RETRACTED ARTICLE: A biomimetic assay platform for the interrogation of antigen-dependent anti-tumor T-cell function

Overcoming tumor-mediated immunosuppression and enhancing cytotoxic T-cell activity within the tumor microenvironment are two central goals of immuno-oncology (IO) drug discovery initiatives. However, exploratory assays involving immune components are often plagued by low-throughput and poor clinical relevance. Here we present an innovative ultra-high-content assay platform for interrogating T-cell-mediated killing of 3D multicellular tumor spheroids. Employing this assay platform in a chemical genomics screen of 1800 annotated compounds enabled identification of small molecule perturbagens capable of enhancing cytotoxic CD8+ T-cell activity in an antigen-dependent manner. Specifically, cyclin-dependent kinase (CDK) and bromodomain (BRD) protein inhibitors were shown to significantly augment anti-tumor T-cell function by increasing cytolytic granule and type II interferon secretion in T-cells in addition to upregulating major histocompatibility complex (MHC) expression and antigen presentation in tumor cells. The described biotechnology screening platform yields multi-parametric, clinically-relevant data and can be employed kinetically for the discovery of first-in-class IO therapeutic agents.

I t is well established that the types and frequencies of immune infiltrates in solid tumors serve as prognostic indicators for progression-free survival in patients 1,2 . Comprehensively, it can be stated that increasing infiltration and/or activity of desirable anti-tumor T-cells (effector T-cells) within the tumor microenvironment results in more durable patient response. Identification of new proteins that confer an immunomodulatory response in malignancy is critically important, in part due to the enormous clinical successes of the checkpoint inhibitors Pembrolizumab (anti-PD-1) and Ipilimumab (anti-CTLA-4) 3,4 . However, efforts to identify new drug targets in this space have been hampered by a lack of reliable and robust in vitro assays coupled with poor predictive powers of in vivo rodent models 5 .
There are many ways to enhance anti-tumor T-cell function. Generation and expansion of tumor-killing effector CD8 + T-cells are crucial for anti-tumor immunity, and this can only result from appropriate T-cell priming (cross-presentation). In cancer, much of the T-cell priming events occur in the tumor's draining lymph node where antigen-presenting cells (APCs) such as dendritic cells (DCs) and macrophages present tumor neoantigens to naive T-cells 6 . It is also desirable to increase penetration of T-cells into the tumor microenvironment, which is often poorly vascularized and characterized by dense extracellular matrix (ECM) and desmoplastic tissue 7 . A third way to increase T-cell tumor cytotoxicity is to facilitate the recognition of the tumor cells by the T-cells. The binding event between the T-cell receptor (TCR) and the peptide-containing major histocompatibility complex (pMHC) on tumor cells is biologically critical. Although a variety of peptides can be recognized by T-cells in the context of cancer 8 , tumor cells will often downregulate the expression of Human Leukocyte Antigen (HLA) genes which encode MHC proteins, thus rendering them invisible to effector T-cells 9 . Therefore, increasing tumor cell "visibility" would be desirable for new immuno-oncology therapeutics.
The necessity for scrutinizing cellular interactions in a 3dimensional (3D) cell model lies in the physical and chemical cues present within the tumor microenvironment that cannot be recapitulated in conventional 2-dimensional cellular assays. Specifically, the multicellular tumor spheroid model is characterized by gradients of oxygen and nutrient penetration, large internal sections of hypoxia and acidic cores 10 , pathologies that are similarly observed in a clinical setting. Tumor cells often exhibit an altered metabolic phenotype called aerobic glycolysis (Warburg effect); the result of which is the production and secretion of lactic acid, which lowers the pH of the extracellular microenvironment and leads to areas of necrosis 11 . Tumor lactic acid production has been shown to be negatively correlated to T-cell proliferation and anti-tumor function 12 . The 3D multicellular tumor spheroid model is a faithful mimic of both hypoxia and lactic acid secretion and, thus, is a useful tool for the interrogation of these pathologies on the ability of effector T-cells to induce tumor cytotoxicity 13 . T-cell tumor infiltration is essential for effector function. Therefore, a 3D scaffolded tumor structure is a prerequisite for observing and modulating this biology 14,15 . Comprehensively, 3D tumor spheroid cell models may represent more faithful surrogates for the study of clinically relevant T-cell anti-tumor function than conventional 2D in vitro methods 16 .
The efforts described here identify new classes of targeted small molecules that enhance anti-tumor T-cell function. Further, we used viral peptide antigens in an HLA-matched co-culture system to mimic a tumor antigen-specific T-cell response. By miniaturizing the assay we were able to profile a large collection of compounds and identify several target proteins that can be modulated to augment cytotoxic T-cell activity. Specifically, CDK4/6 inhibitors and nonspecific BRD inhibitors were shown to act on both T-cell and tumor cell compartments by (1) increasing the release of T-cell cytolytic granules and interferon-gamma (IFNg) and (2) upregulating tumor cell MHC expression, antigen presentation, and IFNg secretion. This resulted in canonical Tcell-mediated tumor killing which was conserved over a variety of different HLA-matched patient donors. Several reports have identified methods that can be used to enhance anti-tumor T-cell function using both 2D 17 and 3D cell models 16 , and this work aims to expand on those efforts in an industrialized drug discovery setting.

