Abstract

Interleukin-7 (IL-7) plays a central role in maintaining T cell development and immune homeostasis, and enhancing the cytokine’s immune-stimulatory functionality has broad therapeutic implications against various oncological malignancies. Herein, we show a computationally designed IL7 superkine, Neo-7, which exhibits enhanced folding efficiency and superior binding affinity to its cognate receptors. To streamline the protein candidate prediction and validation process, the loop region of IL7 was strategically targeted for redesign while most of the receptor-interacting regions were preserved. Leveraging advanced computational tools such as AlphaFold2, we show loop remodeling to rectify structural irregularities that allow for iterative stabilization of protein backbone and lead to identification of beneficial mutations conducive to receptor engagement. Neo-7 superkine shows improved thermostability and production yield, and it exhibits heightened immune-stimulatory and anticancer effect in C57BL/6 J mice. Neo-7 addresses intrinsic developability limitations of IL-7, including inefficient folding, aggregation propensity, and suboptimal receptor engagement, while in vivo pharmacokinetic limitations of wild-type IL-7 were addressed separately through Fc fusion. These findings underscore the utility of a targeted computational approach for de novo cytokine development.

Introduction

Clinical success with immune checkpoint inhibitors (ICIs) and immunogenic cell death (ICD) inducers has invigorated developmental efforts on immunotherapeutic strategies that strengthen the body’s immune system against oncological malignancies (Lim et al., 2024; Galluzzi et al., 2024; Waldman et al., 2020; Tan et al., 2020; Tan et al., 2020; Taefehshokr et al., 2022). The promise of durable tumor suppression by immunotherapeutic approaches has rekindled enthusiasm towards cytokine therapies, which can enhance immune responses and promote a more effective attack on cancer cells by directly stimulating the proliferation and activation of various immune cell types (Park et al., 2024; Fu et al., 2023; Waickman et al., 2016). Cytokine-mediated immune cell activation can override the negative feedback mechanisms and bolster immune cell stimulation, allowing for broad-spectrum tumor suppression and synergistic actions with ICIs and ICD inducers (Holcomb and Zou, 2022; Shen et al., 2020). However, traditional cytokine therapies have been hampered by challenges such as short half-lives, poor pharmacokinetics, and significant toxicity at therapeutic doses (Fu et al., 2023; Silva et al., 2019). These issues have hindered the broader application of cytokines in cancer treatment despite their potential benefits. For instance, IL-2, one of the first cytokines identified with potent anti-cancer properties, is limited by severe adverse effects, restricting its maximum tolerated dose (MTD) to 37 µg/kg, which in turn limits its efficacy (Mizui, 2019). Similar toxicity concerns exist for other interleukins, such as IL-12, IL-15, and IL-21, with clinical MTDs of 1.0, 20, and 100 µg/kg, respectively (Knudson et al., 2020; Steele et al., 2012; Portielje et al., 1999), thus prompting emerging protein design strategies for cytokine variant development with elevated efficacy profiles.

Advances in computational protein design have opened up new opportunities for cytokine engineering, which enables de novo design of cytokine variants with improved stability, reduced toxicity, and enhanced efficacy. These engineered cytokines can be structurally modified for enhanced engagement with their putative cognate receptors, leading to enhanced therapeutic and safety profiles for treatment development. A notable example is the work by Silva et al., 2019, which demonstrated a significant breakthrough in computational protein design (Silva et al., 2019). Using the Rosetta protein design suite, they restructured and redesigned four-helix bundle mimetics of IL-2 and IL-15 (neo-2/15), resulting in proteins that are more potent and stable both in vitro and in vivo, with specifically programmed receptor specificity. Although the seminal study showcased the possibility for researchers to design novel protein with computationally idealized structure in its entirety, such approach demands substantial computing resources and can generate many non-analogous candidates for validation. We envision that a targeted redesign strategy that focuses on protein domains with manifest folding inefficiency can minimize the resources required for computational protein engineering. In the present work, the targeted computational design strategy is demonstrated through the development of an IL-7 variant.

Interleukin-7 (IL-7) is a member of the common gamma chain (γc) receptor family, which includes cytokines such as IL-2, IL-15, and IL-21 (Waickman et al., 2016; Lin and Leonard, 2018; Sim and Radvanyi, 2014). IL-7 plays a crucial role in the survival, differentiation, and proliferation of immune cells, including T-lymphocytes, B-lymphocytes, and natural killer (NK) cells (Park et al., 2024; Winer et al., 2022). In vitro studies have shown that IL-7 promotes the survival and homeostatic proliferation of splenocytes by upregulating survival factors like Bcl-2. Preclinical animal studies and clinical trials have demonstrated broader functions, including the induction and maintenance of memory T cells, promotion of T cell homing to lymphoid organs, increased T cell receptor diversity, and enhanced T cell cytotoxicity by the cytokine (Park et al., 2024; Winer et al., 2022; Lee et al., 2024; Kim et al., 2020). The first-generation IL-7, known as CYT-107, demonstrated potent immune cell reconstitution in phase 2 clinical trials for HIV patients (for CD4+ T cell restoration) and patients with sepsis-induced lymphopenia (Trédan et al., 2018; Perales et al., 2012; Sportès et al., 2010). However, its anti-cancer efficacy was limited, as it increased immune cell count but did not induce a significant anti-tumor response, likely due to its short half-life (8.7–34.6 hr; Park et al., 2024; Winer et al., 2022; Trédan et al., 2018; Fu et al., 2024). The second-generation IL-7 drug, NT-I7 (efineptakin alfa), uses a hybrid human IgG/IgD Fc domain for half-life extension (T1/2=60.8–139.7 hr), which significantly improves its anti-tumor activities. Efineptakin alfa supports the proliferation and survival of CD4+ and CD8+ cells in both humans and mice, showing promising responses in early-phase clinical trials for advanced solid tumors, particularly in combination with immune checkpoint inhibitors (Park et al., 2024; Kim et al., 2020; Fu et al., 2024). Although much effort has been devoted to enhancing IL-7 pharmacokinetics through protein fusion, no variant or computationally designed mimic has been reported on the cytokine to our knowledge.

