Modulation of fibronectin extracellular matrix enhances anti-tumor efficacy of immune checkpoint blockade

Summary

The success of immune checkpoint inhibitors is limited by multiple factors, including poor T cell infiltration and function within tumors, partly due to a dense extracellular matrix (ECM). Here, we investigate modulating the ECM by targeting integrin α5β1, a major fibronectin-binding and organizing integrin, to improve immunotherapy outcomes. Use of a function-blocking murinized α5β1 antibody reduces fibronectin fibril formation, enhances CD8⁺ T cell transendothelial migration, increases vascular permeability, and decreases vessel-associated collagen. These changes culminate in improving the effectiveness of PD-L1 blockade, alone or with chemotherapy, in the E0771 breast cancer model. Clinically, high integrin alpha 5 (ITGA5) expression correlates with worse survival in patients treated with atezolizumab as monotherapy or combined with chemotherapy or anti-angiogenic therapies in numerous clinical trials. Overall, our studies suggest that ECM-modulating approaches could be used as a future strategy to increase the proportion of patients who respond to immune checkpoint inhibition and other immunotherapies.

Graphical abstract

Highlights

• •Blocking mature fibronectin increases CD8⁺ T cell trafficking across the endothelium
• •α5β1 antibody plus PD-L1 blockade improves efficacy in preclinical breast cancer
• •α5β1 and PD-L1 blockade reduces terminally exhausted CD8⁺ T cells in tumors
• •High ITGA5 expression associates with poorer survival in several immunotherapy clinical trials

Khan et al. demonstrate that mature fibronectin extracellular matrix (ECM) disruption, via α5β1 integrin targeting, can enhance immune checkpoint therapy preclinically. They also show that ECM modulation can enhance CD8⁺ T cell adhesion to the endothelium and transmigration. Finally, they find that high ITGA5 gene expression correlates with poor survival, more pronounced in immunotherapy-treated patients.

Introduction

A major benefit of immune checkpoint blockade in oncology includes the remarkable durable and major tumor remissions induced in some patients in a wide spectrum of different tumor types.¹ Early trials involving Programmed Cell Death Protein-1 (PD-1) and Programmed death-ligand 1 (PD-L1) antibody inhibition monotherapy in patients with triple-negative breast cancer (TNBC) showed promising response rates, especially in tumors infiltrated with PD-L1⁺ immune cells.^2,3,4,5 However, of these patients, only a small proportion (1%–10%) showed long-lasting responses.⁶ The current approved immunotherapy options for TNBC include pembrolizumab (a PD-1 antibody) in combination with chemotherapy⁷ and atezolizumab (anti-PD-L1) in combination with nab-paclitaxel (nab-PTX) chemotherapy⁸; however, the latter treatment combination has recently been withdrawn from Food and Drug Administration (FDA) accelerated approval but is still in use in Europe.⁶

TNBC tumors can be separated into four main subtypes when describing the tumor immune microenvironment (TIME): (1) fully inflamed (often responding well to PD-1/PD-L1 immune checkpoint blockade), (2) stroma-restricted, (3) margin-restricted, and (4) “immune deserts/cold tumors,” in which tumors are largely devoid of lymphocytes.⁶ The level of tumor-infiltrating lymphocytes (TILs) is a prognostic biomarker of outcome, and it has been suggested that the levels of TILs should also factor into disease staging of early TNBC.⁹ Ineffective T cell migration and penetration into the tumor mass represents an important obstacle limiting the benefit of immune checkpoint therapy outcomes. Thus, not surprisingly, there is currently a substantial research effort aimed at modifying the TIME in various ways to enhance the number of TILs and thus improve responses to immunotherapy, not only in TNBC but also in other solid tumor types.¹⁰

The hypothesis of physical resistance to T cell infiltration and migration, related to the heterogeneity and aberrant organization of the extracellular matrix (ECM), has led to emerging therapeutic strategies of modulating the ECM to enhance tumor immunotherapy outcomes.^11,12 Two recent studies have described that inhibition of collagen crosslinking leads to increased intratumoral T cell infiltration and improved response to PD-1 blockade through two distinct mechanisms.^13,14 The biomechanical characteristics of a tumor or its “mechanopathology” is emerging as a factor that has implications for drug delivery, immunosuppression, and metastasis.¹⁵ The physical properties of the ECM, such as viscoelasticity, have recently been shown to have direct effects on T cell transcriptional programs and effector function.¹⁶ ECM degradation strategies using tumor necrosis factor alpha (TNF-α)-induced release of proteases have been shown to increase tumor perfusion and immune cell infiltration.¹⁷ In addition, the accumulation of fibronectin and dense ECM has been shown to have inhibitory effects on T cell migration and can restrict them from efficiently interacting with tumor cells.¹⁸ Enzymatic digestion of fibronectin in tumors has been shown to increase chemotherapy delivery as well as chimeric antigen receptor (CAR) T cell infiltration.¹⁹ The α5β1 integrin complex is the major complex involved in fibronectin fibrillogenesis as well as constant fibronectin turnover and remodeling.^20,21 The importance of α5β1 and ECM targeting has recently been described in preclinical models of colorectal cancer.²²

Here, we show that antibody targeting of the intact α5β1 heterodimer can lead to disruptions in mature fibronectin fibril formation and the assembly of other ECM components such as collagen I. This ECM disruption correlates with increased transendothelial migration of activated CD8⁺ T cells. We also report that antibodies targeting α5β1 can enhance the efficacy of PD-L1 antibody-mediated immune checkpoint blockade in mouse primary tumor models and synergize when used as a triplet combination with paclitaxel (PTX) chemotherapy. Moreover, based on a retrospective analysis of multiple clinical trial results, we report that higher expression of integrin alpha 5 (ITGA5) correlates with poorer survival in patients treated with atezolizumab immunotherapy given as monotherapy or when combined with chemotherapy or anti-angiogenic drugs in multiple clinical trials of different tumor types.

Results

Blockade of α5β1 leads to disruption of fibronectin fibril formation and increases transendothelial migration of immune cells in vitro

To investigate the expression of multiple integrins across different cell types, we used single-cell RNA sequencing analysis using a breast cancer dataset in which patients were treated with one dose of PD-1 antibody therapy.²³ This revealed high ITGA5 expression in endothelial cells and lower expression in cancer-associated fibroblasts (CAFs), whereas integrin alpha V (ITGAV) was expressed highly in CAFs and lowly in endothelial cells (Figure S1). Integrin beta 1 (ITGB1) was present in many cell types including high expression in endothelial cells (Figure S1). Given the major role of α5β1 in fibronectin organization, and as it is highly expressed in endothelial cells in breast cancer, we sought to modulate its effects on fibronectin. First we determined whether antibody targeting of human α5β1 (using mouse clone 18C12 specific to the human α5β1 heterodimer)²⁴ could inhibit fibronectin fibril formation by treating human umbilical vein endothelial cells (HUVECs) in vitro. To investigate the disruption of fibronectin fibril incorporation into ECM, we performed biochemical assays in which deoxycholate (DOC) was used to solubilize fibronectin fibrils, which have not yet been incorporated into the mature ECM (whereas fibronectin that has been incorporated is DOC insoluble). This revealed a decrease in both soluble and insoluble fibronectin with α5β1 blockade (Figures 1A and S2A). Using immunofluorescence, we observed that treatment with α5β1 blockade reduced the web-like mature fibronectin network on cells and within the ECM deposited on the plastic of the culture dish surface (decellularized) over a 20-h period (Figure 1B). To investigate whether endothelial permeability was affected by α5β1 blockade, we utilized diffusion of fluorescein isothiocyanate (FITC)-dextran across a HUVEC monolayer, which was increased with α5β1 antibody treatment (Figure 1C). TNF-α (100 ng/mL) was used as a positive control as it is a known vascular permeability inducer (Figure 1C). Next, we investigated whether α5β1 blockade could affect adhesion and transmigration of human immune cells to HUVEC monolayers in response to selected chemokines (Figure 1D). When adding activated human CD8⁺ T cells to HUVECs cultured on fibronectin-coated transwells, α5β1 blockade enhanced adhesion and transendothelial migration. Addition of α5β1 blockade to CXCL12, a chemokine which is a known mediator of T cell chemoattraction,^25,26 or TNF-α stimulation further enhanced transendothelial migration (Figures 1E and 1F). This revealed that α5β1 antibody blockade can increase HUVEC permeability and transmigration of T cells.

