Research ArticleCancer

Selective inhibition of TGFβ1 activation overcomes primary resistance to checkpoint blockade therapy by altering tumor immune landscape

See allHide authors and affiliations

Science Translational Medicine  25 Mar 2020:
Vol. 12, Issue 536, eaay8456
DOI: 10.1126/scitranslmed.aay8456

This isoform is just right

Cancer immunotherapy, including immune checkpoint blockade, has achieved increasing prominence in recent years. Unfortunately, only a fraction of tumors respond to this treatment. Previous studies have suggested that inhibition of transforming growth factor–β (TGFβ) may help overcome resistance to immune checkpoint blockade, but it proved to be too toxic for clinical use. Martin et al. designed a more specific inhibitor, targeting only one isoform of TGFβ. The authors showed that this isoform, TGFβ1, is the most relevant one to target in tumors, then demonstrated the effectiveness and safety of their inhibitor with immune checkpoint blockade in multiple mouse models of cancer.

Abstract

Despite breakthroughs achieved with cancer checkpoint blockade therapy (CBT), many patients do not respond to anti–programmed cell death-1 (PD-1) due to primary or acquired resistance. Human tumor profiling and preclinical studies in tumor models have recently uncovered transforming growth factor–β (TGFβ) signaling activity as a potential point of intervention to overcome primary resistance to CBT. However, the development of therapies targeting TGFβ signaling has been hindered by dose-limiting cardiotoxicities, possibly due to nonselective inhibition of multiple TGFβ isoforms. Analysis of mRNA expression data from The Cancer Genome Atlas revealed that TGFΒ1 is the most prevalent TGFβ isoform expressed in many types of human tumors, suggesting that TGFβ1 may be a key contributor to primary CBT resistance. To test whether selective TGFβ1 inhibition is sufficient to overcome CBT resistance, we generated a high-affinity, fully human antibody, SRK-181, that selectively binds to latent TGFβ1 and inhibits its activation. Coadministration of SRK-181-mIgG1 and an anti–PD-1 antibody in mice harboring syngeneic tumors refractory to anti–PD-1 treatment induced profound antitumor responses and survival benefit. Specific targeting of TGFβ1 was also effective in tumors expressing more than one TGFβ isoform. Combined SRK-181-mIgG1 and anti–PD-1 treatment resulted in increased intratumoral CD8+ T cells and decreased immunosuppressive myeloid cells. No cardiac valvulopathy was observed in a 4-week rat toxicology study with SRK-181, suggesting that selectively blocking TGFβ1 activation may avoid dose-limiting toxicities previously observed with pan-TGFβ inhibitors. These results establish a rationale for exploring selective TGFβ1 inhibition to overcome primary resistance to CBT.

INTRODUCTION

Advances in immunotherapy have transformed the effective treatment landscape for a growing number of patients with cancer. Most prominent are the checkpoint blockade therapies (CBTs), which have now become part of standard-of-care regimens for an increasing number of cancers. Although profound and durable CBT responses are observed across a growing number of cancer types, a substantial fraction of tumors remain unresponsive to these therapies. This is true even at the outset of treatment, suggesting that additional mechanisms may be preventing many patients’ immune systems from targeting and eliminating tumor cells (1). Approaches aimed at extending treatment efficacy to a greater number of patients include combining CBTs with other agents that either affect the same tumor type or modulate seemingly relevant immune system components. Unfortunately, many of these empirical combinations lack a clear, clinically derived, mechanistic rationale, and to date, most have not shown meaningful clinical benefit in comparison to CBT alone (24). A more thorough understanding of the underlying mechanisms driving primary CBT resistance is therefore a growing focus in cancer immunotherapy. Among these efforts are retrospective analyses of patient tumors and clinical trial data, from which transforming growth factor–β (TGFβ) pathway activation has emerged as a critical mediator of primary CBT resistance. Transcriptional profiling of pretreatment melanoma biopsies from patients who are nonresponsive to anti–programmed cell death-1 (PD-1) CBT revealed enrichment of TGFβ-associated pathways (5). More recently, similar analyses in metastatic urothelial cancer also uncovered a correlation between TGFβ-associated transcriptional signatures and lack of response to programmed cell death-ligand 1 (PD-L1) blockade, particularly in tumors where CD8 T cells appear to be excluded from tumor entry (6). These observations in clinical samples are further bolstered by studies in mouse syngeneic tumor models. In the EMT-6 breast cancer model, which is poorly responsive to anti–PD-L1 treatment, combining checkpoint inhibition with TGFβ inhibition resulted in a profound increase in the frequency of complete responses (6). This combination therapy antitumor activity correlated with phenotypic changes in cancer-associated fibroblasts (CAFs) and infiltration of activated CD8 T cells into the tumors. Pairing PD-L1 blockade with a small-molecule inhibitor of the TGFβ type I receptor ALK5 in a murine colorectal cancer model also yielded similar results (7). Collectively, the clinically derived correlative evidence and the preclinical validation results point to TGFβ pathway inhibition in CBT-refractory tumors as a promising approach to increase the number of clinical responses to CBT.

Mammals express three TGFβ growth factors, TGFβ1, TGFβ2, and TGFβ3, each encoded by a distinct gene and all signaling through the same TGFβ receptor complex. All three isoforms are expressed as inactive protein complexes, in which the TGFβ prodomain, also called latency-associated peptide (LAP), wraps around its growth factor, holding it in a latent, nonsignaling state (8). This latent TGFβ complex, also called the small latent complex (SLC), is coexpressed and disulfide linked with one of several TGFβ-binding proteins to form large latent complexes (LLCs). Several LLC-presenting proteins have been identified: Latent TGFβ binding protein-1 (LTBP1) and LTBP3 tether the associated latent TGFβ protein to the extracellular matrix, whereas the transmembrane proteins glycoprotein A repetitions predominant (GARP) and leucine-rich repeat-containing protein 33 (LRRC33) present TGFβs on the surface of regulatory T (Treg) cells or macrophages, respectively (912). In vivo, latent TGFβ1 and latent TGFβ3 are activated by a subset of αV integrins, binding a consensus RGD (Arg-Gly-Asp) sequence on LAP and triggering a conformational change that releases the growth factor (13, 14). TGFβ1 may also be released through proteolytic cleavage of LAP, although the biological relevance of this activation mechanism is less clear (15, 16). The mechanism of latent TGFβ2 activation is unknown because it lacks a consensus RGD motif.

Despite its clear role in disease-relevant biology, therapeutic targeting of the TGFβ pathway has proven challenging. Small molecule–mediated TGFβ type I receptor kinase ALK5 inhibition or antibody blockade of all three TGFβ isoforms resulted in severe cardiac valvulopathies in mice, rats, and dogs (1719), suggesting that broad inhibition of signaling driven by one or more TGFβ isoforms will limit the clinical utility of such agents. In light of the observed preclinical dose-limiting toxicities, the acceptable clinical dosing regimens for these “pan-TGFβ” inhibitors have, to date, proven to be ineffective (2022). As a consequence, no TGFβ pathway–targeting therapy has been approved to date, impeding treatment of TGFβ-driven disease pathogenesis (23).

One potential approach to therapeutically targeting TGFβ activity while avoiding on-target toxicity is specific inhibition of the TGFβ isoform that may be the key driver of disease-relevant processes. Despite using a common set of receptors, mouse TGFβ1, TGFβ2, and TGFβ3 knockout animals have nonoverlapping phenotypes, suggesting distinct biological functions (2426). Mouse and human genetic data suggest that the cardiac phenotypes associated with pan-TGFβ inhibition may be linked to loss of TGFβ2 and, possibly, TGFβ3 activity. Although Tgfb2 knockout mice exhibit a range of developmental phenotypes, including congenital heart defects (25), Tgfb2 haploinsufficient mice develop normally but show dilation of the aortic annulus by ~8 months old (27). In humans, loss-of-function mutations in the TGFB2 gene have been identified in patients with thoracic aortic aneurysm dissections (TAAD) in Marfan and Loeys-Dietz syndromes (27, 28), with nonsyndromic TAAD (29), and with mitral valve disease (29, 30). Predicted partial loss-of-function mutations in TGFB3 have also been associated with aortic defects in patients with Loeys-Dietz (31), as well as with cardiac arrhythmias (32). Whether similar cardiac phenotypes are present in knockout mice is unclear, however, because complete deletion of the Tgfb3 gene causes perinatal lethality due to severe cleft palate and craniofacial defects (24). Together, these data point to potentially important homeostatic roles for TGFβ2 and TGFβ3 in cardiac function, suggesting that specific TGFβ1 isoform inhibition may be a strategy to avoid the cardiotoxicity associated with pan-TGFβ approaches.

