Optogenetically engineered Septin-7 enhances immune cell infiltration of tumor spheroids

Significance

Cancer is a major public health concern and the second-most common cause of death worldwide. Traditional approaches for cancer treatment have been frustrating as they only extend patients’ lifetime along with sacrificing patients’ life quality. During the last decade, chimeric antigen receptor (CAR)-T therapy has emerged and shown great success in certain hematological malignancies including lymphoblastic leukemia and B cell lymphoma. However, the efficacy of using cellular immunotherapy to treat solid tumors is limited due to the poor tumor infiltration of immune cells. Here, we enhance immune cell penetration into tumor spheroids by manipulating septin-7 function in live cells, which might improve cell-based immunotherapies against solid tumors.

Abstract

Chimeric antigen receptor T cell therapies have achieved great success in eradicating some liquid tumors, whereas the preclinical results in treating solid tumors have proven less decisive. One of the principal challenges in solid tumor treatment is the physical barrier composed of a dense extracellular matrix, which prevents immune cells from penetrating the tissue to attack intratumoral cancer cells. Here, we improve immune cell infiltration into solid tumors by manipulating septin-7 functions in cells. Using protein allosteric design, we reprogram the three-dimensional structure of septin-7 and insert a blue light-responsive light-oxygen-voltage-sensing domain 2 (LOV2), creating a light-controllable septin-7-LOV2 hybrid protein. Blue light inhibits septin-7 function in live cells, inducing extended cell protrusions and cell polarization, enhancing cell transmigration efficiency through confining spaces. We genetically edited human natural killer cell line (NK92) and mouse primary CD8⁺ T-cells expressing the engineered protein, and we demonstrated improved penetration and cytotoxicity against various tumor spheroid models. Our proposed strategy to enhance immune cell infiltration is compatible with other methodologies and therefore, could be used in combination to further improve cell-based immunotherapies against solid tumors.

Cell-based immunotherapy has revolutionized pharmaceuticals and emerged as the fifth pillar of cancer treatment (1). The forerunner of cellular immunotherapeutics currently under development is chimeric antigen receptor (CAR) T cell therapies, six of which have been approved by the United States Food and Drug Administration since 2017 (2). CAR T cell therapy works by engineering native patient T-cells to express CARs tailored to recognize tumor-specific antigens (1). CAR T cell therapy has shown tremendous success in some liquid tumors, particularly hematopoietic malignancies (3, 4), whereas it has not performed as well in preclinical studies and clinical trials for solid tumors, with few patients achieving good responses (5). Solid tumors consist of abnormal and heterotypic cells that multiply uncontrollably and communicate through gap and tight junctions, preventing easy infiltration into the center of the tumor (6). Additionally, in contrast to liquid tumors, the mass of cancer cells in solid tumors generates a unique immunosuppressive microenvironment contributing to the failure of CAR T cell therapy. The critical obstacles preventing CAR T cell therapy for solid tumors include: i) the dense extracellular matrix surrounding solid tumors functions as a physical barrier that inhibits immune cell penetration and cytotoxicity; ii) the hostile tumor microenvironment, such as hypoxia and nutrient deficiency, inhibits the vitality and antitumor activity of immune cells; and iii) the high cellular heterogeneity of solid tumors makes it difficult to identify specific and reliable tumor antigens to target (7). Solid tumors make up approximately 90% of adult human cancers and 40% of childhood cancers (8, 9). Due to the high incidence of solid tumors among all cancers and the low efficacy of current treatments, immunotherapy for solid tumors is a crucial and exciting avenue to improving patient survival and cancer cure rates.

Successful recruitment of immune cells to tumor sites and infiltration into the targeted tumors are required for effective therapeutics (10). In addition, increased immune cell penetration into tumors is closely related to improved patient survival (11). Therefore, enhancing immune cell infiltration is a critical step to improve cell-based immunotherapies for solid tumors. Several tumor-targeting light- or pH-inducible nanoplatforms have been developed to simultaneously modify the immunosuppressive microenvironment and promote intratumoral penetration of cytotoxic T-cells through the codelivery of drugs with various functionalities (12–15). For instance, Li et al. (16) designed a pH-sensitive nanoliposome that selectively disassembled and released the drugs losartan (LOS) and polyamidoamine-conjugated doxorubicin (DP) in the acidic tumor microenvironment. LOS improved T cell infiltration by inhibiting the activity of cancer-associated fibroblasts, while DP induced immunogenic cell death, which reversed the immunosuppressive tumor microenvironment and stimulated immune response. Other approaches to assist immune cell infiltration have been explored with different cell types. Neutrophils can cross tissue barriers and penetrate tumors through self-deformation. Utilizing this capability, Hao et al. (17) engineered a tumor-penetrating neutrophil-based cytopharmaceutical, which carried agonists of the stimulator of interferon genes (STING). Activation of the STING pathway reinvigorated the tumor environment and enhanced the infiltration and cytotoxicity of T-cells. Finally, as a member of CD4⁺ T-cells, T helper type 9 cells (Th9) are not only hyperproliferative, highly cytolytic, and less exhausted, but also can recruit/activate other immune cells (18). In a recent study by Chen et al. Th9 cells decorated with tumor-targeting peptides exhibited enhanced infiltration into solid tumors, which led to better antitumor efficacy (10).

All the strategies above were developed to enhance immune cell infiltration, relying on the delivery of chemotherapeutic drugs or tumor-targeted peptides. Here, we propose an approach to improve immune cell penetration by directly manipulating immune cell behavior. By controlling target protein functions in the cell, we hope to change cell behaviors and phenotypes to augment immune cell infiltration into solid tumors. We focused on septin proteins, which are cytoskeletal guanosine triphosphate (GTP)-binding proteins that participate in various essential biological processes, such as cytokinesis, membrane remodeling, and cytoskeleton organization (19–21). A total of 13 mammalian septin proteins (septin 1-12 and septin-14) have been identified and classified into four subgroups, i.e., septin-2, septin-3, septin-6, and septin-7 (19). Septin-7 knockdown T-cells exhibited a fivefold improvement in transmigration efficiency compared to control cells when migrating through 3 and 5-μm pores (22). Since 1) the high spatial density of collagen fibrils (<5 μm apart) surrounding solid tumors significantly inhibits immune cell penetration, and 2) septin-7 knockdown enhances cell penetration through 3 μm pores, we hypothesized that the immune cell infiltration into solid tumors can be improved by controlling septin-7 function in cells.