Figure 1.

Fibronectin/ECM disruption using α5β1 antibodies increases CD8⁺ T cell adhesion and transendothelial migration

(A) Western blot of DOC-soluble or insoluble fractions of HUVEC lysates blotted for fibronectin (anti-FN), treated with isotype control or α5β1 antibodies.

(B) Immunofluorescence microscopy of HUVECs cultured on fibronectin-coated plates treated with isotype control or α5β1 antibodies and stained with fibronectin antibodies, with cells intact or cells removed (decellularized), revealing disruptions in mature fibronectin matrix formation after 20 h of α5β1 antibody treatment. Scale bars represent 50 μm.

(C) HUVEC permeability increased with addition of α5β1 antibodies, TNF-α (100 ng/mL) as positive control, unpaired t test of α5β1 10 μg/mL conditions in comparison to control treatment. ∗∗∗p < 0.0001 n = 6 biological replicates per group, data point represents mean, error bars represent standard error of the mean (SEM).

(D) Schematic diagram of endothelial-immune cell adhesion and transmigration assays (made using BioRender).

(E) CD8⁺ T cells stimulated with CD3e antibodies and interleukin-2 (IL-2) adhere more strongly with α5β1 antibody treatment, when HUVECs are untreated or treated with CXCL12, TNF-α, or both CXCL12 and TNF-α. Unpaired t test, ∗∗∗p < 0.0001, ∗∗p = 0.0060, n = 12 biological replicates, error bars represent SEM.

(F) CD8⁺ T cells stimulated with CD3e antibodies and IL-2 undergo enhanced endothelial transmigration with α5β1 antibody treatment, when HUVEC are untreated or treated with CXCL12, TNF-α, or both CXCL12 and TNF-α. Unpaired t test, ∗∗∗p < 0.0001, n = 12, error bars represent SEM.

Blocking α5β1 increases vessel permeability and disrupts fibronectin fibril formation in vivo

With the goal of modulating the fibronectin ECM in preclinical mouse models, we utilized a monoclonal antibody (hamster clone 10E7) that recognizes the intact mouse integrin α5β1 heterodimer.²⁴ This hamster 10E7 monoclonal antibody was murinized by cloning variable heavy and variable light chains onto a mouse IgG2a backbone to avoid xenogeneic reactions when used in mice. To test the ability of this α5β1 antibody to inhibit fibronectin formation, we utilized mouse lung endothelial cells cultured with an isotype control antibody (against gp120) or a murinized α5β1 antibody 10E7. The results revealed striking reductions in fibronectin fibrils under treatment of α5β1 antibody after 20 h (Figures 2A and 2B). To determine whether the α5β1 antibody could induce vascular permeability in a tumor model, mice bearing orthotopic 66cl4 breast tumors were treated with the α5β1 antibody or control antibody for 2 weeks and then injected intravenously with a near-infrared fluorescent dye to measure vascular permeability 24 h after. This revealed higher accumulation of the dye in tumors of mice treated with α5β1-blocking antibodies (Figures 2C and 2D). Next, we performed DOC-soluble and insoluble fraction analysis of fibronectin by western blotting. This allowed detection of a ratio of insoluble/soluble fibronectin or mature/immature fibronectin. When mice bearing 66cl4 breast tumors were treated with the α5β1 antibody, this significantly reduced the insoluble/soluble fibronectin ratios within tumors (Figures 2E, 2F, S2B, and S2C). As fibronectin is an integral component of the ECM that binds to collagen and is important in collagen fibrillogenesis,^27,28,29 we investigated whether other ECM components were affected by α5β1 and fibronectin disruption. Quantitative analysis of type I collagen immunofluorescence staining showed a significant α5β1 antibody-dependent decrease in vessel-associated collagen I in tumors (Figures 2G and 2H). Taken together, these findings revealed that α5β1 antibody blockade not only reduces the formation of mature fibronectin fibrils but also disrupts the assembly of other important components of the ECM, leading to increased vascular permeability.

Figure 2.

α5β1 antibody targeting reduces fibronectin and collagen deposition and increases intratumoral vascular permeability in mice

(A) Immunofluorescence microscopy of mouse lung endothelial cells stained with fibronectin antibodies after treatment in vitro with α5β1 antibody shows reduced formation of fibronectin fibrils.

(B) Quantification of percentage fibronectin per region of interest (ROI), n = 12, unpaired t test, ∗∗∗p < 0.0001, error bars represent SEM.

(C) α5β1 antibody increased vascular permeability in mice bearing 66cl4 breast tumors, visualized by injection of near-infrared AngioSense dye after 24 h. Scale bars, 5 mm.

(D) Quantification of fluorescence intensity after 24-h time period, n = 9–10 mice per group, unpaired t test, ∗∗∗p = 0.0001, error bars represent SEM.

(E) Representative western blot of tumor lysates from mice treated with IgG isotype antibody or α5β1 antibody bearing 66cl4 breast tumors; fibronectin was detected in deoxycholate (DOC)-soluble and insoluble fractions, normalized to β-actin and a ratio determined.

(F) Statistical analysis of fibronectin insoluble/soluble ratios, n = 6 lysates of tumors from individual mice per group, unpaired t test, ∗p = 0.0144, n = 6 per group, error bars represent SEM.

(G) Immunofluorescence microscopy of breast tumors demonstrates reductions in vessel (endomucin⁺)-associated collagen I with α5β1 antibody treatment. Scale bars represent 50 μm.

(H) Quantification of vessel-associated collagen I H-score, unpaired t test, n = 9 per group, ∗p = 0.0296, error bars represent SEM.

Combination of α5β1 and PD-L1 antibody blockade leads to increased survival in preclinical breast cancer model

Analyses from The Cancer Genome Atlas indicate that fibronectin (FN1) is highly expressed in breast cancer in comparison to normal breast tissue, and the same applies for the TNBC subtype (Figure S3). Since we observed increases in lymphocyte transmigration with α5β1 blockade, we sought to determine whether the combination of PD-L1 and α5β1 antibodies could increase anti-tumor efficacy in preclinical models of TNBC using the orthotopic syngeneic E0771 mouse model. The murinized version of atezolizumab (clone 6E11) was used to block PD-L1, enabling continuous antibody administration in mouse models without triggering hypersensitivity or xenogeneic reactions.^30,31,32 PD-L1 antibody monotherapy did not result in impressive tumor growth inhibition nor did it significantly increase survival resulting in only 1/20 complete regressions (CRs) of tumors (Figure 3A). Monotherapy of α5β1 blockade resulted in 3/20 CRs and causes a significant effect on survival in comparison to IgG control (Figures 3A–3C). However, the combination of α5β1 and PD-L1 antibodies resulted in 8/21 CRs and significantly increased survival when compared against isotype control antibody and against monotherapy groups (Figures 3A–3C). The combination therapy resulted in more than double the number of expected CRs when considering the additive effects of each monotherapy group, suggesting drug synergy. To test for drug synergy, we employed two synergy models to calculate an expected hazard ratio of the combination group (based on observed hazard ratios for PD-L1 and α5β1 monotherapy) (see STAR Methods). Both the Bliss independence and Loewe additivity tests revealed drug synergy with combination therapy (Figure S4A). To determine whether immune memory was induced in mice displaying CRs, we rechallenged mice from the α5β1 and PD-L1 antibody group with E0771 cells in the contralateral mammary fat pad. This resulted in no tumor growth in rechallenged mice compared to robust growth observed in naive mice (Figure S4B). In summary, the combination of α5β1 and PD-L1 antibodies results in synergistic effects, provoking immune memory in the E0771 model.

