Article Text
PDF
Abstract
Background To accelerate the translation of novel immunotherapeutic treatment approaches, the development of analytic methods to assess their efficacy at early in vitro stages is necessary. Using a droplet-based microfluidic platform, we have established a method for multiparameter quantifiable phenotypic and genomic observations of immunotherapies. Chimeric antigen receptor (CAR) natural killer (NK) cells are of increased interest in the current immunotherapy landscape and thus provide an optimal model for evaluating our novel methodology.
Methods For this approach, NK cells transduced with a CD19 CAR were compared with non-transduced NK cells in their ability to kill a lymphoma cell line. Using our microfluidic platform, we were able to quantify the increase in cytotoxicity and synaptic contact formation of CAR NK cells over non-transduced NK cells. We then optimized our droplet sorter and successfully used it to separate NK cells based on target cell killing to perform transcriptomic analyses.
Results Our data revealed expected improvement in cytotoxicity with the CD19 CAR but more importantly, provided unique insights into the factors involved in the cytotoxic mechanisms of CAR NK cells. This demonstrates a novel, improved system for accelerating the pre-clinical screening of future immunotherapy treatments.
Conclusions This study provides a new potential approach for enhanced early screening of immunotherapies to improve their development, with a highly relevant cell model to demonstrate. Additionally, our validation studies provided some potential insights into transcriptomic determinants influencing CAR NK cytotoxicity.
- chimeric antigen receptor - CAR
- natural killer - NK
- lymphoma
Data availability statement
Data are available in a public, open access repository. Sequencing data is available in the National Center for Biotechnology Information's Gene Expression Ombinbus (GEO), accession number GSE243787.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
The efficacy of lymphocytes modified with a CD19 chimeric antigen receptor (CAR) has been previously demonstrated to be effective for the treatment of B cell lymphomas, in both preclinical and clinical settings. It is also understood that natural killer (NK) cells and other lymphocytes are heterogeneous, and can present a range of phenotypes even when undergoing a single culturing method. Additionally, droplet microfluidic methods have previously been demonstrated to allow multiomic analysis at the single-cell level.
WHAT THIS STUDY ADDS
This study presents a novel CD19 CAR construct design in a cytokine-activated NK cell. Most previous work has been in CAR T cells, and each novel CAR design may provide unique benefits. Additionally, this study presents unique findings of potential genetic factors influencing the ability of CAR and non-transduced NK cells to kill tumor cells. And most critical to this study’s findings, a novel approach for improved characterization of immunotherapies using droplet microfluidics is demonstrated.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study provides a new strategy to better understand and improve the quality of immunotherapies by allowing genomic analysis based on functional characteristics. Using this methods, other researchers may be able to enhance the quality of their novel immunotherapies by identifying the genomic or proteomic factors influencing the efficacy of their treatment.
Background
In recent years the development of novel immunotherapeutic strategies for cancer has significantly accelerated. While this progress is exciting and ensures the continuous development of improved cancer therapies, the sheer number of new immunotherapies underscores a critical need to develop methods to identify more effective treatments during their preclinical development.1 2 Understanding factors that determine patient responses and developing novel methods for selecting more potent immunotherapeutic treatments before they reach the clinical trial stage would expedite the development of more effective clinical products. Single-cell platforms are particularly effective as a means of generating quantifiable data reducing extraneous variables that can influence results. An isolated experimental environment can increase reproducibility of results.3 Microfluidic devices provide an optimized method of observing as well as potentially identifying key factors impacting the efficacy of cellular immunotherapy at the single-cell level.4–9 However, these platforms often fall short when it comes to evaluating cell–cell interactions, as such studies typically necessitate the analysis of hundreds to thousands of cells to yield substantial and meaningful data. 10–12 In addition to the inherent capabilities of these devices, such as assessing theraputic efficacy and a more fundamental analysis of biological processes in vitro, microfluidics can also be used to isolate individual cells of interest for downstream analysis. Recent developments in microfluidics have incorporated more complex mechanisms of single-cell trapping, such as dielectric fields,13 surface acoustic waves,14 magnetic fields,15 and optical tweezers.16 While these more advanced technologies provide notable cell selection and analysis potential, their complexity tends to necessitate lower-throughput designs. These platforms are suboptimal for conducting downstream cellular studies such as sequencing or mass spectroscopy as they do not have the throughput required to generate meaningful results.
