EBV+ tumors exploit tumor cell-intrinsic and -extrinsic mechanisms to produce regulatory T cell-recruiting chemokines CCL17 and CCL22

The Epstein-Barr Virus (EBV) is involved in the etiology of multiple hematologic and epithelial human cancers. EBV+ tumors employ multiple immune escape mechanisms, including the recruitment of immunosuppressive regulatory T cells (Treg). Here, we show some EBV+ tumor cells express high levels of the chemokines CCL17 and CCL22 both in vitro and in vivo and that this expression mirrors the expression levels of expression of the EBV LMP1 gene in vitro. Patient samples from lymphoblastic (Hodgkin lymphoma) and epithelial (nasopharyngeal carcinoma; NPC) EBV+ tumors revealed CCL17 and CCL22 expression of both tumor cell-intrinsic and -extrinsic origin, depending on tumor type. NPCs grown as mouse xenografts likewise showed both mechanisms of chemokine production. Single cell RNA-sequencing revealed in vivo tumor cell-intrinsic CCL17 and CCL22 expression combined with expression from infiltrating classical resident and migratory dendritic cells in a CT26 colon cancer mouse tumor engineered to express LMP1. These data suggest that EBV-driven tumors employ dual mechanisms for CCL17 and CCL22 production. Importantly, both in vitro and in vivo Treg migration was effectively blocked by a novel, small molecule antagonist of CCR4, CCR4-351. Antagonism of the CCR4 receptor may thus be an effective means of activating the immune response against a wide spectrum of EBV+ tumors.


Introduction
The Epstein-Barr Virus (EBV) is one of the most ubiquitous known human viruses, with most individuals infected during childhood or adolescence [1,2]. EBV was also the first virus recognized as oncogenic in humans with the discovery of its role in the etiology of nearly 100% of endemic Burkitt's lymphoma (BL). Since that discovery, EBV has been identified as an important etiological factor in other B-cell lymphomas as well as T and natural killer lymphomas, and epithelial carcinomas including nearly 100% of nasopharyngeal (NPC) and approximately 10% of gastric carcinomas [3][4][5]. As of 2010, EBV + tumors were estimated to account for over 140,000 annual deaths globally, with particular impacts in Asia and Africa [6].
In cells latently infected with EBV, the viral genome has the coding potential for 70-80 genes, presenting the potential for numerous foreign antigens to be recognized by the immune system [7,8]. The fact that EBV + tumors are able to develop suggests that these tumors must have mechanisms for immune escape [8][9][10][11][12]. One such mechanism is the recruitment of regulatory T cells (T reg ) into the tumor microenvironment (TME). T reg are a subtype of CD4 + lymphocytes that suppress the activity of cytotoxic CD8 + T cells and dampen antitumor immune responses [13]. High T reg infiltrates in various EBV + tumors have been noted [14][15][16][17]. T reg have been shown to be recruited to the TME via C-C motif chemokine ligand 17 (CCL17) and C-C motif chemokine ligand 22 (CCL22; herein described together as CCL17/22) that are expressed directly by some lymphomas or by infiltrating immune cells within the TME [10,[18][19][20][21]. These chemokines are recognized by T reg -expressed C-C chemokine receptor type 4 (CCR4). In fact, a link between EBV infection and upregulation of CCL17/22 expression in lymphomas has been observed and mechanistically linked to the action of the viral latent membrane protein 1 (LMP1) [14,18,19,21]. LMP1 contributes to the transformation and survival of B cells by multiple pathways, including the activation of NFκB-putatively the mechanism for CCL17/22 expression in these cells [20,22]. While expression of these chemokines by immune cells is widely described, the mechanism for CCL17/22 expression in EBV + tumors of epithelial origin, such as NPC, is less clear.
Here, we provide further evidence for a link between LMP1 expression and CCL17/22 production in EBV + tumors and demonstrate that this production supports the migration of T reg cells in vitro and in vivo. That this migration was largely blocked by a novel small-molecule antagonist of the CCR4 receptor, CCR4-351 [23], highlights the importance of the CCL17/22/ CCR4 chemotactic axis in promoting T reg accumulation in these tumors. While RNA in situ hybridization (ISH) showed a strong link between EBV-positivity and chemokine expression in Hodgkin lymphoma, a mix of tumor cell-intrinsic and -extrinsic expression of these chemokines was revealed in NPC. Human NPC xenografts grown in immuno-deficient mice further demonstrated this mixed tumor-intrinsic and -extrinsic expression of CCR4 ligands. Finally, by evaluating a mouse colon tumor cell line, CT26, engineered to overexpress LMP1, we detected a marked increase in CCL17/22 production by both tumor and dendritic cells accompanied by an influx of T reg . Thus, we have observed varied but convergent mechanisms for CCL17/22 expression that could lead to a TME rich in T reg . Treatments that decrease T reg infiltration or activity, such as a CCR4 antagonist, may be effective immunotherapeutics against multiple EBV + tumor types.