Figure 3.

Combination of α5β1 and PD-L1 antibody treatment in the primary E0771 orthotopic breast cancer model extends median survival

(A) Waterfall plots demonstrating percentage tumor growth from time of treatment initiation. CRs were observed in 0/18 mice for isotype control antibody (black), 3/20 mice for α5β1 antibody (blue), 1/20 mice for PD-L1 antibody (green), and 8/21 mice for PD-L1 + α5β1 antibody (red).

(B) Survival analyses were generated from mice that reached primary tumor endpoint from (A). The same isotype control antibody group is included in each for comparison.

(C) Table displaying hazard ratios (Cox proportional hazards model) and p values from log rank (Mantel-Cox) analyses of survival.

α5β1 antibody blockade affects CD8⁺ T cells and their exhaustion status

To gain an understanding of the TIME within mouse E0771 breast tumors under each treatment group, we performed a shorter treatment and characterized the tumor microenvironment (Figure 4A). After 8 days and 3 doses of antibodies, tumors were harvested, dissociated, and analyzed by flow cytometry. In the PD-L1 + α5β1 antibody combination treatment group, 2/12 CRs and 1/8 CRs for PD-L1 antibody monotherapy were observed, which could possibly lead to skewed analysis of the TIME to tumors that are less responsive to treatment. Nevertheless, flow cytometry revealed trends of increased CD3e⁺ T cells in the combination treatment group (Figure 4C). There was a significant increase in terminally exhausted (PD-1⁺ TIM3⁺) CD8⁺ T cells under PD-L1 antibody monotherapy, but this increase was lost with the addition of the α5β1 antibody (Figures 4B and 4D). Furthermore, when comparing PD-L1 antibody monotherapy with the combination treatments, there was a significant decrease in the ratio of terminally exhausted CD8 T cells (PD-1⁺ TIM3⁺) to non-exhausted T cells (PD-1⁺ TIM3⁻, PD-1⁻ TIM3⁻) (Figure 4E). We also determined whether there were changes in PD-1⁺ CD4⁺ T cells, due to their potential in affecting poor prognosis.^33,34 This revealed trends in increased CD4⁺ PD-1⁺ T cells with PD-L1 antibody therapy (p = 0.677), but this was reduced to control levels with the addition of α5β1 blockade (p = 0.0264) (Figure S5).

Figure 4.

Analysis of the E0771 primary tumor microenvironment exposed to short-term therapy

(A) Individual growth plots showing tumor volumes of E0771 treated with isotype control antibody, α5β1, PD-L1 antibody, or the α5β1 + PD-L1 antibody combination. CRs were observed in 0/8 for IgG isotype, 0/8 for α5β1 antibody, 1/8 for PD-L1 antibody, and 2/12 for PD-L1 + α5β1 antibody. Arrows represent antibody administration.

(B) Contour flow cytometry plots of CD8⁺ T cells stained with PD-1 and TIM-3, revealing PD-1⁺ TIM-3⁺ terminally exhausted CD8⁺ T cells.

(C–E) Flow cytometry analysis of (C) pan-T cells (CD3⁺) as a percentage of live cells, (D) terminally (Term) exhausted CD8 T cells (PD-1⁺ TIM3⁺) as a percentage of live cells, n = 7–10, unpaired t test, ∗p = 0.0217, and (E) ratio of terminally exhausted CD8 T cells (PD-1⁺ TIM3⁺) to non-exhausted CD8 T cells (PD-1⁺ TIM3⁻, PD-1⁻ TIM3⁻), unpaired t test, ∗p = 0.0185, error bars represent SEM.

Combination of PD-L1 and α5β1 antibody blockade with PTX increases the survival of mice with advanced primary tumors

When treating mice bearing E0771 tumors as described in Figure 3, we observed that the α5β1 and PD-L1 antibody combination had the greatest efficacy when therapy was initiated in mice bearing primary tumors of 90 mm³ or smaller. Therefore, we separated the data from Figure 3 into those with tumors less than 90 mm³ tumor volume at treatment initiation, which revealed survival advantages in all groups, with the greatest advantage observed in the combination group (Figure 5A). Additionally, in this analysis, 8/12 complete responses were observed in the combination group, and 2 tumors that initially responded acquired resistance to therapy (Figure S6A). In contrast, tumors larger than 90 mm³ at the start of therapy were not responsive to the combination of α5β1 and PD-L1 antibody blockade (Figures 5B and S6B). This was reflected in hazard ratios and survival analyses (Figure 5C). Chemotherapy, including PTX, has been used clinically as a combination agent in TNBC immunotherapy trials, resulting in improved outcomes, and is used in neoadjuvant treatment settings to “debulk” primary tumors. Thus, we tested whether it could increase efficacy outcomes of the combination antibody treatment when used to treat these larger tumors. First, to investigate whether α5β1 antibody targeting can enhance the efficacy and potentially increase the delivery of PTX chemotherapy, mice bearing E0771 tumors over 90 mm³ were treated with isotype control IgG, PTX alone, or α5β1 antibody plus PTX (Figure S6C). This revealed a marginal, albeit statistically significant, increase in survival with the addition of α5β1 antibody to PTX chemotherapy but, notably, did not lead to any CRs (Figure S6D). Subsequently, we examined the effect of the α5β1 + PD-L1 antibody combination with PTX in more advanced E0771 tumors (i.e., larger than 90 mm³ at the start of treatment). Interestingly, the combination of α5β1 + PD-L1 antibody + PTX outperformed both the PD-L1 + PTX treatment and the α5β1 + PD-L1 antibody groups and resulted in 7/15 CRs of these larger tumors (Figure 5D). The triplet combination also improved overall survival (OS) in comparison to all other groups (Figures 5E and 5F). This presents us with potentially translationally relevant data as PD-1/PD-L1 antibody blockade is currently approved in patients with TNBC in combination with chemotherapy, including PTX,⁷ and thus the addition of α5β1 blockade could conceivably further enhance efficacy outcomes.

Figure 5.

Combination α5β1 and PD-L1 antibody blockade with PTX chemotherapy increases survival in mice bearing E0771 tumors that are less responsive to α5β1 and PD-L1 antibody treatment alone

(A) Mice bearing E0771 tumors less than 90 mm³ at the start of treatment displayed increased survival especially in the PD-L1 and α5β1 antibody combination treatment group (red).

(B) E0771 tumors larger than 90 mm³ at start of treatment responded poorly to each therapy.

(C) Median OS, hazard ratios (cox proportional hazards model), log rank Mantel-Cox test, p values. Black arrows indicate treatment time points of antibodies, and gray arrows indicate PTX.

(D) Percentage tumor growth inhibition of large (>90 mm³) E0771 tumors treated with PTX + PD-L1 (green), PD-L1 + α5β1 antibody (red), or PD-L1 antibody + α5β1 + PTX (purple).

(E) Survival plots from (D).

(F) Table displaying hazard ratios (Cox proportional hazards model) and p values from log rank (Mantel-Cox) analyses of survival.

