Tumor stiffening reversion through collagen crosslinking inhibition improves T cell migration and anti-PD-1 treatment

Only a fraction of cancer patients benefits from immune checkpoint inhibitors. This may be partly due to the dense extracellular matrix (ECM) that forms a barrier for T cells. Comparing 5 preclinical mouse tumor models with heterogeneous tumor microenvironments, we aimed to relate the rate of tumor stiffening with the remodeling of ECM architecture and to determine how these features affect intratumoral T cell migration. An ECM-targeted strategy, based on the inhibition of lysyl oxidase (LOX) was used. In vivo stiffness measurements were found to be strongly correlated with tumor growth and ECM crosslinking but negatively correlated with T cell migration. Interfering with collagen stabilization reduces ECM content and tumor stiffness leading to improved T cell migration and increased efficacy of anti-PD-1 blockade. This study highlights the rationale of mechanical characterizations in solid tumors to understand resistance to immunotherapy and of combining treatment strategies targeting the ECM with anti-PD-1 therapy.


INTRODUCTION
In the last decade, a significant progress has been made in the development of T cell-based immunotherapies (1). The two main T-cell based immunotherapies are adoptive T cell therapy and immune checkpoint inhibitors. Monoclonal antibodies blocking the immune checkpoints cytotoxic T lymphocyte associated antigen 4 (CTLA-4) and programmed death 1 receptor (PD-1) have quickly gone from proof of concepts to FDA approved first and second line treatments for a significant number of tumors even in late stages (2). However, an elevated percentage of patients with solid tumors fail to respond to these therapies. The mechanisms underlying the poor response to immune checkpoint inhibitors are still uncertain, nevertheless recent results suggest that T cell function and distribution in the tumor are affected by numerous immunosuppressive mechanisms (3). It is well established that in progressing tumors T cells exhibit a particular phenotype unable to normally respond to tumor antigens. In addition, in a large proportion of tumors, T lymphocytes are excluded from the tumor cell regions in a so called "excluded-immune profile" (4)(5)(6). Ineffective T cell migration and penetration into the tumor mass might represent an important obstacle to T cell-based immunotherapies. As a support for this notion, various clinical studies have shown that tumors enriched in T cells are more susceptible to be controlled by PD-1 blockade. In contrast, tumors with so-called immune excluded profiles, in which T cell are present within tumors but not in contact with malignant cells, are refractory to PD-1 blockade (4,7). Particularly, the fibrotic state of desmoplastic tumors can cause immunosuppression through multiple mechanisms (8). The hypothesis of a physical resistance to T cell infiltration and migration related to the heterogeneity and aberrant organization of the extracellular matrix with respect to the tumor nests has emerged recently (9,10). By using dynamic imaging microscopy, we highlighted the detrimental impact of collagen fibrils architecture on the migratory behavior of T cells in fresh human tumor explants. Both a guiding strategy combined to a physical hindrance process have been shown to restrain T cells from contacting tumor cells, thus leading to the T cell excluded profile (11,12). Hence, a dense fibrotic stroma could raise physical obstacles to immune cell infiltration similar to the previously established stromal resistance to chemotherapeutics, antibodies, nanoparticles or virus tumor penetration (13,14). In addition, cellular components of tumor-associated fibrosis, particularly the cancer-associated fibroblasts (CAF), can have both direct and indirect effects on T cell infiltration and function (8). Accordingly, one important challenge in the field is to develop strategies targeting tumor fibrosis in order to reverse immune exclusion and to improve T cell-based 4 immunotherapy. Recent studies have been undertaken with this objective. T cells