The majority of cell lines (RENCA, B16F10, 3LL, AB12, PANC1) were acquired from the American Type Culture Collection (ATCC; Manassas, VA), with the exception of the LivMet cell line, which was generously provided by Prof. David Tuveson. LivMet cells are derived from mouse liver metastases that arose in KrasG12D/+ transgenic mice, which developed pancreatic ductal adenocarcinoma (PDA) tumors.

The LPA stable cell line was established in our laboratory. We engineered a green fluorescent protein (eGFP) plasmid vector (9600 bp) containing the mouse PD-L1 (CD274) promoter (1629 bp: 1341 bp downstream and 287 bp upstream of the transcription splice site), the eGFP gene, and puromycin and ampicillin resistance genes ( Supplementary Figure 1 ). This plasmid was custom-designed and synthesized by GenCopeia™ (Rockville, MD) according to our specifications.

GeneCopeia utilized the GeneCopeia Lenti-Pac™ HIV Expression Packaging Kit to co-transfect a GeneCopeia Lenti-Pac HIV Expression Packaging plasmid with the HIV-based lentiviral expression plasmid into GeneCopeia 293T lentiviral packaging cells. This process yielded pseudovirus particles containing the lentiviral expression construct. GeneCopeia provided a titer of 2.82*10^8 TU/ml lentivirus along with the necessary plasmid. Lentiviruses were shipped on dry ice and stored at -80°C until use.

To establish a stable cell line expressing the PD-L1-probe, we integrated the viral expression construct into the genomic DNA of LivMet cells. Initially, 3*10^4 cells were seeded in a 48-well plate and reconstituted with DMEM (Sartorius) containing 10% FCS (Gibco), 1% L-glutamine (Sartorius), 1% penicillin-streptomycin (Gibco), and 1% sodium pyruvate (Gibco). 24 hours later, the medium was replaced with the same medium containing 1% FCS, and cells were infected with 1.5µl of lentivirus (MOI=10). Four hours post-infection, 10% FCS medium was added, and cells were incubated for four days before selection with 4 µl/ml puromycin (Tivan Biotech). Following six additional days of incubation, surviving cells were seeded into ten 96-well plates at a concentration of 0.5cell/well.

Following a week of incubation, using a fluorescent microscopy, all wells were screened, and those containing only one colony were selected for further assessment of eGFP expression. Clones demonstrating high eGFP expression prior to IFNγ stimulation, potentially indicating plasmid insertion in an overexpressed genomic region, were excluded from further examination. Conversely, clones showing minimal or no eGFP expression were split into two groups and subjected to IFNγ stimulation. After stimulation with 20ng/ml mouse IFNγ (PeproTech) for 48 hours, eGFP mean fluorescent intensity (MFI) was measured via flow cytometry. The clone exhibiting the most significant response to IFNγ 48hours post-stimulation (LPA), was identified and chosen for the high throughput screening assay ( Figure 1 ).

A library comprising 1496 FDA-approved drugs was sourced from ApexBio (Houston, TX, USA). Each compound was initially dissolved in 100% DMSO at a concentration of 10 mM. For screening, each compound was used at a working concentration of 10 μM, with and without the addition of 20ng/ml mouse IFNγ. Testing was performed at 24-hour and 48-hour time points post-stimulation, utilizing four separate plates: no IFNγ for 24hours, no IFNγ for 48 hours, IFNγ for 24hours and IFNγ for 48 hours.

Compounds and/or IFNγ were dissolved in the cell medium and added to the culture 24 hours after seeding at 1*10^4 LPA cells per well. The final concentration of DMSO in the culture was 0.1%. Each plate consisted of 8 control wells: 2x LivMet, 2x LivMet+IFNγ, 2x LPA, and 2x LPA+IFNγ. Cells were harvested using 0.25% trypsin-EDTA (Sartorius), which was then neutralized with DMEM supplemented with 10% FCS. Following this, cells underwent three washes with 1xPBS to remove any remaining supernatant.

Finally, cells were resuspended in 1xPBS, and the eGFP MFI was measured using flow cytometry. Compounds that resulted in eGFP MFI values greater than 2 standard deviations compared to the averaged MFI in the specific plate were identified as “hits” that up-regulate PD-L1. Conversely, compounds that decreased eGFP MFI to a level lower than that of control IFNγ-treated LPA cells without any drug were considered “hits” that down-regulate PD-L1. This scoring approach was chosen due to the significantly higher averaged eGFP MFI of a plate compared to the IFNγ-treated control with no drug. Drugs were categorized according to their effect on the expression of IFNγ-dependent eGFP expression (under the control of the PD-L1 promotor) ( Figure 2 ).

All hits were subjected to further validation on LivMet cells across three concentrations to assess dose responses (0.1, 1, and 10μM). Initially, 1*10^4 LivMet cells were seeded per well in a 96-well plate, and after 24hours, IFNγ with either 0.1, 1, or 10μM of each drug were added. Cell supernatants were collected 48 hours following treatment and stored at -20°C until further use. Additionally, cells were stained with an anti-mouse CD274 (PD-L1, B7-H1) antibody (eBioscience; CA, USA) for subsequent flow cytometry analysis. The supernatant was also tested for mouse CXCL10 secreted levels using a DuoSet enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems; MN, USA).

A human primary lung tissue was collected from 5 healthy lung donors. All donors signed an informed consent form, and the study was approved by the institutional ethics committee (see Ethics Declaration Helsinki number- HMO-21-235). Lung tissues were washed with ice-cold 1xPBS plus 1.6% penicillin/streptomycin. Tissue was sliced into 1 mm slices and centrifuged in 1500 rpm for 5 min in 4°C. Supernatant was removed, and cells were re-suspended in a 10 ml cell dissociation buffer comprising: 80% RPMI (Sartorius); 0.4% BSA (Sigma-Aldrich); 0.1% collagenase P (Sigma-Aldrich) and 0.01% DNase (Sigma-Aldrich). Cells were incubated in a 37°C water bath for 30 min and pipetted every 10 min until full cell dissociation of the tissue. Then, we filtered the suspension through a 70-micron filter and enzymes were deactivated with RPMI+10% FCS. After filtration, cells were centrifuged and washed with 10ml RPMI. Following an additional centrifugation round, cells were re-suspended in 500 μl of 1xACK (Gibco) for 2.5min and washed twice with 10ml RPMI. Primary lung cells were seeded in a collagen-coated tissue culture for 48 hours until adherence to plate. Then, medium was replaced with a fresh RPMI medium containing 10% FCS, 1% L-glutamine, 1% penicillin -streptomycin and 1% sodium pyruvate. Cells were collected with 0.25% trypsin-EDTA and 1*10^4 cells per well were seeded in a 96-well plate. 24 hours later, cells were treated with IFNγ and the candidate drugs. After 72hours of incubation, supernatant was collected for assessing CXCL10 and IL-6 levels with ELISA, and cells were washed with 1xPBS and split into two groups. One group was stained with anti-human Epcam (Biolegend), anti-human MHC1 (Biolegend) and anti-human PD-L1 (Biolegend), while the other group was stained with anti-human CD45 (Biolegend), anti-human MHC1 and anti-human PD-L1.

Human pancreatic cancer cell line Panc1 was used to test the effect of 11 selected drugs in-vitro in a human cell line. Panc1 cells were cultured in RPMI medium containing 10% FCS, 1% L-glutamine, 1% penicillin-streptomycin and 1% sodium pyruvate. 2*10^4 cells were seeded per well in a 96-well plate, and after 24hours, 20ng/ml human IFNγ with 1 or 10μM of each drug, in triplicates, were added. Cell supernatants were collected 48 hours following post-treatment and stored at -20°C until further use. Additionally, cells were stained with an anti-human CD274 (PD-L1, B7-H1) antibody (BioLegend) for subsequent flow cytometry analysis. The supernatant was also tested for human CXCL10 secreted levels using a DuoSet enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems; MN, USA).

