Tissues, PBMCs, Cell lines and culture conditions
High grade serous ovarian cancer patient samples were provided by Prof. Dr. Peter Mallmann (Department of Obstetrics and Gynecology Medical Faculty, University of Cologne), pancreatic ductal adenocarcinoma were provided by Prof. Dr. Philipp Stroebel (Institute of Pathology, University Medical Center Göttingen), and glioblastoma samples were provided by Prof. Dr. med. Wolfgang Brück (Institute for Neuropathology, University Medical Center Göttingen) or purchased from BioIVT and used with informed consent by the patients. Fresh frozen healthy organ tissue samples were purchased from BioIVT and ProteoGenex. All pancreatic ductal adenocarcinoma (PDAC) patient derived xenografts (PDX) were obtained from Charles River Discovery Research Services Germany GmbH.
The mouse frozen embedded tissue section is derived from a frozen C57BL/6 mouse spleen provided by the Czech Centre for Phenogenomcis by PD Dr. rer. nat. Radislav Sedlacek and is part of a control group of a bigger experiment characterising mouse mutants using the MICS system which will be published spearately.
Buffy coats and leukaphereses were obtained from the University Hospital in Cologne and Dortmund. Peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats by density gradient centrifugation. Before imaging, PBMCs were washed 3x in PBS and transferred to multiple wells of a glass-bottomed (170 µm coverslip) 24-well plate (with 1x106 PBMCs /well). They were then centrifuged at 1000xg for 10 min (RT), fixed for 10 min using 4% paraformaldehyde (PFA), and finally washed (3x) with PBS buffer.
HEK293T and A2780 cells were purchased from American Type Culture Collection (Manassas, VA) and cultured in DMEM (Biochrom, Nuaillé, France) supplemented with 2 mM glutamine (Lonza, Basel, Switzerland) and 10% FCS (Biochrom, Berlin, Germany). Cell confluency ranged typically between 20–80% during the culture maintenance phase. All human primary cellular products were derived from healthy donors after informed consent and cultivated in TexMACS medium (Miltenyi Biotec, Bergisch Gladbach, Germany) following processing.
Tissue were dissociated using the Tumor Dissociation Kit, human in combination with the gentleMACS™ Octo Dissociator with Heaters (both Miltenyi Biotec). When processing PDX models, mouse cells were depleted using the Mouse Cell Depletion Kit (Miltenyi Biotec). Resulting cell suspensions were analyzed using the MACS® Marker Screen, human (Miltenyi Biotec) a monoclonal antibody panel containing 371 pre-titrated antibodies with 9 isotype controls, or candidate antibodies selected from this panel for subsequent screening steps. All samples were measured on MACSQuant® Analyzer and analyzed using the MACSQuantify™ Software or FlowJow v10.7.1.
Tissue processing for microscopy
Frozen embedded tissue specimen were cryosectioned with a CM3050 cryostat (Leica), 8 µm sections were mounted on SuperFrost® Plus slides (Menzel). For acetone fixation the sections were fixated in -20°C cold acetone for 3 minutes prior to the slide storage at -80°C. On the day of use, the slide was immerged into in -20°C cold acetone for 10 minutes for thawing. After short air drying the appropriate MACSWell™ imaging frame was mounted immediately on the slide and the appropriate initial sample volume of MACSima Running Buffer was added (according to the MACSWell™ imaging frames datasheet). For paraformaldehyde (PFA) fixation, the cryosectioned slices on slides were directly stored at -80°C. On the day of use, the frozen slide was put in a 4% PFA solution and incubated for 10 minutes at room temperature. The slide was washed three times with MACSima Running Buffer. After washing the appropriate MACSWell™ imaging frame was mounted immediately on the slide and the appropriate initial sample volume of MACSima Running Buffer was added (according to the MACSWell™ imaging frames datasheet). Right before the start of the MACSima TM instrument a DAPI pre-staining was performed: the MACSima Running Buffer was removed from the sample to be analysed and stained for 10 min with a 1:10 dilution of a DAPI staining solution (volume depends on working volume for the different MACSwell™ formats, see datasheet). The DAPI staining solution was removed and 3 washing steps were performed (MACSima Running Buffer). Finally, the initial sample volume of MACSima Running Buffern was added.
