CAR T-cell Entry into Tumor Islets Is a Two-Step Process Dependent on IFNγ and ICAM-1

Adoptive immunotherapy with gene-engineered chimeric antigen receptor (CAR) T cells has demonstrated efficacy in several B-cell malignancies but not yet in solid tumors (1). This could be due, at least in part, to the existence of various physical and environmental barriers in solid tumors, which are less present in hematologic cancers. In carcinomas, lymphocytes have to navigate through a dense stroma that surrounds cancer cells. Once within the tumor, T cells must expand, persist, and mediate cytotoxicity in a hostile environment largely composed of immunosuppressive cells and molecules. Using an experimental system of human tumor slices combined with dynamic imaging microscopy, we previously demonstrated that a dense extracellular matrix (ECM; refs. 2, 3) and tumor-associated macrophages (TAM; ref. 4) hinder antitumor functions of T cells in progressing lung tumors by reducing their migration and sequestering them in the stroma. As a result, interactions between T cells and tumor cells are impaired. After overcoming these stromal obstacles, T cells need to infiltrate tumor islets to carry out cytotoxic activity. Up to now, little is known about the elements that promote CAR T-cell interactions with their targets. Much attention has been paid to the expression of the target antigen, and decreased expression of the targeted molecule at the surface of tumor cells has been recognized as a major cause of CAR T-cell resistance (5). The importance of chemokines and their receptors in attracting CAR T cells and promoting their contact with malignant cells has also been recognized and therapeutically exploited in mouse tumor models (6).

Among adhesion molecules, two integrins, lymphocyte function-associated antigen-1 (LFA-1) and CD103 (αE/β7), play a critical role in T cell–tumor cell interactions (7, 8). LFA-1 is an integrin expressed by T cells and other hematopoietic cells and binds to intercellular adhesion molecule 1 (ICAM-1) expressed by antigen-presenting cells, as well as by some tumor cells. When T cells are stimulated by a variety of external stimuli, including antigens and chemokines, the affinity and clustering of LFA-1 increases in an inside-out signaling process that promotes the binding of this integrin to ICAM-1 (9). Whether similar molecules and mechanisms control the antitumoral action of CAR T cells remain to be established. Here, we used fluorescent imaging microscopy and ex vivo assays to compare the activation and distribution of CAR T cells upon contact with malignant B cells and carcinoma cells. The CARs used in this study were specific for CD20, a pan–B-cell antigen, and EGFR, a pan-carcinoma antigen. We describe a two-step entry of EGFR CAR T cells into human tumor islets that is dependent on IFNγ and ICAM-1.

Human studies were carried out according to French law on biomedical research and to principles outlined in 1975 Helsinki Declaration and its modification. Institutional review board approval was obtained (CPP Ile de France II, #00001072, August 27, 2012). Animal studies were approved by the animal experimentation ethics committee of Paris Descartes University (CEEA 34, 17-039) and by the French ministry of research (APAFiS #15076).

Six fresh lung tumors were obtained from anonymized patients diagnosed with clinical stage I–III non–small cell lung cancer (NSCLC) and who underwent primary lobectomy or pneumonectomy. Three tumors with very high autofluorescence of fibers precluding imaging of T cells, with an unclear distinction between tumor cell regions and stroma, were discarded after visual inspection of several microscopic fields. Seven fresh renal tumors were obtained from anonymized patients diagnosed with clear cell renal cell carcinoma (ccRCC) and who underwent laparotomy. All ccRCC tumors were used for subsequent experiments. NSCLC and ccRCC patients did not receive any treatment before the surgery. Nonfixed, fresh tumors obtained after resection were rapidly transported to the laboratory in ice-cold RPMI-1640 (Thermo Fisher Scientific; cat #11875093). Experiments were performed with tumor specimens obtained 2 to 24 hours after tumor resection. Tumors were kept at 4°C until being processed and cut in slices (see the section below). For lung and renal human tumors, a written informed consent was obtained from the patients prior to inclusion in the study.

Chronic lymphocytic leukemia (CLL) blood samples were obtained from 13 untreated patients after written informed consent and validation by the local research ethics committee from the Avicenne Hospital (Bobigny, France), in accordance with the Declaration of Helsinki. The MACSxpress Whole Blood B-CLL Cell Isolation Kit (Miltenyi Biotec; cat #130–104-445) was used for isolation of untouched B cells from anticoagulated whole blood without density gradient centrifugation. Purified B cells were 95% to 99% CD5⁺, the marker of malignant CLL cells. B cells were then frozen at a concentration of 5 × 10⁶ cells/mL in 90% fetal bovine serum (FBS; Thermo Fisher Scientific; cat #16140063) and 10% DMSO (Thermo Fisher Scientific; cat #D12345).

Human cell lines HEK293T (ATCC; cat #CRL-11268, RRID:CVCL_1926), Raji from the Epstein–Barr virus–positive Burkitt lymphoma (ATCC; cat #CCL-86, RRID:CVCL_0511), and BxPC3 derived from pancreatic adenocarcinoma cells (ATCC; cat #CRL-1687, RRID:CVCL_0186) were maintained in culture in complete RPMI medium containing 10% FBS, penicillin (50 U/mL), and streptomycin (50 μg/mL; penicillin–streptomycin from Thermo Fisher Scientific; cat #15140122). Raji and BxPC3 cell lines were authenticated by the ATCC using short tandem repeat DNA profiling analysis (June 2013). The cells were not reauthenticated. The EGI-1 tumor cell line, derived from extrahepatic biliary tract and obtained from the German Collection of Microorganisms and Cell Cultures (DSMZ; cat #ACC-385, RRID:CVCL_1193), was maintained in culture in DMEM supplemented with glucose (1 g/L), 10 mmol/L HEPES (Thermo Fisher Scientific; cat #10564011), 10% FBS, penicillin (50 U/mL), and streptomycin (50 μg/mL). The EGI-1 tumor cell line was authenticated by the DSMZ using short tandem repeat DNA profiling analysis. The cells were not reauthenticated. The stromal cell line hTERT-HSC derived from human activated hepatic stellate cells (RRID:CVCL_M102) were kindly provided by Dr. L. Fouassier, authenticated, and cultured in DMEM supplemented with glucose (4.5 g/L), antibiotics, and 10% FBS. Each cell line was thawed from a lab frozen stock, which were generated from early passages and utilized for each experiment within 4 weeks of culture. All the cell lines utilized were Mycoplasma free determined by qPCR analysis. BxPC3 cells expressing human CD20 (P11836) were obtained by infection with a vesicular stomatitis virus (VSVg)-pseudotyped lentiviral particles encoding human CD20-IRES-GFP. A multiplicity of infection (MOI) of 5 transduction units/mL (TU/mL) lentivirus stock was used for cell transduction. Transduced BxPC3 cells positive for GFP were sorted using a MACSQant Tyto (Miltenyi Biotec).