Our group (Konry lab) has developed a single-cell, droplet-based microfluidic device for rapid screening of novel immunotherapies.3 17–23 We have used this platform for the characterization of cellular interactions between immune and cancer cells under various treatment conditions in several previous studies. To create a comprehensive in vitro screening approach for novel immunotherapies, we have additionally developed a fluorescence-assisted droplet sorting (FADS) platform. This platform allows for high-throughput sorting based on cellular fluorescence within droplets. The key advantage of this design is its ability to sort based on cellular interactions and functional activity. Unlike other traditional sorting approaches that rely on markers of cellular activity, droplet sorting allows the separation of cells based on their ability to kill a target cell. This provides a considerable benefit for the development of immunotherapies, as it allows for downstream analysis based on effector cell functional cytotoxicity. Using a single-cell droplet format reduces influences from external variables and allows easily quantifiable observations on the effects of immunotherapies. Incorporating advanced techniques such as transcriptomic sequencing or mass spectroscopy can provide novel insights into the mechanisms enhancing or inhibiting cytotoxicity of novel immunotherapeutic approaches, including chimeric antigen receptor (CAR) modified T cells (CAR T cells) and CAR natural killer (NK) cells.
In the current study, we applied these platforms to observe the cytotoxicity and cellular interactions of CD19 CAR NK and tumor cells. CAR NK cells have emerged as a powerful tool that addresses many of the drawbacks of other immune cell therapies including CAR T cells. CAR T therapies have proven highly effective in certain cancers leading to their approval in patients with B-cell and plasma cell malignancies.24–30 However, the use of CAR T cells presents several challenges including the risk of cytokine release syndrome, neurotoxicity, and a potential to cause graft versus host disease necessitating T cell receptor locus knockout when used in an allogeneic setting.26 31–33 NK cells use multiple mechanisms of cytotoxicity, which can potentially help to overcome cancer resistance.25 34 35
In this study, we used both NK92 and peripheral blood-derived primary NK cells, comparing NK cells transduced with a CD19 CAR (with 4-1BB and CD3ζ signaling domains) to the same donor’s non-transduced (NT) NK cells. We used the CD19+ Raji lymphoma cell line as a model target cell for our experiments. We applied our previously established single-cell droplet array platform to observe differences in cellular interactions, including cell cytotoxicity and kinetics of effector-target cell contact.18 21 36 37 We proceeded with the development of a FADS platform to separate effector cells based on their ability to kill target cells. We validated the accuracy of our FADS platform and then successfully applied it for the separation of NT and CAR NK cells that were or were not able to effectively kill Raji cells. This was followed by RNA-sequencing on sorted cells. With our combined time-lapse cytotoxicity analysis and transcriptomic data, we were able to determine the potential efficacy of this treatment as well as unique gene expression profiles of the more effective CAR NK cells. This approach can be easily translated to other immunotherapies and provides a potential tool for greatly enhancing the development of personalized medicine.
Methods
Cell culture and expansion
All cells were cultured at 37°C with 5% CO2. Raji cells were purchased from American Type Culture Collection (ATCC, Manassas, Virginia, USA) and cultured in RPMI 1640 media with 10% fetal bovine serum and 1% Antibiotic Mixture (Gibco, Waltham, Massachusetts, USA). NK92 MI were purchased from ATCC and grown in X-Vivo 15 (Lonza, Peabody, Massachusetts, USA) with 5% human AB serum (Gibco). Human primary NK cells were enriched from leukoreduction system chamber from healthy donors using the RosetteSep Human NK Cell Enrichment Cocktail (StemCell) and expanded using NK MACS expansion media (Miltenyi Biotec, Bergisch Gladbach, Germany) with 500 U/mL of recombinant human interleukin-2 (IL-2) (Miltenyi).
CD19 CAR transduction
To overexpress CAR, NK92 MI and primary NK were transduced using lentiviral infection.38 Lentivirus was produced by cotransfecting HEK293 cells with transfer plasmid pHIV-CD19-CAR-GFP plasmid, packaging plasmid pCMV-Δ8.9, envelope pCMV-BaEV and pAdvantage (Promega) using the TransIT-LT1 transfection reagent (Mirus Bio). Culture supernatants were harvested at 24 hours and 36 hours. Primary NK cells were stimulated with 50 ng/mL IL-15 (Miltenyi Biotec, 130-095-762), 10 ng/mL recombinant human IL-12 (R&D Systems) and 50 ng/mL recombinant human IL-18 (R&D Systems) for 18 hours in RP-10, or with IL-2 (Miltenyi Biotec, #130-097-748) for a few days in NKmacs media (Miltenyi Biotec). Stimulated cells were then transduced with BaEV-pseudotyped lentivirus on RetroNectin-coated plates (Takara Bio) with the addition of Vectofusin-1 (Miltenyi).