Antibodies and conjugates for microscopy
The following three types of reagents were used in this work for iterative staining and imaging with MACSima: Fluorescently labelled antibodies, REAdyelease and REAlease staining reagents. Fluorescent antibodies were prepared by chemical linking of one or more fluorescent dyes to a corresponding recombinant (REAfinity™) antibody. Recombinant antibodies are derived from a defined set of genes, ensuring the consistency in antibody structure and performance. A typical example of recombinant fluorescent antibody used in this work is CD4-FITC (CD4 Antibody, anti-human, FITC, REAfinity™, 130-114-531). REAdyelease reagents were prepared by covalent linking and fluorescent labelling of two or more antibody fragments. The chemical linker between the antibody fragments and fluorescent dyes can be disrupted by a specific releasing reagent leading to release of fluorescent dyes and resulting in erase of fluorescence staining. A typical example of REAdyelease conjugate used in this work is CD11c-PE (CD11c Antibody, anti-human, PE, REAdye_lease™ 130-121-314). REAlease reagents were prepared by covalent linking and fluorescent labelling of two or more antibody fragments that are characterized by low epitope binding affinities. Due to multimerization of antibody fragments, REAlease conjugates possess high-avidity and are comparable to conventional antibody-fluorochrome conjugates in terms of labeling properties. The chemical linker between the antibody fragments can be disrupted by a specific releasing reagent. This leads to monomerization of the antibody fragments and their dissociation from the epitope resulting in erase of fluorescence staining and availability of the epitope for restaining. Typical examples of REAlease conjugates used in this work are CD8-FITC (CD8 Antibody, anti-human, FITC, REAlease®, 130-112-070) and CD19-FITC (CD19 Antibody, anti-human, FITC, REAlease®, 130-112-073). To characterize the expression of proteins in the murine spleen and on on cancerous and healthy human tissues antibody panels were designed. A list of all antibodies used in the respective experiment is given in Supplementary Table 1.
Cyclic immunofluorescence staining with the MACSima™ Imaging System
The MACSima™ Imaging System is a fully automated instrument combining liquid handling with widefield microscopy for cyclic immunofluorescence imaging. In brief, staining cycles consisted of the following automated steps: immunofluorescent staining, sample washing, multi-field imaging, and signal erasure (photobleaching or REAlease). Cyclic immunofluorescence with the MACSima is optimally applied on thin tissue cryosections (few micron thick), cultured cells, or suspension cells (either captured in microcavities or centrifuged down onto the glass). We developed MACSwell sample carriers specifically to provide the reaction cavities necessary to perform MICS experiments with the MACSima Imaging Platform. To support either tissue sections of varying size or adherent cells we designed different kinds of sample carriers: MACSwell One, Two, and Four Imaging Frames, and MACSwell 24 Imaging Plates.
Liquid Handling: The liquid handling system is comprised of a syringe pump, peristaltic pumps, a valve head, tubing, and a robotic needle. The valve head connects the needle with the externally mounted buffer bottles (MACSima™ Running Buffer, MACSima™ Storage Solution, MACSima™ System Buffer) and the waste bottle. The valve head allows for switching of the syringe pump to each of the bottles and the needle. To avoid carryover between pipetting steps, needle washing is performed by flushing the needle from both inside and outside with fresh buffer. Additional peristaltic pumps transport the used fluids to the waste bottle. An internal reagent bank has positions for other reagents that can be accessed with the needle (e.g. DAPI, FcR block, REAlease™ Release Reagent). The liquid handling system automates in particular the following steps: (1) preparation of the antibody conjugate staining solution (e.g. resuspension of dried antibody conjugates, dilution/mixing of conjugates), (2) application of the staining solution or the REAlease™ Release Reagent to the biological sample residing in a specific well of a disposable sample plate, and (3) sample washing following the staining or release incubation steps.