Silencing of ICAM-1 in BxPC3 tumor cells was done by short hairpin (sh)RNA–mediated gene-expression knockdown. Briefly, lentiviral particles at an MOI of 20 TU/mL (GeneCopoeia, cat #LPP-HSE009184-LvE004-50) encoding specific shRNAs for the knockdown of ICAM-1 were used according to the manufacturer's directions. Lentiviral particles encoding a nontargeting shRNA vector were used as control at an MOI of 20 TU/mL (GeneCopoeia, cat #HSH009184-CH1). Transduced cells were selected with the above described RPMI culture medium supplemented with puromycin (10 μg/mL; Thermo Fisher Scientific; cat #A1113802).

ICAM-1 knockout (KO) Raji cells and control KO Raji cells were purchased from Synthego (Knockout Cell Pool) using CRISPR–Cas9 technology. ICAM-1 sgRNA (5′-TTACTGCACACGTCAGCCGC-3′) was used. The CRISPR-edited Raji cells were subcloned by limiting dilution. Loss of ICAM-1 protein was confirmed by flow cytometry analysis.

CAR T cells were generated from 3 different healthy donors. Lentiviral transfer vector plasmids encoding second-generation anti-EGFR CARs were generated by cloning scFv sequences derived from nimotuzumab (IMGT/2Dstructure-DB INN 8545H, 8545L) or cetuximab (IMGT/2Dstructure-DB INN 7906H, 7906L) into the CAR expression cassette encoding CD8 hinge and transmembrane domains (P01732), 4-1BB costimulatory domain (Q07011), and CD3z signaling domain (P20963-3) and a 2A site, followed by a truncated LNGFR (P08138) or mCherry for expression monitoring. Generation of transfer vectors encoding the CD20 CAR described previously (10). VSVg-pseudotyped lentiviral particles were produced in the HEK293T packaging cell line using a second-generation plasmid system. PBMCs from healthy donors were isolated from buffy coats by Pancoll (PAN-Biotech) density gradient centrifugation. T cells were isolated from PBMCs using the Human Pan T-cell isolation kit (Miltenyi Biotec; cat #130-096-535), eluted in TexMACS media (Miltenyi Biotec; cat #130-097-196), and subsequently activated by TransAct (Miltenyi Biotec; cat #130-111-160). Activated cells were cultivated in TexMACS media in the presence of IL7 (10 ng/mL; Miltenyi Biotec; cat #130-095-362) and IL15 (10 ng/mL; Miltenyi Biotec; cat #130-095-765), respectively, and transduced on day 1 using an MOI of 5 TU/mL of lentivirus stock was determined on SupT1 cells (ATCC; cat #CRL-1942, RRID:CVCL_1714). Cells were analyzed for LNGFR expression by flow cytometry on day 3 after transduction. TransAct was washed out on day 3, and cells were maintained at 0.5–2 × 10⁶ cells/mL during the expansion phase and cryopreserved (TexMACS 70% (v/v), FBS 20% (v/v), DMSO 10% (v/v)) on day 12. Transduction efficiency was determined on days 7 and 12 (and directly before experiments) by anti-LNGFR (Miltenyi Biotec; cat #130-112-601, RRID:AB_2725864) staining and flow-cytometric analysis. Transduction efficiency with the CD20 CAR or EGFR CAR was between 50% and 85%.

Tumor slices were prepared as previously described (2, 3) and without modification. Samples from human (3 from NSCLC and 7 from ccRCC) and mouse tumors (24 from BxPC3 and 12 from Raji) were embedded in 5% low-gelling-temperature agarose (Sigma-Aldrich; cat #A0701-25G) prepared in PBS (Thermo Fisher Scientific; cat #10010001). Tumors were cut with a vibratome (Leica VT1200S vibratome, RRID:SCR_018453) in a bath of ice-cold PBS. The thickness of the slices was 400 μm for solid tumors and 800 μm for Raji tumors. Slices were transferred to 0.4-μm organotypic culture inserts (Merck Millipore; cat #PICM03050) in 35-mm Petri dishes containing 1 mL RPMI-1640 (without phenol red; Thermo Fisher Scientific; cat #11835063).

Human B Raji lymphoma and human pancreatic BxPC3 tumors were established by transplantation of 10×10⁶ cells injected subcutaneously into the flanks of immunodeficient NSG mice (IMSR; cat #JAX:005557, RRID:IMSR_JAX:005557). When tumors reached a volume of 200 mm³ (∼3 weeks after implantation), mice were sacrificed, and tumors were removed and processed in tumor slices as described above. Tumors were measured by calipers every 3 days, and the tumor volumes were estimated by tumor volume = width × thickness × length/2. Mice were maintained in the Cochin Institute specific pathogen–free animal facility. Animal care was performed by expert technicians in compliance with the Federation of European Laboratory Animal Science Associations.