Here, we uncover evidence that TGFβ1 is the likely driver of changes in the tumor microenvironment, and in particular in the tumor immune contexture, that render such tumors resistant to CBT. We describe SRK-181, a potent and highly selective inhibitor of latent TGFβ1 activation, and we demonstrate that specific inhibition of the TGFβ1 isoform is sufficient to overcome primary resistance to anti–PD-1 in syngeneic mouse tumor models that closely recapitulate features of primary resistance to CBT found in human cancers. Furthermore, in rat toxicology studies previously shown to reveal the above-mentioned dose-limiting toxicities seen with broad pathway inhibition, we show an improved safety profile for TGFβ1-specific inhibition when compared with pan-TGFβ or ALK5-targeted therapies. Together, these data provide a strong rationale for the therapeutic use of selective TGFβ1 inhibition, and in particular SRK-181, to broaden and enhance clinical responses to checkpoint blockade in immunotherapy of patients with cancer.

RESULTS

TGFB1 gene expression is prevalent in many types of human tumors

Previous analyses implicated TGFβ signaling as an important contributor to primary CBT resistance in human tumors (57). In urothelial cancers, TGFΒ1 appears to be one of the most highly correlated TGFβ pathway genes associated with nonresponse to atezolizumab (6), suggesting that this isoform may be particularly important in driving immunomodulatory TGFβ signaling in the tumor microenvironment. To further investigate whether this finding is more broadly relevant across human cancers, TGFΒ1, TGFΒ2, and TGFΒ3 mRNA expression was evaluated using RNA sequencing (RNA-seq) data from The Cancer Genome Atlas (TCGA). Comparative analysis revealed TGFΒ1 as the most prevalent isoform that is expressed in the majority of human cancer types (Fig. 1A). Notable exceptions were breast cancer, mesothelioma, and prostate cancer, where expression of other family members, particularly TGFΒ3, was at least as prevalent as TGFΒ1. When examined at the level of individual samples from seven tumor types for which CBTs are currently approved for use in therapeutic intervention, TGFΒ1 mRNA expression appeared to be greater and more frequent relative to TGFΒ2 and TGFΒ3, again with the notable exception of breast carcinoma (fig. S1A). These and previously published observations in urothelial cancer (6) suggest that TGFβ pathway activity may be driven by TGFβ1 activation in most human tumors. To establish a more direct link between isoform expression and TGFβ signaling, we determined whether correlations existed between TGFβ isoform expression in these seven tumor types and two different gene signatures of TGFβ pathway activation (33, 34). Using either gene set, TGFΒ1 mRNA most consistently and significantly correlated with TGFβ pathway activation (P ≤ 10−6; Fig. 1B and fig. S1B). To further confirm this association of TGFβ1 expression, we assessed the relationship of TGFβ isoform expression to a transcriptional signature of innate anti–PD-1 resistance [IPRES (5)]. Across these seven tumor types, we found significant correlations between TGFΒ1 mRNA expression and IPRES (P ≤ 10−9), whereas either no or limited correlations were observed for TGFB2 and TGFB3 mRNA expression (Fig. 1C). This suggests that TGFβ1 is the most likely isoform that would contribute to primary CBT resistance in these tumor microenvironments. Together, these observations suggested that selective TGFβ1 inhibition may be sufficient to overcome TGFβ pathway–driven primary CBT resistance in most, if not all, tumor types.

Fig. 1 TGFβ1 isoform expression is prevalent in most human tumors and correlates with transcriptional signatures of TGFβ pathway activation.

(A) Heatmap showing percentage of patient tumor samples from TCGA database scoring positive (fragments per kilobase million, FPKM ≥30) for each TGFβ isoform. Tumor types are stratified on the basis of TCGA annotation. (B) Correlation (R, Pearson’s coefficient) analysis of TGFβ isoform expression (FPKM ≥10 cutoff) and a TGFβ pathway activation gene signature from Plasari et al. (33) across selected TCGA-defined tumor types, for which CBT therapies are currently used for therapeutic intervention; significance of association determined by two-sided t test. (C) Similar analysis as described in (B) but using the IPRES transcriptional signature (5) across the same set of tumor types. BLCA, bladder urothelial carcinoma; BRCA, breast invasive carcinoma; ESCA, esophageal carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; SKCM, skin cutaneous melanoma; STAD, stomach adenocarcinoma.

Discovery of SRK-181: A potent and selective inhibitor of latent TGFβ1 activation

To test the hypothesis that TGFβ1 inhibition can surmount primary CBT resistance, we identified SRK-181, a highly selective antibody inhibitor of TGFβ1 activation. Given the high sequence and structural similarity between mature TGFβ1 growth factor and its closely related family members, TGFβ2 and TGFβ3 (table S1), we reasoned that identification of selective, high-affinity antibodies targeting the active TGFβ1 growth factor would likely be a major challenge. However, recent insights into the structure and mechanical aspects of latent TGFβ1 activation by integrins (8, 13) suggested the possibility of targeting latent TGFβ1 complexes to inhibit release of the active growth factor. Such an approach would exploit the lower sequence similarity within the TGFβ1, TGFβ2, and TGFβ3 prodomains, allowing isoform selectivity in both binding and blocking activation (table S1).

A key consideration in this approach is that latent TGFβ1 assembles into disulfide-linked LLCs that present the latent procomplex either in the extracellular matrix or on the cell surface. TCGA data revealed that essentially all tumor types express mRNA encoding all four LLC-presenting molecules, namely, LTBP1, LTBP3, GARP, and LRRC33 (fig. S1C), raising the possibility that multiple TGFβ1 LLCs may be present in the tumor microenvironment. Because of this, we sought to identify antibodies that would specifically bind and inhibit latent TGFβ1 activation in all of these local contexts.

Soluble murine and human forms of each TGFβ1 LLC were designed, expressed, purified, characterized (fig. S2, A and B), and used to screen a naïve human antibody library presented on yeast surfaces. Noncomplexed LLC-presenting molecules were also used in negative selection steps to ensure identification of selective latent TGFβ1 binders. On the basis of its selective binding to all TGFβ1 LLCs and inhibition of latent TGFβ1 activation triggered by integrins, an initial antibody with moderate affinity was further optimized through CDRH1/2 loop shuffling and CDRH3 mutagenesis using the same antibody platform. From this process, we identified SRK-181, a fully human immunoglobulin G4 (IgG4)/kappa antibody with a hinge-stabilizing S228P mutation to prevent Fab arm exchange (35). SRK-181 binds latent TGFβ1 complexes with minimal or no binding to latent TGFβ2 or latent TGFβ3 complexes (Fig. 2A). Furthermore, SRK-181 does not bind any of the three active TGFβ growth factors (fig. S3). Solution equilibrium titration was used to confirm high binding affinities of SRK-181, with dissociation constant (KD) values in the low picomolar range to all four human TGFβ1 LLCs (Fig. 2B). Similar affinities were determined in mouse, rat, and cynomolgus monkey TGFβ1 complexes.

Fig. 2 SRK-181 is an isoform-specific, anti-latent TGFβ1 antibody that inhibits TGFβ1 activation.

(A) SRK-181 was captured on anti-human Fc biosensors, and binding to human LTBP1 LLCs with TGFβ1, TGFβ2, or TGFβ3 was measured by biolayer interferometry. Association phase was measured over 10 min by immersing the sensor into 100 nM antigen solution, and subsequent dissociation was recorded in buffer over 10 min. (B) Affinities of SRK-181 for TGFβ1 LLCs from different species were determined by equilibrium titration. Best-fit KD values were calculated in Prism by nonlinear regression analysis of binding data and are indicated as means ± SE from one to two experiments, each performed in duplicate; n.d., not determined. (C) LN229 cells were transfected with a plasmid encoding human proTGFβ1 or human proTGFβ3 (LLC formation with endogenous LTBP) or cotransfected with plasmids encoding human proTGFβ1 plus either GARP or LRRC33 for LLC formation on the cell surface. TGFβ activity was measured with CAGA12 reporter cells. Data are averages ± SD from two independent experiments, each performed in triplicate. Curves are best fit to dose-response inhibition model. (D) CAGA12 reporter cells were cultured for about 18 hours with TGFβ1 SLC in the absence or presence of human plasma kallikrein (KLK). Data are averages ± SD of one experiment run in quadruplicate and representative of at least three independent experiments. (E) Human CD4+ T effector cells (Teff) and autologous regulatory T (Treg) cells were isolated from freshly isolated human PBMCs. Teff cells were labeled with CellTrace Violet, activated, and cultured in absence or presence of Treg cells and the indicated antibodies (1 or 10 μg/ml) under T cell–activating conditions for 5 days. Teff division was determined by measuring CellTrace Violet dilution using flow cytometry. Shown are individual data points plus mean from one experiment run in duplicate, which is representative of two independent experiments run with cells from two separate donors. **P ≤ 0.01; ***P ≤ 0.001; Student’s two-tailed t test.

