Immunotherapy of triple-negative breast cancer with cathepsin D-targeting antibodies

Background Triple-negative breast cancer (TNBC) treatment is currently restricted to chemotherapy. Hence, tumor-specific molecular targets and/or alternative therapeutic strategies for TNBC are urgently needed. Immunotherapy is emerging as an exciting treatment option for TNBC patients. The aspartic protease cathepsin D (cath-D), a marker of poor prognosis in breast cancer (BC), is overproduced and hypersecreted by human BC cells. This study explores whether cath-D is a tumor cell-associated extracellular biomarker and a potent target for antibody-based therapy in TNBC. Methods Cath-D prognostic value and localization was evaluated by transcriptomics, proteomics and immunohistochemistry in TNBC. First-in-class anti-cath-D human scFv fragments binding to both human and mouse cath-D were generated using phage display and cloned in the human IgG1 λ format (F1 and E2). Anti-cath-D antibody biodistribution, antitumor efficacy and in vivo underlying mechanisms were investigated in TNBC MDA-MB-231 tumor xenografts in nude mice. Antitumor effect was further assessed in TNBC patient-derived xenografts (PDXs). Results High CTSD mRNA levels correlated with shorter recurrence-free survival in TNBC, and extracellular cath-D was detected in the tumor microenvironment, but not in matched normal breast stroma. Anti-cath-D F1 and E2 antibodies accumulated in TNBC MDA-MB-231 tumor xenografts, inhibited tumor growth and improved mice survival without apparent toxicity. The Fc function of F1, the best antibody candidate, was essential for maximal tumor inhibition in the MDA-MB-231 model. Mechanistically, F1 antitumor response was triggered through natural killer cell activation via IL-15 upregulation, associated with granzyme B and perforin production, and the release of antitumor IFNγ cytokine. The F1 antibody also prevented the tumor recruitment of immunosuppressive tumor-associated macrophages M2 and myeloid-derived suppressor cells, a specific effect associated with a less immunosuppressive tumor microenvironment highlighted by TGFβ decrease. Finally, the antibody F1 inhibited tumor growth of two TNBC PDXs, isolated from patients resistant or not to neo-adjuvant chemotherapy. Conclusion Cath-D is a tumor-specific extracellular target in TNBC suitable for antibody-based therapy. Immunomodulatory antibody-based strategy against cath-D is a promising immunotherapy to treat patients with TNBC. Electronic supplementary material The online version of this article (10.1186/s40425-019-0498-z) contains supplementary material, which is available to authorized users.