The in vitro experiments consisted of adding fluorescent T cells onto tumor cells (BxPC3, EGI-1 or Raji) and then monitoring lymphocyte activity with a fluorescent microscope. Twenty-four hours before imaging, BxPC3 and EGI-1 cells (1 × 10⁵ cells/well) were cultured in ibidi μ-slides. To allow Raji cell adhesion to wells of ibidi μ-slides, tumor cells were washed in Ca²⁺-free HBSS (Thermo Fisher Scientific; cat #14175095) and then added (2 × 10⁵ cells/well) in wells 1 hour before the imaging. Nontransduced and CAR T cells were stained at 37°C with either 1 μmol/L CellTracker Green CMFDA Dye (Thermo Fisher Scientific; cat #C2925) for 5 minutes or 125 nmol/L CellTrace Calcein Red-Orange (Thermo Fisher Scientific; cat #C34851) for 10 minutes, and then washed in RPMI-1640 (without phenol red) supplemented with 1% FBS. Finally, nontransduced or CAR T cells (2–10 × 10⁵ cells/well) were added onto tumor cells at E:T ratios of 1:1 and 5:1, depending on the experiment. An inverted widefield microscope (Nikon Eclipse TE2000-U) preheated at 37°C was used for T-cell imaging. In some experiments, tumor cells or CAR T cells were incubated at 37°C with blocking anti–ICAM-1 (clone HA58; BioLegend; cat #353102, RRID:AB_11204426), anti–LFA-1 (clone TS1/18; Thermo Fisher Scientific; cat #MA1810, RRID:AB_223514), anti-IFNγ (R&D Systems; cat #AF-285-NA, RRID:AB_354445), and anti-CD104 (clone ASC-8; Millipore; cat #MAB2058, RRID:AB_94524; 10 μg/mL) for 20 minutes before the experiments. Time points of 30 minutes, 2 hours, and 4 hours were taken.

For imaging CAR T cells in fresh slices of human (NSCLC and ccRCC) and mouse tumors (BxPC3), 0.1 × 10⁶ or 2 × 10⁶ lymphocytes were plated onto sections. Imaging was performed with an upright spinning disk confocal microscope (Leica DM6000 FS, Yokogawa CSU-X1 head unit) equipped with a 37°C thermostated chamber. For dynamic imaging, tumor slices were secured with a stainless-steel ring slice anchor (Warner Instruments; cat #64–1415) and perfused at a rate of 0.3 mL/min with a solution of RPMI (without phenol red), bubbled with 95% O₂ and 5% CO₂. Images were acquired with a 25× water immersion objective (Leica, 25×/0.95 NA) or a 10× objective (Leica 10×/0.3 NA). For four-dimensional analysis of cell migration, stacks of 10 to 12 sections (z step = 5 or 15 μm for the 25× or 10× objectives, respectively) were acquired every 30 seconds for 20 minutes, at depths up to 80 μm. Videos were made by compressing the z information into a single plane with the max intensity z projection of ImageJ software function (ImageJ, RRID:SCR_003070). In some experiments, live vibratome sections were stained for 15 minutes at 37°C with BV421-anti-human EpCAM (clone 9C4; BioLegend; cat #324220, RRID:AB_2563847), AF405-anti-human fibronectin (clone HFN7.1; Novus; cat #NBP2-34633AF405), AF647-anti-human CD3 (clone SK7; BioLegend; cat #344826, RRID:AB_2563441), BV421-anti-human EGFR (clone AY13; BioLegend; cat #352921, RRID:AB_2687195), AF647 or PE-anti-human CD54 (ICAM-1; Miltenyi Biotec; cat #130-103-839, RRID:AB_2658695), PE-anti-human CD106 (VCAM-1; Miltenyi Biotec; cat #130-122-008, RRID:AB_2801824), AF647-anti-mouse/human pY701-Stat1 (clone 4a; BD Biosciences; cat #612597, RRID:AB_399880), eFluor660-anti-mouse Gp38 (podoplanin; clone 8.1.1; Thermo Fisher Scientific; cat #50-5381-82, RRID:AB_11151516). Antibodies were diluted in RPMI (without phenol red) and used at a concentration of 10 μg/mL.

Intracellular Ca²⁺ concentration of T cells was measured as previously described (11) with modifications. In brief, CAR T cells (5 × 10⁶ cells/mL) were incubated for 30 minutes at 37°C with either 1 μmol/L Fura-2 AM (Thermo Fisher Scientific; cat #F1221) or Fluo-4 AM (Thermo Fisher Scientific; cat #F14201). T cells were then washed in HBSS and resuspended in TexMACS complete medium with 3% human AB serum (Sigma-Aldrich; cat #H3667). For in vitro experiments, Fura-2–loaded CAR T cells were added at a 1:1 ratio to the tumor cell layer cultured in ibidi μ-slides. Images were acquired every 10 seconds at 350 nm and 380 nm. Emissions at 510 nm were used for the analysis of Ca²⁺ responses with MetaFluor software (MetaFluor Fluorescence Ratio Imaging Software, RRID:SCR_014294). Ca²⁺ values were represented as a ratio: fluorescence intensity at 350 nm/fluorescence intensity at 380 nm. CAR T cells were considered responsive when the amplitude of their responses reached at least twice that of the background.

For experiments in slices, Fluo-4–loaded CAR T cells (2 × 10⁵ cells/slice) were plated onto fresh slices of human (ccRCC) and mouse tumors (Raji and BxPC3). Images were then acquired using the upright confocal spinning microscope described above. Fluo-4 fluorescence measurements were normalized by dividing the average fluorescence intensity (F) occurring during the course of the experiment to the average fluorescence intensity determined at the beginning of each experiment (F0). Fluorescence measurements in individual CAR T cells were performed using Imaris 7.4 (Imaris, RRID:SCR_007370).

mCherry⁺ BxPC3 tumor cells (control and ICAM-1 shRNA) were cocultured with EGFR CAR T cells in ibidi μ-slides at a 1:1 ratio. Plates were imaged at 0, 6, and 24 hours using an inverted widefield microscope (Eclipse TE2000-U; Nikon). Four images per well were collected at each time point. Total integrated mCherry intensity per microscopic field was assessed as a quantitative measure of live BxPC3 cells. Values were normalized to the t = 0 measurements. Cell death within tumor slices was assessed using a green-fluorescent caspase-3 probe, NucView, that binds DNA upon cleavage by caspase-3 (Biotium; cat #10402). 2 × 10⁶ untransduced or EGFR CAR T cells were applied onto BxPC3 tumor slices that were subsequently incubated with 5 μmol/L of NucView dye. Twenty hours later, slices were rinsed, immunostained (described below), and imaged with an upright spinning disk confocal microscope.