Here, we designed a light-controllable septin-7 hybrid protein (LCS7) by incorporating a blue light-sensitive LOV2 domain with septin-7. We observed extended cell protrusions and increased cell polarity in cells expressing LCS7 upon blue light irradiation. By introducing LCS7 into immune cells, we enhanced cell transmigration efficiency and improved immune cell penetration into tumor spheroids, substantially increasing antitumor efficacy. Our approach expands the immunotherapy arsenal against solid tumors, which can be combined with other methodologies to further improve cell-based immunotherapies.

Results

Manipulating Cell Behaviors through Allosteric Design and Engineering of Target Proteins.

In biology, allostery is a common phenomenon whereby a perturbation that occurs at one site (the allosteric site), such as through ligand binding or light absorption, induces a conformational change of the molecule and, in turn, influences the function/activity at a distal site (the active site) (23–25). By engineering allostery, sensing modules have been integrated into target proteins to allow functional control of proteins (26–30). The rationale is that sensor domains undergo a conformational change upon detection of environmental cues including light, pH, temperature, and molecules (Fig. 1). Introducing this structural perturbation to a target protein through structural rewiring at the allosteric site, results in a sensor-fused protein whose function is regulated by corresponding input signals (Fig. 1). Functional change of target proteins in vivo affects cellular functions and behaviors. Therefore, we use input signals to manipulate cell behaviors to achieve various therapeutic purposes (25, 31–33). To control cell transmigration and enhance immune cell infiltration, we selected the blue light-responsive LOV2 domain as our sensor module and septin-7 as our target protein. By using blue light to switch septin-7 between inactive and active states, we aim to enhance immune cell infiltration and cytotoxicity against solid tumors.

Fig. 1.

The rationale of improving immune cell penetration through cell engineering. Sensor domains detecting various environmental signals could be integrated into a target protein through allosteric networks. The engineered target protein, thereafter, can respond to the input signals, which alter protein conformation and function. The functional change of target protein will further affect cell behaviors, e.g., cell penetration into solid tumors.

Computer-Aided Design of LCS7 Construct and Validation by Discrete Molecular Dynamics (DMD) (34–37) Simulations.

To design LCS7 with blue light sensing ability, we use Ohm (38) to identify the allosteric site for LOV2 insertion on septin-7. This LOV2 insertion site must satisfy two criteria, one of which is that sensor domain insertion should avoid disrupting the original structure and function of septin-7. Therefore, a surface-exposed and evolutionarily nonconservative site is ideal. We identified five potential insertion sites (loop1-5, Fig. 2B) by visually inspecting the three-dimensional structure of septin-7 (AlphaFold ID: AF-Q16181-F1) in PyMOL. We further calculated the normalized solvent-accessible surface area (SASA) and estimated the conservation values (SI Appendix, Table S1). All five insertion sites are surface-exposed as most of the SASA values are larger than 40% (Fig. 2A and SI Appendix, Table S1) (39, 40). Most of the residues at the predicted insertion sites are nonconservative except for loop2, in which both residues are relatively conserved (Fig. 2A and SI Appendix, Table S1).

Fig. 2.

Structural bioinformatics of potential insertion sites in septin-7 and DMD simulations of septin-7-LOV2 dark and lit mutants. (A) Surface exposure, evolutionary conservation, secondary structure (SS), and contact map are used to identify potential LOV2 insertion sites. All the potential sites (loop1 ~ 5) for LOV2 insertion are highlighted in magenta circle. (B) The three-dimensional structure of septin-7 predicted by AlphaFold (ID: AF-Q16181-F1). Five loops indicating potential LOV2 insertion sites are marked in magenta. (C) Representative structures of septin-7-LOV2 dark and lit mutants after equilibrium DMD simulations. The N-terminal region of septin-7 is marked in gray, the C-terminal part of septin-7 is highlighted in pink, and the LOV2 domain is marked in light blue. (D) The distribution of distances between the insertion sites of loop5 for septin-7-LOV2 (dark) and spetin7-LOV2 (lit) mutants during simulations.

The second requirement for establishing a successful allosteric regulation is that the structural disorder induced by the LOV2 domain should affect the target protein conformation hence activating/inactivating its functions. Septin-7 interacts with other septin proteins forming core septin hexamers (septin-2-septin-6- septin-7-septin-7-septin-6-septin-2) and octamers (septin-2-septin-6-septin-7-septin-9-septin-9-septin-7-septin-6-septin-2), which further polymerize into filaments and higher-order structures (41). The assembled septin structures on the cell cortex integrate with actin and microtubule cytoskeletal networks and further, function in cell division, membrane remodeling, and cell polarity maintenance (20, 21). The hetero-oligomer assembly mediated by septin-7 was found to be regulated by a potential phosphorylation site Y319 in septin-7 (42). Another active site T198, the constitutive phosphorylation of which inhibited septin-7-septin-7 interaction, while maintaining the interactions within septin-7-septin-6-septin-2 (43). The disruption of hexamer formation disturbed septin filament formation and ciliogenesis (43). Since both Y319 and T198 are essential for septin oligomer formation, we analyzed the allosteric communication pathways from each of the five insertion sites (loop1 to 5) to these two active sites (38). Among the predicted top 10 allosteric communication pathways (SI Appendix, Tables S2 and S3), loop5 ranked at the first place meaning that loop5 had a stronger allosteric regulation than the other insertion sites (SI Appendix, Fig. S1). Utilizing the loop5 design, we performed DMD simulations to study whether activated and inactivated LOV2 causes a conformational change in septin-7.

We established a septin-7-LOV2 fused structure in silico and used LOV2 dark and lit mutants in DMD to simulate the dark and blue light conditions in experiments. During simulations, residues R265-G277 of septin-7-LOV2 lit mutant stretched out forming a β-sheet structure with LOV2 (Fig. 2C). This secondary structure (SS) formation pulled the two parts together resulting in a more compact septin-7-LOV2 conformation. This structural change may prevent septin-7 from interacting with other septins due to the steric hindrance. When LOV2 is inactivated (dark mutant), the same region is disordered and sits between septin-7 and LOV2 (Fig. 2C). The LOV2 domain is located farther from septin-7 compared to the lit mutant, and therefore, may not interfere with septin-7 oligomer formation with its partners. To illustrate the structural change induced by LOV2 during simulations, we measured the distance between residues at the insertion site (Fig. 2D). The distances of the insertion sites in the septin-7-LOV2 lit mutant are much larger than those of the dark mutant, implying that activation of LOV2 introduces structural disturbance to septin-7, preventing its oligomerization and cellular functions.

