TGF-β1 potentiates Vγ9Vδ2 T cell adoptive immunotherapy of cancer

Summary Despite its role in cancer surveillance, adoptive immunotherapy using γδ T cells has achieved limited efficacy. To enhance trafficking to bone marrow, circulating Vγ9Vδ2 T cells are expanded in serum-free medium containing TGF-β1 and IL-2 (γδ[T2] cells) or medium containing IL-2 alone (γδ[2] cells, as the control). Unexpectedly, the yield and viability of γδ[T2] cells are also increased by TGF-β1, when compared to γδ[2] controls. γδ[T2] cells are less differentiated and yet display increased cytolytic activity, cytokine release, and antitumor activity in several leukemic and solid tumor models. Efficacy is further enhanced by cancer cell sensitization using aminobisphosphonates or Ara-C. A number of contributory effects of TGF-β are described, including prostaglandin E2 receptor downmodulation, TGF-β insensitivity, and upregulated integrin activity. Biological relevance is supported by the identification of a favorable γδ[T2] signature in acute myeloid leukemia (AML). Given their enhanced therapeutic activity and compatibility with allogeneic use, γδ[T2] cells warrant evaluation in cancer immunotherapy.


In brief
gd T cells mediate immune surveillance and represent an attractive option for cancer immunotherapy. Here, Beatson et al. show that when circulating gd T cells are expanded in the presence of TGF-b, they achieve enhanced bone marrow trafficking, tumoricidal activity, and cytokine release, enabling greater efficacy in leukemic and solid tumor models.
INTRODUCTION gd T cells protect against diverse human cancers. Reconstitution of this minority T cell population following hematopoietic stem cell transplantation for leukemia is strongly linked to extended survival. 1 Moreover, the presence of intratumoral Vg9Vd2 T cells is the most predictive leukocyte signature of improved outcomes across 25 cancers. 2 These clinical observations reflect two attributes of gd T cells. First, they undertake the human leukocyte antigen (HLA)-independent detection of meta-bolic, genomic, and signaling hallmarks of malignant transformation. 3 The predominant human circulating gd T cell population expresses a Vg9Vd2 T cell receptor (TCR) that recognizes mevalonate intermediates overproduced in tumor cells and known as phosphoantigens (PAg). 4 gd TCRs can also detect other tumor-associated antigens 5 and stress ligands. 6 A second key property of gd T cells is their ability to coordinate both innate and adaptive immunity. Accordingly, gd T cells can co-stimulate natural killer (NK) cells, 7 promote dendritic cell maturation, 8 and cross-present antigen to CD8 + ab T cells. 9 Circulating gd T cells can be induced to proliferate using aminobisphosphonate drugs (e.g., zoledronic acid [ZOL]), which stimulate Vg9Vd2 T cells. 10 Alternatively, antibody cross-linking of the gd TCR induces the expansion of all gd T cell subtypes. 11 However, adoptive immunotherapy using gd T cell products has only achieved a modest impact, mainly in hematological cancers. 12 This highlights the need for strategies that more effectively harness the versatile antitumor activity of these cells.
Transforming growth factor (TGF) b is a pleiotropic cytokine that can confer regulatory properties on ab and gd T cells. 13 Increasingly, however, it is evident that the effects of TGF-b are context dependent. An interleukin (IL)-17-producing phenotype is favored when IL-6, IL-1b, and IL-23 are also present. 14 TGF-b can also promote the differentiation of CD103 + Vg9Vd2 T cells with enhanced cytolytic activity against E-cadherin-expressing solid tumor cells. 15,16 Given the preeminent role of TGF-b in the tumor microenvironment, 17 this intriguing finding raises the possibility that Vg9Vd2 T cells expanded in TGF-b could exploit a key node of tumor-associated immunosuppression for therapeutic benefit. However, the antitumor activity of TGF-b-conditioned Vg9Vd2 T cells remains untested. Here, we have evaluated this using a range of solid and leukemic cancer models, in addition to exploring the underling mechanisms and relevance to Vg9Vd2 T cells found within human cancers.