engineered to express a chimeric antigen receptor together with heparanase, an ECM degrading enzyme, show enhanced infiltration into xenografted tumors as well as anti-tumor efficacy (15). Recently, a major role for the TGF signaling pathway in promoting T cell exclusion from tumor cells has been demonstrated. In breast mouse tumor models, neutralizing antibodies against TGF were shown to reduce collagen I production, overcoming the T cell excluded profile and increasing the efficacy of anti-PD-L1 antibodies (7,16,17). In cholangiocarcinoma, an immune mesenchymal subtype has been identified, which is associated with TGF signature and poor tumor infiltrating cells (18). Other axes including the CXCR4/CXCL12 in breast metastasis and the focal adhesion kinase in pancreatic ductal adenocarcinoma (PDAC) have also been associated with both desmoplasia and absence of cytotoxic T lymphocytes in tumors from mice (19). Consequently, the inhibition of these axes in preclinical mouse cancer models was shown to reduce fibrosis while significantly increasing T cell infiltration and improving response to checkpoint inhibitors (7,16,20,21). Clinical trials testing such combination are currently ongoing in advanced pancreatic cancer, mesothelioma, urethelial carcinoma and other malignancies (NCT02546531, NCT02758587, NCT02734160, NCT04064190, NCT02947165).
However due to patient and tumor heterogeneity, there is no clear indication on how the T cell distribution in tumors is related to the fibrosis level and to the different architectures of ECM. Thus, there is an urgent need to assist in matching combination approaches to patient populations who could benefit from stromal modulation strategies to improve their response to immunotherapy.
Companion matrix-derived biomarkers and imaging approaches should provide insights into the contribution of the ECM remodeling in shaping the immune milieu of the tumor. Particularly, a critical determinant of fibrotic tumor progression -the tumor mechanics -has been poorly investigated through the prism of immune impact. An important feature of fibrotic tumors is their considerable higher stiffness compared to their neighboring healthy tissues which is highly correlated with cancer progression and metastasis, particularly in breast, colorectal, liver and pancreatic tumors (22,23). The use of non-invasive imaging techniques such as shear wave elastography (SWE) and magnetic resonance elastography (MRE), designed to monitor stiffness of any given tissue, allow an accurate diagnostic and characterization of malignant lesions in vivo (24). The extensive remodeling of the stromal components increasing tumor stiffness can mechanically activate intracellular signaling 5 pathways that promote tumor progression and at the same time can dampen T cell functions including migration and infiltration into tumor islets (25)(26)(27)(28). However, there is a lack of studies correlating the mechanical properties of tumors to their heterogeneous ECM architecture and T cell infiltration capacity. Here we aim at filling this gap through a comprehensive investigation of stiffness evolution in several preclinical mouse models of pancreatic, breast and bile duct carcinomas, presenting different ECM organization, coupled to dynamic imaging of fresh tumor slices to monitor T cell motility. In concert with these imaging biomarkers of both mechanical properties, ECM architecture and T cell migration, we explored the consequences of altering the ECM by inhibition of the lysyl oxidase (LOX), a copper-dependent enzyme responsible for the crosslinking of collagen molecules into fibers that has been seen to be overexpressed in many metastatic tumors and responsible for malignant progression (29,30,31). We highlight that LOX inhibition has different mechanical modulating effects depending on the ECM architecture, with significant improvement in T cell mobility. Despite minor effects in primary tumor growth upon LOX inhibition or PD-1 blockade treatment alone, their combination increases effector CD8 T cell accumulation in tumors and significantly delays tumor progression in a pancreatic cancer model.