Blue Light-Exposed LCS7-Expressing Cells Demonstrate Prominent Cell Protrusions, Increased Cell Polarity, and Spindle-Shaped Morphology.

We inserted the LOV2 domain at all the five predicted insertion sites (loop1 to 5) with GSG linker and synthesized the plasmids for transient expression in mammalian cells. We performed live cell imaging experiments in the dark to study cell morphological changes upon blue light irradiation. Except for the septin-7-LOV2-loop5 design, we did not observe any substantial phenotypic change in cells expressing the other constructs. The experimental data further confirmed our computational prediction and analysis. We focused on the loop5 design as our LCS7. In the dark, COS-7 cells expressing LCS7 spread on collagen grids with a dendritic phenotype (Fig. 3A and Movie S1). Upon blue light irradiation, cells moved around with high dynamics and formed extended cell protrusions (Fig. 3A and Movie S1). To quantify the elongated cell protrusions, we selected six snapshots of cells across 6 h uniformly in both dark and blue light conditions, and measured cell width and height. For wild-type COS-7 cells, cell width and height are similar under both dark and blue light conditions (SI Appendix, Fig. S2), indicating that the application of blue light did not affect cell morphology. In contrast, we observed a significant increase of cell width and height upon blue light illumination in COS-7 cells harboring LCS7 (Fig. 3B). The morphological change of extended cell protrusions is consistent with a previous study when short hairpin RNA (shRNA) was used to knockdown septin-7 expression in immune cells (22).

Fig. 3.

Extended cell protrusions and increased cell polarity observed in live cells expressing LCS7 upon blue light illumination. (A) Confocal images of LCS7-transfected COS-7 cells on collagen grids. The live cell experiment was conducted for 18 h including first 6 h in the dark and the left 12 h under blue light. Snapshots of live cells at different time points are shown. (B) Quantification of normalized cell width and height of LCS7-transfected COS-7 cells on collagen grids. (C) Confocal images of MDA-MB-231 stable cells harboring LCS7 (marked with magenta arrow) on collagen rhomboid grid. The cell imaging experiment was performed for 18 h including 6 h in the dark and 12 h under blue light. (D) Quantification of cell aspect ratio at different time points in both dark and blue light conditions. Note: Collagen grid (A) and collagen rhomboid grid (C) are highlighted in cyan and are visible under blue light. The P values were calculated using the Kolmogorov–Smirnov test.

We further introduced LCS7 to MDA-MB-231 cells and studied cell behavior. In the dark, MDA-MB-231 cells spread on collagen rhomboid grids (Fig. 3C and Movie S2). Once we switched on blue light, cells started to elongate forming a spindle-shaped morphology (Fig. 3C and Movie S2). Prominent protrusions also formed in MDA-MB-231 cells, and a similar phenotype was observed in COS-7 cells. We quantified the aspect ratio of cells at different time points by dividing cell length by cell width (Fig. 3D). Cell aspect ratio increased significantly after blue light irradiation, implicating an increase in cell polarity (44). Based on these results, we propose that in our engineered LCS7 construct, blue light activates LOV2 introducing structural disturbance to the septin-7 protein, which inhibits septin-7 function and disrupts septin oligomerization. This function inhibition further switches cell behavior from a spreading phenotype to a more dynamic and polarized morphology with prominent protrusions.

Inactivation of Septin-7 Function Improves Cell Transmigration Efficiency.

When cells transmigrate through physical barriers (SI Appendix, Fig. S3), protrusion formation is the first step for a cell to migrate (45–47). To enter and pass the confining spaces, cells deform and elongate, resulting in a spindle-like shape with large aspect ratios (44). As we observed elongated protrusions and increased aspect ratio of cells expressing LCS7 under blue light, we wanted to know whether these morphological changes affect cell transmigration. We generated MDA-MB-231 and NK92 stable cell lines constitutively expressing LCS7. We used transwell inserts with 8 μm and 3 μm pore sizes for MDA-MB-231 and NK92, respectively (22, 48). Cells treated with blue light (blue light 1 h off/on and 2 h off/on) showed a significant increase in transmigrated cells (Fig. 4). When cells were illuminated with blue light for 4 h, however, we observed a similar transmigration efficiency that is not significantly different from cells in the dark (Fig. 4C). This might be due to the damaging effects of strong blue light exposure over a long duration (7 μW/cm², SI Appendix, Fig. S4A), which can cause photodamage and impair cell transmigration (49).

Fig. 4.

Enhanced transmigration efficiency of engineered MDA-MB-231-LCS7 and NK92-LCS7 cells. (A) Engineered MDA-MB-231-LCS7 and NK92-LCS7 stable cell lines were used for the transmigration assay. Transwell insert with 8 μm and 3 μm pore size was applied to MDA-MB-231-LCS7 and NK92-LCS7, respectively. Cells were incubated for 4 h and treated with dark or different blue light conditions. (B) Transmigrated MDA-MB-231-LCS7 cells attached to the bottom of the membrane and were fixed and stained. Red indicates MDA-MB-231-LCS7 cells. Yellow marks the pores on the membrane. (C) Quantification of the number of transmigrated MDA-MB-231-LCS7 and NK92-LCS7 stable cells (n = 6). In this experiment, various conditions were investigated including dark for 4 h, blue light for 4 h, blue light 1 h off and 1 h on (4 h total), and blue light 2 h off and 2 h on.

Engineered Immune Cells Expressing LCS7 Exhibit Enhanced Cytotoxicity against Three-Dimensional (3D) Tumor Spheroids.

To explore whether increased transmigration efficiency can assist immune cells in penetrating solid tumors, we established 3D tumor spheroids models using a human breast cancer cell line, MDA-MB-231, and cocultured them with lentiviral transduced NK92 cells expressing LCS7 (NK92-LCS7). Solid tumors in vivo are protected by a dense extracellular matrix that prohibits immune cell infiltration. To mimic the in vivo environment, we supplemented collagen in the cell culture during tumor spheroid formation (Fig. 5A). We found that collagen addition was crucial to the formation of round, uniform, and solid tumor spheroids. We cocultured tumor spheroids with NK92-LCS7 or wild-type NK92 (NK92-WT) and treated them with various dark and blue light conditions (Fig. 5B). We designed a programmable blue light emitting diode (LED) equipment (SI Appendix, Fig. S5) and tested the effect of different blue light frequencies on tumor spheroid killing using a low light intensity (3 μW/cm², SI Appendix, Fig. S4B). This low blue light intensity showed no phototoxicity as the growth of tumor spheroid alone was not affected by the blue light (SI Appendix, Fig. S6).

Fig. 5.

Engineered NK92-LCS7 cells show improved cytotoxicity against MDA-MB-231 tumor spheroids. (A) The approach used to produce MDA-MB-231 or Hela tumor spheroids. Ultralow attachment 96-well plate was used to generate solid tumor spheroids. (B) The images of MDA-MB-231 tumor spheroids on each day after coculturing with NK92-LCS7 cells. The coculturing system was treated with different dark and blue light conditions. BL-1 s means blue light 1 s on and 1 s off, etc. (C) Quantification of the integrated density of MDA-MB-231 tumor spheroids along the incubation time (Left) and the comparison of the percent change in integrated density of tumor spheroids treated with different conditions after 5 d incubation. The P values were calculated using Tukey’s multiple comparison test (n = 10).

The optical density of MDA-MB-231 tumor spheroids started to decrease after 2 d of incubation with NK92-LCS7 under blue light conditions (Fig. 5B), confirming tumor cell killing by NK92-LCS7. When cocultured with NK92-WT in both dark and blue light conditions or incubated with NK92-LCS7 in the dark (Fig. 5 B and C), we observed a transient increase in the integrated density of tumor spheroids, with the integrated density decreasing slowly during the incubation period. Strikingly, the tumor spheroid density decreased more in the center than the periphery after 2 d of coculturing with NK92-LCS7 under blue light conditions. We believe that the dark circle on the outer surface of the spheroids may be a collagen matrix mixed with tumor cells. The collagen-dense environment can influence the interaction between immune cells and cancer cells, thus reducing immune cell activation and removal of tumor cells on the outer surface. In the center of the tumor spheroids, however, few collagens exist and therefore, immune cells can interact with tumor cells and eliminate them with no physical barrier. These phenomena indicate that engineered immune cells can penetrate the dense extracellular matrix and infiltrate into tumor spheroids, thus gaining access to cells deep in the spheroid. We quantified and compared the percent change in integrated density of tumor spheroids after 5 d of incubation with NK92-WT or NK92-LCS7 treated with dark or various blue light conditions (Fig. 5 C, Right). Compared to the NK92-WT control and NK92-LCS7 in the dark, the integrated density decreased significantly for tumor spheroids cocultured with NK92-LCS7 under different blue light conditions. When we treated the coculture with various frequencies of blue light applications, however, no significant difference was observed.

Next, we investigated the generalizability of our approach in enhancing immune cell infiltration and cytotoxicity against tumor spheroids. We fabricated 3D tumor spheroids using human cervical carcinoma HeLa cells or mouse B16.F10-Tag melanoma cells, respectively. We adopted the same method (Fig. 5A) to synthesize HeLa tumor spheroids and cocultured them with NK92-LCS7 or NK92-WT cells. As shown in Fig. 6B, the integrated density of tumor spheroids increased initially and then, decreased slowly during the incubation, a similar trend occurred with MDA-MB-231 tumor spheroids. For HeLa tumor spheroids incubated with NK92-LCS7 under blue light irradiation, the integrated density reduced significantly, exhibiting a massive tumor cell killing and a greater cytotoxicity of NK92-LCS7. After 5 d of incubation, about 95% of the tumor spheroids were eliminated by NK92-LCS7 with blue light treatment, while only 65% of the tumor spheroids were destroyed by NK92-WT in both dark and blue light conditions or by NK92-LCS7 in the dark (Fig. 6B).

Fig. 6.

Engineered immune cells exhibit improved cytotoxicity against Hela or B16.F10-Tag tumor spheroids. (A) The schematic of the hanging drop method applied to establish B16.F10-Tag tumor spheroids and the timeline of coculturing engineered immune cells with B16.F10-Tag tumor spheroids. (B) Quantification of the integrated density of Hela tumor spheroids along the incubation time (Top) and the comparison of the decrease in integrated density of tumor spheroids treated with different conditions after 5 d coculturing (Bottom, n = 10). (C) Quantification of the integrated density of B16.F10-Tag tumor spheroids along the incubation (Top) and the comparison of the decrease in integrated density of B16.F10-Tag tumor spheroids after 5 d incubation (n = 6). The coculturing system was treated with different dark and blue light conditions. Blue light was 3 h on and 3 h off.

Mouse B16.F10-Tag melanoma tumor spheroids were loosely packed with random shapes when growing in the ultralow attachment 96-well plate (Fig. 5A). Therefore, we applied the hanging drop approach to establish melanoma tumor spheroids (Fig. 6A). We optimized the number of seeded cells and the volume of hanging drops (SI Appendix, Fig. S7). A volume of 20 μL RPMI media containing 5,000 B16.F10-Tag cells and 0.25% methylcellulose was able to produce melanoma tumor spheroids with a round and uniform shape. We cocultured B16.F10-Tag tumor spheroids with lentiviral transduced mouse primary CD8⁺ T-cells expressing LCS7 (T-LCS7) or wild-type CD8⁺ T-cells (T-WT). As shown in Fig. 6C, the integrated density of tumor spheroids decreased substantially since the beginning of incubation for all four different conditions. We did not observe an increase in tumor density, which is distinct from Hela and MDA-MB-231 tumor spheroids. As B16.F10-Tag cells do not aggregate as tightly as Hela or MDA-MB-231 cells, the loose structure of B16.F10-Tag tumor spheroids is easier for immune cells to penetrate. Therefore, the solidity of tumor spheroids is an important feature, which protects cancer cells from immune cell attacks. Overall, the integrated density of B16.F10-Tag tumor spheroids incubated with T-LCS7 under blue light illumination decreased faster than those cocultured with T-WT or T-LCS7 in the dark. After 5 d of coculturing, 99% of the tumor spheroids were devoured by T-LCS7 under blue light, while about 90% of the tumor spheroids were eliminated by T-WT or T-LCS7 in the dark.

Discussion

A previous study suppressing septin-7 gene expression using shRNAs reported the formation of excess leading-edge protrusions and an increase in cell transmigration in immune cells (22). However, these unspecific shRNAs also eliminated the entire septin complex, i.e. knockdown of septin-1, septin-6, septin-8, and septin-9 (22). Therefore, the role that septin-7 played in affecting cell behaviors and cell transmigration was still uncertain. Gene silencing via RNA interference or drug-based inhibition of protein function has poor specificity and could cause undesired side effects. Genetic knock-out of septin-7 could be deleterious to cells and biological systems due to its essential role in cell mitosis (50). To minimize the disturbance of cellular homeostasis as well as investigate septin-7 functions in vivo, we designed a light-inducible septin-7 protein construct. By specifically targeting septin-7, we demonstrated the cell morphological change upon septin-7 function inhibition, which helped us to better understand the physiological functions of this protein. Based on our computational studies and experimental data, we propose that in the dark engineered LCS7 functions as the wild-type septin-7 protein. Upon blue light illumination, the activated LOV2 domain introduces large structural perturbation to the septin-7 structure and impairs septin oligomer formation, which eventually leads to extensive cell protrusions, increased cell polarity, improved cell transmigration efficiency, and enhanced immune cell infiltration into tumor spheroids.

Here, we used an optogenetic tool, the LOV2 domain, to control target protein functions. Optogenetic sensors obtain several advantages compared to other sensors including flexibility, noninvasiveness, and high spatial-temporal resolution, which can be used to specifically target cancer cells and avoid affecting healthy tissues (51). Although the activation signal, blue light, has shallow tissue penetration, our strategy can be adapted with near-infrared-light activatable nanoparticles to target deep tissues, which absorb long wavelength near-infrared light and convert it to blue light (52, 53). In the future, other protein-based sensors reacting to various input signals (e.g., small molecules, pH, temperature, metals) could replace LOV2 at the same allosteric site to achieve diverse purposes.

We enhanced immune cell infiltration into tumor spheroids by manipulating cell behaviors and improving their transmigration efficiency through extracellular matrix barriers composed of dense and aligned collagen fibers. Besides this physical barrier outside tumor cell clusters, in vivo biological barriers including the tumor vasculature barrier and the chemokine barrier, also suppress T cell functions and prevent them from entering the tumor bed (11, 54). T-cells are recruited and home to the tumor site through the interactions between T-cells and homing receptor ligands (HRLs) on tumor vasculature (11). Expression deficiency of HRLs in tumors blocks immune cell binding and infiltration. For instance, angiogenic factors including vascular endothelial growth factor A and basic fibroblast growth factor suppress endothelial intercellular adhesion molecule 1 expression, which contributes to tumor evasion from T cell attack (55). Moreover, macromolecules such as programmed-death ligand 1, indoleamine 2,3- dioxygenase, and endothelin receptor B expressed by endothelial cells inhibit T cell homing and function (56). Fas ligand up-regulated on tumor-associated vasculature induces apoptosis of CD8⁺ T-cells and therefore, substantially impairs cytotoxic T cell penetration (57). Chemokines are another essential factor that modulates immune cell trafficking into solid tumors. Chemokine gradients guide cell migration and therefore are important for immune cell recruitment through endothelium and cell migration within the tumor microenvironment (58). By modifying chemokine networks, tumors can suppress T cell infiltration while recruiting tumor-promoting and immune-inhibitory cells including myeloid-derived suppressor cells and tumor-associated macrophages (59, 60). Due to the complex microenvironment of solid tumors, we expect to combine our approach with other strategies to achieve great therapeutic efficacy in treating solid tumors in vivo.

Materials and Methods

Structural Bioinformatics of Septin-7, DMD Simulations, and Simulation Data Analysis.

We performed structural bioinformatics analysis of septin-7 using the three-dimensional structure predicted by AlphaFold (ID: AF-Q16181-F1). We calculated the conservation value for every residue using ConSurf (https://consurf.tau.ac.il/) (61–63). We predicted the SS and calculated the SASA using STRIDE (http://webclu.bio.wzw.tum.de/stride/) (64). We plotted the contact map of septin-7 using Jupyter Notebook.

DMD simulations utilize discrete step function potentials to characterize the interaction between atoms rather than the continuous potentials applied in traditional MD simulations (34–37). Therefore, the DMD simulation engine is faster and more efficient than its peers, allowing the reach of longer timescales in simulations of large biomolecules (34–37). We performed DMD simulations of septin-7-LOV2 (LOV2 inserted at loop5) dark and lit mutants to study the conformational change and allosteric control. We constructed the fused structure of septin-7-LOV2 by following these steps. First, we manually attached the N and C termini of LOV2 (PDB ID: 2V1A) to the loop5 insertion site of septin-7, cut loop5, and renumbered all the residues based on the septin-7-LOV2 design in PyMOL. Then, we used this unconnected structure as a template in SWISS-MODEL (https://swissmodel.expasy.org/) (65) and used the complete septin-7-LOV2 amino acid sequence for homology modeling. We introduced LOV2 dark (C450A) and lit (I510E/I539E) mutations in the sequence during homology modeling to build their corresponding structures. Finally, we put the flavin mononucleotide cofactor back into the septin-7-LOV2 dark and septin-7-LOV2 lit structures in PyMOL and used these structures to initiate DMD simulations. To maintain the key interactions with the flavin cofactor within the dark or lit mutant, we added distance constraints (SI Appendix, Table S7) between atoms during simulations. We conducted DMD simulations for 2 × 10⁶ time steps (~50 ps per step) with a constant temperature at 0.62 kcal/mol·k_B (corresponding to 37 °C) that was maintained by an Anderson thermostat. We used a common heat exchange coefficient of 0.1 that specifies the heat transfer rate between the system and the thermostat-maintained implicit solvent. We used the parameters from the Medusa force field to analyze the original structures of septin-7-LOV2 dark and lit mutants, respectively, and generate the interactions and potentials among atoms. The dimension of the simulation box was a cube with a side length of 300 Å. We saved the modeled structures of the simulation trajectories every 500 time steps. After simulations, we calculated the distance between residues at the insertion site using GROMACS (66).

