Dynamic Profiling of Antitumor Activity of CAR T Cells Using Micropatterned Tumor Arrays

Abstract Cancer immunotherapy based on the engineering of chimeric antigen receptors (CAR) on T cells has emerged as one of the most promising new therapies for patients with B‐cell malignancies. Preclinical assessments of essential CAR T cell functions such as trafficking and cytotoxicity are critical for accelerating the development of highly effective therapeutic candidates. However, current tools for evaluating CAR‐T functions lack sufficient precision. Here, a micropatterned tumor array (MiTA) is described that enables detailed and dynamic characterization of CAR T cell trafficking toward tumor‐cell islands and subsequent killing of tumor cells. It is shown that CAR T cells often merge into large clusters that envelop and kill the tumor cells with high efficiency. Significant differences are also measured between CAR T cells from different donors and between various CAR T cell constructs. Overall, the assay allows for multifaceted, dynamic, high‐content evaluation of CAR T trafficking, clustering, and killing and could eventually become a useful tool for immune‐oncology research and preclinical assessments of cell‐based immunotherapies.


Introduction
Chimeric antigen receptors (CARs) are engineered receptors used to reprogram patient's T cells to specifically target tumor cells. Cancer immunotherapy based on CAR T cells has emerged as one of the most promising new therapies for the treatment of patients with B-cell malignancies. [1][2][3][4][5][6][7][8] The antitumor activity of CAR T cells relies on efficient CAR T cell trafficking to cancer niches, recognition of tumor antigen, and potent cytotoxicity toward tumor cells. These biological processes are dynamic and involve collective interactions of CAR T and tumor cells. Comprehensive preclinical assessments of these key processes are of pivotal importance for ensuring CAR T therapeutic efficacy.
Several in vitro assays can measure CAR T cell cytotoxicity toward tumor cells. Chromium (Cr-51) release assay is the gold standard for quantifying cytotoxicity in the study of tumor cytolysis. [9] However, this assay only provides simple end-point readouts and does not distinguish the CAR T-mediated killing of target cells from other causes of target cell death. Also, the assay is cumbersome to implement and poses safety challenges because it involves the use of radioactive materials. Assays based on the quantification of cytosolic enzymes such as lactose dehydrogenase (LDH) [10] or glyceraldehyde phosphate dehydrogenase (GAPDH) [11] circumvent the need for radioactive materials. However, these assays fail to distinguish the death of target cells from effector cells, since both release cellular enzymes upon lysis. This problem is avoided when firefly luciferase (Fluc) -expressing cells are employed as targets of effector T cells and the release of Fluc into the medium is a specific measure of target cell lysis. However, despite being widely used, these biochemical assays are limited to quantifying only the bulk responses at a single time point which can hardly elucidate the complex and dynamic antitumor activity of CAR T cells.
The integration of real-time monitoring techniques such as time-lapse imaging [12] and electrical impedance sensors [13] with cell culture plates have enabled the dynamic characterization of the cytolysis process. However, the random arrangement of cells in traditional cell culture dishes prohibits the study of CAR T cell trafficking. Lab-on-a-chip technology such as Cancer immunotherapy based on the engineering of chimeric antigen receptors (CAR) on T cells has emerged as one of the most promising new therapies for patients with B-cell malignancies. Preclinical assessments of essential CAR T cell functions such as trafficking and cytotoxicity are critical for accelerating the development of highly effective therapeutic candidates.

However, current tools for evaluating CAR-T functions lack sufficient precision. Here, a micropatterned tumor array (MiTA) is described that enables detailed and dynamic characterization of CAR T cell trafficking toward tumorcell islands and subsequent killing of tumor cells. It is shown that CAR T cells often merge into large clusters that envelop and kill the tumor cells with high efficiency. Significant differences are also measured between CAR T cells from different donors and between various CAR T cell constructs. Overall, the assay allows for multifaceted, dynamic, high-content evaluation of CAR T trafficking, clustering, and killing and could eventually become a useful tool for immune-oncology research and preclinical assessments of cell-based immunotherapies.