Relationship between tumor structure and tumor mechanical properties in different preclinical carcinoma mouse models
One key aspect when testing immunotherapeutic agents is the use of relevant preclinical models that closely mimic the properties of human solid tumors. Human carcinomas derive from epithelial cells and therefore harbor a typical though heterogeneous structure with tumor cells forming compact islets or nests surrounded by the stroma, enriched in ECM proteins, fibroblasts, blood vessels, and immune cells. To unravel the relationship between tumor growth, ECM remodeling, stiffening and immune infiltration, we characterized the tumor structure and the mechanical properties of five different preclinical models, recapitulating the structure heterogeneity of different carcinomas (Table S1): subcutaneous model of cholangiocarcinoma (EGI-1), subcutaneous (MET-1) and transgenic model (MMTV-PyMT) of mouse breast carcinoma, orthotopic (mPDAC) and subcutaneous (KPC) models of mouse pancreatic ductal adenocarcinoma. A multiscale evaluation of the mechanical properties of the tumors was performed. At the macroscale, we measured tumor stiffness during tumor growth using SWE, a noninvasive imaging technique that allows the quantification and mapping of tumor stiffness ( Figure 1A, Table 1). The presence of very stiff regions, defined as areas with an elastic modulus > 40 kPa, in the tumor was quantified together with the average stiffness of the tumor ( Figure 1B, Table   1). At the micron-scale, we evaluated tumor organization and fibrosis using hematoxylin-eosin-safran (HES) ( Figure 1C) and Sirius Red staining. Sirius Red is a highly specific stain for collagen fibers that combined with polarizing microscopy allows differentiating thin collagen fibrils from thick and densely packed collagen fibers (32). Under polarized light, thin fibers show a greenish-yellow birefringence whilst thicker and densely packed fibers give an orange red birefringence. By separating these two colors, it is possible to quantify the amount of thick and densely packed fibers present in the tumor ( Table 1). The fibrillar collagen network was determined using second harmonic generation imaging (SHG), that allows to analyze the architecture and density of fibrillar collagen without having to use detection antibodies ( Figure 1D, Table 1).
In the EGI-1 cholangiocarcinoma model, tumor stiffening and tumor growth have a strong positive correlation ( Figure 1A). The stiffness distribution is highly heterogeneous, presenting 20 % of stiff regions (> 40 kPa) in average that goes up to 50-60 % in tumors with higher volume. In terms of 7 architecture, the tumor and its extensive stroma compartment, occupying around 20 % of the tumor, are well separated. This is a typical trait of desmoplastic tumors and the model accurately reproduces the architecture of human cholangiocarcinoma. Its collagen network is characterized by long (85.9 ± 42.0 µm) and thick (7.4 ± 2.2 µm) collagen fibers that are densely packed (6 % of the tumor) (Table 1, Figure 1D).
The mouse pancreatic ductal adenocarcinoma KPC model (33) exhibits similar features to that of EGI-1. A high positive correlation between tumor stiffness and volume, presenting over 20 % of stiff regions at high tumor volumes, was observed ( Figure 1A). KPC tumors present a typical segregation of ECM and tumor nests, with a high proportion of stroma (~20 %). However, the extension of the stromal areas is lower than that of EGI-1 with higher intercalation with tumor islets. Its collagen network is characterized by shorter (67.1 ± 35 µm) and thinner (4.4 ± 0.9 µm) collagen fibers than that of EGI-1 (Table1, Figure 1D).
The spontaneous orthotopic murine breast cancer model MMTV PyMT, although slower in its growth as compared to subcutaneous tumors also stiffens during tumor progression ( Figure 1A). However, there is a lower density of stiff regions (<16%). Of note, the fact that it is an orthotopic tumor in a genetic model that develops about 10 tumors limits the maximal tumor volume reached for this analysis. Hence, we cannot compare this model with the other models at high tumor volumes for ethical reasons. This spontaneous tumor model also presents a tumor islets-stroma structure, but with a lower amount of stroma (~13 %) as compared to EGI-1 and KPC models. The stroma is more dispersed and intercalated with the tumor compartment. Collagen fibers are characterized for being thin (3.6 ± 1.1 µm) and long (83 ± 41 µm), forming densely packed regions taking up to 4.8 % of the tumor.
The orthotopic murine pancreatic ductal adenocarcinoma (mPDAC) model has a very different profile compared to the other models. The stroma takes up ~ 40 % of the tumor, however it is well nested into the tumor with no spatial segregation of stromal and tumor compartment. mPDAC cells have already undergone the epithelial mesenchymal transition (EMT), which can explain this architecture (34). Cells that have undergone EMT gain the capacity of secreting ECM proteins that modulate the structure of the stroma. The mPDAC collagen network is made up of thin (3.5 ± 0.8 µm) and dispersed collagen fibers, which accounts for the lower presence of densely packed collagen regions (2.7 % of the tumor).
The mean tumor stiffness is lower than that of the other models, partly explained by the limits of the 8 maximal tumor volume reached in this orthotopic model (for ethical reasons) and partly, to the collagen architecture. Tumor stiffness also increases with tumor volume (Figure 1B) in line with previous studies performed in this model (34). Unlike the above mentioned models which exhibit a high correlation between stiffness and tumor growth, the mouse breast carcinoma MET-1 tumor model is characterized by low tumor stiffness (Figure 1A), a limited stroma (~ 6 %), the lack of tumorislet/stroma organization and the presence of thin (3.2 ± 0.9 µm) and dispersed collagen fibers (Table 1, Figure 1D).
This thorough analysis enabled us to confirm the correlation of high tumor stiffness measured noninvasively with collagen accumulation associated to a segregated architecture of thick and densely packed collagen fibers (Sirius red positive) surrounding tumor nests. In contrast, tumors with entangled and thin mesh of collagen present lower rigidity despite overall high collagen content.
Particularly the appearance of stiff regions > 40kPa is seen as a physical biomarker of intratumor heterogeneity and ECM segregation. This analysis maps out potentially relevant preclinical tumor models which might reproduce the fibrotic evolution of human breast, pancreatic and bile duct tumors and their architecture heterogeneity.