Microfluidic sorting and RNA isolation
To sort NK cells based on killing, suspensions of NK and Raji cells were prepared at approximately 3 million cells/mL. CellEvent Caspase 3/7 apoptosis dye (Invitrogen, Waltham, Massachusetts, USA) was added to the suspension of Raji cells to detect killing events. Droplets were generated using the single-cell droplet device and contained NK and Raji at approximately 1:1. Droplets were incubated for 6 hours, then injected into the sorting device, and NK cells that killed Raji were sorted based on fluorescent signal. Fluorescence was detected using a 488 nm laser (Opto Engine LLC, Midvale, Utah, USA) and PMT (Hamamatsu Photonics, Hamamatsu City, Japan). Signal detection and sorting impulse were regulated by a control unit and LabView software (National Instruments, Woburn, Massachusetts, USA), and electronic impulse to the device was amplified with a Trek 609 C-6 High Voltage Amplifier (Advanced Energy, Denver, Colorado, USA). To calibrate the instrument, green fluorescent FluoSpheres (Thermofisher) were delivered through droplets. For all sorting experiments, baseline thresholds for sorting are set at the start of the experiment, to automatically trigger sorting impulse on any fluorescent peak above baseline or background noise. An example of sorting can be observed in online supplemental video 3.
Supplementary video
RNA extraction and transcriptomic sequencings
After collecting droplets from the sorting device, emulsions were removed through gentle pipetting in excess cell media. The solution was allowed to settle, and the aqueous layer was pipetted out and transferred to a new container. Cells were then pelleted, and RNA extracted via RNEasy mini kit QIAGEN, Venlo, Netherlands). Cell pellets were resuspended with Buffer RLT and run through QIAshredder columns (QIAGEN) to homogenize cells and remove any impurities. RNA was then precipitated and cleaned following kit instructions and eluted in 50 µL of RNAse-free water. Samples were then frozen and stored at −80°C. Frozen samples were sent to Novogene (Beijing, China) on dry ice for transcriptomic analysis through their ultra-low RNA input protocol. At Novogene, using Takara SMART-seq V4 kit for amplification and the Illumina NovaSeq platform for sequencing. Data processing and bioinformatics is provided by Novogene. Heatmaps were generated with Novogene’s NovoMagic tool. This software scales normalized read counts by sample, setting the mean read count as 0 and displaying the log2-fold differences between genes. Statistical significance calculated by Novogene, using p value and multiple hypotheses testing to create adjusted p values. Volcano plots and Gene Ontology display log10 p values of gene counts.
Statistical analysis
Analysis of statistical significance was done through GraphPad Prism, using Mann-Whitney U test for contact and Student’s t-Test for cytotoxicity calculations.
Results
Microfluidics platform allows assessment of the increased cytotoxicity and serial killing by CD19 CAR NK cells
Using our single-cell droplet platform, we were able to observe NK cell cytolytic activity, as displayed in figure 1. For our initial experiments, we used the NT and CAR 19 transduced NK92 (CAR NK92) cells to gather initial impressions of cell behavior in droplets and optimize our experimental parameters. We coencapsulated the NK92 cells and Raji in droplets with 1–2 NK cells and 1–4 Raji cells, targeting a 1≤1 effector to target ratio for physiological significance and potential to observe serial killing events. We observed the droplets continuously over 12+ hours in the 4000 droplet array, and selected droplets for analysis based on visual confirmation of E:T ratios (figure 2). As expected, the CAR NK92 consistently killed a higher percentage of the Raji target cells than NT NK92 in each experiment with an average 10% increase (online supplemental figure S1A). When comparing the time of target cell death, no significant differences were observed between the NT NK92 and CAR NK92 (online supplemental figure S1B). This is most likely due to the inherently high cytotoxicity of the NK92 cells, reducing our ability to observe a major advantage of the CAR-transduced cells. While the average time of death was similar between both conditions, a clear subpopulation was visible in the CAR NK time of death data that could kill target cells in the first 2.5 hours of coculture. A similar population did not exist in the NT NK, indicating a subpopulation of NK92 with improved function on CD19 CAR arming. There were also no significant differences observed in synaptic contact times between the NT and CAR NK92 (online supplemental figure S1C). There were a higher number of coencapsulations that did not result in synaptic contact for the CAR NK condition. The small volume and spherical shape of the droplets promote cell–cell contact, suggesting reductions in synaptic contact result from changes in cellular characteristics. It is possible that contact was made between the target and effector cells before imaging began, causing the early death of target cells while no longer in contact. Additional observations were made regarding the efficacy of CAR NK92 by recording encapsulations of single NK92 cells with multiple Raji cells, to assess for serial killing. At 12 hours, CAR NK92 cells demonstrated higher serial killing than the wild type, with approximately 40% of droplets showing serial killing compared with about 22% (online supplemental figure S1D). These results were deemed adequate to proceed with testing primary NT and CAR NK cells.