Results
To explore the connection between EBV biology and the presence of T reg in EBV + tumors, we assayed the expression of LMP1, a viral protein reported to be involved in the upregulation of chemokine expression [19,20], and the related LMP2a by Western blot in 9 Burkitt's lymphoma-or Gastric carcinoma-derived cell lines. Of these, all but KATO III, Ramos, and NCI-N87 have been reported to be EBV + . ELISAs performed on the supernatants of these cell line cultures for the human chemokine proteins, CCL17, CCL20, and CCL22 revealed a very close match between LMP1 expression-observed in Jijoye, Raji, and NC-37 cell lines-and both CCL17 and CCL22 expression by these cells (Figs 1A and S1A). Pearson correlations of 0.96 (p value = 2.3e-6) and 0.95 (p value = 3.4e-5) were found between CCL17 and LMP1, and CCL22 and LMP1, respectively. No correlations with LMP2A were significant (p values > 0.1). CCL20 expression, which has been reported to play a role in EBV-recruitment of T reg [17] , was not detected in any sample and is omitted from the figure. LMP2A was detected in all cell lines previously reported as EBV + . NCI-N87, a gastric carcinoma tumor not previously reported as EBV + , also showed expression. This was further confirmed by PCR (S1B Fig), indicating that it actually is EBV + . Unlike LMP1, LMP2A levels did not mirror those of CCL17/22. CCL22 protein levels were proportional to the density of Raji cells in culture (S1C Fig). These results are consistent with prior observations suggesting a contributing role of LMP1 in the expression of CCL17/22 in cells latently infected by EBV, particularly in B cells.
To further dissect the role of viral proteins in CCL17/22 expression, we transfected Raji cells with LMP1-targeting siRNA (siLMP1), LMP2A-targeting siRNA (siLMP2A), or negative control siRNA. LMP1 protein levels were greatly reduced in the siLMP1-transfected and siLMP1/siLMP2A-cotransfected cells, with no effect of siLMP2A on its own (Fig 1B). Note that an LMP1 reduction by the LMP2A control siRNA was observed-these genes have overlapping transcripts that may explain this effect. LMP2A levels were greatly reduced by siLMP2A or siLMP1/siLMP2A, but not by siLMP1 siRNA. Levels of both CCL17 and CCL22 were significantly decreased by either siLMP1 or siLMP2, with stronger reduction in the siLMP1/ siLMP2A combination (Fig 1C). With the previous results, this suggests that while LMP2A expression is not sufficient for CCL17/22 expression in EBV-infected B cells, it does play a role along with LMP1. Interestingly, it has previously been shown that LMP2A cooperates with LMP1 for its activity [24][25][26].
In order to demonstrate that the CCL17/22 protein produced by Raji cells is functional, we measured in vitro chemotaxis. Control recombinant human CCL22 or supernatant from Daudi or Raji cultures was added to the bottom chambers of migration plates. CCRF-CEM CD4 + T-lymphoblast cells, which express high levels of CCR4, were added to the top chambers followed by quantitation of cell numbers in the bottom chambers after 1 hour. Both recombinant CCL22 and Raji supernatant induced migration of large numbers of CCRF-CEM cells (Fig 2A). This migration was mostly CCR4-dependent, as addition of CCR4-351, a highly specific CCR4 inhibitor, blocked most migration to both recombinant CCL22 and the Raji supernatant. As a more physiologically-relevant assay, the chemotaxis assay was repeated using in vitro polarized human CCR4 CD4 + cells biased to a T reg phenotype (iT reg ). A very similar pattern of CCL22-or Raji supernatant-dependent chemotaxis was observed with iT reg (Fig 2B). This chemotaxis could, again, be inhibited by CCR4-351, although the extent of inhibition of chemotaxis to Raji supernatant was less complete.