To measure the effect of SRK-181 on integrin-mediated latent TGFβ1 activation, we developed a series of cell-based assays enabling TGFβ1 presentation and activation in each of the LLC contexts. Human LN229 glioblastoma cells endogenously express LTBP1 and LTBP3 mRNA (fig. S4A), which, when transfected with a TGFβ1- or TGFβ3-encoding plasmid, can deposit TGFβ LLCs into the extracellular matrix. Because LN229 cells do not express GARP or LRRC33 mRNA (fig. S4A), assays for LLCs containing these molecules were generated by cotransfecting plasmids encoding one of these presentation molecules in conjunction with a TGFβ1 expression construct. LN229 cells also express αVβ8 integrins (36) and can therefore activate latent TGFβ1 complexes (37, 38). Therefore, once TGFβ1 LLCs are deposited into the extracellular matrix or elaborated on the cell surface, they can be activated by endogenous αVβ8 integrins expressed by the same cells. TGFβ1 released by these integrins can then engage its cognate receptor, and this signaling is measured by coculturing with a luciferase-based reporter cell line. All TGFβ1 LLCs were readily activated in these assays. We further note that cotransfection of GARP or LRRC33 into LN229 cells expressing latent TGFβ1 resulted in a substantially higher TGFβ1 signal, confirming formation and activation of cell surface TGFβ1 LLCs that are apparently sufficient to outcompete endogenous LTBPs for intracellular LLC assembly (fig. S4B).

SRK-181 inhibited activation of all human TGFβ1 LLCs in a concentration-dependent fashion, with median inhibitory concentration (IC50) values between 1.15 nM (0.86 to 1.55 nM 95% confidence interval) and 1.42 nM (1.08 to 1.89 nM; Fig. 2C). Inhibition of mouse TGFβ1 activation was similar to human complexes, in line with the observed and similar affinities of SRK-181 for human and murine complexes (fig. S5). Consistent with the lack of SRK-181 binding to the LTBP1-TGFβ3 LLC, little to no LTBP-TGFβ3 inhibition was observed, demonstrating that SRK-181 selectively inhibits TGFβ1 activation (Fig. 2C). SRK-181 also blocked activation of latent TGFβ1 by human plasma kallikrein (Fig. 2D), indicating that SRK-181 may prevent TGFβ1 activation by multiple putative activation mechanisms.

To assess SRK-181–mediated inhibition of a biologically relevant consequence of TGFβ1 activation, we examined its effect on suppressive activity of primary human Treg cells. Sorted CD4+CD25hiCD127lo Treg cells up-regulate surface expression of GARP-TGFβ1 LLC upon T cell receptor stimulation (fig. S6). These activated Treg cells suppressed proliferation of autologous effector CD4 T cells, whereas addition of SRK-181 blocked this suppressive Treg cell activity at concentrations as low as 1 μg/ml (Fig. 2E). These results are consistent with previous observations that Treg cells harness TGFβ signaling to suppress T cell activity (39, 40).

Binding site of SRK-181

To identify possible interaction sites between SRK-181 and latent TGFβ1, which may help further understand its inhibitory mechanism, we performed hydrogen deuterium exchange mass spectrometry (H/DX-MS) analysis. We achieved excellent peptide coverage of 89.6% (fig. S7A), revealing three regions on TGFβ1 SLC that were protected from deuterium exchange by SRK-181 Fab binding (Fig. 3, A and B, and fig. S7, B to D). Region 1 lies within the TGFβ1 prodomain, whereas regions 2 and 3 map to the TGFβ1 growth factor. Region 1 is part of the latency lasso and, consistent with our observation that SRK-181 inhibits kallikrein-mediated activation of latent TGFβ1, contains the proteolytic cleavage sites for both plasmin and kallikrein proteases (8, 15) (Fig. 3B). SRK-181 does not bind to any of the three TGFβ growth factor dimers (fig. S3), implying that any potential interactions with sites on the growth factor are dependent on prodomain interactions. The H/DX protection sites also suggest that SRK-181 blocks integrin-dependent TGFβ1 activation through an allosteric inhibition mechanism because the antibody binds outside the “trigger loop” within the prodomain, which contains the RGD integrin binding site (Fig. 3A). We further confirmed this observation experimentally, showing that SRK-181 does not block latent TGFβ1 interaction with a recombinant αVβ6 integrin ectodomain (fig. S8). This H/DX analysis explains the observed SRK-181 selectivity for latent TGFβ1 versus latent TGFβ2 and latent TGFβ3 complexes because sequence alignment of putative epitope regions 1 to 3 revealed substantial sequence divergence across the three TGFβ isoforms (Fig. 3B).

Fig. 3 SRK-181 binding protects three regions on latent TGFβ1 from hydrogen/deuterium exchange.

(A) Region 1 (red), region 2 (orange), and region 3 (yellow) on latent TGFβ1, which are protected from hydrogen/deuterium exchange (H/DX) upon SRK-181 Fab binding, are mapped onto a model structure of the human TGFβ1 SLC homodimer in surface representation. The TGFβ1 prodomain is shown in blue, the TGFβ1 growth factor is shown in green, and the RGD integrin binding site in the prodomain is highlighted in magenta for orientation. The location of H/DX-protected regions is consistent with SRK-181 binding to the latency lasso region of latent TGFβ1. (B) Amino acid sequences of the three H/DX-protected regions. Region 1 is in the TGFβ1 prodomain, and regions 2 and 3 are on the TGFβ1 growth factor. Highlighted in region 1 are the proteolytic sites for both plasmin and kallikrein in the prodomain. TGFβ2 and TGFβ3 sequences corresponding to the three H/DX-protected regions on TGFβ1 are shown as well. The dots (•) represent amino acid sequence diversity across the three TGFβ isoforms.

Combination treatment effects of SRK-181 with anti–PD-1 on tumor growth in CBT-resistant tumors

SRK-181 enables direct evaluation of the hypothesis that selective TGFβ1 inhibition is sufficient to overcome tumor CBT resistance. For preclinical testing, we sought to identify murine syngeneic tumor models that recapitulate key features of human tumors that exhibit primary resistance to CBT, including (i) limited or no anti–PD-(L)1 single-agent treatment response at doses that are efficacious in other syngeneic tumor models, (ii) evidence for immune exclusion with a dearth of infiltrating CD3+ T cells, (iii) evidence of active TGFβ signaling, and (iv) evidence of TGFβ1 isoform expression.

Using single-agent anti–PD-1 tumor response data and publicly available RNA-seq datasets from whole tumor–derived RNA, we selected three tumor models meeting these criteria to test the TGFβ1 hypothesis: the MBT-2 bladder cancer model (MBT-2), the Cloudman S91 (S91) melanoma model, and the EMT-6 breast cancer model (EMT-6). In each model, analysis of whole-tumor RNA-seq data revealed expression of TGFβ-responsive genes, suggesting TGFβ pathway activation and low effector T cell gene expression compared with myeloid cell markers, consistent with an immune excluded phenotype (fig. S9A). Enzyme-linked immunosorbent assay analysis also demonstrated that TGFβ1 protein is the most highly expressed family member in all three tumor models (fig. S9B). Whereas MBT-2 and S91 tumors contained little to no detectable TGFβ2 and TGFβ3 growth factors, EMT-6 tumors expressed the TGFβ3 isoform, consistent with RNA-seq profiling and published data (fig. S9A) (6). Previous studies with the EMT-6 tumor model and our initial studies of the MBT-2 and S91 models confirmed their resistance to anti–PD-(L)1 treatment (fig. S9C) (6). In addition, TGFβ signaling appears to be involved in mediating checkpoint blockade resistance in the MBT-2 and S91 models because combination of the pan-TGFβ growth factor antibody 1D11 with anti–PD-1 resulted in tumor growth delay or, in some cases, tumor regressions when compared to no responses with anti–PD-1 treatment alone (fig. S9C).

