) GD2 CAR+ T cells in spleens from IMR-5-bearing mice treated with adoptive T cells with and without RPT 23 days after T cell transfer. One-way ANOVA with multiple comparisons test; ∗p < 0.05; ∗∗∗∗p < 0.0001.(D) Stacked bar graph of cell fractions with 0–4 markers (LAG-3, TIM-3, PD-1, and CD39).(E) Fractions of T cell subsets in the spleen based on expression of CD45RA and CD62L.(F–H) Frequency plots of (F) CD4+, (G) CD8+, and (H) B7-H3 CAR+ T cells in spleens from SK-N-AS-bearing mice treated with adoptive T cells with and without RPT 23 days after T cell transfer. One-way ANOVA with multiple comparisons test; ∗∗p < 0.01; ∗∗p < 0.01; ∗∗∗∗p < 0.0001.(I) tSNE plots of tumor-infiltrating T cells and murine myeloid cells from the SK-N-AS-bearing mice 8 days after adoptive T cell transfer. The clusters are color coded with respective labels shown in the plots. The pie charts in each subpanel indicate the proportions of each colored subcluster.(J–L) Stacked bar graph of (J) tumor-infiltrating and (K) splenic T cell fractions with 0–4 markers (TIM-3, PD-1, CD39, and CD69) and (L) memory subsets of tumor-infiltrating T cells defined by CD62L and CD45RA positivity at the study endpoint.Error bars in subpanels are standard deviations. RPT increases CD8 T cell expansion in treated mice (A–C) Frequency plots of (A) CD4+, (B) CD8+, and (C) GD2 CAR+ T cells in spleens from IMR-5-bearing mice treated with adoptive T cells with and without RPT 23 days after T cell transfer. One-way ANOVA with multiple comparisons test; ∗p < 0.05; ∗∗∗∗p < 0.0001. (D) Stacked bar graph of cell fractions with 0–4 markers (LAG-3, TIM-3, PD-1, and CD39). (E) Fractions of T cell subsets in the spleen based on expression of CD45RA and CD62L. (F–H) Frequency plots of (F) CD4+, (G) CD8+, and (H) B7-H3 CAR+ T cells in spleens from SK-N-AS-bearing mice treated with adoptive T cells with and without RPT 23 days after T cell transfer. One-way ANOVA with multiple comparisons test; ∗∗p < 0.01; ∗∗p < 0.01; ∗∗∗∗p < 0.0001. (I) tSNE plots of tumor-infiltrating T cells and murine myeloid cells from the SK-N-AS-bearing mice 8 days after adoptive T cell transfer. The clusters are color coded with respective labels shown in the plots. The pie charts in each subpanel indicate the proportions of each colored subcluster. (J–L) Stacked bar graph of (J) tumor-infiltrating and (K) splenic T cell fractions with 0–4 markers (TIM-3, PD-1, CD39, and CD69) and (L) memory subsets of tumor-infiltrating T cells defined by CD62L and CD45RA positivity at the study endpoint. Error bars in subpanels are standard deviations. Collectively, while only one schedule (RPT 5 days before CAR T cells) was presented, these results emphasize the critical role of dose and appropriate timing in optimizing the effect of RPT and CAR T cell immunotherapy, while preserving CAR T cell function and phenotype. Pre-treatment with [67Cu]Cu-LLP2A activates pro-inflammatory programs in a radio-sensitive NB model To evaluate the mechanisms underlying the therapeutic effect of CARs with RPT, radiation sensitivity was assessed by exposing IMR-5 and SK-N-AS cells to free copper-67 (0.74 MBq, 20 μCi) in vitro, simulating the in vivo radiation exposure during RPT with [67Cu]Cu-LLP2A. We intentionally used this dose to avoid tumor ablation and demonstrate the priming effect of RPT. Results were normalized to cells receiving no radionuclide. IMR-5 cells treated in vitro with free copper-67 showed a gradual decline in cell viability over 7 days (Figure 3A), indicating sensitivity to low-dose radiation. In contrast, the SK-N-AS line demonstrated no change in viability when exposed to low-dose radiation treatment (Figure 3B), consistent with its radio-resistant status. Radiation can both exert direct cytotoxic effects against and enhance the immunogenicity of tumor cells, potentially rendering them more sensitive to CAR T cell therapy. To test this potential effect, we pre-treated NB cells with and without free copper-67 and subsequently subjected them to CAR T cells for 72 h. At low effector-to-tumor (E:T) ratios of 1:8 for IMR-5 (Figure S5A) and 1:30 for SK-N-AS (Figure S5B), when CAR T cells have no anti-tumor effect, copper-67-pre-treated NB cells exhibited more CAR-mediated cell death than untreated NB cells. These findings suggest that low-dose radiation enhances tumor susceptibility to CAR T cell cytotoxicity, particularly under suboptimal immune effector conditions.Figure 3Radiation induces pro-inflammatory programs in radio-sensitive IMR-5 in vitro(A and B) Viability assay of (A) IMR-5 and (B) SK-N-AS after treatment with free copper-67. Cell viability was determined at days 1, 3, and 7 post-treatment. One-way ANOVA test with multiple comparisons; ns, p > 0.05; ∗p < 0.05. Means and standard deviations are plotted.(C) Experimental schema.(D) Principal-component analysis (PCA) plot of IMR-5 samples, color-coded by time point, with circles representing controls and triangles representing treatment groups.(E) Volcano plot with the top differentially expressed genes at 72 h in IMR-5, marked in red. Blue indicates genes of the TNF-α pathway.(F–I) GSEA of IMR-5, evaluating the (F) TNF-α and (G) epithelial-to-mesenchymal transition (EMT) pathway, as well as (H) p53 and (I) MYC targets. NES, normalized enrichment score; FDR q, false discovery rate q value.(J) PCA plot of SK-N-AS RNA-seq samples.(K) Volcano plot of differentially expressed genes at 72 h in SK-N-AS, top genes marked in red. No enrichment of TNF-α-related genes is found.(L and M) GSEA of SK-N-AS for (L) TNF-α and (M) epithelial-to-mesenchymal transition pathways.