Supplemental material
Representation of the single-cell droplet analysis workflow. (A) Representation of single-cell droplet pairing, displaying a CAR NK cell inducing apoptosis in a lymphoma cell. (B) Apoptosis of the target lymphoma cell activates the included apoptosis-sensing fluorescent dye. The droplet now contains an effectively killing CAR NK cell and a fluorescent target cell. (C) Droplets containing coencapsulated cells are generated and collected for incubation. After adequate time for cellular events of interest, droplets are recollected and introduced to the droplet sorting platform for fluorescence-based separation. Droplets are sorted based on functional cell activity, as displayed in A and B. (D) Collected droplets are homogenized, and contents extracted for downstream analysis. In this case, RNA is isolated for transcriptomic sequencing. CAR, chimeric antigen receptor; NK, natural killer.
Visual overview of single-cell microfluidic model design and function. 3D Model displaying single cell device (A) during droplet filling and (B) after device is loaded. (C) Stitched image of a region of the droplet array after loading. (D) Microscope image of cell coencapsulation and droplet generating regions of device at 5× magnification. (E–G) 20× images of single cell droplets containing one NK cell (gray) and one Raji lymphoma cell (green) showing (E) start of imaging, prior to first interaction, (F) synaptic contact between cells and (G) death of target Raji cell. (H–J) 20× images of two Raji target cells with one NK to display serial killing. Images represent (H) both target cells alive, (I) one target cell killed and (J) both target cells killed. NK, natural killer.
Primary NK cells demonstrate increased cytotoxicity and target cell contact on arming with a CD19 CAR
After observing modest but reproducible effects with NK92 cells above, we evaluated CD19 CAR-transduced peripheral blood-derived primary NK cells. CAR NK cells were compared with the NT primary NK in droplets coencapsulated with Raji lymphoma cells. As with the NK92, the CD19 CAR NK cells consistently displayed higher target cell killing than the NT primary NK (figure 3A,B).
Single-cell droplet data comparing primary CAR NK and NT NK cytotoxicity and affinity toward Raji lymphoma cells. (A) Increased cytotoxicity of CAR NK versus NT NK at each hour for all experiments combined. (B) Average experimental target cell viability over 12 hours. n=5 for each condition. (C) Time of death for all target cells in each condition. n for NT=447 target cells, CAR=357 cells. Events in red square signify a potential subpopulation of faster killing CAR NK cells. Statistical significance was calculated by Mann-Whitney U test; **p<0.005. (D) Proportion of synaptic contact time between primary NK and Raji cells. n for NT=502 target cells, CAR=398 cells. CAR, chimeric antigen receptor; NK, natural killer; NT, non-transduced.
The greatest increase in the killing of Raji cells by the CD19 CAR NK cells was in the 7-to-10-hour range (figure 3A), with an average of 15% more target cells eliminated than by NT NK cells. These results indicate distinctly less time needed by CAR NK cells to kill the target cells. This was reinforced by the time of death data which displayed a significantly earlier average death time for target cells in the CAR NK group (figure 3C). Additionally, as with the NK92 data, a distinct subpopulation of NK that can kill target cells in the first 2.5 hours was observed in CAR NK cells but not in NT NK cells (figure 3C). In addition to killing faster, CAR NK appeared to form more lasting synaptic contact between the target cells (figure 3D). During the lifespan of target cells in the droplets, primary CAR NK cells had significantly more time in contact with target cells than the NT NK cells. This data indicate that the CAR arming allows recognition of CD19 on Raji cells and keeps these NK cells in closer proximity to their target cells for longer durations thus facilitating their enhanced cytotoxic response.
Interestingly, the primary NK cells did not demonstrate the increase in serial killing that we observed with the NK92 cells. It is likely that earlier time points may reflect a greater difference in CAR versus NT primary NK cells, as the faster killing observed in droplets was expected to correlate with enhanced serial killing as well. This may also indicate that CAR NK cells work best at or above a 1:1 effector-to-target cell ratio. While the primary CAR NK did produce a statistically significant decrease in time required to kill target cells, the overall effect on cytotoxicity was still relatively modest (approximately 10% average increase compared with control NK at the end of 12 hours). Regardless, based on our time of death data, a distinct subpopulation of CAR NK cells that demonstrated enhanced killing efficiency of the target cells appears to exist (figure 3C).