Microscope: All microscope images on the MACSima are obtained using an epifluorescence setup with one of three different objectives: 20x objective (NA 0.75, 170-micron coverslip glass), long-working-distance 20x objective (NA 0.45, objective slides 1.0 mm), and 2x (NA 0.1). Fluorescence excitation is achieved with custom-designed illumination based on a set of LEDs (infrared, red, green, blue, UV) with LED-specific filters to narrow the excitation light spectrum. Additional excitation/dichroic/emission filter sets define the epifluorescence channels optimized for standard dyes (e.g. DAPI, FITC, PE, APC, and Vio780). Images are captured by a monochromatic scientific CMOS (SCMOS) camera with 106 nm/pixel for the 20x objectives and 1060 nm/pixel for the 2x objective. Autofocusing is achieved in two ways. A hardware autofocus measures the position on the glass surface using infrared light to a precision of < 1 µm. Image-based autofocusing is also possible via optimization of the DAPI image.
Photobleaching: Photobleaching is achieved by focusing an additional set of red, green, and blue LEDs onto a single square spot (3 mm x 3 mm). Illumination with the blue LED (2 W/cm2 at the sample; 2 min), green LED (0.4 W/cm2 at the sample; 2 min) or red LED (1 W/cm2 at the sample; 6 min) led to a > 90% reduction of the FITC, PE or APC intensity.
Stage: Reagent plates and sample holders are mounted on an xy-stage (with positional accuracy on the order of a few µm) to align the biological sample with the optical path of the microscope or with the photobleaching position, as well as to bring the reagents and sample to the needle position.
Image Acquisition and Processing: The image processing pipeline for the MACSima is displayed in Fig. S2. In the first step, a series of single-channel exposures of a fixed position in the sample are combined into a single, statistically optimal, high dynamic range (HDR) image based on a calibrated Gaussian noise model for the IRIS 15 SCMOS camera. Our use of an HDR representation for the images allows for the removal and replacement of all saturated pixels and additionally boosts signal-to-noise in the dimmer portions of the image. In the next steps, the HDR image is corrected in each pixel for the sensor flatfield (pixel-to-pixel differences in quantum efficiency), the optical profile (illumination/detection gradient across the field of view), and hot/cold outlier pixels (by median filtering). Non-local corrections are then performed over the image to correct for distortion (including chromatic effects), to register the images over the cycles, to stitch neighboring (overlapping) images together, and to downsample to the Nyquist frequency. Spectral unmixing of the image (along with corresponding images obtained for the other channels) is then performed based on a calibrated crosstalk matrix. A final subtraction of the pre-stain image removes any residual intensities from remaining autofluorescence or incomplete erasure of the previous cycle staining. While not all images presented in this publication were analyzed with the full pipeline as described above, the most critical processing steps corresponding to the removal of saturated pixels, replacement of hot/cold pixels, flatfielding, and subtraction of the pre-image were applied to all data sets presented here. Further details of the image processing pipeline will appear in a separate publication.
Image data sets (stack of images) for each tissue were imported into the software QiTissue™. The software uses nuclei and cell membrane markers to perform image segmentation identifying individual cells. As it uses all the cell membrane markers known to the QiTissue™ system simultaneously, it can segment most types of cells in a sample. For each tissue similar segmentation parameters were used. Once the cells are identified, the features like mean fluorescent intensities (MFI) are computed for each cell against the background. These intensities are then used for further downstream processing. During the downstream analysis, all computed MFI are scaled between the range 0 and 1 for visualization and comparability.