LOX modulates tumor stiffness and the ECM organization
The panel of tumor stroma structures reported above allows us to investigate the direct effects of ECM modulating agents in situations mimicking the heterogeneity observed in human carcinoma. Thus we sought to determine whether beta-aminopropionitrile (BAPN), an inhibitor for LOX enzymatic activity, could modulate tumors' mechanical properties in concert with the stroma architecture (31). For these experiments, BAPN was administered in the drinking water of mice upon tumor cell implantation and until their sacrifice for most models, except for MMTV-PyMT model that was treated approximately at the time that tumors start to spontaneously develop. We first examined the effect of LOX inhibition on tumor stiffness (Figures 2A and 2C) and on the presence of stiff regions ( Figure 2B). Results show that all models, except for MET-1, undergo reduction in mean stiffness when LOX is inhibited. EGI-1 and KPC models both show the most striking differences. The change is mainly perceived at late stages of tumor development, since in these models, tumor stiffness is positively correlated with tumor growth. In the MMTV-PyMT model, however, significant differences were noted throughout the development of the tumor. For the mPDAC model, tumor stiffness was only evaluated at the end of the 9 BAPN treatment. A significant decrease of mean tumor stiffness is seen in BAPN treated mPDAC tumors.
Notably, BAPN treatment did not affect tumor growth in most models (Figures S1A-S5A), except for mPDAC ( Figure S4A). To verify that the variation in tumor stiffness was not due to difference in tumor volume, the mean tumor stiffness of control and treated mice were compared at different tumor volumes (Figures S2C-S6C). In both KPC and EGI-1 tumors a clear difference in mean tumor stiffness can be seen in tumors with a volume > 400 mm 3 (Figures S1C and S2C).
We also explored whether the presence and proportion of stiff regions was reduced when LOX was inhibited ( Figure 2B). The percentage of control tumor area with a mean stiffness >40kPa increased with time (and tumor volume), indicating that there is not only an increase of overall mean stiffness, but also an increase of the heterogeneity of stiff regions within the non-treated tumor. However, this percentage was significantly reduced in BAPN-treated tumors with marked differences observed in the KPC model and to a lesser extent in EGI-1 and MMTV-PyMT models. In mPDAC, BAPN-treated tumors did not display stiff regions (Figure S4 C-E). The only model that does not respond to LOX inhibition by stiffness reduction is MET-1 (Figure S5  A significant reduction of collagen fiber width distribution was observed in EGI-1, KPC and mPDAC models, whilst fibers in the MMTV-PyMT tumor model did not display a significant change in their width ( Figure 3A). The most substantial differences was seen in the EGI-1 model, where collagen fiber width was decreased by 9,4 % on average (6.7µm versus 7.4 µm). Changes in mPDAC and KPC were less pronounced, with a reduction of 4.4 and 5 % respectively ( Table S2). The inhibition of LOX did not affect collagen fiber length in any of the models (Table S2).
We next assessed the orientation and linearization of collagen fibers in control and BAPN-treated animals. In general, the collagen fibers in normal tissues are typically curly and anisotropic in contrast to the situation observed in progressing tumors in which many of the fibers progressively thicken and linearize. Collagen fiber orientation was described as the coefficient of variation (CV) of the angle for all fibers, the smaller the CV is, the more aligned the fibers are. Fibers in non-treated tumors remained mainly oriented in one dominant direction (Figures 3B, 3C and 3F) with a CV from 1.85 to 0.5 consistent with previous findings (35). LOX inhibition tends to disrupt the alignment of collagen fibers, meaning that they were more dispersed and oriented in different directions with increased CV as compared to control conditions. The most significant effects were seen in KPC and mPDAC model.   Overall, our results demonstrate the rationale of targeting LOX enzymatic activity for normalizing tumor mechanical properties and ECM structure (mostly collagen fibers compaction, segregation and linearity) in tumors exhibiting high tumor stiffness together with mechanical and structural heterogeneity.

