All chemicals not specifically mentioned are from Sigma-Aldrich (highest grade). All inhibitors not specifically mentioned are from Cayman Chemical. The following antibodies were used: Alexa Fluor 647 anti-human CD3 antibody (UCHT1, BioLegend), Alexa Fluor 488 anti-human Granzyme A antibody (CB9, BioLegend), Alexa Fluor 647 or Brilliant Violet (BV) 510 anti-human perforin antibody (dG9, BioLegend), Alexa Fluor 647 anti-human granzyme B antibody (GB11, BioLegend), Brilliant Violet (BV) 421 anti-human CD178 (Fas-L), anti α-Tubulin mAb antibody (DM1A, Cell Signaling Technology), and Alexa Fluor 405 conjugated goat anti-mouse IgG (H+L) cross-absorbed secondary antibody (ThermoFisher Scientific). All isotype controls of fluorescence conjugated primary antibodies are from BioLegend. The following reagents were used: Hoechst 33342 (ThermoFisher Scientific), Alexa Fluor 488 or Alexa Fluor 568 phalloidin (ThermoFisher Scientific), Atto 488 NHS ester ((ThermoFisher Scientific), collagenase type I (ThermoFisher Scientific), FibriCol^® type I collagen Solution (Bovine, Advanced Biomatrix).

For pGK-puro-pCasper-pMax (referred to as pCasper-pMax in the manuscript), the vector backbone used for generation of this plasmid is a kind gift from Ulrich Wissenbach (Saarland University) who previously modified the AMAXA vector (Lonza) by replacing the sequence encoding GFP with a linker sequence encoding a multiple cloning site (pMAX). In a first step, the sequence encoding pCasper was amplified from pCasper3-GR (evrogen #FP971) with the following primers introducing an XhoI recognition site at both ends of the amplicon. Forward primer: 5’-CTCGAGGCCACCATGGTGAGCGAG -3’, reverse primer: rev 5’-GACGAGCTGTACCGCTGACTCGAG-3’. The amplicon was subcloned into the XhoI site of pMAX. In a second step, the sequence encoding puromycin resistance was introduced into the intermediate plasmid under the control of 3-phosphoglycerate kinase promoter (PGK-1). The PGK-1-Puromycin sequence was amplified out of pGK-Puro-MO70 vector backbone (Alansary et al., BBA 2015) using the following primers introducing SacI recognition sites at both ends of the amplicon to insert it into the Sac-I site of the intermediate plasmid. Forward primer: 5’-GAGCTCAATTCTACCGGGTAGGGGA-3’, reverse primer 5’-GCAAGCCCGGTGCCTGAGAGCTC-3’. The final plasmid is named pGK-puro-pCasper-pMAX. Final and intermediate plasmids were controlled by endonuclease digestion patterns and sequencing. As a gift from William Bement (Addgene plasmid # 26741). Histone 2B-GFP was a gift from Geoff Wahl (Addgene plasmid # 11680). LifeAct-mRuby was a kind gift from Roland Wedlich-Söldner (University of Muenster). pmEGFP_a_tubulin_C1 was a gift from Daniel Gerlich (Addgene plasmid # 21039). pmCherry-C1 mCherry-NLS was a gift from Dyche Mullins (Addgene plasmid # 58476).

Peripheral blood mononuclear cells (PBMCs) were obtained from healthy donors as described before (30). Briefly, leukocyte reduction chambers were flushed with Hank’s Balanced Salt Solution and loaded on a standard density gradient Leukocyte separation medium (LSM 1077, PAA). PBMCs were isolated by a density gradient centrifugation (450 g, 30 min), and remaining red blood cells were removed by the lysis buffer (155 mM NH₄Cl, 10 mM KHCO₃, 0.1 mM EDTA, pH=7.3). Human primary CD8⁺ T Cells were negatively isolated from PBMCs using Dynabeads™ Untouched™ Human CD8 T Cells Kit (ThermoFisher Scientific) or Human CD8⁺ T Cell Isolation Kit (Miltenyi Biotec), stimulated with Dynabeads™ Human T-Activator CD3/CD28 (ThermoFisher Scientific) with 17 ng/ml of recombinant human IL-2 (ThermoFisher Scientific). MART-1-specific CD8⁺ T-cell clones were generated by Friedmann et al. (31). All CD8⁺ T cells were cultured in AIM V medium (ThermoFisher Scientific) containing 10% fetal calf serum (FCS) and 1% Penicillin-Streptomycin. For nucleofection, CD3/CD28 beads were removed 48 hours after stimulation and 5 × 10⁶ CTLs were electroporated with 2 μg plasmid using 4D-Nucleofector (Lonza). Medium was changed 6 hours after nucleofection and transfected cells were used 24-36 hours after electroporation.