We then sought to assess whether SRK-181 would render these tumor models sensitive to anti–PD-1 treatment. To minimize SRK-181 immunogenicity in mouse syngeneic tumor models, we produced SRK-181 as a chimeric antibody with the human V domains of SRK-181 fused to mouse IgG1/kappa constant domains. This chimera, called SRK-181-mIgG1, had similar inhibitory activity to the fully human SRK-181 (fig. S9D). First, we investigated the role of TGFβ1 in the highly aggressive MBT-2 tumor model (Fig. 4A). As observed in our initial studies, MBT-2 tumor–bearing animals did not respond to anti–PD-1 (clone RMP1-14) or SRK-181-mIgG1 treatment alone (Fig. 4B). However, the combination of anti–PD-1 and SRK-181-mIgG1 resulted in a reduction in tumor burden in several mice, including 21% (3 mg/kg per week) and 36% (10 mg/kg per week) complete responses (see Materials and Methods for definition), as well as significant survival benefit over the study duration (P < 0.001; Fig. 4, B and C). In total, 4 of 14 animals responded to anti–PD-1/SRK-181-mIgG1 (3 mg/kg per week), and 8 of 14 responded to anti–PD-1/SRK-181-mIgG1 (10 mg/kg per week) compared with 0 of 13 on anti–PD-1 alone.

Fig. 4 Antitumor effects and survival benefit of SRK-181-mIgG1 combination with anti–PD-1 in checkpoint blockade–resistant syngeneic mouse tumors.

(A) Experimental design for the MBT-2 study. MBT-2 cells were subcutaneously implanted, and dosing was initiated 14 days later, when group mean tumor volumes reached 56 to 57 mm3. Study ended on day 32. (B) Tumor growth (mm3) over time during treatment of established subcutaneous MBT-2 tumors. SRK-181-mIgG1 or its isotype control was dosed once weekly at 10 mg/kg except where indicated. Anti–PD-1 or its control antibody was dosed twice weekly at 10 mg/kg. Tumor ulceration is a common feature of this model, and animals with pronounced ulcerations were euthanized per veterinarian discretion and removed from subsequent survival analyses, yet represented here as dotted lines and further described in table S2. Red dashed line is at 1200 mm3, the IACUC-defined study endpoint. Blue dashed line is at 300 mm3 or 25% of endpoint volume, defining responders, which are indicated as a fraction of evaluable animals in each group. Data are representative of two independent studies. (C) Kaplan-Meier survival plots from (B). ***P < 0.001 log-rank test versus anti–PD-1. (D) Experimental design for the Cloudman S91 study. Cloudman S91 tumor cells were implanted, and dosing was initiated 16 days later, when group mean tumor volumes reached 126 to 132 mm3. Dosing continued for 72 days. (E) Tumor growth (mm3) over time during treatment of established subcutaneous Cloudman S91 tumors. SRK-181-mIgG1 or its isotype control were dosed once weekly at 30 mg/kg, except where indicated. Anti–PD-1 or its control antibody was dosed twice weekly at 10 mg/kg. Red dashed line is at 2000 mm3, IACUC-defined study endpoint. Blue dashed line is at 25% endpoint tumor volume (500 mm3), defining responders, indicated as a fraction of evaluable animals in each group. Limb fractures are common in this model and are considered non–treatment related. Animals with broken limbs were euthanized per veterinarian discretion and were not included in subsequent survival analysis but plotted here as dotted lines and further detailed in table S2. Data are representative of two independent studies. (F) Kaplan-Meier survival plots from (E). **P < 0.01; ***P < 0.001 log-rank test versus anti–PD-1. (G) Experimental design for the EMT-6 study. EMT-6 cells were implanted, and dosing was initiated 3 days later, when group mean tumor volumes reached 39 to 41 mm3 and continued throughout the length of the study, which ended on day 56. (H) Tumor growth (mm3) during treatment of established subcutaneous EMT-6 tumors. SRK-181-mIgG1 or its isotype control was dosed once weekly at 10 mg/kg. Anti–PD-1 or its control antibody was dosed twice weekly at 10 mg/kg. The pan-TGFβ antibody 1D11 was dosed twice a week at 5 mg/kg. Red dashed line is at 2000 mm3, IACUC-defined study endpoint. Blue dashed line is at 25% of endpoint tumor volume (500 mm3), defining responders, which are indicated as a fraction of evaluable animals in each group. Data are representative of two independent studies. (I) Kaplan-Meier survival plots from (H). **P < 0.01; log-rank test versus anti–PD-1.

To further assess the extent and duration of antitumor responses observed in the combination study groups, treatment of surviving animals was stopped on study day 32, and the animals were evaluated for possible tumor regrowth. Of the 12 mice that had responded to anti–PD-1/SRK-181-mIgG1 at the end of the dosing period, 10 mice remained tumor-free 7 weeks after treatment discontinuation, suggesting that the response to combination treatment was durable and likely complete. Further, to assess whether these mice developed durable immune memory, we then subcutaneously reimplanted MBT-2 cells in the flank contralateral to the original implantation and monitored subsequent tumor growth. We observed no detectable tumor growth in any of these mice, whereas all mice in a control, age-matched, tumor-naïve group developed measurable tumors within 3 weeks of implantation (fig. S10A). These results indicate that mice classified as tumor-free survivors appear to have established durable antitumor immunological memory.

To determine whether combining anti–PD-1 treatment with TGFβ1 blockade could be effective in other tumor models, we then similarly evaluated single-agent and combination treatments in the Cloudman S91 melanoma model (Fig. 4, D to F). As with the MBT-2 model, anti–PD-1/SRK-181-mIgG1 combination treatment resulted in profound antitumor effects, with up to 75% of mice exhibiting clear responses that resulted in a significant survival advantage at all dose levels (P < 0.01). Five responders from across the anti–PD-1/SRK-181-mIgG1 combination groups in the Cloudman S91 study did retain a small yet stable residual mass at the implantation site over the remaining treatment period. Although we considered evaluation of tumor rechallenge in this model, the postimplantation take rate of Cloudman S91 tumor cells is low, thus limiting the feasibility of this type of follow-up analysis. For this reason, we could only assess the durability of responses after discontinuation of combination treatment in mice with complete or near-complete responses. Six weeks after treatment cessation, mice with no measurable tumor at the time of cessation remained tumor free. Those with measurable tumors at treatment cessation had mixed responses: Many were cleared, some remained small but palpable, whereas others, notably restricted to the lowest-dose groups for SRK-181-mIgG1, began to grow about 20 days after treatment cessation (fig. S10B). These data indicate that some residual tumors retain the ability to escape or suppress host immune response after removal of combination treatment, suggesting that treatment may need to be maintained until full tumor clearance is achieved.

In contrast to MBT-2 and Cloudman S91 tumors, in which TGFβ1 is the predominantly expressed isoform, the EMT-6 breast carcinoma model expresses similar amounts of Tgfb1 and Tgfb3 mRNA and about fivefold lower amounts of TGFβ3 protein than the TGFβ1 isoform (fig. S9, A and B). To assess whether selective inhibition of TGFβ1 activation is sufficient to overcome anti–PD-1 resistance when multiple TGFβ isoforms are present, we conducted combination treatment studies in this tumor model (Fig. 4, G to I). We observed complete responses in 50% of EMT-6 tumor-bearing mice treated with the anti–PD-1/SRK-181-mIgG1 combination, as well as a significant survival benefit compared with anti–PD-1 or SRK-181-mIgG1 single-agent treatments (P < 0.01; Fig. 4, H and I). Complete responders remained tumor free 6 weeks after treatment cessation, again demonstrating the durability of response (fig. S10C). These results are consistent with previously reported observations in this tumor model using anti–PD-L1 combination with a pan-TGFβ antibody (6). Collectively, our observations of antitumor responses in these three syngeneic tumor models suggest that the local activity of the TGFβ1 isoform within the tumor microenvironment is sufficient to drive immunosuppressive effects that cause primary resistance to anti–PD-1 treatment.