LOX inhibition increased intratumoral T cell migration and infiltration
Previous studies performed in our group have proved that the density and orientation of the ECM can have important impact on T cell behavior and their displacement in fresh human lung and ovarian tumor slices (11,38). Motile T cells were mainly found in loose ECM stromal regions whereas fibrotic areas were devoid of lymphocytes. Based on this, we hypothesized that LOX-dependent tumor stiffness could consequently affect T cell migration in tumors and eventually predict the T cell behavior in the various ECM environment. To test this, we performed dynamic imaging of T cell migration on fresh tumor slices from mice treated or not with BAPN, using an approach previously established (39).
EGI-1 is a xenografted tumor model, with implantation of human carcinoma cells into immunesuppressed mice that lack resident T cells. Thus, in order to evaluate T cell migration in this model we isolated human peripheral blood T cells (PBT) and activated them in vitro. We then added the activated PBT onto fresh tumor slices and analyzed their migration using real-time confocal microscopy. As the MMTV-PyMT tumor model is poorly infiltrated in host T cells, we investigated the migration of exogenously purified murine activated PBTs in the same manner as for the human EGI-1 model. In both mPDAC and KPC mice tumor models, resident tumor infiltrating T lymphocytes (TILs) were monitored after staining with directly-coupled anti-CD8 antibodies (39). The three parameters In every tested model, LOX inhibition resulted in an overall increase in T cell migration as compared to control condition. However, different parameters were altered in each model depending on the nature of T cells that were monitored. In EGI-1 BAPN-treated tumors, activated PBT cells displayed longer displacement lengths compared to untreated tumors. This was also true for MMTV-PyMT tumors.
Since activated PBT are not specific to the tumor, the effects observed are due to LOX inhibition and not due to T cells engaging in stable conjugates with cancer cells trough antigen recognition. In EGI-1 tumors, the trajectory straightness of activated PBTs was also significantly increased upon LOX inhibition.
When we evaluated the effect of BAPN on the dynamics of endogenous T cells infiltrated into KPC and mPDAC tumors, the most striking difference was found in the mPDAC model with a 5-fold increase in the displacement of T cells. In terms of cell speed, enhancements were observed in both the KPC and mPDAC models. The trajectory straightness was increased in mPDAC models but not in KPC tumors. In addition, we also evidenced an increased infiltration of resident T cell in tumors when treated with BAPN. Under anti-LOX treatment, a significant 4-fold increase of CD8+ T cell infiltration in both the stroma and tumor islets was observed in the KPC model (Table 2, Figure S10). Table 2 have been obtained with data pooled from all mice either treated or not with BAPN. We decided to extend our analysis at the level of individual mouse and investigated the relationship between T cell motility (speed and displacement) and mean stiffness of control and BAPNtreated tumors. Our data indicate that T cell motility was inversely correlated to tumor stiffness ( Figure   5) in line with Table 2 and Figure S9. However, in three out of the four models tested this correlation is different if one compares control and BAPN-treated tumors. In BAPN-treated tumors, there is a clear 13 inverse linear correlation between T cell migration speed and mean tumor stiffness as evidenced by a steep slope. In comparison, such correlation is less pronounced in control stiff tumors. Similar results were observed when comparing T cell displacement and mean tumor stiffness. These results suggest that above a stiffness threshold that depends on the model, T cells are mostly arrested. In contrast, as tumor stiffness decreased by inhibition of LOX, T cell migration was restored. Thus, T cell motility is highly influenced by small variations in the stiffness of softer tumors as it is the case when LOX's activity is inhibited.