Fluorophore-labeled beads were used to compare the linearity and sensitivity of image cytometry by the MACSima with flow cytometry by the MACSQuant in the following way. Compensation beads from the MACS Compensation Bead Kit, anti-REA (130-104-693) were incubated with different stoichiometric mixtures of CD4-APC (130-113-222) and dark CD4-Biotin (130-113-224) antibodies (clone REA623). The anti-REAfinity antibody conjugated to the bead surface generically recognizes Miltenyi REAfinity antibodies through their kappa light chains, with the specific CD4-recognition domain here playing no role. Saturation of CD4-APC antibodies on the beads was determined from a titration series to occur at roughly 10 µg/mL (95% saturated). Compensation beads were diluted (1:2) and incubated in the dark (10 min, RT) in MACSQuant Running Buffer (130-092-747) to a final volume of 400 µL with stoichiometric mixtures of CD4-APC to CD4-Biotin antibodies at overall saturating conditions equivalent to the following percentages of labeled probe: 100, 10, 1, 0.33, 0.10, 0.030, 0.010, 0.0030, 0.0010 (see Fig. 2a). Samples were shaken during incubation and diluted afterwards with 1 mL of the buffer to stop the incubation process. The beads were then centrifuged at 300xg for 5 min to remove unbound antibodies and washed 1x in the original volume of 400 µL. Half of the volume of each of the samples was transferred to a well of a glass-bottomed (170 µm coverslip) 24-well plate (~ 1x106 beads/well) for image cytometry with the MACSima and the other half was used for flow cytometry with the MACSQuant.
For flow cytometry by the MACSQuant, beads were gated in the software MACSQuantify v.2.13.0 using the following gating strategy. (1) Draw an ellipse gate around the densest region of beads in a FSC-A vs. SCC-A density plot (A = area of detection peak, FSC = forward scatter, SSC = side scatter). (2) Draw a polygon gate around the linear fit events in an FSC-A vs. FSC-H plot (H = height of peak) to purify for singlets. (3) Draw an interval gate around the bead distribution in an APC-A histogram plot.
For image cytometry of the beads by the MACSima, fifteen separate positions in each well were imaged (see cropped example image in Fig. 2a inset) with the images spaced far enough apart (by 1 mm) from one another to avoid inter-image photobleaching during acquisition. Several hundred beads were acquired per position with, consequently, several thousand total beads imaged for each well (each assayed percentage, see Fig. 2a). Imaging consisted of an exposure series (60, 160, 640 ms) in the APC channel (total acquisition photobleaching was < 5%) followed by a single exposure (1500 ms) in the DAPI channel of the weak bead autofluorescence, with the latter image used to create masks around each detected bead that were therefore independent of the degree of APC staining.
The brightest unsaturated image in each APC channel image series was processed in the following way: (1) Subtract camera readout offset. (2) Flatfield by dividing by a readout-offset-subtracted image of APC in solution. (3) Divide by exposure time.
For object mask creation, the bead autofluorescence image was analyzed in ImageJ/FIJI as follows: (1) Subtract camera readout noise. (2) Perform Gaussian blur with σ = 3 pixels. (3) Perform “Auto Local Threshold” with “Median” filter and radius = 15 pixels to obtain binary image. (4) From “Morphological Filters”, dilate as a disk with radius = 2 pixels. (5) Fill holes. (6) Perform watershed. (7) Label all particles with area greater MIN = 4000 and less than MAX = 12000 pixels, and with circularity greater than 0.1. (8) Save object masks.
To remove background in the APC channel from the object mask pixels, the background was interpolated in the following way: (9) Return to the mask resulting from “(5) Fill holes” above. From “Morphological Filters”, dilate as a disk with radius = 17 pixels and then invert to create a mask for the background. (10) Mask the processed APC channel image to reveal only pure background regions. (11) Using a custom-written Python routine, place a uniform grid (with non-overlapping quadrants of size 400x400 pixels) over the background image (5056x2960 pixels) to downsample it by performing a median filter over each quadrant with assignment of the median value to the central position of each quadrant. (12) In Python, interpolate over the median values (quadrant centers) to estimate the background contribution to the object mask pixels using “griddata” (OpenCV) with the method “cubic” for bulk pixels followed by the method “nearest” for boundary pixels. (13) In ImageJ, subtract interpolated background image from the initial processed APC channel image. (14) Integrate background-subtracted bead intensities over each object mask and save the results as a table.