Blockade of latent TGFβ1 activation in combination with anti–PD-1 overcomes immune exclusion

To more thoroughly elucidate the mechanisms by which TGFβ1 signaling drives immune exclusion and prevents responses to anti–PD-1 treatment, we examined changes in tumor immune contexture after anti–PD-1 and SRK-181-mIgG1 combination treatment, in comparison to control and single-agent treatments. MBT-2 tumors were harvested from mice 10 days after treatment initiation and evaluated for changes in select immune cell markers. Although the overall percentage of the CD45+ immune compartment did not change with treatment, the combination of anti–PD-1 and SRK-181-mIgG1 induced a 10-fold increase in the CD8 T cell representation within this compartment relative to isotype control antibody treatment (average of 34 ± 5.6% versus 3.5 ± 1.6%, respectively; Fig. 5A), consistent with an increase in the absolute number of CD8 T cells per milligram of tumor (fig. S11A). Single-agent treatment with anti–PD-1 or SRK-181-mIgG1 did not have an effect on CD8 T cell representation. In addition, analysis of RNA derived from these tumors showed increases in the cytotoxic T cell activation markers perforin and granzyme B, which are consistent with the increase in CD8+ cell number and are indicative of active effector function of these cells (fig. S11B). It is notable that a significant increase in the percentage of CD4+FoxP3+ Treg cells was also observed with combination treatment (P < 0.05). The relevance of this increase is unclear, given the pronounced antitumor effects observed with combination treatment. However, the ratio of CD8:Treg cells was not altered in response to combination treatment (fig. S11C). Intracellular cytokine staining confirmed that these Treg cells expressed low amounts of interferon-γ (IFN-γ) compared with the CD8 T cells, thus further confirming the activated phenotype of the latter (fig. S11D). However, treatment with anti–PD-1 and SRK-181-mIgG1 had no effect on the size of the IFNγ+ CD8 T cell population (fig. S11D).

Fig. 5 Blockade of latent TGFβ1 in combination with anti–PD-1 overcomes immune exclusion.

(A) Flow cytometric analysis of immune cell populations in MBT-2 tumors at day 10 after treatment initiation, five to six animals per group. Mϕ, macrophages; MDSC, myeloid-derived suppressor cells; *P < 0.05; **P < 0.01, ***P < 0.001 Student’s two-tailed t test against anti–PD-1. (B) Representative immunohistochemical staining for CD8 in MBT-2 tumors at day 10 after treatment initiation. Scale bars, 200 μm. (C) Quantification of data exemplified in (B). Area of CD8+ signal per whole-tumor slide scan, n = 5 to 6 animals per group. *P < 0.05, Student’s two-tailed t test against anti–PD-1 group. (D) Representative 40× magnification of immunofluorescence staining for CD8 (cyan) and CD31 (green) in MBT-2 tumors 10 days after treatment initiation. Scale bars, 30 μm. (E) Distribution of CD8+ cells from CD31+ objects from data exemplified in (D) using nearest-neighbor analysis binned by average CD8+ object diameter of 12.4 μm. A distance of “0” is touching a CD31+ object. Data show percentage of CD8+ signal of total tumor CD8+ signal for five to six whole-slide scanned tumors per group.

In addition to the marked increase in cytotoxic CD8 T cell population within the tumors, anti–PD-1/SRK-181-mIgG1 combination treatment also induced a significant reduction in overall CD11b+ cell representation within MBT-2 tumors (P < 0.01; Fig. 5A). This appeared to result from selective reduction in CD11b+F4/80+CD206+ and CD11b+F4/80Gr1+ subpopulations, which correspond to immunosuppressive M2-like macrophages and myeloid-derived suppressor cells (MDSCs), respectively (Fig. 5A and fig. S11A). High expression of Arginase 1 and TGFβ1 LAP by the M2-like macrophages compared with the M1-like macrophages further supports their designation as immunosuppressive myeloid cells (fig. S11E). Collectively, the representation of M2-like macrophages and MDSCs was reduced from an average of 47 ± 3.3% and 10.9 ± 2.7% of the CD45+ cell population in control animals to 14 ± 6.1% and 1.4 ± 0.6% after combination treatment, respectively (Fig. 5A). The M1-like macrophage subpopulation (CD11b+F4/80+CD206) did not appear to change with treatment, indicating that PD-1/TGFβ1 blockade has a selective impact on the immunosuppressive milieu within tumors, beneficially affecting both lymphoid and myeloid compartments.

The mechanism by which combined PD-1 and TGFβ1 inhibition resulted in pronounced CD8 T cell entry and/or expansion into the tumor microenvironment is not clear. Hence, we further examined the relationship between TGFβ pathway activity and immune exclusion using a more detailed histological analysis. First, we confirmed a significant increase in CD8 staining throughout anti–PD-1/SRK-181-mIgG1–treated MBT-2 tumors that was consistent with the flow cytometry data (P < 0.05; Fig. 5, B and C). A similar increase in CD8 T cells after combination treatment with anti–PD-L1/1D11 was previously reported in studies using the EMT-6 tumor model, suggesting that cytotoxic T cell entry and expansion are a common combination response mechanism (6). Closer examination of the treated MBT-2 tumors revealed that many of the CD8 T cells in the anti–PD-1/SRK-181-mIgG1 groups were closely clustered together around what appeared to be blood vessels. CD8/CD31 costaining and digital image quantification indicated that CD8 T cells are enriched in areas adjacent to CD31+ tumor blood vessels (Fig. 5, D and E), suggesting that tumor vasculature may be a route of T cell entry. This was true regardless of treatment and observed in our analysis of the EMT-6 model as well (fig. S11F). Prior studies reported an association between TGFβ signaling and the presence of fibroblast-rich peritumoral stroma, suggesting that the stroma forms a barrier to T cell entry into the tumor (6). Our preliminary observations suggest that an additional, TGFβ1-dependent vascular barrier may also play a prominent role in prevention of CD8 T cell entry into the tumor.

TGFβ1 isoform specificity of SRK-181 results in improved preclinical toxicity profile in comparison to pan-TGFβ inhibition

Having established that selective inhibition of TGFβ1 activation is necessary and sufficient to overcome resistance to PD-1 blockade in multiple tumor models, we next investigated whether SRK-181 treatment would mitigate cardiac toxicity observed with pan-TGFβ inhibition. Female Sprague-Dawley rats were administered four weekly intravenous doses of SRK-181 at 10, 30, or 100 mg/kg and evaluated in a full toxicology analysis. A high-affinity pan-TGFβ anti-growth factor antibody and an ALK5/TGFβRI kinase inhibitor were also included in this study to recapitulate the previously reported observations of treatment-related heart valvulopathy (17, 18).

Consistent with previously reported studies, within 1 week, animals treated daily with the small-molecule ALK5/TGFβRI kinase inhibitor or a single dose of the pan-TGFβ antibody developed heart valvulopathies characterized by heart valve thickening due to hemorrhage, endothelial hyperplasia, mixed inflammatory cell infiltrate, and/or stromal hyperplasia (Fig. 6A). In contrast, no test article–related cardiovascular lesions were noted after SRK-181 treatment at any tested dose (Fig. 6B). In addition, all SRK-181–treated animals underwent gross necropsy assessment, organ weight measurements, clinical pathology assessments, and tissue collection for histopathological evaluation. Full histopathological analysis of all tissues in SRK-181–treated rats did not reveal any treatment-related histopathology, including no evidence for immune activation in any tissue beds. Thus, the no observed adverse effect level (NOAEL) of SRK-181 in this study was the weekly dose (100 mg/kg), the highest dose tested (data file S1). These results suggest that selective inhibition of TGFβ1 activation by SRK-181 appears to be well tolerated and avoids the key dose-limiting toxicity observed with pan-TGFβ inhibition at doses well above those required for therapeutic effect observed in combination with CBT.

Fig. 6 Lack of cardiovascular findings in rats treated with SRK-181.

(A) Hematoxylin and eosin histologic staining of rat heart tissue sections. Left panel: Control rat with normal heart valves. Middle panel: Heart valve displaying hemorrhage and inflammatory cells in rat treated with ALK5 inhibitor LY2109761 (300 mg/kg administered daily). Right panel: Heart valves displaying hemorrhage and endothelial hyperplasia in rat treated with a single intravenous dose of pan-TGFβ antibody (30 mg/kg). (B) Summary of incidence and severity of cardiac lesions observed across different treatment groups (n = 5 per group). Animals were administered four weekly intravenous doses of SRK-181 on days 1, 8, 15, and 22 and euthanized on day 29 for histopathology evaluation. Animals receiving pan-TGFβ inhibitors were dosed daily (LY2109761) or once on day 1 (pan-TGFβ antibody) and examined after 1 week of exposure. The control group received four weekly doses of buffer; po, oral gavage; iv, intravenous; qwk, weekly dosing; qd, daily dosing; Ab, antibody.