Droplet sorting platform efficiently separates cells based on fluorescent intensity
We next sought to develop a method with which to better understand the novel gene expression profiles that may influence the efficacy of the CAR NK cell killing and that could also be applied to other novel immunotherapies in the future. In order to allow sorting of the immune effector cells based on their cytotoxicity profiles, we have created our own FADS device, using a design developed by Mazutis et al but incorporating several of our own modifications.39 Our microfluidic cellular sorter uses an oil injection port and a droplet injection port for reinjecting previously made intact cell encapsulated droplets (figure 4A). Our channel geometry uses passive filters and S turns to control internal droplet and oil flow velocity while minimizing debris infiltration. The electrode alloy is made from a conductive low melting point indium–bismuth–tin alloy which we inject directly into the channel to create a complete circuit. To control the exact resistance between the collection and waste channels in the sorter, our outlet channels are connected to external pumps. This creates a closed system that prevents pressure fluctuations due to ambient atmospheric pressure shifts and easier control of internal fluid velocity and pressure. This platform is capable of sorting 10,000 droplet samples per hour. The entire system is run in a controlled dark environment to minimize external light from effecting experimental parameters. Our photo multiplier tube converts fluorescent signal into voltage using a focusing electrode and a series of dynodes and anodes. The signal is first filtered through a series of dichroic mirrors which act as filters against the background fluorescent signal of the laser and ambient light, isolating the desired emission spectrum. The threshold of detection for voltage signal can be adjusted in real time, and allows detections of minute levels of fluorescence against the ambient signal, as demonstrated in figure 4F. Integration of a high frame rate camera into the platform allows us to isolate and visualize the internal contents of droplets as they flow at high speeds through the device. When a droplet containing a fluorescent cell passes through the sorting junction, an electrical impulse is passed through the system, creating a dielectric field that pulls the droplet toward the collection outlet (figure 4C–G). After screening a population of droplets, we are able to recollect cells from both the actively sorted collection outlet and passively sorted Waste outlet for performing downstream analyses.
Visual overview of droplet sorting platform design and function. (A) Schematics of droplet sorting device. (B) Imaging showing fabricated droplet sorting device ready for use. (C) Microscope image of droplet sorting junction at 5× magnification. (D, E) Representation of sorter in passive state, displaying (D) baseline fluorescent intensity readings and (E) droplet flow to ‘waste’ channel when no sorting impulse is applied. (F, G) Representation of sorter when activated, displaying (F) signal peak from cells with elevated fluorescent intensity above threshold settings and (G) droplet redirection to ‘collection’ channel when sorter is activated.
To validate our sorter’s accuracy, we loaded droplets containing K562 cancer cells labeled with Cell Tracker CFSE Green (ThermoFisher Scientific, Waltham, Massachusetts). Cells were added at an adequately low concentration to ensure that the majority of droplets would be empty, with approximately 10% or fewer containing cells. After running the droplets through the sorter, we recollected the droplets and imaged them to compare cell concentrations (figure 5). About 89% of droplets observed from the actively sorted collection channel contained cells, while only 12% of droplets contained cells coming from the waste channel (figure 5A). When looking at total cell counts from the imaged droplets after recollection, we observed an average of 367 cells in each image from the collection droplets, and only 19 cells per image in the waste droplets (figure 5B). This disparity can be observed in figure 5C,D, demonstrating example images of the recollected droplets. This result confirmed the accuracy of our droplet sorting platform and allowed us to proceed with developing our method to screen CAR NK cells based on cellular activity. Droplets that passed through the collection outlet are subject to higher pressure and thus more likely to merge, resulting in the larger droplets observed in figure 5C. This merging does not impact the viability or genetic material of the cells contained within the droplets.
Verification of droplet enrichment through cellular fluorescence-based sorting. (A) Bar graph representing average cellular encapsulation in droplets from the collection and waste channels after sorting. (B) Bar graph representing the total cell counts observed in images of droplets from either collection or waste after sorting. ****p<0.001 based on Student’s t-test. (C) Image of droplets from the actively sorted, collection channel, containing a high quantity of fluorescent cells. (D) Image of droplets from passively collected waste outlet, containing sparse cellular encapsulation in droplets.
Sorting and transcriptomic profiling of NK cells based on their cytotoxic activity reveal unique gene expression patterns in killer and non-killer populations
With the development of our FADS platform, we next applied it to further understand the underlying biology determining the NT and CAR NK cells’ ability to kill their targets. The droplet sorting platform allowed the separation of NK cells that could or could not kill target cells, which were subsequently sent out for RNA sequencing. The NK cells and Raji cells were first coencapsulated in droplets and then incubated for 6 hours. After incubation, droplets were withdrawn into tubing using a syringe pump and then injected into the sorting device. Using standard fluorescent apoptosis dyes, the killing of Raji cells produced an increased fluorescent signal, activating the sorter and directing droplets to the collection channel (figure 4). All non-sorted droplets were passively directed to the waste channel. All droplets from both channels were collected, and RNA was extracted as described in the Methods section. Both CAR NK and NT NK were collected through this method and then RNA sequencing was performed on these cells. Sequencing data are publicly available on the National Institutes of Health Gene Expression Omnibus.40 We observed considerable differences in the gene expression between the cells that killed target cells versus those that did not in both CAR NK and NT NK (figure 6). In CAR NK killers versus non-killers analysis, 3927 genes were upregulated, while 3411 genes were downregulated (figure 6A). In NT killers versus non-killers, 5435 genes were upregulated, while only 1722 genes were found to be downregulated (figure 6B). Both CAR and NT NK cell killers showed an increase in surface marker expression and signaling-related pathways, such as ion receptors and transporters, and other extracellular molecules (figure 6C,D).