Molecular Cloning of Septin-7-LOV2 Constructs.

We purchased the cloning plasmid Septin-7_pET-28a(+)-TEV with the septin-7 DNA sequence from GenScript and used it as the template for PCRs. Primers (sept7-F and sept7-R, SI Appendix, Table S4) with HindIII and SalI restriction sites were designed and used to amplify the septin-7 sequence from the template. We isolated the resulting PCR product by DNA gel electrophoresis and purified it using Monarch^® DNA Gel Extraction Kit (Cat# T1020L). Then, we digested the PCR product and plasmid pmCherry-C1_SEPT9 i1 (Addgene item# 71622) separately by incubating them with HindIII (New England Biolabs (NEB), Cat# R3104S) and SalI (NEB, Cat# R3138S) enzymes according to the manufacturer’s protocol. We isolated and purified the digested DNA sequences. Eventually, we ligated the septin-7 DNA sequence to vector pmCherry-C1 through DNA ligation. We transformed the ligation mixture into homemade DH5α Escherichia coli competent cells (Zymo Research, Cat#T3002). We grew the transformed E. coli cells on an Luria-Bertani (LB) plate with kanamycin at 37 °C overnight. The next day, we picked single colonies and cultured them in LB medium with kanamycin at 220 rpm, 37 °C overnight. We extracted the plasmids using Monarch^® Plasmid Miniprep Kit (NEB, Cat#T1010L) and sent them for Sanger sequencing (Eton Biosciences). The constructed plasmid, pmCherry-C1_SEPT7, can be used to express septin-7 protein with mCherry tag at the N terminus in mammalian cells.

To integrate the LOV2 domain into septin-7 at different insertion sites (loop1~5), first we amplified the LOV2 DNA sequence using designed primers (SI Appendix, Table S4) and the template [pBabe-PI(WT)-Src(WT)-mCherry, Addgene item # 87356]. Amplified DNA products of LOV2 were isolated and purified by DNA electrophoresis and DNA gel extraction. Then, we used these LOV2 DNA products as the mega primer and used pmCherry-C1_SEPT7 as the template, and performed a PCR. After the PCR, we digested the original template by adding the DpnI (NEB, Cat# R0176S) enzyme based on the manufacturer’s instructions. We transformed this PCR mixture into DH5α competent cells and cultured them overnight on an LB plate with kanamycin at 37 °C. We picked the single colonies, cultured them, extracted the plasmids, and confirmed the DNA sequence using the protocol described above. The generated plasmids were named pmCherry-C1_SEPT7-LOV2-Loop1, pmCherry-C1_SEPT7-LOV2-Loop2, pmCherry-C1_SEPT7-LOV2-Loop3, pmCherry-C1_SEPT7-LOV2-Loop4, pmCherry-C1_SEPT7-LOV2-Loop5, respectively.

To generate stable cell lines, we designed lentiviral vectors encoding septin-7 constructs. We used primers (sept7-Lentil-F and sept7-Lentil-R, SI Appendix, Table S4) to amplify septin-7-LOV2 from plasmid pmCherry-C1_SEPT7-LOV2-Loop5. We isolated and purified the PCR product by DNA electrophoresis and DNA gel extraction. Then, we used NotI (NEB, Cat# R3189S) and AvrII (NEB, Cat# R0174S) restriction enzymes to digest the PCR product and the plasmid pLV-EF1a-IRES-Blast (Addgene item# 85133), respectively. After DNA digestion, we isolated and purified the products by DNA electrophoresis and DNA gel extraction. We ligated the septin-7-LOV2 fragment to the pLV-EF1a-IRES-Blast vector and confirmed the plasmid sequence by sequencing.

We performed all the above PCRs using a Q5 high-fidelity master mix (NEB, Cat#M0492S). We ordered all the primers (SI Appendix, Table S4) from Millipore Sigma. The complete gene sequence of Septin-7-LOV2-mCherry is shown in SI Appendix, Table S5.

Cell Culture.

We grew HeLa (American Type Culture Collection (ATCC) CCL-2), COS-7 (ATCC CRL-1651), and MDA-MB-231 (ATCC HTB-26) cells under 37 °C and 5% CO₂ in complete Dulbecco’s modified eagle medium (BioWhittaker, Cat# 12-604Q) that was supplemented with 10% fetal bovine serum (Corning™, Cat# MT35010CV), streptomycin/penicillin solution (Gibco™, Cat# 15070063), and nonessential amino acid (Cytiva, Cat# SH30238.01). We used B16.F10-Tag cells expressing SV40 large T and small t antigens, which were previously described (67, 68). We confirmed the expression of large T antigen by immunostaining. We cultured B16.F10-Tag cells in RPMI-1640 Media (Gibco, Cat# 61870036) containing 10 mM HEPES (DOT Scientific, DSH75050-1000), 50 μM 2-mercaptoethanol (Thermo Fisher Scientific, Cat# AC125470100), 25 μg/mL sodium pyruvate (Spectrum Chemical, Cat# SO193), 10% fetal bovine serum (Corning™, Cat# MT35010CV), and streptomycin/penicillin solution (Gibco™, Cat# 15070063). We grew mouse primary CD8⁺ T-cells in the above RPMI media supplemented with 30 U/mL IL-2 (Miltenyi, Cat# 130-097-744) and Dynabeads™ mouse T-activator CD3/CD28 (Thermo Fisher Scientific, Cat# 11456D). We cultured NK92 (ATCC CRL-2407) cells in the complete medium prepared by mixing 500 mL MyeloCult™ H5100 (STEMCELL Technologies, Cat# 05150) with 63 mL Gibco™ Horse Serum (Thermo Fisher Scientific, Cat# 16050130) and 10 μg IL-2 (Miltenyi, Cat# 130-097-744).

Transient Expression of Engineered LCS7 and Lentiviral Transduction.

To transiently express LCS7 in cells, we seeded 50,000 COS-7 cells on a 24-well plate and cultured the cells overnight. On the following day, we mixed transfection buffer and reagent with 500 ng of plasmid. Then, we transferred the mixture to cell culture according to the manufacturer’s protocol (Avantor, Cat#76299-634). After 24 h of incubation, we checked transfection efficiency and confirmed protein expression by inspecting the red fluorescence under a fluorescent microscope.