We sought to further validate this migratory connection between T reg and CCL17/22-producing EBV + Raji in an in vivo migration model. We injected Raji or Daudi cells into NOD-S-CID mice to generate tumors that grew comparably over the course of 30 days (S2 Fig). ELISA on 19 day tumors recapitulated these in vitro chemokine expression patterns: strong CCL17/ 22 expression was detected in the Raji tumors while Daudi tumors were mostly negative ( Fig  2C). We transferred human iT reg into a parallel set of mice 20 days post-tumor cell injection. Seven days post-iT reg transfer, we harvested tumors, and quantitated intra-tumoral iT reg frequency. Since tumor and iT reg both expressed human CD45, migrated iT reg were scored as the fraction of hCD45 + cells that were hCD4 + hCD19 -. Migrated iT reg were approximately 3% of hCD45 + cells in Raji tumors but were absent in Daudi tumors ( Fig 2D). Daily treatment with 50 mg/kg CCR4-351 for 7 days starting 3 hours before iT reg transfer resulted in an 81% decrease of migrated iT reg , again demonstrating the key role of the CCL17/22/CCR4 axis in this biology.
EBV-expressed LMP1 mimicry of B-cell CD40 activity is a putative mechanism for upregulation of CCL17/22 in these cells [27,28]. However, upregulation of CCL17/22 is not a reported physiological process in epithelial cells. However, increased T reg have been reported in tumors such as EBV-associated gastric carcinoma (GC) [14] and nasopharyngeal carcinoma (NPC) [16,29]. To put this T reg increase into context, two published NPC RNA-Seq expression data sets were combined with data on thousands of samples from 32 solid tumor types from the Tumor Cancer Genome Atlas (TCGA) and the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) databases. We further subset EBV + GC (GC_EBV) from the TCGA GC samples based on prior annotation [30]. We grouped these samples by tumor type and examined FOXP3, CCL17, and CCL22 expression levels. NPCs and EBV + GC had the first and second highest median expression of FOXP3, a T reg marker, as well as elevated CCL17 and CCL22 compared to the other tumors (Fig 3). Examination of control genes and sample clustering confirmed that the NPC and EBV + GC samples were broadly similar to other tumors (S3 and S4 Figs). The correlation between CCL17+CCL22 and FOXP3 expression within NPC and EBV + GC tumors (Pearson's r = 0.59 and 0.55, respectively) was similar to the correlation across all the TCGA/TARGET samples (r = 0.59).
This high CCL17/22 and FOXP3 mRNA expression in EBV + epithelial tumors is reminiscent of that in EBV + lymphomas [19,21], but this could be due to different underlying human cell lines were probed for LMP1 and LMP2A. Supernatants from these cell lines were assayed by ELISA for CCL17 and CCL22 protein. Chemokines were measured in biological duplicates with error bars at ± 1SD. Westerns were quantitated by densitometry and shown at an arbitrary scale with LMP2A levels multiplied by 500 relative to LMP1 for visibility. Cell lines are ordered by CCL22 expression. (B) Western blots on 50 μg of protein lysate from Raji cells 72 hours post transfection with test or control LMP1, LMP2A, or LMP1+LMP2A siRNA were probed for LMP1 and LMP2A protein production, or for respective loading controls, HSP90 and Actin. (C) CCL17 and CCL22 levels in the supernatants of siRNA-transfected Raji cell lines were measured by ELISA. https://doi.org/10.1371/journal.ppat.1010200.g001