Post sort gene expression data showing similar expression changes in killer versus non-killer CAR and NT NK cells. (A) Volcano plot comparing gene expression of CAR NK cells that effectively killed target cells versus CAR NK that did not. (B) Volcano plot comparing gene expression of NT NK that effectively killed target cells versus NT NK that did not. (C) GO analysis of upregulated pathways and mechanisms of CAR NK killers compared with CAR NK that did not kill target cells. (D) GO analysis of upregulated pathways in NT NK killers compared with non-killers. BP, biological processes; CAR, chimeric antigen receptor; CC, cellular components; GO, gene ontology; MF, molecular factors; NK, natural killer; NT, non-transduced.
We next pulled out specific genes of interest in various pathways and plotted the gene counts for each condition into heatmaps (figure 7). Looking specifically at markers associated with the NK cell-mediated immune response (figure 7A), we saw CAR NK cells that killed target cells had higher expression of the expected cytotoxic pathways. Granzymes (GZMs), perforin (PRF1), and IFNγ (IFNG) were highly expressed in killer CAR NK cells, which are all components of the traditional CAR NK cell cytotoxic response.41 Genes coding for several apoptosis-inducing ligands were also found at higher expression, including TNSFS10 (TRAIL), TNF, and FASLG (Fas ligand). Expression of several TNF superfamily receptor genes (TNFRSF8, 9, and 18) was also higher in the killer CAR NK droplets. TNFRSF14 gene expression was found higher in the non-killer CAR NK possibly due to the increased presence of live Raji cells in this sample. Both FCGR3A (CD16) and NCAM1 (CD56) genes had higher expression levels in the Killer CAR NK. Comparing gene counts directly from several of the key cytotoxicity-driving genes observed (online supplemental figure S2), expression of genes coding for granzyme A, perforin, and TRAIL was most notably elevated in killer CAR NK cells compared with other conditions. Overall, both CAR NK killer and non-killer populations displayed higher expression of cytotoxic factors than in NT NK cells.
Heatmaps displaying log2 adjusted normalized gene counts of CAR and NT NK cells and Raji cells, sorted based on the ability of NK cells to kill target cells. (A) Genes of interest found related to NK cytotoxicity mechanisms. (B) Genes of interest found related to checkpoint pathways. *Genes potentially expressed in both NK cells and Raji cells. (C, D) Summary of the key biological traits observed using RNA-sequencing data from CD19 CAR NK cells (C) versus NT NK cells (D). CAR, chimeric antigen receptor; NK, natural killer; NT, non-transduced.
Interestingly, for NT NK, the killer NK cells did not have the same upregulation of granzyme and IFNγ as we saw in the CAR NK cells but instead showed notably high expression of several TNF superfamily ligands. Increases in the TNF, TNFSF10, 12, and 15, and FASLG gene expression in the killer NT NK droplets compared with non-killers suggest that cytotoxicity in NT NK cells may be driven predominantly by direct ligand-receptor interactions (figure 7A, online supplemental figure S2).42 We saw an increase in the expression of TNFRSF4, 9, and 14 genes in the non-killer NT NK droplets. It is unclear whether this is due to a higher quantity of intact Raji RNA in these droplets, or whether these pathways were potentially contributing to the Raji cell resistance to NK cell killing. We additionally looked at numerous checkpoint ligands and receptors on Raji and NK cells (figure 7B). We observed higher Ceacam1 and several Human Leukocyte Antigens (HLAs) associated with checkpoint receptor activation, including HLA-DR, DP, DQ, and E in non-killer NT conditions.43 44 While this may be influenced by Raji cell death, this result would correlate well with expectations, given NT NK are not equipped with a specialized receptor and therefore subject to standard regulatory mechanisms. Killer NT NK did display higher levels of the TIGIT and KLRC1 (NKG2A) checkpoint receptors. In the CAR NK non-killer droplets, KLRD1 (CD94), CEACAM1, and NECTIN4 gene expression was much higher than in the killer CAR NK cells. Additionally, CD47 and CD80 gene expression was higher in the CAR NK non-killer cells. A visual representation showing the relevant pathways mediating cytotoxicity based on our RNA-sequencing data in CD19 CAR NK and NT NK cells is depicted in figure 7C, D.