microfluidic cell culture and organ-on-chips hold great promise for advancing the therapeutic screening of cancer immunotherapies. [14][15][16][17][18] However, they are labor-intensive and difficult to use. [19] Reproducing sophisticated in vitro microenvironment usually takes days [18][19][20] and the complexity of the microfluidic systems adversely affect the robustness of measurements. [19] To overcome the challenges of aforementioned approaches, we designed an assay for the dynamic profiling of antitumor activity of CAR T cells. We employed micropatterning to precisely pattern multiple myeloma tumor cells into arrays of microscale islands. [21] The tumor-cell islands have uniform size and shape and contain a similar number of tumor cells, facilitating the reproducibility of screening. Spatially segregating the tumor cells into microscale islands toward which CAR T cells have to actively migrate allows the systematic study of trafficking and subsequent tumor killing. The arrays of tumorcell islands are housed in customized microfluidic chambers which eliminates the drifting of CAR T cells caused by convection-induced flow, ensuring robust cell interactions. We found that CAR T cells robustly migrate toward the tumor-cell islands, increasing the local effector-cell density and aggregating into large clusters that envelop the tumor cells and exert cytolytic effects on the tumor cells. The assay detects and quantifies differences in trafficking, clustering, and cytotoxicity of CAR T cells from different donors. Using the assay, we conducted multifaceted characterizations of anti-BCMA and APRIL-based CAR T cell constructs against BCMA+ and BCMA knockout (KO) multiple myeloma MM.1s tumor cells. We demonstrate that APRIL-based CAR T cells efficiently clustered around and eliminated both tumor-cell types, suggesting that this CAR T cell construct could reduce the incidence of antigen-negative escape. With the ease of use, the high throughput and reproducibility and the ability to dynamically map the antitumor activity of CAR T cells, the micropatterned tumor array (MiTA) could become a useful tool for studying cancer immunology and aiding the preclinical evaluations of cell-based cancer immunotherapy.

Micropatterned Tumor Arrays
We patterned a large array of microscale tumor-cell islands that are housed in a microfluidic compartment. First, an array of 1024 spots of an adhesion-promoting material that consists of a mixture of poly-L-lysine and ZETAG were patterned on 1 × 3" glass substrates using an automated liquid dispenser (Figure 1a, panel i). The patterned adhesive spots have a uniform diameter of 185.5 ± 5.2 µm and an average roundness index of 0.99 ± 0.01. Then, we assembled on top of the patterned glass substrate, a polydimethylsiloxane (PDMS) membrane containing 16 microfluidic chambers and a plastic frame defining 16 wells (Figure 1a, panel i). After assembly, the substrate is divided into 16 individual compartments. Each compartment contains 64 spots spaced 500 µm apart, in an 8 × 8 array (Figure 1a, zoom-in in panel i).
To form the tumor-cell islands, we load a suspension of tumor cells inside the microfluidic chambers and allow them to sediment and adhere on the spots. We then remove the nonadhered cells by gentle wash (Figure 1a, panel ii). Patterning Roswell Park Memorial Institute growth medium (RPMI) 8226 tumor cells yields spot arrays with an average 79 ± 7 cells per spot and an average 2.2 ± 0.2 × 10 4 µm 2 area (Figure 1d,e). The variation of the cell number and area between spots stems from the heterogeneity of the cell size. Finally, we load CAR T cells inside the microfluidic compartments and allow them to sediment. Immediately, we start monitoring the interactions between CAR-T and tumor cells using time-lapse imaging (Figure 1a, panel iii). With the assay, we are capable of quantifying the dynamic interactions between CAR-T and tumor cells (Figure 1f) on 4096 spots on four slides, in 64 different conditions in each experiment ( Figure S1, Supporting Information).

