IL-9 as a naturally orthogonal cytokine with optimal JAK/STAT signaling for engineered T cell therapy

SUMMARY

Cytokines and their receptors enable precise tuning of T cell function. Leveraging this biology holds tremendous promise for optimizing antitumor immunity. Arming T cells with a synthetically orthogonal IL-9 receptor (o9R), for instance, permits facile engraftment and potent anti-tumor functions. Exploiting the paucity of wild-time IL-9R expression and the safety of high doses of IL-9, here we showed that, compared to o9R, T cells engineered with wild-type IL-9R exhibited superior tissue infiltration, stemness, and anti-tumor activity. These qualities were consistent with a stronger JAK/STAT signal, which included canonically IL-12-driven STAT4 in addition to STAT1/3/5. IL-9R T cells were exquisitely sensitive to perturbations of proximal signaling, including structure-guided attenuation, amplification, and rebalancing of JAK/STAT signals. Biased IL-9R mutants showed STAT1 acts as a rheostat between stem-like and effector states. In summary, we identify IL-9/IL-9R as a naturally orthogonal cytokine-receptor pair with an optimal JAK/STAT signaling profile for engineered T cell therapy.

Graphical Abstract

eTOC

Cytokines optimize and fine-tune effector T cell responses. Here, Jiang et al. utilize IL-9, with scarce receptor expression in normal tissues, to boost IL-9R engineered T cells for cancer therapy. IL-9 signaling initiates a unique STAT activation pattern, including STAT4, that modulates T cell states for effective anti-tumor responses.

INTRODUCTION

Common γ-chain (γ_c) cytokines are critical regulators of T cell differentiation and function, exerting their effects through JAK/STAT signaling pathways¹. Their potential to enhance adoptive T cell therapy for cancer has long been recognized². Building on an orthogonal IL-2 system in which we swapped intracellular domains (ICD) of γ_c cytokine receptors, we previously found that the chimera incorporating the IL-9R ICD activates STAT1, STAT3 and STAT5, enriches stem-like T cells, improves infiltration into solid tumors, and enhances antitumor activity in immunocompetent hosts without lymphodepletion³.

Given the relative anonymity of IL-9R signaling among the γ_c family, this surprising result prompted us to explore anti-tumor effects of T cells signaling through the wild-type IL-9 receptor. In particular, we were intrigued by the reported low nanomolar affinity of IL-9 for the wild-type IL-9 receptor⁴. However, it was not clear whether there is sufficient IL-9R expression on T cells to permit a response to IL-9. And furthermore, as the only γ_c cytokine not tested in humans, potential toxicities of systemic IL-9 administration are not well understood.

We integrated genomic, transcriptomic and functional datasets and found that IL-9R is rarely expressed across immune and non-immune tissues, including throughout fetal and thymic development and in the tumor microenvironment. In fact, population genetics suggest limited evolutionary constraint on IL-9. Based on these observations, we nominated IL-9 and IL-9R as a naturally orthogonal cytokine-receptor pair. In mice, IL-9 was well-tolerated even at very high doses. Tumor-specific T cells engineered with the wild-type IL-9 receptor yielded superior engraftment, tumor infiltration, stemness and anti-tumor activity compared to o9R. We linked these findings to more potent phosphorylation of STAT1, STAT3 and STAT5, as well as phosphorylation of STAT4, which is not classically associated with common γc signaling.

By attenuating, amplifying or biasing IL-9R JAK/STAT signaling, we showed that anti-tumor efficacy is exquisitely sensitive to both signal strength and STAT stoichiometry. Mutant receptors revealed a T cell intrinsic STAT1 rheostat that skews T cells from a proliferative stem- and memory-like state toward terminally differentiated effectors. We extend these results to human CAR and TCR T cells, positioning IL-9/IL-9R as a natural alternative to synthetic orthogonal systems.

RESULTS

The expression and activity of IL-9R is restricted in T cells and across normal tissues

We examined single cell RNA-sequencing (scRNA-seq) of 28,964 peripheral blood mononuclear cells (PBMCs) from two healthy donors (Figure 1A)⁵. Compared to receptors for other common γc cytokines, IL9R was expressed at low levels and in very few cells, including rare B cells (0.53%) and CD4 T cells (0.26%). We extended the analysis to tissue resident immune cells in a scRNA-seq dataset of 39,560 lamina propria immunocytes (Figure 1B)⁶. Again, IL9R was rarely and weakly expressed rarely, in contrast to IL2RB, IL4R, IL7R, and IL21R. qRT-PCR of activated human CD4 and CD8 T cells confirmed low IL9R expression (Figure S1A). Even across thymic development, IL9R transcript levels are nearly absent (Figure S1B)⁷. Consistent with the low transcript expression, activated human T cells failed to phosphorylate STATs in response to IL-9 (Figure S1C).

Figure 1. Nomination of IL-9 as a naturally orthogonal cytokine for T cell therapy.

(A) scRNA-seq of PBMCs from healthy donors. Top, t-SNE with cell-type annotations. For each γ_c cytokine receptor, a feature plot (scaled expression) and a per-cell-type scatter plot are shown.

(B) As in (A), but for lamina propria lymphocytes from the uninflamed ileum of patients with Crohn’s disease.

(C) Normal-tissue expression scatterplot of γ_c cytokine receptors across 37 GTEx normal tissues. The dashed line marks the nTPM threshold of 0.7.

(D) Dot plot of cytokine receptor expression across malignant and non-malignant cell types, in soft tissue sarcoma (scRNA-seq; n=65,945 cells from 15 patients).

(E) Differential expression of genes (number and magnitude) across lymph node cell types from mice treated with individual cytokines versus PBS control.

(F) Reported median S_het values for γ_c and non-γ_c cytokines. S_het estimates the impact of predicted heterozygous loss-of-function variants on evolutionary fitness; error bars denote 95% highest posterior density. The dashed line at 0.02 indicates the median across canonical transcripts.

(G) Dot plot of γ_c cytokine receptor expression in sorted CD19 CAR T cells (n=66,042) from 15 pediatric patients at multiple post-infusion time points; CD4 (left) and CD8 (right) shown separately.

PBMC: peripheral blood mononuclear cells; TPM: transcripts per million; GTEX: Genotype-Tissue Expression; CAR: chimeric antigen receptor

Extending beyond immune cells to 37 human tissue types, IL9R expression was broadly restricted: only four of 37 tissues exceeded log₁₀(nTPM)>0.7: lung, small intestine, spleen and urinary bladder (Fig. 1C)^8,9. IL9R expression was also restricted during fetal development (Figure S1D).

To evaluate whether IL-9R expression can be induced across T cell differentiation states, we analyzed scRNA-seq data from tumor, adjacent normal tissues and lymph nodes from three patients with non-small cell lung cancer after immune checkpoint blockade¹⁰. IL-9R transcripts were low across twelve distinct T cell subsets and across tissues of origin (Figure S1E-F). Given the inflammatory tumor microenvironment might induce IL9R expression, we analyzed scRNA-seq from soft tissue sarcoma samples (n=65,945 cells, 15 patients) and again observed restricted IL9R expression across major cell types (Fig. 1D).

As a functional assessment of potential toxicity of systemically administered IL-9, we evaluated scRNA-seq data obtained from lymph nodes of mice treated with one of 86 different cytokines (n=3 mice per cytokine) or PBS (n=14 mice)¹¹. Unlike many γc and non-γc cytokines, IL-9 minimally perturbed gene expression relative to control, with ≤1 differentially expressed gene (∣log₂(fold change)∣>0.25, FDR<0.05, Wilcoxon) per cluster (Fig. 1E).

We hypothesized that IL-9 may be less essential for evolutionary fitness than other cytokines. Using population-scale exomes (n=983,578; Regeneron Genetics Center Million Exome project), IL-2, IL-7 and IL-15 showed high indispensability (S_het), whereas IL-9 ranked most dispensable with mean S_het 0.0019 (95% highest posterior density: [0.0017, 0.0022]), below the canonical transcript mean (0.073)(Fig. 1F)^12,13.

Given insufficient IL-9R expression on T cells, exploiting the IL-9/IL-9R axis in T cell therapy would require genetic modification. Consistent with this, scRNA-seq from 66,042 post-infusion CAR T cells from 16 pediatric patients showed absent IL9R expression, unlike IL2RB/IL2R (primarily CD8) and IL4R/IL7R (primarily CD4) (Fig. 1G).¹⁴

IL-9 is well-tolerated and drives potent anti-tumor efficacy of T cells engineered with IL-9R

We engineered half-life extended mouse IL-9 N-terminal mouse serum albumin tagging (MSA-IL9), which, unlike MSA-IL2, was well-tolerated by weight, survival and mobility even at 100μg every other day for three weeks (Fig. 2A-C, Fig. S2A). MSA-IL2 increased serum IFNγ levels, whereas MSA-IL9 did not, consistent with the low surface expression of IL-9R on major immune subsets across tissues (Fig. 2D and S2B).

Figure 2. Exogenous IL-9 is well-tolerated and enhances tissue infiltration, stemness and anti-tumor activity of IL-9R-engineered antigen-specific T cells.

(A) Body weight (percentage change from baseline) of mice (n=8 mice/group) treated with PBS, IL-9 (50 μg or 100 μg), or IL-2 (50 μg) i.p. every other day.

(B) Survival of mice from (A).

(C) Distance traveled over 30 second period in mice from (A), normalized to the PBS group.

(D) Serum IFNγ measured by ELISA six days after initiating treatment (n=3 mice/group, technical duplicates per mouse) as in A-C.

(E) B16-F10 tumor growth (mean ± SEM, n=5/group) after ACT with untransduced pmel T cells or pmel T cells engineered with o9R or IL-9R (0.4x10⁶ transduced cells, i.v.), and corresponding cytokine treatment (5x10⁴ IU i.p., daily for 5 days starting with ACT). Data are representative of three independent experiments.

(F) Survival of mice from (E).

(G) Transduced pmel T cells in the blood five days after ACT (n=5/group).

(H) Transduced pmel T cells (Thy1.1⁺YFP⁺) across indicated tissues five days after ACT (n=5/group).

(I) Transduced (YFP⁺) pmel T cells as percentage of pmel T cells within the tumor 14 days after ACT (left panel) and CD8⁺ T cells per gram of tumor 14 days after ACT (right panel) (n=5/group).

(J) Shown are the proportion of T_SCM cells cells for o9R or IL9R transduced C57BL/6 T cells stimulated for 24h with IL-2 (10nM), IL-9 (10nM), oIL-2 (10μM), or combination. Statistical comparisons refer to the differences in T_SCM frequency across conditions. Data from two independent experiments with 4 technical replicates per condition.

For in vivo experiments, cytokines were tagged with mouse serum albumin (MSA) for half-life extension. *P < 0.05, **P < 0.005, ***P < 0.0005, ****P < 0.0001 (one-way ANOVA for A-G unpaired t test for H-I, two-way ANOVA for J).

We compared the anti-tumor efficacy of H2-D^b/gp100-specific pmel-1 T cells (hereafter, pmel) engineered with IL-9R or our benchmark, synthetic o9R, in B16-F10-bearing mice. IL-9R was co-expressed with YFP (P2A), which was a reliable surrogate for IL-9R surface expression (Fig. S2C). With corresponding cytokine treatment (MSA-IL9 or MSA-oIL2), IL-9R pmel T cells achieved tumor control and survival compared to o9R pmel T cells (Fig. 2E-F, Fig. S2D). MSA-IL9 alone had no effect (Fig. S2E), did not augment untransduced pmel T cells (Fig. S2F), and IL-9R pmel T cells required exogenous MSA-IL9 (Fig. S2G). Efficacy extended to a sarcoma model derived from Kras^G12D/−P53^flox mice and modified to express gp100 (KP-gp100; Fig. S2H).

Superior efficacy associated with improved peripheral expansion of IL-9R pmel T cells, which scaled with higher IL-9 dose (Fig. 2G). We also noted enhanced early tissue infiltration (lung, liver and tumor; Fig. 2H) and sustained tumor enrichment after cytokine withdrawal (Fig. 2I), despite no enrichment in normal tissues (Fig. S2H).

We previously demonstrated³ that o9R expands a stem cell memory (T_SCM, CD62L⁺CD44⁻Sca-1⁺) phenotype. Here, IL-9R signaling more efficiently induced naïve (T_N, CD62L⁺CD44⁻ T cells; Fig. S2I-J), including T_SCM (Fig. 2L, Fig. S2K), with effects confined to IL-9R⁺ (YFP⁺) cells, whereas IL-2 affected both YFP⁺ and YFP⁻ cells (Fig. S2L). In sum, wild-type IL-9R promotes stemness and tumor infiltration, and yields more potent anti-tumor function than o9R.

In addition to robust activation of STAT1, STAT3 and STAT5, IL-9R recruits STAT4 signaling

We compared signaling strength of IL-9R versus o9R. Dose-response curves showed IL-9R produced higher E_max and lower EC₅₀ for STAT1, STAT3 and STAT5 than o9R (Fig. 3A), with no cytokine-independent tonic signaling of IL-9R (Fig. S3A).

Figure 3. IL-9R signaling phosphorylates STAT4, in addition to STAT1, STAT3, and STAT5.

(A) Dose-response curves for STAT1, STAT3, and STAT5 phosphorylation in IL-9R or o9R transduced (YFP⁺) pmel T cells stimulated with either oIL-2 or IL-9 for 20 minutes (technical duplicates; representative of three independent experiments).