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors mechanisms. We profiled 52 Hodgkin lymphoma (HL) biopsies and 15 NPC biopsies by RNA in situ hybridization on the RNAscope platform [31], probing for EBER1, a constitutively expressed EBV transcript, along with CCL17, CCL22, and FOXP3. Twenty-seven percent of the HL samples, representing 15 tumors, were EBV + by EBER1 staining (representative

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors positive staining is shown in Figs 4A and S5). These showed chemokine expression coincident with EBER1 in the HL Reed-Sternberg cells, with 22 out of 26 cores having statistically-significant coexpression of EBER1 with CCL17 and 21 cores with coexpression of EBER1 with CCL22 (FDR < 0.05 by chi-square). The NPC samples also showed chemokine expression associated with EBV-positivity, but quantitation was not possible due to the intensity of the EBER1 staining (representative image in Fig 4B). NPC chemokine expression was not exclusive to EBER1 + cells and it was difficult to determine the fraction of CCL17/22 expression that was tumor-extrinsic. CCL22 expression was more clearly observed in the FOXP3/ CCL22-stained NPC sections ( Fig 4C). FOXP3 + cells were seen in the vicinity of CCL22-positivity, which had a punctate pattern throughout the tumor. To discriminate between chemokine expression sources, human NPC tumors grown as mouse xenografts were probed for EBER1 and human or mouse CCL22 transcripts. Of the four xenografts analyzed, C15, C17, C18, and C666-1, only C15 expresses LMP1 [32][33][34][35]. Most cells in the C15 (Fig 5A), C17 (S6A  Fig 5B) xenografts stained positive for EBER1. Varying levels of hCCL22 expression throughout the xenografts were observed in C15, C17, and C18, while C666-1 showed mostly punctate expression along with some very-strongly CCL22-expressing cells ( Fig 5B). Mouse CCL22 was observed in all samples, but rarer in C666-1, and mostly

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors confined to regions of EBVcells (Fig 5A and 5B, right panels). These data demonstrate a mixed pattern of tumor-intrinsic and -extrinsic CCL17/22 expression in NPCs.
To further dissect EBV + epithelial tumor biology, we engineered the mouse colon tumor line CT26 to express LMP1 (CT26-LMP1) as well as a control for the increased antigenicity of LMP1, CT26 expressing chicken ovalbumin (CT26-OVA). None of the cell lines produced mouse CCL17 or CCL22 protein in vitro (CCL22 shown in Fig 6A). Tumors formed by CT26 cells contained a modest amount of CCL22, which increased in CT26-OVA tumors, and increased further in CT26-LMP1 tumors ( Fig 6A). GFP-marked mouse iT reg were transferred into these tumor-bearing mice for in vivo migration. Seven days post-transfer, virtually no iT reg were observed in CT26 tumors and a modest number in CT26-OVA tumors, while a marked increase in iT reg infiltration occurred in CT26-LMP1 ( Fig 6B). This was completely abrogated by dosing mice with the CCR4 antagonist, CCR4-351. No changes in mouse iT reg migration to spleen were observed in any of these conditions (S9 Fig).
CT26, CT26-OVA, and CT26-LMP1 tumors were subjected to single cell RNA-sequencing and epitope labeling using the 10x Genomics RNA and CITE-Seq platform [36]. Four major classes of cell types were observed in these tumors-tumor, stroma, lymphoid, and myeloid cells-based on examination of cluster-specific genes, and visualization by uniform manifold approximation and projection (UMAP; Figs 7A, S10-S12, S2 and S3 Tables). CCL17/22 expression was observed in only tumor and myeloid cells (Figs 7A, S10 and S12). Myeloid expression was primarily restricted to classical tissue-resident dendritic cells (cDC) and migratory dendritic cells (mDC) as defined by Binnewiess et al [37] and Miller et al [38] (Fig 7A). These mDCs also matched the expression pattern described for "mature DCs enriched in immunoregulatory molecules" (mregDCs) [39] (S13 Fig, S4 Table). CCL17/22 expression was also observed in a small number of macrophages (S14  (Fig 7D). In CT26-LMP1, the small number of mDC cells was responsible for the majority of chemokine expression. This combination of CT26-LMP1 tumor cell-intrinsic expression and tumor cellextrinsic expression from infiltrating myeloid cells, particularly mDCs, is reminiscent of the mixed expression in the NPC xenografts.

