Anti-GD2 Antibodies Conjugated to IL15 and IL21 Mediate Potent Antitumor Cytotoxicity against Neuroblastoma

mAb therapy targeting GD2 is part of the standard of care for patients with high-risk neuroblastoma (1). However, 50% of treated patients still do not achieve long-term remissions (2). To enhance antitumor efficacy, an antibody–cytokine conjugate (immunocytokine, IC) was developed by linking IL2 to the COOH terminus of a humanized anti-GD2 antibody (hu14.18-IL2; refs. 3, 4). Although hu14.18-IL2 eradicated minimal residual disease in mice and patients, those with bulky tumors did not respond, highlighting the limitations of this therapy (5–7).

Soluble IL2 was originally part of the immunotherapy regimen to treat neuroblastoma but has been removed because its therapeutic benefit was called into question owing to severe toxicities associated with higher doses and unestablished efficacy at lower doses (8). IL2 belongs to the γ_c family of cytokines and induces pro-immune effects by engaging the IL2 receptor γ (IL2Rγ) and cytokine-specific receptor subunits (IL2Rα and IL2/15Rβ), which are expressed on various effector cells of antibody-dependent cell-mediated cytotoxicity (ADCC; ref. 9). IL15 and IL21, other members of this family, have overlapping biologic effects with IL2, but their therapeutic use in neuroblastoma has been underappreciated until recently. For example, we and others have shown that IL15/IL15Rα complexes enhance ADCC against neuroblastoma to a greater extent than soluble IL2 (10, 11). However, it is unknown whether ICs linked with IL15 and IL21 can effectively target neuroblastoma and how their activity compares with hu14.18-IL2.

In this study, we found that hu14.18-IL15 induced equivalent ADCC against GD2-expressing neuroblastoma as hu14.18-IL2 in vitro with human NK cells serving as effector cells. Equivalency was also observed in T-cell–depleted patient-derived xenograft (PDX) models in vivo with NK cells as primary effector cells. However, in fully immunocompetent syngeneic mouse models with orthotopic neuroblastoma, hu14.18-IL15 and -IL21 outperformed hu14.18-IL2. Single-cell RNA sequencing (scRNA-seq) data revealed significant differences in the remodeling of the tumor microenvironment (TME), potentially overcoming immunosuppressive mechanisms. In summary, this work demonstrates the importance of IL15 and IL21 γ_c family of cytokines as adjunct agents for mAb therapy in neuroblastoma.

NXS-2-GD2⁺ (Mycn-wild-type, WT) was the kind gift of Dr. Paul Sondel (University of Wisconsin). The PDXs SJNBL012407_X1 and SJNBL046_X (both MYCN-amplified) were provided by the Children's Solid Tumor Network. IMR-5, IMR-32, KCNR, SKNBE2C (all MYCN-amplified), SKNFI, SH-SY5Y, and SKNAS (latter three MYCN-WT) were retrieved from the laboratory repository. All cells tested negative for Mycoplasma (Mycoalert, Lonza) before use. Their identity was confirmed by short-tandem repeat DNA profiling. Thawed cells were used within 3 months.

Stable luciferase (ffLUC)–GFP-expressing cells were generated by lentiviral transduction. Neuroblastoma cell lines were selected with 0.5 μg/mL of puromycin (Thermo Fisher Scientific) and cryopreserved with >60% enrichment for transduced cells. PDX cells were transduced with freshly prepared concentrated lentivirus particles and grown for 24 hours in full RPMI medium and on Matrigel-coated plates (Corning Inc.) before injection into mice. After one passage in mice, cells were cryopreserved.

Fresh human NK cells were isolated from buffy coats of healthy volunteer donors (NIH Blood Bank) via magnetic bead negative selection (Stem Cell Technology). >95% purity of CD45⁺ CD56⁺ CD3^– NK cells was deemed acceptable for further experiments (Supplementary Fig. S1).