Discussion
Here we established a novel droplet-based in vitro method for in-depth phenotypic and transcriptomic characterization of NK cell CARs at the single-cell level. Using our 4000 droplet array device, we were able to see interactions between NK cells and Raji lymphoma cells at 1:1 or 1:1+ effector-to-target ratios with a high number of replicates. Compared with the NT NK cells, CAR NK cells required less time to kill and CAR arming improved synaptic contact with target cells, thus improving the specificity of the NK cytotoxic response. We also observed a tendency towards improved capacity of single CD19 CAR NK cells to kill multiple lymphoma cells in a short timeframe, which correlated well with the faster killing observed in our droplets. Furthermore, we established a novel droplet-sorting approach for understanding transcriptomic differences based on the functional characteristics of the NT and CAR NK cells. We observed fundamental differences in the potential mechanisms of killing in CAR and NT NK cells based on our transcriptomic data which could have implications beyond CD19 CAR NK cells therapy.
Our novel imaging-based single-cell droplet platform allowed us to assess killing efficiency as well as kinetics in NT and CAR-armed NK cells. As expected, we saw increased killing with the CAR transduced cells (both NK92 and primary NK cells) in comparison to NT NK cells. While expected, these results demonstrated the accuracy of using this platform in observing changes in cytotoxicity with different cell populations. Unique to this type of imaging-based platform, we observed serial killing of individual cells which was further improved on CAR arming. In addition to the cytotoxicity data, we also observed differences in the cellular contact durations between the effector and target cells. The primary CAR NK cells had higher contact times with Raji cells than the NT NK cells, insinuating a higher binding affinity. While increases in cytotoxicity of CAR NK were modest, these results are logical given the natural cytotoxicity of NK cells towards foreign cancer cells, as well as limitations with transduction rates producing a heterogeneous effector cell population. The increased potential for serial killing, as well as increased contact affinity for Raji cells both provide unique additional metrics of how CD19 CAR NK cells potentially increase elimination of the CD19+ B cell lymphomas.
We next focused on developing our own method for sorting droplets based on fluorescent signals. The key advantage of a droplet-sorting platform is the ability to isolate individual immune cells based on their activity, such as target cell killing. While protocols have been developed to sort droplets via flow cytometry, the recovery rate is much lower than a platform designed for droplet sorting (40% as opposed to 95%, as found by Hatori et al).45 Additionally, flow cytometry can induce cellular stress, negatively affecting downstream analysis and viability for the expansion.46 Droplets can reduce sorter-induced cell stress and can provide phenotypic sorting capabilities that are impossible in standard, bulk cell-sorting approaches. After validating our droplet-sorting platform’s efficacy, we successfully applied it to separate CAR and NT primary NK cells based on target cell killing and then performed RNA-sequencing. Sorting and sequencing based on the ability to kill target cells require co-encapsulation, which is a unique capability of our single-cell droplet sorter. Due to this coencapsulation requirement, detailed RNA sequencing data on CAR NK cells based on their killing efficiency has not previously been reported. Based on the killing profiles we were successfully able to compare gene expression profiles between killer versus non-killer CAR as well as killer versus non-killer transduced NK cells. We observed several interesting gene expression differences between killer and non-killer CAR NK cell populations. Genes coding for TNF superfamily ligands 12 (TWEAK) and 13B (BAFF) had a higher expression in the killer CAR NK cell population. TWEAK activates NF-κB, and BAFF can suppress B cell proliferation depending on receptor binding.42 We saw higher expression of the TNFRSF14 gene in the non-killing CAR NK. TNFRSF14 expression has been shown to increase apoptosis in cancer cells, contradicting these results.47 It may however be a product of increased numbers of surviving Raji cells expressing this gene in the droplets with non-killing NK cell populations. We also observed higher expression of key checkpoint receptor genes in the killer CAR NK cell population, including HAVCR (TIM3), TIGIT, and KLRC1 (NKG2A). This coincides with the observation of a more mature, active NK population, which typically also correlates with higher checkpoint receptor expression.48 49 The checkpoint ligands CD94, TIM3, Ceacam1 and CD47 were observed at notably higher expression in the non-killing population, suggesting they may be able to inhibit CAR NK. KLRD1, or CD94, when dimerized with NKG2A, acts as a checkpoint to NK cells, and has also been found to induce NK apoptosis.50 Ceacam1 is one of the known ligands for TIM3 and it interacts either in cis or trans configuration.51 This suggests that Ceacam1 expression on either Raji or NK themselves may be potentially inhibiting CAR NK. Nectin4 is a ligand for TIGIT and has previously been shown to affect NK-mediated cytotoxicity toward Raji cells.52 CD47, a surface marker strongly associated with resistance in B cell lymphomas was also observed in notably higher levels in the CAR non-killers.53 54 CD80, which has been found to interact with Programmed cell death ligand 1 (PDL1), was also found higher in the CAR non-killers.55 These findings display pathways that could present targets of cotherapy for CAR NK cells with continued research into their role in CAR NK cytotoxicity.