Endpoint Evaluation of Overall CAR-T Antitumor Efficacy Using MiTA
We designed a second-generation anti-BCMA chimeric antigen receptor consisting of a single chain variable fragment (scFv) connected with a CD8 hinge/transmembrane domain to 4-1BB and CD3ζ intracellular domains (Figure 2a-d). In order to facilitate the evaluation of transduction efficiency with the lentiviral construct, we incorporated the mCherry fluorescent reporter gene after a T2A element at the C-terminal of the CAR sequence (Figure 2a We also compared CAR T cells from four different healthy donors and found that substantial differences in antitumor activity, ranging from 17.1% (donor 3) to 39.3% (donor 2) (Figure 2h), despite a similar ≈50% transduction efficiency. Taken together, the end-point snapshots of MiTA unveiled that CAR T cells eliminated the tumor cells more efficiently and consistently than the unspecific killing by UTD T cells and the antitumor efficacy varied among donors.

Dynamic Profiling of the CAR T Antitumor Activity Using MiTA
The micropatterned tumor array enables us to visualize and quantify the dynamic process of tumor elimination by CAR T cells (Figure 3 and Video S1, Supporting Information). We distinguished two phases during the interaction between CAR T and tumor cells: an initial phase of CAR T cell accumulation at the tumor islands followed by a phase of rapid tumor killing. During the initial phase, which lasts ≈2 h, CAR T cells migrate toward the tumor islands (Figure 3a, CAR). Most tumor cells stay alive (Figure 3b-i, CAR, blue region) and the initial morphology of the islands is retained (Figure 3b

CAR T Trafficking toward Tumor Spots
We quantified the trafficking of CAR T cells at various cell densities (Figure 4). We found that a rapid, initial trafficking phase is usually followed by a slow, plateau phase. The duration of the first phase depends on the E:T ratio. At E:T = 10 (high CAR T density), CAR T cells continuously migrated to the spots in the first 4 h with the CAR T cell area increased by threefold (Figure 4c). After 4 h, the trafficking plateaus and the accumulation rate decreases to ≈0 µm 2 h −1 (Figure 4c right panel, black arrow). At lower E:T ratios, the trafficking is slower and the plateaus are delayed (Figure 4c right panel, 4, 6, and 12 h at E:T = 10, 5, and 2.5). The fold change in the CAR T area is larger at smaller E:T ratios. At 18 h, the area is increased by 4.0 ×, 3.2 ×, and 2.4 × at E:T = 2.5, 5, and 10, respectively (Figure 4d), confirming the robustness of trafficking.
We mapped the dynamic correlation between CAR T cell trafficking and the killing of tumor cells (Figure 4g). We found that at higher cell density (E:T = 5 and 10), CAR T cells exert efficient killing after the trafficking has plateaued. For example, in the first 2 h, while the CAR T area increases 3 ×, the area of tumor cells does not change (Figure 4g, E:T = 10, circles). From 2 to 8 h, the CAR T trafficking reaches the plateau and the area of tumor cells shrinks rapidly to 60% and 30% at 8 and 18 h, respectively.

CAR T Clusters Enhance Tumor-Cell Killing
We observed that CAR T cells often form large clusters on top of target islands when killing the tumor cells (Figure 4a,b). During tumor-cell killing, CAR T cells first formed multiple small clusters on top of the tumor island and then merge into one large cell cluster that engulfs the tumor cells within it (Figure 4b). In the first 6 h, the average cluster area increases from ≈1600 to 18 000 µm 2 while the average number of clusters decreasing from 2 to 1, indicating the multiple small clusters merge into one large cluster (Figure 4e (Figure 4g, triangle). The killing is inefficient, and the CAR T clusters do not grow larger than the tumor-cell area even at 18 h ( Figure S2, Supporting Information).

Heterogeneity in Antitumor Activity of CAR T-Cells from Different Donors
We found distinct dynamic profiles of tumor killing by CAR T cells originating from different healthy donors (Figure 3f-h). Tumor cells were killed equally efficient by CAR T cells from donor 1 and 3. However, tumor cells were killed less efficiently by CAR T cells donor 2 and the shrinkage of tumor islands was slower and delayed (Figure 3f-g, donor 2). Interestingly, tumor cells were killed faster during the first 3 h by the CAR T cells from donor 2, with the highest killing rate of ≈16% h −1 at 3 h (Figure 3h, donor 2). However, the killing slowed down after 3 h, with the rate sharply decreasing to ≈4% at 5 h and then ≈2% at 8 h. As a result, only 60% tumor cells were eliminated at 18 h (Figure 3g). The rate of area shrinkage remains < 4% throughout 18 h distinct from the other two donors (Figure 3h, light blue dashed lines).