For the DTH model we used female BALB/cOlaHsd mice (Envigo). Mice were sensitized over their shaved abdominal skin with 100µl of 2% 4-Ethoxymethylene-2-phenyl-2-oxazalin-5-one (Sigma-Aldrich) dissolved in acetone/olive oil [4:1 (vol/vol)] applied topically (day 0). DTH sensitivity was elicited 6 days later by challenging the mice with 20µl of 0.5% oxazalone reconstituted in acetone/olive oil, with 10µl being administered topically to each side of their right ears. Ear thickness was measured 24 hours after the challenge using a micrometer digital caliper (Mitutoyo Corp; Tokyo, Japan). Thickness of the left ear served as the control for each mouse (26). Positive control mice were subcutaneously injected with dexamethasone (Omega) (100ug/mouse in a total volume of 200µl) on day 5 following challenge (n=5-8 mice per experiment). Negative control mice were treated daily subcutaneously with 1xPBS. Clofazimine (Sigma-Aldrich) was administered by a gavage needle per OS (300mg/kg) on days 2,4,5 and 6 following challenge (n=7). Dinaciclib (Abcam) was injected intraperitonially (400ug/mouse/inj) on days 0, 3 and 6 following challenge (n=6 and 7 in two different experiments). Penfluridol (Sigma-Aldrich) was administered by a gavage needle per OS (10mg/kg/inj) on days 3,4,5 and 6 following challenge (n=5). Nefiracetam (Abcam) was injected intraperitonially (1mg/kg/inj) on days 3, 4, 5 and 6 following challenges (n=6). Baricitinib (Selleckchem) was administered by a gavage needle per OS (10mg/kg/inj) on days 3,4,5 and 6 following challenge (n=7). Ganetespib (Abcam) was administered IV (500ug/mouse) on day 6 following challenge (n=5). Cyclosporin A (Tocris) was administered intraperitonially (100ug/mouse/inj) on days 0, 3 and 6 following challenge (n=7). Glycopyrrolate (Sigma-Aldrich) was injected subcutaneously (1.5mg/kg) on days 3 and 6 following challenge (n=7). Deferasirox (Selleckchem) was administered by a gavage needle per OS (30mg/kg/inj) on days 3,4,5 and 6 following challenge (n=5). Pizotifen (Adooq Bioscience) was injected intraperitonially (10mg/kg/inj) on days 4, 5 and 6 following challenge (n=8). Axitinib (Sigma-Aldrich) was administered by a gavage needle per OS (10mg/kg/inj) on days 3,4,5 and 6 following challenge (n=7). Cinepazide maleate (Sigma-Aldrich) was administered intraperitonially (20mg/kg/inj) on days 3, 4, 5 and 6 following challenge (n=7). Experiments were conducted with the approval of the institutional animal care ethics committee (See Ethics Declaration). The percent change in ear thickness of each mouse was calculated. Subsequently, the fold change of each mouse compared to the averaged percent change of the control group in each experiment was determined. The mean fold change ± standard error of the mean was then calculated. A p-value < 0.05 was considered statistically significant.

Anti-mouse CD274 (PD-L1, B7-H1) antibody PE-conjugated clone MIH-5 (eBioscience™) was used at a 1:50 dilution and was added to the cells following a 1xPBS wash. Cells were incubated for 30min with the antibody and washed and data were collected on a CytoFLEX instrument (Beckman Coulter). MFI of PE was analyzed using CytExpert 2.4 software. The background was defined as the MFI of the isotype control rat IgG2a kappa Isotype Control (eBR2a), PE (eBioscience; CA, USA). eGFP MFI of LPA cells was measured by the laser 488 filter 525-40 of the CytoFLEX instrument and was analyzed using CytExpert 2.4 as well.