Combination of PD-L1 and α5β1 antibody blockade in additional preclinical models

We next expanded the number of breast tumor models treated with the PD-L1 + α5β1 antibody combination strategy. As we had already observed increased tumor vessel permeability and decreased mature fibronectin fibril formation in the 66cl4 breast cancer model in response to α5β1 blockade, we wanted to determine whether combination with PD-L1 antibody would result in improved or synergistic anti-tumor efficacy. Addition of PD-L1 antibody to α5β1 antibody resulted in additive effects, which resulted in significant reductions in tumor growth after 21 days when compared with isotype control antibody treatment (Figure S7A). We next tested the same therapeutic combination strategy in the EMT6 TNBC orthotopic mouse model. The combination of α5β1 and PD-L1 antibody blockade failed to exhibit an additive effect in this model (Figure S7B). In order to determine whether a selected variant EMT6 subline having a higher tumor mutational burden would respond to PD-L1 blockade and the antibody combination, we used the TNBC EMT6-CDDP model.³⁵ EMT6-CDDP responds well to PD-L1 antibody monotherapy but does not induce high numbers of robust CRs.^36,37 The addition of α5β1 antibody to PD-L1 antibody led to a modest additive improvement of anti-tumor efficacy in this case (Figure S7C). In surrogate survival studies of primary tumor treatment, both PD-L1 antibody monotherapy and PD-L1 + α5β1 antibody combination significantly increased survival when compared to control antibody (p values ∗∗0.0047 and ∗∗∗0.0005, respectively); hazard ratios were improved for the combination and resulted in 1/10 CRs (Figure S7D). However, the PD-L1 + α5β1 antibody combination did not improve survival in comparison to PD-L1 antibody monotherapy. In the event that superior PD-L1 antibody monotherapy efficacy was masking the effects of the combination treatment, we tested a fifth model, which has a high tumor mutational burden and moderately responds to PD-L1 antibody monotherapy, namely the CT26 MNNG⁺EMS⁺ subcutaneous colon cancer model. The importance of α5β1 and ECM targeting has recently been described in preclinical models of colorectal carcinoma.²² This subline was previously derived by exposing the CT26 colon tumor cell line in vitro to double mutagen treatment using nitroguanidine (MNNG) plus ethyl methanesulphonate (EMS).³⁷ In this model, the combination of α5β1 antibody and PD-L1 antibody significantly increased survival compared to the PD-L1 or α5β1 antibody monotherapy groups, respectively (Figures S7E and S7F). In summary, it appears that α5β1 antibody addition to PD-L1 antibody can improve efficacy in some additional models tested, especially in those that are more immunogenic. This variability reflects clinical response patterns to most, if not virtually all, cancer drugs or therapies and could be the result of numerous factors.

High ITGA5 gene expression is frequently retrospectively associated with poorer OS in patients treated with atezolizumab (PD-L1 antibody)

As we have shown that α5β1 antibody-mediated reduction in mature FN1 levels leads to increased CD8⁺ T cell transendothelial migration and synergizes with PD-L1 blockade in some tumor models, we sought to determine whether expression of α5β1 could have effects on clinical efficacy outcomes involving immunotherapy, in this case using PD-L1-targeting antibodies. We used gene expression data from pre-treatment biopsies and OS data from multiple clinical trials in which atezolizumab was used as the immunotherapy agent either as monotherapy or in combination with either chemotherapy or an anti-angiogenic vascular endothelial growth factor (VEGF) antibody (bevacizumab). Patients in the atezolizumab-treated cohorts were divided into above or below median expression of the ITGA5 gene (at time of pre-treatment biopsy), and OS was analyzed. In pooled analyses of patients with urothelial cancer (IMvigor210, IMvigor211, and IMvigor010), non-small cell lung cancer (NSCLC) (FIR, POPLAR, BIRCH, OAK, IMpower131, and IMpower150), small-cell lung cancer (SCLC) (IMpower133), and renal cell carcinoma (RCC) (IMmotion150 and IMmotion151), high expression of ITGA5 resulted in significantly poorer OS and increased hazard ratios in treatment groups of atezolizumab monotherapy (Figure 6A) or atezolizumab combination therapy (Figure 6B); however, there was no significant difference in OS between high and low ITGA5 expression in control-treated patients (Figure 6C). In the phase 2 IMvigor210 urothelial carcinoma trial where all patients were treated with atezolizumab, high expression of ITGA5 resulted in significantly poorer OS and increased hazard ratios (Figures 6D and S8A). This was somewhat recapitulated in the phase 3 adjuvant IMvigor010 urothelial carcinoma trial, where high ITGA5 expression in atezolizumab-treated patients resulted in poorer OS (Figures 6D and S8B), whereas this did not reach significance in patients under observational control (p = 0.0924) (Figures 6D and S8C). This was not the case in patients with urothelial cancer that were heavily pre-treated with platinum therapies in the phase 3 IMvigor211 trial (Figures 6D and S9). The phase 2 BIRCH NSCLC trial revealed poorer OS in patient tumors with high ITGA5 expression (Figure 6D). In the phase 3 advanced squamous NSCLC IMpower131 trial, high ITGA5 expression resulted in poorer OS and an increased hazard ratio in patients treated with atezolizumab plus chemotherapy (Figures 6D and S8E), whereas this was not the case in patients treated with chemotherapy alone (Figures 6D and S8F). This differential in OS outcome with respect to ITGA5 expression was not observed in the non-squamous NSCLC IMpower150 (Figures S9C and S9D) and the SCLC IMpower133 trials (Figures 6D, S9E, and S9F). In RCC, while high ITGA5 expression in atezolizumab plus bevacizumab resulted in significant differential OS in the phase 2 IMmotion150 trial in comparison to no significance for atezolizumab alone or sunitinib alone (Figures S9G–S9I), this did not hold true in the phase 3 IMmotion151 trial (Figures S9J and S9K). These analyses provide potential clinical rationale for the modulation of fibronectin through α5β1 in multiple cancer types in combination with atezolizumab therapy; however, this appears to be dependent on tumor type, disease stage, and prior therapy.

Figure 6.

Elevated ITGA5 gene expression is associated with poor OS of patients in groups treated with atezolizumab in multiple phase 2 and phase 3 clinical trials

(A–C) Data from patient tumors were pooled from all trials and separated into above median or below median expression of ITGA5 for patients in groups treated with (A) atezolizumab monotherapy, (B) atezolizumab combination (anti-angiogenic agent or chemotherapy), and (C) no atezolizumab treatment (control chemotherapy, control anti-angiogenic, or observation), which revealed significantly worse median OS in patients with high ITGA5 expression in the context of atezolizumab monotherapy or atezolizumab combination therapy.

(D) Swimmer plots depicting hazard ratios using univariate Cox proportional hazards model and p values using log rank (Mantel-Cox test) of ITGA5 expression for all trials listed. Data represent hazard ratio, and error bars represent 95% confidence intervals of the hazard ratios.

Discussion

The tumor-associated ECM is known to be a contributor to tumor aggressiveness, metastasis, and response to various anti-cancer drugs including immunotherapy.^38,39 Here, we show that blocking α5β1 heterodimers inhibits the formation of mature fibronectin fibrils, leading to ECM disruptions and subsequent increases in CD8⁺ T cell transendothelial migration. When α5β1 antibody blockade is combined with PD-L1 immune checkpoint-targeting antibodies, anti-tumor efficacy and survival can be increased in some cancer models. However, we also report that in some other tumor models, the combination is not sufficient to enhance the efficacy of PD-L1 antibody treatment, which reflects clinical patterns of variable response for patients with cancer types known to be potentially responsive to immune checkpoint antibodies alone or in combination with another type of cancer drug.