Results
Development of the T-cell tumor spheroid-killing platform and screening workflow. In developing an assay platform to enable the identification of agents that enhance T-cell anti-tumor function the aim was to detect compounds that elicit effects on Tcells either intrinsically or extrinsically via tumor cells. There are several conceivable ways experimental compounds might enhance T-cell-mediated anti-tumor function in our model system. Compounds might act directly on T-cells by enhancing antitumor cytokine pathways (e.g., IFNg, TNFa), agonizing cytolytic pathways, or enhancing perforin or cytolytic granzyme synthesis/ secretion ( Fig. 1a: 1 and 2). Alternatively, compounds might act via tumor cells to affect T-cell killing; by increasing MHC class I expression and/or antigen processing/presentation or upregulation of co-stimulatory molecules ( Fig. 1a: 3).
3-dimensional (3D) tumor spheroids were used in this phenotypic screening platform to mimic clinically relevant tumor physiology. Although small tumor spheroids do not typically display necrotic cores comparable to larger-sized spheroids 10 , 1536-well spheroids did display substantial areas of hypoxia and acidic pH ( Supplementary Fig. 1a). T-cell anti-tumor function was quantitated by a reduction of spheroid size or complete dissociation of the 3D tumor spheroid structure. The assay began on day −9 when normal human donor peripheral blood mononuclear cells (PBMCs) were thawed and cultured in the presence of a 1 μM CEF peptide pool (Fig. 1b). The CEF peptide pool is a mix of 32 different peptides derived from either Cytomegalovirus (CMV), Epstein-Barr virus (EBV), or Influenza virus (Flu) and induces T-cell IFNg secretion and expansion in approximately 80% of tested donor PBMCs ( Supplementary  Fig. 1b). PBMC cultures were then expanded in the presence of low dose interleukin-2 (IL-2) for a further 3 days to increase the proliferation of CD8 + CD127 + memory T-cells 18 . PBMC cultures, which consisted mainly of CD4 + and CD8 + T-cells, were then cultured in a "resting" medium for a further 3 days in the presence of IL-2, IL-4, and IL-7 (Fig. 1b) 19 . This resting period was necessary to mitigate nonspecific, antigen-independent cytolytic properties of the activated T-cells ( Supplementary  Fig. 1c). On day −2, 1000 human colorectal tumor cells (HCT116 expressing the enhanced green fluorescent protein-EGFP) were plated into 1536w spheroid plates in the presence of either 1 μM CEF peptide pool or a negative control CEF Scramble pool (to act as a counter screen for antigen-dependence). On day 0, T-cells were stained red and added to the tumor spheroids in a 3:1 ratio of T-cells to tumor cells (effector: target) (Fig. 1b). Experimental test compounds were then added at 30 μM single point dose. The assay proceeded for 3 days before the first imaging readout was collected and then another 2 days before the final readout was collected by high-content confocal microscopy (Fig. 1b).
Characterization of antigen presentation and reactivity. In order to ensure an antigen-dependent mechanism of T-cell antitumor function, it was necessary to quantitate tumor cell MHC expression and antigen presentation as well as T-cell antigen  Fig. 2a), indicating a high level of expression. HCT116 cells expressed high quantities of MHC on their cell surface membranes and treatment with IFNg did not meaningfully increase MHC presentation (Supplementary Fig. 2b). Next, we examined the exogenous loading of tumor cell HLA-A02 by first quantitating the amount of viral peptide expressed on MHC Class I molecules. HCT116 tumor spheroids were incubated with a fluorescently tagged CMV peptide [CEF 20: CMVpp65(495-503)-BODIPY-488], then treated with either a protein transport inhibitor (Brefeldin) or an unlabeled CMV cold competitor. Brefeldin was used as a positive control treatment in order to arrest protein transport within the cells, thus preventing intracellular MHC molecules from presenting peptides on cell surface membranes. The localization of the CMV-BODIPY peptide to the cell membranes of tumor cells was visualized by high-content confocal microscopy (Fig. 2a). Spheroids were holistically displayed as maximum intensity projections of the entire spheroid (Z-Stack Merge) and then as focused sections within the middle of the spheroid (single plane merge) to characterize CMV-BODIPY cellular localization.