RNA was extracted from cells using an RNA-isolation reagent (TriReagent; Sigma-Aldrich) according to the manufacturer’s protocol. RNA was then treated with DNase (TURBO DNA-free, Ambion). mRNA quantity was assessed using Nanodrop (Thermo Scientific). Reverse transcription of RNA to cDNA was performed using the qScript™ cDNA synthesis kit (Quanta Biosciences; Gaithersburg, MD). qPCR was performed with PerfeCta^® SYBR^® Green FastMix^® ROX on a BioRAD CFX384™ Real-Time System according to the manufacturer’s protocol. Cycling conditions were 95°C for 20s, followed by 40 cycles of 95°C for 1s, and 60°C for 20s, 65°C for 5s. mRNA expression of PD-L1 (sense- TAATCAGCTACGGTGGTGCG; anti-sense- CTTCTCTTCCCACTCACGGG), CXCL10 (sense- GAGAGACATCCCGAGCCAAC; anti-sense- GGGATCCCTTGAGTCCCAC) and eGFP (sense- GGTCACGAACTCCAGCAG; anti-sense- CAGAAGAACGGCATCAAGG) were evaluated in triplicates and were normalized to the expression levels of the endogenous control HPRT1 (sense- AGGGCATATCCAACAACAAACTT; anti-sense- GTTAAGCAGTACAGCCCCAAA) and calculated according to the standard formula of 2^(-ΔΔCT), producing results as a relative quantification (RQ).

Data are presented as the mean ± SD. Statistical comparisons of the means were performed using two-tailed unpaired Student’s t-tests. Differences of p ≤ 0.05 were considered statistically significant.

While IFNγ activates the immune response by stimulating CXCL10 expression, it also increases the levels of the immune checkpoint ligand PD-L1 which suppresses the immune response. This dual effect negatively regulates the immune response. To investigate this phenomenon, we sought for cell types that upregulate the expression of PD-L1 and CXCL10 in response to IFNγ stimuli.

Screening of various murine cell lines revealed differences in the expression of PD-L1 and CXCL10 following activation by IFNγ. Among the cell lines tested were RENCA (carcinoma), 3LL (Lewis lung carcinoma), AB12 (mesothelioma), B16F10 (melanoma), and LivMet (a liver metastasis of pancreatic cancer cell line derived from KrasG12D/+ transgenic mice with pancreatic ductal adenocarcinoma (PDA) tumors). Notably, LivMet cells exhibited a substantial upregulation of both CXCL10 and PD-L1 upon IFNγ stimulation (4.26-fold and 20.83-fold, respectively), compared to untreated controls. In contrast, 3LL, RENCA, and AB12 cells showed lower levels of CXCL10 post-stimulation (407pg/ml, 281pg/ml and 509pg/ml, respectively) compared to control untreated cells (456pg/ml, 0pg/ml and 523pg/ml, respectively), while LivMet and B16F10 cells displayed elevated CXCL10 levels (1572pg/ml and 3145pg/ml, respectively) compared to control untreated cells (366pg/ml and 37pg/ml, respectively) ( Figure 3 ). Increased PD-L1 expression following IFNγ stimulation was evident in all tested cell lines.

Given the strong response of LivMet cells to IFNγ, with significant elevations in both PD-L1 and CXCL10 levels, we selected this cell line for our high-throughput screening (HTS) assay. This robust response makes LivMet cells particularly suitable for evaluating the effects of various drugs on IFNγ-dependent PD-L1 and CXCL10 expression. Our goal was to assess whether these cells could reliably exhibit both upregulation and downregulation of these markers in response to different compounds.

Using a combination of flow cytometry, qPCR, and ELISA measurements, we demonstrated that LivMet cells uphold the necessary molecular machinery for IFNγ-dependent PD-L1 and CXCL10 expression. mRNA and protein levels of PD-L1 were highly expressed 48 hours after IFNγ stimulation (44.7-fold, P val=0.006 and 18.22-fold, P val<0.0001; respectively), in addition to CXCL10 mRNA and protein levels which were elevated as well (54.65-fold, P val<0.0001 and 4.4-fold, P val<0.0001; respectively) ( Figure 4 ). This indicates their potential as a tool for assessing the impact of drugs on IFNγ-dependent immunity.