(N and O) Ingenuity Pathway Analysis (IPA) in (N) IMR-5 and (O) SK-N-AS. Radiation induces pro-inflammatory programs in radio-sensitive IMR-5 in vitro (A and B) Viability assay of (A) IMR-5 and (B) SK-N-AS after treatment with free copper-67. Cell viability was determined at days 1, 3, and 7 post-treatment. One-way ANOVA test with multiple comparisons; ns, p > 0.05; ∗p < 0.05. Means and standard deviations are plotted. (C) Experimental schema. (D) Principal-component analysis (PCA) plot of IMR-5 samples, color-coded by time point, with circles representing controls and triangles representing treatment groups. (E) Volcano plot with the top differentially expressed genes at 72 h in IMR-5, marked in red. Blue indicates genes of the TNF-α pathway. (F–I) GSEA of IMR-5, evaluating the (F) TNF-α and (G) epithelial-to-mesenchymal transition (EMT) pathway, as well as (H) p53 and (I) MYC targets. NES, normalized enrichment score; FDR q, false discovery rate q value. (J) PCA plot of SK-N-AS RNA-seq samples. (K) Volcano plot of differentially expressed genes at 72 h in SK-N-AS, top genes marked in red. No enrichment of TNF-α-related genes is found. (L and M) GSEA of SK-N-AS for (L) TNF-α and (M) epithelial-to-mesenchymal transition pathways. (N and O) Ingenuity Pathway Analysis (IPA) in (N) IMR-5 and (O) SK-N-AS. To investigate the mechanisms by which RPT enhances CAR T cell activity, we performed bulk RNA sequencing (RNA-seq) on control and treated NB cells harvested at 24 and 72 h post-treatment (Figure 3C). In the radio-sensitive IMR-5 cell line, RPT induced marked transcriptional reprogramming, with the most pronounced differences observed at 72 h (Figure 3D). Among the top upregulated genes were CHGA, RET, and VGF, associated with differentiation and adrenal lineage identity,26 as well as several components of the TNF-α signaling pathway (Figure 3E), as confirmed by gene set enrichment analysis (GSEA) (Figure 3F). Additional enrichment was observed in pathways related to epithelial-mesenchymal transition (EMT) and the TP53 tumor suppressor pathway (Figures 3G and 3H). MYC target genes were negatively enriched, suggesting that alterations of the MYCN-driven transcriptome may also contribute to the anti-tumor effect, although MYCN protein levels were not altered by radiation treatment at 72 h (Figure 3I; Figure S5C). Network analysis of the differentially expressed genes revealed a tightly interconnected group of pro-inflammatory mediators (e.g., IFNγ, CXCR4, CX3CL1, and TNF), alongside key regulators of the cell cycle, including TP53 and CDKN1A (Figure S5D). Notably, TNF-α is known to be induced in NB by radiation and to activate tumor pro-survival and inflammatory pathways via NF-κB.27 However, in this context, we hypothesized that the activation of the TNF-α pathway in NB may activate T cells in a paracrine manner and contribute to the synergy between RPT and CAR T cell therapy. To further evaluate this effect, we expanded our studies to other molecularly diverse NB cell lines and showed differential upregulation of genes of the TNF-α pathway (Figure S6A). Hypothesizing that TNF-α acts in a paracrine manner to stimulate adjacent T cells, we performed transient TNF-α knockdown (KD) in NB cells and assessed the activation of T cells grown in the supernatant of wild-type and TNF-α-deficient NB cells after irradiation. We found that supernatants from irradiated NB cells with intact TNF-α expression were capable of activating T cells, marked by elevated expression of CD69 and CD25, markers of early and late T cell activation (Figure S6B). In contrast, this effect was diminished in IMR-5 TNF-α-KD conditions, implicating TNF-α as a mediator of T cell activation in the IMR-5 model. In vivo testing using TNF-α blockade was technically challenging as mice treated with 0.5 Gy of EBRT, adalimumab, and GD2 CAR T cells developed rapid systemic toxicity associated with markedly elevated murine IL-6, CCL2, CXCL10, and IL-1β levels, leading to death shortly after adoptive T cell transfer (Figures S6C and S6D). RNA-seq analysis of the radio-resistant SK-N-AS cell line revealed clustering by time point rather than by radiation (Figure 3J). The transcriptional changes observed in IMR-5 were not recapitulated in SK-N-AS. Notably, TNF-α pathway genes were not enriched, and EMT-related gene sets were negatively enriched (Figures 3K–3M). The gene interaction network in SK-N-AS was sparse, with IL17A and IL1A being central nodes (Figure S5E). Ingenuity Pathway Analysis (IPA) further reinforced these distinctions. While copper-67-treated IMR-5 cells exhibited activation of pro-inflammatory, immune-regulatory, and lymphocyte-interactive pathways (Figure 3N), SK-N-AS cells displayed upregulation of neuronal programs, including acetylcholine, GABA, and ion channel signaling pathways associated with synapse formation and neurovascular coupling (Figure 3O). Collectively, these results suggest that low-dose radiation induces a pro-inflammatory transcriptional signature in the radio-sensitive IMR-5 line, thereby enhancing the susceptibility to CAR-mediated killing. In contrast, in the radio-resistant SK-N-AS line, radiation activates neuronal signaling pathways with no evident tumor cell-intrinsic immunomodulatory impact. The synergy observed in vivo for SK-N-AS might be less about direct paracrine activation of CAR T cells by tumor cell-derived TNF-α, as seen in IMR-5, and more about RPT’s effects on the TME. Thus, to further define the mechanisms driving these differential responses, we next characterized the changes in the TME of SK-N-AS in vivo. [67Cu]Cu-LLP2A reshapes the TME by depleting immunosuppressive macrophages and bolstering cytotoxic effectors in a radio-resistant NB model To explore how RPT alters the TME of neuroblastoma, we performed single-cell RNA-seq on orthotopic SK-N-AS tumors 7 days after treatment with 1.5 mCi [67Cu]Cu-LLP2A or cold LLP2A. We chose this dose to maximize the differences induced by RPT. Mice bearing SK-N-AS xenografts were randomized to receive either UT mock or B7-H3 CAR T cells alone or adoptive T cells with [67Cu]Cu-LLP2A. On day 7 post-injection, when T cell infiltration and expansion were evident (Figure S7A), tumors from three animals per group were pooled for analysis. We resolved three principal cell populations, comprising NB cells, human T lymphocytes, and murine myeloid cells, further subdividing each into functional subclusters (Figures 4A and 4B; Figures S7B and S7C). Quantification of these subsets revealed a marked lower frequency of M2-like tumor-associated macrophages (TAMs) in RPT- and CAR-treated tumors compared