DISCUSSION

TGFβ pathway activation recently emerged as a potential contributor to primary CBT resistance (5, 6). In this clinical scenario, CD8 T cells are believed to be excluded from entry into tumors via up-regulation of the TGFβ pathway (6, 41), suggesting that tumors co-opt the immunomodulatory functions of TGFβ signaling to generate an immunosuppressive microenvironment. These insights from retrospective clinical tumor sample analyses and preclinical studies provided the rationale for further investigating the role of TGFβ signaling in primary resistance to CBT.

Several key conclusions can be drawn from the results of our studies. First, analysis of gene expression data from TCGA points to TGFβ1 as the most prevalent isoform in many human tumor types and identifies an association between TGFB1 expression and TGFβ pathway activity as measured by multiple gene signatures. These correlative analyses implicate TGFβ1 expression and activation as the likely driver of immune exclusion that is associated with primary resistance to CBT in human tumors.

Second, the identification and testing of a highly selective inhibitor of TGFβ1 activation support the hypothesis that selective inhibition of TGFβ1 will be sufficient to overcome primary CBT resistance. By targeting the latent TGFβ1 prodomain, SRK-181 achieves exquisite isoform selectivity, inhibiting latent TGFβ1 activation in all known molecular contexts without binding to latent TGFβ2, latent TGFβ3, or any of the three active TGFβ growth factors. Pharmacological evaluation of SRK-181-mIgG1 in syngeneic tumor models that were selected to reflect features of primary CBT resistance in human tumors demonstrated that treatment with this antibody rendered such tumors sensitive to anti–PD-1 treatment. Anti–PD-1/SRK-181-mIgG1 combination treatment resulted in profound and durable antitumor responses in models of urothelial cancer, melanoma, and breast cancer, and promoted the establishment of immunological memory in a model of tumor rechallenge. It is well established that the induction of durable T cell antitumor memory is a function of checkpoint blockade by anti–PD-(L)1 (42, 43). Our observations therefore confirm that TGFβ1 activation blockade allows for the establishment of expected anti–PD-1 effects on antitumor immune activity and memory. The anti–PD-1/SRK-181-mIgG1 combination demonstrated antitumor activity in the EMT-6 breast cancer model, in which TGFβ1 was not the sole isoform present in the tumor. The antitumor response observed with anti–PD-1/SRK-181-mIgG1 combination treatment in the EMT-6 model was consistent with the previously reported antitumor activity of an anti–PD-L1 combination with 1D11, a pan-TGFβ antibody; this combination activity was also clearly shown to be dependent on the presence of T cells (6). This suggests that TGFβ1 is likely positioned to play an immunomodulatory role within the tumor microenvironment, whereas other isoforms may not be relevant to this biology. This observation suggested that selective TGFβ1 inhibition may have therapeutic potential in overcoming primary CBT resistance across a broad spectrum of cancers, irrespective of the expression of other TGFβ isoforms.

Mechanistic analysis of the MBT-2 bladder cancer model demonstrated that simultaneous blockade of PD-1 and TGFβ1 activity induced a profound change in the intratumoral immune contexture, most notably a 10-fold increase in CD8 T cells. These CD8 T cells were likely mediators of the observed tumor cell killing because the key cytolytic genes perforin and granzyme B were also up-regulated in tumors by the combination treatment. We note that anti–PD-1/SRK-181-mIgG1 combination treatment induced an enrichment of Treg cells, a somewhat unexpected finding given the importance of TGFβ signaling for peripheral Treg cell maintenance (44). We interpret this increase as an indicator of a strong immune response because Treg cells are recruited to sites of inflammation. The CD8 T cells adopted an activated phenotype, eliciting a strong antitumor response despite this increase in Treg cell numbers. A potential explanation for why this increase in Treg cells does not compromise the antitumor response could be that Treg cells do not substantially contribute to the immunosuppressive MBT-2 tumor microenvironment and that other suppressive immune cell populations, for example, myeloid cells, play a more important role. Alternatively, and not mutually exclusive, because TGFβ1 is a key mediator of Treg cell–driven immunosuppression, SRK-181-mIgG1 treatment may also abrogate Treg cell activity despite the increase in Treg cell numbers.

In the EMT-6 breast cancer model, Mariathasan et al. found that CD8 T cells appeared to build up along the collagenous matrix and fibroblast-rich layer at the tumor’s leading edge, implying that this matrix contributed to T cell exclusion (6). In contrast, we did not consistently detect a dense fibroblastic area near the leading edge of MBT-2 tumors. An unexpected finding, however, was the histological observation of a close proximity of CD8+ T cells to CD31+ tumor vasculature. This pattern of enrichment supports the hypothesis that the tumor vasculature may be a key route of T cell entry into the tumor. In this paradigm, T cells would then migrate and expand outward radially into the tumor after extravasation from tumor blood vessels. Thus, in addition to the CAF layer, immune exclusion may also be imposed by TGFβ1-responsive vascular endothelial cells or by other cells in proximity to the endothelium. Tumor-associated macrophages have also been described to localize near tumor vasculature (45) and may be involved in generating an immune-excluded tumor microenvironment, either by activating TGFβ1 to affect tumor vascular endothelial cells or by producing other immunosuppressive factors at or near the vasculature. It is likely that T cells enter the tumor via multiple routes, and a better understanding of these mechanisms will aid in identifying additional tumor resistance mechanisms and potential therapeutic targets.

In addition to the impact on cytotoxic T cells within tumors, SRK-181-mIgG1/anti–PD-1 combination treatment also beneficially modulated the immunosuppressive myeloid compartment. CD11b+ myeloid cells comprised nearly 80% of the immune infiltrate in untreated MBT-2 tumors, but their representation was reduced to less than 40% after combination treatment. This was driven by a loss of M2-like immunosuppressive macrophages and an even more profound reduction in MDSCs, whereas the proinflammatory M1-like macrophage population remained unchanged. The observed reduction in immunosuppressive myeloid cells was not only relative to the overall immune infiltrate but also an absolute drop in the numbers of these cells within the tumor tissue. The mechanism driving this immunosuppressive myeloid cell reduction is unclear, as is the role of TGFβ1 signaling in the recruitment, development, or maintenance of these cells in the tumor microenvironment. It is known that TGFβ signaling polarizes macrophages into an M2-like phenotype (46). In addition, evidence suggests that TGFβ is partially responsible for MDSC development and acquisition of suppressive functions (47, 48). The influx of IFN-γ–secreting T cells, paired with the altered tumor microenvironment, may contribute to repolarization and decreased trafficking of immunosuppressive myeloid cells. Regardless of the underlying mechanism, reduction in intratumoral immunosuppressive myeloid cells would be highly desirable as part of a tumor immunotherapy approach because this cell population correlates with poor prognosis and checkpoint blockade therapy resistance in patients (49). Therefore, a therapeutic strategy that targets these important immunosuppressive cell types may improve upon targeting a single immunosuppressive cell type (for example, only Treg cells) in the tumor microenvironment.

TGFβ1 is likely expressed by multiple cell types within the tumor microenvironment, including Treg cells, suppressive myeloid cells, fibroblasts, and tumor cells. Each of these produces TGFβ1 in different LLCs: Activated Treg cells express GARP LLCs on their surface (11, 12), suppressive myeloid cells express LRRC33 LLCs on their surface (9), and fibroblasts likely express and deposit LTBP LLCs into the surrounding extracellular matrix (10). Each source of “activatable” TGFβ1 could play a role in promoting immune exclusion and immune suppression in the tumor, and the relative contributions of each could vary across different tumor types. mRNA profiling of the MBT-2 and EMT-6 tumors used in these studies indicated that LRRC33 and LTBP1 are the most highly expressed LLC components, whereas the Cloudman S91 tumors appear to exclusively express LRRC33. In addition, we have shown here that M2-like macrophages in MBT-2 tumors express TGFβ1 LAP on their surface, most likely as part of an LRRC33-TGFβ1 LLC. Direct confirmation of TGFβ1 expression establishes this M2-like macrophage population as a potentially critical source of TGFβ1 activity within the tumor microenvironment. Expression and activation of latent TGFβ1 on the surface of immunosuppressive myeloid cells require LLC formation with the transmembrane protein LRRC33 (9). SRK-181 directly binds to and inhibits the activation of LRRC33-TGFβ1 LLC and can therefore inhibit the activation of latent TGFβ1 provided by immunosuppressive myeloid cells. Hence, selective inhibition of specific LLCs [for example, the GARP-TGFβ1 LLC on Treg cells (40)] may be necessary but not sufficient for enabling the same profound antitumor response as observed with SRK-181-mIgG1/anti–PD-1 combination treatment.