Tumor cells cocultured with CAR T cells for 1 to 4 hours were fixed in 4% paraformaldehyde (PFA; Thermo Fisher Scientific; cat #J61899.AK) for 10 minutes at 4°C and then washed in PBS and stained with anti–ICAM-1 (Miltenyi Biotec; cat #130-103-839, RRID:AB_2658695) or anti-phosphorylated STAT1 pY701 (clone 4a, BD Biosciences; cat #612597, RRID:AB_399880). Antibodies were diluted in RPMI (without phenol red) and used at a concentration of 10 μg/mL. For phospho-STAT1 intracellular staining, cells were permeabilized with ice-cold 90% methanol for 12 hours at 4°C.

Immunostaining of a BxPC3 tumor slice (8-μm thick cryosection, fixed in acetone) was taken using a prototype of the MACSima Imaging Platform, a cyclic immunofluorescence device developed by Miltenyi Biotec (Miltenyi Biotec; cat #130-121-164). The three displayed images of EpCAM-APC, EGFR-PE, and CD104-APC were obtained from different cycles of a much longer run (72 distinct immunolabelings from 24 3-color stain-erase cycles) on this single tissue slice. All immunostainings were obtained using recombinant, primary antibodies directly labeled with specific fluorophores. Antibodies used are listed in Supplementary Table S1.

Image analysis was performed at the Cochin Imaging Facility (Institut Cochin, Paris). A 3D image analysis was performed on x, y, and z planes using Imaris 7.4 software (RRID:SCR_007370). First, superficial planes from the top of the slice to 15 μm in depth were removed to exclude T cells located near the cut surface. Cellular motility parameters were then calculated using Imaris. Tracks >10% of the total recording time were included in the analysis. When a drift in the x, y dimension was noticed, it was corrected using the “Correct 3D Drift” plug-in in ImageJ. CAR T-cell concentration, motility, and Ca²⁺ responses were quantified in different tumor regions, namely, stroma versus tumor islets and islet–stroma junctions. These regions were identified by visual inspection of immunofluorescence images. Fluorescence intensities were determined along a line scan or in regions of interest using ImageJ. The number of T cells in defined regions was quantified using the Analyze Particles function of ImageJ from fluorescent images that were first thresholded and then converted to binary images.

The cytokines IFNγ and TNFα contained in supernatants derived from 1- to 4-hour BxPC3-CAR T-cell cocultures were quantified using LEGEND MAX Human IFNγ and TNFα ELISA Kits in accordance with the manufacturer's instructions (BioLegend; cat #430107 for IFNγ; BioLegend; cat #430207 for TNFα). Standards and supernatants were added to the plate, and following a 2-hour incubation, the plate was washed, and the detection antibody was added for 2 hours. After washing, streptavidin-HRP was added for 20 minutes, followed by washing and addition of the kit's substrate solution. Following visible color development, the kit's stop solution was added, and the optical density was immediately determined at 450 nm and background at 540 and 570 nm wavelengths using a fluorometer (PerkinElmer). Background readings were subtracted from the readings at 450 nm, and IFNγ and TNFα concentration was determined based on the standard curve.

The membrane expression of CD20 and ICAM-1 on B cells from CLL patients was assessed by flow cytometry. Briefly, cells were washed and stained with PE-anti-human CD54 (ICAM-1; Miltenyi Biotec; cat #130-103-839, RRID:AB_2658695) and APC-anti-human CD20 (clone L27; BD Biosciences; cat #340908, RRID:AB_2868743) for 20 minutes at 4°C. Antibodies were diluted in PBS supplemented with 0.1% BSA (Thermo Fisher Scientific; cat #15561020) and used at a concentration of 10 μg/mL For assessing the viability of Raji B cells deficient or not for ICAM-1, cells were incubated with CD20 CAR T cells at different ratios for 6 hours, washed with PBS, resuspended in Annexin V binding buffer (BioLegend; cat #422201), and incubated with allophycocyanin-conjugated Annexin V (BioLegend; cat #640941) and 7AAD (BioLegend; cat #420404) used at 0.5 μg/mL for 15 minutes at room temperature. The cells were immediately analyzed by flow cytometry. Acquisition was performed on an LSRFortessa flow cytometer (Becton Dickinson). Data were analyzed using FlowJo software (FlowJo, RRID:SCR_008520).

RNA-sequencing (RNA-seq) data from The Cancer Genome Atlas (TCGA) from 487 patients suffering from lung squamous cell carcinoma (TCGA-LUSC) and 177 patients suffering from pancreatic adenocarcinoma (TCGA-PAAD) were downloaded and were accessed as upper quartile–normalized fragments per kilobase of transcript per million mapped reads using FireHose data repository (https://gdac.broadinstitute.org/), for each cancer of interest. RNA-seq data analysis was performed by GenoSplice technology (www.genosplice.com) using its in-house software. The lymphocyte infiltration and Th1 signature gene-expression signatures used have been previously defined (12). Only tumor samples from patients with clinical data information from Thorsson and colleagues (12) were included. Spearman correlation was used to quantify the association between ICAM-1 gene expression and the expression of the other genes, for each tumor type.