Discussion
The treatment of tumors with immune-based therapies, known as immuno-oncology (IO), holds great promise. There are a multitude of approaches from protein therapeutics such as the approved anti-PD-1, anti-PD-L1, and anti-CTLA-4 checkpoint antibodies to cell-based

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors therapies and small molecule therapies. These therapies are each expected to be active only against tumors with particular immunologic or antigenic phenotypes. We would expect that EBV-driven tumors, which should be particularly immunogenic due to the expression of foreign viral antigens, to employ particular immune-evasive strategies that could be targeted by

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors matched IO approaches. Support for this conjecture comes from multiple observations of high levels of infiltrating T reg in different types of EBV + tumors (this work and [14,16,17,29,40]). T reg are a type of CD4 + lymphocyte that tempers inflammation by suppressing the activity of cytolytic CD8 + T cells. A pan-tumor analysis shows that T reg levels track those of CD8 + cells, suggesting an adaptive immune resistance mechanism that acts as a negative feedback process in the TME. Thus, reducing the activity or number of T reg may help to drive antitumor immune responses in these tumor types.
The chemokines CCL17 and CCL22 are potent activators of the chemokine receptor CCR4, and this CCL17/22/CCR4 axis may be the major mechanism for recruiting T reg into tumors (this work and [41][42][43]). Although TGFβ can support the conversion of CD4 + T cells to T reg as well as their subsequent proliferation, existing data suggests that migration rather than conversion/expansion is the key driver of T reg numbers in tumors [23,44]. Here, we show that out of a variety of human EBV + tumor-derived cells lines, only those of B-lymphoma origin that express LMP1 also secreted both CCL17 and CCL22 at high levels in vitro. This is in agreement with other work such as in age-related EBV + B-cell lymphoproliferative disorder (ALPD) [19]. Reducing LMP1 mRNA levels in these cells via siRNA reduced CCL17/22 expression. LMP2A, which has been shown to cooperate with LMP1 for its activity [24,25], also affected CCL17/22 production but did not appear to be sufficient in these cells. Since LMP1 is believed to mimic CD40 activation in B cells [27,28], a process which normally leads to CCL17/22 expression, there is a clear mechanism driving this pathway.
Less clear has been the mechanism for CCL17/22 expression in EBV + tumors of epithelial origin. The EBV + gastric carcinoma cell line we tested, NCI-N87, did not produce either chemokine in vitro and RNA expression data for the NPC cell line C666-1 showed barely detectable chemokine levels [45]. However, we observed that the most common EBV + epithelial tumor types, EBV + gastric carcinoma and nasopharyngeal carcinoma, are among the highest expressors of both chemokines. Strikingly, these appear to also be the highest T reg -infiltrated tumors based on levels of FOXP3 expression across nearly 10,000 disparate samples. Matching these tumor expression data, we observed chemokine expression by RNA ISH in NPCs and in NPC xenografts. Unlike the expression pattern we observed in Hodgkin lymphomas, where CCL17/22 expression was strongly linked to EBV-positivity, the NPCs showed chemokine expression that appeared to be a combination of tumor cell-intrinsic and expression by tumorinfiltrating EBVcells. A mixed human/mouse xenograft system allowed us to more cleanly observe the sources of chemokine expression. In fact, both human and mouse CCL22 expression was observed, and this expression was localized, respectively, to tumor cells and regions of host cell infiltration.
LMP1 is not expressed in all NPCs [46,47] and is detected in only one of the NPC xenografts we tested (C15; [32]). Interestingly, LMP1-negative NPCs have been shown to have genetic alterations that can mimic the effects of LMP1 expression such as the genetic TRAF3 inactivation in C666-1 cells that mimics TRAF3 sequestration by LMP1 [48][49][50]. The betterestablished NFκB-activating role of LMP1 in EBV + lymphomas led us to test the effect of engineered LMP1 expression on CCL17/22 expression in an exemplar epithelial tumor. We engineered the mouse colon cancer cell line CT26 to express LMP1 and assayed its chemokine production. Although no CCL17/22 expression was observed in vitro by either the parental