RESULTS
Expansion of Vg9Vd2 T cells in serum-free medium containing TGF-b1 elicits a distinct immunophenotype with enhanced bone marrow migratory capacity We hypothesized that the antileukemic activity of Vg9Vd2 T cells could be harnessed more effectively if their bone marrow trafficking capacity was amplified. CXCL12 recruits CXCR4-expressing cells to bone marrow, 18 where endothelial E-selectin provides a portal of entry. 19 Consequently, we set out to increase CXCR4 and E-selectin ligand expression by Vg9Vd2 T cells.
When cultured in serum-free medium (SFM) 20 or TGF-b1, 21,22 ab T cells upregulate E-selectin ligands (e.g., cutaneous lymphocyte antigen [CLA]) and CXCR4. 23 To test whether Vg9Vd2 T cells respond similarly, peripheral blood mononuclear cells (PBMCs) were activated with ZOL, or anti-pan gd TCR, and Vg9Vd2 T cells were expanded in SFM containing IL-2 alone (gd [2] cells) or IL-2 + TGF-b (gd[T2] cells). In both cases, gd T cells enriched to equivalent purity ( Figure 1A). As expected, a greater proportion of non-Vd2 cells were present when antibody cross-linking rather than ZOL was used ( Figure S1A). We also found that Vg9Vd2 T cell yield was significantly greater when TGF-b was added ( Figure 1B). While we noted considerable donor-to-donor heterogeneity, CLA ( Figure 1C), CXCR4 ( Figure 1D), and calcium-dependent E-selectin binding activity ( Figure 1E) were significantly greater in gd[T2] cells. TGF-b also retarded differentiation ( Figure 1F), with increased CCR7 and CD27 and reduced CD45RA expression ( Figure S1B). gd[T2] cells were more activated, indicated by elevated CD25 and CD69 and reduced CD62L ( Figure S1C). Moreover, baseline apoptosis, necrosis ( Figure 1G), and camptothecin-induced death ( Figure S1D) were reduced in gd[T2] cells, accompanied by significantly increased expression of anti-apoptotic molecules (e.g., cIAP-1, XIAP) and a reduction in pro-apoptotic molecules (e.g., DR5, FADD) ( Figure S1E) and cytokines (e.g., tumor necrosis factor a [TNF-a], which promotes gd T cell apoptosis 24 ; Figure S1F). Given that TGF-b has been shown to inhibit Vg9Vd2 T cell proliferation, 16 we believe that the reduction in cell death accounts for the enhanced yield of gd [T2] cells. Further phenocopying the effect of TGF-b on ab T cells, 25 CD103 (a E integrin) was markedly upregulated on gd[T2] cells ( Figure 1H). Expression of the alternative E-cadherin ligand and late-stage differentiation/exhaustion marker KLRG1 was proportionately reduced ( Figure 1I). Although FoxP3 was detectable in gd [2] and gd[T2] cells, the levels were lower than in CD4 + regulatory T cells (Treg; Figures S1G and S1H). Moreover, neither gd[T2] and gd [2] cells suppressed the proliferation of activated CD4 + T cells ( Figure S1I). Production of IL-10, a cytokine produced by some regulatory Vg9Vd2 T cells, 26 was also significantly lower by gd[T2] cells (Figure S1J). We were unable to address whether TGF-b exerted these effects directly since the expansion of purified Vg9Vd2 T cells in SFM was not robust in our hands. Our data suggest that gd[T2] cells are more closely related to TGF-b-induced CD103 + cytotoxic Vg9Vd2 T cells 15 than are Vg9Vd2 Tregs. 13 We next evaluated the bio-distribution of infused gd[T2] cells in vivo. Studies were undertaken in mice since interactions involving CXCR4 27 and E-selectin 28 both cross the human/ mouse species barrier. Human gd[T2] cells were engineered to co-express firefly luciferase (ffLuc) and red fluorescent protein (RFP; Figure 1J) and were injected intravenously (i.v.) into NSG mice in which Jurkat-GFP leukemia had been established. To determine the combined effect of SFM and TGF-b, comparison was made with Vg9Vd2 T cells that had been expanded in serum-containing medium with IL-2 alone (gd[2S] cells; Figure 1J). Using bioluminescence imaging (BLI; Figure 1K) and flow cytometry ( Figure 1L), a significant increase in the migration of gd[T2] cells to bone marrow/femora, but not spleen, was seen at 24 and 48 h. At these early time points, antileukemic activity was not evident ( Figure 1M). Orthogonal demonstration of bone marrow entry was obtained by positron emission tomography/computed tomography (PET/CT) imaging of [ 89 Zr]Zr(oxinate) 4 -labeled gd[T2] cells. 29 Signal was clearly visualized in both femora and vertebrae ( Figures S2A and S2B), with confirmation by bio-distribution analysis ( Figure S2C).