We found that the CAR T cells from different donors also displayed different trafficking and clustering profiles (Figure 4h-j). CAR T cells from donor 1 exhibited the strongest trafficking with 2.5 × increase in the area after 18 h, much higher than donor 2 and 3 (Figure 4i). In addition, they showed the best ability to cluster around the tumor cells, with ≈90% cells merging in to a single cluster at 18 h (Figure 4k, blue line). The trafficking of CAR T cells from donor 2 was the fastest but plateaued the earliest among the three donors (Figure 4h, green line). Uniquely, the CAR T area kept decreasing after 3 h, leading to the smallest area at 18 h (Figure 4h). CAR T cells from donor 2 clustered the fastest and formed a single cluster at 4 h, earlier than the other two donors (Figure 4j, green lines and green arrow). CAR T cells from donor 3 exhibited a similar trafficking profile as donor 1 (Figure 4h,j, purple and blue lines). However, their ability to form clusters is weaker than the other two donors with only 60% of cells clustering together at 18 h (Figure 4k, purple). Mapping CAR T trafficking and tumor-cell killing together, we found that the killing efficiency may relate to the area of CAR T cells on the spot (Figure 4l). Despite fast trafficking at the first 2 h (Figure 4l, red dots), the area of CAR T cells from donor 2 is smaller after 4 h than the other two donors. Correspondingly, the killing is less efficient, shown as a higher % of remaining tumor area at 4, 8, and 18 h (Figure 4l green, blue, and yellow triangles vs circles and squares).

Comparing the Antitumor Activity of Two CAR T-Cell Constructs Using MiTA
We employed MiTA to compare the antitumor activity of anti-BCMA and APRIL-based CAR T cells toward BCMA positive and negative multiple myeloma MM.1s tumor cells. "A proliferation-inducing ligand" (APRIL) is a soluble ligand that can bind BCMA and the transmembrane activator and calciummodulator and cyclophilin ligand (TACI), two antigens highly expressed on MM cells. We generated an APRIL-based CAR consisting of a truncated APRIL fused to a spacer domain and to the same endodomain used for anti-BCMA CAR construct (anti-BCAR). Our hypothesis is that dual antigen targeting will enhance the tumor-cell killing, reduce the incidence of antigen negative escape, and overall therapeutic potential. [22] Our assay shows that APRIL-based T cells can efficiently eliminate both BCMA positive and negative MM.1s tumor cells (Figure 5a,b-i,c-e). At 18 h, < 30% tumor cells survived the killing and area of the tumor decreased to ≈35% (Figure 5c). Anti-BCAR T cells eliminate BCMA positive tumor cells more efficiently than APRIL-based CAR T cells (19% surviving tumor cells and 19% remaining area at 18 h), however exhibited significant deficiency in killing BCMA negative tumor cells (Figure 5a,b-ii,c-e). At 18 h, 57% tumor cells had survived the killing and the tumor area had only decreased to 72% (Figure 5c).
We assessed the dynamic profiles of tumor-cell killing for both CAR T constructs (Figure 5d,e). APRIL-based CAR T cells exhibited a similar profile for killing BCMA positive and negative MM.1s tumor cells (Figure 5d,e, dark and light dotted curves) with an interaction phase from 0 to 3 h and rapid elimination phase after 3 h. Anti-BCAR T cells killed BCMA positive tumor cells more efficiently than APRIL-based CAR T cells, which caused immediate and faster shrinkage of tumor area (Figure 5e). However, they exhibited deficient killing of BCMA negative tumor cells. The killing started immediately in the absence of the initial interaction stage seen in other conditions (Figure 5d). The killing rate was slower and the shrinkage of tumor occurred after 9 h, ≈5 h later than other conditions (Figure 5e).