From a development perspective, IL-7 is prone to aggregation and requires additional engineering to enhance its solubility and stability. While prior clinical efforts have primarily addressed IL-7’s short circulating half-life through fusion strategies such as Fc conjugation, comparatively less attention has been paid to intrinsic scaffold inefficiencies of IL-7, including long inter-helical loops, aggregation propensity, and folding dependence on multiple disulfide bonds. To address these challenges, we herein engineer a novel variant of IL-7, known as Neo-7, with enhanced folding efficiency through targeted computational protein design. By removing redundant loops and redesigning the protein backbone using Rosetta loop remodeling, we reconnected the functional domains in a more efficient and economical structural (defined here as minimizing the number of residues required to connect receptor-binding or structural helices). Advanced computational tools were employed to ensure that the Neo-7 variant would fold according to the predicted structural models, resulting in a protein with superior folding efficiency and receptor binding affinity. In further experimental studies, Neo-7 demonstrates improved in vitro production yield and in vivo biological activity, positioning it as a promising candidate for cancer immunotherapy. These advancements underscore the potential of IL-7 and its engineered variants as effective cancer immunotherapies, highlighting the importance of optimizing cytokine properties for clinical application.

Results

Development of a computational pipeline for rational protein design

In redesigning IL-7, we aimed to minimize sequence alterations relative to the wild-type protein. This strategy is based on the understanding that a protein’s sequence is closely tied to its structure and function. By preserving the native sequence elements involved in IL-7’s biological activity, we sought to maximize the likelihood of producing a successful variant without compromising its original function. We began by examining the crystal structure of IL-7 bound to its receptor, IL7Rα (interleukin-7 receptor alpha; PDB ID: 3DI2), which recruits IL-2Rγ to form a heterodimeric receptor complex essential for downstream signaling. Loop remodeling was performed using Rosetta’s loop remodeling modules, followed by validation with AlphaFold, employing both single-sequence monomer and multimer models. During the computational design and validation process, each model predicted by either Rosetta or AlphaFold was meticulously inspected using PyMOL and Discovery Studio 4.5. Residues that were found to clash with neighboring amino acids were further refined through mutations using Rosetta’s fix backbone modules. The gene sequences of the final designs that successfully passed this computational pre-filtering were then synthesized and cloned into a yeast display vector for subsequent in vitro validation (Figure 1).

Schematic illustrating the computational pipelines for Neo-7 engineering from wild type IL7 sequence.

In general, non-receptor interacting loops were deleted from the WT-IL7 sequence and loops connecting the adjacent helices were modeled using Rosetta Loop Remodeler and Rosetta fix backbone design function. The sequence of the designed model was extracted and submitted to AlphaFold (monomer and multimer mode for structure and protein-receptor binding prediction respectively) as a preliminary validation of the Rosetta-remodeled protein. Iterations of the bad models (models that do not fold into the expected structure or models that did not predict to bind to the receptors) back to the design stage were performed. Models that passed the AlphaFold validation proceeded to subsequent in vitro assay using yeast display system and flow cytometry to determine their relative binding affinity to IL-7 receptors in comparison to WT-IL7.

The wild-type IL-7 protein forms a 4-helix bundle structure with three inter-helical loops. Upon visually examining its crystal structure in complex with IL-7 receptor alpha (IL7Rα), we noted that the connectivity between each helix is intricate and inefficient (Figure 2). In the crystal structure of WT-IL7 complexed with IL-7 receptor alpha, IL-7 is composed of four helices. Helices 1 and 3 are found to interact with IL-7 receptor alpha (IL7Rα), while helices 1 and 4 are predicted to interact with the IL-2 receptor common gamma chain (IL2Rγ; McElroy et al., 2009). Helix 2 was observed to serve as a structural support for the helix bundle. However, the connectivity of these helices exhibited complexity and inefficiency. Notably, the loop between helix 1 and 2 spans 29 amino acids (AA), the loop connecting helix 2 and 3 is 6 AA long, and the loop connecting helix 3 and 4 is 35 AA long. We extracted the helices from the PDB structure file and reconnected the helices using the Loop Remodel function from the Rosetta suite. The protein was remodeled clockwise in the order of H1>H3>H2>H4. The new loops connecting H1-H3, H3-H2, and H2-H4 are 7, 8, and 5 AA in length, respectively. The amino acid sequences of the designed loops were subsequently optimized using Rosetta Fix Backbone design function and validated using Alphafold monomer.

Blueprint of Neo-7 design.

Blueprint of the WT-IL7 was shown on the left of the figure. The connectivity of the functioning helixes was connected in a manner that requires extremely long protein loops by design (i.e. helices were not connected to the closest adjacent helixes but to the opposite helix). Loops that were not interacted with the IL-7 receptors were deleted and the helixes were reconnected in a clockwise manner via new protein linkers connecting to the adjacent helixes. The blueprint of the redesigned protein was shown at the right side of the figure. Protein structures are colored as rainbow (from N-to-C terminus with the order of Blue-Green-Yellow-Red).

