IFNγ, and to a Lesser Extent TNFα, Provokes a Sustained Endothelial Costimulatory Phenotype

Background Vascular endothelial cells (EC) are critical for regulation of local immune responses, through coordination of leukocyte recruitment from the blood and egress into the tissue. Growing evidence supports an additional role for endothelium in activation and costimulation of adaptive immune cells. However, this function remains somewhat controversial, and the full repertoire and durability of an enhanced endothelial costimulatory phenotype has not been wholly defined. Methods Human endothelium was stimulated with continuous TNFα or IFNγ for 1-48hr; or primed with TNFα or IFNγ for only 3hr, before withdrawal of stimulus for up to 45hr. Gene expression of cytokines, costimulatory molecules and antigen presentation molecules was measured by Nanostring, and publicly available datasets of EC stimulation with TNFα or IFNγ were leveraged to further corroborate the results. Cell surface protein expression was detected by flow cytometry, and secretion of cytokines was assessed by Luminex and ELISA. Key findings were confirmed in primary human endothelial cells from 4-6 different vascular beds. Results TNFα triggered mostly positive immune checkpoint molecule expression on endothelium, including CD40, 4-1BB, and ICOSLG but in the context of only HLA class I and immunoproteasome subunits. IFNγ promoted a more tolerogenic phenotype of high PD-L1 and PD-L2 expression with both HLA class I and class II molecules and antigen processing genes. Both cytokines elicited secretion of IL-15 and BAFF/BLyS, with TNFα stimulated EC additionally producing IL-6, TL1A and IL-1β. Moreover, endothelium primed for a short period (3hr) with TNFα mostly failed to alter the costimulatory phenotype 24-48hr later, with only somewhat augmented expression of HLA class I. In contrast, brief exposure to IFNγ was sufficient to cause late expression of antigen presentation, cytokines and costimulatory molecules. In particular HLA class I, PD-1 ligand and cytokine expression was markedly high on endothelium two days after IFNγ was last present. Conclusions Endothelia from multiple vascular beds possess a wide range of other immune checkpoint molecules and cytokines that can shape the adaptive immune response. Our results further demonstrate that IFNγ elicits prolonged signaling that persists days after initiation and is sufficient to trigger substantial gene expression changes and immune phenotype in vascular endothelium.


INTRODUCTION
The blood endothelium is a highly active sensor that integrates multiple signals to control physiological vascular and immune function. It is well established that cytokine activation of blood endothelial cells (EC) triggers upregulation of numerous chemokines and adhesion molecules that promote adherence and extravasation of leukocytes. Recently, the capacity of lymphatic endothelium, and particularly lymph node-associated lymphatics, to also control innate and adaptive immune responses has been an increased subject of inquiry (1). Despite significant attention to the vascular endothelial cell's interaction with lymphocytes at the level of adherence, it has been a matter of contention whether blood EC possess sufficient costimulatory capacity to modulate the adaptive immune compartment beyond simply "recruiting" it to sites of injury.
Activation of naïve T cells requires engagement of the TCR by MHC class I or class II with cognate peptide and a specific repertoire of antigen-independent costimulatory molecules. Memory T cells are less reliant on costimulation but also require specific signals. Professional antigen presenting cells (APC) able to fulfill this capacity are of the hematopoietic lineage and include monocytes, macrophages, dendritic cells and B cells.
Like all nucleated cells, endothelial cells constitutively express MHC class I molecules, which is augmented by exposure to TNFa or interferons (IFN). Early experiments confirmed that allogeneic endothelium could activate bulk CD8 T cells (2), in a manner requiring cell-cell contact (3). In addition, human microvascular EC from at least some vascular beds may express HLA class II in situ (4) and respond to type II IFN by upregulation of MHC class II molecules. St. Louis et al. (5) described that IFNg-stimulated EC could trigger partial helper T cell responses and the authors hypothesized that "costimulatory deficient" endothelial cells could contribute to antigen-selective peripheral recruitment but not T cell proliferation. Subsequently it has been shown that although EC lack the capacity to activate naïve T cells, they can enhance activation of memory T cells (6,7). It was proposed that endothelial antigen presentation may function to selectively recruit antigen-specific T cells (8,9). Thus, endothelial immune functions, such as support of T cell activation, proliferation and/or differentiation, are continuing to be elucidated. The emerging paradigm is that EC can function as semi-professional or conditional antigen presenting cells, possessing a restricted repertoire of costimulatory molecules that can alter activation of some T cell subsets (10). Yet, the full endothelial collection of immune molecules that may shape the adaptive immune response has not been defined, particularly which repertoire is expressed under which inflammatory conditions (i.e. the context specific patterns of presentation).
Resolution of inflammation and attenuation of the immune response are critical to curb inappropriate chronic inflammation. Phased activation of endothelial cells is likely to play a role in terminating peripheral inflammation, certainly at the level of recruitment (11) but possibly also throughout effects on the adaptive immune compartment. Importantly, it is not known for how long after cytokine stimulation endothelial cells maintain an enhanced costimulatory phenotype.
In order to provide a comprehensive characterization of endothelial capacity to modulate the adaptive immune response, we focused on molecules that could fulfill the three main signals required for T cell activation: MHC/antigen presentation; cytokines; and costimulatory molecules. We examined multiple endothelial cell types and leveraged public datasets to provide several lines of evidence-both mRNA and protein, across diverse vascular beds of origin-for the expression or absence of these molecules in the endothelial compartment. We compared Th1 cytokines TNFa and IFNg stimulation of human endothelial cells, assessing baseline and inducible expression of 76 cytokines, antigen presentation and costimulatory molecules known to affect T cell activation. Next, we assessed how durable these alterations were using primewithdrawal experiments. Our results reveal overlapping and distinct antigen presenting phenotypes after TNFa and IFNg. Moreover, IFNg exerted an enduring shift in endothelial cell phenotype days after cytokine has been withdrawn, with longterm enhanced expression of HLA class I and class II, cytokine and coinhibitory molecules. Although the main effector molecules induced by IFNg appeared as a late phase response, only 3hr of direct exposure to IFNg was needed to provoke a prolonged endothelial cell phenotype change. Our results provide broader insight into the arsenal of antigen presentation molecules employed by endothelial cells, and particularly highlight the long-lasting effects of IFNg on endothelial cell immune function.

Prime-Rest Experiments
For transcript changes, endothelial cells (HMEC-1) and primary endothelial cells were stimulated with TNFa (20ng/mL) or IFNg (200U/mL) diluted in M199+10% FBS for 1hr, 3hr, 6hr, 18hr and 24hr. For protein changes, primary endothelial cells (HAEC) were stimulated with TNFa (20ng/mL) or IFNg (200U/mL) for 3hr, 6hr, 18hr, 24hr and 40hr. In prime-withdrawal conditions, endothelial cells were primed with TNFa or IFNg for 3hr, then medium was removed, cells were washed, and fresh M199+10% FBS without cytokine was added for the remainder of the experiment. Controls were treated and tested in parallel for each time point, i.e. when EC were primed for 3hr and then withdrawn for an additional 3hr, a parallel time matched 6hr continuous condition was also tested. All time points ended together, and cell material was collected for analysis concurrently.

Targeted Gene Expression Analysis
Stimulated endothelial cells were detached with Accutase (Sigma-Aldrich, #A6964), pelleted and resuspended in RLT Buffer (Qiagen) at 1mL/6500 cells. mRNA counts were measured by Nanostring (Human Immunology Panel v2.0, Nanostring Technologies) and analyzed in NCounter software. mRNA counts were normalized against internal and housekeeping controls. Normalized counts ≥250 were considered positive, and genes were considered changed if the counts differed by 50% (i.e. 1.5-fold) or more compared with baseline.

Cytokine and Chemokine Measurements
Conditioned medium was collected from flow cytometry and mRNA experiments and immediately stored in low protein binding tubes or multi-well plates at or below -20⁰C. Secreted cytokines in the supernatants were measured by ELISA (Human BAFF Quantikine ELISA R&D Systems #DBLYS0B; Human IL-15 Quantikine ELISA, R&D Systems #D1500) and Luminex (IL-6, IL-15, Milliplex MAP Human Cytokine/Chemokine 38-Plex panel, Millipore Sigma).

Statistical Analyses
Heat maps were generated in MORPHEUS (Broad Institute, https://software.broadinstitute.org/morpheus/). Data were graphed in Prism (GraphPad Software). Statistical differences between groups were calculated by unpaired or paired t test, or one-way ANOVA or two-way ANOVA followed by Fisher's LSD in Prism, as indicated in the figure legends.

RESULTS
T cells require three main signals for activation by an APC: cell surface antigen presentation in MHC/HLA; a specific repertoire of secreted cytokines; and costimulatory or coinhibitory ligands. We analyzed the kinetics and patterns of endothelial expression of each of these types of effector molecules.
For activation with TNFa, we analyzed two public transcriptome datasets in HUVEC, one with hourly TNFa stimulation from 1-6hr [GSE27870, n=3] and another of 39 donors stimulated with TNFa for 24hr [GSE144810]. For IFNg stimulation, we analyzed publicly available datasets of IFNgexposed EC for 2hr [GSE19082, HMEC, n=4] 5hr [GSE3920, HUVEC, n=5], or 3hr, 6hr, and 24hr [GSE106524, HLMVEC, n=3-4]. In addition, we generated our own dataset of an extended transcriptional time course of TNFa and IFNg stimulation (1-24hr) and also confirmed these results at the protein level in primary human endothelium from diverse vascular beds of origin.

TNFa and IFNg Induce Expression of More Than 40 Genes Involved in Antigen Presentation and Modification of Adaptive Immunity
We first characterized temporal transcriptional changes in human endothelial cells, after TNFa and IFNg stimulation.
Endothelial cells (HMEC-1) were left untreated or stimulated with TNFa (20ng/mL) or IFNg (200U/mL) for 1hr, 3hr, 6hr, 18hr or 24hr. Forty-two genes encoding antigen presentation, costimulation/coinhibition molecules or cytokines were upregulated in EC by TNFa, IFNg or both ( Figure 1A, Supplemental Table 4). On the other hand, 35 cytokines, antigen presentation and costimulatory genes were either not found to be expressed by endothelium or were expressed but not induced by TNFa or IFNg (Supplemental Table 5).
Ten genes were increased by both TNFa and IFNg to a comparable extent; 11 genes were more highly upregulated by IFNg than TNFa; 12 genes were IFNg responsive but unaltered by TNFa; and 9 genes were predominantly TNFa but not IFNg inducible ( Figure 1A, Supplemental Table 4). Nearly all antigen presentation genes under IFNg control occurred as late phase events, at 18hr or 24hr after stimulation, while many TNFainduced genes appeared earlier. Based on these kinetics, we confirmed mRNA induction of cytokines and costimulatory molecules in human primary endothelial cells from 6 other vascular beds (coronary artery, cardiac microvascular, pulmonary artery, lung microvascular, renal glomerular and dermal microvascular), after stimulation with TNFa 4hr ( Figure 1B) or IFNg 24hr ( Figure 1C). Fold change compared to an untreated parallel condition is presented in Figures 1B, C; discrete mRNA counts for each cell type are provided in Supplemental Figures S4-S10.

Time Course of TNFa-Induced Cytokines, Antigen Presentation Machinery and Costimulatory Molecules
Classical HLA genes HLA-A and B2M (b2-microglobulin) did not increase significantly in the first 6hr of TNFa treatment, yet the chaperone gene TAP1 and the immunoproteasome gene PSMB10 were significantly elevated as early as 3-4hr (Supplemental Figures S1A-D, GSE27870 in HUVEC n=3). In our own extended time course, HLA class I transcripts (HLA-A, HLA-B, HLA-C) (Figure 2A) as well as class I antigen presentation accessory molecules and immunoproteasome subunits (ERAP2, TAP1, TAP2, TAPBP, PSMB8, PSMB9, PSMB10) were increased by TNFa after 6hr, increasing to 24hr ( Figure 2B). At the protein level on primary aortic endothelium, HLA class I was constitutively expressed and further increased by TNFa as early as 18hr, rising through 48hr ( Figure 2C, left panel). Similar responses were seen across endothelial cells from 6 diverse vascular beds ( Figure 2C, right panel; Supplemental Figure S3). TNFa also upregulated endothelial non-classical HLA-E and HLA-F expression (2.0-fold at 18hr, data not shown), while HLA class II and related genes were not affected in primary aortic endothelium (n=5) and endothelium from other vascular beds (n=6) (data not shown and Figure 2D, Supplemental Figure S3).
Endothelial cells elaborated interleukins and other cytokines in response to TNFa stimulation. Two cytokines, CXCL12 (SDF-1) and IL32, were constitutively expressed by EC. Tonic CXCL12 (SDF-1) expression was substantially reduced by TNFa as early as 1hr ( Figures 3A, B) and remained durably suppressed through 24hr ( Figure 3A). Constitutive IL32 mRNA was also decreased in TNFa-activated endothelial cells (at 3hr), but then became elevated above baseline at 24hr (Figures 3A, C GSE144810). Raw mRNA counts are shown in Supplemental Figure S4A, S4D.
mRNA for the IL-6 family cytokine LIF as well as for PDGFB was increased early at 3hr but declined rapidly ( Figures 3D, E).  Figure S4J).
Although we detected a slight increase in transcript for CD274 (PD-L1) in HMEC-1 ( Figure 5A), this could not be reproduced in other datasets with primary umbilical vein endothelium ( Figure 5B) or other primary endothelium . HMEC-1 endothelial cells were stimulated with TNFa (20ng/mL) or IFNg (200U/mL) for 1hr, 3hr, 6hr, 18hr or 24hr. mRNA counts were measured by Nanostring and normalized to housekeeping genes and controls. Heat map shows relative gene expression over time, with normalization across genes/rows. Hierarchical clustering of rows is by one minus Pearson correlation. (B). Primary human endothelial cells (HAEC, HCAEC, HCMVEC, HPAEC, HPMVEC, HDMVEC, HRGEC, HLSEC) were stimulated with TNFa (20ng/mL) for 4hr, and transcript levels were measured by Nanostring. The fold change in normalized mRNA counts was calculated compared to untreated conditions for each EC type (n=8). (C). Primary human endothelial cells (HPAEC, HRGEC, HPMVEC, HLSEC) were stimulated with IFNg (200U/mL) for 24hr, and transcript levels were measured by Nanostring. The fold change in normalized mRNA counts was calculated compared to untreated conditions for each EC type. Endothelial cells did not express CD80 or CD86, two B7 ligand costimulatory molecules that are critical for activation of naïve T cells, in response to TNFa or IFNg (Supplemental Figures S2A-E). In addition, CD58 (LFA-3), which is known to be constitutively expressed on endothelium (3,23), was similarly unchanged by TNFa or IFNg treatment.
Taken together, endothelium responds to TNFa with augmented expression of numerous cytokines, including IL-6, IL-15 and BAFF, and costimulatory molecules, PD-L2, CD40, ICOSL, and 4-1BB, after TNFa stimulation, in addition to HLA class I molecules. Therefore endothelial cells express a constellation of molecules that can influence activation and differentiation of infiltrating leukocytes.
IFNg-induced class II accessory gene CIITA and CD74 (invariant chain) transcripts appeared as early as 3-6hr, which preceded HLA class II expression (HLA-DRA, DPA, DPB, HLA-DO, HLA-DM) at 18-24hr (Figures 7B, E-G and Supplemental Figure S7). Enhanced HLA class I protein expression on primary endothelial cells was seen as early as 18hr, and continued to increase through 48hr of continuous IFNg exposure (aortic EC: Figure 7H; 6 primary vascular beds: Figure 7I). HLA-DR protein expression was comparatively delayed, with low and inconsistent upregulation at 18-24hr and highest cell surface expression at 48hr (aortic EC: Figure 7H; 6 primary vascular beds: Figure 7I).
Among costimulatory molecules, IFNg promoted upregulation of CD274 (PD-L1) and PDCD1LG2 (PD-L2) transcripts by 3hr, with later and more modest CD40 and TNFRSF14 (HVEM) ( Figure 9A). We confirmed in 4 other FIGURE 3 | TNFa induced endothelial expression of cytokines. (A) HMEC-1 endothelial cells were stimulated with TNFa (20ng/mL) for 1hr, 3hr, 6hr, 18hr or 24hr. Time course of TNFa induced cytokines in HMEC-1. Time course of TNFa-suppressed cytokine mRNA for TGFBI, IL32 and CXCL12 in HMEC-1. One representative experiment is shown. Results are presented as absolute number of normalized mRNA counts for each gene, normalized to housekeeping controls. (B). Expression values of CXLC12 from GSE27870 are plotted, for HUVEC stimulated with TNFa for 1hr, 1.5hr, 2hr, 3hr, 4hr, 5hr and 6hr. Values at each time point were compared to baseline (0hr) by One way ANOVA followed by uncorrected Fisher's LSD t test. *p < 0.05; ***p < 0.001; ****p < 0.0001. Results are shown as mean ± SEM (n=3). (C) Expression values of IL32 from GSE144810 are plotted, comparing HUVEC (n=39) left untreated or stimulated with TNFa for 24hr. Value distributions were compared by unpaired t test. Results are presented with each measured value, and the line at the median. ****p < 0.0001. (D) Time course of TNFa early transientinduced cytokine mRNA for PDGFB and LIF in HMEC-1. One representative experiment is shown. Results are presented as absolute number of normalized mRNA counts for each gene, normalized to housekeeping controls. (E) Expression values of PDGFB from GSE27870 are plotted, for HUVEC stimulated with TNFa for 1hr, 1.5hr, 2hr, 3hr, 4hr, 5hr and 6hr. Values at each time point were compared to baseline (0hr) by One way ANOVA followed by uncorrected Fisher's LSD t test. *p < 0.05; **p < 0.01. Results are shown as mean ± SEM (n=3). (F) Expression values of PDGFB from GSE144810 are plotted, comparing HUVEC (n=39) left untreated or stimulated with TNFa for 24hr. Value distributions were compared by unpaired t test. Results are presented with each measured value, and the line at the median. ****p < 0.0001. (G) Expression values of LIF from GSE144810 are plotted, comparing HUVEC (n=39) left untreated or stimulated with TNFa for 24hr. Value distributions were compared by unpaired t test. Results are presented with each measured value, and the line at the median. ****p < 0.0001. (H) Time course of TNFa late phase induced cytokine mRNA for TNFSF10, TNFSF15, IL6 and IL1B in HMEC-1. One representative experiment is shown. Results are presented as absolute number of normalized mRNA counts for each gene, normalized to housekeeping controls. (I) Expression values of IL6 from GSE27870 are plotted, for HUVEC stimulated with TNFa for 1hr, 1.5hr, 2hr, 3hr, 4hr, 5hr and 6hr. Values at each time point were compared to baseline (0hr) by One way ANOVA followed by uncorrected Fisher's LSD t test. *p < 0.05; **p < 0.01; ****p < 0.0001. Results are shown as mean ± SEM (n=3). (J) Expression values of IL6 from GSE144810 are plotted, comparing HUVEC (n=39) left untreated or stimulated with TNFa for 24hr. Value distributions were compared by unpaired t test. Results are presented with each measured value, and the line at the median. ****p < 0.0001. (K) Expression values of TNFSF10 from GSE27870 are plotted, for HUVEC stimulated with TNFa for 1hr, 1.5hr, 2hr, 3hr, 4hr, 5hr and 6hr. Values at each time point were compared to baseline (0hr) by One way ANOVA followed by uncorrected Fisher's LSD t test. **p < 0.01; ****p < 0.0001. Results are shown as mean ± SEM (n=3  Figures S9C, F, G). Similar increases at the protein level were seen using 6 primary endothelial cells from different vascular beds ( Figure 9E, PD-L1: 3.40 ± 2.1-fold at 24hr, 3.64 ± 2.2-fold at 48hr; Figure 8H, PD-L2: 1.97 ± 0.5-fold at 24hr, 2.65 ± 1.2-fold at 48hr, n=6; raw MFI in Supplemental Figures S9B, E, G). Cell surface CD40 was also enhanced on aortic endothelial cells (1.94 ± 0.46-fold at 24hr, 2.51 ± 0.4-fold at 48hr, Figure 9J and S10C) and on 6 other primary endothelial cell types as early as 18hr ( Figures  9K, S10B).
Collectively, these results demonstrate that IFNg activation of endothelial cells results in late display of a restricted set of cytokines, IL-15 and BAFF, and costimulatory and coinhibitory molecules CD40, PD-L1 and PD-L2, in addition to extensive HLA class I and HLA class II antigen presentation machinery.

Short Exposure to TNFa Enhances Late Phase HLA Class I Expression by Endothelial Cells, But Not Cytokine or Costimulatory Molecule Expression
After defining the pattern of cytokines and costimulatory molecules expressed by cytokine-activated endothelial cells, we next assessed the timing of endothelial return to quiescence. After 3hr of TNFa cytokine priming, endothelial cells were withdrawn from cytokine for an additional 3-45hr. Expression of costimulatory factors was measured, compared with the same times in the continuous presence of cytokine.
We observed three distinct patterns after TNFa removal: 1) transcripts recovered to baseline after TNFa withdrawal (TNFSF13B, IL6; CD274, TNFRSF14, TNFRSF9, CD83, ICOSLG, CD40); 2) TNFa prime/withdrawal yielded intermediate induction, elevated above baseline but lower than continuous (TNFSF15, IL15, TNFSF10; HLA class I, proteasome components); or 3) TNFa prime/withdrawal resulted in  Time course of TNFa induced costimulatory molecule mRNA for TNFSF14, ICOSLG, TNFRSF9, TNFSF4, CD40, and CD276 in HMEC-1. Results are presented as absolute number of normalized mRNA counts for each gene, normalized to housekeeping controls. (B) Expression values of TNFRSF9 from GSE27870 are plotted, for HUVEC stimulated with TNFa for 1hr, 1.5hr, 2hr, 3hr, 4hr, 5hr and 6hr. Values at each time point were compared to baseline (0hr) by One way ANOVA followed by uncorrected Fisher's LSD t test. *p < 0.05; ****p < 0.0001. Results are shown as mean ± SEM (n=3). (C) Expression values of TNFRSF9 from GSE144810 are plotted, comparing HUVEC (n=39) left untreated or stimulated with TNFa for 24hr. Value distributions were compared by unpaired t test. Results are presented with each measured value, and the line at the median. ****p < 0.0001 compared to control. (D) Primary human (HAEC, HCAEC, HCMVEC, HPAEC, HPMVEC, HLSEC) endothelial cells were treated with TNFa (20ng/mL) for 4hr or 18hr. Cell surface expression of 4-1BB was measured by flow cytometry (n=6). **p < 0.01 by paired t test comparing 4hr to 18hr. (E) Primary human aortic endothelial cells (n=3 donors) were stimulated with TNFa for 24hr and 48hr. Cell surface 4-1BB was measured by flow cytometry. Results are presented as average fold increase in MFI of each molecule relative to untreated cells. (F) Expression values of ICOSLG from GSE27870 are plotted, for HUVEC stimulated with TNFa for 1hr, 1.5hr, 2hr, 3hr, 4hr, 5hr and 6hr. Values at each time point were compared to baseline (0hr) by One way ANOVA followed by uncorrected Fisher's LSD t test. *p < 0.05; ***p < 0.001; ****p < 0.0001. Results are shown as mean ± SEM (n=3). | Short priming with TNFa followed by withdrawal/rest in the absence of cytokine. (A) HMEC-1 endothelial cells were treated with TNFa (20ng/mL) for 3hr, 6hr, 18hr, or 24hr. In some conditions, medium was replaced after 3hr with medium without TNFa, and cells were allowed to rest for an additional 3hr, 15hr or 21hr. mRNA counts were measured by Nanostring and normalized to housekeeping genes and controls. Heat map shows relative gene expression over time, with normalization across genes/rows. Hierarchical clustering of rows by one minus Pearson correlation. (B-I) Primary human aortic endothelial cells were treated with TNFa (20ng/mL) for 3hr, 24hr or 48hr; or primed with TNFa for 3hr followed by an additional 21hr or 45hr without cytokine. Cell surface expression of (B) HLA-ABC,  and 4-1BB ( Figure 10C), remained elevated above baseline as long as 48hr later, albeit significantly lower compared with chronic TNFa stimulation. However, cell surface ICOSL ( Figure 10D), CD40 ( Figure 10E), and PD-L2 ( Figure 10F), and secreted IL-15 ( Figure 10G) all declined to untreated levels after TNFa was withdrawn. As with chronic stimulation, no changes in HLA-DR or PD-L1 were seen (Figures 10H, I).
Representative histograms are shown in Supplemental Figure S14.
Brief Priming With IFNg Is Sufficient to Trigger Long-Term Augmentation of HLA Class I, HLA Class II, Cytokines and Costimulatory Molecules We performed the same experiments testing priming of endothelial cells with IFNg for 3hr, followed by removal. Surprisingly, nearly all IFNg-induced costimulatory molecules and cytokines persisted long after IFNg withdrawal. Two patterns of transcript expression after IFNg withdrawal were observed: 1) IFNg prime/withdrawal caused intermediate gene expression, above untreated but lower than continuous (CD274, CD40; TNFSF13B, TNFSF10; HLA class II, CD74, CIITA; IDO1); or 2) IFNg prime/withdrawal resulted in equivalent potentiation as chronic stimulation (IL32, IL15, CXCL12; TNFRSF14, PDCD1LG2; HLA class I, TAP1, TAP2, TAPBP, PSMB8, PSMB9, PSMB10) ( Figure 11A). Only suppression of constitutive CXCL12 and IL32 was rapidly reversed when IFNg was removed ( Figure  11A). Normalized mRNA counts for genes of interest over time are shown in Supplemental Figures S15 and S16.
Most strikingly, late phase HLA class I and proteasome induction were equal whether IFNg was continuously present or had been removed after 3hr ( Figure 11A). At the protein level, cell surface HLA class I and class II expression were equivalent to continuous IFNg stimulation, up to 45hr later ( Figures 11B, C). Similarly, short IFNg priming triggered delayed expression of PD-L1 ( Figure 11D) and PD-L2 ( Figure 11E) that was comparable to chronic IFNg presence (PD-L1: continuous 2.48-fold vs. rest 3.05-fold; PD-L2: continuous 1.41-fold vs. rest 1.89-fold). Only CD40 ( Figure 11F) was slightly reduced at 48hr after IFNg pulse compared to persistent IFNg stimulation.
When EC were primed with IFNg for 3hr, then medium was replaced with cycloheximide without IFNg for an additional 21hr, cell surface expression of HLA-ABC and HLA-DR was abolished ( Figures 11G, H), suggesting that new transcriptional events after 3hr are required for protracted HLA expression by IFNg.
Lastly, secreted BAFF protein was increased 2.64 ± 1.3-fold at 24hr, and 6.37 ± 5.2-fold at 48hr of IFNg stimulation. Although no cytokine secretion was detected at the 3hr time point, there was no significant difference in the amount of BAFF secreted by EC 21hr and 45hr later when they had only been exposed to IFNg for 3hr (2.02 ± 0.65-fold at 24hr, 5.82 ± 1.9-fold at 48hr) ( Figure  11I). Similarly, IL-15 was elaborated up to 45hr later from endothelial cells pulsed for only 3hr with IFNg, with no significant difference compared to chronic IFNg presence (2.04 ± 0.2-fold at 48hr) ( Figures 11J and S16K).
These results demonstrate that brief priming of endothelial cells with type II interferon elicits a protracted costimulatory phenotype with elevated HLA class I and class II, costimulatory molecules and cytokine expression.

DISCUSSION
In this study, we characterized inducible endothelial expression of the three major signals required by T cells for activation: antigen presentation, cytokines and costimulatory molecules. We also investigated whether an altered endothelial cell phenotype was durable, or if it rapidly resolved in the absence of inflammatory stimulus. These questions are particularly relevant in the context of solid organ transplantation, where allogeneic T cells encounter HLA mismatched donor vascular endothelial cells and cause rejection of the transplanted tissue. T cell activation and peripheral antigen recognition within self HLA is also important in autoimmunity and atherosclerosis.
Endothelial activation by cytokines canonically leads to increased leukocyte-endothelial interactions due to upregulation of adhesion molecules and chemoattractants. In addition to actively coordinating recruitment of leukocytes, endothelial cells are strategically positioned to shape the immune response relative to local cues at the site of diapedesis. The question of endothelial antigen presenting capacity is particularly relevant in transplantation, where resident vascular cells from the donor express foreign HLA molecules and directly encounter the recipient's adaptive immune system. In the 1980s in situ HLA expression by endothelium in transplanted organs was therefore heavily investigated and confirmed (24)(25)(26)(27)(28). This was followed by seminal in vitro work from the Pober and Lechler labs showing that coculture of bulk allogeneic T cells with endothelium promoted T cell proliferation and acquisition of activation markers (23,(29)(30)(31), in a contact-dependent manner (3) and in the absence of exogenous mitogens (32). The effect was observed with both CD4+ and CD8+ T cells (2,33). Later studies resolved that endothelial cells could not prime naïve T cells and indeed may even elicit nonresponsiveness in the CD45RA+ population (6,(34)(35)(36). Instead, memory T cells were specifically activated by allogeneic endothelium (37)(38)(39)(40). The current paradigm is that endothelial cells are able to promote proliferation and activation at least of memory T cells, with further specific effects on differentiation and skewing still being revealed (particularly of Treg and Th17) (40).
In this study, we provide a broader classification of endothelial costimulatory molecule patterns under cytokine stimulation. For example, we report inducible endothelial expression of cytokines that are less well-described in the vascular compartment; some of which have been shown only a few times on EC and with unclear functional relevance.
Endothelial cells express high levels of HLA class I molecules. In situ, microvascular endothelial cells may also express HLA class II, although it is typically lost in cell culture. Like many other cells, can upregulate HLA class II molecules in response to type II interferon, as well as further enhance cell surface HLA  class I in response to TNFa or IFNg. HLA-DQB1 expression by endothelial cells is of high interest particularly in solid organ transplantation, where it is the most common target of donor specific alloantibodies associated with poor long-term graft outcomes. Moreover, polymorphisms in the HLA-DRB1 and DQB1 genes are strongly associated with and often causative in numerous autoimmune diseases. MHC accessory molecules critical for antigen processing and presentation are also upregulated by inflammatory cytokines in the vascular compartment. For example, we found low basal expression of TAP1 and TAP2, transporters essential for import of cytosolic peptide antigens into the endoplasmic reticulum; but these were substantially upregulated by TNFa and IFNg. Moreover, components of the immunoproteasome (PSMB8, PSMB9, PSMB10) essential for optimal proteolytic antigen processing for binding in the MHC class I groove (41), were specifically upregulated, while genes encoding the constitutive proteasome (PSMB7, PSMD7) were highly expressed and either unaffected or downregulated by cytokine treatment. Under IFNg stimulation, HLA class II-associated antigen presentation machinery were increased, including the master transcription factor CIITA, the invariant chain CD74, and the peptide chaperones HLA-DM and HLA-DO. Collectively, these changes point to a dramatic physiological shift in the function of endothelium towards antigen presentation u nder stimulation with Th1 proinflammatory cytokines. Which contact-dependent molecules are involved in the endothelial activating effect on T cells? Endothelial cells mostly lack expression of the B7 family CD28 ligands CD80 and CD86 (29,42), under either resting and stimulated conditions, which may account for the inability to stimulate naïve T cells that are heavily dependent on this signal. Yet, even this is controversial, with some early studies finding endothelial expression of CD86 (43,44); and more recently showing CD86 in specialized vascular compartments (45). In our results, no CD80 or CD86 expression was detected in cultured human endothelial cells, either resting or activated by TNFa or IFNg.
Initially endothelial ICAM-1 and LFA-3 were identified as important for the proliferative effect on T cells (30,46). LFA-3 (CD58) is constitutively expressed by endothelium (47), acting as a ligand for CD2 in costimulation for CD28-T cells (48) and mediating endothelial-T cell activation (23). While we did observe LFA-3 expression across endothelia, neither TNFa nor IFNg increased its expression. Therefore LFA-3 may function in basal T cell-endothelial cell interactions but not in inflammatory T cell activation.
Endothelial expression of CD40 is well-described (49). Engagement of endothelial CD40 by CD40L on activated T cells elicits a potent proinflammatory stress response in the endothelium (50,51). As others have reported (52,53), we found that both TNFa and IFNg modestly increased expression of CD40 transcript and protein on endothelial cells, where induction was greater with TNFa than IFNg. Thus, in addition to signaling to the T cell, increased expression of CD40 may further exacerbate endothelial cells to an inflammatory activation state.
CD83 was transiently increased in endothelial cells by TNFa but not IFNg, with relatively low but detectable mRNA counts. Interestingly, CD83 is a marker of mature dendritic cells, which appears to have a negative regulatory role on the function of both T and B cells and therefore may be critical in tolerance and resolution of inflammation (54,55). Other positive immune checkpoint molecules include ICOSL, OX40L, and 4-1BB. 4-1BB expression in the vasculature was a driver of inflammation in a murine model of atherosclerosis (56), and ICOSL expression by endothelium also mediated T cell costimulation (57). In our experiments, the stimulatory ICOSL and 4-1BB were induced on endothelial cells by TNFa but not IFNg.
Both TNFa and IFNg augmented endothelial expression of ligands for PD-1, a critical immune inhibitory checkpoint for T cells. Importantly, our data show that TNFa did not induce PD-L1, only PD-L2; while IFNg highly upregulated both. PD-L2 has a slightly higher affinity for PD-1 compared with the more broadly expressed PD-L1 (58) and had been thought to be restricted to antigen presenting cells (59). However, we and others have observed inducible PD-L2 expression on human endothelium. Functionally, PD-L1 and PD-L2 suppressed syngeneic CD8+ T cell activation by endothelial cells (60,61).
Finally, among soluble signals produced by endothelial cells, both TNFa and IFNg suppressed constitutive expression of CXCL12 (SDF-1) and IL32. IL-32 is relatively recently identified, and growing evidence supports its function as a proinflammatory cytokine. IL32 was cytokine-inducible in EC, and also likely has a function in vascular inflammation (62,63) as well as effects on the adaptive immune compartment, although its functions are still being elucidated. Within the lymph node, CXCL12 was uniquely expressed by blood but not lymphatic endothelium (1). CXCL12 (SDF-1) also enhances TCR stimulation as an additional costimulatory signal (64). It will be intriguing to determine the biological consequence of a switch from constitutive endothelial expression of these factors to other cytokines when initiating a pro-inflammatory program.
Other better described inducible cytokines produced by inflamed EC include IL6, LIF, TNFSF15 (TL1A) and TNFSF10 (TRAIL). TNFSF10 (TRAIL) presentation by APCs alters their immunostimulatory function, inhibiting TCR signaling when TRAIL-R on T cells is engaged (65). TNFa, but not IFNg, triggered weak endothelial expression of IL6 and LIF. IL-6 skews Th2 development and is also critical for Tfh and B cell antibody responses, while leukemia inhibitory factor (LIF) is an IL-6 family cytokine that plays a central role in T cell lineage development, particularly Treg vs. Th17 divergence (66).
We were surprised to find that activated endothelial cells elaborated BAFF/BLyS and IL-15, which we detected both at the mRNA level and secreted protein in the supernatant. Both TNFa and IFNg elicited expression of the well-known prosurvival factor for B cells BAFF. TNFSF13B (BAFF) secretion has been demonstrated from EC in the bone marrow, as well as in the settings of malignancy, autoimmunity and viral infection (17)(18)(19)67). Stromal endothelial cell-derived BAFF supported survival of leukemic cells in CLL (67), but there are few other reports of endothelium as a significant source of this cytokine. IL-15 was also reportedly produced by EC under the conditions of viral or autoimmune disease (20,21). It was recently demonstrated that EC transpresent IL-15 to CD8 T cells (22). In addition to effects on B cells, BAFF also stimulates T cells, including potentiation of TCR activation, T cell survival and proliferation (68,69). BAFF was recently implicated in the formation of tertiary lymphoid structures in the autoimmune disease lupus nephritis (70), and similarly BAFF expression correlated with formation of TLS in giant cell arteritis (71). IL-15 is an important cytokine for NK cells, and drives central memory CD8+ T cells. Therefore, local production of these factors may propel leukocyte activation and possibly influence the formation of TLS within peripheral tissues.
Our results illustrate the temporal phases of endothelial cell phenotype. It is intriguing that that the antigen presentation and costimulatory molecules induced by TNFa and IFNg appeared in the later stages of activation, whereas many chemokines and adhesion molecules characteristically appear earlier (3-6hr) as part of an acute and transient phase. This is suggestive of a multiphase response beginning with pro-adhesive recruitment functions and culminating in costimulatory or coinhibitory effects on the leukocyte compartment.
The significance of high endothelial HLA class I expression in the context of certain positive immune checkpoint molecules (i.e. in the context of TNFa) compared with high HLA I and HLA II but fewer available costimulatory molecules (i.e. IFNg context) is unknown. Moreover, the cumulative biological impact of these combined signals remains to be elucidated. Experiments are ongoing in our lab to elucidate how these cues are functionally integrated to tune activation and skewing of the adaptive immune compartment. Potentially the context (TNFa vs. IFNg) may drive disparate outcomes of inflammation or tolerance, stimulation or immunoregulation of the adaptive response. The specific functional contribution of each costimulatory molecule and cytokine expressed by endothelial cells on T cell activation and function has also yet to be fleshed out, and work is ongoing to elucidate the functional significance of endothelial costimulatory molecules on T cell activation.
Inappropriate prolongation and failure to initiate healing and return to non-inflamed homeostasis lies at the heart of numerous chronic diseases, including vascular diseases. Resolution requires termination of pro-inflammatory signaling, either through direct requirement for stimulation (i.e. passive, turns off when stimulus is gone) or active negative feedback and desensitization. We present novel data showing that the endothelial cell costimulatory phenotype remains altered for at least two days after last exposure to inflammatory cytokines. TNFa and IFNg patterns were distinct, where TNFa-inducible changes were mostly self-limiting over time, and the majority of which rapidly declined when TNFa was removed. IFNg-triggered changes manifested as late phase responses, and intriguingly IFNg effects on endothelial cells were extended without the requirement for continual stimulation. Unlike TNFa, withdrawal of the pro-inflammatory IFN signal does not alone contribute to resolution; but rather a stable program results in moderately persistent inflammation. We infer therefore that the perturbations caused by IFNg require more time to contract and restore endothelial cell homeostasis.
Prior studies mostly employed human umbilical vein endothelium (HUVEC), which are widely available and easy to propagate, but differ in important ways from arterial and capillary endothelial cells lining the vasculature of organs and tissues. We provide several lines of evidence by analyzing public datasets and testing multiple different human endothelial cell types, at both the transcript and protein level, that endothelial cells express an array of costimulatory molecules and cytokines. It is possible that there are tissue specific patterns as well, particularly at sites of entry to immune privileged tissue (cornea, brain, testes) and within the liver, which is widely viewed as tolerogenic. For example, liver sinusoidal endothelium may have a limited capacity to stimulate naïve as well as memory T cells, with a more tolerogenic effect compared with other vascular beds (42,(72)(73)(74). The results presented herein do support that costimulatory molecules are broadly induced on multiple endothelial cell types. Our initial study was not designed to address site-specific differences in endothelial heterogeneity of costimulatory molecules, but this will be an important outgrowth to pursue.
By technical necessity most EC-T cell costimulation studies have employed an allogeneic system, in which endothelial cells express mismatched MHC molecules that can activate T cells through direct allorecognition. In an indirect allorecognition model in vivo, it was shown that EC can effectively present alloantigen to MHC matched T cells (75). It will be important for future work to model both direct allorecognition with genetically mismatched cells, such as in the context of transplantation, as well as HLA syngeneic antigen presentation by endothelial cells, as in atherosclerosis, malignancy and infection.

CONCLUSIONS AND IMPLICATIONS
Although endothelial cells lacked CD28 ligands (CD80, CD86) and did not produce many of the typical cytokines needed for Th skewing (IFNs, IL-12, IL-21, IL-4), they could be provoked to express PD-1 ligands, CD40 and other cytokines that can bias T cell activation. That systemic and tissue resident vascular endothelial cells possess a wide constellation of costimulatory molecules suggests that adaptive immunity takes shape not only in secondary lymphoid organs but also locally in the periphery. The role of the local vasculature in initiating or propelling activation of the adaptive immune compartment will need to be a consideration in therapeutic approaches employing costimulatory agonism or blockade, such as cancer immunotherapy, autoimmunity and organ transplant immunosuppression.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
All experiments using human endothelial cells were approved by the UCLA Institutional Review Board (IRB#17-000477). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

AUTHOR CONTRIBUTIONS
NV is responsible for experiment conception and design, data collection and analysis, and manuscript preparation. The author contributed to the article and approved the submitted version.
FUNDING Support for this was provided in part by the Norman E. Shumway Career Development Award from the International Society of Heart and Lung Transplantation and Enduring Hearts (to NV).