Tumor spheroids treated with the drug vehicle (DMSO) displayed punctate and cell membrane-localized CMV-BODIPY peptides, consistent with MHC-based cell membrane localization (Fig. 2a, DMSO, blue arrows). Treatment of spheroids with Brefeldin resulted in accumulation of CMV-BODIPY within cells and poor trafficking of the peptides to cell membranes (Fig. 2a, Brefeldin, white arrows). Treatment of spheroids with an unlabeled CMV cold competitor did not affect CMV-BODIPY transport to cell membranes to a noticeable degree, though we did observe some accumulation of CMV-BODIPY peptides within cells (Fig. 2a, CMV cold competitor). To functionally quantitate CD8 + T-cells specifically reactive to a common CMV peptide within the CEF pool (CMVpp65 495-503), we used antibody tetramers, soluble MHC molecules loaded with the CMVpp65 peptide antigen. Tcells stimulated with CEF pool were quantitated through tetramer flow cytometry and shown to be highly reactive against CMVpp65 peptide, but not against the negative control EBV peptide (Fig. 2b). PBMC donor #4666 was highly reactive to the CMVpp65 viral peptide and was chosen to be used in the subsequent high-content screening campaign.
To prepare cytotoxic T-cells, we expanded T-cells from PBMC cultures in three phases. First, T-cells were stimulated using the CEF viral peptide pool which initiated CD8 + CD127 + T-cell expansion (Fig. 2c, day 3). T-cells were then further proliferated in the presence of IL-2 for three days which selectively increased CD8 + T-cell expansion over CD4 + T-cell expansion (Fig. 2c, day 6). The third phase of T-cell expansion required culturing T-cells in "resting" medium, which contained IL-2, IL-4, and IL-7 (Fig,  2c, day 9) and resulted in the further expansion of the desired CD8 + CD127 + population. Failure to expand T-cells in a resting medium resulted in aggressive cytolytic behavior upon exposure to tumor spheroids in a non-antigen-dependent manner ( Supplementary Fig. 1c). However, a 4-day expansion in T-cell resting media mitigated these antigen-independent cytolytic effects enabling the identification of T-cell-enhancing therapeutics.
To ensure that antigen alone was responsible for this observed T-cell anti-tumor function, we performed a cell-killing assay using T-cells that had been primed with the CMVpp65 viral peptide and HCT116 colorectal cancer cells engineered to express the Firefly Luciferase (FL) reporter gene. HCT116-FL cells were grown in the presence of a limiting dilution of either CMVpp65 peptide or a scramble negative control peptide. T-cells were added in a 3:1 ratio (T-cells: tumor cells) and co-culture proceeded for 48 h. Analysis of HCT116-mediated Firefly Luciferase enzyme indicated T-cells killed tumor cells in a CMV-dose-dependent manner; this cytotoxicity was not observed in tumor cells that had been loaded with scramble control peptides ( Supplementary Fig. 3).

Screen for enhancers of T-cell-mediated anti-tumor function.
To find target proteins whose functions might be modulated to enhance T-cell-mediated tumor cytotoxicity, we employed the spheroid-killing IO platform in a chemical genomics screen of 1770 annotated compounds with known or suspected targets or mechanisms of action (MoAs) 20 . The compound collection screened during this initiative was a subset of the library described in 20 and consisted of low molecular weight targeted perturbagens covering more than 16 different protein and enzyme classes ( Supplementary Fig. 4a). The majority of abnormal spheroid phenotypes observed could be broadly categorized as either Hit Class I or Hit Class II (Fig. 3a). Hit Class I compounds reduced the size of the spheroid but maintained its overall shape factor compared to the vehicle control (DMSO; control picture); whereas treatment with Hit Class II compounds resulted in the complete loss of both spheroid shape factor and spheroid area (Fig. 3a). Hit Class II compounds, such as CDK4/6 inhibitors, demonstrated severe deterioration of spheroid structural integrity compared to the DMSO (vehicle) negative control in which the spheroid grew to a large and characteristically rounded shape (Fig. 3b). Interestingly, T-cell proliferation was often correlated to anti-tumor activity indicated by increased numbers of T-cells observed in hit compound-treated samples versus non-hit compound-treated samples ( Supplementary Fig. 4b).
Increased T-cell proliferation was often observed in DMSO-treated samples, which is why T-cell count calculations were performed against the median average of the screened compounds and not the DMSO vehicle controls ( Supplementary Fig. 4b).
The high-throughput screen was run concurrently with a counter screen for the purpose of distinguishing compounds whose activities were dependent on tumor cell antigen presentation and T-cell recognition. The primary screen was run using both T-cells and tumor cells loaded with identical tumor antigens (CEF Peptide Pool), the counter screen was run using negative control antigens (CEF Scramble Pool), and all of the hit compounds were profiled on tumor spheroids alone to identify those compounds that were not dependent at all on the presence of T

-cells (No T-cells).