Structural modeling of Neo-7 in AlphaFold single sequence mode

The synthesized Neo-7 sequence was cloned into a yeast-display vector to assess its expression and binding capability to IL7Rα. Despite the concordance between Rosetta and AlphaFold monomer predictions, no binding was observed between displayed Neo-7 and IL7Rα (Figure 3C). A degree of binding to IL2Rγ was detected, possibly reflecting partial folding of the displayed protein in the yeast display platform. We also postulated that the high similarity between the helices of Neo-7 and WT-IL7 amino acid sequences might lead to sampling bias, leading AlphaFold to predict a similar folding of Neo-7 to WT-IL7. To investigate this, we utilized AlphaFold Single Sequence (AlphaFold SS), an online platform capable of performing AlphaFold without relying on a PDB template and multiple sequence alignment (MSA) database. The resulting structure model from AlphaFold SS indicated misfolding of helix 2 (providing structural support) and helix 3 (binding to IL7Rα) when superimposed onto WT-IL7 helices (Figure 3A). However, the helices responsible for binding to the gamma receptor aligned well with the WT template. Subsequently, we addressed the misfolded helices by redesigning their connecting loops (loop 1 and 2) using Rosetta fix backbone design and validated the structures with AlphaFold SS (Figure 3B).

Validation of Neo-7 designs from Rosetta loop remodeling and fix backbone design.

(A) AlphaFold validation of the first loop design version of Neo-7 (Neo-7 LDv1) using the default (left) and single sequence mode (right). (B) AlphaFold validation of the second loop design version of Neo-7 (left; Neo-7 LDv2) and Neo-7 LDv2 with mutations (right) favored by Rosetta fix backbone design. (C) Crystal structure of human IL-7 in complexation to human IL-7 receptor alpha (PDB ID = 3DI2). (D) Superimposition of Neo-7 structures (with or without additional disulfide bridge) predicted by AlphaFold. (E) Yeast display and flow cytometry validation of IL-7/Neo-7 bindings towards the IL-7 receptors. The yeast-displayed protein (different redesigned IL-7s) carries a HA-tag while the recombinant IL-7 receptors carry either a HIS tag (IL-7 receptor alpha) or a FC-tag (common-IL-2 family receptor gamma; for detection of IL-2Rγ binding, yeast cells were first incubated with recombinant IL-7Rα, washed, and subsequently incubated with IL-2Rγ.) The signal intensity of the X-axis (conferred by the binding of anti-HA mab) correlates with the expression level of the displayed protein while the signal intensity of the Y-axis (conferred by the binding of the anti-HIS/anti-FC mAb to the recombinant receptors bound to the displayed proteins) correlates with the binding affinity of the displayed proteins towards the IL-7 receptors.

Disulfide stapling on Helix 1 and stabilizing mutations are essential for binding of IL-7 and Neo-7 to IL-7 receptor-α

Since the helices were adapted from the WT-IL7, we hypothesized that those beneficial mutations to stabilize IL-7 or enhance its binding to the cognate receptors should be applicable to our designed constructs. Prior to the transfer of the beneficial mutations discovered from our previous study (Figure 3—figure supplement 1), we first constructed a multimer model consisting of Neo-7, murine IL7Rα, and the common IL2Rγ receptors using AlphaFold Multimer. The resulting model was used as a structure template for Rosetta fix backbone design, which the amino acid variants of the mutants and their wild-type counterpart (Table 1) were keyed into the Rosetta fix backbone residue files for calculation. Only those mutations that were favored by Rosetta are selected to maximize the stability of the protein and to preserve their theoretical folding in the subsequent in vitro yeast display assay.

Site-directed mutagenesis residues recommended by Rosetta to improve the folding stability of the Neo-7s.

Residue number	Wild type	Mutant	Remarks	Rosetta’s choice
5	K	R	Cytokine-receptor binding interface	R
6	Q	P	Cytokine-receptor binding interface	Q
44	T	I	Cytokine core-surface interface	I
45	I	T	Borderline of cytokine binding interface	T

The Neo-7 design, incorporating Rosetta-favored mutations, demonstrated enhanced protein-display level and binding to IL2Rγ (Figure 3E). However, despite the alignment between computational protein design and prediction programs, weak or insignificant binding of the designed Neo-7 to murine IL7Rα was observed in the yeast display system. A closer examination of the crystal structure of WT-IL7 with IL7Rα revealed a kink on helix 1 is crucial for proper IL7-IL7Rα binding, stabilized by a disulfide bridge on the N-terminus and loop 3 of WT-IL7 (Figure 3C). The structure of Neo-7 generated from AlphaFold lacked this kink, possibly due to the absence of the disulfide bridge. To address this, we hypothesized that regenerating the kink through mutations of core amino acids between helix 1 and 4 into cysteine might recapitulate the native kink conformations observed in WT-IL7 (Figure 3C). Introducing G4C and L96C mutations into Neo-7 validated this hypothesis, as the resulting disulfide-stapled Neo-7 model predicted by AlphaFold SS exhibited the reappeared kink on helix 1 (Figure 3D). Encouraged by this result, we proceeded to the next stage of in vitro evaluation via yeast display using the designed Neo-7.

Meanwhile, to assess the importance of the kink for IL7Rα binding affinity in WT-IL7, we deleted the first 7 amino acids at the N-terminus of WT-IL7 and investigated its binding to murine IL7Rα in the yeast display system. Interestingly, the results (Figure 3E) demonstrated that N-terminal truncation fully diminished the binding of IL-7 to IL7Rα. Conversely, for the Neo-7 design, both the disulfide stapling and Rosetta-favored mutations were found essential for binding to IL7Rα. This indicates that both the helix kink from the disulfide design and core stability conferred by the transferred mutations are required for the generation of a stable and functional Neo-7. Importantly, the enhanced binding affinity towards IL7Rα also led to improved binding towards the common IL2Rγ, relative to other variants in the Neo-7 series.

Transfer of beneficial mutations on the cytokine-receptor binding interface further enhanced the binding of Neo-7 to IL7-Receptor α

Encouraged by the successful expression of a functional Neo-7 (Neo-7 Loop design 2 with R5K-T44I-I45T mutations, herein referred to as Neo-7) design in the yeast display platform, we modeled the structure of Neo-7 in the presence of murine IL7Rα and IL2Rγ. After careful examination of all the mutation points (Figure 4A), we hypothesized that two mutation points might further improve the binding affinity of Neo-7 towards IL7Rα due to the following reasons: (1) Both residues were located closely to a hydrophobic patch of IL7Rα and replacing the polar glutamine and threonine into a more hydrophobic proline or isoleucine is favorable for binding through hydrophobic interactions (2) Proline was disfavored by Rosetta probably due to its specific characteristics as a helix breaker (Table 1). However, after careful inspection of the structure, we observed that proline was located at the starting point of helix 1 that was further stabilized by disulfide stapling and therefore voided the concerns on the helix breaking properties of proline under this design. (3) At position 45, threonine is favored by Rosetta in comparison to the wild type isoleucine residue. Such a decision might arise from the additional consideration of the stability of Neo-7 monomer in aqueous solution as an exposed hydrophobic residue in the aqueous environment is energetically unfavorable. However, after comparing the binding conformations of threonine and isoleucine with IL7Rα, we decided that isoleucine was observed to have better shape complementarity that could occupy the hydrophobic pocket at the Neo-7- IL7Rα interface. Subsequently, we constructed the relevant mutants for Neo-7, sampling the effects of mutations located at the Neo-7-IL7Rα interface towards their binding to IL7Rα. Interestingly, the Q6P mutations showed the most prominent enhancements in the yeast display system, while the T45I mutation showed moderate improvement for IL7Rα. Double mutations of Q6P and T45I further improved the binding affinities of Neo-7 to both IL7Rα and IL2Rγ, suggesting an additive effect of the binding improvement (Figure 4B). Two best variants, Neo-7 with Q6P mutation and Neo-7 with dual mutations (Q6P+T45 I), were selected for subsequent in vitro bioactivity assay.

Validation of amino acid mutations that confer to the binding affinity of Neo-7 towards IL-7 receptor alpha and gamma.

(A) Inspection of structural and binding interactions of residue mutations Q6P and T45I on Neo-7 towards the murine IL-7R alpha. (B) Yeast display and flow cytometry validation of the binding ability of IL-7/Neo-7 variants toward the IL-7 receptors.

Recombinant Neo-7 shows higher expression and thermostability than WT IL7

After validating Neo-7’s ability to bind both IL-7Rα and IL-2Rγ receptors, we aimed to express recombinant Neo-7 in Escherichia coli. Wild-type cytokines from the IL-2 family (IL-2, IL-7, IL-15, and IL-21) are often challenging targets for E. coli expression due to their complex folding requirements, which necessitate multiple disulfide bonds to stabilize their structures. For instance, WT-IL-7 contains three disulfide bridges, and its soluble expression in E. coli yields no detectable properly folded IL-7, as analyzed by FPLC after pre-purification using Ni-NTA affinity chromatography (Figure 5A–B). In contrast, Neo-7 was engineered with only one disulfide bridge and designed for more efficient folding and packing compared to WT-IL-7 (Figure 5C–D). Indeed, Neo-7 demonstrated superior performance in terms of yield and purity. The Q6P variant of Neo-7 exhibited the highest purity (78.51%) and yield (1.56 mg/200 mL culture), while the double mutation variant (Q6P+T45 I) showed lower purity (58.24%) and yield (0.36 mg/200 mL). In comparison, WT-IL-7 was expressed as inclusion bodies, solubilized using 6 M guanidine HCl, and refolded via dialysis, resulting in a final yield of 0.23 mg/200 mL culture with a purity of 34.66% (Figure 5I). We also assessed the thermostability of the purified recombinant proteins using the SYPRO assay. As expected, both Neo-7 variants exhibited improved thermostability (71.12°C and 70.39°C for the single and double mutation variants, respectively) compared to WT-IL-7 (56.35°C; Table 2). These results confirm that our novel designs confer better protein packing and higher stability than WT-IL-7.

Characterization of E. coli expressed IL-7 and Neo-7s and the in vitro biological activities.

FPLC profile of E. coli expressed (A) WT-IL7 (B) refolded WT-IL7 (C) Neo-7-Q6P and (D) Neo-7-Q6P-T45I. Percentage of purity is calculated from the SEC-FPLC peak profile via Cytiva Unicorn 7 software after affinity chromatography purification. SPR (Biacore) characterization of the binding kinetics of (E) Neo-7-Q6P (F) Neo-7-Q6P-T45I and (G) WT-IL7 towards murine IL-7R alpha. (H) 2E8 proliferation assay to investigate the biological activity of the IL-7/Neo-7s expressed by E. coli. Error bars represent standard deviation (n=3).

Yield, purity, and thermostability data of the E. coli expressed IL-7/Neo-7s, yield was presented as mg of recombinant protein per 200 mL of E. coli culture.

| | Culture volume | Expression mode | Yield after Ni-NTA purification | Yield after FPLC purification | Purity from FPLC (%) | Thermostability (melting temperature) | | | ––––––– | ————— | —————————–– | —————————– | ––––––––––– | ————————————— | | Neo-7 (Q6P) | 200 mL | Soluble | 3.76 | 1.56 | 78.26 | 71.12 | | Neo-7 (Q6P+T45 I) | 200 mL | Soluble | 1.67 | 0.36 | 58.24 | 70.39 | | WT-IL7 | 200 mL | Refolding | 7.41 | 0.23 | 23.24 | 56.35 |

Recombinant Neo-7 shows enhanced binding affinity to murine IL-7 receptor alpha than WT IL7

We then characterized the binding affinities of Neo-7 variants to mouse IL-7 receptor alpha (mIL-7Rα) in a quantitative manner using surface plasmon resonance (SPR). The dual mutant variant of Neo-7 exhibited the most potent binding to mIL-7Rα, with a dissociation constant (KD) of 3.12 nM, followed by the Q6P variant (KD = 5.22 nM) and WT-IL-7 (KD = 13.4 nM; Figure 5E–G). Interestingly, through the SPR curve, we observed different binding kinetics between WT-IL-7 and the Neo-7 variants. WT-IL-7 binds to IL-7Rα through a fast-on-fast-off kinetic model. In contrast, Neo-7 variants displayed a slower on and off rate, with an approximately five times lower on-rates (ka) compared to WT-IL-7 (1.03 x 106 and 1.38 x 106 M–1s–1 for Neo-7-Q6P and Neo-7-Q6PT45I, respectively, versus 5.86 x 106 M–1s–1 for WT-IL-7). However, Neo-7 variants also demonstrated significantly slower off-rates (kd) (6.313 x 10–3 and 4.296 x 10–3 s–1 for Neo-7-Q6P and Neo-7-Q6PT45I, respectively, compared to 7.844 x 10–2 s–1 for WT-IL-7). This slower dissociation rate compensates for the slower association rate, resulting in more potent overall binding affinities. To assess the biological functions of Neo-7 variants, we then used the IL-7-dependent cell line 2E8 for a cell-based proliferation assay (Figure 5H). The ability of the Neo-7 variants in stimulating 2E8 proliferation is correlated with their binding kinetics determined by the SPR assay. At high concentrations (10 ng/mL), the differences in proliferative abilities were insignificant, likely due to saturation of the cytokine. However, at lower concentrations (1 ng/mL), the differences became more pronounced, with Neo-7 variants demonstrating stronger pro-survival signals compared to WT-IL-7. Encouraged by the promising binding affinity data from SPR and 2E8 bioassay, we then submitted the sequence of WT-IL7 and Neo-7s into the Immunomedicine group server for protein immunogenicity prediction. The predicted output showed that WT-IL7 is the most immunogenic with six potential antigenic fragments while the designed Neo-7s had three segments which all overlapped with the WT-IL7 antigenic fragments, indicating no additional immunogenic peptide fragments were introduced throughout the engineering processes of Neo-7 (Figure 5—figure supplement 1A-D).

Fc-fused Neo-7s show superior yield and purity profile in CHO-expression system

Subsequently, we designed a Fc-fused neo-7/WT-IL7 for in vivo anti-tumor assay and expressed the recombinant protein in a CHO-cell based system. Immunomodulatory cytokines are attractive anti-tumor agents due to their immuno-stimulating potential, but the in vivo efficacies are largely limited by their poor pharmacokinetic properties that resulted in rapid clearance from the circulatory system. We therefore fused WT-IL7 and the selected neo-7 variants with the Fc fragments of human IgG1 as a half-life prolonging strategy (Kwon et al., 2024; Foss et al., 2024; Martin et al., 2013). The fused Fc fragments were engineered with LALAPG mutations to attenuate Fc-mediated effector functions to prevent unintended cellular cytotoxicity in the host (Mausser et al., 2023). A structural illustration generated by AlphaFold simulation is shown in Figure 5—figure supplement 1E. During recombinant protein expression, Fc-fused WT-IL7 exhibited a poorer purity profile (31.84%) compared to the Neo-7 Q6P variant (80.88%) and the double mutation variant (55.15%), indicating that the superior production yield and purity of Neo-7 extend beyond prokaryotic systems to mammalian CHO-S cell lines (Figure 6A–C). Following FPLC purification, we assessed the bioactivities of the Fc control, Fc-fused IL-7, and Neo-7 variants using a murine spleenocyte proliferation assay, with flow cytometry evaluating viability and cell count at Day 3 and Day 7 post-treatment (Figure 6D–E). All treatments maintained spleenocyte viability and numbers relative to the Fc control, with a slight but not statistically significant increase in CD45+ cells in the Fc-Neo-7-Q6PT45I group. Our findings demonstrate that Neo-7 variants offer superior yield and purity compared to WT-IL7 in both bacteria (Table 2) and mammalian expression systems (Table 3).

Characterization of CHO-S cells expressed FC-fused cytokines and their in vitro and in vivo biological activities.

FPLC profile of CHO-S expressed (A) WT-IL7 (B) Neo-7-Q6P and (C) Neo-7-Q6P-T45I. Percentage of purity is calculated from the SEC-FPLC peak profile via Cytiva Unicorn 7 software after affinity chromatography purification. (D–E) Murine splenocyte proliferation assays performed at day 3 and day 7 following treatment with Fc-control (gray), Fc-WT-IL7 (red), Fc-Neo-7-Q6P (blue), or Fc-Neo-7-Q6P-T45I (green). In vivo immune stimulatory ability of the Fc-fused cytokines on murine PBMCs at day 0 to day 12 post-treatment. The data are presented as a count of (F) total viable CD45+ cells (G) viable CD45+ CD3+ CD4+ T cells (H) viable CD45+ CD3+ CD8+ T cells (I) viable CD45+ CD3- NK1.1+NK cells. Treatment groups are colored as Fc-control (gray), Fc-WT-IL7 (red), Fc-Neo-7-Q6P (blue), and Fc-Neo-7-Q6P-T45I (green). All data were presented as individual data plots with error bars (SEM) (n=3). Statistical differences among groups were determined using one-way ANOVA with Turkey’s multiple comparison test. Significance levels are defined as follows *p < 0.05; **p = 0.01–0.05; ***p = 0.0001–0.001; and ****p < 0.0001.

Yield and purity data of the CHO-S expressed Fc-fused IL-7/Neo-7s, yields were presented as mg of recombinant protein per 100 mL of CHO cell culture.

| | Culture volume | Expression host | Yield after protein A affinity purification/100 mL culture | Yield after FPLC purification/100 mL culture | Purity from FPLC curve (%) | | | ––––––– | ————— | ––––––––––––––––––––––––––––– | –––––––––––––––––––––– | –––––––––––––– | | Neo-7-Fc | 130 mL | CHO | 15.73 | 10.44 | 80.88 | | Neo-7 T43I-Fc | 130 mL | CHO | 16.66 | 6.42 | 55.15 | | WT-IL7-Fc | 130 mL | CHO | 17.22 | 3.37 | 31.84 |

Fc-fused Neo-7s shows superior ability to promote T-cell expansions in B6 mice

After validating the biological functions of the Fc-fused cytokines in vitro, we then investigated their in vivo effects on the immune cell composition in mice peripheral blood. Each mouse received a single intraperitoneal dose of 10 mg/kg Fc-fused IL-7 or Neo-7 variants. Blood samples were collected before treatment (D0) and at 3, 7, and 12 days post-treatment to analyze immune cell composition. We focused on CD45+ cells as an indicator of total immune cells and further examined the subpopulations and numbers of CD3+ CD4+ T cells, CD3+ CD8+ T cells, and CD3- NK1.1+ NK cells, which are the primary effector cell types contributing to the therapeutic effects of cytokine therapies. As shown in Figure 6F–I, all treatments exhibited a similar pattern of increased CD45+ immune cells, peaking 7 days post-treatment. Both Fc-fused Neo-7 variants induced significantly higher levels of CD45+ immune cells compared to Fc-WT IL-7 treatment, particularly in the CD8+ T cell subpopulation.

Fc-fused Neo-7s shows stronger anti-tumor activity in comparison to the WT counterpart

Given the superior immune stimulatory effects of Fc-fused Neo-7s compared to Fc-WT-IL7, we hypothesized that these effects could translate into enhanced anti-tumor activity. To test this, we treated mice bearing subcutaneous MC38 tumors (a syngeneic model for colorectal cancer) with a single dose of 10 mg/kg Fc-fused IL-7 or Neo-7s and monitored tumor growth. All Fc-fused IL-7 treatments showed significant tumor suppression compared to the Fc control (human IgG-1 Fc with LALAPG mutation), with Fc-fused Neo-7 (Q6PT45I) being the most potent, followed by Fc-fused Neo-7 (Q6P) and Fc-fused WT-IL-7 (Figure 7A, Figure 7—figure supplement 2).

In vivo anti-tumor activity of Fc-fused WT-IL7/Neo-7s towards MC38 syngeneic models.

(A) Tumor growth curve of MC38 tumor after treatment with the Fc-fused cytokines. Proportion of total leukocytes and different immune cell types within the MC38 tumor of mice at day 7 post-treatment. (B) Proportion of intratumoral CD4 +and CD8+ T cells (C) Proportion of other immune cell types present in the TME at day 7 post-treatment. All data were presented as individual data plots with error bars (SEM). Statistical differences among groups were determined using two-way ANOVA and one-way ANOVA with Turkey’s multiple comparison test for tumor growth curve and TILs analysis, respectively. Significance levels are defined as follows *p < 0.05; **p = 0.01–0.05; ***p = 0.0001–0.001; and ****p < 0.0001.

To explore the possible anti-tumor mechanisms, we conducted a replicate anti-tumor assay and analyzed leukocyte counts and populations in the peripheral blood and tumor of treated mice. We found that the count and proportion of total immune cells and CD3+ CD8+ T cells within the tumor correlated directly with the anti-tumor potency of the treatment groups (Figure 7C). In the peripheral blood of tumor-harboring mice, all Fc-cytokine fusions increased the total immune cell count compared to the Fc control, in the order of Fc-Neo-7-Q6P-T45I, Fc-Neo-7-Q6P, and Fc-WT-IL-7 (Figure 7—figure supplement 1). This trend was also observed for CD4+ T cells, CD8+ T cells, CD19+ B cells, CD11c+ dendritic cells, and F4/80+ macrophages. Among the immune cells analyzed, CD8+ T cells showed the most significant increase, indicating they might be the primary targets for the immune stimulatory effects of IL-7/Neo-7s (Figure 7B). No significant differences were observed in the count of Ly6C+ Ly6g- CD11b+ monocytes in the peripheral blood (Figure 7—figure supplement 1A). However, differences were observed between the single (Q6P) and double (Q6P-T45I) mutation variants of Fc-fused Neo-7 in the numbers of CD4+ FoxP3+ Treg and CD3- NK1.1+ NK cells within the peripheral blood. The Q6P variant showed the highest number of Tregs and NK cells among all treatment groups, followed by WT-IL7 and the Q6P-T45I variant. Additionally, only mice treated with Fc-WT-IL7 had significantly higher numbers of Ly6g+ CD11b+ neutrophils compared to other treatments (Figure 7—figure supplement 1A). These differences in immune cell regulation might be attributed to variations in binding affinity or kinetics of WT-IL-7 and Neo-7s towards IL-7 receptors on immune cells. Further studies are required to unravel the specific mechanisms underlying this phenomenon.