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors CT26 cell line or by CT26-LMP1, strong chemokine expression was seen in vivo with CT26-LMP1 when the cell lines were grown as syngeneic tumors. Examining the tumors by single cell RNA-Seq revealed expression from the CT26-LMP1 tumor cells themselves as well as a major contribution from dendritic cells, especially from migratory DCs. These migratory DCs are a class of myeloid cells that bring antigen from peripheral tissues to lymph nodes [51], and their increased expression of the suppression-related chemokines CCL17 and CCL22 may play an important role in changing immune responses to EBV-infected tumors. DC migration is complex and only partially understood, with a variety of chemokines, including CCL19 and CCL21, and small molecules such as leukotriene B4 and oxysterols (which are bound by the intriguingly-named Epstein Barr Virus Induced Gene 2 receptor) acting as possible DC attractants [52,53]. Although we did not observe significant changes in the best known chemokine DC attractants, their general low levels of expression and restriction to specific cell compartments may have led to a lack of sensitivity in detection of such expression changes.
Edwards et al. grew AGS, a human EBVgastric cancer cell line, in vitro and in vivo in mice, along with an LMP1-expressing engineered version of AGS and an EBV-infected version of AGS as well as the human NPC xenografts C666, C15, and C17, and compared these cell lines and conditions by transcriptional and protein profiling [54]. While these experiments comparing the properties of EBV-related cell lines in vivo and in vitro are broadly similar to the experiments described herein, Edwards et al. did not identify the same biological processes we did. While one would hope that an upregulation of CCL17/22 would have been corroborated in that study, it is important to consider the major differences between these two studies: Edwards et al. looked at human cells in immune-deficient mice while we analyzed a mouse cell line in immune-competent mice; Edwards et al. used bulk, rather than single cell, RNA Sequencing that would likely miss changes due to rarer cells such as chemokine-expressing DCs; the differential gene expression analysis reported in Edwards et al. focused on the human gene expression changes as opposed to changes in the tumor-infiltrating mouse cells. Edwards et al. serves as an informative study of the changes to tumor-intrinsic pathways triggered by EBV, such as in vitro regulation of gene transcription by miRNAs [54].
Although the experiments reported here and mechanistic studies [19,20,50] show how EBV-derived LMP1 can lead to increased chemokine expression, the mechanism for the tumor-extrinsic increase in CCL17/22, shown here in a variety of settings, is less clear. We have shown that LMP1 expression leads to both an increase in the number of chemokineexpressing DCs and the level of chemokine expression in these cells. In contrast, no significant increase in mDC or cDC number was observed in the antigenicity-control model, CT26-OVA, and a more modest increase in chemokine expression was observed. Whether the difference between CT26-OVA and CT26-LMP1 is one of quantity, with perhaps LMP1 being more antigenic than OVA, or quality, where LMP1 participates in a specific biological pathway, is unknown. Regardless of which or both of these mechanisms are in play, we postulate that the net effect of EBV-LMP1 is a marked increase in CCL17/22 production that fosters an immune-suppressive environment beneficial to the EBV + tumor. Further, regardless of mechanism, antagonism of CCR4 by CCR4-351, a novel, oral specific small-molecule inhibitor, completely blocked the new infiltration of T reg -polarized cells in our mouse model.
In summary, multiple lines of evidence suggest that suppression of productive inflammation by T reg may be a common mechanism employed by EBV + tumors of both lymphocytic and epithelial origins. In both cases, tumor-produced chemokines CCL17 and CCL22 trigger T reg migration by activating the CCR4 chemokine receptor, and, in both cases, the viral protein LMP1 may be central to this process. In lymphomas, LMP1 has been shown to coopt existing B cell-intrinsic pathways to directly upregulate the chemokines. In contrast, chemokine production in epithelially-derived EBV + tumors is likely due to a combination of both tumor-intrinsic and tumor-extrinsic mechanisms. LMP1 and/or other viral proteins may lead to the indirect production of CCL17/22 in EBV + tumors via recruitment of infiltrating cells such as dendritic cells followed by tumor-intrinsic CCL17/22 expression in response to the activity of these infiltrating immune cells. These data suggest that blocking the T reg CCR4 receptor, such as with the selective antagonist CCR4-351, may be an effective way to potentiate antitumor inflammation and be an important part of a pan-EBV + tumor therapy.

Ethics statement
Propagation in nude mice was done with the approval of the Gustave Roussy Ethics Committee for Animal Experimentation (APAFIS#1605-2015090216498538v2 -November 26, 2015).

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors with enhanced chemiluminescence reagent. Densitometry was done with Li-Cor Image Studio Lite software.

Human T reg in vitro generation
Human T reg -polarized CD4 + cells (induced T reg ; iT reg ) were generated as previously described [42]. Routinely, >90% of CD4 + cells expressed Ccr4 and 30-60% expressed FoxP3. iT reg suppressed CD8 + T cell activation at levels comparable to natural T reg isolated from human PBMCs.