to CAR treatment alone, accompanied by higher frequencies of human T cells, suggesting better infiltration, expansion, or survival of CAR T cells (Figure 4C). These findings align with changes in cell abundances in the TME in SK-N-AS tumors (Figure 2I). Despite their murine origin, myeloid cells showed strong predicted crosstalk with human T cells, notably through the immunomodulatory CD74-MIF ligand-receptor pair (Figure 4E; Table S1). Furthermore, myeloid depletion with liposomal clodronate in addition to B7-H3 CAR T cells significantly improved murine survival and tumor control, supporting a role for murine myeloid cells in modulating human B7-H3 CAR T cell activity in this model (Figure 4D). IPA revealed that CD4+ CAR T cells in RPT-treated tumors upregulated translational machinery important for T cell expansion, whereas those from untreated lesions engaged more inflammatory signaling cascades (Figure 4F). In non-irradiated controls, CD8+ CAR T cells were enriched for cytokine production, necroptosis, and apoptotic pathways, which are signatures of heightened but potentially unsustainable, self-limiting immune activation (Figure 4G). However, CD8+ CAR T cells from RPT-treated tumors exhibited upregulation of cAMP, mitochondrial dysfunction, and mTOR signaling pathways, each implicated in T effector differentiation and activation (Figure 4G). Effector molecule and cytokine profiling of tumor tissues from treated mice revealed increased levels of human granzyme A, granzyme B, perforin, and granulysin, consistent with enhanced T cell cytotoxic activity RPT/CAR vs. CAR-alone groups. Human IFNγ and murine IL-6 levels did not differ between the groups. These findings support that RPT augments T cell effector function, resulting in increased effector molecule production (Figure S7D). Together, these findings suggest that [67Cu]Cu-LLP2A not only reduces suppressive macrophage populations but also fine-tunes CAR T cell function and differentiation, favoring a more durable and effective phenotype, rather than rapid and intense activation followed by premature exhaustion.Figure 4[67Cu]Cu-LLP2A modulates immune-suppressive macrophage and T cell effector populations in the TME(A) PCA plot of resolved cell populations from integrated single-cell RNA-seq samples.(B) Dot plot of key marker genes used to annotate major cell clusters. The color of the dots indicates expression gene levels, while the size of the dot represents the percentage of cells in each respective cluster with detectable gene expression.(C) Stacked bar graph of the frequencies of each cellular subset by the treatment group.(D) Survival curves of mice with orthotopic SK-N-AS treated with UT mock or B7-H3 CAR T cells with Clodrosome or Encapsome as control to deplete myeloid cells. Logrank test; ∗∗p < 0.01; (n = 5-9/group).(E) Heatmap to visualize the number of receptor-ligand interactions across subsets.(F and G) IPA of (F) CD4+ CAR+ and (G) CD8+ CAR+/cytotoxic T cells comparing samples from RPT-treated and non-treated mice. [67Cu]Cu-LLP2A modulates immune-suppressive macrophage and T cell effector populations in the TME (A) PCA plot of resolved cell populations from integrated single-cell RNA-seq samples. (B) Dot plot of key marker genes used to annotate major cell clusters. The color of the dots indicates expression gene levels, while the size of the dot represents the percentage of cells in each respective cluster with detectable gene expression. (C) Stacked bar graph of the frequencies of each cellular subset by the treatment group. (D) Survival curves of mice with orthotopic SK-N-AS treated with UT mock or B7-H3 CAR T cells with Clodrosome or Encapsome as control to deplete myeloid cells. Logrank test; ∗∗p < 0.01; (n = 5-9/group). (E) Heatmap to visualize the number of receptor-ligand interactions across subsets. (F and G) IPA of (F) CD4+ CAR+ and (G) CD8+ CAR+/cytotoxic T cells comparing samples from RPT-treated and non-treated mice. [67Cu]Cu-LLP2A-treated tumors contain enriched cytotoxic T cell clusters within pro-inflammatory niches in radio-resistant NB To investigate the spatial distribution of CAR T cells within the TME following RPT, we used multiplex fluorescence imaging on tumor tissues collected on days 8 (IMR-5) and 9 (SK-N-AS) post-adoptive transfer. Staining panels included markers for human T cells (CD4 and CD8), effector function (granzyme B [GMZB]), inhibitory receptors (CTLA-4), murine TAMs (CD206), and myofibroblasts or vessel-associated structures (α-SMA). In the subcutaneous IMR-5 model, the density of tumor-infiltrating CD4+ and CD8+ T cells, as well as α-SMA+ myofibroblasts or vessel-associated structures, remained largely unchanged across treatment groups (Figure 5A). However, a notable increase in CD206+ M2-like macrophages was observed in tumors from mice treated with both RPT and CAR T cells compared to CAR alone. In the orthotopic SK-N-AS model, tumors treated with the combination of RPT and CAR T cells showed markedly higher infiltration of cytotoxic CD8+ GZMB+ and CD4+ GZMB+ T cells (Figure 5B; Figure S8). We observed that T cells clustered in groups in samples from RPT-treated mice (Figure 5C). As cytotoxic CD8+ T cells represent the primary mediators of anti-tumor immunity in this context, we quantified spatial proximity by calculating the number of neighboring cells within a 30 μm radius around each CD8+ GZMB+ cell in the TME (Figure 5D). The resulting data were visualized in heatmaps representing average cellular density per niche. Across both models, the use of RPT was associated with increased aggregation of effector cells, with cytotoxic CD8+ GZMB+ T cells often located adjacent to other activated CD8+ T cells, myofibroblasts, vessel-associated structures, and CD206+ TAMs (Figures 5E and 5F). In SK-N-AS tumors, this pattern shifted slightly, with fewer neighboring endothelial cells but a pronounced increase in adjacent CD4+ T cells, suggestive of an effective, coordinated anti-tumor response. These findings underscore model-specific remodeling of the TME by RPT. In subcutaneous IMR-5 tumors, RPT appears to promote T cell cytotoxic effector aggregates, while in SK-N-AS, the therapy enhances the presence and spatial organization of CD4+ and CD8+ T cell niches, which may synergize through spatial colocalization to reinforce anti-tumor responses.Figure 5Cytotoxic CD8+ cells colocalize with CD4+ T cells to form inflammatory niches in RPT-treated radio-resistant NB(A and B) Enumeration of different immune cell subsets and endothelial cells in (A) IMR-5 tumors and (B) SK-N-AS tumors across therapy groups.(C) Representative images of immune cell niches in the TME with magnified areas highlighting the differences in immune cell representation. Scale bars, 50 μm.(D) Analysis schema to quantify nearest neighbor cell abundance.(E and F) Neighboring cells within a 30 µm radius from CD8+ GZMB+ T cells were counted and represented in a heatmap for (E) IMR-5 tumors and (F) SK-N-AS tumors. Cytotoxic CD8+ cells colocalize with CD4+ T cells to form inflammatory niches in RPT-treated radio-resistant NB (A and B) Enumeration of different immune cell subsets and endothelial cells in (A) IMR-5 tumors and (B) SK-N-AS tumors across therapy groups. (C) Representative images of immune cell niches in the TME with magnified areas highlighting the differences in immune cell representation. Scale bars, 50 μm. (D) Analysis schema to quantify nearest neighbor cell abundance. (E and F) Neighboring cells within a 30 µm radius from CD8+ GZMB+ T cells were counted and represented in a heatmap for (E) IMR-5 tumors and (F) SK-N-AS tumors. Discussion Neither CAR T cells nor RPT alone has achieved durable remissions in solid tumors, particularly in patients with high tumor burden, highlighting the urgent need for more effective strategies.3,4,5,7 In this study, we show that pairing CAR T cells with RPT can eradicate both radio-sensitive and radio-resistant NB in preclinical systems, broadening the potential patient population beyond those traditionally eligible for either modality. Critically, the mechanism supporting this combined effect is fundamentally different depending on the tumor’s intrinsic characteristics, such as the distinct responses to radiation and genetic background. In the radio-sensitive NB model, RPT exerts direct tumoricidal effects on tumor cells and induces a TNF-α-centric inflammatory response that amplifies CAR T cell activation via paracrine pathways, whereas in the radio-resistant model, the same regimen primarily shows lower frequencies of suppressive TAMs and restructures the TME toward pro-inflammatory niches rich in CD4+ and CD8+ effectors. Our RPT agent, a peptidomimetic ligand for VLA-4, binds to various solid tumor and leukocyte cell surfaces.22,28,29,30,31,32,33,34,35,36 Given that repeat biopsies are rarely performed in pediatric patients with relapsed/refractory NB, we envision using the 64Cu-labeled counterpart to noninvasively image VLA-4 using positron emission tomography (PET) and select those most likely to benefit from this treatment.22 On-target, off-tumor hematologic toxicity remains a concern; however, the lymphodepletion routinely administered prior to CAR infusion could be delivered by RPT,22 simultaneously clearing or preventing the emergence of suppressive populations and “priming” the tumor niche for CAR engraftment. In a recent preclinical study, 177Lu-NM600, an alkylphosphocholine mimetic, was combined with a GD2 TRAC CAR, demonstrating synergistic anti-tumor activity in a heterotopic NB model.37 Notably, while 1.8 Gy RPT resulted in complete tumor regression in all treated mice, increasing the dose to 3.6 Gy yielded a complete response in only two-thirds of the animals. Though radiation is known to increase CD8+ T cell expansion,38 the authors attributed this reduced efficacy to direct radiotoxic effects on CAR T cells at the higher RPT level. Consistent with these findings, our study also showed that lower activity of RPT was more effective. Specifically, treatment with 0.5 mCi (18.5 MBq) of [67Cu]Cu-LLP2A resulted in superior therapeutic outcomes compared to 1.0 mCi (37 MBq), underscoring the critical importance of precise dosing and potentially timing when combining RPT with CAR T cell therapy. While no significant toxicities were reported in the aforementioned study, escalating the administered activity led to notable toxicity in NSG mice in our study, likely due to their compromised DNA repair capacity. The observed effects highlight the need for caution when translating this approach into the clinic. In the patient setting, tumor antigen load and the number of circulating leukocytes may significantly influence RPT distribution, acting as competing sinks that affect tumor uptake and overall exposure. Personalized RPT dosimetry guided by the diagnostic isotope analog,33 a concept referred to as “theranostics,” could provide a tailored assessment of radiation delivery, helping to optimize the therapeutic index while also anticipating individual risk profiles for immune-related toxicities such as cytokine release syndrome. The previously referenced study attributed the enhanced efficacy of RPT and CAR T cell co-therapy to the generation of polyfunctional CD8+ GD2 TRAC CAR T cells and upregulation of Fas in irradiated NB cells.37 In contrast, our findings suggest that the mechanism of action varies with tumor radiosensitivity. In radio-sensitive tumors, RPT exerted direct cytotoxicity, remodeled the myeloid compartment, and triggered TNF-α-mediated paracrine activation of T cells. Activation of TNF-α has been previously described in this context in NB but postulated to be an escape mechanism by which tumor cells activate anti-apoptotic proteins via NF-κB and survive the radiation insult.27 Conversely, in radio-resistant models, RPT remodeled the TME toward a T cell-rich environment. These observations indicate that the therapeutic effect of RPT is context dependent, shaped by both tumor-intrinsic properties and TME composition. Understanding these dynamics will be crucial for enhancing therapeutic effects and identifying mechanisms of resistance. While NRAS and TP53 mutations are rare in NB compared to adult cancers, 10%–50% of relapsed NBs harbor alterations in these genes,39,40,41 rendering most of them refractory to DNA-damaging agents.42,43 Even if RPT does not directly kill these tumor cells, reshaping stromal or immune compartments could still create a permissive environment for CAR-mediated killing. Although our models exhibit only intermediate to low VLA-4 expression,22 they demonstrate sufficient ligand engagement to remodel the tumor niche. In alignment with previous work,44,45 we found enriched immune neighborhoods that are induced by radiation. Interestingly, low-dose irradiation of distant sites, such as the gut, has been shown to modulate systemic immunity through microbiota shifts.46 Exploring such “abscopal” effects alongside CAR T cells may yield additional synergy, though with a potential risk of compounding inflammatory toxicities. We selected copper-67 for its favorable half-life and β-emission range (0.1–1.0 mm). To achieve even more localized radiation deposition, future work could test α-emitters, which offer shorter path lengths and higher relative biological effectiveness. Likewise, designing smaller, tumor-specific ligands could widen the therapeutic window in NB. Notably, our ligand has limited ability to cross the blood-brain barrier and may spare neural tissue, thereby potentially minimizing long-term neurotoxicity, but may also leave a sanctuary for relapse.33 In that regard, the inherent trafficking ability of CAR T cells may prove invaluable in clearing occult tumor cells in the central nervous system.47 Limitations of the study The xenogeneic models used in this study limit the ability to conduct long-term evaluations of immune memory and toxicity due to the development of xenogeneic graft-versus-host disease. Future studies could address these limitations by employing syngeneic orthotopic model systems to enable more detailed mechanistic investigations. By weaving together direct radiocidal effects, inflammatory priming, and TME remodeling, our strategy, merging RPT and cell therapy, holds promise for expanding CAR T cell therapy into traditionally refractory solid tumors. Carefully calibrated RPT dosing, patient-specific imaging and personalized dosimetry, and continued exploration of other isotopes and targeting moieties will be essential steps on the path to clinical translation. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Rosa Nguyen (hongharosa.nguyen@nih.gov). Materials availability This study did not generate new unique reagents. Data and code availability •The RNA and single-cell RNA-seq data (GSE330931) were deposited in a public database.•This paper does not report a custom code.•Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request. •The RNA and single-cell RNA-seq data (GSE330931) were deposited in a public database. The RNA and single-cell RNA-seq data (GSE330931) were deposited in a public database. •This paper does not report a custom code. This paper does not report a custom code. •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request. Acknowledgments This work was supported by the 10.13039/100031022Center for Cancer Research (CCR; ZIA BC 012066 to R.N. and ZIA BC 011800 to F.E.E.), 10.13039/100000002National Institutes of Health iCURE program (J.M.R.-G), 10.13039/100000002National Institutes of Health grant R01CA275766 (R.B.P. and R.N.), K08CA241319 (R.B.P.), R01CA204018 (C.J.A.), P30CA047904 (J.R.N. and R.B.P.), and New Jersey Pediatric Hematology Oncology Research Center of Excellence (NJ PHORCE) (X.Z.). We thank the 10.13039/100000054NCI CCR Animal Resource Program/NCI Biological Testing Branch (Dr. M. Custer and K. Divi) for providing mice for our experiments and Dr. A. Mendoza for his assistance with animal regulatory work. Some figures were generated using the Biorender.com software. Author contributions Conceptualization, F.E.E., R.B.P., and R.N.; methodology, C.J.A., F.E.E., R.B.P., and R.N.; investigation, C.R., R.S.E., G.D., J.M.R.-G., W.L., J.B., K.B., D.N., R.O., I.P., Z.S., H.K., Y.S., A.G.C., A.J., A.W., D.E.C., X.Z., R.B.P., and R.N.; visualization, C.R., R.S.E., W.L., X.Z., R.B.P., and R.N.; funding acquisition, F.E.E., R.B.P., and R.N.; supervision, J.R.N., F.E.E., R.B.P., and R.N.; writing – original draft, C.R. and R.N.; writing – review and editing, all authors. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesCD45RA (HI100)BioLegend304146, RRID:AB_2564079CD3 (OKT3)BioLegend317340, RRID:AB_2563408CD45RO (UCHL1)BioLegend304242, RRID:AB_2564159CD62L (DREG-56)BioLegend304830, RRID:AB_2629555CD4 (OKT4)BioLegend317416, RRID:AB_571945CD8 (Rpa-t8)BioLegend301032, RRID:AB_893422G4S (E7O2V)Cell Signaling Technology38907, RRID:AB_3626304CD69 (FN50)BioLegend310931, RRID:AB_2561370LAG3 (11C3C65)BioLegend369320, RRID:AB_2716125CD39 (A1)BioLegend328224, RRID:AB_2564319TIM-3 (F38-2E2)BioLegend345018, RRID:AB_2563859PD-1 (EH12.2H7)BioLegend329904, RRID:AB_940479CD25 (m-a251)BioLegend356137, RRID:AB_2632780CD44 (IM7)BioLegend103030, RRID:AB_830787GD2 (14.G2a)BioLegend357304, RRID:AB_2561885B7-H3 (MIH42)BioLegend351003, RRID:AB_10719959CD56 (W22097A)BioLegend318310, RRID:AB_604106PDL1 (10F.9G2)BioLegend124315, RRID:AB_10897097CD163 (S15049F)BioLegend156707, RRID:AB_2910324CD11B (M1/70)BioLegend101207, RRID:AB_312790F4/80 (T45-2342)BioLegend565613, RRID:AB_2734770LY6C (HK1.4)BioLegend128016, RRID:AB_1732076LY6G (1A8)BioLegend127622, RRID:AB_10643269CD206 (C068C2)BioLegend141715, RRID:AB_2561991MHCII (I-A/I-E)BioLegend107641, RRID:AB_2565975N-Myc Antibody (B8.4.B)Santa CruzSc-53993, RRID:AB_628014TotalSeq-C0046 anti-human CD8BioLegend344753, RRID:AB_2800922TotalSeq-C0072 anti-human CD4BioLegend300567, RRID:AB_2800725Isotype controlBioLegend400113,RRID:AB_326435CD49d (9F10)BioLegend304304,RRID:AB_314430QuantiBRITE beadsBD340495,RRID:AB_2868736Bacterial and virus strainsOne Shot™ Stbl3™ Chemically Competent E. coliThermo Fischer ScientificC737303Biological samplesLenti-X 293TTakara X632180IMR-5NCI POB repositoryN/ASK-N-ASNCI POB repositoryN/AKCNRNCI POB repositoryN/ASH-SY5YNCI POB repositoryN/ASHEPNCI POB repositoryN/ASK-N-FINCI POB repositoryN/ANBSDNCI POB repositoryN/ASK-N-BE2CNCI POB repositoryN/ANGP.GPC2Dr. Sabine Heitzenender (Stanford)N/ASJNBL012407_×1CSTNN/AChemicals, peptides, and recombinant proteinsInterleukin-2Frederick BRB Preclinical RepositoryN/ADynabeads™ Human T-Activator CD3/CD28 for T cell Expansion and ActivationThermo Fisher Scientific11131DDStandard Macrophage Depletion Kit (Clodrosome® + Encapsome®)EncapsulaNanoSicencesCLD-8901InVivoSIM anti-human TNFα (Adalimumab Biosimilar)BioXcellSIM001RecombiMAb human IgG1 isotype control, anti-hen egg lysozymeBioXcellCP174Critical commercial assaysOneGlo assayPromegaE6110CellTiter-GloPromegaG7572CCK8 KitGenBiosystemA311Opal 6-Plex Detection KitAkoya BioscienceNEL871001KTDeposited dataRNA seqThis paperGSE330931Single-cell RNA-seq dataThis paperGSE330931Experimental models: Cell linesSJACT030812_D1CSTNN/AExperimental models: Organisms/strainsNSG PDX modelNCI Animal Resource Program and Charles River LaboratoriesN/ARecombinant DNAGD2 scFv sequenceQVQLQESGPGLVKPSQTLSITCTVSGFSLASYNIHWVRQPPGKGLEWLGVIWAGGSTNYNSALMSRLTISKDNSKNQVFLKMSSLTAADTAVYYCAKRSDDYSWFAYWGQGTLVTVSS GGGGSGGGGSGGGGS ENQMTQSPSSLSASVGDRVTMTCRASSSVSSSYLHWYQQKSGKAPKVWIYSTSNLASGVPSRFSGSGSGTDYTLTISSLQPEDFATYYCQQYSGYPITFGQGTKVEIKRN/AMGA271 scFv sequenceDTEVQLVESGGGLVQPGGSLRLSCAASGFTFSSFGMHWVRQAPGKGLEWVAYISSDSSAIYYADTVKGRFTISRDNAKNSLYLQMNSLRDEDTAVYYCGRGRENIYYGSRLDYWGQGTTVTVSSGGGGSGGGGSGGGGSDIQLTQSPSFLSASVGDRVTITCKASQNVDTNVAWYQQKPGKAPKALIYSASYRYSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQYNNYPFTFGQGTKLEIKN/ApsPAX2Addgene12260, RRID:Addgene_12260pMD2.GAddgene12259, RRID:Addgene_12259R980-M19-663Frederick National Laboratory for Cancer Research, Frederick, MDN/ASoftware and algorithmsFlowJo softwareversion 10.10.0Living Image softwarePerkinElmerversion 4.3.1GraphPad Prismversion 10InForm softwareAkoya Bioscienceversion 3.1.0OtherLSRFortessaBD BiosciencesN/AIVIS Lumina III SystemPerkinElmerN/A REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesCD45RA (HI100)BioLegend304146, RRID:AB_2564079CD3 (OKT3)BioLegend317340, RRID:AB_2563408CD45RO (UCHL1)BioLegend304242, RRID:AB_2564159CD62L (DREG-56)BioLegend304830, RRID:AB_2629555CD4 (OKT4)BioLegend317416, RRID:AB_571945CD8 (Rpa-t8)BioLegend301032, RRID:AB_893422G4S (E7O2V)Cell Signaling Technology38907, RRID:AB_3626304CD69 (FN50)BioLegend310931, RRID:AB_2561370LAG3 (11C3C65)BioLegend369320, RRID:AB_2716125CD39 (A1)BioLegend328224, RRID:AB_2564319TIM-3 (F38-2E2)BioLegend345018, RRID:AB_2563859PD-1 (EH12.2H7)BioLegend329904, RRID:AB_940479CD25 (m-a251)BioLegend356137, RRID:AB_2632780CD44 (IM7)BioLegend103030, RRID:AB_830787GD2 (14.G2a)BioLegend357304, RRID:AB_2561885B7-H3 (MIH42)BioLegend351003, RRID:AB_10719959CD56 (W22097A)BioLegend318310, RRID:AB_604106PDL1 (10F.9G2)BioLegend124315, RRID:AB_10897097CD163 (S15049F)BioLegend156707, RRID:AB_2910324CD11B (M1/70)BioLegend101207, RRID:AB_312790F4/80 (T45-2342)BioLegend565613, RRID:AB_2734770LY6C (HK1.4)BioLegend128016, RRID:AB_1732076LY6G (1A8)BioLegend127622, RRID:AB_10643269CD206 (C068C2)BioLegend141715, RRID:AB_2561991MHCII (I-A/I-E)BioLegend107641, RRID:AB_2565975N-Myc Antibody (B8.4.B)Santa CruzSc-53993, RRID:AB_628014TotalSeq-C0046 anti-human CD8BioLegend344753, RRID:AB_2800922TotalSeq-C0072 anti-human CD4BioLegend300567, RRID:AB_2800725Isotype controlBioLegend400113,RRID:AB_326435CD49d (9F10)BioLegend304304,RRID:AB_314430QuantiBRITE beadsBD340495,RRID:AB_2868736Bacterial and virus strainsOne Shot™ Stbl3™ Chemically Competent E. coliThermo Fischer ScientificC737303Biological samplesLenti-X 293TTakara X632180IMR-5NCI POB repositoryN/ASK-N-ASNCI POB repositoryN/AKCNRNCI POB repositoryN/ASH-SY5YNCI POB repositoryN/ASHEPNCI POB repositoryN/ASK-N-FINCI POB repositoryN/ANBSDNCI POB repositoryN/ASK-N-BE2CNCI POB repositoryN/ANGP.GPC2Dr. Sabine Heitzenender (Stanford)N/ASJNBL012407_×1CSTNN/AChemicals, peptides, and recombinant proteinsInterleukin-2Frederick BRB Preclinical RepositoryN/ADynabeads™ Human T-Activator CD3/CD28 for T cell Expansion and ActivationThermo Fisher Scientific11131DDStandard Macrophage Depletion Kit (Clodrosome® + Encapsome®)EncapsulaNanoSicencesCLD-8901InVivoSIM anti-human TNFα (Adalimumab Biosimilar)BioXcellSIM001RecombiMAb human IgG1 isotype control, anti-hen egg lysozymeBioXcellCP174Critical commercial assaysOneGlo assayPromegaE6110CellTiter-GloPromegaG7572CCK8 KitGenBiosystemA311Opal 6-Plex Detection KitAkoya BioscienceNEL871001KTDeposited dataRNA seqThis paperGSE330931Single-cell RNA-seq dataThis paperGSE330931Experimental models: Cell linesSJACT030812_D1CSTNN/AExperimental models: Organisms/strainsNSG PDX modelNCI Animal Resource Program and Charles River LaboratoriesN/ARecombinant DNAGD2 scFv sequenceQVQLQESGPGLVKPSQTLSITCTVSGFSLASYNIHWVRQPPGKGLEWLGVIWAGGSTNYNSALMSRLTISKDNSKNQVFLKMSSLTAADTAVYYCAKRSDDYSWFAYWGQGTLVTVSS GGGGSGGGGSGGGGS ENQMTQSPSSLSASVGDRVTMTCRASSSVSSSYLHWYQQKSGKAPKVWIYSTSNLASGVPSRFSGSGSGTDYTLTISSLQPEDFATYYCQQYSGYPITFGQGTKVEIKRN/AMGA271 scFv sequenceDTEVQLVESGGGLVQPGGSLRLSCAASGFTFSSFGMHWVRQAPGKGLEWVAYISSDSSAIYYADTVKGRFTISRDNAKNSLYLQMNSLRDEDTAVYYCGRGRENIYYGSRLDYWGQGTTVTVSSGGGGSGGGGSGGGGSDIQLTQSPSFLSASVGDRVTITCKASQNVDTNVAWYQQKPGKAPKALIYSASYRYSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQYNNYPFTFGQGTKLEIKN/ApsPAX2Addgene12260, RRID:Addgene_12260pMD2.GAddgene12259, RRID:Addgene_12259R980-M19-663Frederick National Laboratory for Cancer Research, Frederick, MDN/ASoftware and algorithmsFlowJo softwareversion 10.10.0Living Image softwarePerkinElmerversion 4.3.1GraphPad Prismversion 10InForm softwareAkoya Bioscienceversion 3.1.0OtherLSRFortessaBD BiosciencesN/AIVIS Lumina III SystemPerkinElmerN/A Experimental model and study participant details Mouse models All in vivo procedures were performed under protocols approved by the institutional IACUC (POB-003-2). NSG mice (4–6 weeks old, both sexes) were inoculated subcutaneously with 1 × 106 IMR-5 cells or orthotopically in the para-adrenal fat pad with 5 × 105 SK-N-AS.luc.eGFP or 1.5 × 106 IMR-5.luc.eGFP cells.48 For myeloid depletion experiments, 200 μL/mouse of clodrosome or empty liposome (control) was administered by intraperitoneal injection on Day −2 and Day 2, around the adoptive T cell