(Continued from previous page)
Results: High CTSD mRNA levels correlated with shorter recurrence-free survival in TNBC, and extracellular cath-D was detected in the tumor microenvironment, but not in matched normal breast stroma. Anti-cath-D F1 and E2 antibodies accumulated in TNBC MDA-MB-231 tumor xenografts, inhibited tumor growth and improved mice survival without apparent toxicity. The Fc function of F1, the best antibody candidate, was essential for maximal tumor inhibition in the MDA-MB-231 model. Mechanistically, F1 antitumor response was triggered through natural killer cell activation via IL-15 upregulation, associated with granzyme B and perforin production, and the release of antitumor IFNγ cytokine. The F1 antibody also prevented the tumor recruitment of immunosuppressive tumor-associated macrophages M2 and myeloid-derived suppressor cells, a specific effect associated with a less immunosuppressive tumor microenvironment highlighted by TGFβ decrease. Finally, the antibody F1 inhibited tumor growth of two TNBC PDXs, isolated from patients resistant or not to neo-adjuvant chemotherapy.
Conclusion: Cath-D is a tumor-specific extracellular target in TNBC suitable for antibody-based therapy. Immunomodulatory antibody-based strategy against cath-D is a promising immunotherapy to treat patients with TNBC.
Keywords: TNBC, Human antibody-based therapy, Immunomodulation, Protease, Tumor microenvironment, Phage display Background Breast cancer (BC) is one of the leading causes of death in women in developed countries. Triple-negative breast cancer (TNBC), defined by the absence of estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER-2) overexpression and/or amplification, accounts for 15-20% of all BC cases [1]. Chemotherapy is the primary systemic treatment, but resistance to this treatment is common [1]. Hence, tumor-specific molecular targets and/or alternative therapeutic strategies for TNBC are urgently needed. With the discovery of antigens specifically expressed in TNBC cells and the developing technology of monoclonal antibodies, immunotherapy is emerging as a novel promising option for TNBC [2].
Human cathepsin D (cath-D) is a ubiquitous, lysosomal, aspartic endoproteinase that is proteolytically active at low pH. Cath-D is overproduced and abundantly secreted by human epithelial BC cells [3] with expression levels in BC correlating with poor prognosis [4,5]. Cath-D affects both cancer and stromal cells in the breast tumor microenvironment by increasing BC cell proliferation [3,6,7], fibroblast outgrowth [8,9], tumor angiogenesis [10,11], tumor growth and metastasis [6]. Human cath-D is synthesized as a 52-kDa precursor that is converted into an active 48-kDa single-chain intermediate in the endosomes, and then into a fully active mature form, composed of a 34-kDa heavy chain and a 14-kDa light chain, in the lysosomes. Its catalytic site includes two critical aspartic residues, residue 33 on the 14-kDa chain and residue 231 on the 34-kDa chain.
Cath-D deficiency in humans is associated with neuronal ceroid lipofuscinosis, one amongst the most common pediatric neurodegenerative lysosomal storage diseases, indicating its non-redundant essential role in protein catabolism and cellular homeostasis maintenance [22]. Consequently concomitant inhibition of intracellular and extracellular cath-D with cell-permeable chemical drugs [23] could be toxic. Therefore, we hypothesized that human antibodies targeting extracellular cath-D, which is massively secreted in BC, could circumvent this toxicity and be better therapeutic agents.
Here, we first validated the potential value of cath-D as a tumor-specific extracellular target in TNBC and its suitability for antibody-based therapy. Then, we generated two human anti-cath-D scFv fragments cloned in the human IgG1 λ format (F1 and E2) that efficiently bind to human and mouse cath-D, even at the acidic pH of the TNBC microenvironment. F1 and E2 accumulated in TNBC MDA-MB-231 tumor xenografts, inhibited tumor growth and improved mice survival without apparent toxicity. Using this xenograft model, we found that the Fc function of F1 was essential for maximal tumor inhibition. Mechanistically, F1-based therapy triggered an antitumor response via natural killer (NK) cell activation, and antitumor cytokine production. Furthermore, F1 treatment prevented the recruitment of tumor-associated macrophages (TAMs) and myeloid-derived suppressor cells (MDSCs) within the tumor, a specific effect associated with a less immunosuppressive tumor microenvironment. Finally, F1 inhibited tumor growth of TNBC patient-derived xenografts (PDXs). This preclinical proof-of-concept study validates the feasibility and efficacy of an anti-cath-D immunomodulatory antibody-based strategy to treat patients with TNBC.

Cell lines, ELISA, immunoprecipitation and western blotting
The MDA-MB-231 cell line was previously described [6]. Cells were cultured in DMEM with 10% fetal calf serum (FCS, GibcoBRL). To produce conditioned medium, cells were grown to 90% confluence in DMEM medium with 10% FCS, and conditioned medium was centrifuged at 800 x g for 10 min. For sandwich ELISA, 96-well plates were coated with M2E8 antibody in PBS (500 ng/well) at 4°C overnight. After blocking non-specific sites with PBST/1% BSA, conditioned medium was added at 4°C for 2 h. After washes in PBST, serial dilutions of F1 or E2 were added at 4°C for 2 h and interaction revealed with an anti-human Fc antibody conjugated to HRP (1/2000; 355 ng/well). Cath-D was quantified in TNBC cytosols by sandwich ELISA, as described above, after coating with the D7E3 antibody in PBS (200 ng/well) and with the M1G8 antibody conjugated to HRP (1/80), and using recombinant pro-cath-D (1.25-15 ng/ml) for reference [6]. TNBC cytosols were previously prepared and frozen [25]. GST-cath-D fusion proteins were produced in the E. coli B strain BL21 as described [9]. The resulting proteins were separated on 12% SDS-PAGE and analyzed by immunoblotting.

Study approval
Mouse experiments were performed in compliance with the French regulations and ethical guidelines for experimental animal studies in an accredited establishment (Agreement No. D3417227). The study approval for PDXs was previously published [26]. For TMA, TNBC samples were provided by the biological resource center (Biobank number BB-0033-00059) after approval by the Montpellier Cancer Institute Institutional Review Board, following the Ethics and Legal national French dispositions for the patients' information and consent. For BC cytosols, patient samples were processed according to the French Public Health Code (law n°2004-800, articles L. 1243-4 and R. 1243-61), and the biological resources center has been authorized (authorization number: AC-2008-700; Val d' Aurelle, ICM, Montpellier) to deliver human samples for scientific research. All patients were informed before surgery that their surgical specimens might be used for research purposes.