To compare the sensitivity of MACSima image cytometry with MACSQuant flow cytometry on a real biological sample, peripheral blood mononuclear cells (PBMCs) were examined as follows. A buffy coat from a healthy anonymous donor was obtained from the German Red Cross Dortmund, and the PBMCs were isolated by density gradient centrifugation. The cells were stained with CD45-VioBlue (130-110-637) as a general marker for PBMCs and were additionally stained with different stoichiometric ratios of a fluorescent CD3-APC antibody (130-113-135) and a non-fluorescent CD3-Biotin antibody (130-113-137). More specifically, a 1.5 mL suspension of fresh PBMCs containing 2x107 cells/mL was prepared using autoMACS Running Buffer (130-091-221). Furthermore, buffer was added to separate Eppendorf tubes along with CD45-VioBlue (1:50) and mixtures of the CD3-APC and CD3-Biotin antibodies at the stoichiometries listed in Table 1 and at overall high saturation (the standard dilution of 1:50 for the CD3 antibodies implies a high degree of saturation of the targeted CD3 epitopes), considering a 1:3 cell suspension dilution and a final volume of 300 µL. Then, 100 µL of the prepared PBMC suspension was added to the 200 µL contained in each separate tube. Cells were stained in the dark and shaken during incubation (10 min, RT). Afterwards, samples were diluted with 1 mL of buffer to stop the incubation process. The cells were then centrifuged at 300xg for 5 min to remove unbound antibodies, resuspended in 200 µL of 4% paraformaldehyde (PFA) and fixed for 10 min in the dark. Subsequently, cells were centrifuged at 600xg (fixed cells are lighter) for 5 min to remove PFA, and washed 1x in 300 µL of buffer, with half (150 µL) used for MACSima image cytometry and the other half for MACSQuant flow cytometry.
For the MACSQuant, CD45 + cells were gated using density plots according to the following strategy. (1) Draw a polygon gate around the linear fit events in a FSC-A vs. FSC-H plot. (2) Disregard platelets, red blood cells, and debris (low FSC-A signals) with a rectangular gate drawn in a FSC-A vs. SSC-A plot. (3) Select CD45 + cells with a rectangular gate in a CD45-VioBlue-A vs. SSC-A plot. Then, assess the selected cells for their CD3-APC-A (integrated) intensities.
For the MACSima, 150 µL of the cell suspension from each tube was transferred to a well of a glass-bottomed (170 µm coverslip) 24-well plate and centrifuged at 100xg for 10 min to stick them to the bottom of the well. Image acquisition was identical to that used for the beads, with the exception that fewer cells were acquired per field. Image analysis was also largely identical to that used for the beads (see Bead Experiments above), aside from using the CD45-VioBlue staining to create the cell masks instead of autofluorescence, omission of the watershed, and allowance for objects with areas up to MAX = 15000 pixels. For both MACSQuant and MACSima datasets, the individual integrated cell intensities, I, were converted by the following formula to an arcsinh scale that is asymptotically equivalent to log10, but importantly also allows for negative values:
y = arcsinh(aI - b)/ln(10),with the normalization factors a and b independently chosen for each dataset such that the mean (in arcsinh units) of the blank distribution equals 0 and the mean of the highest stained population equals 2 (see Fig. 2c).
Gene synthesis in combination with an optimization algorithm for codon usage in humans (ATUM, Newark, CA) was used to design genes of interest. E. coli DH5alpha were used for cloning and plasmid generation. Plasmids were purified using DNA isolation kits (Quiagen).
Lentiviral vector production
Second generation self-inactivating VSV-G-pseudotyped lentiviral vectors were produced by transient transfection into adherent HEK293T cells. One day before transfection, 1.6x107 HEK293T cells were seeded per T175 flask. Each T175 flask was then transfected with a total of 35 µg plasmid DNA composed of pMDG2 (encoding VSV-G), pCMVdR8.74 (encoding gag/pol), and the respective transgene-encoding transfer vector using MACSfectin reagent (Miltenyi Biotec). All transfection reactions were performed with a DNA:MACSfectin ratio of 1:2. Following overnight incubation, sodium butyrate (Sigma-Aldrich) was supplemented at a final concentration of 10 mM. After 48 hours the medium was collected, cleared by centrifugation at 300xg and 4°C for 5 min and filtered through 0.45 µm-pore-size PVDF filters. Concentration of the viral stock was performed by centrifugation at 4°C and 4,000xg for 24 hours. Pellets containing lentivirus were air-dried and resuspended at a 100-fold concentration with 4°C cold PBS. Lentiviral aliquots were stored at 80°C. In order to generate EPCAM positive A2780 cells human EPCAM lacking intracellular domains was cloned in the transfer vector. To generate eGFP positive A2780 cells a cassette containing eGFP was cloned in the transfer vector.