Results reported in
Overall, these results suggest that the excessive accumulation and linearization of collagen in ECM limits T cell migration within several rigid tumors with desmoplastic evolution and that LOX-inhibiting BAPN treatment can both reverse tumor stiffening and improve T cell infiltration and migration to tumor cells. We also identify SWE tissue stiffness as a predictive physical marker of T cell motility and infiltration in desmoplastic tumors.

LOX inhibition improves response to anti-PD-1 therapy.
Even though the inhibition of LOX was followed by an increase in CD8 T cell number and migration, this finding was not accompanied with major effects on tumor growth in 4 out of the 5 tumor models tested (Figures S1-S5). In different settings, an increase in intratumoral T cell motility is not sufficient to reduce tumor growth if T cells are still impaired in their capacity to respond to tumor antigens (39).
Consequently, we decided to assess whether LOX inhibition could improve the response to immune checkpoint inhibitors. KPC tumor bearing mice were treated with BAPN combined with anti-PD-1 antibodies. Mice were treated or not with BAPN from tumor cell injection up to their sacrifice and were treated with anti-PD-1 antibodies when the tumor volume was around 80 -150 mm 3 . At this point the mice received 4 doses i.p. injection of anti-PD-1 or isotype control antibody at 4 days intervals (i.p. injection). As shown in Figure 6A, while the anti-PD-1 alone was not able to reduce tumor growth, the combination of BAPN with the checkpoint inhibitor significantly delays tumor progression. We then profiled the immune cell population in these tumors. We found that BAPN treatment alone significantly decreases the number of polymorphonuclear neutrophils (Figure 6B), but increases the presence of MHCII+ tumor associated macrophages (TAMs) (Figure 6C) while the combination therapy expanded the percentage of GrzmB CD8+ T cells (Figure 6D) and the ratio of CD8+ to Treg cells (Figure 6E).
We also analyzed the amount of cytokines in supernatants of whole-tumor slices derived from these 14 experiments. Results show that the combination therapy led to an increase in TNF and RANTES, supporting further the increase of T cell infiltration and activation in this condition. In both BAPN and BAPN combined with anti-PD-1 conditions we observed an increase of GrzmB+ levels compared to the control condition.
Overall, while ECM and stiffness normalization achieved though LOX inhibition increases T cell infiltration and migration, this strategy also improves the efficacy of anti-PD-1 blockade

DISCUSSION
Despite the success of targeting the stromal compartment in tumors (7,11,15,19,20,41), in particular tumor ECM, there are still a series of challenges that remained to be addressed. In the first part of this study we tackle two of these challenges. One of them is finding ways to accurately assessing the architecture of the stroma. In this paper, we propose a thorough analysis combining non-invasive imaging techniques for a macroscopic characterization of tumor stiffness with advanced microscopy techniques to elucidate the collagen network structure, one of the most important components of the tumor ECM. We established a link between the architecture of the collagen network in different desmoplastic tumors with the tumor mechanical properties. This allowed us to extract mechanical and structural reference values and allowed to later evaluate the effect of an ECM targeted therapy. Such an approach could be translated to patients given the availability of SWE techniques that could be correlated to histological observations of the TME.
Accurately modeling tumor complexity and heterogeneity found in patients' tumors using a single preclinical model, is a real challenge. In the first place, many preclinical models do not always reproduce the carcinoma structure found in their respective human tumors. Another important feature of human solid tumors, often absent in mouse tumor models, is their considerable higher stiffness compared to their neighboring healthy tissues which is highly correlated with cancer progression and metastasis. These two features can often be found in human xenograft models that, on the other hand, are not suitable to study immune reactions. For all of these reasons, in this study we have set up and characterized five different solid tumor models covering 3 types of carcinoma, in order to be able to tackle the different aspects of the TME as well as cover the heterogeneity found in patients. It is well established that the number of T cells found within a tumor as well as their ability to migrate and reach cancer cells is key for an effective antitumoral response. These last few years a lot of efforts have been made in identifying cells and factors controlling the migration of T cells within tumors. The notion that prevailed is that growing tumors are composed of cells and factors, such as macrophages and hypoxia, hostile for T cells to migrate (40,42). The importance of the ECM in controlling the distribution and migration of T cells in tumors has also emerged. In human lung and ovarian tumors we found that T cells preferentially accumulate and migrate in stromal regions that exhibited a loose matrix architecture but not in dense regions (11,38). Likewise, in aged skin, dermal fibroblasts harbor a phenotype similar to CAFs and produce ECM matrices that limit T cell displacements (43). However, in triple-negative breast cancers and in pancreatic tumors T cells were still found in dense networks of collagen fibers (35,44). Since most of the aforementioned studies were correlative, we decided in this study to specifically alter the ECM network by LOX inhibition and investigated the consequences on T cell motile behavior. Our results confirmed the importance of the ECM and tissue stiffness in controlling the migration of T cells in tumors. However, not all tumor models react similarly to BAPNtreatment and differences were observed on T cell motility parameters. In KPC and mPDAC models, LOX-inhibition results in an increase in TIL migration speed, whilst in the EGI-1 and MMTV-PyMT models the impact was mainly on T cell displacement. These differences in behavior can be explained by i) the fact that in the former case we analyzed resident TIL and in the later ex vivo purified and activated PBT that are not specific to the tumor and by ii) the fact that LOX inhibition causes differential effects in each of the models that can translate to differences in the impact on T cell migration. Regardless of the explanation, T cell intratumoral motility was inversely correlated to tumor stiffness as measured non-invasively using SWE (Figure 5) and this was true in all tested models.
However, this relationship is not linear and in most models T cells strongly decelerate when a threshold in stiffness is reached. In soft tumors such as those induced by LOX inhibition, T cells manage to migrate. Conversely, in stiff non-treated control tumors T cell migration is impeded. These data fits well with results obtained in vitro in a range of 3D collagen matrices showing that T lymphocytes have the ability to adapt their morphology to the structure of the tissue up to a certain limit (45). In dense collagen matrices, T cell motility is halted. Our analysis supports the idea that elastography measurements could provide valuable companion markers to evaluate the need for antistromal strategy in order to normalize tumor stiffness and consequently improve T cell migration.
In this study we did not take into consideration possible effects of LOX inhibition on tumor blood vessels. Previous studies have reported a reduced angiogenesis after LOX and LOX-like protein inhibition and an increased perfused vessel density in the case of overexpression of LOX (46,47).