In vivo studies
MDA-MB-231 cells (2× 10 6 ; mixed 1:1 with Matrigel) were injected subcutaneously in 6-week-old female athymic mice (Foxn1 nu , ENVIGO). When tumors reached a volume of about 50 mm 3 , tumor-bearing mice were randomized and treated with F1 (15 mg/kg), E2 (15 mg/kg), rituximab (15 mg/kg), or NaCl by intraperitoneal injection 3 times per week. Tumors were measured using a caliper and volume was calculated using the formula V = (tumor length × tumor width × tumor depth)/2, until the tumor volume reached 2000 mm 3 . For PDX models, approximately 5 × 5 × 5 mm of B1995 and B3977 tumor fragments were transplanted in the inter-scapular fat pads of in 6-week-old female Foxn1 nu mice. When tumor volume reached a volume of about 150 mm 3 , mice were randomized in two treatments groups: F1 (15 mg/kg) or saline solution by intraperitoneal injection 3 times per week. Tumor volumes were measured as described above.

Immunohistochemistry
For cath-D immunostaining, TNBC TMA and PDX primary tumor sections were incubated with anti-cath-D mouse antibody (clone C-5) at 0.4 μg/ml for 20 min after heat-induced antigen retrieval using the PTLink pre-treatment (Dako) and the High pH Buffer (Dako) and endogenous peroxidase quenching with Flex Peroxidase Block (Dako). After two rinses in EnVision™ Flex Wash buffer (Dako), sections were incubated with a HRP-labeled polymer coupled to secondary anti-mouse antibody (Flex® system, Dako) for 20 min, followed by 3,3′-diaminobenzidine as chromogen. Sections were counterstained with Flex Hematoxylin (Dako) and mounted after dehydration. Sections were analyzed independently by two experienced pathologists, both blinded to the tumor characteristics and patient outcomes at the time of scoring. In tumor and normal epithelial breast cells with peripheral membrane labeling, cath-D signal was scored as absent (0%), low (< 20%), moderate (20-50%), or high (> 50%). Extracellular granulations observed in the stroma were considered as extracellular cath-D staining. Extracellular stromal cath-D signal was defined as: absent (0%), low (< 20%), moderate (20-50%), or (high > 50%). For IHC of MDA-MB-231 xenografts, tumor samples were collected and fixed in 10% neutral buffered formalin for 24 h, dehydrated, and embedded in paraffin. For F4/80 immunostaining, xenograft sections (4-μm thick) were incubated with an anti-F4/80 antibody for 30 min, followed by a rabbit anti-rat antibody (Thermo Scientific, 31,218) before the Envision® system (Dako) as described above. F4/80 staining images were digitalized with the NanoZoomer slide scanner (Hamamatsu) and analyzed with the Aperio Imagescope software.

Homology modeling and docking
Homology models were built using Modeller [27]. The heavy and light chain (VH and VL) were modeled separately, using as template the closest homolog with the same CDR length. VH and VL models were then reassembled based on the relative orientation in the template used for VH modeling. Docking of each molecular model on cath-D was made using PRIOR [28]. Figures were prepared using the PyMOL Molecular Graphics System (Version 2.0 Schrödinger, LLC).

Gene expression data analysis
Recurrence-free survival with a 10-year follow-up was calculated using the on-line Kmplot tool accessed on October 22, 2017 with the 200766_at Affymetrix probe ( [29], www. http://kmplot.com). Analysis was restricted to the 255 patients with TNBC present in the database at this date and with the best cut-off option. The cut-off value was 1919 with a probe range from 50 to 6518. The group with high cath-D mRNA expression at the time 0 represented 38% of the population. Differences were evaluated with the Log-rank test.

Quantitative RT-PCR
Reverse transcription of total RNA was performed at 37°C using the Moloney murine leukemia virus reverse transcriptase (Invitrogen, Carlsbad, CA) and random hexanucleotide primers (Promega, Madison, WI). Real-time quantitative PCR analyses were performed on a Light Cycler 480 SYBR Green I master and a Light Cycler 480 apparatus (both from Roche Diagnostics, Indianapolis, IN). The PCR product integrity was verified by melting curve analysis. Quantification data were normalized to the amplification data for the reference gene encoding ribosomal protein S9 (RPS9). The sequences of the primers for IL-15, GZMB, PRF1, IFNγ, CD206, F4/ 80, TGFβ, and RPS9 are presented (Additional file 1: Table S1).

Isolation of tumor-infiltrating cells and FACs analysis
Tumors were digested with a mixture of collagenase IV (1 mg/ml) (Sigma) and DNase I (200 U/ml) (Sigma) in Hank's Balanced Salt Solution (HBSS) containing 2% FCS at 37°C for three incubations of 15 min/each. The mixtures were then mechanically separated using the Gentle MACs procedure. After digestion, tumor suspensions were passed through a 70 μm nylon cell strainer, centrifuged and resuspended in FACS buffer (PBS pH 7.2, 1% FBS, 2 mM EDTA and 0.02% sodium azide). Cells were blocked with FACS buffer containing 1% (v/v) of Fc Block

Statistical analysis
A linear mixed regression model was used to determine the relationship between tumor growth and number of days after xenograft. The variables included in the fixed part of the model were the number of days post-graft and the treatment group; their interaction was also evaluated. Random intercepts and random slopes were included to take into account the time effect. The model coefficients were estimated by maximum likelihood. A survival analysis was conducted, and the event considered was a tumor volume of 2000 mm 3 . Survival rates were estimated using the Kaplan-Meier method and survival curves were compared with the Log-rank test. Statistical analysis was conducted with the STATA 13.0 software. The Student's t test was used to evaluate difference. Statistical significance was set at the 0.05 level.