Raji and NALM-6 cells were cultured in RPMI-1640 medium (ThermoFisher Scientific) containing 10% FCS and 1% Penicillin-Streptomycin. NALM-6 pCasper cells were generated by Knörck et al. (32) and were cultured in RPMI-1640 in the presence of puromycin (0.2 µg/ml). SK-Mel-5 cells were transfected with pCasper-pMax using jetOPTIMUS^® DNA Transfection Reagent (Polyplus-transfection) following the manufacturer’s instructions and then cultured in MEM medium (ThermoFisher Scientific) containing 10% FCS and 1% penicillin-streptomycin. All cells were cultured at 37°C with 5% CO₂.

Collagen hydrogels were prepared following previous protocols (33). Briefly, bovine collagen type I stock solution (10 mg/ml) was neutralized with 0.1 N NaOH solution on ice to reach pH 7.0-7.4. 10×PBS was added into the neutralized collagen to a dilution factor of 1:10. The collagen solution was further diluted with PBS to the final concentrations. Cells were resuspended in the collagen solution and the mixture was left for 1 hour at 37°C with 5% CO₂ (if not mentioned otherwise) for fibrillation in 96-well plates. In order to increase the stiffness of the collagen matrix, the collagen stock solution was first diluted to a concentration of 3 mg/ml with 0.1% acetic acid with 100 mM ribose (34). These collagen solutions were kept at 4°C for 5 days, and then used for cell encapsulation as described before.

For killing assays, we used either NALM-6 cells stably expressing apoptosis reporter pCasper-pMax or SK-Mel-5 cells transiently transfected with pCasper-pMax (referred to as NALM-6-pCasper or SK-Mel-5-pCasper, respectively) as target cells. NALM-6-pCasper were pulsed with staphylococcal enterotoxin A (SEA, 0.1 µg/ml) and SEB (0.1 µg/ml) at 37°C with 5% CO₂ for 40 min prior to killing assays. Target cells were resuspended in chilled collagen solution, and transferred in 96-well plates. After centrifugation at 4°C (200 g, 7.5 min), collagen was solidified in the incubator for 1 hour. CTLs were then added from the top if not mentioned otherwise. For the inhibitor-treatment, CTLs were added on top of solidified collagen in medium containing the corresponding inhibitor or vehicle. The effector to target (E:T) ratio for pre-mixed condition is 1:1. For the condition that CTLs were added on top of the collagen, the E:T ratio is 5:1 or 1:1 for CTL : NALM-6-pCasper and MART1 specific CTLs: SK-Mel-5-pCasper, respectively.

Images were acquired by ImageXpress (Molecular Devices) with Spectra X LED illumination (Lumencor) at 37°C with 5% CO₂ for 12 to 24 hours. As described previously (35), fluorescence of pCasper-pMax was acquired using LEDs 470/24 for excitation and the following filter sets (Semrock): Ex 472/30 nm, Em 520/35 nm for GFP and Em 641/75 nm for RFP/FRET. A 20× S Fluor 0.75 numerical aperture objective (Nikon) was used. The killing efficiency was calculated as (1-N_exp(t)/(N_exp(t₀)×N_live(t)/N_live(t₀))×100%.

(N_live: number of live target cells in the control wells without CTLs; N_exp: number of live target cells in the experimental wells; t₀: the first time point of the measurement; t: end time point of the measurement).