PLOS PATHOGENS
Multiple modes of CCL17 and CCL22 expression by EBV + tumors

In vitro chemotaxis
Assays were performed using the ChemoTX migration system with a 5 μm pore size PCTE membrane (106-5, Neuro Probe). CCRF-CEM cells were resuspended at 2x10 6 cells/mL in human serum. CCR4-351 (300 nM) or DMSO were added to a DMSO concentration of 0.25% (v/v) followed by a 30-minute preincubation. 29 μL of recombinant hCCL22 (diluted to 0.9 nM in 1xHBSS with 0.1% BSA) or supernatant from cultured cells was dispensed in the lower wells. PCTE membrane was placed onto the plates and 50 μL of the CCRF-CEM cell/compound mixture was transferred on top. Plates were incubated at 37˚C, 100% humidity, 5% CO 2 for 60 minutes, then the membranes were removed and 15 μL Cell Titer Glo was added to lower wells. Luminescence was measured using an Envision plate reader (PerkinElmer).

Mouse in vivo T reg migration
After CT26, CT26-LMP1, and CT26-OVA tumors reached 200-300 mm 3 , mice were given 50 mg/kg CCR4-351 or vehicle orally. Three hours later, mice were injected intravenously with GFP + iT reg at 97% purity and 27% CCR4-positivity. Tumors were harvested after 7 days, during which CCR4-351 was dosed orally daily, and incubated in digestion buffer with DNAse and lysed in Miltenyi C tubes using the Gentle MACS Octodissociator. A single cell suspension was prepared from spleens using syringes, filtered and stained for: TruStain FcX anti-mouse CD16/32 antibody (101320), CD4 APC Cy7 (100414)

Bulk RNA-Seq
Solid tumor TCGA and TARGET RNA-Seq datasets were downloaded from the UCSC Xena data hub [57] on June 18 th , 2017. NPC datasets GSE102349 [58] and GSE68799 were downloaded from NCBI GEO and processed with Kallisto [59]. Counts across all data sets were quantile-normalized using preprocessCore [60]. EBV status for GC was obtained from cBio-Portal [61]. S1 Table shows tumor-type abbreviations.

Single cell RNA-Seq
After tumors reached 200-300 mm 3 , they were collected and digested with collagenase buffer in a 37˚C bath, pipetting every 10 minutes to dissociate. Cells from 5 tumors were pooled per sample. Cell surface protein feature barcoding (CITE seq) antibody (BioLegend TotalSeq-A)

NPC Xenografts
C15, C17, and C18 are patient-derived xenografts (PDX) from cells propagated solely in nude mice [65,66]. The C666-1 NPC tumor line was first established as a PDX and later as a cell line propagated in vitro [67] and recently re-implanted in nude mice by one of us (PB).

Statistics
Significance tests used for data in plots are two sample Student's t-tests unless otherwise noted. Significance for LMP proteins and chemokines reported for Fig 1A was calculated by the R "stats::cor.test" function using Pearson correlations and two-sided hypothesis testing. A repeated measures ANOVA was used for data in Fig 2D. Significance codes for p-values are: ���� = p < 0.0001, ��� = p < 0.001, �� = p < 0.01, � = p < 0.05. Raji and Daudi Xenografts were established in NOD/SCID mice to measure chemokine production and iT reg migration (Fig 2C and 2D).  Table) were applied to the three DC-like clusters identified in this study. A robust Ztransform was used on the relevant genes across all myeloid cells, then the average Z scores for the genes in each signature were plotted for myeloid cluster 1 (migratory DC, mDC), myeloid cluster 7 (classical resident DC, cDC), and myeloid cluster 8 (plasmacytoid DC, pDC). The cDC cluster stood out for Binnewies signature 3 (resident CD11b + cDC2), Binnewies signature 7 (resident CD8a + cDC1), and the Miller cDC signature. The mDC cluster stood out for Binnewies signature 1 (migratory CD103 + cDC1), Binnewies signature 2 (Langerhans cell), Miller mDC, and the Maier mregDC signature. The pDC assignment is supported by expression of Tlr7 and Tlr9 in this cell cluster.