Disruption of fibronectin formation in vitro led to increased adherence and transendothelial migration of activated CD8⁺ T cells in the presence or absence of the cytokine CXCL12. The addition of CXCL12 increased transmigration from baseline levels, and the addition of α5β1 antibody increased this by ∼2-fold. CXCL12 can bind to fibronectin,⁴⁰ and such CXCL12 sequestration in the ECM could potentially disrupt the chemokine gradient needed for effective transendothelial migration of lymphocytes in this model, which can then be reversed by exposure to α5β1 antibody.

The murinized antibody targeting intact mouse α5β1 heterodimers (clone 10E7) replicated the reduction in fibronectin fibril formation on endothelial cells as reported with human α5β1-targeting antibodies. This surrogate mouse antibody also led to increased vascular permeability in tumors, in line with previous studies showing increased permeability when ITGA5 is genetically deleted.⁴¹ It is possible that the combination of α5β1 and PD-L1 antibodies leads to increased intratumoral accumulation of both antibodies in the tumor microenvironment, due to α5β1-mediated increases in vascular permeability, and could potentially increase the intratumoral accumulation of chemotherapy. This too could be a mechanism to partially explain the treatment effectiveness of the combination of the two antibodies plus PTX chemotherapy. Additionally, α5β1 antibody monotherapy reduced vessel-associated collagen I, suggesting that fibronectin disruption can lead to further modulation of other ECM components. Collagen I, acting through the leukocyte-associated immunoglobulin-like receptor 1 receptor expressed on T cells, has roles in CD8⁺ T cell exhaustion and has a potential role in PD-1/PD-L1 blockade resistance.¹³ Collagens have also been shown to have inhibitory effects on natural killer cells.⁴²

The therapeutic combination of α5β1 and PD-L1 antibodies led to increased survival and a number of complete tumor regressions in the E0771 breast cancer model; these effects were more pronounced in smaller tumors at treatment initiation (<90 mm³). This also resulted in rejection of E0771 tumor rechallenges; however, as we did not rechallenge with a non-related tumor line, we cannot rule out the unlikely possibility of non-specific rejection. Previous studies have also shown that modulation of the ECM (by inhibiting collagen fiber alignment) in the E0771 model can result in inhibition of primary tumor growth by T cell-mediated mechanisms.⁴³ The α5β1 + PD-L1 antibody combination also resulted in reversion of PD-L1 antibody-mediated increases in exhausted CD8⁺ T cells, to baseline levels. Additionally, others have demonstrated the reduction in exhausted CD8⁺ T cells (PD-1⁺ TIM-3⁺) when T cells are co-cultured with fibroblasts containing an α5 genetic deletion, as well as an increase in exhausted T cells when they are cultured in high-density collagen matrices.²² Additionally, biomechanical signals from tumor-associated stiff ECM have been shown to have direct effects on CD8⁺ T cell exhaustion.⁴⁴ Therefore, it is possible that the roles the ECM plays in driving CD8⁺ T cell exhaustion were mitigated by inhibition of α5β1-mediated mature fibronectin fibril formation.

When combined with PTX, PD-L1 and α5β1 antibodies improved anti-tumor efficacy in the E0771 model, even when treating larger tumors that were less responsive to the antibody combination treatment. The addition of PTX provides potentially translationally relevant information as PD-L1/PD-1 antibodies are administered with PTX for the treatment of patients with TNBC.⁷ The addition of α5β1 antibody or fibronectin modulation could potentially improve responses in patients with TNBC treated with PD-1/PD-L1 blockade and chemotherapy.

Further evaluation of our ECM modulation and immunotherapy strategy revealed varied effects, where the combination failed in the EMT6 parental line model but led to additive anti-tumor effects in the 66cl4 and the immunogenic EMT6-CDDP and CT26 MNNG⁺EMS⁺ models. The mechanisms for these differences in response are currently not known, and this presents a caveat to our work. However, this is also the case clinically; for example, with PD-1/PD-L1 blockade, certain tumor types such as NSCLC and melanoma respond well,⁴⁵ whereas in prostate cancer, this is not the case despite the presence of PD-L1 expression.⁴⁶

It is possible that other processes are interrupted by the targeting of α5β1 and the subsequent reduction in fibronectin fibrillogenesis. For example, reductions in fibronectin within the tumor microenvironment could lead to reductions in VEGF bound to fibronectin that has accumulated in the ECM, and VEGF bound to the ECM is known to increase its signaling potential.⁴⁷ As it is known that VEGF has diverse effects on mediating immunosuppression, this could have an additive effect in targeting the ECM in the context of immunotherapy.^48,49 Fibronectin has also been shown to bind with high affinity to over 25 different growth factors and cytokines without inhibiting their activity,⁵⁰ increasing the complexity of ECM disruption in tumors. Regarding the other processes that α5β1 targeting may have effects on, toxicity may be of concern. However, when phase 1 clinical trials were conducted testing the safety of α5β1 antibodies as monotherapy or in combination with bevacizumab, no toxicities were observed, and a maximum tolerated dose of the α5β1 antibody was not reached.⁵¹ Additionally, we observed no changes in the ECM of organs such as liver or kidneys in mice treated with α5β1 antibody vs. isotype control (Figure S10).

A recent study detailing α5β1 targeting by peptides and antibodies against the α5 subunit showed synergistic combination with PD-L1 antibody blockade, and our study contributes more evidence to the potential of this treatment strategy.²² In the study by Lu et al.,²² the authors reported that some effects of α5β1 targeting are due to disrupting fibronectin formation by fibroblasts, whereas in our study, we describe the additional contribution that endothelial cells have in α5β1 and fibronectin disruption leading to effects on T cell infiltration. In our study, there are likely also effects of blocking α5β1 on other cell types other than endothelial cells, such as fibroblasts and T cells themselves, which could potentially contribute to therapy efficacy. The study by Lu et al. involved colorectal cancer (CRC), which is a tumor type that does not respond to immune checkpoint blockade, with the exception of tumors with a high neoantigen burden due to DNA mismatch repair (MMR) deficiencies.^52,53 In our study, we conducted experiments on the CT26 MNNG⁺EMS⁺ CRC line, which has a high tumor mutational burden due to prior chemical mutagen treatment³⁷ and hence is a better surrogate for CRC with MMR deficiencies. In this case, the combination of PD-L1 antibody and α5β1 antibody improved survival, presenting a potential combination therapy to enhance efficacy in patients with CRC. We also detail the potential use of α5β1 blockade in combination with PD-L1 antibody immunotherapy with or without chemotherapy, providing a potential addition to improve standard-of-care treatment in patients with TNBC.

High gene expression of ITGA5 in patients treated with atezolizumab (as monotherapy or in combination with chemotherapy or VEGF antibodies) in multiple different cancer types was associated with decreased OS in some clinical trials described. This poses the possibility that in certain tumor subsets, the presence of ITGA5 worsens survival in the context of immune checkpoint blockade treatment and provides a rationale for combination treatment targeting ITGA5 in these patients. As immune checkpoint blockade therapy outcomes are dependent on T cell infiltration and activation, it is possible that high levels of ECM contribute to the dampening of CD8⁺ T cell activation and may account for some of the observed effects seen in clinical trials that included atezolizumab treatment.