To generate stable cell lines expressing LCS7, we adopted lentiviral transduction. First, we produced a lentivirus containing LCS7. Specifically, we transfected HEK cells with the lentiviral packaging plasmid psPAX2 (Addgene, item# 12260), the envelope plasmid pMD2.G (Addgene, item# 12259), and the engineered plasmid pLV-EF1a-IRES-Blast containing LCS7. On the next day, we changed cell media and kept incubating HEK cells under 37 °C and 5% CO₂. We collected the cell media containing lentivirus for the following 2 d. We filtered lentiviral supernatant through 0.45 μm sterile polyvinylidene fluoride membrane (Basix™, Cat# 13-1001-05) to remove cell debris and stored lentivirus solution under −80 °C. Then, we performed lentiviral transduction into different cell lines. We concentrated lentivirus by mixing lentiviral supernatant with 1/3 volume of PEG 6000 solution for 4 h under 4 °C. The PEG 6000 solution was prepared by dissolving 80 g PEG 6000 (Thermo Fisher Scientific, Cat# 192280010) and 14 g NaCl (EMD Millipore, SX0420-3) in ultrapure water and 20 mL 10× phosphate-buffered saline (PBS) (Corning, 46-013-CM). After the chemicals were completely dissolved, we adjusted the pH to 7.0 ~ 7.2 and the final volume to 200 mL. The solution was sterilized through filtration of 0.2 μm membrane. Then, we centrifuged the lentiviral solution with PEG 6000 at 1,600 g, 4 °C for 1 h. After centrifugation, we removed the supernatant and resuspended the lentivirus with the corresponding cell media. To generate MDA-MB-231 stable cell lines, we added lentivirus and 8 μg/mL polybrene (Sigma-Aldrich, TR-1003-G) into cell media (complete Dulbecco's modified eagle medium (DMEM)) and incubated overnight. To transduce NK92 cells, we mixed lentivirus and 8 μg/mL polybrene with dispersed NK92 cells and then centrifuged the cells at 1,000 g for 1 h. To transduce mouse primary CD8⁺ T-cells, we added lentivirus and 8 μg/mL polybrene into the cell culture that contained IL-2 and Dynabeads and cultured the cells overnight. We verified the efficiency of transduction by flow cytometry.

Preparation of Elastic Micropatterns.

We fabricated two-dimensional collagen grids based on a previous protocol (69). Specifically, we labeled rabbit polyclonal anti-collagen-1 antibody (Abcam, Cat# ab34710) with biotin (Sigma-Aldrich, Cat# H1759) and Alexa Fluor 488 (Invitrogen™, Cat# A20000) following the manufacturer’s instructions. After the reaction, we transferred the mixture to a Slide-A-Lyzer™ MINI Dialysis Device (Thermo Fisher Scientific, Cat# 69560) and performed dialysis overnight in cold PBS. We kept the labeled anti-collagen antibody under 4 °C for future usage. To make collagen grids, we added 3 μL of labeled anti-collagen antibody solution on top of a composite elastomeric stamp printed with parallel 15 μm spaced lanes (69). We covered the antibody droplet with a 15 mm round glass coverslip for 2 h at room temperature to make an effective stamp surface coating. Then, we rinsed the stamps in ultrapure water and dried them under flowing nitrogen gas. We printed the first anti-collagen antibody pattern horizontally onto a square glass coverslip and then printed a second layer vertically onto the same coverslip to make a grid pattern with 90° angle. To make the rhomboid grids, we crossed the two sets of collagen lanes at a 22° angle. Then, we added 10 μL of polymerizing polyacrylamide (PAA) gel with G′ = 8.6 kPa, which was prepared according to a reported protocol, onto an activated 35 mm glass-bottom Petri dish (MatTek Corp., Part# P35G-1.0-20-C) and immediately covered the PAA gel with the above micropatterns. The Petri dish was previously activated by 3-(trimethoxysilyl) propyl methacrylate (Sigma-Aldrich, Cas#2530-85-0) in acetic acid and ethanol solution (69). After PAA gel polymerizing, we incubated the PAA “sandwich” with ultrapure water for 10 min and gently removed the glass coverslip. Finally, we aspirated the water, added 1 mL of 1 mg/mL rat collagen type-1 (Corning, Cat#354249) in cold PBS, and incubated under 4 °C overnight. On the following day, we aspirated the collagen solution and rinsed the micropatterns with cold PBS three times. We stored the micropatterns in cold sterile PBS before usage.

Cell Imaging and Data Analysis.

We seeded 100,000 of transfected COS-7 or MDA-MB-231 cells onto the micropatterns per Petri dish in complete DMEM. We placed the cells in the incubator for 4 h to let them attach to the micropattern. Then, we changed the culture media and removed the unattached cells before live cell imaging experiments.

We performed high-resolution cell imaging experiments using a 10× objective on a Leica SP8 confocal microscope with sequential lasers (405, 488, 561, and 647 nm). We used 561 nm (0.7% laser intensity) to visualize COS-7 or MDA-MB-231 cells expressing LCS7, which included a mCherry tag and applied 488 nm (0.2% laser intensity) to image the micropatterns as well as activate the LOV2 domain. We conducted the live cell imaging experiments in a Tokai Hit stage top incubator under 37 °C and 5% CO₂. We recorded the cell behavior in the dark for 6 h, and then switched the blue light on and imaged the live cells for 12 h. We utilized ImageJ 1.52 k to measure the width and height of each cell and used the Leica software to produce cell images and videos.

Transmigration Assay.

We used transwell inserts with 8 μm (Greiner Bio-One, Item# 662638) and 3 μm (Greiner Bio-One, Item# 662631) pore sizes for MDA-MB-231 and NK92 cells, respectively, since MDA-MB-231 cells are much bigger than NK92 cells. We seeded 100 μL of 10,000 lentiviral transduced MDA-MB-231 cells in DMEM into the transwell insert and put it in the incubator for 10 min. Then, we added 600 μL of complete DMEM containing fetal bovine serum in the well of the 24-well plate, which was at the bottom of the transwell insert. For NK92 transmigration assay, we used NK92 media instead of DMEM. The media at the bottom contained serum, while the media in the transwell insert was serum-free. We incubated the cells for 4 h and treated them with dark or different blue light conditions. We used a blue flashlight with a wavelength of around 450 nm and a light intensity of 7 μW/cm² for the experiments (SI Appendix, Fig. S4A). MDA-MB-231 cells attached to the bottom surface of the transwell insert after transmigration. We removed the untransmigrated cells in the insert using cotton swabs and fixed the transmigrated cells with a 4% paraformaldehyde solution (Thermo Fisher Scientific, J19943.K2). We stained the transmigrated cells using Alexa Fluor™ 594 phalloidin (Thermo Fisher Scientific, Cat# A12381). We counted the number of transmigrated cells under a fluorescent microscope. NK92 cells are distributed in the media after transmigration. Therefore, we centrifuged the media, collected the transmigrated cells, and counted the number of transmigrated cells using a hemocytometer.

Light Control Apparatus.

We fabricated a structure to hold blue LED light panels using 3D printing technology (SI Appendix, Fig. S5A). To control the timing and intensity of applied blue light, we used an Arduino Mega 2560 Rev3 microcontroller board (manufactured by Elegoo Inc.). As our light source, we used two 8-by-8 square arrays of red, blue and green LED lights (individually addressable, 5V, 5 × 5 mm, WS2812B LED, manufactured by Kuman). The light was powered and controlled by general purpose input/output pins on the Arduino board, as shown in the circuit diagram (SI Appendix, Fig. S5B). We programmed the Arduino board with the Arduino integrated development environment app, using the FastLED library (https://fastled.io/). An example of the code used to control blue light with a frequency of 4 h on and 4 h off is shown in SI Appendix, Table S6.

Mice.

We purchased C57BL/6 (H-2^b) mice from the Jackson Laboratory (Bar Harbor, ME). T-cell receptor (TCR)-I mice express SV40 T antigen epitope I-specific TCRα and TCRβ chains as a transgene (70) and are available from the Jackson Laboratories as line B6.Cg-Tg(Tcra TcrB)416Tev/J. Approximately 95% of T_CD8 express the TCR transgenes. TCR-I mice were crossed onto the B6.PL-Thy1^a/CyJ background to allow detection using anti-CD90.1 antibody staining. We maintained all mice at the animal facility of the Milton S. Hershey Medical Center. We performed animal studies in accordance with the guidelines established by the Pennsylvania State University College of Medicine Institutional Animal Care and Use Committee under an approved protocol.

Isolation of TCR-I T-Cells.

We isolated TCR-I T-cells from the spleens and lymph nodes of TCR-I mice by mechanical disruption as previously described (71), except that red blood cell lysis was not performed. We isolated T-cells from single-cell suspensions by negative selection using the EasySep Mouse T cell Isolation Kit (STEMCELL Technologies, Cat# 19853) according to the manufacturer’s instructions. In brief, we washed cells with EasySep buffer containing PBS with 2% fetal bovine serum and 1 mM ethylenediaminetetraacetic acid and then resuspended cells at 10⁸ cell/mL in EasySep buffer. We added rat serum at 50 μL/mL followed by the addition of 50 μL/mL Isolation Cocktail for 10 min at room temperature. We added RapidSpheres™ at 75 μL/mL and incubated for 2.5 min at room temperature followed by the addition of buffer and magnetic isolation of negatively labeled cells using “The Big Easy” magnet. We stained cells postsort to determine T cell purity by flow cytometry (SI Appendix, Fig. S8) and then froze cells in aliquots using 90% fetal bovine serum/10% dimethylsulfoxide. We stored cells in liquid nitrogen until use. Flow cytometry was performed using a FACSympony A3 analyzer in the Penn State College of Medicine Flow Cytometry Core (RRID:SCR_021134) and data analyzed using FlowJo (V.10.9.0).

Coculture of Engineered Immune Cells with Tumor Spheroids.

To generate MDA-MB-231 or Hela tumor spheroids, we used Nunclon™ Sphera™ ultralow attachment 96-well plate with U-shaped bottom (Thermo Fisher Scientific, Cat# 174925). Specifically, on day 0, we seeded 10⁴ cells in 100 μL complete DMEM for each well, centrifuged cells at 290 g for 3 min, and put the plate back in the incubator. On day 1, we gently added 100 μL of the media containing 6 μg/mL collagen I (Enzo Life Sciences, ALX-522-435-0100) into each well at room temperature and centrifuged the plate at 100 g for 3 min. We cultured the tumor spheroids for another 3 d. On day 4, we changed the media in each well to NK92 media and added 3 × 10⁵ engineered or NK92-WT cells. The effector over target ratio is 10:1. We cocultured the tumor spheroids with NK92 cells for another 5 d.

To make B16.F10-Tag tumor spheroids, we adopted the hanging drop method. On day 0, we seeded 5,000 cells in 20 μL RPMI media containing 0.25% methylcellulose (R&D Systems, HSC001) on the lid of a 100 mm petri dish. We added 10 mL sterile water to the petri dish turned over the lid and put it back. The cell droplet would hang on the lid and the water would prevent the droplet from evaporating. We cultured the B16.F10-Tag cells for 4 d. On day 4, we transferred the B16.F10-Tag tumor spheroids to a normal 96-well plate that was previously covered with 1% agarose at the bottom. Then, we added 1.5 × 10⁵ engineered or wild-type T-cells to each well and cocultured them for 5 d. For all the coculturing experiments, we used the blue LED device (SI Appendix, Fig. S5A) with a wavelength of around 470 nm and a light intensity of 3 μW/cm² (SI Appendix, Fig. S4B).

Statistical Analysis.

We used the Kolmogorov–Smirnov test to calculate the P values of two samples. For multiple populations, we adopted a one-way ANOVA and Tukey’s multiple comparison post hoc test. We used a two-sided 95% CI for all the statistical analyses. We performed the statistical analysis using R version 3.5.2. We generated and assembled all the figures using Jupyter Notebook and Adobe Illustrator.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Live cell imaging of COS-7 cells transfected with LCS7 on collagen coated grids. The live cell was imaged for 18 hours. Blue light was turned on after the first 6 hour. Collagen grids are marked in cyan and only visible under blue light.

Movie S2.

Live cell imaging of MDA-MB-231 stable cells expressing LCS7 on collagen coated rhomboid grids. The live cell was imaged for 18 hours. Blue light was turned on after the first 6 hour. Collagen rhomboid grids are highlighted in cyan and only visible under blue light.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.