The third key conclusion that we can draw from our studies is that selective inhibition of TGFβ1-driven pathway activity results in improved preclinical safety compared with broad inhibition of all isoforms. Pleiotropic effects associated with broad TGFβ pathway inhibition have hindered therapeutic development of TGFβ pathway inhibitors. Most experimental therapeutics to date (for example, inhibitors of the ALK5/TGFβRI kinase or pan-TGFβ antibodies) lack TGFβ isoform selectivity and therefore broadly block all TGFβ signaling, likely contributing to the dose-limiting toxicities observed in nonclinical and clinical studies (17, 18, 20, 50, 51). Genetic data from both knockout mice and human loss-of-function mutations suggest that these toxicities, particularly the effects on cardiac vasculature, may be due to TGFβ2 and/or TGFβ3 inhibition (30, 52, 53). Here, we demonstrate that SRK-181 has an improved safety profile in a 4-week repeat-dose rat toxicology study, avoiding the pleiotropic effects associated with pan-TGFβ inhibition, with NOAEL at the highest dose tested, which is well above doses required to elicit robust antitumor responses when combined with PD-1 blockade.

This study has two important limitations. First, the efficacy and safety profiles of SRK-181 presented here were determined in rodents. Human tumors are considerably complex and heterogeneous. Human clinical trials will be required to assess the safety of SRK-181 and the efficacy of SRK-181 in combination with anti–PD-(L)1 therapy in the treatment of solid tumors exhibiting primary resistance to anti-PD-(L)1 therapy. Second, the exact cell types responding to TGFβ1 signaling within the tumor have not been defined. It will be of high interest to investigate the link between TGFβ1 signaling and intratumoral immunosuppressive myeloid cells and to better understand how tumor vasculature may be involved in generating TGFβ signaling–dependent immune exclusion.

Given the widely acknowledged importance of TGFβ pathway activation in several disease processes, various approaches have been taken in an attempt to more selectively target the disease-relevant TGFβ pathway activation and avoid the previously mentioned pleiotropic side effects of broad TGFβ pathway inhibition. These include the identification of increasingly isoform-selective anti-TGFβ growth factor antibodies, discovery of an antibody that selectively targets the Treg cell–associated GARP-TGFβ LLC, engineering of TGFβ ligand traps derived from TGFβ receptors, and development of drugs that target TGFβ-activating integrins (40, 5459). Biologics targeting the active TGFβ1 growth factor directly, including antibodies and ligand traps, need to be able to outcompete the local and high-affinity interaction between the TGFβ1 growth factor and the heteromeric TGFβ receptor complex (55), whereas targeting the latent TGFβ1 complex with SRK-181 prevents the release of the growth factor from the prodomain. The noncompetitive mechanism of SRK-181–mediated inhibition may prove advantageous for potent and consistent in vivo TGFβ1 pathway blockade.

Together, the results presented herein demonstrate that highly selective inhibition of TGFβ1 activation overcomes a key mechanism of primary resistance to CBT that has been observed in the clinical setting and recapitulated in preclinical models. Combination treatment with PD-1 blockade and SRK-181-mIgG1 demonstrated marked antitumor responses in syngeneic tumor models that were otherwise refractory to anti–PD(L)-1 antibody treatment. The exquisite TGFβ1 selectivity of this antibody circumvents previously recognized toxicities of less selective TGFβ inhibitors that have limited their clinical utility. Our results, along with the clinical observation of TGFβ association with primary CBT resistance, suggest that treatment with SRK-181 may meaningfully expand the number of patients who could benefit from CBT and provide a strong rationale for clinical testing of this approach.

MATERIALS AND METHODS

Study design

The work presented here was designed to characterize SRK-181, a fully human antibody of the human IgG4/kappa subtype that selectively inhibits latent TGFβ1 activation. We first investigated the binding affinity and TGFβ1 isoform selectivity of SRK-181 and its inhibitory activity in assays using transfected cell lines and primary human Treg cells. For in vitro assays, a minimum of three independent experiments were run with technical duplicates or triplicates, unless otherwise noted. The binding site of SRK-181 on latent TGFβ1 was determined using H/DX MS. SRK-181 was investigated in three different syngeneic mouse tumor models, which recapitulate key features found in human tumors with primary CBT resistance, including lack of response to anti–PD-1, evidence for immune exclusion, expression of TGFβ1, and TGFβ pathway activation. We investigated both the impact of SRK-181-mIgG1 in combination with anti–PD-1 on tumor control and animal survival, as well as changes to tumor immune cell infiltrates by flow cytometry and IHC. Appropriate group sizes were determined on the basis of our previous experience with the models and in consultation with the study director. Animals euthanized due to tumor necrosis, ulcerations, or for poor health as per veterinarian discretion were excluded from survival analysis but identified by dotted growth curves on tumor growth graphs and in table S2. Investigators were not blinded during the studies. Last, the preclinical safety profile of SRK-181 was investigated in a 4-week rat toxicology study and compared with less selective pan-TGFβ inhibitors.

Syngeneic mouse models

Mouse studies were run at Charles River Discovery Services North Carolina, an AAALAC (Association of Assessment and Accreditation of Laboratory Animal Care International)–accredited facility, in compliance with the Guide for Care and Use of Laboratory Animals and Institutional Animal Care and Use Committee (IACUC) guidelines. For mouse in vivo experiments, data are representative of at least two independent studies with 5 to 15 animals per group as noted. Mice bearing established tumors were randomized into treatment groups on day 1 based on tumor size such that the mean tumor volume and ranges of tumor volumes were matched across groups. Information on tumor responses to anti–PD-1 in different mouse syngeneic tumor models was from Charles River (www.criver.com/products-services/discovery-services/vivo-pharmacology/oncology-pharmacology-models/syngeneic-models). For the MBT-2 model, 8- to 12-week-old C3H/HeN (Charles River) female mice were anesthetized with isoflurane to implant 5 × 105 MBT-2 tumor cells subcutaneously in the flank. Animals were distributed into groups of 15 for efficacy studies and 10 for the mechanistic study, with group mean tumor volumes of 56 to 57 mm3 such that all groups had matched starting volume means and ranges. For Cloudman S91, 8- to 12-week-old DBA/2 (Charles River) female mice were anesthetized with isoflurane to implant 5 × 105 Cloudman S91 tumor cells in 50% Matrigel subcutaneously in the flank. Animals were distributed into groups of 12 when group mean tumor volumes reached 126 to 132 mm3 such that all groups had matched starting volume means and ranges. For the EMT-6 model, 8- to 12-week-old female BALB/c mice (Charles River) were implanted with 5 × 106 EMT-6 tumor cells subcutaneously in the flank. Animals were distributed into groups of 10 with a group mean starting volume of 39 to 41 mm3, such that all groups had matched starting volume mean and range. All animals were treated with two antibodies, either anti–PD-1 or control rat IgG1 plus either SRK-181-mIgG1 or its isotype control antibody. Control rat IgG1 (rIgG1) or anti–PD-1 (RMP1-14; BioXCell) was dosed at 10 mg/kg twice a week. SRK-181-mIgG1 or control mIgG1 was dosed at the indicated doses once a week. When multiple SRK-181-mIgG1 doses were tested, its isotype control was administered at the highest dose. The pan-TGFβ antibody 1D11 (BioXCell) was dosed at 5 mg/kg twice weekly. In mechanistic studies, SRK-181-mIgG1 or its isotype control was dosed on days 1 and 8, and anti–PD-1 or its isotype control was dosed on days 1, 4, and 8. All antibodies were dosed intraperitoneally in phosphate-buffered saline (PBS). Tumor size was measured twice a week by caliper, and animals were euthanized by CO2 asphyxiation when tumors reached 1200 mm3 (MBT-2) or 2000 mm3 (Cloudman S91, EMT-6) or upon ulceration. Tumor volumes were calculated asvolume=w2×l2where w is the tumor width and l is the tumor length in millimeters. Tumors that reached a volume of 25% or less of the study endpoint (300 mm3 for MBT-2 and 500 mm3 for Cloudman S91 and EMT-6) were classified as responses. Complete responses were classified as a tumor with a volume of less than 13.5 mm3 for three or more consecutive measurements. Tumor-free survivors had no palpable tumors at study end. Animals euthanized due to tumor necrosis, ulcerations, or poor health as per veterinarian discretion were removed from subsequent survival analyses but are indicated by dotted tumor growth curves in the figures and identified in table S2.