To facilitate the screening of a vast library of small molecules, we engineered an endogenous molecular sensor integrated into LivMet cells. This sensor, constructed within a lentiviral vector, pEZX-LvPF02 (GeneCopeia), features a reporter gene, i.e., eGFP, controlled by the murine PD-L1 (CD274) promoter and incorporating puromycin resistance (see Supplementary Figure 1 ). The expression levels of eGFP serve as an indication for PD-L1 activation. LivMet cells were transduced with the lentiviral vector, followed by single-cell cloning to isolate the most responsive clone exhibiting IFNγ-dependent eGFP expression (See Materials and Methods). The novel LivMet cell line is termed “LPA”.

To assess the efficacy of different compounds in modulating IFNγ-dependent PD-L1 activation, we utilized HTS approach on the newly established LPA cell line. Compounds altering PD-L1 expression were also evaluated for their impact on CXCL10 secretion. First, we validated IFNγ-dependent PD-L1 activation in LPA cells by examining eGFP expression. Fluorescence microscopy analysis revealed approximately a 40% increase in eGFP expression in LPA cells post-IFNγ stimulation ( Figure 5A ). Moreover, flow cytometry and qPCR analyses demonstrated that the expression levels of both PD-L1 and eGFP in LPA cells were elevated following IFNγ stimulation, with mRNA expression levels increasing by 42.6-fold and 2.45-fold, and protein levels by 20.83-fold and 3.9-fold, respectively ( Figures 5B–E ).

After confirming the suitability of LPA cells for identifying small molecules targeting IFNγ-mediated PD-L1 expression, we screened 1496 drugs from the DiscoveryProbe™ FDA-approved drug library (ApexBio) to evaluate their effect on IFNγ-dependent PD-L1 expression. Initially, all candidate drugs were screened at a concentration of 10 μM, and the eGFP MFI was evaluated 48 hours post-IFNγ stimulation. Additionally, drugs were assessed for their effect on eGFP MFI without IFNγ stimulation. This analysis revealed 130 hits, of which 48 drugs downregulated IFNγ-dependent eGFP expression, and 82 drugs upregulated it.

Following the identification of 130 potential drug candidates, we proceeded with validating their effect, down- or up-regulation of PD-L1 expression. This was achieved by performing a drug dose response on unmodified LivMet cells, in increasing drug concentrations of 0.1, 1 and 10μM, and PD-L1 staining, assessed via flow cytometry ( Figure 6 ). Of the 130 hits, 104 drugs were validated to alter PD-L1 protein levels in LivMet cells, demonstrating an 80% validity of the screening assay. Subsequently, 10 drugs among the validated 104 were found to be cytotoxic for LivMet cells (with less than or equal to 10% of live cells at 1μM concentration), and thus, were excluded from further analysis.

The quality of the screening was assessed by two factors: The Z’ factor and the signal-to-noise ratio. The Z’ factor, calculated as Z’ = 1 − 3 SD of positive control + 3 SD of negative control/|mean of positive control − mean of negative control|, was determined to be 0.694, indicating an excellent assay quality. Similarly, the signal-to-noise ratio (S/N = (mean signal – mean background)/SD of background) was calculated to be 71.2, further confirming the robustness of the assay and the effective separation of the distributions in the screening method.

We excluded drugs with minor effects on PD-L1 protein levels (Log2FC between -0.2 and 0.2), resulting in a selection of 50 drugs. Subsequently, these 50 drugs were further evaluated for their impact on CXCL10 expression in the supernatant of treated LivMet cells. The 50 hits were further classified into four groups based on their potential therapeutic utility ( Table 1 ):

Group 1: Comprises four drugs that downregulate both PD-L1 and CXCL10. These drugs hold promise for treating conditions such as viral infections, graft-versus-host disease (GVHD), inflammation, and autoimmunity, where promoting a less immunogenic environment is desirable.

Group 2: Consists of 38 drugs that upregulate both PD-L1 and CXCL10 expression. Group 3: Four drugs that upregulate PD-L1 without affecting CXCL10.

Groups 2 and 3 present potential candidates for treating immune-related diseases or disorders like hyperinflammatory syndrome in viral infections (such as in COVID-19), GVHD, and cancer, where the recruitment of cytotoxic T cells is pivotal, and PD-L1 overexpression enhances immunogenicity, warranting the use of anti- PD-L1 therapeutics in combination with identified drugs.

Group 4: Comprises four drugs that upregulate PD-L1 but downregulate CXCL10. These drugs are potential candidates for treating immune-related diseases or disorders such as autoimmunity, inflammation, and hyperinflammatory syndrome in viral infections (such as in COVID-19), where the aim is to suppress the recruitment and activation of cytotoxic T cells.

After their validation in mouse tumor cells (LivMet), we focused next on human cells. We assessed all four drugs of Group 1 (Ruxolitinib, Baricitinib, Dinaciclib and Ganetespib) that were shown to downregulate IFNγ-dependent PD-L1 and CXCL10 expression in mice cells, in addition to 7 representatively selected drugs from the second and third groups, in human normal primary lung cells. Representative drugs that upregulate IFNγ-dependent PD-L1 expression and either upregulate (Group 2: Glycopyrrolate, Axitinib, Nefiracetam, Cinepazide maleate) or have no effect on (Group3: Deferasirox, Cyclosporin A, Clofazimin) IFNγ-dependent CXCL10 expression. Group 4 was out of our focus in this study.

Out of the 11 tested drugs, we found that only the drugs of Group 1, Ruxolitinib, Baricitinib, Dinaciclib and Ganetespib, demonstrated significant inhibition of IFNγ-dependent PD-L1 (59.4 ± 0.62, 70.7 ± 1.63, 46.2 ± 1.93 and 41.29 ± 1.68% inhibition, respectively) and CXCL10 expression (97 ± 2.9, 99 ± 0.2, 100 ± 0.25 and 98 ± 0.62% inhibition, respectively) in Panc1 human pancreatic cancer cells ( Figure 7 ) compared to cells treated only with IFNγ. This effect was previously observed in mice cancer cells.

We further tested the effect of the Group 1 drugs on human primary lung cells. Treatment with Ruxolitinib (a JAK/STAT pathway inhibitor which is known to have an effect on IFN-γ pathway), Dinaciclib and Ganetespib significantly inhibited IFNγ-dependent PD-L1 expression in human normal primary lung CD45+MHC1+ immune cells (72.5, 18.2 and 12.75% inhibition, respectively) and in human primary EPCAM-MHC1+ lung epithelial cells (81.3, 19.6 and 64.11% inhibition, respectively), and completely eliminated CXCL10 expression (100% inhibition for all drugs) compared to their expression levels in control IFNγ-treated cells.

The delayed-type hypersensitivity (DTH) model serves as a rodent model for studying inflammation mediated by soluble antigens, primarily involving the activation of CD4+ or CD8+ T cells. These reactions are characterized by the release of mediators from activated T cells, which subsequently stimulate local endothelial cells and recruit macrophages, resulting in localized inflammation and swelling.

To assess the impact of the hits on inflammation and immune cell recruitment in-vivo, we utilized the DTH mouse model in BALB/c mice. The mice were treated with ten selected compounds (out of the 11 tested on human cells). Our results showed that the only compounds significantly reducing ear swelling and immune cell infiltration in-vivo, were the three compounds that decreased both IFNγ-dependent PD-L1 and CXCL10 expression levels across all cell types tested in-vitro (because JAK/STAT pathway inhibitors are well known to have this effect on IFNγ pathway, we chose only one of the two JAK/STAT pathway inhibitors that we found in the screening as a positive control).

Compared to the control mice, we found three candidate drugs, all from Group 1, that decrease ear swelling: Baricitinib, 32% decrease (± 3%, P<0.0001); Ganetespib, 25% decrease (± 5%, P=0.01); and Dinaciclib decreased swelling by 60% (± 6%, P<0.001) ( Figure 8 ). These are comparable to the decrease in ear swelling by that Dexamethazone (± 10%, P=0.0001), used as positive control. Thus, we speculated that these compounds could serve as promising candidates for the treatment of inflammation, conditions characterized by cytokine storm, such as COVID-19, and autoimmune disorders.

Of the other seven compounds tested, none exhibited a significant effect on ear swellin.