Predictably, most of the compounds that demonstrated spheroid cytotoxicity were not at all dependent upon the presence of T-cells. These nonspecific cytotoxic compounds, such as microtubule inhibitors and histone deacetylase (HDAC) inhibitors, were able to dissociate tumor spheroids and kill tumor cells (Fig. 4a, Cytotoxic). Other classes of compounds that killed tumor spheroids were those that interfered with MAPK signaling. HCT116 cells are mutant KRAS-driven colon cancer, so compounds that antagonize RAF, MEK, or ERK signaling would be expected to act selectively against this tumor type 21 . Intriguingly, many compounds whose targets and mechanisms act through ECM remodeling could be distinguished by their unique phenotypes. Compounds that antagonize matrix metalloproteinases (MMPs) or integrin receptors led to the complete structural dissociation of the tumor spheroids without any noticeable associated cellular toxicities (Fig. 4a, ECM). To distinguish between spheroid destruction and tumor cell killing we profiled hit compounds on T-cell: HCT116-FL spheroid cocultures. Luciferase assays performed on treated HCT116-FL spheroids revealed consistencies between the phenotypic imaging data and the spheroid viability data. Compounds that inhibited HDAC or ERK proteins severely affected tumor cell viability at low doses (Fig. 4b). Tubulin inhibitor compounds were less toxic to HCT116-FL spheroids and MMP inhibitors demonstrated minimal viability effects on spheroids, consistent with the imaging analyses (Fig. 4b).
There were, however, a number of compounds from the screen that demonstrated T-cell-dependent activities. For example, compounds that negatively affected NF-κB or IL-4 signaling showed very strong antigen-dependent anti-tumor activities, as scramble antigen controls did not produce these phenotypes (Fig. 5a, Anti-inflammatory). As mentioned previously, another class of compounds that was shown to enhance T-cell tumor cytotoxicity was the cyclin-dependent kinase (CDK) group of kinase inhibitors. Compounds that inhibited CDK4 and/or CDK6 were able to slow tumor spheroid growth in the absence of Tcells, consistent with the behavior of cytostatic cell-cycle-targeting compounds (Fig. 5a, Cytostatic, No T-cells). However, in the presence of T-cells, CDK4/6 inhibitors were able to significantly reduce tumor spheroid size and these phenotypes were exacerbated when T-cells were added to tumors loaded with their cognate viral antigens (Fig. 5a, Cytostatic). In addition to CDK inhibitors, bromodomain (BRD) inhibitors also demonstrated antigen-dependent anti-tumor phenotypes. Compounds targeting other proteins with previously-unreported Tcell-modifying activities also scored as hits in the screen. Many of these compounds acted through nonspecific (non-antigenrelated) T-cell mechanisms and demonstrated tumor-killing behaviors in the CEF Peptide Pool, as well as the CEF Scramble Pool negative control and to the best of our knowledge, have not been previously reported to have immunomodulatory properties (Fig. 5a, Novel Targets). To confirm these compounds were inducing T-cell-mediated cytotoxicity on tumor spheroids we performed identical assays using HCT116-FL spheroids loaded with matched CEF Pool peptides. All of the T-cell-dependent compound classes identified from the phenotypic screen demonstrated potent viability effects on HCT116-FL spheroids, indicating tumor cell death in addition to tumor spheroid destruction (Fig. 5b).
Characteristics of lead compounds and T-cell phenotypes. Few compounds that scored in the screen demonstrated antigendependent T-cell mechanisms. Therefore, to expand on these compounds and target classes, we selected small molecules from our compound libraries that have similar structures or similar targets as the hits from the high-throughput screen. 72 of these SAR compounds (structure-activity relationship) were tested for T-cell-mediated anti-tumor activity. One particularly interesting compound that elicited antigen-dependent anti-tumor T-cell function was cyclopamine, a well-known Sonic Hedgehog (SHH) pathway inhibitor, and Smoothened (SMO) antagonist. Cyclopamine treatment of T-cell/tumor spheroid cultures led to robust killing and dissociation of tumor spheroids in a dosedependent manner (Fig. 6a). Further, antigen-matched cultures demonstrated a 7-fold increase in cyclopamine potency over nonantigen-matched (Fig. 6a, IC (Fig. 6a, No T-cells). Finally, several experimental therapeutics with unknown targets exhibited potent anti-tumor T-cell efficacies when tested kinetically and in dose-response. GNF001, a small molecule with an unknown target, yielded potent T-cell-mediated tumor killing in a time-and antigen-dependent manner (Fig. 6b). This compound showed slight anti-tumor activity after 3 days but marked anti-tumor efficacy after 5 days (Fig. 6b, IC 50 curves).