A2780 THY1 knockout cells were generated in-house by CRISPR/Cas9-mediated knockout and subcloning. The respective THY1 gRNA used was GACGAAGGCTCTGGTCCACT. A2780 cells were electroporated at a density of 10^5 cells/ml in a CliniMACS Prodigy electroporator. Ribonucleoprotein particle complexes were assembled according to the manufacturer’s protocol (IDT).
Isolation of T cells and generation of CAR T cells
Peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats or leukapheresis products by low-density centrifugation on Pancoll (Pan-Biotech, Aidenbach, Germany) and enriched for Pan T cells by negative magnetic selection (Miltenyi Biotec). The enriched Pan T cells were resuspended at a density of 1x106 cells/mL in TexMACS medium containing IL-7/15 and stimulated with T cell TransAct (all from Miltenyi Biotec). One day after activation, T cells were lentivirally transduced with a multiplicity of infection (MOI) of 1:5 with the respective concentrated vector. Cell numbers were determined every 2–3 days and fresh TexMACS medium supplied with IL-7/15 was added to maintain a cell concentration of 1x106 cells/mL until day 12. Analysis of CAR-expressing cells was routinely performed on day 6 after activation using flow cytometry and CAR detection via Biotin-PE. Afterwards, cells were expanded in the presence of IL-7/15until downstream processing.
CAR T cell killing assays
All functionality assays were performed with CAR T cell populations on day 12–14 following T cell TransAct activation and in TexMACS medium without additives. The frequency of CAR-expressing T cells was equalized before all functional assays and untransduced T cells served as a control for allogeneic reactivity. To analyze CAR T cell-mediated cytotoxicity by target cell death, real-time monitoring using the IncuCyte S3 system (Essen BioScience, Ann Arbor, MI) was performed. Target cells were harvested using Trypsin (3min, 37°C) and cell numbers as well as target expression (THY1 REA897-APC, EPCAM REA764-Vioblue; staining was performed according to the manufactures protocol) were determined at MACS Quant 10. Afterwards target cells were seeded to ensure proper target cell adherence.
Assays were set up by seeding 25,000 GFP-transgenic target cells per well of flat bottomed 96 well plates and – following overnight incubation – adding CAR T cells to the cultures at a 1:2 (E:T) ratio. Subsequently, adapter molecules (THY1 REA897-Biotin, EPCAM REA764-Biotin) were supplemented at varying doses. Cultures of target cells only, target cells with mock-transduced T cells, and target cells with adapter CAR T cells without adapter molecule or unspecific adapter molecule were used as controls. Phase contrast and green fluorescence images were captured with 10x magnification every two hours for 3–6 days. Analysis of images was performed using the software provided by the manufacturer.
For all studies using human primary tissue from glioblastoma, ovarian or pancreatic cancer, written informed consent was obtained following the guidelines of the approved Universitätsmedizin Göttingen, and Cologne Review Board protocol, respectively.
Peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats of healthy anonymous volunteers that were purchased from the German Red Cross Dortmund. All blood samples were handled following the required ethical and safety procedures.
All animal experiments were approved by the Governmental Review Committee on Animal Care in NRW, Germany and performed according to guidelines and regulations. Animals were maintained under specific pathogen-free conditions according to the recommendations of the Federation of European Laboratory Animal Science Association. All procedures were carried out in accordance with the European Communities Council Directive European Communities Council (86/609/EEC) and (2010/63/EU).
Healthy whole blood samples were taken from voluntary healthy donors that gave their written consent before.