Statistical analysis was carried out using GraphPad Prism (GraphPad Prism, RRID:SCR_002798). Averages are expressed as mean ± SEM. Significant differences between two series of results were assessed using the unpaired two-tailed Student t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001. Correlations were examined by the Spearman correlation method.

First, we used fluorescent imaging microscopy to visualize the activity of CD20 CAR T cells when contacting CD20⁺ tumor cells. CD20 CAR T cells, loaded with the fluorescent dye fura-2, were added onto a monolayer of the B lymphoma cell line Raji. The intracellular Ca²⁺ rise in T cells following CAR engagement is one the earliest signs of T lymphocyte activation. The interaction between CD20 CAR T cells and Raji B cells was rapidly followed by large and sustained Ca²⁺ increases (Fig. 1A; Supplementary Movie S1). By comparison, untransduced T cells transiently interacted with several Raji cells without increasing their Ca²⁺ (Fig. 1A, right). Next, we monitored the motile behavior and Ca²⁺ responses of CD20 CAR T cells in Raji tumor tissues (Fig. 1B and C). For these experiments, Raji cells were implanted subcutaneously into immune-deficient mice (NSG). Three weeks later, tumors were vibratome-sliced and CD20 CAR transduced or untransduced T cells were plated onto fresh slices. Whereas untransduced T cells migrated very actively within tumor slices in an apparently random manner without stopping, CD20 CAR T cells were static, although not totally arrested (Fig. 1B). Unlike untransduced lymphocytes, CD20 CAR T cells exhibited a rapid increase in their intracellular Ca²⁺ upon infiltration into Raji tumor slices, although more transiently than that measured in vitro (Fig. 1C; Supplementary Movie S2).

We then monitored Ca²⁺ responses of CD20 CAR T cells during their interaction with malignant B cells from 13 CLL untreated patients. Our results indicated that the percentage of Ca²⁺-responding CAR T cells ranged from 15% to 50%, depending on the patient (Fig. 2A). This variability did not seem to be related to CD20 expression because there was no clear correlation between the proportion of activated CD20 CAR T cells and target antigen density, as assessed by flow cytometry (Fig. 2A). This result suggests that other surface proteins expressed by tumor cells dictate CD20 CAR T-cell responsiveness. We then investigated the role of the adhesion molecule ICAM-1, which binds to LFA-1 expressed by T cells to control the formation of a productive immune synapse (9). To this end, Ca²⁺ responses of CD20 CAR T cells induced by malignant B cells from CLL patients expressing various levels of ICAM-1 were measured. A clear positive correlation was observed between the percentage of activated CAR T cells and ICAM-1 density (Fig. 2B), as underlined by the tumors of two patients with low (pink) and intermediate (green) ICAM-1 expression. Blocking anti–ICAM-1 or anti–LFA-1 decreased the percentage of Ca²⁺-responding CD20 CAR T cells in contact with Raji B cells (Fig. 2C). We then evaluated the contribution of ICAM-1 to the cytotoxicity of CD20 CAR T cells. ICAM-1–negative Raji cells were generated by knocking out ICAM-1 using CRISPR–Cas9 gene editing. The killing efficacy of CD20 CAR T cells against ICAM-1 wild-type (WT) and ICAM-1 KO Raji cells was compared by measuring the percentage of early (Annexin V⁺/7AAD⁻) and late (Annexin V⁺/7AAD⁺) apoptotic Raji cells. ICAM-1 deletion decreased cytotoxicity of CD20 CAR T cells (Fig. 2D). These results demonstrate that ICAM-1 expression by malignant B cells plays a crucial role in the responsiveness of CD20 CAR T cells to their targets.

After characterizing the responses of CD20 CAR T cells during their interactions with malignant B cells, imaging experiments were then performed with EGFR CAR T cells and the human tumor pancreatic cell line BxPC3, which expresses EGFR. Untransduced T cells or T cells expressing CD20, an irrelevant CAR, were used as controls. Of note, BxPC3 cells did not undergo epithelial–mesenchymal transition and, therefore, still expressed epithelial markers such as EpCAM. In culture, these cells grew in clusters with conspicuous peripheral and central areas (Fig. 3A). We observed that EGFR CAR T cells did not form many productive conjugates with pancreatic tumor cells (Fig. 3A; Supplementary Fig. S1A). Most CAR T cells showing Ca²⁺ increases were in contact with tumor cells located at the periphery of the clusters (Fig. 3A, right; Supplementary Movie S3). In contrast, few CAR T cells were activated by interaction with tumor cells in the center of the islets. This differential spatial response was observed with T cells expressing either high- (cetuximab) or low- (nimotuzumab) affinity CARs for EGFR. The only difference between the two CARs was that Ca²⁺ spikes were more oscillatory with the low-affinity CAR (Supplementary Fig. S1B). Such peripheral CAR T-cell activation was also noted with the cholangiocarcinoma cell line EGI-1, which showed a spatial organization similar to that of BxPC3 (Supplementary Fig. S1C). The propensity to preferentially activate CAR T cells at the islet outskirts was independent of the target antigen because similar results were obtained with BxPC3 transduced with CD20 and CAR T cells redirected toward CD20 (Fig. 3B; Supplementary Movie S4).

We next assessed if this peripheral activation of CAR T cells was observed on fresh slices from BxPC3 tumors isolated from NSG mice. Indeed, EGFR CAR T cells were enriched at the tumor–stroma border compared with nontransduced T cells, which exhibited a homogeneous distribution in all compartments (Fig. 3C; Supplementary Movie S5). EGFR CAR T cells were relatively static at this border (Fig. 3C, right) and underwent frequent Ca²⁺ responses (Fig. 3D) in sharp contrast with T cells located in the stroma and in tumor islets (Supplementary Movie S6). The same pattern was observed when EGFR CAR T cells were added onto fresh human lung tumor slices, with a structure representative of a carcinoma composed of compact tumor islets surrounded by a fibrous stroma (Fig. 3E, left). In terms of CAR T-cell distribution, lymphocytes were enriched in the stroma and at the edge of the islets in contact with peripheral tumor cells (Fig. 3E, middle). CAR T cells in the tumor islet periphery were relatively static compared with untransduced lymphocytes in the same regions (Fig. 3E, right; Supplementary Movie S7). This peculiar enrichment and arrest of EGFR CAR T cells at the tumor–stroma junction were also confirmed in slices made from human renal cell carcinomas (Supplementary Fig. S1D; Supplementary Movie S8). Overall, these results revealed that EGFR CAR T cells preferentially form productive conjugates with tumor cells localized at the periphery of the clusters, but not with those localized in the center of tumor islets.