CTLs were fixed with pre-chilled 4% paraformaldehyde (PFA) after recovery from collagen degraded with collagenase. Then cells were washed twice with PBS/0.5% BSA, permeabilized and blocked with 0.1% saponin in PBS containing 5% FCS and 0.5% BSA, and then stained with the indicated primary antibody or Alexa Fluor 488 Phalloidin for 30 min at room temperature followed by staining of Alexa Fluor 405 labeled secondary antibody if the primary antibody was not fluorophore-conjugated. Flow cytometry data were acquired using a FACSVerse™ flow cytometer (BD Biosciences) and were analyzed with FlowJo v10 (FLOWJO, LLC).

As described previously (33), collagen with 10 × 10⁶ CTLs/ml polymerized in the capillary at 37°C with 5% CO₂ for 2 hours. Afterwards, the samples were scanned with light-sheet microscopy Z1 (Zeiss) at 37°C for 30 min with an interval of 30 sec and a z-step size of 1 µm. A 20× objective (W Plan-Apochromat, N.A. 1.0) was used. Excitation was realized by two lasers, 488 and 561 nm. Emission was filtered via Em525/40nm and Em 585 LP filters. The images were acquired with ZEN 2014 SP1 Hotfix 2 software. Trajectories of CTLs and nuclear irregularity index (NII) were determined and analyzed with Imaris 8.1.2 (containing Imaris, ImarisTrack, ImarisMeasurementPro, ImarisVantage from Bitplane AG). The nuclei or the cell bodies were detected automatically by ImageJ/Fiji based on the corresponding fluorescence, and parameters (circularity and Feret’s diameter) were analyzed with ImageJ/Fiji. Nucleus deformability is the average of nucleus sphericity change between two neighbor time points.

CTLs (5 × 10⁶ cells/ml) were resuspended in collagen solution with or without Calcein labeled target cells (5 × 10⁶ cells/ml). Cell/collagen mixture (3 µl) was pipetted as a droplet onto the center of an Ibidi μ-dish (Ibidi GmbH). Then a Sigmacote^® (Merck) coated glass coverslip (5 mm, Orsatec GmbH) was carefully placed on top to flatten the droplet (calculated thickness around 150 µm). The Ibidi μ-dish was closed with the lid and incubated at 37°C with 5% CO₂ for 1 hour. After collagen polymerization, the glass coverslip was removed from collagen matrix. For migration assay, CTLs were either non-fluorescent (in presence with target cells) or stained with 1 µg/ml Hoechst 33342 in AIMV (10% FCS) medium for 30 min (without target cells), and then were incubated in fresh AIMV (10% FCS) medium for another 30 min in presence of inhibitors or vehicles as indicated in the figure legends. Raji cells and unlabeled CTLs were mixed. The images were acquired with Zeiss Observer Z.1 for 30 min or 3 hours at 37°C with 5% CO₂ with a Zeiss Colibri LED illumination system and 20× objectives (Fluar 20×/0.75 M27 Air). Images were taken using an AxioCamM1 CCD camera and AxioVision 4.1.8. The images of cell migration and nuclear irregularity index (NII, measured as nuclear circularity) were analyzed using Imaris 8.1.2. Migration trajectories in presence of target cells were analyzed by Fiji. Nuclear deformationability is defined as standard deviation of NII. To quantify CTL search efficiency, CTLs were randomly selected from the CTLs that were observed for the whole period. The probability for CTLs finding at least one target within 3 hours was quantified. From these CTLs, the time of CTL contacting the first target cells was quantified, if this CTL could find at least one target cell within 3 hours. Nucleus deformability is the average of nucleus circularity change between two neighbor time points.

For the fixed sample, CTLs were fixed at indicated time points with 4% PFA and permeabilized with 0.3% Triton-100 with 5% FCS in PBS, followed by staining with indicated antibodies or fluorescent dyes according to the manufacturers’ instructions. Images were acquired by confocal microscopy LSM 710 with a 63× objective (N.A. 1.4) and a Nikon E600 camera using ZEN software. The nuclear irregularity index (NII, measured as nuclear circularity from maximum intensity projection) was analyzed with ImageJ. The distance between nucleus and MTOC was analyzed with Imaris 9.6.