This could partly explain why we observe a significant increase in T cell infiltration in KPC tumors upon
BAPN treatment. However, other recent studies argue the opposite as an increase in collagen crosslinking and matrix stiffness resulted in an increase in angiogenic sprouting. Conversely, the inhibition collagen cross-linking in tumors resulted in reduced vasculature density and permeability (48).
Given the low efficacy of T cell based immunotherapies in solid tumors any method to increase its effect on tumor regression is of interest. With the exception of desmoplastic melanomas, features of wound healing and fibrosis are usually detrimental to anti-PD-1 responses (49,50). Accordingly, a number of anti-fibrotic strategies have been recently implemented in combination with immune checkpoint inhibitors (10). One of the most promising targets appears to be TGF. In preclinical mouse tumor models, TGF inhibition with immune checkpoint blockade induces complete and durable responses in otherwise unresponsive tumors. However, due to TGF pleiotropic effects, concerns regarding blockade of this cytokine arose (7,16).
Our study indicates that LOX represents another valuable target as its inhibition in the transplanted KPC model increases the efficacy of anti-PD-1 treatment, while monotherapy with either agent alone is ineffective. Moreover, the combination treatment was associated with a tumor microenvironment 18 shifted towards antitumoral effector cells and components whereas immunosuppressive cells were reduced. The reason of this reprogramming is not known for the moment. Apart from T cells, other immune cells can be subjected to regulation by the ECM. For instance, recent crosstalk between tumor-associated macrophages and ECM have been reported (51).
Although the clinical use of BAPN has been impeded by concerns regarding toxicities, other strategies to inhibit LOX in cancer and fibrotic disease patients are currently ongoing (52). Our work confirms LOX as a molecular target to improve T cell migration dynamics as well as to ameliorate the immunosuppressive microenvironment. It paves the way for clinical trials combining LOX inhibitors with PD-1/ PD-L1 blockade, possibly in biomarker-selected cohorts of patients with high tumor stiffness evaluated with non-invasive imaging approaches. Human and mouse CD8+ T cells were isolated and cultured as described in the Supplemental experimental procedures.