Transcriptomic sequencing was also performed on NT NK-Raji droplets sorted based on killing. Interestingly, NT killer NK cells appeared to be driven by different cellular factors than the CAR NK based on our RNA sequencing data. Killer NT NK cells did have similar GO term hits, as well as increases in CD56 and CD16 expression, indicating a potentially more mature phenotype. However, the killer NT NK lacked the drastic increases in granzyme, and interferon production found in killer CAR NK. Instead, killer NT NK cells appeared to have more consistent increases in TNF superfamily ligands (figure 7A). Expression of genes coding for TNF, FasL, and TNFSF10, 12, 13B, and 15 was all higher in the killer versus non-killer NT NK cell populations. These ligands all have the potential to induce apoptosis in target cells, depending on the receptor engagement.42 Expression of several checkpoint ligands was higher in the non-killer NK cell droplets. HMBGB1, a soluble ligand of TIM3, MHC-II (HLA-Ds), a ligand of LAG-3, and HLA-E, a ligand of NKG2A, were all higher in the non-killer droplets most likely reflecting increased concentrations of surviving Raji cells in these droplets.
There are several challenges presented in this study that may be improved in future work. We hypothesized that some contact events might be missed in the window of time between device loading and droplet imaging in the time-lapse microscopy experiments. To overcome this, droplets could be imaged at periodic intervals during the setup, to ensure no contact or killing events are missed. This study is also limited by an inadequate number of sequencing replicates to draw major biological conclusions, as the focus of this work was establishing and validating our novel method of screening immunotherapies. This study was also impacted by the inability to separate effector and target cells prior to sequencing, and by a heterogenous expression of CAR in the NK cells. Sequencing effector and target cells in a single sample can produce challenges in attributing expression levels of genes that are contained by both cell types. Genes known to be expressed in only one of the sequenced cell types can be targeted for analysis to overcome this. Incorporating single-cell RNA sequencing approaches would allow the identification of cell type, greatly enhancing the quality of transcriptomic analysis. Single-cell sequencing is however considerably higher cost and requires longer analysis. In some cases, bulk RNA sequencing may prove an adequate approach for this method, particularly in cases where effector and target cell have fewer conserved genes. Our results are also limited by the use of a single cancer cell line (Raji cells) in place of primary patient samples. In future studies, we plan to repeat these experiments with a higher number of replicates and with multiple donors and primary tumor cells, to derive more clarity of the cytotoxicity and suppression of CAR NK cells. Additionally, we plan to incorporate additional treatments for combination therapy with the CAR NK cells based on our sequencing observations, such as checkpoint inhibitor antibodies.
Our single-cell droplet system provides detailed interaction and killing kinetics as well as unique downstream data analysis. Our droplet array device revealed killing data within expectations, as well as novel serial killing, and effector-target contact data. Using our droplet sorter, we investigated the potential mechanisms driving the enhanced killing mediated by the CAR NK cells compared with the NT revealing several checkpoint pathways as well novel differences in the gene expression pathways killer versus non-killer NK cell/NK cell CARs. We believe our findings support the use of this single-cell droplet microfluidic array and sorting approach for screening immunotherapies with greater sensitivity and biological insight. Droplet platforms are uniquely suited for observing metrics like contact, serial killing, and downstream analysis based on killing capability. Applying findings generated by this approach could further improve the efficacy of numerous immunotherapies by uncovering previously unknown factors in the interactions between immune effector cells, cancer cells, and novel immunomodulatory and/or more traditional chemotherapy approaches.
Supplementary video
Supplementary video
Data availability statement
Data are available in a public, open access repository. Sequencing data is available in the National Center for Biotechnology Information's Gene Expression Ombinbus (GEO), accession number GSE243787.
Ethics statements
Patient consent for publication
Ethics approval
Not applicable.
References
Footnotes
X @romeerizwan
MRS and MF contributed equally.
Contributors Conceptualization: TK, RR, MS; development of droplet platforms: YY, MF, MRS, TK; Investigation: MRS, MF, YY, JT; development/production of cells: AA, MS, IK, YA, EB; supervision: TK, RR, MS; writing—original draft: MRS; writing—review and editing: MRS, TK, RR, MS, AA, MF, JT; Guarantor: TK, RR.
Funding Funding for this manuscript was the National Institute of Health grants R33 (1R33CA223908-01) and R01 (5R01GM127714-04) and the National Science Foundation (CBET-1803872) grant awarded to Dr Tania Konry. This work was additionally funded by the DFCI-NU Joint Program in Cancer Drug Development Awards (351144), awarded to Drs Tania Konry and Rizwan Romee.
Competing interests This paper was produced in part using droplet based technology that is currently optioned to a spin out Feromics Inc in which the author Tania Konry has a majority interest.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.