Nonclinical toxicology study

The rat toxicology study was run at Covance Laboratories, Greenfield, Indiana, an AAALAC-accredited facility, in compliance with the Guide for Care and Use of Laboratory Animals. Seven- to 8-week-old female Sprague-Dawley rats (Charles River Laboratories) were group housed in polycarbonate cages containing appropriate bedding and water valves in a controlled environment (20° to 26°C; 30 to 70% relative humidity; 12-hour light and 12-hour dark cycles) and were offered Certified Rodent Chow (Envigo) and tap water ad libitum. Rats were randomly assigned to treatment groups (n = 5 animals per group). The study included a vehicle-treated control group, positive control groups treated with either the small-molecule ALK5 inhibitor LY2109761 (MedChem Express, at 200 and 300 mg/kg) or the high-affinity pan-TGFβ antibody 12.7 (60) (at 3, 10, or 30 mg/kg), and SRK-181–treated groups at 10, 30, or 100 mg/kg. LY2109761 was formulated in 1% (w/v) carboxymethylcellulose, 0.25% (v/v) Polysorbate 80, and 0.05% (v/v) Dow Corning Antifoam 1510-US in purified water. LY2109761 was administered once daily by oral gavage, and the pan-TGFβ antibody (in PBS pH 7.4) was administered once via intravenous injection. Animals receiving a control pan-TGFβ inhibitor were euthanized on day 8. SRK-181 was administered intravenously for 4 weekly doses (on days 1, 8, 15, and 22), animals were euthanized on day 29, and serum was collected to confirm exposure. Mean SRK-181 serum concentrations at study termination reached 2.3 mg/ml for the 100-mg/kg dose group, which is equivalent to concentrations that are about 10,000-fold higher than the IC50 values of SRK-181 inhibitory activity, as determined in cell-based assays. General clinical observations of animals were performed twice daily, and cageside observations were conducted after dose to assess acute toxicity. Other observations performed included an assessment of food consumption and measurement of body weight once weekly. These also included clinical pathology (hematology, serum chemistry, and coagulation) and anatomic pathology (gross and microscopic) evaluations. A comprehensive set of tissues were collected at necropsy for microscopic evaluation. Tissues were preserved in 10% neutral buffered formalin, trimmed, processed routinely, and embedded in paraffin. Paraffin blocks were microtomed, and sections were stained with hematoxylin and eosin (H&E). In particular, the heart was trimmed by longitudinally bisecting along a plane perpendicular to the plane of the pulmonary artery to expose the right atrioventricular, left atrioventricular, and aortic valves. Both halves were submitted for embedding. Each heart hemisection was embedded in paraffin with the cut surface down. Blocks were sectioned to obtain at least three heart valves. The tissue sections were examined by light microscopy by a board-certified member of the American College of Veterinary Pathologists. The study summary is available in data file S1 (SRK-181 is referred to as 6993-hIgG4), with additional test article names unrelated to SRK-181 redacted.

Statistical analysis

GraphPad Prism (versions 7 and 8) and R with the gene set variation analysis (GSVA) package were used for statistical analyses. IC50 values were calculated by nonlinear regression analysis (three parameter fit) of dose-response data in Prism. KD values were calculated by nonlinear regression analysis of equilibrium titration data in Prism. Data shown are mean ± SD, unless noted otherwise. Statistical tests and P values, indicated by asterisks, are noted for each figure in the legend. Original data are provided in data file S2.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/12/536/eaay8456/DC1

Materials and Methods

Fig. S1. TGFβ isoform mRNA expression and pathway activation in individual tumors.

Fig. S2. Characterization of recombinant latent TGFβ1 complexes.

Fig. S3. SRK-181 does not bind active TGFβ growth factors.

Fig. S4. Characterization of LLC-presenting molecule expression in LN229 cells.

Fig. S5. SRK-181 inhibits activation of mouse latent TGFβ1 LLC.

Fig. S6. Activation of peripheral human Treg cells induces GARP and TGFβ1 LAP expression on their cell surface.

Fig. S7. SRK-181 Fab binding protects three regions on TGFβ1 SLC from hydrogen/deuterium exchange.

Fig. S8. SRK-181 and integrin αVβ6 can bind to TGFβ1 SLC simultaneously, suggesting an allosteric mechanism of inhibition.

Fig. S9. Selection of syngeneic mouse tumor models that recapitulate profiles from CBT-resistant human tumors.

Fig. S10. Induction of prolonged tumor control with SRK-181-mIgG1/anti–PD-1 combination treatment in multiple tumor models.

Fig. S11. Further characterization of the treatment effect of SRK-181/anti–PD-1 in MBT-2 tumors.

Table S1. Percent amino acid sequence identity across human TGFβ isoforms.

Table S2. Animals euthanized early due to health reasons or found dead.

Table S3. qPCR reagent list.

Data file S1. Nonclinical toxicology study summary.

Data file S2. Original data.

References (6270)

REFERENCES AND NOTES

Acknowledgments: We are grateful to the Charles River Discovery Services for running our syngeneic mouse tumor models. We acknowledge Adimab (Lebanon, NH) for their contributions to the antibody discovery and engineering work. We thank Covance Laboratories for running our rat toxicology study. We thank S. J. Eyles (UMass Institute of Applied Life Sciences Mass Spectrometry Core Facility, Worcester MA) for assistance with H/DX-MS data collection. We acknowledge G. Chang for early contributions to the project and thank A. Donovan and N. Mahanthappa for critically reviewing the manuscript. Funding: This study was funded by Scholar Rock, Inc. Author contributions: C.J.M. designed, supervised, analyzed, and interpreted syngeneic mouse tumor studies; analyzed gene expression data; and conducted Treg assays. A.D. designed, coordinated, and supervised antibody discovery and engineering efforts and interpreted the data. C.L. and C.C. developed and executed cell-based assays. A.K. designed and supervised the rat toxicology study. S.W. coordinated bioinformatics efforts, analyzed and interpreted gene expression data, and supervised cell-based assay efforts and bioanalytics. K.B.D. designed, executed, analyzed, and interpreted the H/DX experiment; purified and characterized recombinant proteins; and performed biochemical competition experiments. C.T.B. developed IHC methods, performed IHC analysis of mouse tissues, and analyzed data. A.N., F.T.D., F.C.S., and C.B. expressed, purified, and characterized recombinant proteins. A.S. and N.D. analyzed mouse tumors and interpreted results. J.W.J. performed biolayer interferometry experiments and interpreted results. S.L. developed equilibrium titration protocols, executed experiments, and analyzed the data. R.R.F. developed pharmacokinetics methods and analyzed rat serum samples. P.R. contributed to bioinformatic analysis and interpretation of tumor gene expression. A.D.C. supervised protein science efforts, developed protein expression and purification strategies, expressed and purified proteins, and interpreted the data. A.B. and G.J.C. oversaw the research, provided strategic guidance, and discussed and interpreted the data. T.S. initiated the research, designed antigens, led overall project, and discussed and interpreted the data. C.J.M., A.B., and T.S. wrote the manuscript with editorial input from A.D., A.K., S.W., K.B.D., and G.J.C. Competing interests: All authors were Scholar Rock employees at the time of the study and stock and/or stock option holders, except for P.R. P.R. was a consultant to Scholar Rock and received consulting fees and travel reimbursement from Scholar Rock. A.B. is the chief scientific officer of Scholar Rock, Inc. C.J.M., A.D., C.L., C.C., S.W., K.B.D., J.W.J., S.L., A.D.C., A.B., G.J.C., and T.S. are named inventors on an international patent application related to this work (WO/2020/014460; High-Affinity, Isoform-Selective TGFβ1 Inhibitors and Use Thereof) filed by Scholar Rock, Inc. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. Sequences of SRK-181 have been published in the above-referenced patent application. The GSVA code used for data analysis is available on Zenodo (61).
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article