The humanized anti-GD2 antibody hu14.18, an IgG1 isotype antibody, which contains a WT F_c, was conjugated at the COOH terminus via a (G₄S)₃ linker to human IL15 (cross-reactive to mouse) or murine IL21 (not cross-reactive to human) to obtain research-grade protein for experimental use (Creative Biolabs; Supplementary Fig. S2). Hu14.18-IL2 (human IL2, which is cross-reactive to mouse) was obtained from the Biological Resources Branch at the Frederick National Laboratory (BRB Frederick). We confirmed intact conjugation of the antibody and respective cytokines using a sandwich ELISA assay with the anti-idiotypic antibody 1A7 (BRB Frederick) as capture antibody as previously described (12). We also confirmed detection of antibody and cytokines in the tumors and serum at 48 hours (Supplementary Fig. S3).

Freshly isolated human NK cells were incubated in descending dilutions of ICs or soluble human IL15 or IL2 (BRB, Frederick). To quantify the level of cellular activation, CD69 expression was measured by flow cytometry after 24 hours of co-culture.

NK cells and ffLUC-expressing tumor cells were co-cultured in the presence or absence of ICs (1 μg/mL) at varying effector-to-tumor (E:T) ratios as previously described (13). After 24 hours, the specific lysis of the tumor cells was measured using the ONE-Glo Luciferase Assay System (Promega).

One day before the assay, IMR-5-ffLUC-GFP cells were added to flat-bottom 96-well plates at a density of 10,000 cells per well. The next day, human NK cells and ICs (1 μg/mL) were added at an E:T ratio of 5:1. Cell confluence was monitored every four hours using the Incucyte system (Sartorius) by recording the area emitting fluorescent GFP signal. Every 24 hours thereafter, 10,000 IMR-5-ffLUC-GFP cells were added to each well to rechallenge the NK cells. The experiment was terminated when controls reached confluence.

For neuroblastoma PDX studies, 4- to 6-week-old female athymic nude mice were obtained from the NCI CCR Animal Resource Program/NCI Biological Testing Branch. For syngeneic studies, 4- to 6-week-old female A/J mice were purchased from The Jackson Laboratory. All animal studies were approved by the Institutional Animal Care and Use Committee of the National Cancer Institute (protocol number: PB-023).

Tumor-bearing mice were injected with d-luciferin potassium salt (150 mg/kg, intraperitoneal), anesthetized with isoflurane, and imaged on an IVIS Lumina XR System (PerkinElmer) 5 minutes after d-luciferin injection (1 minute acquisition time). Region of interest analysis was performed using the Living Image software (PerkinElmer; version 4.3.1).

Athymic nude mice lack T cells and have decreased levels but functional monocytes and NK cells, thus enabling the engraftment of PDX tumors and the study of ADCC as previously shown (10). In this model, chemotherapy comprised intraperitoneal injections of irinotecan (1.6 mg/kg) and oral temozolomide (16.5 mg/kg) for 5 days. Immunotherapy was given to respective mice one week after chemotherapy as five daily intravenous injections (50 μg per dose) as previously described (10). Mice received two treatment courses with each course lasting 3 weeks (third week of rest), emulating clinical regimens (14, 15). First, PDX cells were implanted orthotopically (2.5 × 10⁵) into athymic nude mice (16). Typically, three weeks after tumor implantation surgery, animals meeting enrollment criteria with a bioluminescence imaging (BLI; >10⁷), were randomized to receive either (i) vehicle, (ii) daily chemotherapy, (iii) chemotherapy followed by hu14.18-IL2, (iv) chemotherapy followed by hu14.18-IL15, or (v) chemotherapy followed by hu14.18-IL21. The experiment was terminated on day 65 after tumor injection. BLI signals did not correlate with tumor burden in the PDX model, particularly in mice with large tumors. Therefore, we recorded tumor weights (primary endpoint) and the occurrence of metastases (secondary endpoint) to determine the efficacy of the treatments.

A/J mice underwent orthotopic injection of NXS-2-ffLUC-GFP cells (5 × 10⁵; ref. 3). Eight days later, when they established a defined BLI signal (>10⁶), mice received daily injections of ICs (25 μg per dose) for 5 consecutive days intravenous. The tumor weight was determined on day 30 after tumor implantation or survival of treated mice was monitored until day 100. Survival endpoints were death, >20% weight loss from baseline, or severe moribund status.

Samples were stained using 10⁶ cells. Gates were drawn with fluorescence-minus-one controls. Compensations and voltages were set with single-color controls. The following antibodies were used for detection of surface epitopes: CD45 (clone 30-F11), CD3 (OKT3), CD4 (RM4–4), CD8 (53–6.7), F4/80 (BM8), CD25 (3C7), CD49b (DX5), CD45 (HI30), CD69 (FN50), CD56 (5.1H11), TIGIT (A15153G), and GD2 (14.G2a; all BioLegend). Intracellular molecules were detected with Foxp3 (MF14), IFNγ (XMG1.2), and Perforin (S16009A) antibody. The PE Phycoerythrin Fluorescence Quantitation Kit (BD Biosciences) was used to determine the density of GD2 expression in tumor cells. Data were collected on a Fortessa LSR machine and analyzed in FlowJo v10.

Cytokine bead assays were conducted to quantitate the secreted cytokines in the supernatant of co-cultures following the manufacturer's instructions (BioLegend).

A/J mice were injected orthotopically with NXS-2 cells (5 × 10⁵). Three weeks later, mice were randomly stratified into four different groups to receive (i) Vehicle, (ii) hu14.18-IL2, (iii) hu14.18-IL15, (iv) hu14.18-IL21. ICs were administered intravenously as one dose of 25 μg. About 48 hours later, tumors were processed into single-cell suspensions and enriched via CD45 magnetic bead–positive selection (Stem Cell Technologies). The viability of samples was confirmed to be >80% before further processing. Three individual mouse samples per treatment group were pooled. About 10,000 cells from each treatment group were loaded with the goal of capturing 6,000 cells using the 10X Genomics 3′ v3.1 chemistry. We followed the manufacturer's instructions for library generation. The cDNA libraries were sequenced on the NextSeq2000 with a target depth of approximately 50,000 reads per cell.

The 10X Genomics “cellranger” pipeline was applied to align reads to the mouse mm10 genome with default settings and to generate a single-cell gene expression matrix. Ambient RNA was removed using the SoupX algorithm by setting the contamination rate at 30% (17). Doublets were detected by the scrublet package and removed for the downstream analysis in Seurat (v4.0.1; refs. 18, 19). Cells with low (<50 genes) and aberrantly high numbers of genes (>7,000 genes) were removed. Cells that contained more than 10% mitochondria genes were removed. Sample integration across the treatment groups was performed using the standard anchor-based workflow in Seurat. For initial clustering and annotation, we applied the KNN graph based on the weighted RNA similarities, to calculate the Jaccard index (neighborhood overlap) between every pair of cells with a high resolution and merging of clusters. Results were presented on Uniform Manifold Approximation and Projection plots for visualization using the Seurat package. Tumor cells were identified by the green fluorescent protein sequence. CD45⁺ immune cells were annotated using canonical gene markers. Lymphoid and myeloid cells were separated and reclustered to obtain more refined cell clusters. Differential gene expression was calculated for all pairs of clusters and therapy groups.

Two groups were compared using the Student t test (normally distributed data) or Mann–Whitney U test (skewed data). For experiments with more than two groups, the statistical significance was assessed using a one-way ANOVA (normally distributed data) followed by Tukey post hoc comparison tests or a one-way ANOVA on ranks (skewed data) followed by the Dunn test. For survival analysis, Kaplan–Meier curves were generated, and a two-sided log-rank test was performed to examine the differences between survival curves with significance level adjusted for multiple comparisons with the Bonferroni method. All experiments were repeated at least once.

Materials and data are available from the corresponding author upon reasonable written request. The scRNA-seq data were deposited in GEO (GSE206628).

To measure the activating effects of the ICs on human NK cells, we co-cultured NK cells with descending concentrations of hu14.18-IL15 and hu14.18-IL2. Soluble IL2 and IL15 served as positive controls, supplementing the media at concentrations commonly used to expand NK cells in culture (13, 20). After 24 hours, we analyzed the cells for the expression of CD69 as an activation marker. Both ICs were able to activate NK cells in a dose-dependent manner compared with controls (Fig. 1A). Slightly higher CD69 levels were noted with hu14.18-IL2 than -IL15. The IC-mediated activation was accompanied by secretion of the effector molecules perforin and granzyme A into the culture supernatant at 48 hours (Fig. 1B).