We observed different trafficking dynamics for CAR T cells with different constructs (Figure 5f). The trafficking of APRIL-based CAR T cells toward BCMA positive tumor spots is the fastest, leading to the largest CAR T area at 9 h among the four conditions (Figure 5f, dark red curve). After 9 h, the trafficking plateaus. The trafficking of APRIL-based CAR T cells toward BCMA negative tumor cells and anti-BCAR T cells toward BCMA positive tumor cells is slower and plateaus later than 9 h (Figure 5f, light red and dark green curves). Finally, the trafficking of anti-BCAR T cells toward BCMA negative tumor cells was the slowest, progressing in a linear fashion from 3 to 18 h (Figure 5f, light green curve).
Although a similar number of CAR T cells arrive at the spots at 18 h (Figure 5f), the clustering of CAR T cells is different for the four conditions (Figure 5g-i). When CAR T cells form a single large cluster on the tumor spots, killing is efficient (Figure 5g-I,h,i). When CAR T cells aggregate into multiple, small clusters, killing is inefficient (Anti-BCAR vs BCMA negative tumor- Figure 5g-II,h,i). The largest clusters on the spots in Anti-BCAR versus BCMA negative tumors are significantly smaller than all other conditions (Figure 5h). During efficient killing, 37-63% of the spots have one single large CAR T cluster. In contrast, only 13% of spots have one single cluster in the anti-BCAR CAR T versus BCMA negative tumor condition. Taken together, our results suggest that APRIL-based CAR can form single, large clusters on both BCMA positive and negative tumor spots. However, anti-BCMA cells failed to form clusters on BCMA negative tumors.
We mapped the correlation between the trafficking of CAR T cells and the corresponding killing which revealed distinct dynamic profiles during efficient and inefficient killing (Figure 5j-l). The results demonstrate distinct dynamics among the four conditions and highlighted the deficient killing of anti-BCAR versus BCMA negative (Figure 5j, highlighted with a red line) and efficient killing in the other conditions. The clustering of CAR T cells also correlates with the killing efficiency ( Figure 5k). Tumor cells were eliminated efficiently when CAR T cells formed large clusters (Figure 5k, efficient). In deficient killing situations, CAR T cells only form smaller clusters with a significantly smaller ratio of CAR cluster area to tumor area (Figure 5l).
In summary, MiTA enables high-content analysis and multifaceted comparison of the antitumor activity between different CAR T constructs. Our data show that APRIL-based CAR T cells can effectively migrate to tumor spots, form clusters, and eliminate both BCMA positive and negative tumor cells, while anti-BCAR T cells failed to do so toward BCMA negative tumor cells. This result suggests that APRIL-based CAR T cells could reduce the incidence of antigen-negative escape and thus have stronger therapeutic potential.

Discussion
We developed a micropatterned tumor array (MiTA) that enables high-content and dynamic profiling of the collective antitumor activity of CAR T cells against multiple myeloma tumor cells. [23] Spatially patterning tumor cells into islands induces strong CAR-T trafficking toward tumor targets of similar size and area and allows for simultaneous characterizations of the recruitment of effector cells and elimination of target cells. The microfluidic compartments minimize the mechanical perturbation acting on loosely adherent T cells and prevents artificial cell interactions induced by cell drifting. The integration of the microfluidic compartments in a multiwell plate format facilitates multiplexed and high-throughput screening of CAR T cells which could expedite the testing of different tumor-cell lines, CAR T constructs, and drug candidates.
Compared to widely used biochemical assays that only provide end-point results, MiTA provides comprehensive information regarding CAR T cell trafficking and subsequent tumor killing. Compared to conventional cell-based assays that probe the interactions of effector and target cells that are randomly distributed on a surface, MiTA enables monitoring of collective interactions of CAR T cells with spatially patterned tumorcell group, which revealed the potential impacts of CAR T cell recruitment and clustering on tumor cluster elimination. Compared to microfluidic and organ-on-a-chip models, MiTA is more straightforward to setup and enables simultaneous characterizations of antitumor activities on a large number of structurally similar tumor islands, which may promote the robustness of screening.