transfer. For in vivo TNFα blockade, NSG mice bearing heterotopically implanted IMR-5 tumors in the right hindlimb received localized limited-field EBRT (1.5 Gy), followed 3 days later by intraperitoneal injection of 250 μg of adalimumab or an isotype control. After an additional day, 2 × 106 GD2+ CAR T-cells or matched numbers of untransduced mock T-cells were administered intravenously. Tumor growth was monitored, and animals were randomized by tumor volume or IVIS (in vivo imaging system) bioluminiscense signal prior to treatment. Mice received either cold LLP2A or [67Cu]Cu-LLP2A. Five days later, CAR T-cells or untransduced T-cells were administered intravenously. Tumor progression was tracked biweekly using calipers or weekly by IVIS imaging via 3 mg of D-Luciferin via intraperitoneal injection. Animals were removed from the study upon disease progression or when meeting humane endpoint criteria. In vitro models IMR-5, SK-N-AS, KCNR, SH-SY5Y, SHEP, SK-N-FI, NBSD, SK-N-BE2C (all NCI POB repository), and NGP.GPC2 (kindly provided by Dr. Sabine Heitzenender, Stanford University) were grown in RPMI media (Lonza) supplemented with 10% fetal bovine serum (FBS) and 100 U/mL penicillin/streptomycin (all Gibco). Mycoplasma-free status was confirmed (ATCC Universal Mycoplasma Detection Kit). Short tandem repeat analysis was used for identity testing. IMR-5 and SK-N-AS were transduced to stably express luciferase-green fluorescent protein (GFP). Method details CAR constructs and CAR T cell manufacturing CAR T cell constructs specific to B7-H349 and GD2 antigens3,50 were generated using previously validated sequences. Peripheral blood mononuclear cells from healthy adult donors (NIH Blood Bank) were activated with anti-CD3/CD28 Dynabeads (1:1 ratio) in the presence of IL-2 for 48 h before viral transduction at a multiplicity of infection of 5. Cultures were expanded until Day 8–10, when they were harvested for downstream applications48 (Figure S9). CD4, CD8, and CAR percentages were determined in the manufacturing product prior to experimentation. In vitro cell viability assay with RPT To assess direct tumor cell killing, NB cell lines were incubated with free copper-67 at doses of 20 μCi for up to seven days. Cell viability was quantified on Days 1, 3, and 7 using the Cell Counting Kit-8, following the manufacturer’s instructions. In vitro CAR T cell kill assay To test the combination of CAR T cell and low-dose radiation effects in vitro, luciferase-tagged NB cells were pre-exposed to 20 free μCi copper-67. After three days, surviving cells were collected, replated, and co-cultured with either CAR T-cells or untransduced T cell controls at low effector-to-target (E:T) ratios. Viability was assessed at 24, 48, and 72 h using the OneGlo luminescence assay (Promega), and luminescence was determined using a microplate reader. In vitro 24-h T cell activation assay NB cells were first treated with a TNFɑ-targeting or control siRNA. Cells were subsequently irradiated with 0.5 Gy (50 Rad) and cultured for 72 h. Conditioned media from these cultures were used to activate T-cells in vitro for 24 h. The activation status was evaluated by flow cytometric analysis of CD25 and CD69 expression. Labeling of CB-TE1A1P-PEG4-LLP2A with copper-67 (0.35 mCi/μg) LLP2A conjugated to 1,4,8,11-tetraazacyclotetradecane-1-(methane phosphonic acid)-8-(methane carboxylic acid)-(ethylene glycol)-4 [CB-TE1A1P-PEG4-LLP2A] was synthesized by a commercial vendor. 67CuCl2 was obtained from the Idaho Accelerator Center (Idaho State University) and National Isotope Development Center. 67CuCl2 (5–10 μL in 0.5 M HCl) was diluted with 0.5 M ammonium acetate buffer (pH 4, 100 μL). The CB-TE1A1P-PEG4-LLP2A solution was mixed with 67Cu-acetate in 0.5 M ammonium acetate buffer (pH 6.5 for CB-TE1A1P). The mixture was incubated at 70°C for 30 min and subsequently tested for purity by radio-HPLC with a Jupiter C18 column (300 Å pore size, 5 μm particle size, 150 × 4.6 mm (Figures S10A and S10B). Biodistribution/dosimetry About 72 h post [67Cu]Cu-LLP2A intravenous injection, blood, heart, lungs, thymus, muscle, femur, pancreas, spleen, kidneys, liver, stomach, small intestine without content, brain, skin, and tumor were harvested, weighed, and counted on a gamma counter. The percentage of injected activity per gram of tissue (%IA/g) was calculated by normalizing the data to the total activity injected into each corresponding mouse. Regions of interest (ROIs) of both normal and tumor tissues were segmented from PET/CT images using the VivoQuant software. Monte Carlo dosimetry calculations were performed using Torch. Uptake in each organ was expressed as absorbed dose per unit administered activity (Gy/MBq). Flow cytometry IMR-5 or SK-N-AS cells (5 × 105) were stained using GD2, B7-H3, CD49d (VLA-4), or isotype control for 12 min at room temperature and subsequently washed twice before analysis. PE-QuantiBRITE beads were used to determine the antigen molecules per cell. Spleen or tumor cells (1–3 × 106) were stained with Zombie Live/Dead dye for 12 min at room temperature and subsequently washed, incubated with the antibody panel for 12 min, and washed again twice. Gates were drawn using fluorescence-minus-one (FMO) controls, and compensations and voltages were determined using single-color controls. Gating strategies are shown in Figures S11 and S12. Data were collected on an LSR Fortessa flow cytometer (BD Biosciences) and analyzed with FlowJo v10. Multiplex immune fluorescence analysis Formalin-fixed paraffin-embedded (FFPE) tissue sections were cut at 4-micron thickness and mounted on charged slides. Tissues were incubated at 60°C for 2.5 h, deparaffinized using xylene, and rehydrated with ethanol prior to fixing with neutral buffered formalin for 15 min. Akoya Bioscience’s Opal 6-Plex Detection Kit was used according to the manufacturer’s instructions for sequential staining of each marker in each panel. Following Opal detection, fluorophore combinations were used for the immunophenotyping panel: CTLA-4 (1:100) Opal 570 (1:150), CD8 (1:150) Opal 520 (1:150), CD4 (1:150) Opal 690 (1:150), Granzyme B (1:200) Opal 620 (1:150), a-SMA (1:500) Opal 480 (1:150), and CD206 (1:500) Opal 780 (1:18). Finally, cells were counterstained with DAPI for 3 min, mounted with ProLong anti-fade mountant, and stored at 4°C in the dark until imaging. Akoya Bioscience’s PhenoimagerHT 2.0 platform was used for whole-slide scanning. Image analysis was performed using the InForm software. First, a machine learning algorithm was trained, using selected regions of interest (ROIs) from all samples to segment tissues and cells, enabling the identification of individual cells and quantification of marker expression on a per-cell and per-cell-compartment basis. Batch analysis was applied to whole-slide images, previously annotated on Phenochart 2.2.0, and the resulting data were exported to R. The open-source package phenoptrReports was used to extract phenotyping data, calculate cell densities, and perform spatial analyses based on the outputs generated in InForm 3.1.0. For validation of the spatial results obtained through the InForm software, we performed an independent analysis using the outputs of the InForm tissue segmentation algorithm. Images were reconstructed from the individual ROIs in Python using the Pillow Image module (PIL fork) and converted to a binary mask. Using the phenotype data from the output file, the positions of cells were classified as part of tumor tissue as point patterns in an x-y coordinate system to visualize their distribution in space. Point patterns were plotted in R with the ggplot2 package. Western blot analysis Whole cell proteins were extracted using RIPA buffer supplemented with protease/phosphatase inhibitor following the manufacturer’s protocol. Protein concentrations were measured using the BCA protein assay. 10 μg protein samples with Laemmli sample buffer were separated by electrophoresis on 4–20% SDS/PAGE gels and transferred to a PVDF transfer membrane. After blocking in 5% non-fat dry milk at room temperature, the membranes were incubated with primary antibody overnight at 4°C and subsequently with secondary antibodies for 1 h at room temperature. Cytokine detection For cytokine detection, murine or human cytokine/chemokine bead panels (BioLegend) were used as per the manufacturer’s instruction and quantified by flow cytometry. RNA-seq analysis Total RNA was isolated from in vitro samples and subjected to whole transcriptomic profiling. RNA extraction was done using the RNeasy Mini Kit, following the manufacturer’s direction. RNA quantity and quality were determined using the Agilent 4200 TapeStation system at the Genomics Core of the Center for Cancer Research at the NIH. RNA samples with an RNA integrity score higher than 9 were used for library construction with the TruSeq Stranded Library Preparation Kit and sequenced on a NovaSeq X Plus sequencer at the Sequencing Core Facility at the Frederick National Laboratory for Cancer Research. Sequencing adaptors and low-quality bases were trimmed before the reads were aligned to the reference genome (human-GRCh38) and then annotated using STAR.51 Library complexity was measured using Picard’s MarkDuplicate function. Next, gene expression levels were determined using RSEM,52 and the raw counts were used in DESeq2 for differentially expressed gene analysis.53 Single-cell RNA-seq analysis Fresh tumor samples were harvested from mice treated with or without RPT in combination with mock or B7-H3 CAR T-cells. Tumor samples were processed into single-cell suspensions and stained with TotalSeq-C0046 anti-human CD8, and TotalSeq-C0072 anti-human CD4 antibodies. cDNA libraries were prepared using the 10× Genomics Chromium 5′ v3 chemistry with Feature Barcode Technology targeting 12,000 cells per sample. Sequencing was performed on an Illumina NovaSeq X Plus sequencer with a target depth of ∼50,000 to 100,000 reads per cell. Computational analysis of scRNA-seq data Raw reads were demultiplexed using DRAGEN BCL Convert v4.1.23 and aligned to a combined reference (GRCh38 + mm10 + custom CAR sequence) using CellRanger v8.0.1 (STAR 2.7.2a) for further analysis in Seurat (v5.1.0, R package).54,55 SoupX was used to remove ambient RNA contamination.56 Doublets were removed with Scrublet.54 Low-quality cells (<300 genes detected, >7500 genes detected, >70% of reads mapped to mitochondrial genes, <500 unique molecular identifiers (UMI) detected, or >50000 UMI detected) were excluded. Data integration across different treatment groups was performed with the reciprocal principal component analysis (RPCA) in Seurat.57 The data was normalized with the ‘NormalizeData’ function, applying the following parameters: normalization.method = “LogNormalize” and scale.factor = 10000. The ‘FindVariableFeatures’ function with ‘vst’ method was used to identify highly variable genes. Additionally, reads of feature barcode sequences were added and used to normalize the dataset following Seurat’s multimodal analysis vignette. Gene expression matrices were scaled and centered with the ‘ScaleData’ function, applying default parameters. The ‘RunPCA’ function was applied for PCA dimensionality reduction. Using the ‘FindNeighbors’ function, we created the shared nearest neighbor (SNN) graph. The ‘FindClusters’ function with a resolution of 0.8 was applied for cluster mapping. Clusters were manually annotated by evaluating highly enriched genes identified with the ‘FindAllMarkers’ function. PCA was applied for data visualization. Differential gene expression was calculated for all clusters. Receptor-ligand interactions were inferred using the cellphoneDB software v258 in human an murine cells, applying a statistical analysis and threshold of 0.4. Quantification and statistical analysis Shapiro-Wilk tests determined data distribution. For normally distributed data, one-way ANOVA followed by Tukey’s post hoc test was used; for non-parametric data, Kruskal-Wallis and Dunn’s post hoc tests were applied. Experiments were repeated with at least two independent donors and models. Significance was defined at p < 0.05. Graphs and statistical tests were produced using GraphPad Prism 10.4.2.