When performing cellular studies involving multiple cell types, it is critical to establish which cellular compartment is responsible for eliciting any observed phenotypes. Therefore, we interrogated the effects of our lead compounds on both monoculture T-cells and tumor spheroids. Compounds were applied to T-cells or tumor spheroids for 48 h, then washed out, before combining cells into the co-culture model described earlier. Compounds that antagonized CDK4/6 activity were observed to elicit their activities via both cell types, as T-cell anti-tumor activity was observed under both tumor spheroid-only treatment and T-cellonly treatment (Supplementary Fig. 5). However, a dose-response for tumor cytolytic phenotypes was observed for CDK4/6 inhibitors only under tumor spheroid treatment; only the highest dose of this compound on T-cells was able to enhance T-cellmediated anti-tumor function ( Supplementary Fig. 5). BRD inhibitors and the proprietary compound GNF001 likely worked through tumor spheroid-mediated mechanisms as T-cell antitumor function was seen only in the spheroid-treated samples compared to vehicle-treated cultures (DMSO) ( Supplementary  Fig. 5). Compounds that antagonized NF-κB or IL-4 inflammatory pathways did not elicit potent phenotypes under To understand the mechanism of action of these compounds, we treated either T-cells or tumor spheroids in monoculture with lead compounds for 48 h and quantitated cytokine secretion pathways and cell surface markers by FACS. Although none of the compounds except for cyclopamine were able to significantly increase the proliferation of CD3 + T-cells, we did observe the significant expansion of the CD8 + /CD25 + T-cell fraction under CDK4/6 inhibitor, NF-κB inhibitor, and cyclopamine treatment (Fig. 7a). More importantly, however, were the increases in T-cell reactivity to the CMV tetramer antibody (Fig. 7a, CD8 + /CMVtet + ) observed for nearly all compounds tested. These T-cell phenotypes indicated increased recognition of the CMV viral peptide, which may explain a potential mechanism by which tumor spheroids were more susceptible to T-cell cytolytic function under compound treatment. Intriguingly, treating T-cells with CDK4/6 inhibitors led to a drastic increase in interferon-gamma secretion, even in the absence of tumor spheroid target cells, indicating significant agonism of this antitumor cytokine pathway (Fig. 7a, CD8 + /IFNg + ). In conjunction with the FACS-based T-cell phenotyping experiments, we also interrogated HLA-A02 cell surface expression and proinflammatory cytokine secretion by tumor spheroids under monoculture compound treatment conditions. Nearly all compounds tested in this manner were observed to increase the expression of HLA-A02 molecules on tumor cell surface membranes (Fig. 7b, HLA-A02 + ). Additionally, CDK4/6 inhibitors and GNF001 also were shown to significantly increase interferon-gamma secretion in tumor spheroids (Fig. 7b, IFNg + ). Surprisingly, TNF-alpha secretion was reduced in tumor spheroids in nearly all samples indicating antagonism of this cytokine pathway under compound treatment (Fig. 7b, TNFa + ). How reduced TNFa secretion in tumor spheroids affects T-cell anti-tumor function has yet to be determined.
Establishing conserved efficacies over patient cohorts is of critical importance with experimental therapies. Therefore, we sought to characterize the effects of one lead compound over a variety of normal donor PBMCs. Kinetic studies using the CDK4/ 6 inhibitor (CDK4/6i) compound over 9 HLA-A02 normal donor PBMCs yielded heterogeneous results. Donor T-cell/spheroid cocultures were treated with either 50 or 5 μM CDK4/6i for 96 h and tracked for spheroid size under antigen-matched (CEF) or antigen-mismatched (Scr) conditions. Three donors demonstrated anti-tumor efficacies only under high dose of compound treatment and only under antigen-matched conditions (Fig. 8a, Antigen-Dependent Responders). Three donors exhibited antitumor cytotoxicity under high dose treatment, independent of antigen recognition (Supplementary Fig. 6a). Two donors showed no response to CDK4/6i treatment under any conditions (Fig. 8a, Non-Responders). These data reflect a heterogeneous response rate of immunomodulatory agents comparable to clinical settings.
For a more detailed analysis of T-cell phenotypes during coculture with tumor spheroids under CDK4/6i treatment, we extracted T-cells from these cultures and profiled them for functional markers of active tumor killing. Nearly all of the CD8 + T-cells were positive for the degranulation marker CD107 indicating a high degree of tumor killing (Fig. 8b). We observed modest yet significant increases in Granzyme-B (GranB) and Interferon-γ (IFNg) in the CEF-treated samples over the Scr-treated samples indicating potent cytokine expression under antigenmatched T-cell conditions. Further T-cell FACS-based phenotyping ( Supplementary Fig. 6b) revealed an increase in CD8 + CD107 + , CD8 + CMVtet + , CD8 + GranB + and/or CD8 + IFNg + cells in the CEF-treated spheroids over Scr-treated spheroids in three donors that also showed an antigen-mediated T-cell anti-tumor response ( Fig. 8b; D17044427, D19054173, D17044460, CEF). T-cells from two donors that did not demonstrate CEF-specific T-cell killing or any anti-tumor efficacy did not display similar T-cell phenotypes ( Fig. 8b; D19054445, D19054242). These data indicate that donor T-cells which have contributed to anti-tumor cytolytic function display canonical T-cell degranulation and cytokine markers which are proportional to the observed level of tumor cell killing.