The key role played by ICAM-1 in controlling the activation of CD20 CAR T cells during their contact with malignant B cells prompted us to assess the expression of this adhesion molecule at the surface of BxPC3 pancreatic and EGI-1 cholangiocarcinoma tumor cell lines. Our data indicated that ICAM-1 was hardly detected in tumor islets formed by BxPC3 and EGI-1 tumor cells, even at their periphery. This contrasts with the endogenous expression of ICAM-1 by the Raji B lymphoma cell line (Fig. 4A) and suggests that a defect in the expression of ICAM-1 by solid tumors may explain why EGFR CAR T cells are relatively inefficient in forming productive conjugates with carcinoma cells.

Next, we performed immunostaining experiments using the MACSima approach to study the distribution of 25 membrane proteins on a single BxPC3 tumor cryosection. We found that EpCAM, as well as EGFR, presented homogeneous distributions, with no enrichment at the periphery of tumor islets of BxPC3 tumors (Fig. 4B; Supplementary Fig. S1E). In contrast, CD104 (integrin β4) and CD49f (integrin α6) exhibited a clear enrichment on the edge of the islets (Fig. 4B, top). A line-scan analysis showed a 2- to 3-fold enrichment of CD104 and CD49f at the periphery, whereas all the other proteins presented an even distribution (Fig. 4B, bottom). These results suggest that the preferential enrichment of CD104 and CD49f at the periphery of the tumor cell clusters might modulate the capacity of EGFR to be detected by CAR T cells. CD104 dimerizes exclusively with CD49f to form the integrin α6β4, which functions as a receptor for the basement membrane protein laminin. We thus tested the consequences of targeting CD104 with a blocking antibody on the ability of EGFR CAR T cells to be activated at the periphery of BxPC3 clusters. Our results indicated a partial, but significant, decrease in the percentage of CAR T cells showing Ca²⁺ responses at the margin of the BxPC3 monolayer (Fig. 4C). Overall, these results demonstrate that carcinoma cell lines BxPC3 and EGI-1, unlike Raji B cells, express low ICAM-1. The data also suggest that the preferential expression of CD104 at the periphery of BxPC3 clusters partially controls the initial responses of EGFR CAR T cells when interacting with target cells.

ICAM-1 expression is known to be controlled by various inflammatory cytokines, including TNFα and IFNγ, produced by activated T cells (13). We thus wondered if EGFR CAR T cells could induce the upregulation of ICAM-1 on BxPC3 cells. To this end, we performed coculture experiments between EGFR CAR T cells and BxPC3 carcinoma cells for 4 hours. ICAM-1 expression on target cells was increased by EGFR CAR T cells but not by CD20 CAR T cells (Fig. 5A). Recombinant IFNγ, which was used as a positive control, partially mimicked the effect of EGFR CAR T cells on ICAM-1 upregulation. In comparison, recombinant TNFα was much less effective than IFNγ at inducing ICAM-1 expression on BxPC3 cells (Supplementary Fig. S2A). Conversely, VCAM-1 expression on the surface of BxPC3 tumor cells was not increased by EGFR CAR T cells (Supplementary Fig. S2B). EGFR CAR T cells, but not CD20 CAR T cells, were able to induce phosphorylation of STAT1, a key molecule for IFNγ signaling, in BxPC3 tumor cells (Fig. 5B). Phopho-STAT1 was detected as early as 1 hour after adding EGFR CAR T cells. CAR T cells were as effective as recombinant IFNγ to induce phospho-STAT1. This result highlights the role of IFNγ produced by activated EGFR CAR T cells in this process. Indeed, we could detect the presence of IFNγ in the supernatant of CAR T cell–tumor cell cocultures (Supplementary Fig. S2C). EGFR CAR T cell–induced ICAM-1 upregulation on BxPC3 cells was reduced by neutralizing anti-IFNγ (Fig. 5C). Our results showing that activated CAR T cells drive the expression of ICAM-1 in an IFNγ-dependent fashion led us to investigate the expression of ICAM-1 in human renal carcinomas, in relation to resident CD8⁺ T cells. As predicted, in regions enriched for CD8⁺ T cells, tumor cells expressed high ICAM-1 (Fig. 5D). Slices from human renal carcinomas increased ICAM-1 expression in response to IFNγ treatment (Fig. 5E). Finally, analysis of TCGA data sets showed a positive correlation between activated CD8⁺ T cells, IFNγ, and ICAM-1 transcripts in human lung squamous cell carcinomas (Fig. 5F; Supplementary S2D) and pancreatic adenocarcinomas (Supplementary Fig. S2E). Overall, these data suggest that activated T cells, via secretion of IFNγ, are able to induce carcinoma cells to upregulate ICAM-1.