Cath-D within the tumor microenvironment is eligible for antibody-mediated targeted therapy in TNBC patients
First, we investigated the clinical significance of the expression of CTSD (the gene encoding cath-D) in a cohort of 255 patients with TNBC using an online survival analysis [29]. High CTSD mRNA level was significantly associated with shorter recurrence-free survival (HR = 1.65 [1.08-2.53]; p = 0.019) (Fig. 1a), suggesting that cath-D overexpression could be used as a predictive marker of poor TNBC prognosis. Then, we assessed cath-D expression in cytosols of primary BC samples (HR + /HER2 + ; HR − /HER2 + ; HR + /HER2 − , HR − /HER − ) by cytosolic assay. Our results revealed a mean cath-D level of 43.5 pmol/mg protein in the TNBC (HR − /HER − ) subtype that was not significantly different from that of the other BC subtypes (Additional file 2: Figure S1), and was in the range of the cut-off values reported in clinical studies on all combined BC subtypes [4,5]. Next, to assess whether cath-D in TNBC was an accessible molecular target for anti-cath-D antibodies, we re-analyzed cath-D status in previously published datasets used for biotin-based affinity isolation and proteomic analysis of accessible protein biomarkers in human BC tissues [30] and (Additional file 3: Figure S2). We found that extracellular and/or membrane-associated cath-D could be detected only in the TNBC tumor sample and not in the adjacent normal breast tissue (Fig. 1b). We validated these proteomic data by anti-cath-D immunohistochemistry (IHC) analysis of a Tissue Micro-Array (TMA) that included 123 TNBC samples (Additional file 4: Figure S3). We detected extracellular cath-D in the microenvironment of 98% of TNBC samples ( Fig. 1c and d), and membrane-associated cath-D at the cancer cell surface of 85.7% of samples ( Fig. 1e and f). Conversely, extracellular and membrane-associated cath-D was not detectable in normal breast tissues ( Fig. 1g and h). Together with the previously published data, our results show that cath-D is a tumor cell-associated extracellular biomarker and strongly suggest that it could be a good candidate for antibody-based therapy in TNBC.

Selection of novel anti-human cath-D scFv fragments by phage display
For future potential clinical application especially in patients with TNBC, we decided to engineer a collection of novel, fully human, anti-cath-D antibodies. For this purpose, we probed the phage antibody expression library Husc I [31,32] with recombinant human 34 + 14-kDa cath-D, and isolated polyclonal human antibodies in scFv format showing specific binding to immobilized cath-D by ELISA. After enrichment by four rounds of bio-panning (Additional file 5: Figure S4A), we selected five monoclonal antibodies based on their binding to recombinant human 52-kDa pro-cath-D and 34 + 14-kDa mature cath-D (Additional file 5: Figure S4B). We purified these five his-tagged scFv fragments by affinity chromatography (Additional file 5: Figure S4C), and determined by ELISA that the purified antibodies still bound to secreted human 52-kDa pro-cath-D and cellular cath-D (Additional file 5: Figure S4D) from MDA-MB-231 cells (cell line derived from an invasive ductal cell carcinoma that represents one of the most common TNBC models). These scFv fragments also recognized mouse cellular cath-D (81.1% of identity with human cath-D) from MEF cells (Additional file 5: Figure S4E). We then used the three scFv antibodies (F1, E2 and E12 scFv) with the highest binding to human and mouse cath-D to produce fully human IgG1 λ (F1, E2, E12).