After collagen polymerization in the capillary, collagen was stained with Atto 488 NHS ester in PBS (50 µM) at room temperature for 15 min. Afterwards, the collagen matrix was washed by PBS twice. Matrix structure of collagen was visualized by light-sheet microscopy with a 20× objective (W Plan-Apochromat, N.A. 1.0). Collagen pore size was measured in the middle slice of the z-stack by Fiji (BIOP version) with Max Inscribed Circles plugin as described elsewhere (36).

Rheology measurements with different concentrations of bovine collagen were performed using DHRIII Rheometer (TA Instruments). 50 µl of the neutralized collagen solution (pH 7.0-7.4) was placed between two parallel plates of 12 mm diameter pre-heated to either 25°C or 37°C. The shear moduli were measured at frequency ω – 3 rad/s at 37°C as described previously (37). All experiments were performed in triplicates.

CTLs were embedded in 5 mg/ml collagen in a 96-well plate. After collagen polymerization, 100 µl AIMV medium with 10% FCS, 1 µg/ml propidium iodide, and nocodazole or DMSO was added. Images were acquired by ImageXpress with Spectra X LED illumination (Lumencor) using LEDs 542/27 for excitation. The filter set was Ex 542/27 nm and Em 641/75 nm. A 20× S Fluor 0.75 numerical aperture objective (Nikon) was used. The images were acquired at 37°C with 5% CO₂ every 1 h for 12 h.

Research carried out for this study with healthy donor material (leukocyte reduction system chambers from human blood donors) is authorized by the local ethic committee [declaration from 16.4.2015 (84/15; Prof. Dr. Rettig-Stürmer)].

Data are presented as mean ± SD. GraphPad Prism 6 Software (San Diego, CA, USA) was used for statistical analysis. If the number of data points is smaller than 8, the differences between two columns were analyzed by the Student’s t-test. Otherwise, the data were first examined for Gaussian distribution. If the dataset fit Gaussian distribution, the differences between two columns ware analyzed with the Student’s t-test, otherwise with the Mann-Whitney test.

To investigate the impact of collagen density on the killing efficiency of CTLs, we used collagen matrices prepared at three different collagen concentrations (2, 4, and 5 mg/ml). Primary human CD8⁺ T cells were stimulated with anti-CD3/anti-CD28 antibody-coated beads to obtain CTLs. Target cells (NALM-6-pCasper) were embedded in the collagen matrices and CTLs were settled on top of the gels, as depicted in Figure 1A (schematic diagram). Target cells stably expressing apoptosis reporter pCasper-pMax, a GFP-RFP FRET pair linked by a sequence containing caspase recognition site (DEVD), allowing the detection of cell death (32). Apoptotic target cells switches the fluorescence to green as the linker between the GFP-RFP FRET pair is cleaved and necrotic target cells show a complete loss of fluorescence (35) ( Supplementary Figure 1A ). Using a high-content imaging setup (ImageXpress), we observed that 85.4 ± 10.9% of target cells in 2 mg/ml collagen were killed by either apoptosis or necrosis after 24 hours. The fraction of apoptotic and necrotic target cells eliminated by CTLs dropped to 42.1 ± 8.8% and 7.3 ± 6.3% in collagen matrices in 4 and 5 mg/ml collagen matrices, respectively. ( Figures 1A, B ). The same trend was observed at earlier time points (12 hours, Supplementary Figure 1B ).

To exclude a possible influence of the CTL infiltration step from the gel surface into the gel interior, similar experiments were performed by embedding the CTLs with the target cells in the collagen gel. The cytotoxic efficiency of CTLs was similar to the experiment that included infiltration ( Figures 1A–D ). These results indicate that CTL killing efficiency was solely diminished by the dense collagen network.

To unravel the underlying mechanisms for a reduced CTL killing efficiency in dense collagen matrices, we first examined lytic granule and FAS/FASL pathways. To test the expression of cytotoxic proteins, we recovered CTLs from collagen matrices using collagenase. Control experiments showed that the collagenase treatment does not alter the protein level on T cells, e.g. CD3 expression ( Supplementary Figure 2A ). No significant differences in the expression of cytotoxic proteins (perforin, granzyme A and granzyme B) were observed in CTLs extracted from the collagen hydrogels of the three different concentrations ( Supplementary Figure 2B ). FasL expression was also examined, which was at very low levels for all three concentrations ( Supplementary Figure 2C ). Interestingly, in dense collagen, the fraction of CTLs conjugated with targets ( Figure 1E ) and the fraction of target cells conjugated with CTLs ( Figure 1F ) were significantly reduced. These results indicate that the impaired CTL killing efficiency in dense collagen is not owed to changes in the main components of the killing machinery but rather to a reduced search efficiency.

We next investigated the search efficiency of CTLs in collagen in detail using live-cell imaging. Figure 1G shows that a migrating CTL (highlighted by the blue track) in a 2 mg/ml collagen needs around 45 min to find the first target, whereas CTLs in 4 and 5 mg/ml ECM need about 100 min (lower panels of Figure 1G ). In 2 mg/ml collagen matrices, most CTLs (92.0 ± 4.0%) found at least one target cell within 3 hours; whereas this likelihood was reduced to 66.8 ± 2.1% or 54.4 ± 8.2% in 4 and 5 mg/ml matrices ( Figure 1H ). Analyzing only the CTLs that found target cells, in 4 and 5 mg/ml collagen matrices, these needed about 90 min to find the first target but only about 60 min in 2 mg/ml collagen ( Figure 1I ). These results suggest that the probability of CTLs to find their targets is decreased in dense collagen matrices.

The mobility of CTLs in the matrix is a key factor for optimal search efficiency. To quantify motility, we analyzed displacement, migration velocity, and persistence (indicating how directed the migration is) of CTLs inside the collagen matrices using light-sheet microscopy. In dense collagen matrices (4 and 5 mg/ml), the displacement of CTLs (distance between the starting point and the end point) was reduced ( Figure 1J , Supplementary Figure 3, Supplementary Movies 1–3 ). Analysis of trajectories in 3D collagen shows that velocity and persistence of migrating CTLs decreased with increasing collagen concentration ( Figures 1K, L , Supplementary Figure 3 ). It was also observed that CTL migration was impaired in the high concentration of collagen when target cells are present ( Supplementary Figures 4A–C ). Analysis of velocity and persistence distributions reveals a higher fraction of CTLs with low velocity and low persistence in dense collagen (5 mg/ml, Figures 1M, N ). Moreover, migration velocity is negatively correlated with the time of CTL searching target cells ( Supplementary Figure 4D ). In summary, our findings suggest that in a dense matrix, CTL migration is hindered, and this is likely the reason for longer search time and reduced cytotoxic efficiency.

We examined the correlation between CTL killing efficiency and the physical properties of collagen matrix, including pore size and the stiffness. We fluorescently labeled collagen to visualize its structure ( Figure 2A ) to determine pore size ( Supplementary Figure 5 ) and found the pore size was decreased with increasing density ( Figures 2A, B ). Collagen stiffness was determined with the storage modulus G’ measured by rheology, which increased from 0.81 kPa to 3.64 kPa with collagen density between 2 and 5 mg/ml ( Figure 2C ). To study which of these two features contributes to the impaired CTL migration in collagen matrix, we modified the collagen matrix. To widen the range of stiffness keeping the pore size constant, we also prepared collagen matrices with 100 mM ribose (34), which showed increased storage modulus up to 1.24 kPa ( Figure 2D ), without affecting the pore size ( Figure 2E ). We observed no changes in CTL velocity with increasing stiffness ( Figure 2F ), while the persistence was reduced ( Figure 2G ). In line with this result, the cytotoxic efficiency of CTLs was reduced in collagen matrices prepared in the presence of ribose ( Figure 2H ). These results indicate that stiffness is involved in regulating CTL migration persistence, which could influence CTL killing efficiency.