The micropatterned tumor array exhibits high-content information on the dynamic interaction between CAR T cells and tumor islands. The dimensionality of information can be expanded further to decipher this process with greater details. In addition to the area and number of cells and clusters, one could characterize shape factors such as aspect ratio, circularity, etc. as well as the correlation between the tumor and CAR T cluster shapes. Ultimately, multiple-dimensional data may be introduced into a machine learning algorithm for better stratifying the efficiency of CAR T cells against tumor cells.
Studying the interaction between different CAR T cells and tumor-cell types revealed a signature profile for the antigenspecific killing of tumor cells. Efficient killing driven by antigen-specific binding is characterized by an initial, slow accumulation phase and a subsequent rapid killing phase. In the initial phase, CAR T cells migrated from surrounding to the tumor-cell island but exerted a limited cytolytic effect on the tumor cells. Later, the CAR T cells on the island merged into large clusters and exerted a strong cytolytic effect. In contrast, with inefficient antigen-specific binding (anti-BCAR vs BCMA negative MM.1s), the interaction lacks the initial phase and the killing is overall slower and less efficient.
The CAR T cell trafficking and clustering around tumor-cell islands highlights the complex interactions involved in efficient killing of tumor cells. Our results confirm that trafficking is a robust phenomenon that is independent of CAR T density. Moreover, trafficking boosts the local ratio of effector to target cells on the niche, facilitating the killing of tumor cells. This finding echoes a recent in vivo observation in a mouse model of B cell lymphoma which showed that the density of CAR T cells increased by tenfold in 3 d in the bone marrow and the tumor clearance was correlated with the CAR T density. [24] Together, our in vitro data and the in vivo model confirm the importance of CAR T trafficking in promoting tumor-cell killing.
When exerting the cytolytic effect on a tumor island, CAR T cells merge into clusters around tumor cells and collectively shrink the tumor island. We found that the size and morphology of the CAR T cell clusters are correlated with the efficiency of clearance of the tumor cells on the island. The formation of a single large CAR T cluster on the island is always associated with better tumor clearance. The formation of multiple smaller clusters, either due to lower CAR T density or the absence of tumor antigen is related to deficient tumor clearance. These findings imply that efficient clustering of CAR T cells may play an important role in clearing tumor-cell clusters. CAR T cell clustering has been recently reported in a mouse model of B cell lymphoma. CAR T cells formed large cell clusters around malignant B cells in the blood circulation 15 min after injection. [24] Whether the cluster formation in vivo promotes tumor-cell killing or follows the same dynamics observed in vitro remains to be investigated.
Our data show that APRIL-based CAR T cells efficiently killed both BCMA positive and negative MM.1s, while anti-BCMA CAR T cells failed to kill BCMA negative MM.1s. These in vitro data match the results from in vivo experiments which demonstrated that anti-BCMA CAR T cells are unable to clear MM1.s BCMA KO cells engrafted in Non-Obese Diabetic, Severe combined immunodeficient, Interleukin 2 gamma null (NOD-SCIDg chain/(NSG) mice) (manuscript under review). These results suggest that our platform could support the validation of CAR T cell efficacy. The trafficking and clustering of APRIL-CAR T cells occurred in a similar fashion toward both tumor cells, while anti-BCMA CAR T cells showed a deficient ability to form clusters on BCMA negative MM.1s. These suggest the potential of April-CAR T cells to reduce the incidence of antigen-negative escape, without compromising other key cell functions.
Although it permits multifaceted dynamic characterizations of CAR T cells with ease of use and high throughput, MiTA is not without limitations. For example, the recruitment of CAR T cells and their interactions with tumor cells happen on a 2D surface which may differ from those in a physiologically relevant 3D microenvironment. This limitation can be overcome by incorporating more features in MiTA. For example, dispensing CAR T cells embedded in hydrogel on top of tumor island array could realize the monitoring of CAR T cell recruitment and cytolytic activities in 3D. Co-patterning tumor cells with other cellular components such as bone marrow stromal cells could provide a more sophisticated in-vivo like microenvironment. Implementing more features in MiTA will shift it toward a more physiologically relevant model but complicate the preparation and the operation of the system at the same time. The versatility of MiTA platform allows possible system modification to adapt to the requirements of different studies and screenings.
The dynamic profiles of CAR T cell trafficking, clustering, and tumor elimination vary among healthy donors. The differences may stem from the intrinsic variations in T cell populations among donors or variations induced during CAR T cell manufacturing. The functions of CAR T cells among patients are likely to be poorer and vary even more. Deciphering the link between the variability and the corresponding clinical outcome will facilitate the production of more effective CAR T cells and could ultimately serve as a biomarker of response or a measure of T cell "fitness." [25] Ultimately, mapping the dynamic information from in vitro assays, multiomics data of patients, and the clinical outcome could create a landscape that aids the development of more efficient and personalized CAR T cell therapies.

Experimental Section
Construction of CAR, T Cell Culture, and Transduction: Anti-BCMA and APRIL CAR constructs were synthesized and cloned into a thirdgeneration lentiviral plasmid backbone under the regulation of a human EF-1α promoter. Anti-BCMA CAR bears a CD8 hinge and transmembrane domain, 4-1BB costimulatory domain, and CD3 zeta signaling domain. APRIL CAR bears a 4-1BB transmembrane and costimulatory domains and CD3ζ signaling domain. Both vectors also contained a second transgene coding for the fluorescent reporter mCherry to facilitate enumeration of transduction efficiency. Human T cells were purified (Stem Cell Technologies, Catalog #15 061) from anonymous human healthy donor leukopacs purchased from the Massachusetts General Hospital (MGH) blood bank under an Institutional Review Board (IRB)-exempt protocol. For primary T-lymphocyte expansions, bulk human T-cells were activated (day 0) using anti-CD3/CD28 Dynabeads (LifeTechnologies), followed by transduction with a lentiviral vector encoding the CAR 24 h later as described. [26] T cells were cultured in media supplemented with rhIL-2 (20 IU mL −1 ) beginning on day 0 of culture and were maintained at a constant cell concentration (0.5 × 10 6 mL −1 ) by counting every 2-3 d. T cells were de-beaded at day 10 of culture and functional assays were performed at day 11, after resting overnight.
Cell Lines and Culture Conditions: Two B-lymphoblast myeloma cell lines, RPMI 8226 and MM.1s, were purchased from American Type Culture Collection (ATCC). Cells were engineered to constitutively express click beetle green (CBG) luciferase/enhanced GFP (eGFP) and then sorted on a FACSAria (BD) to obtain a pure population (CBG-GFP+) (≥99%). RPMI 8226 cells were cultured in RPMI media containing 10% fetal bovine serum (FBS), penicillin, and streptomycin. MM.1s cells were cultured in RPMI media supplemented with 20% FBS, penicillin, and streptomycin. MM.1s BCMA knockout cells were generated used CRISPR/Cas9 technology.
Flow Cytometry: Anti BCMA-APC antibody was used to detect BCMA expression by flow cytometry (clone 19F2, BioLegend). Cells were stained for 30 min in the dark at 4 °C and washed twice in phosphatebuffered saline with 2% FBS. 4′,6-diamidino-2-phenylindole was added to gate in the viable cells before the acquisition. Samples were run on a Fortessa X-20 (BD) and data analyzed with FlowJo (Version 10).
Microspotting: Poly-L-lysine solution at 0.1% (w/v) (Sigma-Aldrich) and high-molecular-weight cationic ZETAG solution were mixed at a volume ratio of 100:1. The mixture was spiked with FITC-tagged poly-L-lysine (Sigma-Aldrich) for visualization. Using an automatic liquid dispenser (Picospotter, Poly-Pico Technologies LTD), the solution was dispensed into 16 8 × 8 spot arrays on a 3 × 1 in. ultraclean glass slide (SuperChip Microarray Slides, Thermo-Fisher Scientific). The spots were dried at room temperature for overnight. For optimal adhesion of cells on the spots, the slide should be used between15 and 48 h after spotting the material.