In the analysis of tumor-infiltrating leukocytes, we found that only CD8+ T cells were significantly enriched in the tumor compared to the Fc control (except for CD19+ B-cells in the Fc-Neo-7-Q6P group), despite a substantial increase of other immune cell types (CD4+ T cells, macrophages, and NK cells) in the peripheral blood (Figure 7B–C). This indicates that CD8+ T cells are the primary immune effectors that infiltrate and contribute to the anti-tumor properties of the Fc-Neo-7 fusions. The CD8+ TILs also differed in their effector status, where the perforin-positive CD8+ populations were the highest in the Fc-Neo-7-Q6PT45I group, followed by Fc-Neo-7-Q6P, Fc-WT-IL7, and Fc control. Furthermore, only the Fc-Neo-7 fusions showed a significant reduction of the proportion of the monocytes and neutrophils within the tumor compared to the Fc control, a phenomenon not observed in mice treated with Fc-WT-IL7. The reduction of intra-tumoral monocytes and neutrophils-to-CD8 T cell ratio is a favorable outcome for immunotherapy, as both of these immune cell types have been reported to act as immunosuppressors, inhibiting the cytotoxic activity of CD8+ T cells within the tumor microenvironment (Figure 7C; Sarkar et al., 2023; Gallina et al., 2006; Ouyang et al., 2024).

Since IL-7 is also known to promote the generation and maintenance of memory T cells, we also investigated the naive and memory populations of T cells using CD44 and CD62L markers (Figure 7—figure supplement 1B-C). Most of the CD4+ T cells circulating in the peripheral blood of the Fc control group were naive T cells (CD62L+ CD44-). However, treatment with Fc-IL-7/Neo-7s increased the number of effector memory (EM) populations (CD62L- CD44+). For CD8+ T cells, treatment with the Fc-cytokine fusion greatly enriched the CD8+ T cell population with a central memory (CM) phenotype (CD62L+ CD44+), followed by naive and effector memory T cells. In the MC38 tumor, most infiltrating T cells were identified as effector memory T cells. The majority of CD4+ tumor-infiltrating T cells exhibited an EM phenotype, with one exception showing a higher number of naive CD4+ T cells in mice treated with Fc-WT-IL7. For tumor-infiltrating CD8+ T cells, nearly all were positive for the memory marker CD44, with the main populations being EM CD8+ T cells (CD62L- CD44+), followed by CM CD8+ T cells (CD62L+ CD44+).

Fc-Neo-7 (Q6P-T451) greatly enhances expression of genes related to T-cell proliferations and effector functions without up-regulation of exhaustion-related genes

To further investigate the differences between the treatment effects of Fc-Neo-7 (Q6PT43I), Fc-IL-7, and the Fc control, we conducted an RNA-sequencing experiment on the splenic CD8+ T cells of the treated mice. CD8+ T cells were chosen due to their significant responsiveness to the treatments and their abundance in the spleen, ensuring sufficient RNA for sequencing. Principal component analysis (PCA) of the RNA-seq data revealed distinct RNA expression profiles for each treatment group, with clear clustering and segregation among the groups (Figure 8A). Gene ontology (GO) analysis of the differentially expressed genes showed that both Fc-IL-7 and Fc-Neo-7 upregulated genes related to immune system processes, immune responses, and inflammatory responses relative to the Fc control. Comparing Fc-IL-7 and Fc-Neo-7, we found that genes related to cell proliferation (cell cycle, cell division, and DNA replication) were the most prominently upregulated, corroborating our in vivo mice PBMC proliferation assay data (Figure 8B).

RNA sequencing of splenic CD8+T cells isolated from Fc-IL7/Neo-7s-treated mice.

(A) Gene ontology analysis of the gene expression data from RNA sequencing (n=3; three independent biological donors for each group). (B) Gene Set Enrichment Analysis (GSEA) of splenic CD8+T cells treated by Fc-Neo-7 versus Fc-WT-IL7. (C) Principal component analysis and (D) Gene expression heatmap derived from Z-scores calculated from the RNA sequencing data. The gene expression heatmap is derived from Z-scores calculated from the RNA sequencing data, with expression levels color-coded from high (red) to low (blue).