Generation of anti-cath-D human antibodies
Sandwich ELISA using pro-cath-D secreted from MDA-MB-231 cells showed that F1 (Fig. 2a, left panel) and E2 (Fig. 2a, right panel) retained good binding capacities (EC 50 = 0.2 nM and 1.2 nM, respectively). Conversely, E12 in the IgG1 format lost its binding activity (data not shown). Moreover, F1 and E2 binding to pro-cath-D was comparable at pH values from 7.5 to 5.5 (Fig. 2b), suggesting that they are active also in the highly acidic tumor microenvironment. We also confirmed F1 and E2 good selectivity towards pro-cath-D compared with other aspartic enzymes, such as pro-cathepsin E, pepsinogen A and pepsinogen C (Additional file 6: Figure S5). We next characterized the cath-D epitope recognized by F1 and E2. Molecular docking performed on the three-dimensional structure of mature cath-D (PDB ID 1LYA) [33] strongly suggested that F1 and E2 scFv interact mainly with the 34-kDa cath-D chain (in red) (Fig. 2c, top panels). Moreover, the third complementary determining region of the heavy chain (CDRH3) of both F1 and E2 scFv, which is crucial for antibody specificity, might protrude into the proteinase active site (Fig. 2c, bottom panels). By competitive ELISA, we confirmed that F1 and E2 epitopes overlapped (Fig. 2d). Finally, using GST-cath-D fusion fragments, we showed that both F1 and E2 immunoprecipitated the 52-, 48-and 34-kDa forms of GST-cath-D, but not the 4-kDa GST-cath-D pro-fragment and the 14-kDa light chain GST-cath-D (Fig. 2e, left panel). These results indicated that the cath-D epitopes of F1 and E2 are located mainly on the 34-kDa part of the protein (Fig. 2e, right  panel).  Fig. 3a).
The bio-distribution profiles confirmed that 177 Lu-F1 and 177 Lu-E2 gradually accumulated in the tumors from 24 h and up to 96 h (Additional file 7: Figure S6). The percentage (mean ± SD) of injected activity/g tissue detected in tumors (%IA/g) at 72 h was 8.2% ± 4.3% for F1 and 8.1% ± 2.2% for E2 (Additional file 7: Figure S6). Moreover, at 72 h, 177 Lu-F1 and 177 Lu-E2 were present also in blood (6.4 ± 2.1 and 11.3% ± 6.1%, respectively), and liver (10.5 ± 5.2 and 10.7% ± 4.7%, respectively). However, their concentration in blood and liver decreased rapidly due to physiological elimination. These results indicate that F1 and E2 localize and accumulate in TNBC MDA-MB-231 xenografts. Treatment with F1 or E2 significantly delayed tumor growth compared with control ( Fig. 3b; P < 0.001 for F1, P = 0.002 for E2). At day 55, tumor volume was reduced by 58% in the F1 (P = 0.0005) and by 49% (P = 0.0026) in the E2 group compared with control (Fig. 3c). Moreover, the overall survival rate, reflected by a tumor volume inferior to 2000 mm 3 , was significantly longer in mice treated with F1 or E2 than in controls, with a median survival of 72 and 64 days for the F1 and E2 groups respectively, compared with 57 days for control animals ( Fig. 3d; Kaplan-Meier survival analysis, P = 0.0005 for F1, P = 0.0016 for E2). These results show that anti-cath-D human antibodies as monotherapy delay very efficiently tumor growth in nude mice xenografted with MDA-MB-231 cells.

Anti-cath-D antibody-based therapy prevents macrophage recruitment within MDA-MB-231 tumor xenografts
To further investigate the in vivo mechanisms underlying the antitumor effect of F1 and E2, we treated nude mice xenografted with MDA-MB-231 cells with F1, E2 or the anti-human CD20 IgG1 rituximab, as negative isotype control (same schedule as before), and then sacrificed all mice at the treatment end. F1 and E2 led to a significant inhibition of tumor growth compared with rituximab (P = 0.001 for F1, P = 0.002 for E2) (Fig. 4a). At the end of the experiment (day 44), tumor volume was reduced by 76% in the F1 group (P = 0.0001) and by 63% (P = 0.0012) in the E2 group, compared with the rituximab group (Fig. 4b). Moreover, although F1 and E2 cross-react with mouse cath-D, mice treated with the anti-cath-D antibodies gained weight and displayed normal activities (Additional file 8: Figure S7), suggesting minimal off-target effects for these human antibodies. Then, we investigated the effect of F1 and E2 monotherapy on tumor cell proliferation, apoptosis, and angiogenesis by IHC. Ki67, a marker of proliferating cells (Additional file 9: Figure S8A and B), activated caspase 3, a marker of apoptosis (Additional file 9: Figure S8C and D), and the angiogenesis marker CD31 (Additional file 9: Figure S8E and F) were similarly expressed in tumors from the three groups of mice. As antibody-based immunotherapy is often associated with immune modulation of the tumor microenvironment [34], we assessed the impact of anti-cath-D antibodies on tumor-infiltrating immune cells, particularly on myeloid cells that are present in Foxn1 nu nude mice. Staining with the anti-macrophage F4/80 antibody revealed that macrophage infiltration in the tumor core was reduced by 64.8% in the F1 and by 41% in the E2 group, compared with the rituximab group The anti-cath-D antibody F1 prevents M2-like macrophages and MDSC recruitment, leading to a less immunosuppressive tumor microenvironment in MDA-