Microfabrication of the Microchamber Membrane: The PDMS membrane was fabricated with the standard soft lithography process. Briefly, a master mold was fabricated in a negative photoresist (SU-8, Microchem) with a height of 300 µm on a 4 in. silicon wafer. PDMS base and curing agent (PDMS, Sylgard 184, Elsworth Adhesives) were mixed thoroughly at a ratio of 10:1 and cast on the wafer. To adapt membrane to the commercialized well frame, we fabricated the membrane with a thickness of 1.5 mm by casting PDMS mixture (13.5 g) on the 4 in. wafer. The wafer with the mixture was degassed in a vacuum chamber and then transferred to an oven (80 °C) to cure overnight. After curing, the membrane was diced and peeled from the wafer. Inlets and outlets were created at the four corners of each square chamber with a 1 mm diameter biopsy punch (Harris Uni-Core). The membrane was then treated with oxygen plasma to hydrophilize the PDMS surface.
Device Assembly and Operation: The spotted slide and the PDMS membrane were manually aligned and assembled with a commercialized 16-well chamber frame (ProPlate, Grace Bio-Labs). Tumor-cell sample (20 µL, 25million mL −1 ) were loaded in a microchamber with a pipette. After 5 min, cells that were not on the spots were removed by flushing the chamber with RPMI + 10% FBS (200 µL). After patterning the tumor cells, fibronectin solution (R&D systems) (20 µL, 10 µg mL −1 ) was loaded in the microchamber and incubated at 37 °C and 5% CO 2 for 2 h. The microchamber was washed with media (100 µL) and then covered with media (200 µL) to prevent evaporation during the experiments. CAR T cell suspension (20 µL) at desired concentrations was loaded into the chamber. Finally, the 16-well chamber frame was sealed with a transparent sticky film and was ready for the time-lapse imaging.
Time-Lapse Microscopy: Time-lapse fluorescent microscopy was employed to image the migration and antitumor activity of CAR T cells. Images were taken at 100-300 locations using a 10× or 20× objective with a time interval of 15-30 min using a fully automated Nikon TiE microscope (Micro Device Instruments). The microscope is equipped with a heat chamber which provides 37 °C, 5% CO 2 and humidity for long term imaging. Files in .nd2 format were imported into Fiji ImageJ for analysis.
Image Processing and Data Analysis: Time-lapse images were processed in Fiji ImageJ. The area and roundness of the fluorescent poly-L-lysine spots were measured by setting an automatic threshold to the images, followed by Analyze Particles function. The number of tumor cells on a spot was measured automatically with Trackmate module in ImageJ. The "estimated cell sizes" were set to 14 and 10 µm and "intensity thresholds" to 3 and 0.5 for RPMI 8226 and MM1s tumor cells, respectively. These parameters were selected based on the cell size and GFP intensity of each cell line. It was verified that these parameters generate accurate cell counts by comparing the number counted automatically and manually. After confirming the accuracy, the parameters were fixed for all the experiments. The percentage of remaining tumor cells was calculated as the number of cells at a time point over the initial number of cells. The area of tumor cells and CAR T cells on a spot were measured in ImageJ using a macro. The macro set automatic thresholds to a stack of fluorescent images in either Triangle or Huang modes and measured the area of GFP (tumor cells) or mCherry (CAR T cells) positive objects in the image stack. The data were imported into Excel. The graphs were plotted using excel, R and GraphPad Prism.
To calculate the ratio of effector to target cells (E:T ratio), the area density of CAR T cells was calculated first according to the concentration of CAR-T suspension and the dimensions of the chamber. Then, the number of CAR T cells in a 500 × 500 µm 2 region was calculated. This area was chosen because the 64 tumor islands were arranged into an 8 × 8 array spaced 500 µm apart. The ratio of CAR-T/tumor cells calculated the number of CAR T cells in the 500 × 500 µm 2 region over the average number of tumor cells on a spot.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.