(B) Volcano plots depicting differential gene expression by RNA-seq in IL-9R or o9R transduced C57BL/6 T cells treated with IL-9 (10 nM), IL-2 (10 nM) or oIL2 (10 μM) for 48 hours. Comparisons shown below the x-axis. Significance (red) indicates adjusted p < 10⁻⁵ and absolute fold change ≥2.

(C) Differentially phosphorylated proteins between IL-9R transduced C57BL/6 T cells stimulated for 20 minutes with either IL-2 (10nM) or no cytokine (left) or IL-9 (10nM) versus no cytokine (right). Significance (red) indicates adjusted p<0.05 and log₂(fold change)≥ 0.5.

(D) Dose-response curve for pSTAT4 in IL-9R transduced (YFP⁺) pmel T cells treated for 20 minutes with IL-9 or IL-12, or o9R transduced T cells treated with oIL-2 (technical duplicates; representative of two independent experiments).

(E) Waterfall plot of inferred transcription factor enrichment from RNA-seq of IL-9R T cells treated with IL-9 or IL-2 (top 15 for each).

(F) Heat map the Biocarta IL-12 Pathway gene set using RNA-seq from (B). Samples and genes clustered hierarchically without supervision.

(G) In vitro expansion of IL-9R (YFP⁺) pmel T cells treated with 10nM cytokine starting on day 3 after activation (n=3 technical replicates/group). Representative of three biological replicates and three independent experiments.

(H) Representative CD44 and CD62L expression of T cells from (G) after 24h cytokine exposure.

(I) Quantification of naïve CD62L⁺CD44⁻ (left) and stem-like CD62L⁺CD44⁻Sca-1⁺ (right) T cells from (H); two independent experiments.

*P < 0.05, **P < 0.005, ***P < 0.001, ****P < 0.0001 (one-way ANOVA for G; two-way ANOVA for G, I).

RNA-seq 48h after stimulation showed IL-9R T cells (plus IL-9) and o9R T cells (plus oIL-2) elicit relatively similar transcriptomes when compared with their IL-2-treated controls (Fig. 3B, Fig. S3B-D). While 7673 differentially expressed genes (DEGs) were observed between IL-9R + IL-9 and IL-9R + IL-2 groups, and 7166 DEGs between o9R + oIL-2 and IL-9R + IL-2 groups, only 3031 DEGs were noted between IL-9R + IL-9 and o9R + oIL-2 groups (Fig. 3B, Fig. S3B-D). Thus, IL-9R and o9R activate similar gene expression patterns, consistent with their shared ICDs, yet their transcriptomic profiles are distinguishable due to differences in signal strength.

Phosphoproteomic analysis (20 minute stimulation) revealed broad changes with IL-2: 228 differentially abundant phosphoproteins, consistent with its known activation of multiple signaling pathways including JAK/STAT, ERK and AKT (Fig. 3C, Fig. S3D, Table S1)^15,16. However, the phosphoproteomic footprint of IL-9 was highly restricted, with only 4 differentially abundant phosphoproteins. The JAK/STAT pathway accounted for three of the four differentially abundant phosphoproteins. We confirmed that IL-9 did not phosphorylate ERK or AKT by phosphoflow (Fig. S3E-F).

Unexpectedly, pSTAT4 was the most differentially abundant phosphoprotein upon IL-9 stimulation (Fig. 3C)¹. To our knowledge, STAT4 has not been canonically implicated in IL-9 or, more broadly, γ_c signaling¹. Phosphoflow confirmed pSTAT4 activity comparable to IL-12 signaling (Fig. 3D), the canonical activator of STAT4 in T cells¹⁷ and a well-described promoter of anti-tumor functions in T cells (Fig. 3D)^17-20. Transcription factor inference based on RNA-seq data supported STAT1, STAT2, STAT3 and STAT4 engagement with IL-9R versus IL-2 (Fig. 3E). Both IL-9R and o9R signaling increased expression of 15 of 18 Biocarta IL-12 Pathway genes compared to IL-2 (Fig. 3F).

Unlike IL-12 (STAT4-only), IL-9 activates STAT1, STAT3, STAT4 and STAT5. This distinction was evident in the functional outcomes of IL-9 and IL-12 signaling. Functionally, IL-9 provided modest proliferation to IL-9R T cells (less than IL-2, more than IL-12; Fig. 3G). This was likely attributable to a higher STAT5 signal from IL-2²¹. IL-9 also enriched naïve and T_SCM cells compared to IL-12 (Fig. 3H-I), consistent with its stronger STAT3 signal²².

In sum, IL-9R robustly activates STAT1, STAT3 and STAT5 and also recruits STAT4, producing a transcriptional and phenotypic program blending features of IL-12 and γc cytokines and contributing to IL-9R T cell efficacy.

Either attenuation or amplification of IL-9R signaling disrupt its anti-tumor properties

To test how proximal signaling magnitude affect efficacy, we first attenuated IL-9 by mutating the γ_c-contact glutamine within helix D (Q115T; Fig. S4A, predicted complex in Fig. 4A). In other γ_c cytokines, a mutation in this glutamine residue attenuates signaling²³. IL-9^Q115T reduced E_max and EC₅₀ of STAT1, STAT3. STAT4 and STAT5 (Fig. 4B) and diminished T_SCM induction versus IL-9. Attenuation of IL-9 diminished the anti-tumor efficacy of IL-9R pmel T cells against B16 melanoma (Fig. 4C, Fig. S4C). Blood expansion and persistence were initially similar, but by day 41, IL-9R pmel T cells were undetectable with IL-9^Q115T, while IL-9R pmel T cells in the IL-9^WT group persisted (Fig. 4D, Fig. S4D). In fact, IL-9R pmel T cells were detectable 115 days after initial adoptive transfer, at which point a third cycle of IL-9^WT treatment resulted in a third peak of IL-9R pmel T cells that exceeded levels of the initial expansion.

Figure 4. Structure-based attenuation or amplification of IL-9/IL-9R signaling diminishes anti-tumor qualities of T cells signaling through the native receptor complex.

(A) Structural prediction of the interleukin-9 (IL-9) receptor complex based on AlphaFold2 showing IL-9R (green), IL-9 (blue), and the γ_c (pink). The inset demonstrates the interaction of the glutamine at amino acid position 115 of IL-9 (Q115) with the γ_c; this residue was mutated (IL-9^Q115T) to attenuate signaling.

(B) Dose-response curves for pSTAT1, pSTAT3, pSTAT4, and pSTAT5 in IL-9R pmel T cells stimulated with IL-9^WT (blue) or IL-9^Q115T (light gray) for 20 minutes. Shown are technical duplicates; representative of two independent experiments. EC50 and Emax values are provided in the accompanying table.

(C) B16-F10 tumor growth after ACT with IL-9R pmel T cells. Mice were treated with either IL-9^WT (n=7 mice) or IL-9^Q115T (n=9 mice) (10 doses, every other day). Data are representative of two independent experiments.

(D) Peripheral blood IL-9R transduced pmel T cell counts over time in mice treated as in ©; vertical dashed lines indicate cytokine dosing. In the IL-9^Q115T group, cells were undetectable by day 41 or after three additional cytokine doses on days 44-48, only the IL-9^WT group received doses beyond day 48. Data are representative of two independent experiments.

(E) Schematic of IL-9R variants with either three (IL-9R^3x) or five (IL-9R^5x) repeated phosphotyrosine (pY) elements within the intracellular domain (created with Biorender.com). Sequence of the phosphotyrosine element shown in the legend.

(F) Phosphorylation of indicated STATs at E_max (100nM) for IL-9R^WT, IL-9R^3x, and IL-9R^5x transduced pmel T cells stimulated with IL-2 or IL-9 for 20 minutes. Data is representative of three biological replicates.

(G) B16-F10 tumor growth after ACT with either IL-9R^WT (n=7 mice) or IL-9R^3x transduced pmel T cells (n=8 mice). Mice were also treated with IL-9 (5x10⁴ IU i.p., every other day for 5 doses starting with ACT). Shown are individual tumor growth curves (left, middle) and group mean ± SEM (right). Data are representative of two independent experiments.

*P < 0.05, **P < 0.005, ***P < 0.001, ****P < 0.0001. (two-way ANOVA for C; Welch's t-test for D, G; one-way ANOVA for F). See also Figure S4.

To amplify IL-9R signaling, we leveraged the fact that IL-9R signaling is driven by a single phosphotyrosine site (Y405) downstream of the Box 1 motif²⁴. We repeated a 39 amino acid sequence flanking Y405 three or five times (IL-9R^3x or IL-9R^5x) (Fig. 4E). These increased the E_max of IL-9 for pSTAT1, STAT3 and STAT4 (not pSTAT5), with significance for IL-9R^5x (Fig. 4F, Fig. S4E), and minimally augmented T_SCM enrichment in vitro (Fig. S4F).

Despite subtle amplification of the IL-9R signal, IL-9R^3x reduced anti-tumor activity of pmel T cells compared to IL-9R^WT, with earlier outgrowth of B16-F10 tumor and fewer complete responses. (1/8 versus 6/7; Fig. 4G). Thus, both weaker and stronger IL-9R signaling impaired efficacy, suggesting an optimal JAK/STAT signaling window.

IL-9R intracellular domain mutants skew STAT phosphorylation profiles and alter in vivo proliferation and anti-tumor efficacy

Because IL-9R activates multiple STATs, we explored how stoichiometry—not just magnitude--contributes to its anti-tumor functions. We confirmed that Y405 is the primary hub for IL-9R STAT signaling, as a Y405F substitution ablated STAT1, STAT3 and STAT4 activation and diminished STAT5 activation (Fig. S5A)²⁴. Amino acids immediately adjacent to a phosphotyrosine site can influence STAT binding and activation²⁵, so we generated ten single amino acid mutants of the C-terminal proline or glutamine residue of Y405 (Fig. 5A), each yielding distinct pSTAT stoichiometry (Fig. 5B).

Figure 5. IL-9R intracellular domain mutants skew STAT phosphorylation and alter proliferative capacity and anti-tumor efficacy.

(A) Schematic of the IL-9 receptor complex, highlighting the phosphotyrosine (pY) site within the IL-9R intracellular domain (ICD) and three adjacent residues. A panel of ten single amino acid substitutions of the proline/glutamine residues adjacent to the pY site (created with Biorender.com).

(B) Heat map of pSTAT MFI (log-scaled and normalized to IL-9R^WT [YLPQ]) at E_max for C57BL/6 T cells transduced with wild-type IL-9R or one of ten mutants. Transduced pmel T cells (technical duplicates) were stimulated with IL-9 for 20 minutes. Data are representative of two independent experiments. Raw MFI values are shown in each cell.

(C) Dose-response curves of pSTATs among transduced IL-9R^WT, IL-9R^AQ, and IL-9R^PR pmel T cells (YFP⁺) stimulated with recombinant IL-9 for 20 minutes (technical duplicates). Data are representative of three biological replicates. Emax and EC50 are shown in the table.

(D) Relative in vitro expansion of YFP⁺ IL-9R^WT, IL-9R^AQ, IL-9R^PR pmel T cells cultured with IL-9 on day 3 post-activation (10 nM; n=3 technical replicates/group). Data are representative of three biological replicates.

(E) Proliferation index of C57BL/6 T cells engineered with IL-9R^WT, IL-9R^AQ, and IL-9R^PR over nine days post-activation (transduced day 1), measured by dilution of CellTrace Violet dye (n=6 replicates/group). Cells received IL-2 or IL-9 (10 nM) on day 3.

(F) Peripheral blood quantification of IL-9R^WT, IL-9R^AQ, or IL-9R^PR transduced (Thy1.1⁺YFP⁺) pmel T cells in B16-F10 tumor-bearing mice on the indicated days after ACT (n=6-7 mice/group). IL-9 (5x10⁴ IU i.p., every other day) was given with ACT and continued for 5 doses. Representative of at least three independent experiments.

(G) Quantification of intratumoral IL-9R^WT, IL-9R^AQ, or IL-9R^PR transduced (YFP⁺) pmel T cells seven days post ACT (n=6-7 mice/group). Representative of two independent experiments.

(H) B16-F10 tumor growth in mice treated with IL-9R^WT (n=7 mice) or IL-9R^PR (n=6 mice) pmel T cells. Data are representative of at least three independent experiments.

(I) B16-F10 tumor growth in mice treated with IL-9R^WT or IL-9R^AQ pmel T cells (n=10 mice/group). IL-9 treatment as per (F). Data are representative of at least 4 independent experiments.

*P < 0.05, **P < 0.005, ***P < 0.001, ****P < 0.0001 (two-way ANOVA for D-F; Welch's t-test for G-I).

To study the contribution of STAT1, we selected IL-9R^PR (higher pSTAT1 but similar pSTAT3, pSTAT4 and pSTAT5 compared to IL-9R^WT) and IL-9R^AQ (reduced pSTAT1 with similar pSTAT3, pSTAT4 and pSTAT5 compared to IL-9R^WT). Two other mutants, IL-9R^KQ and IL-9R^SQ, also produced lower STAT1 phosphorylation than IL-9R^WT while maintaining phosphorylation levels of the remaining STATs, but for simplicity we proceeded with IL-9R^AQ. We confirmed that IL-9R^PR increased pSTAT1 E_max and IL-9R^AQ reduced pSTAT1 E_max in T cells compared to IL-9R^WT (Fig. 5C). We also confirmed that IL-9R^AQand IL-9R^PR phosphorylated STAT3, STAT4, and STAT5 at similar levels. Compared to IL-9R^WT, IL-9R^AQ and IL-9R^PR produced a similar pSTAT5, but weaker pSTAT3 and pSTAT4 signals.