MB-231 xenografts
As the immunomodulatory effect of antibody-based therapy could depend on Fc-mediated mechanisms [35], we engineered an aglycosylated Fc-silent version of the F1 antibody (F1Fc) in which the mutation N297A prevents binding to FcγRs [36]. We first confirmed that F1Fc binding to cath-D was comparable to that of F1 (Additional file 10: Figure S9). We then treated mice harboring MDA-MB-231 tumor cell xenografts with F1Fc, F1 or rituximab (CTRL) as before. F1 treatment significantly reduced tumor growth compared with rituximab ( Fig. 5a; P < 0.001). Conversely, F1Fc effect on tumor growth was reduced compared with F1 (Fig. 5a). At the end of the experiment (day 54), tumor volume was reduced by 63.1% (P = 0.01) in the F1 group and only by 32.9% (not significant) in the F1Fc group compared with the rituximab group (Fig. 5b). Thus, the Fc effector functions of F1 are essential for We next analyzed tumor immune infiltrates at day 54 by FACS analysis with a specific focus on TAMs and MDSCs, associated with tumor progression and relapse in BC [37,38]. In agreement with the previous IHC results ( Fig. 4c and d), the percentage of F4/80 + CD11b + macrophages within the immune CD45 + cell population was significantly decreased by 67% in F1-treated animals (P = 0.044 compared with the rituximab group) and only by 33% in the F1Fc-treated group (not significant) (Fig.  5c). Moreover, linear regression analysis showed that the percentage of macrophages was significantly correlated with tumor volume in all animals (three treatment groups together) (R 2 = 0.5425, P < 0.0001) (Fig. 5d), suggesting that in this model, tumor progression was associated with macrophage enrichment and that the F1 antibody prevented their infiltration. In many tumors including BC, TAMs are M2 polarized, which is associated with protumorigenic functions [37,39]. At day 54, the expression of CD206 mRNA, a M2-associated marker [40], was significantly downregulated by 56.6% (P = 0.05) and 62.9% (P = 0.04) in MDA-MB-231 tumor xenograft RNA samples from the F1-and F1Fc-treated group, respectively, compared with control (rituximab) (Fig. 5e). This suggests that anti-cath-D antibody monotherapy prevented tumor infiltration by M2 macrophages and that this could have contributed to limit tumor growth.
In addition, the percentage of Gr1 + CD11b + MDSCs within the immune CD45 + cell population also was significantly decreased by 53.4% (P = 0.008) in F1-treated mice and by 29.6% in F1Fc-treated mice (not significant) compared with control (rituximab) (Fig. 5f). The percentage of tumor-infiltrating MDSCs was positively correlated with tumor volume in the whole population (three groups together) ( Fig. 5g; P = 0.0125). Because of the changes of TAMs and MDSCs, F1 treatment may alter immunosuppressive factors in the tumor microenvironment. Indeed, mRNA expression of the inhibitory cytokine transforming growth factor β (TGFβ) was reduced by 51% in tumors from F1-treated mice (P = 0.0099 compared with the rituximab control) and by 30.6% in the F1Fc group (not significant) (Fig. 5h). This strengthened the effect of anti-cath-D antibody therapy on immunosuppressive M2 macrophages and MDSCs. Our data highlight the strong impact of anti-cath-D antibody therapy on the tumor immune microenvironment, leading to a less immunosuppressive microenvironment in MDA-MB-231 xenografts.
The anti-cath-D antibody F1 antitumor response is triggered via NK cell activation NK cells are needed for the efficacy of antibody-based immunotherapies by triggering antibody-dependent cell-mediated cytotoxicity [41]. To determine the potential implication of NK cells in anti-cath-D antibody therapy, we quantified by FACS analysis the CD49b + CD11b + NK cell population in tumors at the end of treatment (day 54) and found that it was comparable in the F1-, F1Fc-and rituximab-treated groups (Fig. 5i). RT-qPCR analysis of the expression of IL-15, a cytokine associated with NK cell activation [42], showed that it was upregulated (up to 209%, P = 0.0013 compared with rituximab) in the F1 group, but not in the F1Fc-treated group (P = 0.0127 compared with F1) (Fig. 5j). This (See figure on previous page.) Fig. 5 Anti-cath-D antibody-based therapy prevents M2-like macrophage and MDSC recruitment, and triggers antitumor response via NK cell activation in MDA-MB-231 xenografts. a Tumor growth. Nude mice bearing MDA-MB-231 tumors of 50 mm 3 were treated with F1 (n = 9), F1Fc (n = 8), or rituximab (CTRL; n = 9) (15 mg/kg) for 35 days. At day 54, mice