We went on to evaluate ADCC against the GD2^high neuroblastoma cell lines, IMR-32, IMR-5, and KCNR. The cytotoxic capacity of NK cells was significantly increased at various E:T ratios by the addition of either one of the ICs (Fig. 1C). We then extended our testing to six neuroblastoma cell lines with GD2^high or GD2^low expression (Fig. 1D). ICs increased NK cell-mediated tumor lysis of GD2^high cell lines, whereas lytic levels of GD2^low neuroblastoma were similar with or without ICs (Fig. 1E). These findings demonstrate that hu14.18-IL15 and hu14.18-IL2 both activate NK cells and significantly enhance their cytotoxic potential against GD2^high but not GD2^low neuroblastoma in vitro.

To assess NK cell function during continuous tumor exposure, we conducted a tumor rechallenge experiment. Human NK cells were co-cultured with GFP-expressing IMR-5 cells ± hu14.18 or ICs at an E:T ratio of 5:1. IMR-5 is very suitable for this imaging-based assay, given that the morphology of the cell line is flat and the GFP signal strong (Fig. 2A). Every 24 hours thereafter, the starting number of tumor cells was added to each well, and cytolysis was monitored by quantifying the GFP⁺ area. The efficacy by which NK cells eliminated tumor diverged after the second tumor rechallenge (Fig. 2B). Tumor confluence continued to increase with NK cells alone, whereas hu14.18 allowed NK cells to restrict tumor growth but rendered them unable to eliminate tumor cells. NK cells harnessed with ICs effectively lysed tumor cells with each consecutive rechallenge. After the fifth tumor rechallenge, residual NK cells had attained a predominantly CD56^bright status and expressed high levels of CD69, suggestive of NK cell activation (Fig. 2C). In contrast, hu14.18 without human cytokines was associated with a CD56^dim phenotype. The expression of TIGIT, a marker of functional NK cell exhaustion, was higher in hu14.18-IL2– than hu14.18-IL15–treated cells and altogether elevated when compared with hu14.18 and NK cells alone. The supernatant after the final rechallenge showed significantly higher levels of granulysin, granzyme B, perforin, sFASL, and IFNγ with hu14.18-IL2 than hu14.18-IL15. NK cells expanded in the presence of IL2-conjugated antibodies (Fig. 2D and E). Our findings demonstrate that both ICs can maintain the cytolytic capacity of NK cells against neuroblastoma. Hu14.18-IL2 led to more robust NK cell expansion and cytokine secretion compared with hu14.18-IL15 but upregulated TIGIT, suggestive of early signs of functional exhaustion.

We have previously demonstrated that athymic nude mice can engraft with orthotopic neuroblastoma PDXs and serve as a preclinical model for chemo-immunotherapy testing (10). Like our previous experience, we noted again that ICs alone do not induce ADCC in mice with bulky tumor (Supplementary Fig. S4), consistent with clinical observations that ICs have better activity in patients with minimal residual disease than bulky tumor (5, 21). Hence, we compared the antitumor activity of the ICs combined with chemotherapy to chemotherapy with irinotecan and temozolomide alone (Fig. 3A) using two MYCN-amplified GD2^high PDXs (Fig. 3B). ICs combined with chemotherapy mediated complete tumor regression in two thirds of mice engrafted with SJNBL012407_X1 (Fig. 3C–F). In mice bearing SJNBL046_X, all except for one mouse in the hu14.18-IL15 arm had residual tumor (Fig. 3G–I). Mice receiving chemotherapy with ICs had less residual tumor burden in both models than animals in the chemotherapy arm. No differences in efficacy were noted comparing the three ICs. Most treated tumors at the end of the study retained high levels of GD2 expression (Fig. 3F and I), potentially rendering them sensitive to additional courses with chemotherapy and ICs. Altogether, these findings mirror the results from our in vitro studies, suggesting that the newly developed ICs have equivalent ADCC as hu14.18-IL2 in T-cell–deficient systems where NK cells serve as the sole effector cell population.