In vitro, mouse IL-9R^AQ T cells expanded more than IL-9R^WT T cells (Fig. 5D), whereas IL-9R^PR T cells expanded less. IL-9R^AQ T cells had a higher proliferative index than compared IL-9R^PR T cells as measured by dilution of CellTrace Violet dye (Fig. 5E and Fig. S5B), and we did not observe an increase in apoptotic or dead cells with either mutant (Fig. S5C-D), consistent with the cytostatic effects of STAT1^26,27.

Phenotypically, IL-9R^AQ T cells favored a central memory (T_CM) phenotype (CD62L⁺CD44⁺) in vitro, and aligned most with IL-2-treated T cells, while IL-9R^PR skewed away from T_CM (Fig. S5E). Across T_SCM and T_EFF(CD62L⁻CD44⁺) subsets, receptors ordered according to their pSTAT1 Emax (IL-9R^PR > IL-9R^WT> IL-9R^AQ).

The improved proliferation of IL-9R^AQ T cells was pronounced in vivo where the cells expanded ~5-10-fold greater in the blood five and ten days after adoptive cell transfer (ACT) compared to IL-9R^WT (Fig. 5F); IL-9R^PR T cells expanded less. More IL-9R^AQ T cells were also observed in the tumor seven days after ACT (Fig. 5G).

Compared to IL-9R^WT pmel T cells, IL-9R^PR pmel T cells resulted in significantly reduced tumor control, fewer complete responses, and worse survival (Fig. 5G; Fig. S5F), consistent with their diminished proliferative capacity. IL-9R^AQ pmel T cells, despite their increased peripheral expansion and superior tumor infiltration, did not surpass the anti-tumor efficacy of IL-9R^WT pmel T cells (Fig. 5H, Fig. S5G).

T cell intrinsic STAT1 acts as a rheostat between stem and effector fates of tumor-infiltrating T cells

We performed scRNA-seq on tumor-infiltrating pmel T cells engineered with IL-9R^WT, IL-9R^AQ or IL-9R^PR, eight days after adoptive transfer (Fig. 6A, Fig. S6A). As expected, the frequency of engineered (YFP⁺) pmel T cells among infiltrating CD8⁺ T cells was highest in the IL-9R^AQ group (17.0%), followed by IL-9R^WT (13.4%) and IL-9R^PR (0.7%) (Fig. S6B).

Figure 6. IL-9R variants reveal STAT1 as a rheostat skewing T cells from a stem and memory state toward a terminal effector state.

(A) Schema for scRNA-seq experiment. IL-9R^WT, IL-9R^AQ, or IL-9R^PR pmel T cells were adoptively transferred into B16-F10 tumor-bearing mice on day 7 after inoculation (n=7-8 mice/group), IL-9 (5x10⁴I.U. every other day) began with ACT. Transduced pmel T cells (Thy1.1⁺YFP⁺) were FACS-sorted for library preparation and scRNA-seq. See related Figure S6A - S6B.

(B) UMAP of scRNA-seq from n=6,706 cells pmel T cells from (A) showing ten annotated clusters. See related Figure S6C-E.

(C) Propeller composition plots showing the proportion of each cluster by treatment group (IL-9R^WT, IL-9R^AQ, or IL-9R^PR).

(D) Violin plots of a KLRG1^hi effector versus naïve mouse T cell gene sets (GSE10239) summarized as Seurat module scores by treatment group.

(E) Violin plots of a acute infection versus malignancy gene sets (GSE60501) summarized as Seurat module scores by treatment group.

(F) Ridgeplots of pseudotime scores by treatment group.

(G) UMAP annotated with pseudotime trajectories; the Tscm-like (Tcf7) cluster was selected as the root (white circles with black outlines). Black circles with white outlines represent nodes of the differentiation trajectory.

(H) Schematic for linking phosphoflow and RNA-seq. o9R, IL-9R^WT, IL-9R^AQ, IL-9R^PR, or IL-9R^5x pmel T cells were exposed to cytokine for 20 minutes (phosphoflow) or 48h (RNA-seq). MFIs at E_max for each pSTAT were merged with transcriptomes to identify genes most correlated with each pSTAT.

(I) Histograms of phosphorylation of each pSTAT for transduced (YFP⁺) T cells after 20 minutes of cytokine exposure at Emax: IL-2 (10nM), IL-9 (10nM) and oIL-2 (10μM).

(J) Venn diagram of the top 100 genes most correlated with the phosphorylation of each STAT.

(K) Scatterplot relating pSTAT1 module score (y-axis; built from the top 100 pSTAT1-correlated genes) to pSTAT1 phosphorylation levels (x-axis) in vitro. Shown are biological triplicates (RNA-seq) colored by sample condition; phosphorylation values are technical replicate means. Gray line indicates the linear regression fit.

(L) Violin plot depicting projection of pSTAT1 module score from (J)-(K) onto dataset from (A).

(M) Overlay of the pSTAT1 module score from (K) onto UMAP from (B).

(N) Waterfall plot summarizing transcription factor (TF) enrichment from vitro RNA-seq data comparing IL-9R^WT vs IL-9R^AQ groups (left) and IL-9R^PR vs IL-9R^WTgroups (right). Increasing STAT1 and related TF activity tracks with gene expression changes across IL-9R^AQ → IL-9R^WT → IL-9R^PR (shown in red).

(O) Ridgeplots summarizing STAT1, STAT3, STAT4, and STAT5a regulon activity (AUC) inferred with SCENIC, shown by group with pairwise two-sided Wilcoxon rank-sum tests).

Most cells mapped to effector subtypes (Fig. S6C, Table S2), including early activated effectors (T early activated, expressing Pdcd1, Lag3, Havcr2, and Vsir, along with Gzmb and Prf1), proliferative effectors (Teff-1/prolif and Teff-2/prolif), and effectors with high Stat1 expression (Teff-5). Two clusters consisted of mixed effector, naïve and either precursor exhausted (Teff-3/pex/naive) or effector memory (Teff-4/naïve/em) T cells. We also identified a cluster with features of both effector memory and resident memory T cells (Tem/rm), marked by Il7r, Itga1, Itga4, Cxcr3, Ccl5 and Gzma expression (Fig. S6D). A cluster of stem cell memory-like T cells (T_SCM-like) had uniquely high expression of Tcf7, as well as Il7r (Fig. S6D). We also identified clusters of proliferative T cells (Tprolif) and T cells enriched for a myeloid differentiation gene set (Fig. S6E)²⁸. Transcription factor and gene regulatory network inference using the Single-Cell rEgulatory Network Inference and Clustering (SCENIC) package aided annotations (Table S2).

Despite the tumor TCR stimulus, T cell fates diverged by cytokine receptor. IL-9R^PR pmel T cells (high pSTAT1) localized to Teff-5 (Stat1/Gzmb/Prf1); IL9R^AQ pmel T cells (low pSTAT1) enriched in Tem/rm and Tscm-like (Fig. 6C, Fig. S6D) and were less represented among effector clusters; IL-9R^WT pmel T cells occupied an intermediate state, enriched for both early activated effectors (Teff-1/activated) and Tem/rm and Tscm-like clusters. Consistent with this gradient of effector phenotype (IL-9R^PR > IL-9R^WT > IL-9R^AQ), the overall expression of a gene set distinguishing KLRG1^hi effector versus naïve T cells was lowest in IL-9R^AQ, and highest in_IL-9R^PR pmel T cells (GSE10239; Fig. 6D)²⁹. A similar trend was seen for the expression of Prf1 and Gzmb (Fig. S6F).

We did not find clusters fitting a canonical definition of exhaustion as most cells lacked expression of Tox and Eomes, consistent with the early timepoint. Instead, sn acute-infection-versus-malignancy gene set (GSE60501)³⁰ was most highly expressed among IL-9R^AQ and lowest among IL-9R^PR pmel T cells (Fig. 6E). Pseudotime analysis anchored on the Tcf7⁺ Tscm-like cluster further supported an earlier differentiation state for IL-9R^AQ, intermediate for IL-9R^WT, and later for IL-9R^PR (Fig. 6G, Fig. S6G). In summary, compared to IL-9R^WT, IL-9R^AQ drives a memory and stem-like state at the expense of effector differentiation, while IL-9R^PR drives a purely effector state with early features of exhaustion.

Phenotypic analysis of T cells seven days after ACT into mice bearing B16-F10 melanoma confirmed more T_N/T_CM cells with IL-9R^AQ across tumor, spleen, and tumor draining lymph node (tdLN) (Fig. S6H). The phenotypic analysis was limited to IL-9R^WT and IL-9R^AQ T cells in the spleen, draining lymph node (tdLN) and tumor, and IL-9R^3x T cells in the tdLN, due to the poor expansion and tumor-infiltration in other groups (Table S3). We also observed a small increase in frequency of T_EFF cells among IL-9R^WT versus IL-9R^AQ cells in the tumor (a similar trend was observed in the tdLN; Fig. S6I). Tumor-infiltrating IL-9R^WT pmel T cells also demonstrated greater frequency of Granzyme B (p=0.03) and TNF-α (p=0.09) expression than IL-9R^AQ pmel T cells (Fig. S6J)., explaining their similar anti-tumor function despite fewer cells (Fig. 5H, Fig. S5G).

To link signaling to transcription, we correlated STAT phosphorylation (20 minutes) with bulk RNA-seq (48 hours) across six conditions (IL-9R, IL-9R^AQ, IL-9R^PR or IL-9R^5x plus IL-9, o9R + oIL-2, and IL-9R + IL-2; Fig. 6H-I). The top 100 pSTAT-correlated genes demonstrated the entirely unique gene programs induced by pSTAT1 and pSTAT5 (Fig. 6J, Table S4). Sixty-six of 100 genes most correlated with pSTAT3 and pSTAT4 were shared. pSTAT1 had the strongest gene expression correlation, reflecting the range and even distribution of pSTAT1 activation across the sample set (Fig. 6K, Fig. S6K).

Gene expression modules based on the top pSTAT-correlated genes were projected onto the scRNA-seq data from Fig. 6A. The pSTAT1 module distinguished IL-9R^PR from IL-9R^WT and IL-9R^AQ pmel T cells, validating the fidelity between the in vitro and in vivo transcriptomic effects of IL-9R signaling (Fig. 6L). We did not find differences in pSTAT3, pSTAT4, or pSTAT5 gene modules between the groups (Fig. S6L), though these were less sensitive for detecting differences given the lower correlation between gene expression and STAT phosphorylation.

With exogenous IFN-β, we selectively increased pSTAT1 in IL-9R^WT and IL-9R^AQ T cells to those of IL-9R^PR T cells (Fig. S6M). To evaluate the functional impact of STAT1 modulation, we focused on the T_CM subset, as our previous experiments showed that IL-9R^AQ, IL-9R^WT, and IL-9R^PR T cells exhibited graduated T_CM phenotypes both in vitro and in vivo (IL-9R^AQ > IL-9R^WT > IL-9R^PR). The increase in pSTAT1 reduced the T_CM phenotype in IL-9R^AQ and IL-9R^WT T cells to levels observed in IL-9R^PR T cells (Fig. S6N), supporting the STAT1-driven differentiation observed in our single cell analyses.

We also sought to understand if the low pSTAT1 signal of IL-9R^AQ T cells contributed to their less-differentiated state. Indeed, transcription factor inference from the in vitro RNA-seq data identified STAT1 and related TFs (IRF1, IRF2 and IRF9), as the dominant distinguishing feature between IL-9R^WT and IL-9R^AQ T cells (Fig. 6N). SCENIC analysis of the scRNA-seq dataset also demonstrated a modest increase in the STAT1 regulon in tumor-infiltrating IL-9R^WT compared to IL-9R^AQ pmel T cells (Fig. 6O). In fact, among STAT1, STAT3, STAT4 and STAT5, STAT1 was the only regulon enriched in IL-9R^WT compared to IL-9R^AQ pmel T cells.

We noted that STAT3, STAT4 and STAT5 regulons were enriched in tumor-infiltrating IL-9R^AQ pmel T cells compared to both IL-9R^WT and IL-9R^PR pmel T cells. Furthermore, genes differentially expressed by IL-9R^AQ T cells were enriched for genes from Hallmark IL2/STAT5 signaling and IL6/JAK/STAT3 signaling gene sets (adjusted p-value = 0.001 and 0.07, respectively; Fig. S6O). This is consistent with the stem, memory^22,31 and proliferative qualities^21,32 of IL9R^AQ pmel T cells (Fig. 6N), but is also partly counterintuitive considering that IL-9R^AQ results in stronger pSTAT3 and pSTAT4 (and similar pSTAT5) compared to IL-9R^WT. We speculate that differences between STAT profiles based on proximal signaling versus in vivo transcriptomics may relate to complex downstream interplay and competition between STAT proteins (see discussion).

Overall, T cell intrinsic STAT1 titrates fate from a less-differentiated, stem-like state (IL-9R^AQ, STAT1-low) to a terminally differentiated effector (IL-9R^PR, STAT1-hi), with IL-9R^WT (STAT1-mid) encompassing both stem-like and effector states.

Potency and sensitivity of proximal IL-9R signaling for human CAR T cells

In human T cells, IL-9R produced stronger phosphorylation of STAT1, STAT3 and STAT5 than human orthogonal chimeric o9R (ho9R), and induced pSTAT4 (Fig. 7A). In repetitive tumor killing assays with NY-ESO-1 TCR (1G4) effectors and human melanoma M407^mCherry tumor cell targets, hIL-9R + IL-9 outperformed ho9R + hoIL-2 (Fig. S7A).

Figure 7. The balance of JAK/STAT signaling through IL-9R and IL-9R variants, and their impact on anti-tumor activity, is conserved in human CAR T cells.

(A) Dose-response curves of each pSTAT in human T cells transduced with human IL-9R or o9R (YFP⁺) and stimulated with either human IL-9 or oIL-2 for 20 minutes. Data are representative of two independent donors.

(B) Schematic of Nalm6 human leukemia orthotopic model in NSG mice treated with human CAR T cells targeting CD19 (CD19-BBz) and co-expressing human IL-9R. Mice were also treated with human IL-2 or IL-9 (or no cytokine).

(C) Weight of NSG mice (n=5 mice/group) in response to treatment in (B).

(D) Bioluminescence (photons/second) of tumors in mice treated as in (B).

(E) Survival curves of mice from (D).

(F) Schematic of 143B human osteosarcoma orthotopic solid tumor model in NSG mice treated with human CAR T cells targeting Her2 (Her2-BBz) and co-expressing IL-9R (or not) and treated with human IL-9 (or not).

(G) Tumor growth as measured by leg volume (n=5 mice/group) as treated per (F).

(H) Survival curves of mice from (G).

(I) Dose-response curves of each pSTAT in human T cells transduced with human IL-9R, IL-9R^PR or IL-9R^AQ and stimulated with IL-9 for 20 minutes. Shown are technical duplicates and representative of two independent donors. EC50 and Emax are shown in the table.

(J) Absolute number of (YFP⁺) CAR T cells in blood twelve days after tumor inoculation (8 days after ACT).

(K) Tumor growth as measured by leg volume of mice (n=5 mice/group) treated as in (F).

(L) Survival curves of mice from (K).

*P < 0.05, **P < 0.005, ***P < 0.0005, ****P < 0.0001 (two-way ANOVA for D, G, I, J; Mantel-Cox test for E, H, K; one-way ANOVA for L).

We co-engineered CD19-BBz CAR T cells with human IL-9R (hIL-9R; Fig. S7B). In repetitive killing assay with NALM6 tumor cells, IL-9R plus IL-9 outperformed the same cells cultured with no cytokine or IL-2 between five and nine rounds of tumor killing (Fig. S7C). In Nalm6 leukemia-bearing NSG mice, IL-2 caused rapid weight loss; IL-9 did not (Fig. 7B-C). Dual CD19-BBz CAR plus IL-9R T cells resulted in prolonged tumor control and survival (Fig. 7D-E), whereas IL-9 had no effect in the absence of IL-9R-engineered T cells.

Against an orthotopic solid tumor (143B osteosarcoma) expressing Her2 (Fig. 7F), CAR T cells targeting Her2 (H2-BBz) were minimally effective alone (Fig. 7G-H), but H2-BBz hIL9R T cells in mice treated with IL-9 improved tumor control and survival (Fig. 7G-H).

Signaling patterns of IL-9 mutant receptors, IL-9R^AQ and IL-9R^PR, were generally conserved between mouse and human systems (Fig. S7D-E): IL-9R^AQ reduced and IL-9R^PR increased pSTAT1 (Fig. 7I); pSTAT5 levels were similar; and pSTAT3 and pSTAT4 were modestly lower than IL-9R^WT. Unlike the mouse system, we observed slight differences pSTAT3 and pSTAT4 levels between human IL-9R^AQ and IL-9R^PR mutants (although for STAT3 this was only observed in selected donors). In vitro, IL-9R^AQ T cells expanded more, and IL-9R^PR expanded less, than IL-9R^WT (Fig. S7F). In vivo, the expansion and enrichment of CAR T cells co-engineered with IL-9R^AQ was superior to IL-9R^WT or IL-9R^PR (Fig. 7J, Fig. S7G). Treatment with IL-9 improved tumor control with CAR T cells co-engineered with IL-9R^WT and IL-9R^AQ, but not IL-9R^PR (Fig. 7K), consistent with the inferior anti-tumor functions of mouse IL-9R^PR pmel T cells. Survival with IL-9R^WT CAR T cells exceeded IL-9R^PR (p=0.078) and IL-9R^AQ CAR T cells, though survival after IL-9R^AQ CAR T cells was confounded by toxicity from vigorous CAR expansion.

Consistent with the in vivo data, IL-9R^WT and IL-9R^AQ outperformed IL-9R^PR in the repetitive killing of M407^mCherry with NY-ESO-1 TCR effectors (Fig. S7H). IL-9R^AQT cells expanded more after each round of killing (Fig. S7I), yet IL-9R^WT maintained similar anti-tumor efficacy (Fig. S7F), consistent with a higher frequency of Granzyme B expression among IL-9R^WT co-transduced cells compared to IL-9R^AQ (Fig. S7J) after three rounds of tumor killing. These findings mirrored our results with mouse T cells and further supported the higher effector capacity of IL-9R^WT versus IL-9R^AQ on a per cell basis.

Altogether, human IL-9R signaling, its single amino acid mutational tuning, and resultant expansion/efficacy mirror the mouse system, supporting its role as a driver of anti-tumor functions in T cells that is exquisitely sensitive to its optimal balance of JAK/STAT signaling.

DISCUSSION

Despite clinical successes^33,34, T cell therapies for solid tumors remain constrained by the need for large-scale manufacturing and toxic lymphodepletion, suboptimal trafficking, chronic antigen exposure and hostile tumor microenvironments. Cytokines offer a regulatable approach to modulate engineered T cells in vivo, but dose-limiting toxicities have limited their therapeutic potential^35-38.

Here, we report that IL-9 stands out among γ_c cytokines because its private receptor (IL-9R) shows remarkably restricted normal tissue expression. This likely explains why high doses of half-life extended IL-9 were well-tolerated and induced negligible lymph node transcriptional perturbation. These data do not preclude important roles for rare IL-9R expressing cells³⁹: IL-9 participates in type 2 immunity and helminth defense^40-43 and can amplify airway inflammation^42,44,45. Even so, IL-9 deficient mice are broadly healthy with normal immune functions^46,47. And based on population genetics, IL-9 scored lowest among canonical cytokines with respect to its role in evolutionary fitness.

Leveraging the paucity of IL-9R expression, coadministration of IL-9 with small numbers of IL-9R engineered tumor-specific T cells generated robust anti-tumor responses without preconditioning lymphodepletion, outperforming our benchmark o9R system. These findings were consistent across multiple mouse and human models, as well as data from Castelli, et al (REF). IL-9R T cells demonstrate superior signaling, transcriptomic, and biological outputs compared to the synthetic o9R system. The native, high affinity IL-9/IL-9R interaction likely contributes to stronger proximal signaling than intentionally low-affinity synthetic pairs (o9R and oIL-2)^4,48.

The stronger IL-9/IL-9R signal revealed the unexpected phosphorylation of STAT4, providing a route to deliver STAT4 effector programs without the toxicities that have limited IL-12^17,37. At the same time, IL-9 preserves STAT5-associated proliferation and effector capacity^21,49 and STAT3-associated stemness²².

Our perturbation studies uncovered a “goldilocks” window: both attenuation and amplification of iL-9R signaling impaired anti-tumor efficacy. Likewise, T cells were exquisitely sensitive to changes in pSTAT stoichiometry⁵⁰. Leveraging two IL-9R single amino acid variants with divergent pSTAT1 signals, we examined the role of T cell intrinsic STAT1 in anti-tumor functions. STAT1 has anti-proliferative effects in T cells^26,27, but has also been linked to clonal expansion, memory formation⁵¹, and effector function. Amplification of pSTAT1 (IL-9R^PR) diminished expansion and reduced anti-tumor efficacy, while attenuation of pSTAT1 (IL-9R^AQ) resulted in superior expansion, particularly in vivo, accompanied with better tumor infiltration.

Subtle shifts in pSTAT stoichiometry have potent effects on cell fate, even in the presence of an active TCR stimulus. Transcriptomic differences driven by IL9R variants were captured along a pSTAT1 axis, with high pSTAT1 driving the development of a terminally differentiated effector state (IL-9R^PR), low pSTAT1 yielding a proliferative, stem- and memory-like state (IL-9R^AQ), and IL-9R^WT occupying an intermediate state. This is consistent with work demonstrating that interferon signaling, through IRF2, promotes CD8⁺ T cell exhaustion⁵², and also provides a plausible explanation for why JAK/STAT inhibition in the context of chronic interferon signaling may restore T cell effector functions⁵³. However, pSTAT1 may not be entirely detrimental to anti-tumor immunity: lower pSTAT1 signals (IL-9R^AQ) restrained effector differentiation, and IL-9R^AQ pmel T cells did not improve anti-tumor efficacy despite superior T cell expansion.

The reduced pSTAT1 signal produced by IL-9R^AQ was also associated with an enrichment in STAT3 and STAT5 activity inferred from single cell transcriptomic data, consistent with their stem, memory^22,31 and proliferative qualities^21,32 of IL9R^AQ pmel T cells. However, this was partly unexpected because IL-9R^AQ produces a weaker pSTAT3 signal than IL-9R^WT by phosphoflow (and similar pSTAT5 signal). We speculate that the discrepancy between STAT activity assessed by phosphorylation and STAT activity inferred from scRNA-seq data may be related to downstream competition between simultaneously active STAT proteins, a phenomenon described in the fate of Th17 cells⁵⁴. The weaker pSTAT1 signal of IL-9R^AQ may result in less competition with pSTAT3, such that even a relatively weak pSTAT3 signal generates a stronger downstream biologic effect.

In summary, our study uncovers and leverages the uniquely restricted normal tissue expression of IL-9R. The IL-9/IL-9R is a natural cytokine-receptor pair with orthogonal qualities, and in models of engineered T cell therapy, outperforms its synthetic counterpart (o9R). The activity of IL-9R can be linked to its higher affinity and more potent signaling, which unexpectedly includes STAT4, not previously considered a primary target of γ_c signaling. The highly potent anti-tumor functions of T cells signaling through IL-9R – validated in multiple mouse and human CAR T cell models against hematologic and solid tumors – can be traced to an optimal strength and balance of IL-9R signals, with pSTAT1 tuning cells between a stem- or memory-like state and a terminal effector state.

Limitations of the Study

The safety of systemic IL-9 in humans will require clinical trials, including consideration of potential tumor-promoting roles⁵⁵. Our results must be considered in the context of the JAK/STAT profile of the IL-9R where multiple STATs are simultaneously active. Although our manuscript highlights the major role for pSTAT1 in distinguishing between IL-9R^WT, IL-9R^AQ and IL-9R^PR, we cannot exclude minor roles for other STATs.

STAR METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

Experimental procedures in mouse studies were approved by the Institutional Animal Care and Use Committee (IACUC) at the Stanford University (animal protocol ID 34340 and 33698) and performed in accordance with the guidelines from the animal facility of Stanford University. Mice were housed in animal facilities approved by the Association for the Assessment and Accreditation of Laboratory Care. Five- to eight-week-old C57BL/6 (C57BL/6J) or pmel-1 TCR/Thy1.1 (pmel; B6.Cg-Thy1a/Cy Tg(TcraTcrb)8Rest/J) transgenic mice were purchased from the Jackson laboratory and maintained in the Stanford University-Research Animal Facility (RAF) Facility. NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were purchased from the Jackson Laboratory and maintained in the Stanford University Lorry Lokey (SIM1) facility by the Stanford Veterinary Service Center before transfer to the Stanford University Comparative Medicine Pavilion for experimental procedures. All T cell adoptive transfers were performed using female donors.

Cell lines and cell culture

The B16-F10 mouse melanoma cell line was purchased from ATCC and cultured with RPMI 1640 with GlutaMAX (Thermo Fisher Scientific) containing 10% fetal bovine serum (FBS, Omega Scientific) and penicillin-streptomycin (100 μg ml⁻¹, Thermo Fisher Scientific) (RPMI-C). The KP-gp100 mouse soft-tissue sarcoma cell line was derived from KP3172, a tumor cell line derived from a Cre-induced sarcomas arising in the hind leg of KP (Kras^LSL-G12D, p53^flox) mice. KP3172 was retrovirally transduced with a construct containing mouse gp100 with mutations KVP at amino acids 25-27 to generate KP-gp100; the cell line was maintained in RPMI-C. Primary mouse T cells derived from C57BL/6 or pmel transgenic mice were cultured in RPMI-c plus 50 μM 2-mercaptoethanol (Gibco), non-essential amino acids (Gibco), sodium pyruvate (Gibco), 20 mM HEPES (Gibco), and GlutaMAX (Gibco).

Primary human T cells were cultured in complete AIM-V media (10% fetal bovine serum, 100 U/mL penicillin, 100 μg/mL streptomycin, 2mM GlutaMAX, 10mM HEPES (Thermo Fisher Scientific) and 100 U/mL recombinant human IL-2; Peprotech). HEK293T and HEK293GP cells were purchased from ATCC and cultured in high-glucose DMEM (Gibco) supplemented with 10% FBS, 2mM GlutaMAX, 10mM HEPES, and penicillin-streptomycin (DMEM-c). Expi293F cells were purchased from Thermo Fisher Scientific and cultured in Expi293 Expression Medium (Thermo Fisher Scientific). M407 was acquired from Dr. Antoni Ribas (UCLA) as part of a material transfer agreement, lentivirally transduced with a construct expressing mCherry (M407^mCherry), and cultured in DMEM-c. Nalm6-FFluc cells and 143B-GFP-FFluc cells were generous gifts from Crystal Mackall and cultured in RPMI-C and DMEM-C, respectively. Cell lines were periodically authenticated (ATCC) and periodically tested for mycoplasma infection using a mycoplasma detection kit (Vazyme).

METHOD DETAILS

Protein production

DNA encoding mouse and human IL-2 and IL-9, and mouse IL-9^Q115T were cloned into the mammalian expression vector pD649, which includes a C-terminal 8xHis tag for affinity purification. DNA encoding mouse serum albumin (MSA) was purchased from Integrated DNA Technologies (IDT) and cloned into pD649 as an N-terminal fusion. Mammalian expression DNA constructs were transfected into Expi293F cells using the Expi293 Expression System (Thermo Fisher Scientific) for secretion and purified for the clarified supernatant by nickel affinity resin (Ni-IMAC, Thermo Fisher Scientific) followed by size-exclusion chromatography with a Superdex-200 column (Cytiva) and formulated in sterile phosphate-buffered saline (PBS) for injection. Endotoxin was removed using the Proteus NoEndo HC Spin column kit following the manufacture’s recommendations (VivaProducts) and endotoxin removal was confirmed using the Pierce LAL Chromogenic Endotoxin Quantification Kit (Thermo Fisher Scientific). Proteins were concentrated and stored at −80°C until use.

Mammalian Expression Vectors

cDNA encoding mouse orthogonal IL-2Rβ extracellular domain (ECD) and gene block cDNA encoding mouse ICDs of IL-9R (IDT) were cloned into the retroviral vector pMSCV-MCS-IRES-YFP by PCR and isothermal assembly (ITA). Human orthogonal IL2Rβ-ECD-IL9R-ICD (ho9R) were similarly cloned into the pMSCV vector. cDNA encoding the native mouse and human IL9R-P2A-YFP were synthesized (VectorBuilder) and cloned into the pMSCV vector. CD19-4-1BBζ and HER2-4-1BBζ constructs were generous gifts from Crystal Mackall.⁵⁶ NY-ESO-1 TCR (1G4) was acquired from Dr. Antoni Ribas as part of a material transfer agreement.

Single amino acid mutations of the IL-9R, including those at and near Y405 (mouse) and Y407 (human), and the single amino acid mutation of IL-9 (Q115T), were made through Q5 site-directed mutagenesis (New England Biolabs). To generate IL-9R receptors with repetitive phosphotyrosine motifs (IL-9R^3x and IL-9R^5x), gene blocks encoding amino acids from position 386 to position 425 were repeated three times or five times in tandem, followed by the remainder of the IL-9R intracellular domain.

Retrovirus production

For production of retrovirus to engineer mouse T cells, HEK293T (ATCC) cells were seeded at 4.5 x 10⁵ cells per well in a 6-well tissue culture treated plate. 6-well tissue culture treated plates were coated with 0.01% poly-l-lysine (Sigma-Aldrich) prior to seeding. For each well, 2.5 μg of plasmid (1.25 μg of pMSCV retroviral vector plus 1.25 μg of pCL-Eco packaging vector) was added to 250 μl of Opti-MEM I Reduced Serum Medium (Thermo Fisher Scientific), followed by 5.5 μl of Lipofectamine^™ 3000 reagent (Thermo Fisher Scientific) and 5 μl of P3000^™ enhancer reagent. After 18 to 20 h, the medium was replaced with fresh DMEM-c with 20 mM HEPES and cultured for another 24 hours before viral collection at 48 hours post-transfection.

For production of retrovirus to engineer human T cells, HEK293GP (ATCC) cells were cultured in DMEM-c. Prior to plating of HEK293GP cells, 10cm tissue-culture treated dishes were coated with 5 mL of 0.01% poly-L-lysine. Cells were subsequently plated at a density of 6.5 x 10⁶ ml⁻¹ per dish in 10 mL total DMEM-c and incubated overnight at 37°C. After 24 hours, cells were transfected with 9 μg vector plasmid, 1.5 mL of Mix A (1.5 mL Opti-Mem, 4.5 μg RD114), and 1.5 mL of Mix B (1.5 mL Opti-MEM, 33uL Lipofectamine 3000, and 2uL P3000 per μg vector plasmid) per plate. After 24 hours, fresh media was replenished. Virus was collected at 48- and 72-hours post-transfection. Viral supernatants were filtered through a 0.22 μM Whatman filter (Cytiva). If not used immediately, virus was frozen for storage in −80°C.

Activation and retroviral transduction of primary mouse T cells

For retroviral transduction of mouse T cells, splenocytes from the five- to ten-week-old mice were mechanically digested and filtered through a 70 μM strainer (Fisher Scientific). Red blood cells were lysed with eBioscience RBC Lysis Buffer (Thermo Fisher Scientific) for 5 minutes at 4°C. Splenocytes were resuspended in PBS with 2% FBS and 1 mM Ethylenediaminetetraacetic acid (EDTA, Thermo Fisher Scientific), and enriched for CD3⁺ cells by magnetic bead separation, according to the manufacturer's protocol (StemCell). C57BL/6-derived mouse T cells were activated with a 1:1 bead:cell ratio of mouse CD3/28 Dynabeads^™ (Thermo Fisher Scientific) in fresh T cell medium with 100 U ml⁻¹ rmIL-2 overnight. On day of transduction, CD3/28 Dynabeads^™ were removed prior to transduction by magnetic bead separation. Isolated pmel T cells were activated with 100 U ml⁻¹ recombinant mouse IL-2 (rmIL-2) (Peprotech) and 1 μg ml⁻¹ human gp100 peptide (Anaspec) the day before transduction. One day before transduction, six-well non-tissue culture treated plates (Thermo Fisher Scientific) were coated with 37.5 μg ml⁻¹ retronectin (Takara Bio) in PBS per well and placed at 4°C overnight. The following day, retronectin-coated plates were blocked with 0.5% BSA (Sigma-Aldrich) in PBS for 30 minutes at 25°C and washed out with PBS. Viral supernatant (2 ml) was added to each well and spun at 2000g for two hours at 32°C without brake. Viral supernatant was carefully removed, 3 ml of activated T cells (1 x 10⁶ ml⁻¹) were added to each well with 100 U ml⁻¹ rmIL-2, and spun at 2000g for 10 minutes at 32°C without brake. Cells were then cultured for 18-22 hours at 37°C. Cells were collected via gentle pipetting and resuspended at 1 x 10⁶ ml⁻¹ in fresh mouse T cell medium with 100 U ml⁻¹ rmIL-2 and expanded overnight before further downstream cellular assays. For in vivo tumor assays, transduced cells were used either immediately after collection or one day post-transduction. YFP served as a surrogate marker for the expression of all mouse constructs used and was checked on day of use (Table S5).

Activation and retroviral transduction of primary human T cells

Healthy donor Leukocyte Reduction System (LRS) chambers were purchased from the Stanford Blood Center according to an institutional review board (IRB)-exempt protocol. Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque Plus (Cytiva) density gradient centrifugation and frozen in a Mr. Frosty^™ Freezing Container in aliquots of 10 x 10⁶ PBMCs mL⁻¹ of CELLBANKER 2 serum-free cell cryopreservation medium (AMSBIO). Aliquots were stored frozen at −80°C. Human T cells were isolated from cryopreserved PBMCs using the EasySep^™ Human T Cell Isolation Kit (StemCell) according to the manufacturer’s protocol. T cells were activated using Dynabeads^™ Human T-Activator CD3/CD28 (Thermo Fisher Scientific) for T Cell expansion and activation at a 3:1 ratio of beads:cells and cells were incubated at 37°C for three days at a concentration of 1 x 10⁶ cells ml⁻¹. Transduction was performed on days 3 and 4 after activation. Twelve-well, non-tissue culture treated plates (Thermo Fisher Scientific) were coated with 18.75 μg mL⁻¹ retronectin (Takara Bio) in PBS per well and left at 4°C overnight or for two hours at 25°C. On the day of transduction, the retronectin was transferred to a new 12-well non-tissue culture treated plate, which was subsequently stored at 4°C overnight. The original retronectin-treated plate was then blocked with 1 mL of 1% bovine serum albumin (BSA; Sigma-Aldrich) in PBS per well for 10-20 minutes. Afterward the BSA was removed and 1 mL of 48 hour retroviral supernatant was added per well. For co-transductions, 1 mL of supernatant per construct was added per well (for a total of 2 mL per well). Following the addition of viral supernatant, the plate was spun at 2000 x g for 2 hours at 32°C. Subsequently, viral supernatant was aspirated, and the activated Day 3 T cells were added at a concentration of 0.5 x 10⁶ cells ml⁻¹ in 1 mL per well. The plate was then incubated at 37°C overnight. The transduction was repeated the following day using 72 hour supernatant. On Day 5 post activation, the DynaBeads were magnetically removed from T cells and cells were resuspended in fresh media at a concentration of 0.5 x 10⁶ cells ml⁻¹. On Day 7 post activation, T cells were passaged and resuspended in fresh media at a concentration of 0.5 x 10⁶ cells ml⁻¹. On Day 10, transduction efficiency was assessed by flow cytometry and T cells were used in subsequent assays.

Quantitative Reverse-Transcriptase PCR (qRT-PCR)

Total RNA was extracted from human cells with the RNeasy Mini Kit (Qiagen) as directed by the manufacturer. 0.6 μg of total RNA was reversed transcribed to synthesize cDNA, using High Capacity cDNA Reverse Transcription Kit (ThermoFisher). Real-time quantitative PCR was run in duplicate using iTaq SYBRGreen (Bio-Rad), following manufacturer’s instructions. PCR amplification was monitored using QuantStudio5 detection system (AppliedBiosystems). Data were expressed as relative mRNA abundance normalized to GAPDH expression level in each sample. The primer sequences are listed in the key resources table.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Phospho-Stat1 (Tyr701) (58D6) Rabbit mAb (Alexa Fluor® 647 Conjugate)	Cell Signaling Technologies	Cat: 8009S
Phosflow™ Alexa Fluor® 647 Mouse Anti-Stat3 (pY705) (Clone 4/P)	BD Biosciences	Cat: 557814
Phosflow™ Alexa Fluor® 647 Mouse Anti-Stat4 (pY693) (Clone 38/P)	BD Biosciences	Cat: 558137
Phosflow™ Alexa Fluor® 647 Mouse Anti-Stat5 (pY694) (Clone 47)	BD Biosciences	Cat: 612599
Phosflow™ Alexa Fluor® 647 Mouse Anti-Mouse Stat6 (Clone pY641)	BD Biosciences	Cat: 558242
Cd8a Monoclonal Antibody (53-6.7), PE-eFluor 610, eBioscience™	Thermo Scientific	Cat: 61-0081-82
BD OptiBuild™ BV786 Rat Anti-Mouse CD8b.2 (Clone 53-5.8)	BD Biosciences	Cat: 740843
Anti-CD90.1 Mouse Monoclonal Antibody (Brilliant Violet® 605) (clone: OX-7)	Biolegend	Cat: 202537
Anti-CD90.1 Mouse Monoclonal Antibody (Brilliant Violet® 711) (clone: OX-7)	Biolegend	Cat: 202539
BD OptiBuild™ BUV395 Rat Anti-Mouse CD274 (Clone: MIH5)	BD Biosciences	Cat: 745616
BD OptiBuild™ BUV563 Rat Anti-Mouse CD115 (CSF-1R) (Clone: T38-320)	BD Biosciences	Cat: 748478
BD OptiBuild™ BUV615 Rat Anti-Mouse Siglec-H (Clone: 440c)	BD Biosciences	Cat: 751024
BD Horizon™ BUV661 Rat Anti-CD11b (Clone: M1/70)	BD Biosciences	Cat: 612977
F4/80 Monoclonal Antibody (BM8), eFluor™ 450, eBioscience™	Thermo Scientific	Cat: 48-4801-82
Primary Antibodies › CD11c Antibodies Invitrogen CD11c Monoclonal Antibody (N418), eFluor™ 506, eBioscience™	Thermo Scientific	Cat: 69-0114-82
Brilliant Violet 570™ anti-mouse/human CD45R/B220 Antibody	Biolegend	Cat: 103237
BD Horizon™ BV605 Rat Anti-Mouse Ly-6C (Clone AL-21)	BD Biosciences	Cat: 563011
CD80 (B7-1) Monoclonal Antibody (16-10A1), PE-Cyanine5, eBioscience™	Thermo Scientific	Cat: 15-0801-82
PE/Fire™ 700 anti-mouse CX3CR1 Antibody	Biolegend	Cat: 149051
PE/Cyanine7 anti-mouse CD86 Antibody	Biolegend	Cat: 105014
LIVE/DEAD™ Fixable Blue Dead Cell Stain Kit, for UV excitation	Thermo Scientific	Cat: L34962
BD OptiBuild™ BV480 Hamster Anti-Mouse CD279 (PD-1) (Clone: J43)	BD Biosciences	Cat: 746784
BD Horizon™ BV510 Rat Anti-Mouse CD44 (Clone: IM7)	BD Biosciences	Cat: 563114
BD Horizon™ BV650 Rat Anti-Mouse CD62L (Clone:MEL-14)	BD Biosciences	Cat: 564108
CD27 Monoclonal Antibody (LG.7F9), Super Bright™ 702, eBioscience™	Thermo Scientific	Cat: 67-0271-82
BD OptiBuild™ BV786 Rat Anti-Mouse CD134 (Clone:OX-86)	BD Biosciences	Cat: 740945
PE anti-mouse CD129 (IL-9R) Antibody	Biolegend	Cat: 158704
APC Anti-mouse TCR γ/δ Antibody	Biolegend	Cat: 118115
BD Horizon™ R718 Rat Anti-Mouse CD127 (Clone: A7R34)	BD Biosciences	Cat: 567540
BD Horizon™ BUV737 Rat Anti-Mouse CD19 (Clone: 1D3)	BD Biosciences	Cat: 612782
CD4 Monoclonal Antibody (GK1.5), NovaFluor™ Blue 610-70S, eBioscience™	Thermo Scientific	Cat: M001T02B06
Primary Antibodies › CD103 (Integrin alpha E) Antibodies Invitrogen CD103 (Integrin alpha E) Monoclonal Antibody (2E7), PerCP-eFluor™ 710, eBioscience™	Thermo Scientific	Cat: 46-1031-82
CD8a Monoclonal Antibody (53-6.7), NovaFluor™ Yellow 690, eBioscience™	Thermo Scientific	Cat: M003T02Y05
BD Horizon™ BUV805 Rat Anti-Mouse CD366 (TIM-3) (Clone: RMT3-23)	BD Biosciences	Cat: 569859
Brilliant Violet 421™ anti-mouse CD25 Antibody	Biolegend	Cat: 102043
CD45 Monoclonal Antibody (30-F11), Alexa Fluor™ 532, eBioscience™	Thermo Scientific	Cat: 58-0451-82
PerCP anti-mouse I-A/I-E Antibody	Biolegend	Cat: 107624
BD Horizon™ BB700 Mouse Anti-Mouse NK-1.1 (Clone: PK136)	BD Biosciences	Cat: 566502
PE/Fire™ 810 anti-mouse Ly-6G Antibody	Biolegend	Cat: 127673
Primary Antibodies › CD3 Antibodies Invitrogen CD3e Monoclonal Antibody (145-2C11), APC-eFluor™ 780, eBioscience™	Thermo Scientific	Cat: 47-0031-82
SC2807-Ab-TurboCHO-Express 2.0 rAb Type:hlgG (custom FM63 idiotype)	GenScript	n/a
Recombinant Human ErbB2/Her2 Fc Chimera Protein, CF	R&D	Cat: 1129-ER-01M
PE anti-human CD19 Antibody	Biolegend	Cat: 302208
BD OptiBuild™ BV421 Biosimilar Anti-Human HER2 Trastuzumab297.rMAb	BD Biosciences	Cat: 756616
BD Horizon™ BV421 Mouse IgG2b, κ Isotype Control, Clone 27-35	BD Biosciences	Cat: 562748
PE Mouse IgG2b, κ Isotype Ctrl Antibody	Biolegend	Cat: 401208
Annexin V (Alexa Fluor 647)	Thermo Fisher Scientific	Cat: A23204
Brilliant Violet 605™ anti-mouse CD95 (Fas) (cloneSA367H8)	BioLegend	Cat:152612
Alexa Fluor® 700 anti-mouse/human CD44 Antibody (Clone: IM7)	BioLegend	Cat: 103026
Zombie Violet™ Fixable Viability Kit	BioLegend	Cat: 423113
BD OptiBuild™ BV786 Rat Anti-Mouse CD8b.2 (Clone: 53-5.8)	BD Bioscience	Cat: 740843
Ly-6A/E (Sca-1) Monoclonal Antibody (D7), PerCP-Cyanine5.5, eBioscience™	Invitrogen/eBioscience	Cat: 45-5981-82
PE/Cyanine7 anti-mouse CD62L Antibody	BioLegend	Cat: 104418
Alexa Fluor® 647 anti-mouse/human CD44 Antibody (Clone: IM7)	BioLegend	Cat: 103018
Anti-CD62L Rat Monoclonal Antibody (APC (Allophycocyanin)/Cy7®) [clone: MEL-14]	BioLegend	Cat: 104428
Anti-CD45 Rat Monoclonal Antibody (PB (Pacific Blue)) [clone: 30-F11]	BioLegend	Cat: 103126
APC/Cy7 anti-mouse TNF-α [MP6-XT22]	BioLegend	Cat: 506344
Granzyme B Monoclonal Antibody (GB11), APC	Thermo Scientific	Cat: GRB05
Bacterial and virus strains
NEB® 5-alpha Competent E. coli, High Efficiency	New England Biolabs	Cat: C2987H
One Shot Stbl3 Chemically Competent	Thermo Fisher Scientific	Cat: C737303
Q5® Site-Directed Mutagenesis Kit, Without Competent Cells	New England Biolabs	Cat: E0554S
Biological samples
PBMC from Heathy Human Subjects	Stanford Blood Center	N/A
Chemicals, peptides, and recombinant proteins
gp100 (25–33), human	Anaspec	Cat: AS-62589
Recombinant Murine IL-2	Peprotech	Cat: 212-12
Recombinant Murine IL-9	Peprotech	Cat: 219-19
Retronectin	Takara	Cat: T100B
Recombinant Human IL-2	Peprotech	Cat: 200-02
Recombinant Human IL-9	Peprotech	Cat: 200-09
DNase I, Grade II, from bovine pancreas	Sigma-Aldrich	Cat: 10104159001
Collagenase D,from Clostridium histolyticum	Sigma-Aldrich	Cat: 11088866001
Tumor Dissociation Kit, mouse	Miltenyi-Biotec	Cat: 130-096-730
Lung Dissociation Kit, mouse	Miltenyi-Biotec	Cat: 130-095-927
Liver Dissociation Kit, mouse	Miltenyi-Biotec	Cat: 130-105-807
ACK Lysis Buffer	Thermo Scientific	Cat: A1049201
MSA-mIL2	Kalbasi Lab
MSA-moIL2	Garcia Lab
MSA-mIL9	Kalbasi Lab
MSA-mIL9Q115T	Kalbasi Lab
MSA-hoIL2	Garcia Lab
MSA-hIL2	Kalbasi Lab
MSA-hIL9	Kalbasi Lab
Critical commercial assays
Ebioscience™ 1X RBC Lysis Buffer	Thermo Scientific	Cat: 00-4333-57
EasySep™ Mouse T Cell Isolation Kit	STEMCELL Technologies	Cat: 19851
Dynabeads™ Mouse T Activator CD3/CD28 for T Cell Expansion and Activation, Monoclonal	Thermo Scientific	Cat: 11453D
Dynabeads™ Human T-Activator CD3/CD28 for T Cell Expansion and Activation	Thermo Scientific	Cat: 11-131-D
Aim V™ Medium, liquid (research grade)	Thermo Scientific	Cat: 12055091
Zombie Violet™ Fixable Viability Kit	Biolegend	Cat: 423113
CellTrace™ Violet Cell Proliferation Kit, for flow cytometry	Thermo Scientific	Cat: C34557
Ebioscience™ Flow Cytometry Staining Buffer	Thermo Scientific	Cat: 00-4222-26
Lipofectamine 3000 Transfection Reagent Kit	Thermo Scientific	Cat: L3000015
Poly-L-lysine solution,mol wt 150,000-300,000, 0.01%, sterile-filtered, BioReagent, suitable for cell culture	Sigma-Aldrich	Cat: P4832-50ML
Cytofix™ Fixation Buffer	BD Biosciences	Cat: 554655
BD Phosflow™ Perm Buffer III	BD Biosciences	Cat: 558050
Expi293™ Expression System Kit	Thermo Scientific	Cat: A14635
Pierce™ High Capacity Ni-IMAC Resin, EDTA Compatible	Thermo Scientific	Cat: A50586
Annexin V Binding Buffer	Biolegend	422201
eBioscience FoxP3/Transcription Factor Staining buffer Set	Thermo Scientific	Cat: 00-5523-00
eBioscience Flow Cytometry Staining Buffer	Thermo Scientific	Cat: 00-4222-26
Deposited data
Bulk RNA-seq (Related to Figure 3 and Figure 6)	Kalbasi Lab	Zenodo: 14537359
Single cell RNA-seq (Related to Figure 6)	Kalbasi Lab	Zenodo: 14537359
Experimental models: Cell lines
B16-F10 mouse melanoma cell line	ATCC	Cat: CRL-6475
KP-gp100 mouse sarcoma cell line	Kalbasi Lab	N/A
NALM6 Firefly Luciferase (NALM6-WT) human leukemia cell line	Mackall Lab	N/A
143B human osteosarcoma cell line	Mackall Lab	N/A
HEK293T cell line	ATCC	Cat: CRL-3216
HEK293GP cell line	Majzner Lab	N/A
Expi293F cell line	Thermo Fisher Scientific	Cat: A14527
M407 human melanoma cell line	Ribas Lab	N/A
Experimental models: Organisms/strains
C57BL/6J	The Jackson Laboratory	Cat: #000664
B6.Cg-Thy1a/Cy Tg(TcraTcrb)8Rest/J (pmel-1)	The Jackson Laboratory	Cat: #005023
NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG)	Stanford Veterinary Service Center	N/A
Recombinant DNA
pMSCV-mIL9R-P2A-YFP (mIL9R-WT)	Kalbasi Lab	N/A
pMSCV-mIL9R-mut(YLAQ)-P2A-YFP (mIL9R-AQ)	Kalbasi Lab	N/A
pMSCV-mIL9R-mut(YLPR)-P2A-YFP (mIL9R-PR)	Kalbasi Lab	N/A
pMSCV-mIL9R-3xpY-P2A-YFP (mIL9R-3xpY)	Kalbasi Lab	N/A
pMSCV-mIL9R-5xpY-P2A-YFP (mIL9R-5xpY)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLFQ)-P2A-YFP (mIL9R-YLFQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLGQ)-P2A-YFP (mIL9R-YLGQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLKQ)-P2A-YFP (mIL9R-YLKQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLLQ)-P2A-YFP (mIL9R-YLLQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLNQ)-P2A-YFP (mIL9R-YLNQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLRQ)-P2A-YFP (mIL9R-YLRQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLSQ)-P2A-YFP (mIL9R-YLSQ)	Kalbasi Lab	N/A
pMSCV-mIL9R(YLWQ)-P2A-YFP (mIL9R-YLWQ)	Kalbasi Lab	N/A
MSCV-mIL9R(FLPQ)-P2A-YFP (mIL9R-FLPQ)	Kalbasi Lab	N/A
pMSCV-mo9R-T2A-YFP (mo9R)	Garcia Lab	N/A
pCL-Eco	Addgene	Plasmid #12371
pMSCV-hIL9R-P2A-YFP (hIL9R-WT)	Kalbasi Lab	N/A
pMSCV-hIL9R-mut(YLAQ)-P2A-YFP (hIL9R-AQ)	Kalbasi Lab	N/A
pMSCV-hIL9R-mut(YLPR)-P2A-YFP (hIL9R-PR)	Kalbasi Lab	N/A
MSGV1-FM63-CD8HTM-41BBZ (CD19-BBZ CAR)	Kalbasi Lab	N/A
4D5-CD8HTM-41BBZ (HER2-BBZ CAR)	Kalbasi Lab	N/A
pRD114	Kalbasi Lab	N/A
pMSGV_NY-ESO-1_1G4 (NY-ESO-1 TCR 1G4)	Kalbasi Lab	N/A
pLVX-EF1a-mCherry	Kalbasi Lab	N/A
Software and algorithms
Incucyte Live-Cell Analysis System	Sartorius
Snapgene v7.2.1	SnapGene
Graphpad Prism v10.2.3	GraphPad
FlowJo 10.10.0	BD Biosciences
Ami Aura Imaging Software	Aura
R 4.3.2	Open Source (R Foundation for Statistical Computing)	https://www.r-project.org/

Phosphoflow signaling assays

Actively growing mouse or human primary T cells were rested at a concentration of 2 x 10⁶ cells mL⁻¹ in T cell medium without IL-2 for 12-18 hours before signaling assays. Cells were plated in a 96-well round bottom plate in T cell medium with reduced FBS (2%) for two hours prior to the assay. Cells were stimulated by addition of recombinant cytokines for 20 minutes at 37°C, and the reaction was terminated by fixation with 2.1% paraformaldehyde (BD Cytofix) for 30 minutes at 37°C. Cells were washed and permeabilized with ice-cold methanol (BD Phosflow Perm Buffer III) for 30 minutes on ice or stored at −20°C overnight. Cells were washed with staining buffer (eBioscience) before staining with pSTAT antibodies for 30 min to 1 hour at 4°C in the dark. Cells were washed and analyzed on a CytoFlex (Beckman Coulter) or Novocyte Quanteon/Penteon (Agilent Technologies). Data represent the median fluorescence intensity (MFI), and points were fit to a log(agonist) versus dose–response model using Prism 10.2.3 (GraphPad). For the gating strategy, see Supplementary Figure S3E. Summary of all EC50 and Emax values are provided in Table S6.

Tandem mass tag (TMT)-based global phospho-proteomics

C57BL/6-derived mouse T cells were activated and transduced as previously described. Transduced T cells were sorted (2.5 x 10⁷ per replicate) based on expression of YFP⁺ using an Aria II cell sorter (BD Biosciences) and stimulated for 20 minutes at 37°C using the respective cytokines: rmIL-2 and rmIL-9 in T cell medium. Stimulated T cells were washed twice with PBS. Supernatant were removed, pellet was snap-frozen in liquid nitrogen and stored at −80°C before downstream analyses at the IDeA National Resource for Quantitative Proteomics.

Total protein from each sample was reduced, alkylated, and purified by chloroform/methanol extraction prior to digestion with sequencing grade trypsin and LysC (Promega). The resulting peptides were labeled using a tandem mass tag 10-plex isobaric label reagent set (Thermo), combined into two multiplex sample groups, then enriched using High-Select TiO2 and Fe-NTA phosphopeptide enrichment kits (Thermo) following the manufacturer’s instructions. Both enriched and un-enriched labeled peptides were separated into 46 fractions on a 100 x 1.0 mm Acquity BEH C18 column (Waters) using an UltiMate 3000 UHPLC system (Thermo) with a 50 min gradient from 99:1 to 60:40 buffer A:B ratio under basic pH conditions, then consolidated into 18 super-fractions for the enriched and un-enriched sample sets (Buffer A = 0.1% formic acid, 0.5% acetonitrile; Buffer B = 0.1% formic acid, 99.9% acetonitrile). Both buffers were adjusted to pH 10 with ammonium hydroxide for offline separation. Each super-fraction was then further separated by reverse phase XSelect CSH C18 2.5 um resin (Waters) on an in-line 150 x 0.075 mm column using an UltiMate 3000 RSLCnano system (Thermo). Peptides were eluted using a 75 min gradient from 98:2 to 60:40 buffer A:B ratio. Eluted peptides were ionized by electrospray (2.4 kV) followed by mass spectrometric analysis on an Orbitrap Eclipse Tribrid mass spectrometer (Thermo) using multi-notch MS3 parameters. MS data were acquired using the FTMS analyzer in top-speed profile mode at a resolution of 120,000 over a range of 375 to 1500 m/z. Following CID activation with normalized collision energy of 31.0, MS/MS data were acquired using the ion trap analyzer in centroid mode and normal mass range. Using synchronous precursor selection, up to 10 MS/MS precursors were selected for HCD activation with normalized collision energy of 55.0, followed by acquisition of MS3 reporter ion data using the FTMS analyzer in profile mode at a resolution of 50,000 over a range of 100-500 m/z. Proteins were identified and reporter ions quantified by searching the UniprotKB database using MaxQuant (Max Planck Institute) with a parent ion tolerance of 3 ppm, a fragment ion tolerance of 0.5 Da, a reporter ion tolerance of 0.001 Da, trypsin/P enzyme with 2 missed cleavages, variable modifications including oxidation on M, Acetyl on Protein N-term, and phosphorylation on STY, and fixed modification of Carbamidomethyl on C. Protein identifications were accepted if they could be established with less than 1.0% false discovery. Proteins identified only by modified peptides were removed. Protein probabilities were assigned by the Protein Prophet algorithm [Anal. Chem. 75: 4646-58 (2003)]. TMT MS3 reporter ion intensity values are analyzed for changes in total protein using the unenriched lysate sample. Phospho(STY) modifications were identified using the samples enriched for phosphorylated peptides. The enriched and un-enriched samples are multiplexed using two TMT10-plex batches, one for the enriched and one for the unenriched samples.

Following data acquisition and database search, the MS3 reporter ion intensities were normalized using ProteiNorm (Graw et al). The data was normalized using VSN (Huber et al) and analyzed using ProteoViz to perform statistical analysis using Linear Models for Microarray Data (limma) with empirical Bayes (eBayes) smoothing to the standard errors (Storey et al, Ritchie et al). A similar approach is used for differential analysis of the phosphopeptides, with the addition of a few steps. The phosphosites were filtered to retain only peptides with a localization probability > 75%, filter peptides with zero values, and log2 transformed. Limma was also used for differential analysis. Proteins and phosphopeptides with an FDR-adjusted p-value < 0.05 and an absolute fold change > 2 were considered significant.

Mouse T cell in vitro expansion, proliferation (CellTrace) and apoptosis/cell death assays

Actively growing mouse T cells were rested for 24 hours post-transduction in T cell medium with 100 U ml⁻¹ rmIL-2. On day of assay, T cells were washed with T cell medium lacking cytokine. IL-9R transduced T cell proliferation was assessed by seeding 100,000 cells per well in a round bottom 96-well plate in the presence of equipotent doses of recombinant murine IL-2 (rmIL-2), rmIL-9, or rmIL-12. T cells were fed with fresh media and cytokine 48 hours after seeding, and every subsequent 48 hours until the end of the assay. Counts were also acquired on the day of replenishment by staining an aliquot of cells with Acridine Orange/Propidium Iodide (AO/PI, The DeNovix Company) and analyzing on a cell counter (The DeNovix Company). An aliquot of cells were also stained with Zombie Violet Live Dead (BioLegend) and analyzed for YFP⁺ on the NovoCyte Penteon (Agilent Technologies). For CellTrace assays, resuspension of dye was followed as recommended by the manufacturer. Proliferation index was calculated in FlowJo by modeling dye dilution peaks to determine the number of cells in each generation, then dividing the total number of divisions (sum of cells in each generation multiplied by their division number) by the number of cells that underwent at least one division. This yields the average number of divisions per responding (divided) cell, excluding undivided cells from the denominator. To measure apoptosis, cells were removed every 48 hours post cytokine treatment. Cells were subsequently resuspended at 1 x 10⁶ cells/mL and stained with Zombiet Violet (BioLegend) and Annexin V (Thermo Fisher) in Annexin Binding buffer (BioLegend) at 25°C for 15 minutes. After the incubation, cells were diluted with 2x Annexin Binding buffer, placed on ice, and immediately analyzed on the Novocyte Quanteon (Agilent Technologies).

ELISA

Serum IFNγ ELISA was conducted using the Mouse IFNγ ELISA MAX kit (BioLegend) using the manufacturer's recommendations. In short, the day prior to running the ELISA, the Capture Antibody was diluted in Coating Buffer and 100 μL of this capture solution to all relevant wells of a 96-well plate. The plate was sealed and incubated overnight at 4°C. On the day of the assay, the plate was washed 4 times with 300 μL of Wash Buffer per well to remove capture solution. After the wash was complete, the plate was then blocked with 200 μL of Assay Diluent per well, and then incubated at 25°C for 1 hour on a plate shaker (500 rpm). Following the blocking, the plate was washed 4 times with Wash Buffer, and then 100 μL of standard dilutions and samples were added in duplicates. The plate was then incubated at 25°C for 2 hours with shaking. After the incubation, the plate was washed 4 times as previously described and 100 μL of diluted Detection Antibody solution was added to each well and incubated at 25°C for 1 hour with shaking. Following the detection incubation, the plate was washed 4 times and add 100 μL of diluted Avidin-HRP solution was added and incubated at 25°C for 30 minutes with shaking. After the incubation, the plate was washed 5 times with Wash Buffer, with the buffer soaking in each well for 30 seconds per wash to minimize background. After the washes were complete, 100 μL of TMB Substrate Solution was added and incubated at 25°C in the dark until the wells turn blue. The reaction was stopped by adding 100 μL of Stop Solution to each well and then immediately read on a fluorescent microplate reader (Thermo Fisher Scientific) (absorbance read at 450 nm).

Coculture, cytotoxicity and proliferation assays for human T cells

Human T cells transduced with a CAR or NY-ESO-1 TCR, and co-transduced with IL-9R or an IL-9R variant, were resuspended at 1 x 10⁶ cells ml⁻¹ in RPMI-c with either no cytokine, 1 nM recombinant human IL-2, or 10 nM recombinant human IL-9 (Peprotech). T cells were serially diluted to achieve specified effector:target ratio and triplicates were cocultured with 5 x 10⁴ cells ml⁻¹ (50μL) NALM6-FFluc or 5 x 10³ M407^mCherry cells in a 96 well plate. 50 μL of media was added to bring the total volume of each well to 200 μL. The plate was imaged in the Incucyte Live-Cell Analysis System (Sartorius) for 72 hours, and killing was quantified by measuring the Total Green or Red Object Integrated Intensity metric of GFP⁺ or mCherry⁺ tumor cells. Cytotoxicity index was calculated by normalizing to the fluorescence intensity at the first time point for each triplicate. Data were analyzed in GraphPad Prism.

For repetitive killing assays using non-adherent tumor cells, additional tumor cells were added in the same effector:target ratio as originally plated once tumor killing plateaued, while maintaining cytokine concentrations. For repetitive killing assays using non-adherent tumor cells, T cells were harvested from each well after tumor killing plateaued, counted, and replated at the same E:T ratio in a new 96 well plate with fresh tumor cells.

In vitro proliferation assays were performed in triplicate in 96-well round-bottom plates. T cells were counted and percent live YFP⁺ cells were assessed by flow cytometry every 48 hours to calculate fold expansion, then re-plated in fresh media with cytokine at their original concentration. Flow cytometry data were analyzed using FlowJo and graphs were generated in GraphPad Prism.

In vivo mouse tumor studies

For in vivo tumor growth experiments using immunocompetent mice, early-passage cell lines were used (fewer than 10 passages). B16-F10 or KP-gp100 tumor cells (5 x 10⁵ cells) were resuspended in 100 μl of PBS and injected subcutaneously in the right flank of 6–10-week-old C57BL/6J mice. Prior to adoptive cell transfer, mice were randomized based on average tumor size. 2 x 10⁶ – 5 × 10⁶ non-sorted, transduced T cells (derived from pmel mice) were adoptively transferred five or six days after tumor inoculation unless otherwise indicated. Specifically, T cells were resuspended in 50 μl of PBS per mouse and administered by retroorbital intravenous injection. Where indicated, mice received treatment with cytokines: mouse serum albumin (MSA)-bound mouse IL-2 (MSA-IL2), MSA-IL9, MSA-IL9^Q115T, MSA-oIL2 (5 x 10⁴ units per dose, intraperitoneal) for five doses (or longer, where indicated), administered every other day starting on the day of adoptive cell transfer (unless otherwise indicated). Tumor size (length x width) was measured with calipers three times a week and volume was calculated as (length x width²)/2). Peripheral blood (10 μl) was collected at indicated time points from the tail vein for quantification of adoptively transferred pmel T cells by flow cytometry. Mice were euthanized when the total tumor volume exceeded 2,000 mm³, as per IACUC guidelines.

We also performed xenograft studies of human T cells and human tumors in immunodeficient NSG mice. For the leukemia model, NALM6-FFluc cells (1 x 10⁶ ml⁻¹ in 200 μL sterile PBS) were injected via tail vein into each mouse. On Day 3 post tumor inoculation, luciferin-based imaging was performed to confirm tumor engraftment. Upon confirmation, T cells were injected via tail vein at a dose of 1 x 10⁶ cells or 3 x 10⁶ cells in 200 μL per mouse. Beginning on Day 3 as well, mice received a dose of 25,000 I.U. of MSA-hIL-9 in 100 μL of sterile PBS (i.p.). Mice were injected with cytokine every other day for a total of ten doses. Luciferase-based imaging was performed twice per week for at least three weeks to assess tumor growth. Specifically, mice were anaesthetized with isoflurane and 200 uL of 13mg mL⁻¹ D-luciferin was injected i.p. per mouse, incubated for 4 minutes, and imaged for 30 seconds with heavy binning or automatic exposure detection. Mice were closely monitored and sacrificed upon reaching any one of the morbidity criteria, as outlined under IACUC guidelines (tumor flux exceeding 1 x 10¹¹ p/s, rear leg paralysis, or other conditions that necessitated euthanasia).

For the osteosarcoma model, the right leg of each mouse was shaved, and width and depth were measured prior to tumor inoculation. 1 x 10⁶ 143B-FFluc cells in 100 μL sterile PBS were injected intramuscularly into the right leg of each mouse. On Day 4 post tumor inoculation, luciferin-based imaging was performed to confirm tumor engraftment. Upon confirmation, T cells were injected via tail vein at a dose of 7 x 10⁶ cells in 200 μL per mouse. Beginning on Day 4 as well, mice received a dose of 25,000 I.U. of MSA-hIL-9 intraperitoneally. Mice were injected with cytokine every over day for a total of 10 doses. Their right leg was measured three times a week. Mice were closely monitored and sacrificed upon reaching any one of the morbidity criteria, as outlined under IACUC guidelines (tumor size, inability to move their right leg, or other conditions that necessitated euthanasia).

In vivo cytokine dose calculations

For in vivo experiments, we dosed cytokines based on activity (international units; I.U.). For MSA-mIL2, MSA-hIL2 or MSA-oIL2 (3A10), we considered 1 μg of protein to equal 3280 I.U. For MSA-mIL9, we considered 1 μg to equal 1118 I.U. For MSA-hIL9, we considered 1 μg MSA-hIL9 equal to 1428 I.U.

In vivo toxicology studies

For measuring the toxicity of cytokines in vivo, mice were weighed and video recorded individually for 30 second intervals at baseline and every 48 hours as indicated. Videos were processed using EthoVision Software to quantify mobility.

Analysis of T cell biodistribution and phenotype

Spleen, tumor-draining lymph nodes, tumor, liver, and lung were harvested and dissociated into single-cell suspensions at the indicated timepoints post adoptive cell transfer. Spleens and lymph nodes were weighed, dissociated mechanically, and filtered through a 70-μM cell strainer. Tumors were minced into 2-4 mm pieces and lungs were perfused with PBS and separated by lobe. Tumor, lung and liver tissues were enzymatically and mechanically dissociated in a gentleMACS Octo Dissociator (Miltenyi Biotec) for 40, 30 and 30 minutes, respectively, using enzymes included in the tissue-specific dissociation kit (Miltenyi Biotec). All single-cell suspensions of dissociated tissues were strained in 70 μM-filters. Dissociated tumors were resuspend in 5 ml Percoll 40% (sigma-Aldrich) and then laid on 3 ml Percoll 67% in a 15ml polypropylene tube. Tubes were spun at 2000 rpm for 20 min at room temperature. Mononuclear cells were collected from the interface of the 40:67% Percoll gradient and washed with complete RPMI prior to antibody staining. Peripheral blood (10 μl) was obtained through tail-vein sampling prior to collection of other organs. Red blood cells were lysed with eBioscience RBC Lysis Buffer (1 mL per spleen; Thermo Fisher Scientific) for 5 minutes at 4°C and washed with PBS. After red blood cell lysis, cells were washed with PBS.

T cells collected from in vitro or in vivo experiments were initially stained with viability dye in PBS for 15 minutes at 4°C. After washing, cells were surface stained with antibody panel diluted in staining buffer (Flow Cytometry Staining Buffer, Ebioscience) for 30 mins at 4°C, then washed. In some cases, cells were fixed in 1.6% paraformaldehyde (PFA) for 15 minutes at 25°C and washed and stored prior to analysis. For the intracellular staining, after fixation, cells were permeabilized (Fixation and Permeabilization Kit, eBiosciences) and stained with antibody panel diluted in staining buffer for 2h at 4°C according to the manufacturer’s protocol. Cells were washed and resuspended in staining buffer. Analysis was performed on an LSR II (BD Biosciences), Symphony A5, (BD Biosciences), Fortessa (BD Biosciences), Aurora (Cytek Biosciences), or NovoCyte Penteon (Agilent Technologies). For analysis of transduced T cells from tissues of mice, 350 cells were required for immunophenotypic analysis; groups with less than three biological replicates meeting this criteria were excluded.

Single cell RNA-sequencing and analysis

Single cell RNA-seq of sorted live thy1.1⁺YFP⁺ pmel T cells was performed using the 10X Genomics platform using 10x Genomics reagents according to the manufacturer’s instructions. The 3′ Gene Expression libraries were sequenced using a NovaSeq 6000 with a sequencing depth of 120M paired reads (240M total reads) per sample to achieve >20K reads per cell. FASTQ files were processed using the CellRanger count pipeline [CellRanger version 7.0.1 (10x Genomics)] using the mm10 mouse reference genome. Raw gene expression matrices were constructed into a Seurat object and imported into R software using Seurat (version 5). To filter out low-quality cells, we removed cells that had either a low or high number of detected genes and cells that had more than 10-15% of mitochondrial UMI counts in the scRNA-seq data and remaining 6706 cells were used for downstream analysis. To normalize the gene expression levels, we utilized the LogNormalize method implemented in Seurat.

We performed principal component analysis (PCA) to reduce the dimensionality of scRNA-seq data. The top 30 principal components (PCs) were selected to construct the UMAP embeddings. The FindClusters function was used to identify cell clusters. Total of 10 major cell clusters were obtained, and cluster specific markers were identified using "FindAllMarkers" or "FindMarkers" functions in Seurat. Only positive markers expressed in at least 1% of cells were considered. The nonparametric Wilcoxon rank-sum test was used to obtain the p value for comparisons, and the adjusted p value based on Bonferroni correction was calculated. Genes with adjusted p < 0.05 were considered as enriched in a particular cluster.

The “AddModuleScore” function in Seurat with default parameters was used to score sets of genes in the single cell data. We calculated the average expression levels of each gene set on a single cell level, subtracted by the aggregated expression of control gene sets. All analyzed features are binned based on averaged expression, and the control features are randomly selected from each bin. The scored expression was used for plotting.

Trajectory analysis of scRNA-sequencing data

To simulate and understand the differentiation trajectories between cells, we used “monocle3” package (version 1.3.7). The T cell clusters and differential gene expression analysis with Seurat were extracted from the scRNA-seq data and used to construct the monocle object. Pseudotime trajectories were constructed with default parameters and visualized with the plot_cells() function. Cells from the “Tscm-like (Tcf7)” cluster were used as the reference starting point (root nodes) from which potential differentiation pathways resonate. The identified paths were mapped on to the UMAP projection for visualization. Ridge plots show the pseudotime scores across treatment groups or clusters.

Transcription factor activity inference from scRNA-seq data

Top regulons (gene sets regulated by the same transcription factor) and their activity were inferred and evaluated using the SCENIC pipeline (version 1.3.1), thereby defining the top regulon activity in every cell. The input files consisted of the expression matrix and metadata. Then, the co-expression network was calculated by GRNBoost2, and the regulons were identified by RcisTarget. Next, the regulon activity for each cell was scored by AUCell. Scaled expression of regulons were used for the plotting.

Analysis of in-house bulk RNA-seq data

Libraries were prepared using KAPA mRNA HyperPrep Kit from total RNA. Paired-end sequencing was performed on a Novaseq 6000/X Plus (40M reads per sample). Read alignment was performed using STAR aligner on GRCm38 mouse reference transcriptome and transcript level quantification using Salmon. Quality control of the paired-end bulk RNA-seq data was performed using the FastQC program. Adapter trimming was carried out by Cutadapt method. Initial quality control of the count data was carried out by PCA and sample-sample distance after variance stabilizing normalization. DESeq2 was used to perform the differential gene expression analysis on samples.

Transcription factor activity inference from bulk RNA-seq data

To infer TF activity in bulk RNA-seq data we ran the Univariate Linear Model (ulm) method using the decoupleR package. For each sample in our dataset (mat) and each TF in our network (net), we fit a linear model that predicts the observed gene expression based solely on TF-Gene interaction weights. Once fitted, the obtained t-value of the slope is the score. If it is positive, we interpret that the TF is active and if it is negative we interpret that it is inactive.

Gene set scoring for bulk RNA-seq data

We calculated the gene set score of the bulk RNA-seq data based on the average normalized expression of the genes. From this, the enrichment score of randomly selected control genes with similar expression levels are subtracted. These control gene sets are defined by first binning all genes into bins (n= 24) of aggregate expression level and then, for each gene in the gene set of interest, 100 genes from the same expression bin as that gene are randomly selected. The score is then set to range from zero, meaning no enrichment compared to random sets of genes with similar expression, to one, reflecting the highest average expression of all genes within the gene set of interest. This gene set score is calculated per sample for bulk RNA-seq samples. Functional geneset enrichment analysis was performed using fGSEA package in R to infer enriched biological pathways in the treatment groups.

Linking bulk RNA-seq and phosphoflow data to generate pSTAT gene modules

We utilized a set of samples with matched RNA-seq and phosphoflow data. The RNA-seq data was obtained from in vitro experiments, while the phosphoflow data consisted of MFI measurements at maximum stimulation (Emax) for phosphorylated STAT proteins (pSTAT1, pSTAT3, pSTAT4, and pSTAT5). These two datasets were merged to create a combined matrix for subsequent analysis. To investigate the relationship between gene expression and STAT phosphorylation, we computed Pearson correlation coefficients between the RNA-seq counts and each pSTAT measurement. This analysis was performed using the cor.test function from the R stats package. The correlation test was conducted for each gene against each pSTAT, resulting in a comprehensive correlation matrix. For each pSTAT, we selected the top 100 genes exhibiting the highest absolute correlation coefficients. These genes were then visualized using scatter plots and the coefficient of determination (R²) was calculated for each pSTAT-gene set relationship to quantify the proportion of variance in pSTAT levels explained by gene expression. We created gene modules for each pSTAT using the top 100 correlated genes identified in the previous step and applying the AddModuleScore function in Seurat. These gene modules were then projected onto our scRNA-seq dataset as violin and feature plots.

Analysis of publicly available PBMC and lamina propria scRNA-seq datasets

The raw expression data was obtained from the Single Cell Portal (Broad Institute) for Human PBMC (“Single Cell Comparison: PBMC data”)⁵ and Lamina Propria (“ICA: Ileum Lamina Propria Immunocytes”)⁶. Raw gene expression matrices were constructed into a Seurat object and imported into R software using Seurat (version 5). Data processing was performed as described in sections above. t-Distributed stochastic neighbor embedding (t-SNE) in Seurat was used for data visualization. Azimuth (version 0.5.0) in Seurat was used to annotate the clusters obtained at resolution 0.2 (PBMC) and 0.5 (lamina propria). To specifically visualize the expression of γ_c cytokines and their receptors, we plotted feature plots and dot plots to examine the expression across different cell types in PBMC and Lamina Propria.

Analysis of publicly available bulk RNA-seq data from Human Protein Atlas

RNA GTEx (genotype tissue expression) tissue gene data was downloaded from The Human Protein Atlas⁵⁷ (v19.proteinatlas.org/about/download). Transcript expression levels summarized per gene in 36 tissues based on RNA-seq. The tab-separated file includes Ensembl gene identifier ("Gene"), analysed sample ("Tissue"), transcripts per million ("TPM"), protein-transcripts per million ("pTPM") and normalized expression ("NX"). The data was obtained from GTEx and is based on The Human Protein Atlas version 19.3 and Ensembl version 92.38. The expression values for IL2RB, IL4R, IL7R, IL9R, and IL21R across different tissues were used for visualization. We defined a threshold of 0.7, below which expression is not detected or negligible.

Analysis of publicly available scRNA-seq data of mouse lymph nodes in response to cytokines

Data was obtained from previously published work on the single cell transcriptomic responses of mouse immune cells (from lymph nodes) in response to a large panel of different cytokines¹¹. Data were downloaded as single cell data objects that were reanalyzed to quantify the overall magnitude of transcriptomic responses of each cytokine across different cell types. We created a dot plot with two metrics: the number of differentially expressed genes (DEGs) and the magnitude of change (based on Euclidean distance) across the entire transcriptome. The number of DEGs was the total number of genes in each cytokine signature. The overall magnitude of cytokine-induced differential expression was computed as the Euclidean distance between the centroid vectors of cytokine-treated cells and PBS-treated cells. A distinct color ramp was used for each cell type to emphasize that cell types have different properties (e.g., different numbers of genes expressed on average) and were independently analyzed. Cytokine–cell type combinations with five or more cells sampled were included in this analysis.

Analysis of publicly available scRNA-seq data from human CAR T cells sorted from patients

We utilized publicly available data from a scRNA-seq dataset of sorted CAR T cell populations at various timepoints after patient administration¹⁴. Across all patients and time points, the authors sequenced 66,042 post-infusion CAR T cells, with an average of 11,549 cells per patient (SD = 7,335) and 20,532 cells per time point (SD = 37,898). Notably, the month 6 post-infusion time point only consists of 7 CAR T cells. We downloaded Seurat objects containing different samples at various time points, fetched the expression of relevant cytokine receptors across different cell types and generated dot plots of the data.

Analysis of soft tissue sarcoma single-cell RNA-seq samples

Previously published synovial sarcoma 10x Genomics and SMART-seq2 scRNA-seq data were downloaded from GEO (GSE131309)⁵⁸. The undifferentiated pleomorphic sarcoma (UPS) and myxofibrosarcoma (MFS) samples were obtained GEO (GSE212527)⁵⁹. Log-normalization was performed independently for each sample using the Seurat function NormalizeData. After this step, the Seurat objects for each independent sample were merged preserving the individual normalization of the data. Sample integration and clustering were performed separately for each histotype, accounting for batch effects and differences in sequencing technologies. The top 2000 highly variable genes of the merged and normalized expression matrices were identified using the function FindVariableFeatures, and then centered and scaled using the ScaleData function. Principal component analysis was then run using the top 2000 highly variable genes previously identified. To integrate multiple samples, the harmony R package(3) (version 0.1.0) was employed for batch correction. Harmony was ran using both the sample of origin and the sequencing technology variables as arguments of “group.by.vars” and based on the first 25 PCA dimensions previously identified. UMAP embeddings were then obtained using the first 25 Harmony dimensions, and clusters were obtained by calculating the k-nearest neighbors (k-NN) and a shared nearest-neighbor graph using the Louvain algorithm implemented in the FindClusters function of the Seurat package with a resolution of 1.0. Endothelial cells, myeloid cells, lymphoid cells, and CAFs were annotated using the differentially expressed genes discovered with the FindMarkers function in Seurat, which was run with the default parameters except for only.pos = T and min.pct = 0.25. Canonical cell identity markers obtained from the literature were used for cell type annotations. A small number of cells could not be uniquely assigned to a specific cell identity and were removed from further downstream analysis. UMAP plots were generated with Seurat and scCustomize (4). Density plots were generated with the R package Nebulosa (5).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical comparisons between cell clusters, treatment conditions and regulon activity from the scRNA-seq data were performed using R. The Kruskal-Wallis test was used to compare the medians of groups. P < 0.05 was considered statistically significant. Statistical analysis for other experimental data was performed using GraphPad Prism 9 (GraphPad software), except where indicated. All values and error bars are shown as mean ± SEM except as indicated. Comparisons of two groups were performed by using two-tailed unpaired Student’s t test. Comparisons of multiple groups were performed by using one-way analysis of variance (ANOVA) with Tukey’s multiple-comparisons test unless otherwise indicated. In some cases with unequal variance, Kruskal-Wallis with Dunn’s test was performed to compare multiple groups. Experiments that involved repeated measures over a time course, such as tumor growth were performed by using two-way ANOVA with Tukey’s multiple-comparisons post-test. Survival data were analyzed using the Log-rank (Mantel-Cox) test. For analysis of in vivo experiments, outliers were identified across groups using the ROUT method.

Supplementary Material

1

Document S1. Figures S1-S7.

2

Table S1. Differentially phosphorylated proteins, related to Figure 3C.

3

Table S2. Differential gene expression and regulon analysis, related to Figure 6.

4

Table S3. Frequency of CD8⁺Thy1.1⁺YFP⁺ T cells from tissues, related to Figure 6.

5

Table S4. Correlations between STAT phosphorylation and gene expression, related to Figure 6H-K.

6

Table S5. Summary of YFP expression in transduced T cells across experiments, related to Figures 3-5 and 7.

7

Table S6. Summary of EC50 and Emax values across experiments, related to Figures 3-5 and 7.

8

Table S7. Source data including experimental replicates, related to Figures 1-7.

Highlights.

• IL-9/IL-9R form a potent and orthogonal cytokine-receptor pair for T cell therapy.

• In addition to STAT1/3/5, IL-9 unexpectedly activates STAT4.

• The impact of IL-9R signaling is highly sensitive to STAT strength and balance.

• STAT1 acts as a rheostat between stem or memory and terminal effector states.