were sacrificed. *, P < 0.001 for F1 versus CTRL; P = 0.077 for F1Fc versus CTRL; P = 0.069 for F1 versus F1Fc (mixed-effects ML regression test). b Mean tumor volumes at day 54. Mean ± SEM; *, P = 0.011 for F1 versus CTRL; P = 0.231 for F1Fc versus CTRL, P = 0.189 for F1 versus F1Fc (Student's t-test). c TAM recruitment. The percentage of F4/80 + CD11b + TAMs was quantified by FACS and expressed relative to all CD45 + immune cells (n = 9 for CTRL; n = 9 for F1; n = 8 for F1Fc); *, P = 0.044 for F1 versus CTRL; P = 0.3 for F1Fc versus CTRL (Student's t-test). d Linear regression analysis of TAM and tumor volumes. R 2 = 0.5425; ***, P < 0.0001; n = 26. e Quantification of CD206 mRNA expression. Total RNA was extracted from MDA-MB-231 tumor xenografts at the end of treatment, and CD206 expression analyzed by RT-qPCR and shown relative to F4/80 (n = 9 for CTRL; n = 9 for F1; n = 8 for F1Fc); P = 0.05 for F1 versus CTRL; P = 0.04 for F1Fc versus CTRL (Student's t-test). f MDSC recruitment. The percentage of Gr1 + CD11b + MDSCs was quantified by FACS analysis and expressed relative to all CD45 + cells (n = 9 for CTRL; n = 9 for F1; n = 8 for F1Fc); **, P = 0.008 for F1 versus CTRL; P = 0.079 for F1Fc versus CTRL (Student's ttest). g Linear regression analysis of MDSC and tumor volumes. R 2 = 0.23315; *, P = 0.0125; n = 26. h Quantification of TGFβ mRNA expression. Total RNA was extracted from MDA-MB-231 tumor cell xenografts at the end of treatment and TGFβ expression analyzed by RT-qPCR. Data are relative to RPS9 expression (n = 9 for CTRL; n = 9 for F1; n = 8 for F1Fc); **, P = 0.009 for F1 versus CTRL; P = 0.1 for F1Fc versus CTRL (Student's t-test). i NK recruitment. The percentage of CD49b + CD11b + NK cells was quantified by FACS and expressed relative to all CD45 + cells (mean ± SEM; n = 9 for rituximab (CTRL); n = 9 for F1; n = 8 for F1Fc); P = 0.7 for F1 versus CTRL; P = 0.8 for F1Fc versus CTRL; P = 0.8 for F1 versus F1Fc (Student's t-test). suggests a causal relationship between the F1 antitumor response and NK cell activation. In agreement, IL-15 mRNA level was inversely correlated with tumor volume in the entire population (three groups together) by linear regression analysis ( Fig. 5k; P = 0.0013). We then quantified granzyme B (GZMB) and perforin 1 (PRF1) mRNA levels, as a read-out of NK cell activity. GZMB was strongly upregulated (up to 220%) in the F1 group (P = 0.0002 compared with rituximab). Although significant, the up-regulation remained modest in the F1Fc group compared with the control group and was significantly reduced compared with the F1-treated group ( Fig. 5l; P = 0.0076). Similarly, PRF1 expression was increased by 500% in the F1 group compared with control (P = 0.033) and slightly less in the F1Fc group (Fig. 5m). Finally, the mRNA expression of the antitumor cytokine IFNγ was upregulated by 494.8% in the F1 group compared with control ( Fig. 5n; P < 0.0001). This upregulation was significantly reduced in the F1Fc-treated group compared with the F1 group (Fig. 5 n; P = 0.0078). Altogether, our results strongly suggest that the antitumor response of the anti-cath-D antibody F1 in MDA-MB-231 xenografts is in part triggered by Fc-dependent mechanisms via NK cell activation through IL-15 upregulation, associated with granzyme B and perforin production and the release of IFNγ.
The anti-cath-D antibody F1 inhibits growth of patientderived xenografts of TNBC Finally, we tested F1 effect in mice harboring PDXs of TNBC [26]. First, quantification by sandwich ELISA in whole cytosolic extracts of five representative TNBC PDXs showed that cath-D concentration varied from 18 to 77 pmol/mg of total protein (Fig. 6a). These values were in the same range as those detected in whole cytosolic extracts prepared from MDA-MB-231 tumor xenografts (Fig. 6a), and from 40 TNBC samples (Fig. 6b). Immunostaining of the B1995 and B3977 primary tumors with an anti-cath-D antibody confirmed that cath-D was detected in tumor cells and microenvironment (Fig. 6c), as previously observed with the TNBC TMA ( Fig. 1c and e). These results indicate these PDX models are representative of the disease, at least concerning cath-D expression. We then engrafted athymic nude mice with PDX B1995 or PDX B3977, the two PDXs showing the fastest growth in nude mice (average passage duration for the first three passages: 46 days for B1995 and 42 days for B3977) (Fig. 6d). Tumor volume increase was significantly slowed down in mice treated with F1 compared with control (Fig. 6e).