To increase the pore size of the matrix maintaining a constant collagen concentration, the fibrillation step was performed at room temperature [instead of 37°C (38)]. Noticeably, collagen matrices obtained at 2 mg/ml collagen concentration showed similar shear moduli, while 4 mg/mL matrices with larger pores showed an increased stiffness ( Figure 2I ). This could be attributed to slower fibrillation kinetics of collagen at RT (39, 40). Nevertheless, in matrices with larger pore sizes ( Figures 2J, K ) the cytotoxic efficiency of CTLs was enhanced ( Figure 2L ). Together, these results suggest that smaller pore size and higher stiffness in dense collagen matrices lead to impaired CTL killing efficiency as a result of hindered migration, whereby CTL migration persistence is mainly determined by matrix stiffness while the velocity is likely determined by the pore size of the matrix.

As the stiffest organelle in cells, the nucleus is essential for decision making of migration direction for immune cells migrating through restricted space (24, 26, 27). Therefore, we examined the morphology of the nucleus in impaired CTL migration in dense matrices. First, we noticed that the nuclear morphology was deformed to an hour-glass shape in migrating CTLs ( Figure 3A , Supplementary Movies 4–6 ). To quantify the extent of this nuclear deformation, we analyzed the diameter of cross-sections (the shortest intersection of the hour-glass shape), as well as the sphericity (how closely an object resembles a sphere). The cross-section diameter was decreased in matrices with higher collagen density ( Figures 3B, C ). Furthermore, the nuclei of migrating CTLs in dense collagen matrices were more frequently deformed to an hour-glass shape than their counterparts in low-density collagen ( Figure 3D ). Nuclear irregularity index decreased in dense collagen (4 and 5 mg/ml) ( Figure 3E ). It is reported that in dendritic cells, the nucleus is drastically deformed when migrating through spatially restricted areas (25). Together, our results suggest that the extent of nuclear deformation is increased in dense collagen likely due to its decreased porosity.

We next examined whether nuclear deformation correlated with CTL migration in 3D. As shown in the exemplary cells, at the time points when the nucleus was deformed, the real-time velocity was high. This phenomenon was observed for all three concentrations ( Figure 3F , Supplementary Movies 4–6 ). The respective analyses show that the migration velocity of CTLs is positively correlated to nuclear deformability as determined by the change in nuclear irregularity index (sphericity) in each density ( Figure 3G ). This observation is in good agreement with a recent report, showing that cell migration velocity positively correlated to the change in nucleus shape in 2D (28). Together, these findings suggest that deformability of the nucleus is a key factor to determine CTL migration in 3D.

Cytoskeleton is an essential regulator for nuclear deformation induced by mechanical forces (41, 42) as well as for cell migration (43). Therefore, we examined the contribution of key cytoskeletal components to the regulation of nuclear deformation and CTL migration in 3D. The expression of F-actin ( Figure 4A ) and microtubules ( Figure 4B ) in collagen-embedded CTLs was modestly upregulated in the dense matrices (4 mg/ml and 5 mg/ml). We analyzed the intracellular distribution of F-actin and microtubules using immuno-staining. Confocal images and live-cell imaging show that F-actin was mainly located in the CTL cortex, whereas the microtubule network was enriched around the microtubule-organizing center (MTOC) at the uropod and nucleus-surrounding areas ( Figure 4C , Supplementary Movie 7 ) as reported by the others (44). Noticeably, the average distance of microtubule network and the nucleus was decreased upon the increase in collagen density ( Figures 4D, E , Supplementary Figures 6A, B ). Moreover, short distance between microtubule network and the nucleus benefits to keep the nucleus volume and sphericity in dense collagen ( Supplementary Figures 6C–G ). As it has been reported that the volume of the nucleus is limited by the microtubule network (42), we hypothesized that the microtubule network is involved in regulating CTL nuclear deformability.