Further examination of the gene expression heatmap (Figure 8C) showed strong differentiation in the gene expression profile among the two Fc-Neo-7 variants with Fc-WT-IL7. Gene clusters associated with T cell proliferation, fitness, and functionality revealed that Fc-Neo-7 treatment upregulated cell cycle-related genes (Cdk2, Cdk4, Cdk6, Mki67) and pro-survival signals (Bcl-2, Brca1, survivin/BirC5) (Kim et al., 2020; Xue et al., 2010; Eum et al., 2024; Jacobs et al., 2010). In addition, a general increase in the expression of genes related to T-cell metabolism was also observed, including those regulating oxidative phosphorylation, fatty acid oxidation, amino acid (glycine, serine threonine pathway), and nucleotide metabolism, indicating an active proliferation and metabolism stage of the CD8 T-cells post-FC-Neo-7 treatments (Hope and Salmond, 2021; Han et al., 2021; Shyer et al., 2020). The serine to glycine one-carbon metabolism pathway was essential to support T cell proliferation by fueling purine biosynthesis and improving mitochondrial respiration efficiency (Ma et al., 2017; Sugiura et al., 2022). Fc-Neo-7 also primed T cells for enhanced functionality, as evidenced by the upregulation of genes responsible for cytotoxicity (granzymes A, B, and C) and surface receptors for IL-2 (IL2Ra, IL2Rb) and IL-12 (IL12rb), which prepared T cell differentiation into effector types. Notably, while IL-2-induced T cell proliferation often leads to exhaustion and activation-induced cell death (ACID), Fc-Neo-7 treatment resulted in reduced expression of inhibitory molecules (Tgfbr2, Lag3, CD83, PD-1) except TIGIT (Anderson et al., 2016; Gumber and Wang, 2022). This suggests a potential combination therapy of Fc-Neo-7 with anti-TIGIT to further enhance CD8+T cell effector activity by blocking the immunosuppressive TIGIT pathway. In terms of exhaustion markers, Fc-Neo-7-treated CD8+T cells exhibited marked reductions in the expression of genes such as Tim-3 (Havcr2), Tox, and CTLA4 (Anderson et al., 2016; Gumber and Wang, 2022). These findings were also confirmed by GSEA analysis, which showed significant enrichment of gene sets related to cell cycle and DNA replication (T cell proliferation), oxidative phosphorylation, and fatty acid metabolism (naive T cell metabolism). In addition, a lower expression of the genes related to the PD1-PDL1 inhibitory axis was also observed (Figure 8D).

Discussion

Cytokines are potent immunomodulatory proteins that serve as attractive therapeutic agents in numerous disease applications. However, clinically, their application is limited due to the need to balance the clinical potency to toxicity and poor pharmacokinetics (Fu et al., 2023; Winer et al., 2022; Tang and Harding, 2019). In this study, we addressed these issues through (1) proper selection of a relatively safe cytokine, IL-7, and (2) redesign of the scaffold of IL-7 via computational protein design to improve the stability and developability of the final molecule, Neo-7. Throughout the design process, we observed that the original structure of the scaffold of WT IL-7 is complex and very inefficient, involving long inter-helices loops that required additional disulfide stapling to ensure proper folding. Since only the helices of IL-7 are required for receptor binding, we chose to truncate all the interconnecting loops and remodel the helices connectivity via Rosetta loop remodeling program. The difference between the design of IL2/15 mimics (Neo 2/15) and this study lies in the design approaches. Neo2/15 was a completely de novo designed protein which required redesigning the mimetics from scratch, while the Neo-7 designed in this study still retains the sequence on the functional-related moiety (receptor binding helices) and thus minimizes the design steps and computational power required (Silva et al., 2019). However, it should be noted that our study has a somewhat different objective in that the Neo2/15 is a product from the breakthrough invention of the de novo protein design algorithm from the Baker lab, while the aim of this study is to search for a viable design idea that utilizes the availability of computational protein design (CPD) toolkits such as AlphaFold and Rosetta protein design suite for the redesign of a therapeutic-relevant cytokine.

In our approach, we avoided the need for extensive rounds of directed evolution and affinity maturation process. During the design process, we utilize AlphaFold2 to serve as a validation point for our designed sequence. We observed that the single sequence mode of AlphaFold2 provided a better discriminative power on the designed sequence as the default mode of AlphaFold2 (with reference to the MSA library) provided higher false positive results (bad sequence generating very similar structure to the WT-IL7 helices), possibly due to the highly similar sequences between the WT-IL7 and Neo-7. In addition, we also discovered that the presence of a kink on helix 1 is crucial for the interaction between the IL-7Rα and IL-7/Neo-7. The re-addition of the kink in helix 1 via disulfide stapling and mutation transfer guided by structural visual inspection generated two Neo-7 variants that are more potent in IL-7R binding affinity in comparison to the WT-IL7. Encouragingly, both the Neo-7 variants are also more thermostable compared to the WT-IL-7 and showed a much higher yield and purity. When expressed in E. coli (as single entity) and in CHO cells (as Fc-Fusion), indicating that the simpler protein scaffold of Neo-7s is much better handled by the cellular machineries during overexpression. Interestingly, the redesigned scaffold of Neo-7s led to a different binding kinetics towards IL-7Rα in comparison to the WT-IL7. The Neo-7s adopted a slow-on-slow-off model, which we ascribed to the more stable fold as a more flexible (less stable) fold in the WT-IL7 could probably increase the chances to bind to the IL7Rα via an induced-fit mechanism, but the overall flexibilities of the protein will also lead to stronger destabilization of the protein-receptor complex and therefore lead to a faster off-rate (McElroy et al., 2009; Broom et al., 2015). In contrast, a more stable fold adopted by the Neo-7s would require longer time for the protein-receptor association but also a slower dissociation rate once the complex has been formed (Celej et al., 2003; Majewski et al., 2019).

Moreover, the comparison of the in vitro and in vivo efficacies of the Fc-Neo-7s and Fc-IL-7 also highlighted the importance of in vivo stabilities during animal studies. In vitro murine spleenocyte proliferation assays showed no significant differences between the samples treated with Fc-Neo-7s and Fc-IL-7, but the in vivo PBMC analysis showed a significant increase in the count of CD8+ T cells (peaked at day 7), possibly due to the better stability of Fc-Neo-7s in the mice circulatory system. Encouragingly, the in vivo immune stimulation activities of the Fc-IL-7/Neo-7s also translated into anti-tumor activities where the efficacy of tumor growth delay was correlated with the binding affinities to IL-7Rα determined in the yeast display and SPR assays. Further analysis of the TILs also revealed a significantly higher infiltration of the CD8+T cells in the tumor, and the higher proportion of Perforin + CD8+ T cells also suggests that the T cells induced by Fc-Neo-7s treatment