gd[T2] cells demonstrate enhanced in vitro antileukemic activity
We next evaluated the efficacy of gd[T2] cells in leukemic models. Owing to the ease of production, ZOL-activated cells were used. Greater killing of U937 and KG-1 cells was mediated by gd[T2] cells compared to gd [2] cells (Figures 2A and 2B), potentiated by target pre-exposure to pamidronate (PAM) or ZOL. 30 Jurkat cells proved highly sensitive to killing by both gd [T2] and gd [2] cells ( Figure 2C). Activated gd[T2] cells released more interferon g (IFN-g) and IL-2 ( Figures 2D-2I) and maintained cytotoxic function upon repeated leukemic cell addition ( Figures 2J and 2K), unlike gd [2] cells.
Many cytotoxic drugs sensitize tumor cells to killing by gd T cells. 31,32 We observed that sublethal exposure to the antileukemic drug, Ara-C ( Figure S3A) rendered both U937 and KG1 cells more susceptible to killing by gd[T2] T cells, with further enhancement by ZOL ( Figures S3B and S3C). In contrast to PAM or ZOL, sensitization by Ara-C did not increase IFN-g release ( Figures S3D and S3E) or upregulate NKG2D ligands on leukemic cells ( Figures 3F and 3G). However, cleaved caspase 3 levels in U937 and KG1 cells were increased, suggesting that Ara-C lowered the threshold for Vg9Vd2 T cell-mediated killing ( Figures S3H and S3I Figure 3A). While pre-treatment with ZOL had a marginal effect, the combination of ZOL + Ara-C boosted response to gd[T2] cells, improving both disease control ( Figure 3A) and survival ( Figure 3B). By contrast, neither ZOL + Ara-C alone or combined with gd [2] cells was effective ( Figures 3A and 3B). We next evaluated sensitization by Ara-C alone and found that this also potentiated the efficacy of gd[T2] cells against Jurkat leukemia ( Figure S4A), leading to prolonged survival ( Figure S4B).
To test this in a more challenging setting, we selected the KG1 leukemia model. Despite in vitro sensitivity, KG1 xenografts are poorly responsive to gd[T2] or gd [2] cells ( Figure 3C). Nonetheless, when mice were sensitized with ZOL + Ara-C, gd[T2] cells delayed disease progression ( Figure 3C) and enhanced survival ( Figure 3D). gd[T2] cells also delayed disease progression (Figure S4C) and prolonged the survival of mice with an established U937 xenograft ( Figure S4D). gd[T2] cells undergo enhanced activation against solid tumor cells but require regional delivery for efficacy Given the enhanced antitumor activity of gd[T2] cells in models of hematological malignancy, we wondered whether enhanced function would also be seen in solid tumor models. When compared to gd [2] Figure 4B), accompanied by the increased release of several cytokines and chemokines (Figure S5B). Notably, the production of pro-tumorigenic IL-17 was negligible ( Figure S5B). 33 Neither cytotoxicity ( Figure 4C) nor cytokine release ( Figures 4D and 4E) by gd[T2] cells was suppressed by exogenous TGF-b1. Moreover, gd[T2] cells demonstrated preferential cytotoxic activity against transformed compared to non-transformed cell types ( Figure 4F).
To test in vivo function, mice with established orthotopic MDA-MB-231 tumors were treated with ZOL followed by i.v. gd [2] or gd [T2] cells after 24 h and IL-2 at 72 h. However, no therapeutic activity was elicited by either Vg9Vd2 T cell population in vivo ( Figure S5C). Given their altered trafficking properties, we hypothesized that gd[T2] cells had preferentially migrated to bone marrow rather than to tumor. To investigate this, we studied the trafficking of ffLuc/RFP-engineered gd[T2] cells after i.v. delivery to tumor-bearing mice. Using BLI, we observed the migration of infused cells to femora ( Figure S5D). Moreover, RFP + cells were retrieved from bone marrow but not tumor at 48 h (Figure S5E), indicating that influx at the site of disease was inadequate. These data also reaffirm the potential utility of gd[T2] cell immunotherapy in the treatment of bone marrow malignancy.
To circumvent the need for gd[T2] cell migration to solid tumors, we tested whether intraperitoneal (i.p.) delivery may prove useful in ovarian cancer treatment, given its tendency for locoregional rather than distant spread. In NSG mice with established i.p. ffLuc SKOV-3 tumors, treatment with gd[T2] cells alone or with ZOL resulted in improved disease control (Figures 5A-5C) and survival ( Figure 5D), compared to gd [2] cells. To test this in a second system, xenografts were established using ffLuc + Kuramochi cells, which represents the most representative cell line model of high-grade serous ovarian cancer. 34 Tumors were unstable, poorly vascularized, and partially sensitive to ZOL (Figure 5E). Nonetheless, co-treatment with gd[T2] cells led to a further reduction in disease burden ( Figure 5E).

Mechanistic investigation of gd[T2] cells
Bulk RNA sequencing (RNA-seq) was performed on flow sorted gd[T2] and gd[2] cells on days 9 and 15 of culture (n = 3 independent donors; Tables S1 and S2) to identify differentially expressed genes and pathways between these cells. A total of 55 genes (fold change [FC] R4 or %0.25 and false discovery  (Table S1), 29 of which were unique to this time, while 26 were still present on day 15. On day 15, 109 genes were noted, 82 of which were unique to this time ( Figure 6A). TGF-b1-responsive genes (e.g., ITGAE, LDLRAD4, MRC2, SMAD3) were modulated in the expected manner at both time points, confirming the validity of this approach. Nevertheless, there was considerable donor-todonor variability in gene expression, consistent with the heterogeneity of these ex vivo expanded Vg9Vd2 T cell cultures ( Figure 6B).
Principal-component analysis (PCA) of the expression profiles at day 9 showed substantial similarity between gd [2] and gd[T2] samples ( Figure 6C), suggesting that phenotypic effects of TGFb were not mature by this time. By contrast, gd[T2] samples exhibited clear clustering and separation from gd [2] samples by day 15 ( Figure 6D). Given these findings, we advanced the day 15 RNA-seq dataset for further analysis. The top 24 differentially expressed genes on day 15 were designated ''gd[T2] cell signature'' ( Figure 6E) and included upregulation of ITGAE (a E integrin; CD103), IL9, and IL9R, as noted previously. 16 Despite differences in culture methods and RNA analysis methodology, when we compared the 229 genes with FDR < 0.05 in the Peters dataset and our dataset, the Spearman correlation between the log FCs was found to be 0.81 (p < 2.2 eÀ16 ), emphasizing the strong similarity of these 2 cell populations. We also observed significantly elevated CCR4, CCR7, and CXCR4 expression in gd[T2] cells. Several inhibitory receptors were downregulated, notably prostaglandin (PG)E 2 receptors (PTGER2 and PTGER4), CD300A, and KLRG1 ( Figures 6A and 6B; Tables S1 and S2). Using gene set enrichment analysis (GSEA), we identified the overrepresentation of genes associated with TGF-b signaling, naive phenotype, IL-2/STAT5 signaling, glycolysis and gluconeogenesis, and fatty acid metabolism in gd[T2] cells, compared to gd [2] cells ( Figure 6F; Table S3).
Many of these transcriptional changes were validated at the protein level. IL-9 was produced at variable levels by gd [2] cells, and this was consistently increased in gd[T2] cells ( Figure 7A), as was IL-9 receptor expression ( Figure 7B). Despite the considerable donor-to-donor variability, IL-9 production by gd [2] and gd [T2] cells were strongly correlated (Spearman r = 0.475, p = 0.0092), suggesting that some donors have a greater intrinsic capacity to generate IL-9-producing gd[T2] cells. PTGER2 was identified by immunoblotting as a 53-kDa doublet, as described. 35 Levels were variably reduced in gd[T2] compared to gd[2] cells obtained from 3 different donors ( Figure 7C), while cell surface CD300a expression was also significantly reduced in gd[T2] cells ( Figure 7D).
Three candidates were prioritized for further investigation, namely IL-9, CD103, and PTGER2. Antibody-mediated blockade of IL-9 during the expansion of gd[T2] cells had a negligible effect on their cytotoxic activity or activation-induced IFN-g release ( Figures S6A and S6B). Conversely, the addition of IL-9 during the expansion of gd [2] cells also failed to alter these parameters ( Figures S6C and S6D), arguing against an important role of this autocrine loop in the in vitro phenotypic effects of TGF-b. While antibody-blocking experiments targeted against CD103 proved inconclusive, retroviral expression of CD103 expressed in gd [2] cells (gd[2-ITGAE] cells) promoted a trend toward enhanced cytotoxicity ( Figure 7E) and IFN-g release ( Figure 7F) in co-cultures with E-cadherin + solid tumor cells. These data are consistent with a prior study of TGF-b-conditioned Vg9Vd2 T cells. 15 Importantly, however, a trend in the opposite direction was observed when gd[2-ITGAE] cells were co-cultured with leukemic cells that lack E-cadherin ( Figures 7E and 7F). These data do not support a role for CD103 in the enhanced antileukemic activity of gd[T2] cells. Since the expression of both PTGER2 and PTGER4 was reduced in gd[T2] cells, we next investigated their sensitivity to suppression by PGE 2 . We focused on hematological models given the lack of involvement of CD103 or IL-9 in the enhanced antileukemic activity of gd[T2] cells and the reported production of PGE 2 by leukemic blasts. 36 When PGE 2 was added to leukemic cell co-cultures, gd [2] cell cytotoxicity and IFN-g release were inhibited in a dose-dependent manner. By contrast, there was significantly less impact on gd[T2] cell function ( Figures 7G and 7H). Given the role of CD103 in the killing of solid tumor cells, we next assessed the role of another integrin, LFA-1, in leukemic cell killing. We found that anti-CD11a neutralizing antibodies significantly inhibited the killing of KG-1 targets by gd[T2] but not gd [2] cells ( Figure 2L), implicating enhanced LFA-1 activity in this process.
A gd[T2] cell signature is associated with improved prognosis in acute myeloid leukemia (AML) Finally, we used bioinformatic approaches to explore whether gd [T2]-like cells are present in human cancer and influence disease outcomes. Using the Cancer Genome Atlas Genomic Data Commons (TCGA GDC) datasets, we found that TRDV2 transcripts were detected in all 22 cancers evaluated ( Figure S7A, top panel), with the highest expression in AML (n = 128 leukemias; Figure S7A, center panel). Moreover, there was a clear survival benefit for AML patients when high TRDV2 and IL9 expression coincided ( Figure S6E). As previously reported, 2 TRDV2 expression was associated with improved prognosis across all cancers (n = 8,369 cancers; Figure S7B).
Given the disproportionately high percentage of TRDV2 + AMLs and thymomas, we next assessed the prognostic association of transcripts encoding TCR subunits found in Vg9Vd2 cells, making comparisons with Vd1 and ab TCR subunits. Only high levels of TRDV2 and TRGV9 transcripts were associated with significantly improved survival in AML. By contrast, all five transcripts showed a positive survival association in thymoma, with three reaching significance ( Figure S7C).
Next, we tested whether the 24 gene gd[T2] cell signature described above ( Figure 6E) was specifically associated with Vg9Vd2 cells in AML and thymoma. A correlation analysis was performed between the Z scores for each of our 24 signature genes and the correlation coefficient for each signature gene and the relevant TCR transcript, derived from TCGA datasets ( Figure S7D). By this means, we set out to ascertain whether the 24 genes that comprise the gd[T2] cell signature were similarly associated (in terms of direction and strength) with either ab or gd TCR genes, both in our in vitro cultured cells and those hypothesized to be present in AML or thymoma. More important, we found significant positive associations between the gd[T2] cell signature and both TRGV9 and TRDV2 in AML, but not thymoma. No such associations were seen in either disease with TDRV1, TRAC, or TRBC2.
These findings support the hypothesis that the phenotype of Vg9Vd2 T cells found in AML resembles that of gd[T2] cells. To further test this, we assessed TGFB1 transcripts in all 22 cancers indicated above and found that levels were significantly higher in AML than all other tumors ( Figure S7A, lower panel). The abundant co-expression of TRDV2 and TGFB1 transcripts in AML suggests that a suitable environment may be present in this cancer to favor TGF-b conditioning of Vg9Vd2 T cells. Next, we examined whether the gd[T2] cell signature had prognostic significance across a broader panel of 38 cancers. We assessed the impact of each individual gene using PRECOG (prediction of clinical outcomes from genomic profiles). 2 This revealed a strong association between the gd[T2] cell signature and positive outcome in AML and neuroblastoma, while a weaker association was found in B cell acute lymphoblastic leukemia ( Figure S7E). In keeping with this, gd[T2] cells not only elicited superior antitumor activity in myeloid leukemic models (see above) but also were significantly more cytolytic against the SH-SY5Y neuroblastoma cell line ( Figure S5F). Next, we assessed the prognostic association of the combined 24 gene gd[T2] cell signature. Higher expression was linked to longer overall survival in AML, but only if leukemias were positive for TRDV2 expression (Figure S7F). A similar trend was seen in thymoma and also the remaining 20 tumors in which lower expression of TRDV2 was present ( Figure S7F). Finally, we assessed the prognostic significance of FoxP3 status on survival in AML. While survival was significantly reduced in TRDV2 LO leukemias that were also FoxP3 HI rather than FoxP3 LO , no such effect was observed when comparing TRDV2 HI FoxP3 HI or TRDV2 HI FoxP3 LO leukemias ( Figure S7G). This suggests that FoxP3 is not detrimental to disease outcome when Vd2 gd T cells are present. While these data do not definitively establish that gd[T2] cells are present in human cancer, they point strongly toward this possibility in TGF-b-rich cancers such as AML, and link this with improved disease outcomes.