In vivo studies
All animal experiments were performed in agreement with institutional animal use and care regulations after approval by the animal experimentation ethics committee of Paris Descartes University (CEEA 34, 16-063). For details on protocols used for every mouse models, see Supplemental methods.
Tumor growth was followed with a caliper, and tumor volume (V) was calculated as follows: xenograft volume = (xy 2 )/2 where x is the longest and y, the shortest of two perpendicular diameters.
For LOX inhibition, animals were treated with BAPN (3mg/mL, Sigma) in the drinking water, which was changed twice per week. In all implantable models (KPC, mPDAC, MET-1 and EGI-1) animals were
Slices (350 μm) were cut with a vibratome (VT 1000S; Leica) in a bath of ice-cold PBS.
In EG-1 and MMTV-PyMT models, T cell migration was assessed on previously purified and activated T cells introduced into fresh tumor slices. In KPC and mPDAC models, resident CD8 T cell migration was evaluated after staining for 15 minutes at 37°C with PerCP-e710 anti-mouse CD8a (53-6.7 clone, eBioscience). For details see Supplemental Information.

Dynamic imaging analysis
A 3D image analysis was performed on x, y, and z planes using Imaris 7.4 (Bitplane AG). First, superficial planes from the top of the slice to 15 μm in depth were removed to exclude T cells located near the cut surface. Cellular motility parameters were then calculated. Tracks >10% of the total recording time were included in the analysis.
Second harmonic generation microscopy.

Cytokine detection in tumor slice supernatants.
Fresh KPC tumor slices were prepared as previously described and kept at 37 °C in 24-well plates with 0.5 mL RPMI per well. Four to five slices were put in culture for each mouse. Eighteen hours later supernatants were collected and centrifuged at 300 × g to eliminate suspension cells. Cell-free supernatants were frozen and stored at 80 °C. Granzyme B, TNF and RANTES (CCL5) release was assayed by Luminex technology (Bio-Plex 200 from Bio-Rad) with a customized Milliplex kit (Merck Millipore).

Statistical analysis
Results were analyzed using the GraphPad Prism 5.0 statistical software. Data are shown as means ± standard error of the mean (SEM). For comparisons between two groups, parametric Student t-test or 22 non-parametric Mann-Whitney test were used. For comparisons between more than two groups, a parametric One-Way analysis of variance (ANOVA) test was followed by a posteriori Kruskal-Walis test.

SUPPLEMENTAL INFORMATION
Supplemental information includes 10 figures, 2 movies, supplemental experimental procedures, 2 supplemental tables, one supplemental reference.  Figure 4B shows tissues'

ACKNOWLEDGMENTS
Young's modulus mean values ± standard deviation to highlight the difference between control and BAPN tissues. Statistical significance was determined using two-tailed Student's t-test for overall values (n=6).          following manufacturers protocol. At day 7 after activation T cells were used for migration experiments.

Mouse tumor models
EGI-1 subcutaneous model mPDAC orthotopic model 51 10 3 cells suspended in 50 μL of PBS were injected orthotopically in the pancreas of 6-week-old FVB/n mice (Janvier, France). Tumor growth was followed through ultrasound imaging using a VEVO2100 (Visualsonics).

Shear wave elastography.
The mice were anesthetized with 2% isoflurane and their body temperature was maintained at physiological level using a heating plate. B-mode images and SWE images were acquired simultaneously. The B-mode image allowed us manually determine the region of interest (ROI) corresponding to the tumor contours. SWE mode was performed using the penetration mode with a color scale ranging from 0 (transparent) to 40 kPa (red), arbitrarily chosen in the beginning of the study according to the expected stiffness values. The area, the diameter and a set of stiffness values (mean, minimum, maximum and standard deviation) were recorded for the ROI as previously defined. SWE images were also analyzed using an in-house MATLAB code to recover the stiffness map. T cells were imaged with a DM500B upright microscope equipped with an upright spinning disk confocal microscope (Leica) equipped with a 37 °C thermostatic chamber. For dynamic imaging, tumor slices were secured with a stainless steel slice anchor (Warner Instruments) and perfused at a rate of 0.8 mL/min with a solution of RPMI without Phenol Red, bubbled with 95% O2 and 5% CO2.

Imaging of CD8 T cell motility
Ten minutes later, images from a first microscopic field were acquired with a 25× water immersion

Flow cytometry
The anti-mouse antibodies used were the following: