Bifunctional immune checkpoint-targeted antibody-ligand traps that simultaneously disable TGFβ enhance the efficacy of cancer immunotherapy

A majority of cancers fail to respond to immunotherapy with antibodies targeting immune checkpoints, such as cytotoxic T-lymphocyte antigen-4 (CTLA-4) or programmed death-1 (PD-1)/PD-1 ligand (PD-L1). Cancers frequently express transforming growth factor-β (TGFβ), which drives immune dysfunction in the tumor microenvironment by inducing regulatory T cells (Tregs) and inhibiting CD8+ and TH1 cells. To address this therapeutic challenge, we invent bifunctional antibody–ligand traps (Y-traps) comprising an antibody targeting CTLA-4 or PD-L1 fused to a TGFβ receptor II ectodomain sequence that simultaneously disables autocrine/paracrine TGFβ in the target cell microenvironment (a-CTLA4-TGFβRIIecd and a-PDL1-TGFβRIIecd). a-CTLA4-TGFβRIIecd is more effective in reducing tumor-infiltrating Tregs and inhibiting tumor progression compared with CTLA-4 antibody (Ipilimumab). Likewise, a-PDL1-TGFβRIIecd exhibits superior antitumor efficacy compared with PD-L1 antibodies (Atezolizumab or Avelumab). Our data demonstrate that Y-traps counteract TGFβ-mediated differentiation of Tregs and immune tolerance, thereby providing a potentially more effective immunotherapeutic strategy against cancers that are resistant to current immune checkpoint inhibitors.

G enetic mutations accruing from the inherent genomic instability of tumor cells present neo-antigens that are recognized by the immune system. Cross-presentation of tumor antigens at the immune synapse between antigenpresenting dendritic cells and T lymphocytes can potentially activate an adaptive antitumor immune response that is mediated by CD4 + T-helper cells (T H 1) and CD8 + cytotoxic effector cells, and sustained by tumor-reactive central memory T cells 1 . However, tumors continuously evolve to counteract and ultimately defeat such immune surveillance by co-opting and amplifying mechanisms of immune tolerance to evade elimination by the immune system [1][2][3] . This prerequisite for tumor progression is enabled by the ability of cancers to produce immunomodulatory factors that create a tolerogenic immune cell microenvironment 3 .
Transforming growth factor-β (TGFβ) is a multifunctional cytokine that is overexpressed in a majority of cancers 4 . The highaffinity binding of TGFβ to TGFβ receptor II (TGFβRII) recruits TGFβ receptor I into a heterotetrameric complex that initiates SMAD-mediated transcriptional activation or repression of several genes that control cell growth, differentiation, and migration 5 . Besides promoting epithelial-to-mesenchymal transition, invasion, and metastases of tumor cells, TGFβ has a critical role in regulating the adaptive immune system [6][7][8][9] . TGFβ suppresses the expression of interferon-γ (IFN-γ), restricts the differentiation of T H 1 cells, attenuates the activation and cytotoxic function of CD8 + effector cells, and inhibits the development of central memory T cells [8][9][10][11] . Most significantly, TGFβ induces the differentiation of regulatory T cells (Tregs), a sub-population of immunosuppressive CD4 + T cells that express the interleukin-2 α-chain (CD25) and the forkhead box P3 (FOXP3) transcription factor [12][13][14][15][16][17][18] . TGFβ induces the expression of FOXP3, the signature transcription factor that determines and maintains the functional program of the Treg lineage [19][20][21][22][23] . FOXP3, in turn, induces the expression of cytotoxic T lymphocyte antigen-4 (CTLA-4), an immune-inhibitory receptor that restrains co-stimulation of T cells, and Galectin-9 (GAL-9), a ligand that engages the T-cell immunoglobulin domain and mucin domain-3 (TIM-3) immune-inhibitory receptor, and triggers exhaustion or apoptosis of effector T cells [24][25][26][27][28] . GAL-9 further interacts with TGFβ receptors to drive FOXP3 expression in a positive-feed forward autocrine loop involving SMAD3 activation to induce and maintain Tregs 29 . This ability of TGFβ to skew the differentiation of CD4 + T cells away from a T H 1 phenotype toward a Treg lineage has significant clinical implications, as the functional orientation of tumor-infiltrating immune cells has a major impact on the outcome of patients with cancer 30 . Whereas T H 1 cells, cytotoxic CD8 + T cells and central memory T cells are uniformly and strongly associated with a longer disease-free survival, infiltration of tumors with Tregs has been correlated with a poor prognosis in patients with several types of cancer [30][31][32][33][34][35] .
A potential limitation of T-cell co-stimulation by current immune checkpoint inhibitors is a tumor milieu enriched with TGFβ, which strongly correlated with FOXP3 expression in our analysis of The Cancer Genome Atlas (TCGA) data set of diverse human cancers, including melanoma and breast cancer. We hypothesized that autocrine and paracrine TGFβ signaling in the localized microenvironment of tumor-infiltrating T cells could skew them toward Tregs and attenuate the activation of T H 1 and CD8 + immune effector cells, thereby limiting the therapeutic efficacy of CTLA-4 or PD-1/PD-L1 antagonists 44,45 . As Tregs express and employ TGFβ and Gal-9 to maintain their own phenotype and function, enhancing the efficacy of immune checkpoint inhibitors requires a strategy to specifically break this hyperactive autocrine loop in tumor-infiltrating Tregs. To test this hypothesis and address this therapeutic challenge, we invented bifunctional antibody-ligand traps (Y-traps) comprising an antibody targeting either CTLA-4 or PD-L1, which is fused at the C terminus of the heavy chain (HC) to a TGFβRII ectodomain sequence to sequester and disable autocrine/paracrine TGFβ in the target cell microenvironment (a-CTLA4-TGFβRIIecd and a-PDL1-TGFβRIIecd) 46,[47][48][49] . We find that a-CTLA4-TGFβRIIecd is significantly more effective in reducing and counteracting tumor-infiltrating Tregs, activating antitumor immunity, and inhibiting tumor progression compared with the CTLA-4 antibody, Ipilimumab. Likewise, a-PDL1-TGFβRIIecd exhibits superior antitumor efficacy compared with PD-L1 antibodies (Atezolizumab or Avelumab). Our data demonstrate that Y-traps simultaneously disable immune checkpoints and counteract TGFβ-mediated differentiation of Tregs and immune tolerance, thereby providing a more effective immunotherapeutic strategy against cancers that fail to respond to current immune checkpoint inhibitors.

Results
TGFβ signaling correlates with FOXP3 expression in cancers. We used iPANDA, a bioinformatics software suite for qualitative analysis of intracellular signaling pathway activation based on transcriptomic data 50,51 , to assess the level of TGFβ signaling in TCGA data sets of different types of cancer and investigate whether the TGFβ pathway activation in tumors is correlated with the level of expression of FOXP3, the signature transcription factor of the Treg lineage. Analysis of transcriptomic data from a skin cutaneous melanoma (SKCM) data set (n = 472), using skin biopsy of healthy women (n = 122) as a reference, showed that upregulation in TGFβ signaling strongly correlated with increased messenger RNA expression levels of TGFB1 and FOXP3 (Fig. 1a, b). The strong correlation between TGFβ pathway activation and FOXP3 expression was also noted in a TCGA breast cancer data set (n = 776), using normal breast tissue as a reference (Fig. 1c, d). Among breast cancers, TGFβ pathway activation and corresponding elevation of FOXP3 was especially striking in triplenegative breast cancer (TNBC) (Fig. 1c), an aggressive subtype that lacks expression of hormone receptors (estrogen receptor (ER)/progesterone receptor (PR)) and HER2/neu, and has a higher risk of metastases and death within 5 years of diagnosis. Although TGFβ and PD-L1 can cooperate to induce expression of FOXP3, expression of PD-L1 (CD274) mRNA did not exhibit a corresponding or consistent correlation with FOXP3 mRNA expression (Fig. 1d). The strong correlation of TGFβ activation with FOXP3 expression supports a crucial role of autocrine/ paracrine TGFβ signaling in induction and maintenance of Tregs in diverse cancers.
ligand-binding sequence of the extracellular domain of TGFβRII via a flexible linker peptide (Fig. 2a, b). Protein identification of the purified antibody from CHO-K1 cell supernatants was performed by liquid chromatography Fourier transform tandem mass spectrometry (LC-MS/MS) to confirm the amino acid sequence of the HC of a-CTLA4-TGFβRII (Fig. 2b). SDSpolyacrylamide gel electrophoresis (PAGE) under reducing (R) and non-reducing (NR) conditions was used to compare the fulllength (FL), HC, and light chain (LC) of a-CTLA4-TGFβRII and a-CTLA-4 antibody (Fig. 2c). MS analysis confirmed the expected higher molecular weight of the HC of a-CTLA4-TGFβRII (65.697 kDa) compared with the HC of a-CTLA-4 antibody (49.256 kDa). The bifunctional ability of a-CTLA4-TGFβRII to simultaneously bind CTLA-4 and TGFβ1 was confirmed by enzyme-linked immunosorbent assay (ELISA), wherein a-CTLA4-TGFβRII was added to CTLA-4-Fc-coated plates, followed by recombinant human TGFβ (rhTGFβ1) that was detected by a biotinylated anti-human TGFβ1 antibody (Fig. 2d, e). Unlike a-CTLA-4, a-CTLA4-TGFβRII exhibited the additional ability to compete with a TGFβ capture antibody for binding to TGFβ1 (Fig. 2f).
a-CTLA4-TGFβRII counteracts Tregs and T H 17 differentiation. The FOXP3 transcription factor governs the differentiation and function of Tregs. The transcription factors SMAD3 and nuclear factor of activated T-cells (NFAT) are required for activation of a FOXP3 enhancer, and both factors are essential for induction of FOXP3 in primary T cells. TGFβ-activated SMAD-2/ 3 cooperates with interleukin (IL)-2-activated NFAT to induce FOXP3 expression and promote the conversion of naïve CD4 + T cells to FOXP3-expressing Treg cells (induced Tregs or iTregs) that mediate immune tolerance 19 (Fig. 2a). Consistent with these observations, treatment with rhTGFβ1 induced the phosphorylation of SMAD-2/3 and increased expression of FOXP3 in human peripheral blood mononuclear cells (PBMCs) costimulated with anti-CD3/anti-CD28-coated beads and rhIL-2 (Fig. 3a,  left panel).
a-CTLA4-TGFβRII is designed to exploit the FOXP3-mediated expression of CTLA-4 on Tregs to decorate the targeted cells with a decoy TGFβRII ectodomain that captures and disables TGFβ in their localized microenvironment (Fig. 2a, b). We examined the ability of a-CTLA4-TGFβRII to inhibit TGFβ-induced SMAD-2/3 phosphorylation and expression of FOXP3 in human T cells. Human PBMC were stimulated with rhIL-2 and anti-CD3/anti-CD28-coated beads in the presence of rhTGFβ1 with or without either a-CTLA4-TGFβRII or a-CTLA-4. Unlike a-CTLA-4, a-CTLA4-TGFβRII counteracted TGFβ-induced SMAD-2/3 phosphorylation and FOXP3 expression in co-stimulated T cells (Fig. 3a, right panel).
As FOXP3 is instrumental for the suppressive function of Tregs, the relative ability of a-CTLA4-TGFβRII and a-CTLA-4 to counteract Treg-mediated suppression of tumor-reactive T cells was also examined using tumor-infiltrated BM from a patient. Anti-CD3/anti-CD28 and rhIL-2 activated CD3 + marrow-  infiltrating lymphocytes (aMILs) 52,53 were CFSE labeled and added to autologous BM that had been pulsed with either tumor cell lysate (tumor-specific antigen) or nonspecific antigen in the presence or absence of autologous CD4 + /CD25 + Tregs isolated from the same patient's BM. Following culture of these cells for 3 days with or without either a-CTLA4-TGFβRII or a-CTLA-4, tumor antigen-reactive T cells (CD3 + /CFSE low /IFNγ + ) were quantified by immunophenotype analyses (Fig. 3d). As expected, the addition of autologous Tregs suppressed the activation of tumor antigen-reactive T cells (CD3 + /CFSE low /IFNγ + ) in anti-CD3/anti-CD28-activated aMILs stimulated with tumor antigenpulsed autologous BM. a-CTLA4-TGFβRII was far more effective than a-CTLA-4 in counteracting Treg-mediated suppression and restoring activation of tumor antigen-specific T cells in the presence of autologous Tregs (Fig. 3d). These data demonstrate that a-CTLA4-TGFβRII is more effective than a-CTLA-4 in counteracting Tregs in the tumor microenvironment.
The differentiation of CD4 + T cells into T H 1, T H 17, or iTreg cell lineages is determined by the cytokine milieu 54 . Whereas IFN-γ drives T H 1 differentiation, TGFβ is required for differentiation of both iTreg and T H 17 cells. Although TGFβ cooperates with IL-2 to induce iTreg differentiation 13,55,56 , TGFβ promotes T H 17 differentiation in the presence of proinflammatory cytokines, such as IL-6 [57][58][59][60][61] . In contrast to T H 1 cells that are strongly associated with good clinical prognosis for all cancer types, T H 17 cells are associated with tumor-promoting inflammation and autoimmune pathology 30,[62][63][64] . As a-CTLA4-TGFβRII can render the targeted T cells incapable of responding to TGFβ signals in their immediate milieu, we examined whether a-CTLA4-TGFβRII also skews the differentiation of CD4 + T cells away from T H 17 cells toward an IFN-γ-expressing T H  Antibody: TGF-β ratio TGF-β (pg ml   CTLA-4, a-CTLA4-TGFβRII was able to abrogate the expression of IL-17 in CD4 + T cells and switch them to an IFN-γ-expressing T H 1 phenotype (Fig. 3e).
a-CTLA4-TGFβRII is more effective than a-CTLA-4 and a-PD1. Besides CTLA-4, engagement of PD-1 by PD-L1 expressed on tumor cells or T cells also inhibits antitumor T cells. Although monoclonal antibodies (mAbs) against PD-1, such as Nivolumab or Pembrolizumab, are effective in some patients, the vast majority of cancers fail to respond to either PD-1 blockade or even dual checkpoint inhibition with a-CTLA-4 and a-PD-1. Therefore, we investigated the ability of a-CTLA4-TGFβRII to elicit antitumor immunity and inhibit the growth and metastases of cancers that are refractory to current checkpoint inhibitors, such as TNBC. Approximately 15-25% of patients with breast cancer have TNBC, an aggressive type that does not respond to hormonal agents or targeted therapy and has an increased risk of metastases. As TNBC is representative of a tumor type that exhibits a TGFβ/FOXP3 signature of Tregmediated immune tolerance (Fig. 1c), we used human immune reconstituted NSG mice bearing the bioluminescent human MDA-MB-231-luc (D3H2LN) TNBC cell line that expresses elevated PD-L1 (Fig. 5a, inset) and TGFβ (531 pg per 10 6 (Fig. 5a-c). In contrast, treatment with a-CTLA4-TGFβRII was significantly more effective at inhibiting the progression of MDA-MB-231-luc tumors compared with untreated controls (p < 0.00001, Student's unpaired t-test), or animals treated with either a-CTLA-4 alone (p < 0.001, Student's unpaired t-test) or the combination of a-CTLA-4 and a-PD-1 mAbs (p < 0.0001, Student's unpaired t-test) (Fig. 5a, b). In addition, a-CTLA4-TGFβRII exhibited significantly better antitumor efficacy compared with either a-TGFβ (p < 0.001, Student's unpaired t-test) or a combination of a-CTLA-4 and a-TGFβ (p < 0.04, Student's unpaired t-test) (Fig. 5a, b), and was more effective in inhibiting lung metastases (Fig. 5c). Consistent with its superior antitumor efficacy, a-CTLA4-TGFβRII was more effective in reducing Tregs, elevating tumor-reactive IFN-γexpressing CD8 + cells and increasing the CD4 + and CD8 + central memory T cells compared with the combination of a-CTLA-4 and a-PD-1 mAbs (Fig. 5d).
Design and bifunctional target binding of a-PDL1-TGFβRII.

PD-L1 is overexpressed on tumor cells as well as tumorinfiltrating T cells, where it cooperates with TGFβ to inhibit Tcell activation and induce and maintain immunosuppressive Treg cells. Although
TGFβ and PD-L1 can cooperate to induce FOXP3, our analysis of both TCGA data sets showed that the correlation of TGFβ pathway activation with FOXP3 expression was substantially stronger than its correlation with PD-L1 (CD274) mRNA (Fig. 1). These data suggest that PD-L1/PD-1 checkpoint and TGFβ signaling exercise independent, yet cooperative mechanisms of immune tolerance, thereby supporting a therapeutic rationale for simultaneously counteracting both axes in the tumor immune microenvironment. Anti-PDL1-TGFβRIIecd (a-PDL1-TGFβRII) is a bifunctional antibody-ligand trap that was designed to target PD-L1 and simultaneously inactivate TGFβ by fusion of an extracellular domain sequence of TGFβRII to the C terminus of the HC of anti-PD-L1 antibody via a flexible linker sequence, (GGGGS) 3 (Fig. 6a, b). SDS-PAGE of two different anti-PD-L1 antibodies (Atezolizumab and Avelumab) and their corresponding anti-PDL1-TGFβRII products (Ab1 and Ab2) under R and NR conditions showed the expected higher molecular weight of the HC of anti-PDL1-TGFβRII. Size exclusion-high-performance liquid chromatography (SEC-HPLC) analysis showed a single peak corresponding to purified a-PDL1-TGFβRII with no aggregation (Fig. 6c).
The comparative ability of a-PDL1-TGFβRII and a-PD-L1 to bind PD-L1 was evaluated by ELISA assay, wherein biotinylated recombinant human PD-L1 (rh B7-H1-biotin) was added to plates coated with a-PDL1-TGFβRII or a-PD-L1 antibody, and detected by horseradish peroxidase (HRP)-Avidin. Plates coated with nonspecific IgG-TGFβRII showed no binding to PD-L1 and served as a negative control. Each a-PDL1-TGFβRII (Ab1 and Ab2) exhibited specific binding to rhPD-L1 with an efficiency that was similar to the respective a-PD-L1 (Fig. 6d). The comparative ability of a-PDL1-TGFβRII and a-PD-L1 to bind TGFβ was evaluated by ELISA assay, wherein rhTGFβ1 was added to plates coated with a-PDL1-TGFβRII or a-PD-L1 and then detected by biotinylated anti-TGFβ1 and HRP-Avidin. Plates coated with nonspecific IgG-TGFβRII and rhTGFβRII-Fc served as positive controls to analyze the binding ability of the test samples to TGFβ. In contrast to the respective a-PD-L1 that failed to bind TGFβ, each corresponding a-PDL1-TGFβRII exhibited binding to TGFβ with an efficiency that was similar to the positive controls (Fig. 6e). The ability of a-PDL1-TGFβRII to simultaneously bind both PD-L1 and TGFβ was also evaluated by a bispecific ELISA assay, wherein a-PDL1-TGFβRII or a-PD-L1 was added to PD-L1-Fc-coated plates, followed by rhTGFβ1 that was detected by a biotinylated anti-human TGFβ1 antibody. In contrast to the respective a-PD-L1, each corresponding a-PDL1-TGFβRII (Ab1 and Ab2) exhibited simultaneous binding to PD-L1 and TGFβ (Fig. 6f).
a-PDL1-TGFβRII is more effective than a-PD-L1 antibodies. The comparative antitumor efficacy of a-PDL1-TGFβRII, a-PD-L1, nonspecific IgG-TGFβRII, and the combination of a-PD-L1 and IgG-TGFβRII against human cancers expressing both PD-L1 and TGFβ was evaluated in either A375 (Fig. 7a, b) or MDA-MB-231-Luc (Fig. 7c, d) bearing NSG mice reconstituted with human CD34 + hematopoietic stem cells (HSCs). Moreover, in the TNBC model, independent experiments were conducted to compare two different a-PDL1-TGFβRII antibody-ligand traps with their respective a-PD-L1 antibodies (Atezolizumab and Avelumab) (Fig. 7c). In vivo tumor growth curves (mean ± SEM) in both tumor models demonstrated that treatment of tumor-bearing mice with a-PDL1-TGFβRII was significantly more effective at inhibiting the progression of A375 (p < 0.01, Student's unpaired t-test) (Fig. 7a) or MDA-MB-231-luc (p < 0.004, Student's unpaired t-test) (Fig. 7c) tumors compared with the respective a-PD-L1 alone, IgG-TGFβRII alone, and the combination of a-PD-L1 and nonspecific IgG-TGFβRII. Consistent with its superior antitumor efficacy, treatment with a-PDL1-TGFβRII resulted in significant inhibition of FOXP3 + expressing Tregs (p < 0.05, Student's unpaired t-test) (Fig. 7b, d: left) and a greater elevation in percentage of tumor-reactive IFNγ-expressing CD8 + cells (p < 0.01, Student's unpaired t-test) (Fig. 7b, d: right) compared with treatment with the a-PD-L1 alone, IgG-TGFβRII alone, and even their combination. Mice treated with a-PDL1-TGFβRII maintained serum hepatic enzymes within a normal range of liver function and demonstrated no loss of body weight during the course of experiment.

Discussion
Cancer immunotherapy is currently focused on targeting immune inhibitory checkpoints that control T cell activation, such as CTLA-4 and PD-1 40,42,43,[65][66][67] . Monoclonal antibodies that block these immune checkpoints can unleash antitumor immunity and produce durable clinical responses in a subset of patients with advanced cancers, such as melanoma and non-small-cell lung cancer 42,43,67 . However, these immunotherapeutics are currently constrained by their inability to induce clinical responses in the vast majority of patients. A key limitation of checkpoint inhibitors is that they narrowly focus on modulating the immune synapse but do not address the key molecular determinants that are primarily responsible for immune dysfunction in the tumor microenvironment 3,6-8,10,54 . Our data indicate that elevated expression of TGFβ is a root cause of such T-cell dysfunction in the tumor microenvironment. We find that autocrine and paracrine TGFβ signaling fundamentally affects tumor-infiltrating T cells by skewing the differentiation of T H 1 cells toward a Treg phenotype, attenuating the activation of CD8 + effector cells and limiting the development of central memory cells. As Tregs express and employ TGFβ to maintain their own phenotype and function 13,18,19,44,[59][60][61]68 , counteracting these deleterious cells and restoring beneficial T H 1 cells is contingent upon making them impervious to such autocrine signaling. This poses the therapeutic challenge of specifically breaking this TGFβ-driven autocrine loop in tumor-infiltrating Tregs. Systemic TGFβ antagonists fall short of interrupting autocrine signaling in Tregs, as they lack preferential localization to T cells and fail to efficiently compete with the native TGFβRII receptor for binding TGFβ.  (atezolizumab and avelumab). c SEC-HPLC analysis of purified a-PDL1-TGFβRII; d ELISA showing the comparative ability of a-PDL1-TGFβRII and a-PD-L1 antibody to bind PD-L1. Biotinylated recombinant human PD-L1 (rh B7-H1-biotin; 0-100 ng ml −1 ) was added to plates coated with a-PDL1-TGFβRII or a-PD-L1 antibody (1 μg ml −1 ), followed by detection with HRP-Avidin. Plates coated with nonspecific IgG-TGFβRII showed no binding to PD-L1 and served as a negative control to analyze the binding ability of the test samples. e ELISA showing the comparative ability of a-PDL1-TGFβRII and a-PD-L1 antibody to bind TGFβ. Recombinant human TGFβ (rhTGFβ1; 0-2,000 pg ml −1 ) was added to plates coated with a-PDL1-TGFβRII or a-PD-L1 antibody (1 μg ml −1 ), which was detected by biotinylated a-TGFβ1 and HRP-Avidin. Plates coated with nonspecific IgG-TGFβRII and rhTGFβRII-Fc served as positive controls to analyze the binding ability of the test samples to TGFβ. f ELISA showing the ability of a-PDL1-TGFβRII to simultaneously bind both PD-L1 and TGFβ. Anti-PDL1-TGFβRII or a-PD-L1 antibody (0-100 ng ml −1 ) was added to PD-L1-Fc coated plates (1 μg ml −1 ), followed by rhTGFβ1 (100 ng ml −1 ) that was detected by a biotinylated anti-human TGFβ1 antibody.  36 . However, a-CTLA-4 fails to counteract autocrine/paracrine TGFβ signaling, thereby resulting in NFAT/ SMAD3-mediated upregulation of FOXP3 19 and a paradoxical increase in tumor-infiltrating Tregs in the TGFβ-enriched immune microenvironment found in the majority of cancers 45 . Our data demonstrate that a-CTLA4-TGFβRII effectively addresses this challenge by exploiting the FOXP3-driven expression of CTLA-4 to not only disable the CTLA-4 checkpoint, but also decorate the targeted Tregs with a decoy TGFβRII ectodomain that traps TGFβ at the surface of the T cell, thereby rendering them virtually unresponsive to autocrine or paracrine TGFβ in their immediate milieu. As a result, a-CTLA4-TGFβRII counteracts autocrine/paracrine TGFβ/SMAD3-dependent expression of FOXP3, thereby reducing the differentiation and suppressive activity of Tregs. By skewing CD4 + T cells away from FOXP3 + Tregs or T H 17 cells to a T H 1-helper phenotype, a-CTLA4-TGFβRII enables effective activation of antitumor CD8 + effector T cells. An especially attractive feature of a-CTLA4-TGFβRII is its targeted ability to trap TGFβ at the surface of the T cell in a CTLA-4-directed manner, thereby interrupting the TGFβ-autocrine loop that drives FOXP3-mediated expression of CTLA-4. This distinguishing feature allows a-CTLA4-TGFβRII to enjoy a better therapeutic index compared with non-targeted TGFβ antagonists or even combinatorial therapy with a CTLA-4 antibody and a systemic TGFβ antagonist that is not directed to the T cell microenvironment. This unique ability to counteract Tregs and correct immune tolerance in a TGFβ-enriched tumor immune microenvironment enables a-CTLA4-TGFβRII to be significantly more effective in activating antitumor immunity and inhibiting tumor progression compared with a CTLA-4 antibody, a PD-1 antibody, or even the combination of both mAbs. Interestingly, a-CTLA4-TGFβRII was able to exhibit superior single agent activity against PD-L1-expressing tumors and the addition of a-PD1 antibody did not significantly enhance its antitumor efficacy in the breast cancer model. Although the highly effective counteraction of Tregs and immune tolerance by a-CTLA4-TGFβRII was sufficient to inhibit tumor growth, this might have obscured any potential value of combination therapy with a-PD1 during the course of the experiment. As IFN-γ-mediated upregulation of PD-L1 has been shown to be a mechanism of adaptive immune tolerance 39,69 , it remains possible that PD-1/PD-L1 blockade could potentially enhance the antitumor activity of a-CTLA4-TGFβRII over a more extended treatment period or in other tumor models.
Whereas CTLA-4 is highly expressed on Tregs, PD-L1 is overexpressed on tumor cells as well as tumor-infiltrating T cells, where it cooperates with TGFβ to inhibit T-cell activation and induce and maintain Tregs 70 . As PD-L1 and TGFβ entrain independent but cooperative mechanisms of immune tolerance, autocrine and paracrine TGFβ signaling in the tumor immune microenvironment may also limit the therapeutic efficacy of PD-1/PD-L1 antagonists. Consistent with this notion, our data demonstrate that a-PDL1-TGFβRII is significantly more effective in inhibiting tumor progression compared with the corresponding a-PD-L1 antibody due to its bifunctional ability to not only block PD-L1/PD-1 interaction, but simultaneously interrupt autocrine/paracrine TGF-β signaling in the localized microenvironment of PD-L1 expressing tumor-infiltrating immune cells and tumor cells.
Although humanized NSG mice used in this study exhibit a functionally validated surrogate human immune system 71 , this model supports the growth of human cancer cell line and PDXs even when they are not specifically HLA matched to the human CD34 + HSC used for immune reconstitution 72 . The absence of tumor rejection or inhibition of tumor progression in this model demonstrates that there is no spontaneous anti-allogeneic or tumor-specific immune response against such xenografts. However, as HLA-A*02 is the most highly prevalent HLA-A allele in patients with melanoma and breast cancer (including tumors cells used in this study), NSG mice were reconstituted with HLA-A*02 CD34 + HSCs. This was designed to ensure that HLA-A2restricted TILs recognize HLA-A2-expressing xenografts, enabling generation of HLA-A2-restricted cytotoxic T cells. As such, our models were used to assess the comparative ability a-CTLA4-TGFβRII or a-PDL1-TGFβRII and their respective parent antibodies (a-CTLA-4 and a-PD-L1) to counteract immune tolerance in the TME and activate antitumor immune responses.
Although these include HLA-A2-restricted responses against tumor antigens, they could exclude HLA-restricted T-cell responses to antigens presented by other class I or class II HLA loci that may not be matched. The elicited immune responses may not be restricted to tumor antigens, but also potentially encompass anti-allogeneic responses. As such, our tumor models stringently compare T-cell-mediated antitumor immune responses between treatment groups and controls under the same conditions, rather than estimate the absolute efficacy of each independent treatment. Our preclinical studies indicate that both antibody-ligand traps (a-CTLA4-TGFβRII and a-PDL1-TGFβRII) have a superior therapeutic index compared with their parent immune checkpoint inhibitors that are currently in clinical use. Although no adverse events were observed in mice treated with either a-CTLA4-TGFβRII or a-PDL1-TGFβRII, any novel immunotherapeutic strategy that seeks to counteract Treg cells and unleash antitumor immunity carries a potential risk of autoimmune sequelae in patients. As such, the clinical translation of this approach requires well-designed phase I dose-escalation trials to carefully evaluate the safety of each novel agent, determine the maximum tolerated dose, and identify the optimal therapeutic dose and schedule that can elicit an antitumor immune response without prohibitive immune-related adverse events. As elevated TGFβ is an especially common denominator of immune dysfunction in many types of cancer, these Y-traps may provide an effective immunotherapeutic strategy against cancers that fail to respond to current immune checkpoint inhibitors by simultaneously disabling immune checkpoints and counteracting TGFβmediated immune tolerance.

Methods
Correlative analysis of TGFβ pathway and FOXP3 expression. RNA sequencing (RNA-Seq) data for 472 melanomas and 776 breast tumors were retrieved from TCGA. As unlike the breast cohort, TCGA melanoma collection lacks data from the healthy individuals, we have carefully selected a tissue specific normal control cohort (accession number GSE85861) form NCBI GEO repository. RNA-Seq data preprocessing and normalization steps were performed in R version 3.1.0 using DEseq package from Bioconductor. To adjust for the possible batch and processing effect, we have employed the XPN algorithm (R package, CONOR) 73 . The resulting matrix contained mRNA expression information for over 20K genes across all analyzed samples. Normalized gene expression data were loaded into iPANDA 50,51 . The software enables calculation of the Pathway Activation Score (PAS) for each of the 374 pathways analyzed, a value that serves as a quantitative measure of differential pathway activation. A collection of 374 intracellular signaling pathways (which cover a total of 2,294 unique genes) strongly implicated with various solid malignancies was obtained from the SABiosciences (http://www.sabiosciences.com/ pathwaycentral.php), and used for the computational algorithm as described previously 50,51 . Calculated PAS values for TGFβ pathway were used for correlative analysis with TGFB1, FOXP3, and PDL1 expression levels seen in the same patients.
Design of a-CTLA4-TGFβRII and a-PDL1-TGFβRII. Anti-CTLA4-TGFβRII was designed by fusing the C terminus of the HC of a human anti-CTLA-4 antibody (Ipilimumab) with a ligand-binding sequence of the extracellular domain of TGFβRII (TGFβRII ECD) via a flexible linker peptide, (GGGGS) 3 . Anti-PDL1-TGFβRII was designed to simultaneously target both PD-L1 and TGF-β by fusing the C terminus of the HC of human anti-PD-L1 antibody (Atezolizumab and Avelumab) with a ligand-binding sequence of the extracellular domain of TGFβRII (TGFβRII ECD) via a flexible linker peptide, (GGGGS) 3 . The amino acid sequences were codon optimized with GeneOptimizer (Life Technologies). Anti-gp120-TGFβRII antibody was used as a non-specific IgG-TGFβRII control. Amino acid sequences of all fusion antibodies used in this study are provided in Supplementary Material. The complementary DNA for the antibody HC and the cDNA for the antibody LC were gene synthesized and subsequently cloned into separate plasmids (pEvi3; evitria AG, Switzerland) under the control of a mammalian promoter and polyadenylation signal. Plasmid DNA was amplified in Escherichia coli and DNA was purified using anion exchange kits for low endotoxin plasmid DNA preparation. The plasmid DNAs for HC and LC were subsequently co-transfected into CHO K1 cells with eviFect (evitria AG, Switzerland), and the CHO cells were cultured in eviMake (evitria AG, Switzerland), a serum-free, animal-componentfree medium. Production was terminated once viability reached 75%, which occurred at day 8 after transfection. The antibody-containing supernatant was then harvested and antibody was purified at 20 o C by Protein A affinity chromatography on a Bio-Rad BioLogic FuoFlow FPLC machine with subsequent gel filtration as polishing and re-buffering step. The purified antibody was re-buffered into phosphate-buffered saline, sterile-filtered, aliquoted, and frozen at − 80 o C. Protein identification of the purified antibody from CHO cell supernatants was performed by LC-MS/MS to confirm the amino acid sequence and size of the HC of a-CTLA4-TGFβRII and a-PDL1-TGFβRII (Mass Spectrometry and Proteomics Facility, Johns Hopkins University School of Medicine). SDS-PAGE under R and NR conditions was used to compare the FL, HC, and LC of a-CTLA4-TGFβRII with a-CTLA-4, and a-PDL1-TGFβRII with a-PD-L1.
Bifunctional target-binding ability of fusion antibodies. The ability of anti-CTLA4-TGFβRII antibody to simultaneously bind both CTLA-4 and TGFβ was evaluated by a 'double-sandwich' ELISA, wherein anti-CTLA4-TGFβRII or anti-CTLA-4 antibody (1 μg ml −1 ) was added to CTLA-4-Fc-coated plates, followed by rhTGFβ1 (0-2,000 pg ml −1 ) that was detected by a biotinylated anti-human TGFβ1 antibody (R&D Systems). The positive standard curve (TGFβRII-Fc-coated plate) was used to analyze the binding ability of the test samples. The ability of a-CTLA4-TGFβRII to bind TGFβ1 was also evaluated by competition ELISA. The ELISA plate was coated with the capture antibody (a-TGFβ, 1 μg ml −1 ), followed by rhTGFβ1 in the presence of either a-CTLA4-TGFβRII or a-CTLA-4 (Antibody : TGFβ1 ratio 1 : 1 to 100 : 1) for 1 h at room temperature. Each experiment was performed twice, with triplicate wells for each indicated condition.
TGFβ-binding ability of a-TGFβ and IgG-TGFβRII. The ability of a-TGFβ (1D11) and nonspecific IgG-TGFβRII (anti-gp120-TGFβRII) antibody to equally bind TGFβ in vitro was evaluated by a standard ELISA assay (Supplementary Figure 1). rhTGFβ1 (0-2000 pg ml −1 ) was added to the plates coated with either TGFβRII-Fc (R&D Systems), a-TGFβ (Bioxcel), or IgG-TGFβRII (1 μg ml −1 each), and binding to rhTGFβ1 was detected by a biotinylated anti-human TGFβ1 antibody (R&D Systems). TGFβRII-Fc-coated plates were used as a TGFβ-binding positive control (Supplementary Figure 1). To demonstrate that both agents were administered at doses sufficient to saturate systemic TGFβ in vivo, sequestration of serum TGFβ was assessed in A375 tumor-bearing NSG immune-reconstituted mice treated with either a-TGFβ or IgG-TGFβRII (5 mg kg −1 per week) for 4 weeks (Supplementary Figure 2). At the endpoint, serum was collected from tail bleed and levels of TGFβ were detected using the TGFβ-1 Human ELISA Kit (ThermoFisher Scientific) following the manufacturer's protocol.
Analysis of Treg suppressor function. Patient-derived tumor-infiltrated BM (myeloma-BM) was stained with anti-CD3 and Glycophorin A to enumerate T cells. Following activation for 7 days with rhIL-2 and anti-CD3/anti-CD28 beads, the cells were magnetically separated and stained with anti-CD3. Concurrently, autologous Tregs were isolated from the same patient's peripheral blood lymphocytes using anti-CD4/anti-CD25 beads (Miltenyi Biotechnology). The activated T cells were CFSE labeled (Life Technologies) and added to autologous BM that had been pulsed for 30 min in medium with or without either tumor-specific antigen (tumor cell lysate) or nonspecific antigen (nonspecific cell lysate) and plated in the presence or absence of the selected autologous Tregs. Following culture for 3 days with or without either a-CTLA4-TGFβRII or a-CTLA-4 antibody (5 μg ml −1 ), the cells were stained with anti-CD3 and anti-IFNγ. Tumor antigenspecific T cells were considered as CD3 + /CFSE low /IFNγ + .
Tumor cell lines and treatments. A375 and SK-MEL-5 human melanoma cell lines were purchased from ATCC and maintained according to ATCC guidelines. MDA-MB-231 is a metastatic human TNBC cell line with mesenchymal-like morphology (Basal B-like). MDA-MB-231-Luc (D3H2LN) is a TNBC subline with enhanced primary tumor growth and lung metastases that was derived from a metastatic deposit of bioluminescent MDA-MB-231 cells stably expressing firefly luciferase. MDA-MB-231 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and penicillin/streptomycin. All cell lines were periodically monitored for mycoplasma at Johns Hopkins Genetic Resources Core Facility using the MycoDtect kit (Greiner Bio-One) and authenticated using genetic fingerprinting (Identifiler, Applied Biosystems) before use.
Immunophenotype analysis of human tumor-bearing mice. Tumors and BM were collected from tumor-bearing mice in each group for immunophenotype analysis of T cells. Tumor samples were subjected to collagenase digestion for 20 min at 37°C followed by red blood cell lysis. Tregs in tumor-infiltrating or BM T cells were measured by flow cytometric analysis of CD4 + CD25 + CD127 low FOXP3 + cells, as described above. Tumor-infiltrating or BM T cells were stained with anti-human CD3-PE, anti-human CD8-PE-CY TM 7, anti-human CD45RO-APC, and anti-human CD62L-FITC, and analyzed by flow cytometry to quantify T cells with a central memory phenotype (CD45RO high CD62L high ). To evaluate tumor-specific IFN-γ expression in CD3 + /CD8 + T cells, BM cells were plated in 96well plates (2 × 10 5 cells per well) in the presence of tumor cell lysate, nonspecific control peptide, or medium alone. Cells were cultured for 72 h followed by incubation with Golgi stop for 4 h. Cells were stained extracellularly with anti-CD3-FITC and anti-CD8-APC antibodies, permeabilized, stained intracellularly with anti-IFN-γ-PE or its corresponding isotype control, and then analyzed by flow cytometry. All the antibodies were from BD Biosciences.
Bioluminescent imaging of primary and metastatic tumors. Tumor burden in mice bearing MDA-MB-231-Luc was assessed by visualization of in vivo luciferase activity using a Xenogen IVIS Spectrum system. Images were acquired at 10 min post injection of 50 mg kg −1 i.p. dose of luciferin. To detect metastases, the lower portion of each animal was shielded before re-imaging to minimize bioluminescence from the primary tumor. Lungs were harvested and imaged ex vivo to confirm in vivo observations. Photon flux was used to quantify the differences in tumor burden between treatment groups.
Human cell and tissue samples. Approval for research on human subjects was obtained from The Johns Hopkins University Institutional Review Board. This study qualified for exemption under the U.S. Department of Health and Human Services policy for protection of human subjects (45 CFR 46.101(b)) (IRB 03-11-12-06e). BM, peripheral blood, and tumor samples were obtained from patients and normal donors under informed consent in accordance with the Declaration of Helsinki with approval from the Institutional Review Board at Johns Hopkins University. Plasma was removed by centrifugation and stored at − 80°C. Lymphocytes were obtained by Ficoll-Hypaque density gradient centrifugation (GE Healthcare).
Animal use and care. The animals were maintained in accordance with guidelines of the American Association of Laboratory Animal Care and a research protocol approved by the Johns Hopkins University Animal Use and Care Committee.
Statistical analyses and interpretation. All data are presented as the mean ± SEM. Student's unpaired t-test (two-sided) was used to analyze differences between two groups. When appropriate, the Bonferroni correction was applied to account for multiple comparisons. Results with p < 0.05 were considered significant. Statistical analyses were conducted using GraphPad InStat (GraphPad Software). Pvalues are summarized as: *P ≤ 0.05, **P ≤ 0.01, and ***P ≤ 0.001, unless stated otherwise.