DISCUSSION
We have shown here that TGF-b1 operates through multiple mechanisms to enhance the antitumor activity of ex vivo expanded Vg9Vd2 T cells. Culture in the presence of TGF-b led to a clear increase in viability, associated with a favorable ratio of anti-apoptotic to pro-apoptotic protein expression. 37 Despite the known inhibitory effect of TGF-b1 on Vg9Vd2 T cell proliferation, 16,37 the cell yield from gd[T2] cultures was increased. gd[T2] cells were also less differentiated, a finding linked to superior clinical outcomes using engineered ab T cells. 38 Moreover, gd [T2] cells were uniquely desensitized to two key immunosuppressive factors in the tumor microenvironment, namely TGF-b itself and PGE 2 . PGE 2 potently suppresses the proliferation, cytokine production, and cytotoxic capacity of gd T cells. 39 Given its pleiotropic actions, several additional factors are likely to contribute to the superior antitumor activity of TGFb-conditioned Vg9Vd2 T cells, and many candidates were identified by our RNA-seq analysis. These include the upregulation of Ephrin A1 and A4 receptors, which favor memory T cell migration across high endothelial venules. 40 CD300A inhibits cellular responses upon binding to lipids exposed during cell death 41 and was downregulated in gd[T2] cells, potentially protecting them from the inhibitory effects of apoptotic or necrotic cells. LPAR6 encodes a receptor that binds the immunosuppressive intermediate, adenosine, and was also reduced in gd[T2] cells. Finally, TGF-b has been shown to repress mammalian target of rapamycin (mTOR) signaling to promote a less exhausted T cell metabolic state. 42 We observed that gd[T2] cells expressed gene sets that favored enhanced glucose generation, consumption, and fatty acid metabolism. Consequently, further metabolic characterization of these cells is warranted.
Although short-term killing activity was unchanged, gd[T2] cells consistently achieved superior cytolytic function at 24 h. In the case of E-cadherin POS solid tumors, this resulted in part from in part through more efficient integrin-mediated target cell interaction, rather than potentiation of their pro-apoptotic machinery. The potential contributory roles of insensitivity to TGF-b and PGE 2 and cytokine-mediated cell death (e.g., necroptosis induced by TNFa/IFN-g, both of which are overproduced by activated gd[T2] cells) also warrant further study. It should also be noted that all of the data shown in this study were generated using immortalized cell lines, providing proof of concept. Further testing against patientderived leukemic and tumor xenografts would be useful, although heterogeneity of these models requires consideration.  (F) Supernatants were collected from co-cultures described in (E) after 72 h and analyzed for IFN-g (means ± SDs; unpaired Student's t test). (G) Effect of exogenous PGE 2 on cytotoxicity (means ± SEMs, n = 10 from 5 independent donors) by gd [2] and gd[T2] cells when co-cultivated with U937 and KG-1 cell lines for 72 h at 1:1 E:T ratio. Data were normalized with respect to cytotoxicity by gd [2] and gd[T2] cells in the absence of added PGE 2 (2-way ANOVA). NS, not significant.
(H) Effect of exogenous PGE 2 on IFN-g production (means ± SEMs, n = 8 from 4 independent donors) by co-cultures described in (G). Data were normalized with respect to IFN-g production by gd [2] and gd[T2] cells in the absence of added PGE 2 (2-way ANOVA). ''All'' in (E) and (F) refers to pooled data from solid tumor and leukemic cell lines, respectively.
Given their enhanced bone marrow trafficking capacity, we posited that adoptive immunotherapy using gd[T2] cells would be effective for hematologic malignancies. We observed the strong therapeutic activity of these cells in leukemic model systems. However, the altered biodistribution of gd[T2] cells may have compromised their use in solid tumor immunotherapy, despite the frequent production of CXCR4 ligands 47 and expression of E-selectin on the vasculature of both human and mouse carcinomas. 48 This limitation could be circumvented by regional (i.p.) delivery in models of epithelial ovarian cancer (EOC), enabling the superior anticancer activity of gd[T2] cells to again be harnessed.
Under some circumstances, TGF-b1 confers a Treg phenotype on gd T cells. 13,37,49-51 We observed such a gene signature in gd[T2] cells using GSEA. More important, however, gd [T2] cells lacked functional Treg activity. Where TGF-b1 has promoted gd T-reg differentiation, cultures included fetal calf serum, IL-15, and/or vitamin C, none of which were used here. Recently, it was shown that TGF-b-conditioned Vg9Vd2 T cells exhibited transient and low-level FoxP3 expression without demethylation of the FoxP3 Treg-specific demethylated region or convincing Treg function, unless vitamin C was also present. 51 In comparing Vg9Vd2 tumor-infiltrating lymphocytes (TILs) across cancers, we and others 52 found that their abundance was greatest in AML. Our bioinformatic analyses suggest that gd TILs found in AML resemble gd[T2] cells and are associated with improved survival. While thymomas also contain a large number of Vg9Vd2 TILs, they do not share a gd[T2]-like phenotype or have a similar impact on outcomes. This may be the result of the lower abundance of TGFB1 transcripts in this tumor. Although clear evidence of an IL-9 autocrine loop was found in gd [T2] cells, this was not implicated in their enhanced superior in vitro function, either against leukemic or solid tumor cell lines. Nonetheless, analysis of TCGA datasets in AML provide clinical evidence that an IL-9-rich environment in combination with Vg9Vd2 TILs was linked with the best prognosis, suggesting that this cytokine contributes importantly in vivo to gd[T2] cell biology. On a cautionary note, IL-9 itself has been linked to pro-leukemic effects in some settings. 53 While our data provide support for the use of gd[T2] cells in the immunotherapy of a range of cancers, AML represents a particularly attractive option for clinical evaluation of this strategy. A substantial proportion of AML samples are naturally susceptible to Vg9Vd2 T cells. 54,55 Consequently, AML patients in disease remission may benefit from consolidation immunotherapy using healthy donor-derived allogeneic gd[T2] cells. Given the variability in the yield, immunophenotype, function, and diversity of these cells, 56-58 carefully selected healthy donors will be required for the production of cell products. Conditioning of patients with fludarabine, Ara-C, and ZOL would benefit from the combined lymphodepleting, antileukemic, and gd T cell-sensitizing actions of these drugs, providing an intriguing platform for the evaluation of this immunotherapeutic strategy. To aid in clinical development, there is a need to define predictive biomarkers that correlate with the enhanced antileukemic potency of these heterogeneous cell products, a task that will require considerable further study.