Discussion
Augmenting anti-tumor immunity through the enhancement of CD8 + cytolytic T-cell function is a central initiative in the field of immuno-oncology (IO). The study of functional immunity in vitro can be challenging, and IO cellular assay platforms may not take into account important characteristics of the tumor microenvironment, such as hypoxia and low interstitial pH, which may drastically affect anti-tumor T-cell function. Further, the only way to truly mimic clinical T-cell-tumor cell interaction ex vivo is through the use of primary patient tumor material and autologous PBMCs; a limited prospect when conducting industrial drug discovery screening campaigns that require large amounts of cells. The assay platform presented here attempts to address both of these shortcomings through the use of 3D tumor spheroids and HLA-A02-matched T-cells and tumor cells.

L E
Additionally, the assay platform relies on an antigen-dependent mechanism of T-cell-tumor cell interaction, mimicking T-cell education in the draining lymph node, thereby making it highly mimetic of clinical tumor immunology.
When profiling an advanced library of annotated small molecules, it can be expected that compounds targeting key cellular housekeeping proteins will elicit cytotoxic phenotypes; for example, the large number of tubulin inhibitors that scored as hits in the screen. Similarly, given the KRAS-dependency of the HCT116 human colorectal cancer cell type used, we expected the high proportion of MAPK pathway-targeting kinase inhibitors to show anti-tumor activity.  0 0.2 0.3 0.5 0.7 0.8 1.0 1.2 1.3 1.5 1.7 1.8 2.0 2.2 2.3 2.5 2.7 2.8 3.0 3.2 3.3 3.5 3.7 3.8

A R T I C L E
We focused our attention on those compounds that elicited a tumor cytotoxic phenotype only in the presence of T-cells and only within an antigen-matched cellular context. Several compounds targeting the NF-κB pathway as well as antagonists of proinflammatory cytokines (e.g., IL-4) demonstrated potent anti-tumor phenotypes in an antigen-dependent manner. Although much is known about NF-κB signaling and its antagonism in CD4 + regulatory T-cells, NF-κB signaling is less characterized in tumor-associated cytotoxic CD8 + T-cells. NF-κB inhibition has been shown to decrease Foxp3 expression on CD4 + T-cells and increase IFNg and IL-2 secretion 22 , so NF-κB inhibition in CD8 + T-cells might act through similar cytokine release mechanisms 23 . Surprisingly, NF-κB inhibitors demonstrated anti-tumor efficacy only in the co-culture cellular model, where no anti-tumor activities were observed when either T-cells or tumor spheroids were treated separately. Consistent with this observation, treating T-cells with NF-κB inhibitors did not significantly affect their proliferation, differentiation, or proinflammatory cytokine secretion. Treatment of tumor spheroids with NF-κB inhibitors slightly enhanced HLA-A02 expression on cell surface membranes and decreased IFNg and TNFa secretion, though these phenotypes would not explain the strong anti-tumor efficacies observed during T-cell/ spheroid co-culture treatment. It is tantalizing to speculate that co-culture of T-cells with tumor spheroids may activate NF-κB signaling in one of these cell types that is not present in monoculture and that activation of this paracrine signaling mechanism is responsible for the observed efficacies of NF-κBtargeting compounds. IL-4, a key component of chronic inflammation, has been shown to be pro-tumorigenic and inhibition has been shown to promote the anti-tumor activity of cytotoxic T lymphocytes 24 . Therefore, our observation that small molecule-mediated IL-4 inhibition enhances T-cell cytolytic function is consistent with reported results.
Strikingly, a number of characteristically cytostatic compounds were able to significantly enhance anti-tumor T-cell function. Particularly, treatment with CDK4/6 inhibitors had a mild cytostatic effect on monoculture tumor spheroids but marked cytotoxic effects were observed when those compounds were combined with T-cells. Given the recent finding that CDK4/6 inhibition augments tumor immunity by increasing the functional capacity of tumor cells to present antigen and reducing the activity of immunosuppressive regulatory T-cells 25 , we explored whether those phenotypes were present in our cellular model during CDK4/6 inhibitor treatment. Consistent with previous reports, we detected significant increases in HLA-A02 expression on cell surface membranes of tumor spheroids when exposed to CDK4/6 inhibitory compounds and this activity may have been a consequence of enhanced IFNg secretion which was concomitantly increased in these cells during drug treatment. Similarly, CDK4/6 inhibitors were able to significantly enrich T-cell reactivity to the CMV tetramer antibody as well as agonizing IFNg expression, indicating this class of compound may simultaneously increase antigen presentation by target cells and antigen recognition by cytolytic T-cells.
In a mechanism similar to CDK inhibition, bromodomain (BRD) protein inhibitors displayed strong anti-tumor function when combined with T-cells. Reported immunomodulatory activities of BRD inhibitors include (1) expansion and enhancement of cytotoxic CD8 + T-cell populations 26 and (2) increased CD8 + T-cell penetration, IFNg secretion, and degranulation in vivo 27 . Although we did not observe the expansion of cytolytic T-cell populations or IFNg expression under BRD inhibitor treatment, we did detect significant increases in both tumor cell HLA-A02 cell surface expression and T-cell CMVpp65 antigen reactivity. These data indicate that similar to CDK4/6 inhibitors, BRD antagonists may simultaneously enhance T-cell-mediated surveillance of antigen-presenting cells while decreasing immune escape by tumor cells.
One compound that exhibited potent anti-tumor function in an antigen-and T-cell-dependent manner was cyclopamine. Cyclopamine is a well-known teratogen with anti-cancer properties 28 . Cyclopamine is a Sonic Hedgehog (SHH) pathway inhibitor, binding Smoothened (SMO) protein with high affinity, and shutting down SHH signaling 29 . Cyclopamine immunomodulatory activity has been shown to be dependent upon oxygenation state, where enhanced T-cell proliferation, tumor cytotoxicity, and IFNg secretion only occurred under hypoxic conditions 30,31 . During our study, we revealed a potent ability of cyclopamine to induce antigen-dependent T-cell-mediated tumor cytotoxicity. Through FACS-based T-cell phenotyping, we observed modest yet significant increases in both total CD3 + and CD8 + /CD25 + cytolytic T-cell proliferation under cyclopamine treatment as well as upregulation of the checkpoint molecule PD-1. It is conceivable the hypoxic environment of spheroid cores was responsible for the observed efficacy of cyclopamine in our assay, a phenomenon that would not have been detected in a conventional 2D cell-killing assay, yet has far-reaching clinical implications.
Several small molecules with targets not previously reported to be involved in T-cell biology or immunomodulation also scored as hits on the screen. One of these compounds, GNF001, yielded a highly desirable therapeutic profile; showing no detectable anti-tumor function in the absence of T-cells. This compound is speculated to affect its anti-tumor function via the tumor cell compartment due to the significant increases in IFNg expression induced in tumor cells under compound exposure. In support of this hypothesis, we observed the anti-tumor efficacy of this compound only during tumor spheroid monoculture treatment, but not in T-cell monoculture treatment. More thorough MoA interrogations are needed to understand the mechanisms by which GNF001 enhances antitumor T-cell function via tumor cells.
Response to immunomodulatory chemotherapeutics in the clinic is heterogeneous. Appropriate patient stratification for clinical testing is crucial for maximizing therapy response. By profiling nine different PBMC donors in a mini "ex vivo clinical trial", we were able to identify people that would potentially respond to CDK4/6 inhibition in a model of colorectal cancer. These donors could be categorized as either antigen-dependent responders, antigen-independent responders, or non-responders, similar to heterogeneous IO-response rates observed in clinical settings. Moreover, T-cell phenotyping from these donor cultures revealed a tight correlation between activation of T-cell killing markers and an anti-tumor response, indicating the utility of this assay platform for profiling patient T-cell response to therapy.
Taken together, our unique IO assay platform provides evidence that T-cell anti-tumor function can be profiled in a highthroughput industrial drug discovery setting and that small molecule targeted inhibitors can augment intrinsic properties of T-cells either by upregulation of antigen recognition machinery and/or secretion of IFNg. Alternatively, through this study, we revealed that many of these targeted inhibitors are also capable of upregulating tumor cell MHC expression and IFNg secretion, which may, consequently, enhance tumor immune surveillance and recognition by cytolytic T-cells. Comprehensively, these data reveal T-cell-and tumor cell-mediated immunomodulatory functions of targeted small molecules that may be exploited clinically to synergize with checkpoint inhibitors and improve patient care. peripheral blood mononuclear cells (PBMCs) that were matched to human colorectal cancer cells by their HLA-A02 alleles. Further, in order to create an expanded population of effector CD8 + memory T-cells, PBMCs were stimulated with a pool of virally derived peptides that induce viral antigen-dependent T-cell expansion. The same viral peptides were used to load tumor cell HLA-A02 molecules, ensuring interactions between T-cells and tumor cells were dependent upon antigen presentation by the tumor cells and recognition by memory T-cells through their TCRs. This method served as a surrogate for tumor neoantigen presentation and recognition by T-cells that might occur in patients. Primary human peripheral blood mononuclear cells (PBMCs) were obtained from a healthy 38-year-old donor and were characterized by PCR as HLA-A02 positive (Donor ID 4666, Hemacare). Frozen PBMCs were thawed and cultured in RPMI + 10% human serum + 1× pen/strep/anti-fungal at~1 Mil cells/ml. PBMCs were stimulated with 1 µM CEF Viral Peptide Pool (Anaspec) for 48 h. On day 2, recombinant human (rh) IL-2 (Peprotech) was added at 0.5 ng/ml for 72 h, and cultures were replenished with rhIL-2 every 24 h. On day 5, rhIL-2 (0.5 ng/ml), rhIL-4 (0.5 ng/ml), and rhIL-7 (10 ng/ml) (all from Peprotech) were added, and cultures were incubated for a further 4 days, with cytokine replenishment every 48 h. Two days prior to harvest of T-cells, HCT116-EGFP cells (human colorectal cancer cells expressing the enhanced green fluorescent protein; engineered from parental HCT116 cells acquired from ATCC) were dispensed into 1536-well Ultra-Low Attachment Spheroid Microplates (Corning) using an automated liquid dispenser (GNF systems) at 1000 cells per well in 4 µl of T-cell media with either 1 µM CEF Peptide Pool or a negative control CEF Scramble Pool (JPT). Spheroid formation proceeded for 48 h. After 9 days of culture, PBMCs (heretofore referred to as T-cells) were harvested, counted, and stained with 0.5 µM CellTracker Deep Red Dye (ThermoFisher Scientific) according to the manufacturer's instructions. T-cells (3000 cells/3 µl in T-cell media) were then dispensed into 1536-well Ultra-Low Attachment Spheroid Microplates containing the HCT116-EGFP spheroids. At this time, 70 nl of experimental test therapeutic compounds were dispensed using Echo Acoustic Transfer (Labcyte). Plates were incubated in a temperature-controlled tissue culture incubator at 37°C with 5% CO 2 . After 3 days, plates were imaged using the ImageXpress Micro Confocal High-Content Imaging System (Molecular Devices) with a 488 nm laser for the HCT116-EGFP spheroid and a 561 nm laser for the T-cell population. Wells were read using a Z-stack series of 27 individual planes over a 200 µm vertical distance and images were analyzed using a maximum intensity projection of all 27 planes. Plates were again imaged on day 5 using the same plate parameters.
During experiments when either T-cells or tumor spheroids were treated with compounds separately in monoculture, the following method was used. T-cells were treated with experimental compounds for 48 h, then cultures were washed 3× with PBS before being resuspended in T-cell media and applied to tumor spheroid cultures as described earlier in the methods. For tumor spheroid monoculture treatment, tumor spheroids were treated with experimental compounds for 48 h, then washed 3× with PBS before the addition of T-cell media and T-cell cultures as described earlier in the methods.
Confocal tomography was conducted with an ImageXpress Micro Confocal (Molecular Devices, San Jose, CA) using a 60 µm Nipkow spinning disk. GFP and Cy5 filter sets optimized for LED excitation were used to acquire eGFP and CellTracker DeepRed signals. A sampling depth of 220 µm was obtained with a 10×0.45NA Plan Apochromat objective. Maximum intensity projections were created for high-content analysis (HCA), which was performed using MetaXpress Custom Module Editor. The HCA pipeline measured the total area of tumor spheroids and T cells using the segmentation of HCT116-GFP and CellTracker DeepRed, respectively. Additionally, the pipeline measured the quantity of HCT116 spheroids detected. This facilitated the identification of improper cell seeding and spheroid fractures. Quantitative HCA results were passed through a custom KNIME (KNIME, Zurich, Switzerland) data pipeline for treatment annotation and calculation of CMV peptide treated to scramble peptide treated HCT116 tumor spheroid area ratio.
Analysis of tumor spheroid 3D pathologies. HCT116 parental cells were used to create ultra-HTS 1536-well spheroids as described earlier in the methods. Spheroids were then stained with Image-iT™ Green Hypoxia Reagent to highlight areas of low oxygen and pHrodo™ Red, SE to highlight areas of low pH (Ther-moFisher Scientific) according to the manufacturer's instructions. Spheroids were then imaged using high-content microscopy as described earlier in the methods.
CMV viral peptide tracking assay. HCT116 human colorectal cancer cells (ATCC) were cultured in McCoy's 5 A media + 10% FBS + 1× pen/strep/antifungal according to standard growth conditions. Cells were harvested, counted, and dispensed into a 384-well Black/Clear Bottom Polystyrene Spheroid Microplates (Corning) at 10,000 cells/well in 30 µl media in the presence or absence of 10 µg/ml recombinant human interferon-gamma (IFNg). Cells were incubated in a temperature-controlled tissue culture incubator at 37°C with 5% CO 2 for 48 h. After 48 h of spheroid growth, 2 µg/ml Brefeldin-A (Sigma-Aldrich) or a volume equivalent of the dimethyl sulfoxide (DMSO) vehicle was added and spheroids were incubated for 30 min. After 30 min, 1 nM CMV-BODIPY-SE (Anaspec) viral peptide was added with or without a cold CMV peptide competitor

A R T I C L E
Statistics and reproducibility. Independent sample t-tests were used to calculate probability between two different groups. Experiments typically consisted of three or more biological replicates and studies were often repeated three or more times to determine the robustness of data. For HTS assays, four or more biological replicates were often profiled per experiment.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data sets generated during and/or analyzed during HTS campaigns of the current study are not publicly available due to proprietary activities of compounds with known targets or molecular mechanisms of action but are available from the corresponding author on reasonable request. High-throughput compound screening data are currently stored on the GNF (Novartis) cloud-based data storage system. Non-HTS data has been supplied with the manuscript under Supplementary Data 1.