The accelerated approval of atezolizumab in combination with nab-PTX has recently been re-evaluated and is no longer FDA approved; additionally, the proportions of patients with TNBC who respond to approved pembrolizumab plus chemotherapy still need to be improved; in the neoadjuvant setting, ∼60% of patients show pathological complete responses with pembrolizumab plus chemotherapy vs. ∼50% treated with chemotherapy alone.⁵⁴ Therefore, there is an obvious incentive to increase the number of patients who can respond to immunotherapy, especially in patients with TNBC. Taken together, our results suggest that modulation of the ECM, and specifically fibronectin, may be one such strategy to increase the clinical benefits in patients especially when their tumors express high levels of ITGA5 and/or FN1. More broadly, targeting of fibronectin and the ECM may be a viable approach in other solid tumor types not only to improve tumor responses to immune checkpoint blockade but also to enhance tumor infiltration of CAR T cell or TIL infusion therapy in solid tumors.

Limitations of the study

Our study is limited in its clinical utility at this current stage; there remain many open questions such as whether fibronectin and ECM modulation would have effects on CD8⁺ cytotoxic T cells in patients. While we observed correlations of high ITGA5 expression associated with reduced OS in patients treated with atezolizumab, this was not the case in all tumor types or in all trials. Clinical trials would need to be conducted to address the benefit of combining an α5β1 antibody or other ECM-modulating therapies with a standard-of-care immunotherapy regimen. Additionally, we have not evaluated the exact mechanisms why the α5β1 antibody plus PD-L1 antibody combination works well in some models but not at all in others. This too would be useful information to inform which patients would best benefit. Finally, we did not explicitly investigate the effects that ECM modulation and immunotherapy have on metastasis, an important point when considering potential clinical translation and application.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Weilan Ye (ye.weilan@gene.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

No custom code was used. Any additional information required to reanalyze the data reported in this work is available from the lead contact. The following clinical trial datasets are made available upon request at the European Genome-Phenome Archive under accession codes: IMvigor210 EGAS00001002556, IMvigor010 EGAS00001004997, IMmotion151 EGAC00001001813, OAK EGAC00001002120, POPLAR EGAC00001002120, and IMmotion150 EGAC00001001748.

Acknowledgments

This study was funded by a Genentech (USA) sponsored research agreement awarded to R.S.K. A Canadian Institute of Health (CIHR) Banting Postdoctoral fellowship was awarded to K.A.K. Some support was also from a grant awarded to R.S.K by the CIHR. We thank Shan Man for her expertise in surgeries and tumor implantation. We thank Cassandra Cheng for her excellent secretarial assistance. We also thank the following core facilities for their support: Sunnybrook Histology core facility (Petia Stefanova), the Centre for Flow Cytometry & Scanning Microscopy (CCSM) (Kevin Conway and Paul Oleynik), and Comparative Research for their vital core services at Sunnybrook Research Institute.

Author contributions

Conceptualization, K.A.K., M.C.M., I.P.M., W.C.-M., W.Y., and R.S.K.; methodology, K.A.K. and M.C.M.; formal analysis, K.A.K., M.C.M., W.C.-M., R.P., R.J., G.L., P.H., and P.X.; investigation, K.A.K., M.C.M., W.C.-M., G.L., P.H., P.X., H.Y.K., A.G., R.W., and A.C.; resources, W.Y.; data curation, K.A.K., R.J., and R.P.; writing – original draft, K.A.K.; writing – review and editing, K.A.K., I.P.M., W.Y., and R.S.K.; visualization, K.A.K. and R.P.; supervision, K.A.K., M.C.M., W.Y., and R.S.K.; funding acquisition, W.Y. and R.S.K.

Declaration of interests

During the course of these studies, R.S.K. was a scientific advisory board (S.A.B) member of Novelty Nobility (Seoul, Republic of Korea), a consultant for Pharmabcine Inc (Daejeon, Republic of Korea) and a consultant to Eyepoint Pharmaceuticals (USA), and is currently an S.A.B member of OncoHost (Haifa, Israel). W.Y., R.P., R.W., A.G., R.J., and M.C.M. are or were employees of Genentech (a Roche subsidiary), and some hold Roche stocks. W.Y. has issued (US-8962275-B2) and pending patents regarding human integrin α5β1 antibodies.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-CD3e (for activation)	eBiosciences	RRID:AB_468854
Anti-Fibronectin	Abcam	RRID:AB_447655
Fibronectin RabMab antibody KO validated	Abcam	RRID:AB_2941028
Anti Collagen I	Abcam	RRID: AB_446161
Human α5β1 (hu18C12) antibody	Genentech	–
Mouse α5β1 (10E7) antibody	Genentech	–
Gp120 (3E5.2H12) antibody	Genentech	–
Mouse PD-L1 (6E11) antibody	Genentech	–
Mouse MOPC-21 antibody	BioXcell	RRID:AB_1107784
FcR block	BD Biosciences	RRID:AB_394657
CD45 APC-Cy7	BD Biosciences	RRID:AB_396774
CD3 FITC	Biolegend	RRID:AB_312671
CD4 PerCP Cy5.5	Biolegend	RRID:AB_893324
CD8a BV650	Biolegend	RRID:AB_2563056
CD366 TIM-3 PE	Biolegend	RRID:AB_345377
PD-1 PE-Cy7	Biolegend	RRID:AB_10696422
Cy3-AffiniPure Donkey Anti-Rabbit IgG (H + L)	Jackson ImmunoResearch	RRID:AB_2313568
Biological samples
Various primary tumors from mice	This paper	–
Chemicals, peptides, and recombinant proteins
TNFα	R&D systems	# 210-TA
Human VEGF-A165	R&D systems	# 293-VE
Endothelial cell growth supplement (ECGS)	Alfa Aesar	# AAJ64516MF
Human FGF-2	R&D systems	#233-FB
Human recombinant IL-2	R&D systems	# 202-IL-010
Fibronectin	Sigma Aldrich	#F2006
CXCL12	R&D Systems	#350-NS-050
FITC–dextran (40 kDa)	Sigma-Aldrich	#FD-40
AngioSenseIVM	Perkin Elmer	#NEV10054EX
EndoGRO VEGF media	Millipore	#SCME002
Cell Tracker Green CMFDA Dye	Thermoscientific	#C7025
Critical commercial assays
CD8+ T cell isolation kit, human	Miltenyi Biotec	130-096-495
Experimental models: Cell lines
E0771	ATCC	RRID: CVCL_GR23
EMT6	ATCC	RRID:CVCL_1923
HUVEC (EndoGRO TM HUVEC)	Millipore	SCCE001
mLungEC (mouse lung endothelial cells)	Angiocrine Biosciences	mVera-Lng-01
66cL4	Dr. F. Miller55	RRID:CVCL_9721
CT26 MNNG+EMS+	Kuczynski et al.37	–
EMT6-CDDP	Teicher et al.35	–
Experimental models: Organisms/strains
C57Bl/6 mice	Jackson Laboratories	RRID:IMSR_JAX:000664
BALB/c mice	Jackson Laboratories	RRID:IMSR_JAX:000651
Software and algorithms
R survival	R	RRID:SCR_021137
R Seurat	R	RRID:SCR_016341
Qupath 0.4.0	Qupath	RRID:SCR_018257
Graphpad Prism 5	Graphpad	RRID:SCR_002798
No custom code used
Other
Bassez et al. breast cancer scRNA-seq data	http://biokey.lambrechtslab.org	Accession number: EGAD00001006608
IMvigor210	European Genome-Phenome Archive	Accession number: EGAS00001002556
IMvigor010	European Genome-Phenome Archive	Accession number: EGAS00001004997
IMmotion151	European Genome-Phenome Archive	Accession number: EGAC00001001813
OAK	European Genome-Phenome Archive	Accession number: EGAC00001002120
POPLAR	European Genome-Phenome Archive	Accession number: EGAC00001002120
IMmotion150	European Genome-Phenome Archive	Accession number: EGAC00001001748

Experimental models and study participant details

Cell culture

Human umbilical vein endothelial cells (HUVECs) were cultured in EndoGro VEGF (Millipore # SCME002). Mouse Lung endothelial cells (mLungECs, mVera-Lng-01, Angiocrine Biosciences), were cultured on 0.1% gelatin coated plates in advanced DMEM/F12 (Sigma Aldrich) with 20% FCS, 1% L-glutamine, 1% penicillin, 1% streptomycin, 10 mM HEPES pH 7.0, 50 μg/mL heparin, MEM non-essential amino acids (Gibco), 50 μg/mL endothelial cell growth supplement (ECGS) (Alfa Aesar, cat no # AAJ64516MF), 20 ng/mL recombinant human FGF-2 (R&D Systems, 233-FB) and 10 ng/mL of recombinant human VEGF-A₁₆₅ (R&D Systems # 293-VE). E0771 (female, also referred to as EO771 C57BL/6, ATCC, RRID: CVCL_GR23), EMT6 (female, BALB/c ATCC, RRID: CVCL_1923) EMT6-CDDP (female, from B. Teicher³⁵), 66cL4 (female, also referred to as 66cL-4 RRID:CVCL_9721) and CT26 MNNG⁺ EMS⁺ were cultured in DMEM with 5% FCS. All cells were cultured in 5% CO₂ at 37°C in a humidified incubator. Cell line authentication was performed for EMT6 and E0771 using Mouse Cell Line Authentication Testing (LabCorp) and analyzed using CLASTR v1.4.4. Cells were routinely tested and confirmed to be negative for mycoplasma by in-house Mycoplasma screening PCR using a nine-primer pool or using Lonza MycoAlert Mycoplasma Detection Kit (LT07-318).

Animal experiments

All mouse experiments were carried out in accordance with the guiding principles of the American Physiology Society and were approved by the Institutional Animal Care and Use Committee (IACUC) at Genentech, Inc and/or with the approval of the institutional Animal Care Committee of Sunnybrook Research Institute in accordance with the Canadian Council on Animal Care (CCAC) guidelines. All mice used in experiments were between 6 and 12 weeks of age and were housed under standard conditions with 12-h day-night cycles and ad libitum access to standard chow and water. BALB/cJ (Jackson Laboratories), C57BL/6J (Jackson Laboratories) were used. For breast cancer studies female mice were used. For CT26 MNNG+EMS+ female BALB/c were used to match the cell line.

Method details

HUVEC permeability assays

HUVEC (Millipore Sigma #SCCE001) were seeded at 1 × 10⁵ per well of 24-well plates (Sigma-Aldrich) in complete EndoGRO Medium (Millipore Sigma). Three days later, culture medium was changed to EndoGRO Basal Medium followed by 16 h of incubation. Then, α5β1 antibodies (hu18C12)²⁴ antibodies (10 μg/mL) were added. After an incubation of 30 min, 50 μL of 5 mg/mL FITC–dextran (40 kDa) (Sigma-Aldrich Cat no. FD-40) was added into each insert, and 50 μL of medium was taken from the receiver tray of each well at different time points and measured by a microplate reader (excitation filter, 485 nm; emission filter, 535 nm).

Leukocyte adhesion and transmigration assays

Fibronectin (10 μg/mL) (Sigma cat no. F2006) in PBS was used to coat wells (40 μL per well) in an Incucyte ClearView 96 well plate, incubated at room temperature for 1 h (Sartorius, Cat no. 4582). HUVECs were seeded at 7500 cells per well in EndoGro Basal media, centrifuged at 25g for 3 min and allowed to culture for 24 h. Human CD8⁺ T-cells were purified from PBMCs using negative magnetic cell selection (# 130-096-495, Miltenyi) and activated for 5 days in CD8⁺ T cell activation media (RPMI-1640 supplemented with 10% FBS, Pen/Strep, Glutamine, 10 ng/mL rh-IL2 (R&D cat no # 202-IL-010) and 50 ng/mL anti-human CD3 (eBioscience 16-0037-81). On day of assay activated CD8⁺ T-cells were incubated in 0.5 μM of Cell Tracker Green CMFDA Dye (Thermoscientific #C7025) for 30 min at 37°C. Pretreatment of HUVEC was performed for 3 h, TNFα 100 ng/mL, isotype control antibody or α5β1 (hu18C12) antibody 5 or 10 μg/mL. CD8⁺ T-cells 5000 per well were incubated with HUVECs in EndoGro VEGF (Millipore # SCME002) with 10 ng/mL hIL-2. EndoGro basal media with or without CXCL12 (R&D cat no. 350-NS) 166.67 ng/mL was then added to the reservoir plate and the inserts containing HUVECs and CD8 T-cells added to this plate. After 1 h an inverted centrifuge of the plate was performed to mimic shear force and remove loosely adhered cells (25g for 3 min), imaging on the Incucyte (10x magnification) at this point determined the adhesion index = number of adhered cells/number of cell input. Images were then acquired over a 12-h period every 2 h on the Incucyte to determine the transendothelial migration index = (Number of adhered CD8⁺ T-cells) - (CD8⁺ T-cells no longer in view)/Number of adhered CD8⁺ T-cells.

In vivo permeability assay

Tumor bearing mice were treated with 10mg/kg of α5β1 10E7 or control antibody (gp120) (n = 9–10 treatment group) and at 0 h and 24 h were injected intravenously with the fluorescent blood pool marker AngioSenseIVM (PerkinElmer, cat no. NEV10054EX). Distribution of AngioSense IVM within tumors was measured by visualizing fluorescence (650 nm excitation and/700nm emission) with a Kodak 4000 FX Pro imaging system (CareStream Health) and quantifying fluorescence intensities within regions of interest placed over tumor or adjoining tissue normalized to time = 0. At each indicated time point, animals were anesthetized under isoflurane, with body temperature maintained at 37°C, and were imaged.

Biochemical analysis of DOC soluble and insoluble fibronectin fibrils

The fibronectin Deoxycholate (DOC) insoluble/soluble assay was performed as described previously.⁵⁶ Tumor tissues, or HUVEC were lysed in DOC buffer (2% (w/v) sodium deoxycholate, 20 mM Tris pH 8.8, 2 mM phenylmethysulfonyl fluoride (PMSF) 2 mM EDTA, 2 mM iodoacetic acid, 2 mM N-ethylmaleimide) for the DOC soluble fraction, then centrifuged and the insoluble material was solubilised in SDS buffer (1% (w/v) SDS, 20 mM Tris pH 8.8, 2 mM phenylmethysulfonyl fluoride (PMSF) 2 mM EDTA, 2 mM iodoacetic acid, 2 mM N-ethylmaleimide) for the DOC insoluble fraction. Samples were then analyzed by western blot using fibronectin antibodies (Abcam #ab23750 1 μg/mL), and an insoluble/soluble ratio was determined by normalising to β-actin expression.

Fibronectin fibril staining in vitro

HUVEC or mouse lung endothelial cells (mLungECs) were cultured on acid treated glass coverslips for 2 days and then treated with 10 μg/mL of IgG isotype control (against gp120, Genentech clone 3E5.2H12) or α5β1 10E7 antibody for 20 h. Glass coverslips were washed with PBS twice and incubated with frozen methanol for 5 min. Next coverslips were blocked in Dako blocking solution with 10% donkey serum for 30 min. Then incubated with FN1 antibodies for 1 h at room temperature; after PBST washes, antibodies were detected using donkey anti rabbit Cy3 1:150 dilution (Jackson Immunoresearch cat #711-166-152). Whole coverslips were scanned on a Zeiss AxioObserver with Colibri LED light source and random regions of interest were selected using Qupath⁵⁷ and fibronectin levels assessed. For HUVECs anti-fibronectin rabbit polyclonal antibody was used at 1:50 dilution (abcam # ab23750); for mLungEC Recombinant anti-Fibronectin RabMab antibody KO validated [EPR23110-46] (abcam #ab268020, lot #GR3367702-24) was used at 1.2 μg/mL (1:500 dilution).

In vivo primary tumor therapy studies

Animal experiments were carried out with the approval of the Institutional Animal Care Committee in accordance with Canadian Council on Animal Care (CCAC) guidelines, or in accordance with the Institutional Animal Care and Use Committee (IACUC) at Genentech, Inc. Female BALB/c or C57Bl/6 mice were purchased from Jackson Laboratories. For orthotopic breast cancer models female mice aged 6–8 weeks were implanted in the inguinal mammary fat pad with 100,000 cells of EMT6/P or EMT6-CDDP (BALB/c mice) or 100,000 E0771 cells (C57Bl/6 mice). For CT26 MNNG⁺EMS⁺ 500,000 cells were implanted into BALB/c female mice subcutaneously. Tumors were measured using vernier calipers. For EMT6-CDDP models when tumor volume reached a mean volume of 100–150 mm³ and E0771 60–120 mm³ mice were randomised so that each treatment group had the same or near exact mean tumor volume, and antibody treatment was administered via intraperitoneal (i.p.) injection. PD-L1 antibody clone 6E11 was dosed at 5 mg/kg, α5β1 antibody clone 10E7 at 10 mg/kg, and isotype controls gp120 and MOPC-21 used at 10 mg/kg (control for α5β1 antibody) and 5 mg/kg (control for PD-L1 antibody) respectively, twice weekly. In monotherapy groups, the corresponding isotype antibody was also included as a control for the antibody not being administered, for example, in α5β1 antibody monotherapy group the 10E7 α5β1 antibody at 10 mg/kg and the MOPC-21 isotype control at 5 mg/kg was administered. Paclitaxel (Accord Healthcare Inc., DIN: 02391465) was administered by i.p. injection at 30 mg/kg every two weeks. For surrogate survival analyses mice were euthanized when they reached tumor size limit or displayed signs of sickness, Kaplan-Meier survival plots were generated using the date of endpoint. Median overall survival was determined using Prism, log rank (Mantel-Cox) tests were performed on Prism. Hazard ratios were calculated using the cox proportional hazards (coxph) model using the Survival package in R.

Synergy analyses

The Bliss independence model assumes that no drug interactions are occurring therefore an expected outcome of a combination is based upon the addition of its observed effects of monotherapies.⁵⁸ Hazard ratios determined using the coxph model were used to determine an expected hazard ratio using the following formula: HR_{combo-expected} = HR_α5β1 x HR_PD-L1. Synergy score = HR_{combo-expected} - HR_{combo-observed} when this value is less than zero synergy is assumed. The Loewe additivity model assumes that each drug acts on the same biological pathway, but they have different potencies. Hazard ratios were used to determine expected HR with the formula: HR_{combo-expected} = (HR_α5β1 x HR_PD-L1)/HR_α5β1 + HR_PD-L1. These analyses were also expressed as a percentage change in risk from expected HR vs. observed HR.

Flow cytometry

Primary tumors were resected from mice and digested in an enzymatic solution containing: 1% BSA, 12,500 units collagenase II ∼2 mg/mL (Worthington), 12,500 units collagenase IV ∼2 mg/mL (Worthington), and DNase I, 10 μg/mL (cat. no. LS006333, Worthington) for 45 min and then a single-cell suspension gained. After neutralisation in DMEM containing 10% FCS, digests were passed through a 70 μm cell strainer and resuspended and washed in PBS. 1 million cells per staining experiment were incubated with Zombie violet viability dye diluted 1:1000 in PBS for 30 min in the dark at room temperature. After washing with FACS buffer, cells were incubated with FcR block (5 μg/mL) (BD Biosciences cat. no. 553142) for 30 min and then incubated with primary antibodies at 4°C for 1 h. Primary antibody panel: CD45-APC-Cy7 (BD Biosciences #557659), CD3 FITC (Biolegend #100306) 5 μg/mL, CD4 PerCP Cy5.5 (Biolegend #100434), CD8a BV650 (Biolegend #100742), TIM-3 PE (Biolegend #119703), PD-1 PE-Cy7 (Biolegend #135216), at 1:100 dilution for 2 μg/mL. Fluorescence minus one (FMO) control conditions were included and used for the gating strategy (Figure S5A). Flow cytometry data was acquired on an LSR II (BD Biosciences) and analyzed using FlowJo v10.8.1.

Clinical trial related bioinformatics analyses

Overall survival was defined from the time of primary tumor diagnosis to death or the last day of follow-up (censor). Patients were stratified based on the median ITGA5 expression into “ITGA5 high” and “ITGA5 low” groups within each trial. p values for Kaplan–Meier analyses were derived using the log rank test. The hazard ratios between high and low expressors were calculated by a univariate Cox proportional hazards model in R. Patient consent was obtained through the clinical trial recruitment process.

scRNAseq analysis

A raw count matrix was obtained from Bassez et al. study as a rds file and metadata from cohort 1 (cohort A).²³ We performed analyses as described in Bassez et al., briefly, we used the Seurat package in R as well as dplyr, ggplot2, clusterProfiler, and org.Hs.eg.db packages for annotation and creating figures. A Seurat object was created from the raw count matrix to which the metadata was linked. We then performed a standard workflow of normalization, variable finding, scaling and principal component analysis with 50 components. Seurat was used to find neighbors using 20 principal components, clustering at 0.5 resolution and create a UMAP at 20 principal components. Cell types were annotated in the metadata and included the broad labels of cancer cells, endothelial cells, fibroblasts, myeloid cells, plasmacytoid dendritic cells, B-cells, T-cells and mast cells. Expression levels are represented as log-transformed counts per 10,000 transcripts.

Gomori Trichrome

Kidneys and livers were harvested from BALB/c mice bearing EMT6-CDDP tumors treated with IgG isotype or α5β1 antibody at 21 days post implantation. Tissues were fixed in zinc formalin (Sigma #Z2902) overnight (∼16 h). Tissues were paraffin embedded and sectioned at 6 μm. Trichrome staining was performed using Gomori aniline blue one-step (Newcomer Supply #1403C). Sections were deparaffinized and hydrated in distilled water, incubated in preheated Bouin’s solution (Sigma-Aldrich, cat. # HT10132-1L) at 56°C for 1 h. Slides were washed in running tap water until yellow coloration was removed, then rinsed in distilled water. Slides were stained in Weigert Iron Hematoxylin (Sigma-Aldrich, cat. # HT1079-1Set) for 10 min and rinsed in running tap water for 10 min. Then stained in Gomori One-step for 20 min at room temperature. Sections were differentiated in 0.5% acetic acid for 2 min and then dehydrated, cleared in xylene and mounted with Cytoseal 60. Whole slides were scanned using a Tissuescope LE (Huron Digital Pathology). Qupath was used to analyze Gomori stained collagen.⁵⁷ A thresholder was created at full resolution using a Gaussian prefilter in the hematoxylin (blue) channel, positive pixels were defined as above 1. Random regions of interest (12 for kidney and 15 for liver) were generated within the area of the tissue and a mean % coverage of ECM was calculated per ROI.

Quantification and statistical analysis

Statistical tests were carried out using Graphpad Prism or R. Error bars represent standard error mean. Log rank tests were carried out to compare survival curves, unpaired t-tests were used for analysis between individual treatment groups. N numbers displayed in figure legends denote biological replicates unless stated otherwise. Details for statistics can be found in figure legends.

Published: September 8, 2025