Limitations of the study
We have not dissected the effects of TGF-b on non-d2 gd T cells present when these cells are expanded using antibody activation. Moreover, the compatibility of this platform with genetic targeting approaches such as chimeric antigen receptors remains untested.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, John Maher (john.maher@kcl.ac.uk).

Materials availability
Reagents generated in this study will be made available on request, but we may require a payment and/or a completed Materials Transfer Agreement if there is potential for commercial application.

Data and code availability
RNA-seq data were deposited at NIH Gene Expression Omnibus (GEO) and are publicly available as of the date of publication. There was no new code developed as part of this study. Any additional information required to re-analyze the data reported in this work paper is available from the lead contact upon request.

Mice
Experiments were performed with mice aged 6-10 weeks. All in vivo experimentation adhered to UK Home Office guidelines, as specified in project license numbers 70/7794 and P23115EBF and was approved by the King's College London animal welfare and ethical review body (AWERB). SCID Beige (CB17.Cg-Prkdc scid Lyst bg-J /Crl) and Nod SCID gc null (NSG; NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ) mice were purchased from Charles River Laboratories. Both male and female mice were used in equal numbers except in models of breast and ovarian cancer in which female mice were used. An equivalent number of male and female donors were used otherwise and we did not detect any clear impact of gender on the findings of our study. Animals were housed in individually ventilated cages within the Biological Services Units at King's College London. Mice were randomly allocated to experimental groups based on similar average tumor burden prior to treatment.
Cell lines and tissue culture Tumor, leukemic and immortalized cell lines were obtained from US or European Cell Banks or stocks that were validated using short tandem repeat analysis and were subject to regular mycoplasma testing. Cell lines were maintained in either Dulbecco's Modified Eagle's Medium or RPMI 1640 supplemented with 10% FCS, l-glutamine and antibiotic-antimycotic solution. Primary human pulmonary endothelial cells were maintained in endothelial cell growth medium. Cells were engineered to co-express ffLuc/RFP or a GFP reporter gene using an SFG retroviral vector as described. 59 Cells were maintained at 37 C in a humidified atmosphere of 5% CO 2 .
Human study oversight Blood samples were also obtained from healthy donors following approval of the study protocol by a National Health Service Research Ethics Committee (09/H0804/92 and 18/WS/0047).

METHOD DETAILS
Culture and genetic modification of primary human Vg9Vd2 T cells After isolation by density gradient separation, PBMC were plated at a density of 3x10 6 cells/ml in TexMACS SFM supplemented with GlutaMax and antibiotic-antimycotic solution. Where specifically indicated, medium also contained 10% human AB serum. Activation of Vg9Vd2 T cells was achieved using ZOL (1 mg/ml) or immobilized pan-gd TCR antibody (0.8mg/mL). On the day of activation, IL-2 (100U/ml) and, in some cases, TGF-b (5ng/mL) and/ or IL-9 (concentrations specified) were added. Where indicated, cultures were supplemented with a blocking IL-9 antibody (neutralizing concentration 2-5mg/mL per 5ng/mL IL-9; https://www. novusbio.com/products/il-9-antibody_af209#datasheet, accessed 12-09-2017). Cytokines, inhibitors and/or blocking antibodies were replenished every 2-3 days with addition of medium as appropriate over a total culture period of at least 15 days. Retroviral transduction of Vg9Vd2 T cells was performed 7 days after activation using PG13-derived viral particles, pre-loaded on RetroNectin coated plates. 60 The ITGAE gene was synthesized and was fused via a downstream furin cleavage site (RRKR) and Thosea Asigna (T2A) ribosomal skip peptide to a membrane anchored epitope tag in which a human CD8a leader peptide was joined to a 9e10 myc epitope (EQKLISEEDL) followed by codons 114-180 of human CD28 to achieve membrane anchoring. This construct was delivered to gd T cells using the SFG retroviral vector. 60 Cell Reports Medicine 2, 100473, December 21, 2021 e5 Article ll OPEN ACCESS

Protein analysis
Flow cytometric analysis was performed using FACScalibur cytometer with on an LSR FORTESSA analyzer, recording at least 5 3 10 5 events. Compensation settings were established using single stained samples. Intracellular staining was performed after fixation and permeabilization of cells with 4% paraformaldehyde and 0.1% saponin. Where viability of cells was assessed following camptothecin exposure, 10 6 cells were pre-exposed to camptothecin (12mM) for 6h at 37 C prior to analysis. Cells were first gated based on forward (FSC-A) and side (SSC-A) scatter (measuring cell size and granularity respectively) to exclude debris. Single cells were then selected using SSC-A versus SSC-W parameters. Dead cells were excluded using a viability stain.
To test E-selectin binding, cells were incubated with 1mg recombinant E-selectin-Fc in the presence of FACS buffer (0.5% BSA in PBS) plus 5mM EDTA or 5mM Ca 2+ (as indicated) overnight at 4 C. On the next day, anti-human IgG Fc-FITC secondary antibody was added for 30 m at 4 C prior to washing and analysis by flow cytometry.
Human cytokines were quantified by ELISA or Luminex 30-plex cytokine array kit, as described by the manufacturers, using CLAR-IOSTAR or Flexmap 3D platforms respectively. Expression of proteins involved in apoptotic cell death was determined using a Proteome Profiler Human apoptosis array kit, as recommended by the manufacturers and analyzed using ImageJ. Cell lysates from matched day 15 cells were prepared for immunoblotting analysis using a RIPA buffer as per manufacturer's instructions. 10mg of cell lysate, quantitated using the BCA assay, was reduced and separated by SDS-PAGE (100V) before being transferred onto PVDF membrane. Membranes were blocked in 5% milk powder in tris-buffered saline with 0.1% tween 20 (TBST) for 1h at 4 C, before being probed with primary antibodies (at concentrations recommended by the manufacturer) overnight at 4 C. Membranes were washed in TBST before being probed with appropriate secondary horseradish peroxidase-labeled antibodies for 1h at RT. Membranes were washed with TBST and developed using ECL. Pixel density in relevant bands was quantified in captured electronic Images using ImageJ software.
To test effect of PGE 2 on IFN-g production, gd [2] or gd[T2] cells were co-cultivated with ffLuc-expressing U937 and KG-1 for 72h at 1:1 E:T ratio in the presence of the indicated concentration of PGE 2 . IFN-g was measured in harvested supernatant by ELISA. Percentage inhibition of IFN-g production was calculated using the formula: [IFN-g] in the absence of PGE 2 -[IFN-g] in the presence of PGE 2 x 100% divided by [IFN-g] in the absence of PGE 2