To compare the efficacy of all three ICs in an immunocompetent neuroblastoma mouse model, we implanted MYCN-WT GD2^high NXS-2-ffLUC cells into A/J mice and started treatment with ICs when animals established a defined tumor signal in the para-adrenal region on BLI (Fig. 4A and B). Using longitudinal BLI, we found that 3/7 (43%) mice treated with hu14.18-IL2 had a decline in BLI signals to near background levels by day 30 post tumor implantation compared with 5/7 (71%) treated with hu14.18-IL15 and 7/8 (88%) treated with hu14.18-IL21 (Fig. 4C and D). Corresponding tumor weights were also significantly smaller after treatment with the new ICs compared with vehicle or hu14.18-IL2 (Fig. 4E). As noted in our previous experiments (Fig. 3F and I), GD2 expression again remained high in residual tumor cells, suggesting that animals may benefit from additional doses of ICs (Fig. 4F). Furthermore, we found extended survival in animals that had received hu14.18-IL15 (Fig. 4G and H). Among the nine animals that survived beyond day 100, four were treated with hu14.18-IL15 and remained tumor-free. Animals that died before day 100 succumbed due to large tumor burden or small tumors invading the adjacent aorta leading to bleeding and death (Fig. 4I). Our findings demonstrate that hu14.18-IL15 and -IL21 outperformed hu14.18-IL2 in an immunocompetent host by controlling tumor growth within an early time window after IC administration, leading to long-term survival beyond 100 days in treated mice. Our results suggested that immune cells other than NK cells may influence the therapeutic efficacy of ICs. Thus, to further elucidate the contribution of tumor-infiltrating immune cells, we next sought to analyze the TME using multi-color flow cytometry and scRNA-seq.

We have shown that we can detect tumor-bound ICs at 48 hours after intravenous injection of one dose (Supplementary Fig. S3). We, therefore, analyzed immune infiltrates in NXS-2 tumors of treated A/J mice 48 hours after one intravenous dose of IC using multicolor flow cytometry. The largest fraction of tumor-infiltrating immune cells comprised F4/80⁺ macrophages (74%) followed by NK (9%), CD4⁺ T (5%), and CD8⁺ T cells (4%; Supplementary Fig. S5A). CD8⁺ T cells exhibited 1.6-fold higher levels of perforin after hu14.18-IL21 treatment compared with no treatment, and a similar trend was noted in NK cells after administration of hu14.18-IL2 (Supplementary Fig. S5B). Most strikingly, the fraction (>1.7-fold) of Tregs was significantly elevated in animals treated with hu14.18-IL2 compared with all other treatment groups (Supplementary Fig. S5C), approximating 50% versus 30% of CD4⁺ T cells, respectively. To extend these observations, we next conducted scRNA-seq of tumor-infiltrating leukocytes.

Three tumor-bearing animals per group were treated and after 48 hours, we dissociated the para-adrenal tumors, generated single-cell suspensions, and obtained 10⁶ pooled cells from all animals for each group. These samples were further enriched for CD45⁺ tumor-infiltrating leukocytes, and their transcriptomes were captured using the droplet-based 10X Genomics platform. After quality control and filtering, we obtained a total of 23,209 single-cell transcriptomes that comprised even proportions from each IC treatment group but contained fewer cells from control animals (Fig. 5A). We used Ptprc to distinguish immune cells from tumor cells (Fig. 5B). We found that roughly 50% of the samples accounted for immune cells in the IC groups, however, proportionally more leukocytes were derived from control animals, altogether yielding comparable total immune cell numbers (Fig. 5C; Supplementary Fig. S6). To define the main subsets of tumor-infiltrating leukocytes, we performed graph-based unsupervised clustering of Ptprc-expressing cells and used conventional markers to define T (Cd3e, Cd8a, Cd4), NK (Ncr1), B (Ms4a1), and dendritic cells (DC; Siglech), polymorphonuclear/granulocytic (PMNs; S100a9), and macrophages/monocytes (Csf1r; Fig. 5D; Supplementary Figs. S7 and S8). Quantification of these subsets showed that CD8⁺ T cells and macrophages/monocytes comprised the largest populations of immune cells (Fig. 5E), which is consistent with our findings from the multi-color flow cytometry analysis (Supplementary Fig. S5A).

Next, we performed separate analyses for myeloid (i.e., PMNs, DCs, monocytes, and macrophages) and lymphoid (i.e., T and NK cells) cell populations. A total of 17 cell clusters were defined by unsupervised clustering of the myeloid compartment (Fig. 5F), of which we excluded two clusters that contained predominantly ribosomal RNA and lacked lineage-defining genes (total of 170 cells, <4%). On the basis of gene signatures, we manually annotated each cluster and grouped them into M1-like (pro-inflammatory and cytotoxic) and M2-like (anti-inflammatory) tumor-associated macrophages (TAM), Mki67⁺ myeloid cells, inflammatory monocytes, PMN-MDSCs (myeloid-derived–suppressor cells), classic DCs, monocyte-derived DCs, myeloid DCs, plasmacytoid DCs, and eosinophils (Fig. 5G; Supplementary Table S1). The M1-like clusters were characterized by high levels of the chemokines Cxcl9 and Cxcl10. The M2-like clusters exhibited upregulation of Arg1, Retlna, Ccl8, and Cxcr1. Differentially expressed genes in PMN-MDSCs included S100a8/9, Cxcl2, Cxcr2, Trem2, and Il1b. Compared with untreated controls and hu14.18-IL2 and -IL21 treatment, animals receiving IL15 conjugated antibody had a marked increase in Cxcl9- and Cxcl10-expressing M1-like TAMs (43% vs. ∼30%; Supplementary Table S2) in the TME, though measurement of the precise cell abundance should be interpreted in the context of the technical limitations of scRNA-seq. Both chemokines are known to regulate the recruitment of Cxcr3-expressing stem-like CD8⁺ T cells to the TME and correlate with positive clinical responses to immune checkpoint blockade therapy (22–24). Furthermore, emerging data have demonstrated that M2-biased TMEs are associated with a poorer outcome in neuroblastoma (Supplementary Fig. S9; refs. 25, 26). Comparing the three ICs, we also noted a trend toward fewer Cxcl2-expressing PMN-MDSCs in mice treated with hu14.18-IL15 (5% vs. 7 and 10%; Fig. 5H). CXCL2 plays an essential role in suppressing the effector function of antigen-specific tumor-infiltrating lymphocytes, thereby countering responses to immunotherapy (27). Taken together, our findings demonstrate that ICs substantially alter the composition of the immune cell infiltrate in the TME. Hu14.18-IL15 caused a selective expansion of M1-like TAMs and led to the recruitment of fewer PMN-MDSCs. These findings are consistent with superior ADCC and survival in mice treated with hu14.18-IL15 compared with the other therapy groups.

The same analytical approach that was applied to the myeloid compartment was used to interrogate the single-cell transcriptome of T and NK cells using. Unsupervised clustering revealed 20 independent cell clusters (Fig. 6A). On the basis of canonical genetic signatures (Supplementary Table S3), we assigned annotations manually. Tregs were identified by the expression of the transcription factors Ikzf2 and Foxp3 and NK cells by Ncr1. T_cm cells expressed high levels of Il7r and Tcf7 but lacked differential expression of genes associated with effector function (28). A small subset of cells was characterized as tissue-resident memory (T_RM) cells based on the high expression of Ly6c1/2, Cd69, and Cxcr6 (29). The cluster containing exhausted CD8 T cells was marked by enriched genes encoding chemokines (e.g., Ccl3, Ccl4), cytokines (e.g., Ifng), cytotoxic molecules (e.g., Prf1), and markers and transcription factors associated with functional exhaustion such as Havcr2 and Tox (30, 31). Finally, two large clusters were identified as proliferating lymphocytes and CD8⁺ cytokine secreters/responders. Comparing the different treatment groups (Fig. 6B; Supplementary Fig. S10), we found that ICs significantly decreased the numbers of exhausted CD8⁺ T cells in the TME compared with vehicle-treated controls (Supplementary Table S4). Furthermore, we noted an expansion of NK cells with hu14.18-IL2 but found larger proportions of CD8⁺ cytokine secreters/responders with hu14.18-IL15 and -IL21.

Next, we merged all clusters into 6 distinct subgroups based on cell type-specific genes (Fig. 6C–E). We then compared the proportions of the main cell populations across the treatment groups. Our scRNA-seq analysis corroborated previous flow cytometry findings (Fig. 6C), demonstrating almost twice as many Tregs with hu14.18-IL2 than with the other ICs (Fig. 6F). CD8 T cells were evenly represented across all four therapy groups; however, significantly elevated proportions of NK cells were associated with hu14.18-IL2 (Fig. 6G–H). Curiously, when we compared the fractions of proliferating cells within each cell type, we noted higher percentages of Mki67-expressing CD8 T and NK cells with hu14.18-IL2 but no difference among Tregs (Fig. 6I). This suggests that hu14.18-IL2 increases NK cell proportions through proliferative expansion, whereas Tregs may increase through recruitment to the TME.

To illuminate the transcriptional changes across the therapy groups, we performed separate differential gene expression analyses in CD8⁺ T cells and NK cells comparing ICs with untreated controls (Supplementary Fig. S11). The transcriptional changes in CD8⁺ T cells indicated that the cells undertake preparative measures for cytotoxic activity. For example, several mitochondrial genes (e.g., Uqcr11, Atp5g) were upregulated to increase energy generation. The expression of cathepsins was induced, possibly to process pro-granzymes into proteolytically active forms (32). Genes that protect against oxidative stress were likewise upregulated (Fig. 6J; ref. 33). Similarly, NK cells induced the expression of Txn1, Prdx1, Cyba (protective of oxidative stress), and cathepsins upon IC treatment. Comparisons between the IC groups showed very few transcriptomic differences in both CD8⁺ T and NK cells. Hu14.18-IL15 and -IL21 had slightly higher levels of Prf1 (1.2- and 1.4-fold increase) compared with hu14.18-IL2, respectively. Altogether, our data showed that treatment with ICs can increase the number of cytokine-secreting T cells and decrease the fraction of exhausted cytotoxic T cells in the TME most likely by converting them into functionally intact cells. Quantification of the cellular subsets also revealed that hu14.18-IL2 induced the proliferation of NK cells and recruited immune-suppressive Tregs to the TME. These cellular changes, including the shifts in the myeloid compartment toward a pro-immune environment, may contribute to superior ADCC induced by hu14.18-IL15 compared with the other two ICs.

In the present study, we demonstrate that hu14.18-IL15 and hu14.18-IL21 exhibit robust preclinical activity against neuroblastoma and can remodel the TME, potentially overcoming immunosuppressive mechanisms. We and others previously demonstrated that pre-complexed IL15/IL15Rα can induce impressive responses in humanized aggressive orthotopic neuroblastoma models (10, 11). However, because most patients experience significant systemic side effects with the administration of IL15 or IL15/IL15Rα complexes, more localized delivery strategies may be beneficial (34–36). We designed new ICs linking two other γ_c cytokines, IL15 and IL21, to hu14.18 and compared their ADCC with that of hu14.18-IL2.

Although it has been known that IL2 stimulates Tregs at lower doses (37, 38), IL15 and IL21 primarily act on CD8⁺ T cells and NK cells (9), which renders them attractive alternative adjunct agents for immunotherapy. Aside from expanding CD8⁺ T cells, we found that hu14.18-IL15 also acted on the myeloid compartment by decreasing the recruitment of MDSCs to the TME and increasing pro-immune M1 TAMs. These changes occurred with the added benefit of limiting the numbers of Tregs when compared with hu14.18-IL2. These cooperative effects can enhance the recruitment of cytotoxic T and NK cells and inhibit tumor angiogenesis via the CXCL9/CXCL10 axis, two chemokines that we found to be highly expressed in M1 TAMs (22). CCL3 and its paralogue CCL4 (also known as MIP-1α and MIP-1β), differentially expressed by CD8⁺ T cells in our dataset, have diverse regulatory roles in the TME (39). As such, they can complement a pro-inflammatory response via the recruitment of antigen-primed DC to the TME to induce tumor rejection (40, 41) or by attracting CD8⁺ T cells and macrophages to the primary tumor site to elicit cytotoxicity (42). Interestingly, we found equivalent ADCC between hu14.18-IL2 and hu14.18-IL15 in NK cell-based experimental in vitro systems and T-cell–deficient PDX models with NK cells as effector cells. These observations led us to believe that immune-suppressive mechanisms mediated by MDSCs and Tregs and pro-immune effects executed by M1 TAMs and CD8⁺ T cells in the TME were dominating factors in driving ADCC response in our study. So-called cytokine "muteins” with differential receptor affinities may be useful for the design of future ICs and allow a more selected engagement of desired effector subsets (43). Our work also emphasizes the need to use immune-competent syngeneic murine models complimentary to T-cell–deficient humanized tumor models for preclinical studies.

Among the causes of immunotherapy failure, antigen escape is a well-known mechanism after adoptive cellular therapies (44) and mAb treatment (45–47). Although the emergence of GD2-negative relapses after monoclonal anti-GD2 antibody therapy has been reported in the clinic (48–50), it is unknown how many patients recur due to antigen loss. Interestingly, in neither of our in vivo model systems did we find substantial antigen loss in residual tumors post antibody administration. The measured GD2 expression approximated levels exhibited by the positive control. Thus, it is possible that repeated dosing of ICs could be therapeutically exploited given that NK cells maintain their cytolytic capacity during tumor rechallenge and the relative stability of GD2 expression on neuroblastoma cells.

Given the timing of our scRNA-seq analysis, we acknowledge that our mechanistic findings represent early changes and that analysis at later time points may be needed to illuminate the secondary and tertiary immunologic shifts in the TME upon treatment with ICs. Though cross-reactive between human and mice, one limitation of our ICs is possible immunogenicity of human cytokines in immunocompetent mice when given as repeated dosing cycles.

It is important to acknowledge that proteolytic cleavage of the cytokines can contribute to the occurrence of systemic toxicities as reported previously with hu14.18-IL2 (7). Therefore, it may be anticipated that patients can develop IL-15–related dose-limiting toxicities (e.g., hypotension, thrombocytopenia, and elevation of AST and ALT) during phase I dose-escalation studies (34). Our binding and distribution studies in mice showed that all three ICs were detected on tumor cells derived from the tumor tissue and in the serum at 48 hours (Supplementary Fig. S3). However, the levels of each cytokine were different in tumor tissue and serum, with IL2 being low on the tumor cells but elevated in the circulation, IL21 elevated in both locations, and IL15 being high on the tumor cells but low systemically. These differences corroborate the improved ADCC with hu14.18-IL15 and -IL21 compared with hu14.18-IL2 and could indicate differences in systemic exposures upon translation, which requires future confirmation in non-human primates and/or humans.

In summary, our study demonstrates that IL15 and IL21 are potent adjunct agents for the generation of ICs. Our work highlights that iterative preclinical testing of immunotherapies requires the use of immunocompetent models to emulate the complex immunologic interactions in the TME and gain accurate insights into mechanisms mediating preclinical responses. Our promising results require confirmation in the clinical context. Given the robust preclinical activity of hu14.18-IL15, we have identified an efficacious IC for further consideration for clinical testing in patients with GD2-expressing neuroblastoma. Additional testing in other preclinical models of GD2-expressing pediatric tumors such as osteosarcoma is also warranted.

R. Nguyen reports grants from the Department of Defense and other support from the NIH intramural research program, NCI during the conduct of the study. M. Sun reports grants from Government-National Cancer Institute during the conduct of the study. C.J. Thiele reports grants from Department of Defense during the conduct of the study. No disclosures were reported by the other authors.

R. Nguyen: Conceptualization, resources, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. X. Zhang: Formal analysis, writing–review and editing. M. Sun: Data curation, writing–review and editing. S. Abbas: Formal analysis, writing–review and editing. C. Seibert: Data curation, writing–review and editing. M.C. Kelly: Conceptualization, writing–review and editing. J.F. Shern: Formal analysis, writing–review and editing. C.J. Thiele: Formal analysis, writing–review and editing.

This study was supported by the Department of Defense Peer Reviewed Cancer Research Program (CA191207 to R. Nguyen) and the intramural research program of the National Institutes of Health (all authors). We want to thank the NCI CCR Animal Resource Program/NCI Biological Testing Branch (Dr. M. Custer and Kathy Divi) for providing mice for our experiments, the NIH CC Blood bank (Thomas Lewis) for providing healthy donor blood cells, the Biological Resources Branch at the Frederick National Laboratory for giving us cytokines and hu14.18-IL2, Drs. N. Taylor, Z. Liu, J. Gulley, and D. Cervi for reviewing the article and giving us thoughtful suggestions, Dr. A. Mendoza for his assistance with animal regulatory work, Dr. D. Arroyo for his scientific input, and the Children's Solid Tumor Network for providing PDX lines for our studies. The CCR Single Cell Analysis Facility was funded by FNLCR contract 75N91019D00024. Sequencing was performed in conjunction with the CCR Genomics Core. This work used the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). Some figures were generated with Biorender.com.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.