We next investigated the consequences of CAR T cell–induced ICAM-1 expression on carcinoma cells. We hypothesized that this process created a positive feedback loop by promoting CAR T-cell interaction with cancer cells. To address its kinetics, we followed the distribution of EGFR CAR T cells at different times (30 minutes, 2 hours, and 4 hours) after adding onto BxPC3 tumor cells (Fig. 6A). Our data showed that at 30 minutes and 2 hours, most of the EGFR CAR T cells were located at the periphery of the tumor cell clusters, confirming our previous finding (Fig. 3). However, 4 hours after plating, EGFR CAR T cells were preferentially found within tumor cell clusters, unlike irrelevant CD20 CAR T cells (Fig. 6A, right). This shows that CAR T cells progressively acquired the ability to interact with tumor cells localized in the center of the islets. A similar enrichment into tumor islets was observed with CD20 CAR T cells cultured with CD20-expressing BxPC3 cells (Supplementary Fig. S3A). To better mimic carcinoma organization, BxPC3 tumor cells were cocultured with human immortalized hepatic stellate cells. In this model recreating a structure of tumor cell regions surrounded by mesenchymal cells, EGFR CAR T cells interacted first with peripheral tumor cells before being enriched in the center of tumor islets (Supplementary Fig. S3B). This cell distribution was again associated with an ICAM-1 upregulation by BxPC3 tumor cells.

The importance of LFA-1–ICAM-1 interaction in this process was investigated in several ways. First, we assessed the distribution of LFA-1 and ICAM-1 on EGFR CAR T cells in contact with BxPC3 after 4 hours of culture. Our data revealed enrichment of LFA-1, but not of ICAM-1, at the synapse formed between T cells and their targets (Fig. 6B). Next, BxPC3 cells were genetically modified using ICAM-1–targeted shRNA or control scrambled shRNA. Immunofluorescence confirmed the specific silencing of ICAM-1 (Supplementary Fig. S3C). Knockdown of ICAM-1 substantially decreased the enrichment of EGFR CAR T cells within tumor cell regions after 4 hours (Fig. 6C). Similar inhibition of CAR T-cell distribution was also observed with blocking anti–ICAM-1 (Fig. 6C, right). EGFR CAR T cells demonstrated reduced killing capacity (Fig. 6D) in response to ICAM-1 knockdown BxPC3 cells compared with WT cells. An inhibition of CAR T-cell enrichment into tumor islets could also be observed when IFNγ was neutralized with anti-IFNγ (Supplementary Fig. S3D). Conversely, the treatment of BxPC3 with IFNγ for 12 hours promoted rapid enrichment, as fast as 1 hour, of EGFR CAR T cells into tumor islets (Fig. 6E). This treatment also rendered central BxPC3 and EGI-1 carcinoma cells more prone to activate CAR T cells, as evidenced by an increased percentage of Ca²⁺-responding CAR T cells with IFNγ-treated tumor cells (Fig. 6E, right). EGFR CAR T cells plated onto slices from ICAM-1⁺ human renal tumors also exhibited Ca²⁺ responses, unlike untransduced T lymphocytes (Supplementary Fig. S4; Supplementary Movie S9). This supports the idea that ICAM-1 expression induced in tumor cells by activated T cells promotes T-cell enrichment in tumor islets.

We confirmed this two-step process (initial CAR T-cell peripheral activation and the subsequent interaction between T cells and tumor cells) using fresh BxPC3 tumor slices. In line with our previous results, EGFR CAR T cells were preferentially enriched at the tumor–stroma junction rapidly after adding EGFR CAR T cells to BxPC3 tumor slices (Fig. 7A). Twenty hours later, most CAR T cells were found within tumor islets that expressed high ICAM-1. We also found that EGFR CAR T cells at 20 hours exhibited increased ability to kill tumor cells, as evidenced by a pronounced caspase-3 activity monitored in BxPC3 tumor slices using a fluorescent dye (Fig. 7B).

In this study, advanced imaging techniques applied to different tumor models, including fresh human tumor slices, allowed us to unravel major determinants that regulated the interaction and activation of CAR T cells within tumors. Imaging approaches have been previously used to explore the kinetics of CAR T-cell activation and immune synapses formed with targets cells (14–18). However, most analyses were performed with individual tumor cells and, therefore, were in the absence of the architecture found in human carcinomas (compact tumor islets surrounded by a stroma). Herein, using tumor models mimicking carcinoma structures, we demonstrated that EGFR CAR T-cell infiltration into tumor islets was regulated by a two-step process (Fig. 7C). An initial CAR T-cell activation at the tumor cell periphery is followed by a progressive enrichment into the center. The first step is an increase in ICAM-1 expression by tumor cells induced in an IFNγ-dependent pathway by initial CAR T-cell activation at the periphery of the islet. The second step is the progressive interaction of CAR T cells with tumor cells in tumor islets.

Our results, obtained in two in vitro carcinoma cell line models and confirmed in ex vivo tumor slices from human lung and renal carcinomas, indicated that the spatial organization of carcinoma cells is important and dictates the responsiveness of EGFR CAR T cells. Only tumor cells located at the periphery of tumor islets are initially permissive to CAR T-cell interaction and activation. We found that, unlike EGFR, which was homogeneously distributed, two proteins, CD104 and CD49f, that form the α4β6 integrin were enriched at the outer edge of tumor islets. Our first attempts to target CD104 with a blocking antibody suggest that this protein might participate in the initial EGFR CAR T-cell activation during interactions with a subset of cancer cells that express CD104. Neoplastic cells within individual carcinomas often exhibit considerable phenotypic heterogeneity in their epithelial versus mesenchymal-like cell states. Interestingly, CD104 has been shown to be preferentially expressed on cancer stem cells harboring a mesenchymal phenotype (19). Based on our results, we propose that these cancer cell subsets with mesenchymal features, located at the periphery of tumor islets, are more prone to stimulate CAR T cells through the expression of CD104 and eventually other proteins. The confirmation of the role of CD104 in this process and the identification of its ligand at the T-cell surface will be an important topic to address in the future.