Cytotoxicity assays
Co-cultivation assays between T cells and 1 3 10 4 target cells were established for intervals and at effector to target (E:T) ratios as specified in individual experiments. Where indicated, target cell destruction was quantified by in vitro MTT or luciferase assays, as described. 59 Alternatively, viability of target cells in monolayer cultures was monitored by real time impedance measurement using an xCELLigence RTCA MP (ACEA Biosciences, San Diego CA), as recommended by the manufacturers. Where indicated, target cells were sensitized with indicated concentrations of ZOL, PAM and/ or Ara C for 24h prior to addition of Vg9Vd2 T cells. Residual viable cells were normalized to tumor or untreated control cells that were cultured alone (set at 100%).
In the case of cytotoxicity assays using SH-SY5Y cells, tumor cells were sensitized with 1 mg/mL ZOL for 24h, washed once in serum-free media. gd [2] or gd[T2] cells were added at a 5:1 E:T ratio. After 24h, media was aspirated from wells and cells were fixed with 4% glutaraldehyde for 20 minutes at room temperature. Cells were washed twice with PBS and subsequently stained with 0.05% methylene blue for 20 minutes at room temperature, with gentle shaking. Cells were washed three times in a tray of running water then air-dried upside down, overnight at room temperature. The following day, methylene blue dye was extracted by addition of 3% HCl and incubation for 30 minutes with gentle shaking, at room temperature. Cell viability was determined by measuring absorbance at 655nm on the FLUOstar Omega plate reader (BMG Labtech). All OD values were corrected by subtraction of background values generated using the media alone control, and viable cell numbers were calculated using a standard curve prepared with tumor cells alone.
To test the effect of CD11a blockade on cytotoxicity, KG-1 cells were plated at 1x10 5 cells in 24 plates in TexMACS media. ZOL (1mg/mL) was applied to appropriate wells for 24h. gd [2] or gd[T2] cells were added at a 5:1 E:T ratio. After 24h, all cells were removed and stained with a fixable viability dye and anti-CD3. Live KG-1 cells were defined by lack of uptake of the viability dye and negative staining for anti-CD3. Percentages of live KG-1 cells remaining after co-culture with a CD11a neutralizing antibody were corrected against live KG-1 cell percentages when using an isotype control.
To test effect of PGE 2 on cytotoxicity, gd [2] or gd[T2] cells were co-cultivated with ffLuc-expressing U937 and KG-1 for 72h at 1:1 E:T ratio in the presence of the indicated concentration of PGE 2 . Leukemic cell viability was assessed by luciferase assay, normalized to leukemic cells alone (100% viability). Percentage inhibition of cytotoxicity or was calculated using the formula: % cytotoxicity in the absence of PGE 2 -% cytotoxicity in the presence of PGE 2 x 100 divided by % cytotoxicity in the absence of PGE 2 In vivo measurement of anti-leukemic and anti-tumor activity Tumor and leukemic cells were transduced with SFG ffLuc/RFP 59 and were purified by flow sorting prior to engraftment in mice as indicated for individual experiments. Mice with similar average tumor burden were randomly assigned to groups for blinded treatment with the indicated agents. Bioluminescence imaging was performed as described. 59 In all experiments, animals were inspected daily and weighed weekly. Mice were culled if symptomatic as a result of tumor progression or weight loss of R 20%.  4 was performed as described. 29 In brief, 26 3 10 6 T cells were washed twice with PBS and re-suspended in 4 mL PBS to which 100ml of [ 89 Zr]Zr(oxinate) 4 (70-80MBq dissolved in 30% dimethylsulfoxide) was added. Following incubation for 25 minutes at room temperature with regular gentle mixing by swirling, cells were washed with PBS and re-suspended in 300 mL PBS at a final cell density of 8.9 3 10 7 cells per mL.
Animals were injected i.v. with 5 3 10 6 89Zr-labeled Vg9Vd2 T cells (5MBq). Mice were imaged by PET-CT for 60 minutes at 24 and 48 hours after T cell injection. Images were reconstructed and activity within regions of interest (ROI) were calculated. Bio-distribution analysis was performed by ex vivo tissue gamma counting.
RNA sequencing analysis ZOL-activated gd [2] and gd[T2] cells were expanded from 3 separate healthy donors as described above. On days 9 and 15 of the culture, Vg9Vd2 T cells were flow-sorted to purity and RNA extracted using TRIzol. Ribosomal RNA was removed using Ribo-Zero and libraries were prepared using NEBNext. RNA-Seq was performed with a minimum sequencing depth of 100,000 reads. The quality of the tags was inspected with fastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/, accessed 15/06/2020). Raw reads were trimmed and filtered to remove adaptor contamination and poor-quality bases using trimmomatic. 61 The resulting FASTQ were mapped to the GRCh38 assembly of the human genome using Hisat2 with default parameters. 62 The number of reads mapping to the genomic features annotated in Ensembl 63 with a MAPQ score higher than or equal to 30 was calculated for all samples using htseq-count with default parameters. 64 Features with no mapped reads in at least one sample or with less than 10 reads on average across all samples were not included in downstream analyses. Differential gene expression analysis between sample groups were performed in R using the Wald test as implemented in the DESeq2 package. 65 p values were adjusted for multiple testing according to the Benjamini and Hochberg procedure. 66 The raw and adjusted p values were re-estimated empirically with fdrtool, 67 when the histograms of the initial p value distributions showed that the assumptions of the Wald test were not met.
The Cancer Genome Atlas Genomic Data Commons (TCGA GDC) determination of the gd[T2] cell signature and survival analysis TCGA GDC datasets were downloaded from Xena Browser (https://xenabrowser.net, accessed May 20 th , 2020) and TCR transcript survival and correlation analysis performed on the Xena Functional Genomics Explorer. 68 We selected the 24 differentially expressed genes with the lowest estimated false discovery rate (FDR = 8.44e-14) from the RNA-Seq analysis of gd [2] and gd[T2] samples, thereby generating a gd[T2] cell signature ( Figure 6B). Using RNA-Seq TCGA datasets derived from a range of cancer types, the impact of gd[T2] cell signature on survival was tested and downloaded using Kmplot (Semmelweis University, Budapest; https:// kmplot.com/analysis/, accessed May 22 nd , 2020).
Gene set enrichment analysis (GSEA) GSEA (v4.1.0) was downloaded from the Broad Institute. 69 Expression data was inputted as mean (of the 3 donors) transcripts per million (TPM) for class A (gd [2] cells) and class B (gd[T2] cells). The analysis was run on the following MSigDb gene sets databases, 70 Figure 6 and Table S3.

PREdiction of Clinical Outcomes from Genomic profiles (PRECOG)
The gd[T2] cell signature was used to calculate survival outcomes using prediction of clinical outcomes from genomic profiles (PRE-COG), 2 a tool which calculates prognostic survival scores through a meta-analysis by integrating 165 cancer gene expression datasets and clinical outcomes across $26,000 patients presenting with 39 malignancies. PRECOG scores for genes within the gd[T2] cell signature with a negative z score (i.e., downregulated in gd[T2] cells compared to gd [2] cells) were transformed by multiplying by À1. Other Vg9Vd2 T cell marker genes available on PRECOG were subsequently included to evaluate survival outcome predictions based on the presence of these cells. Downstream analyses and plots were conducted in R (version 3.6.1), using the tidyverse 71 packages (version 1.3.0) as well as the plotly package (version 4.9.2.1).

QUANTIFICATION AND STATISTICAL ANALYSIS
Normality of all experimental data was tested using the Shapiro-Wilk test prior to statistical analysis. Statistical analysis was performed using two-tailed Student's t test, one-way or two-way ANOVA (normally distributed datasets), or Wilcoxon signed rank test (other datasets). To test correlation between IL-9 production by gd [2] and gd[T2] cells from individual donors a Spearman test was performed (data not normally distributed). Survival data were analyzed using the Log-rank (Mantel-Cox) test. All statistical analysis was performed using GraphPad Prism version 9.0 (GraphPad software, San Diego, CA) or Excel for Mac 2020 (Berkshire, UK).