To investigate this hypothesis, we used nocodazole, an inhibitor of tubulin polymerization, to abrogate the functionality of the microtubule network. We found that nuclear circularity was significantly decreased by nocodazole treatment as illustrated in the exemplary cells ( Figure 4F ) and the quantitative analysis ( Figure 4G ), indicating that with microtubules depolymerized, the nucleus is more deformable.

Combined with the finding that nuclear deformation is correlated with CTL migration, we postulated that disruption of the microtubule network should impact CTL migration, especially in dense ECM. Analysis of migration of CTLs in collagen matrices shows that in CTLs treated with nocodazole migration velocity was enhanced at all matrix densities ( Figure 4H ); whereas persistence was only increased in collagen with 5 mg/ml ( Figure 4I ). In comparison, disruption of the actin network by latrunculin-A or abrogation of myosin IIA by blebbstatin almost abolished CTL velocity and persistency for all three collagen densities ( Figures 4H , 5I ). Inhibition of a myosin IIA-upstream kinase Rock, or focal adhesion kinase also drastically impaired CTL velocity ( Supplementary Figure 7A ) but did not drastically change 3D migration persistence ( Supplementary Figure 7B ). Furthermore, live-cell imaging shows that the nucleus in the nocodazole-treated CTLs was more deformable than their DMSO-treated counterparts ( Figure 4J ). For CTLs, despite nocodazole-, vehicle-, or other inhibitors-treated, the migration velocity was positively correlated with the extent of nuclear deformation ( Figure 4K , Supplementary Figure 8 ). In summary, we conclude that disruption of the microtubule network enhances CTL migration especially in dense collagen matrices, which is correlated with enhanced deformability of nucleus in CTLs.

Considering the dependence of CTL migration velocity and persistence on the microtubule-network, we hypothesized that interference with microtubules should improve the impaired killing efficiency in dense ECM. Results of 3D killing assay show that indeed in dense matrix (4 mg/ml and 5 mg/ml), disruption of the microtubule network ameliorated CTL killing ( Figures 5A–C ). Moreover, to further confirm this phenomenon, we used primary human CTL clones specific for MART1 (also known as Melan-A) (31). A melanoma cell line SK-Mel-5, which endogenously, present MART1 on their surface (45), was used as target cells. The time lapse shows that in good agreement with previous results, nocodazole-treatment significantly enhanced killing efficiency of CTL clones to remove SK-Mel-5 ( Figures 5D, E ). Together, our findings suggest that the microtubule network is as a promising target to improve CTL killing against tumor cells in dense ECM.

We further confirmed that also in presence of target cells, disruption of the microtubule-network with nocodazole treatment (10 μM) promoted 3D CTL migration ( Figure 5F ), but not reduced cytotoxic protein secretion and expression ( Supplementary Figure 9 ). Concomitantly, the likelihood for microtubule-disrupted CTLs showed increased velocity ( Figure 5G ) and had a higher probability to find their target cells ( Figure 5H ). Only analyzing CTL which were successful to find at least one target, nocodazole-treated CTLs needed similar times to locate their first targets compared to DMSO-treated CTLs ( Figure 5I ), there was only a slight but insignificant reduction of search time for the first target. However, on average, the number of target cells found by nocodazole-treated CTLs was significantly higher than by control CTLs ( Figure 5J ). In addition, we examined CTL viability after 12 hour-nocodazole treatment, which is the condition we used to examine killing efficiency. We found that under this condition, CTL viability in 5 mg/ml collagen was slightly reduced ( Supplementary Figure 10 ). Together, we conclude that enhancement of CTL killing efficiency in dense collagen by nocodazole-treatment is due to the amelioration of CTL migration and infiltration, not due to improvement of CTL survival.

Finally, we tested vinblastine, a microtubule-inhibitor applied as a chemotherapeutic to disrupt tumor cell mitosis (46). We treated CTLs with vinblastine to determine CTL killing efficiency in dense collagen. We found that vinblastine increases both killing efficiency of primary CTLs ( Figure 5K ) and human MART1-specific CTL clones ( Figure 5L ). In summary, we conclude that disruption of the microtubule network significantly enhances CTL migration and killing efficiency in dense collagen.