We have shown in several models that ICAM-1 expression on malignant cells (both leukemias and carcinomas) regulates the interaction of CAR T cells with their targets. Our results are in line with previous studies demonstrating the importance of ICAM-1 in the formation of a productive synapse between T cells and either antigen-presenting cells or target cells or between CD19 CAR T cells and malignant B cells from acute lymphoblastic leukemia patients (20–23). EGFR CAR T cells were not activated and unable to interact with ICAM-1–low expressing carcinoma cells located at the center of the islets. Such tumor cells had a high EGFR expression. Thus, the expression of the target antigen on carcinoma cells was not sufficient to activate CAR T cells. Conversely, the expression of ICAM-1 alone was not sufficient to arrest CAR T cells on target cells. These results are in agreement with those obtained from nonmodified T cells. Highly motile T cells plated on phospholipid bilayers containing ICAM-1 decelerate upon antigen recognition (24). We know now that this stop signal is dependent on Ca²⁺ increases and on the “inside-out” signaling process emanating from the TCR to activate LFA-1 (25, 26). Given the distinct nature of CAR constructs, a key open question pertains to the ability of CAR T cells to properly activate integrins upon tumor cell recognition. This is especially relevant in regard to studies showing that many aspects of CAR signaling are unique and distinct from endogenous TCR signaling (27).

Downregulation of adhesion molecules, including ICAM-1, on tumor-associated endothelial cells is an effective mechanism by which tumors may prevent immune cell trafficking into tumors (28–30). Based on our data and others, we propose that the absence of ICAM-1 on tumor cells, especially in noninflamed tumors, is an element that has to be taken into account in the therapeutic failure of CAR T cells observed in a number of malignancies. Therefore, ICAM-1 expression on tumor cells should be considered as one of the predictive markers for CAR T-cell efficacy in human tumors. Our data also indicated that when abundant, activated CAR T cells can overcome an initial low density of ICAM-1 by inducing its expression on tumor cells through the production of IFNγ. In other words, large enough numbers of CAR T cells can convert an immune cold into a hot tumor that then is permissive to lymphocytes interacting with cancer cells in tumor nests. Interestingly, studies have demonstrated that IFNγ produced by CAR T cells reprograms host immune cells to reinforce T-cell antitumor activities (31, 32).

Our results showing a key role of CAR T cell–induced IFNγ in ICAM-1 upregulation by cancer cells can be altered in a number of different ways in solid tumors. The amount of IFNγ produced by CAR T cells is dependent on T-cell ability to reach tumor cells. However, it is well established that the tumor microenvironment can suppress migration of T cells, diminishing the formation of T cell–tumor cell conjugates. Our study supports the rationale of combining CAR T cells with strategies that target stromal features, e.g., a dense extracellular matrix and TAMs, to improve T-cell migration (33).

The control of IFNγ production, and the response of cancer cells to this cytokine, are other aspects to take into account. Having engineered T cells producing high IFNγ upon target cell recognition is considered ideal for ICAM-1 upregulation. In that respect, T cells expressing CARs with a CD28 costimulatory domain have been shown to release higher quantities of IFNγ than T cells expressing 4-1BB–costimulated CARs (34). A comparison of the effects of CD28 and 4-1BB costimulatory domains in CAR T-cell activation and interaction with tumor cells will be an important area for the future. However, a high amount of IFNγ can also have a negative influence on CAR T cells due to the increased expression of immunosuppressive molecules like PD-L1 (35). To this end, PD-1/PD-L1 blockade may be an effective strategy for improving the potency of CAR T-cell therapies (36). Finally, not all tumor cells respond to IFNγ due to the presence of molecular aberrations in the IFNγ signaling pathways (37). Therefore, the possibility to increase ICAM-1 expression in an IFNγ-independent manner is of great interest. In this regard, systemic thermal therapy has been shown to be associated with IL6 production, leading to ICAM-1 expression by tumor blood vessels and, thus, favoring the entry of T cells into the tumor site (38).

J. Mittelstaet reports grants from European Union's Horizon 2020 research and innovation program under grant agreement number 667980 during the conduct of the study. N. Mockel-Tenbrinck reports grants from CARAT during the conduct of the study. A. Kinkhabwala reports grants from Miltenyi Biotec during the conduct of the study. D. Damotte reports grants from AstraZeneca/MedImmune and nonfinancial support from Roche outside the submitted work. M. Alifano reports personal fees from AstraZeneca outside the submitted work. A.D. Kaiser reports grants from Horizon 2020 European Union funding for research and innovation and personal fees from Miltenyi Biotec B.V. and Co. KG during the conduct of the study. No disclosures were reported by the other authors.

C. Kantari-Mimoun: Conceptualization, formal analysis, investigation, writing–original draft. S. Barrin: Formal analysis, investigation, methodology. L. Vimeux: Formal analysis, investigation, methodology. S. Haghiri: Formal analysis, investigation, methodology. C. Gervais: Formal analysis, investigation. S. Joaquina: Formal analysis, investigation. J. Mittelstaet: Resources, methodology. N. Mockel-Tenbrinck: Resources, methodology. A. Kinkhabwala: Formal analysis, investigation, methodology. D. Damotte: Resources, investigation. A. Lupo: Resources, investigation. M. Sibony: Resources. M. Alifano: Resources. E. Dondi: Resources, investigation. N. Bercovici: Resources, investigation, writing–review and editing. A. Trautmann: Investigation. A.D. Kaiser: Conceptualization, resources, investigation. E. Donnadieu: Conceptualization, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration.

The authors wish to thank Pierre Bourdoncle and Thomas Guilbert of the Cochin Imaging Facility (Institut Cochin, Paris) for advice and assistance with microscopes and help in data analysis; Fathia Ouaaz and Elisa Peranzoni for valuable discussions and critical reading of the manuscript; Marion Guérin for help with phospho-STAT1 staining and critical reading of the manuscript; Sandra Dapa for help with lentivirus production; Maximilian Wichert and Salomé Carcy for help with imaging experiments; Florence Cymbalista and Vincent Levy for providing CLL B cells; and Laura Fouassier for providing human cell lines. This study was supported by funding from the European Union's Horizon 2020 research and innovation program under grant agreement number 667980 and by the French Ligue Nationale Contre le Cancer (Equipes labellisées).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.