Discussion
Here we described the discovery, characterization, and preclinical development of F1, a fully human anti-cath-D IgG1 antibody isolated from a human scFv phage display library. Our study indicates the anti-cath-D human antibody F1 is a candidate drug that could be translated into the clinic for the treatment of patients with TNBC. Our findings demonstrated F1 therapeutic efficacy in three different TNBC models (MDA-MB-231 cell xenografts and two TNBC PDXs). Moreover, they also confirmed that tumor-specific extracellular non-receptor oncoproteins are reliable molecular targets for antibody-based cancer therapy. Indeed, although monoclonal antibodies have a remarkable efficacy as anti-cancer therapeutics [34], only a limited number of soluble targets have been explored so far, and antibody therapy is mainly focused on receptor antigens on cancer cells as "druggable" targets. In BC, the most relevant example is trastuzumab (Herceptin®) to treat HER2 + tumors.
Our data show that cath-D is a pertinent target for antibody-based therapy in TNBC. A recent study reported that cath-D is overexpressed in 71.5% of the 504 TNBC analyzed and proposed a prognostic model for TNBC outcome based on node status, cath-D expression and Ki67 index [43]. Here, we found that high CTSD mRNA expression was significantly associated with shorter recurrence-free survival in a cohort of 255 TNBC samples [29]. Our re-analysis of previously published proteomic data [30] allowed describing the extracellular and cell-surface expression of cath-D in TNBC. Moreover, in the TNBC TMA, we detected extracellular cath-D in the tumor microenvironment in 98.4% of samples and at the surface of cancer cells in 85.4% of samples. Conversely, in normal breast, we observed cath-D expression only inside luminal breast cells. Altogether, our data strongly suggest that extracellular cath-D could be considered as a therapeutic target and a biomarker in TNBC.
To engineer anti-cath-D antibodies, we used the antibody phage display technology, an efficient methodology for the isolation of human monoclonal antibodies for therapeutic applications [34]. Importantly, the human scFv antibody format is suitable for incorporation of the binding specificity into therapeutic proteins and can be reformatted into intact IgG1, as shown here. Anti-human cath-D mouse monoclonal antibodies have been produced also using the hybridoma strategy [24]. However, their therapeutic effect was not investigated. Vetvicka's group also developed antibodies against the amino acid stretch (aa 27-44) within the human cath-D pro-peptide. These antibodies inhibit human BC cell xenograft growth in nude mice [44].
Here, we validated the concept of targeting extracellular cath-D in the TNBC microenvironment using anti-cath-D human antibodies in the IgG1 format that allows future clinical applications. In vivo, F1 and E2 inhibited tumor growth and improved the overall survival without apparent toxicity. Moreover, F1, our most potent antibody, inhibited growth also of two TNBC PDXs (B1995, and B3977). PDX B1995 was isolated from a patient with TNBC after neo-adjuvant therapy, whereas PDX B3977 was isolated from a patient with TNBC resistant to neo-adjuvant chemotherapy. Thus, our findings strongly suggest that F1 may represent a new therapeutic strategy to treat patient with TNBC resistant to chemotherapy.
Monoclonal antibodies mediate tumor cell inhibition through multiple mechanisms of action, including direct target inhibition through antigen-binding (Fab) fragment engagement and immune cell modulation through their constant region Fc. Using the F1Fc mutant we showed that F1 Fc function is essential for maximal tumor growth inhibition in athymic nude mice xenografted with TNBC MDA-MB-231 cells. The IgG1 Fc portion can recruit and activate components of the complement system and innate immune effector cells, resulting in destruction of antibody-targeted cancer cells through complement-dependent cytotoxicity, antibody-dependent cellular phagocytosis, or antibody-dependent cellular cytotoxicity (ADCC) [34]. Evidence from multiple in vivo models suggests that antibody binding to cellular FcγRs is the major pathway for ADCC, which is classically mediated by NK cells [45]. Here, we demonstrated that F1 leads to NK cell activation via IL-15 upregulation, associated with increased granzyme B and perforin production to eliminate tumor cells, as well as higher levels of the antitumor IFNγ cytokine. Conspicuously, these effects were limited when using the aglycosylated F1Fc antibody in which the mutation N297A prevents binding to FcγRs. Aglycosylated antibodies retain antigen binding, as observed for F1Fc, but show defects in binding to FcγRs, and ADCC induction [36]. Altogether these results indicate that NK cells are implicated in F1 antitumor response, suggesting ADCC occurrence in vivo. ADCC is one of the immunotherapeutic strategies currently under investigation to awake the innate anti-cancer response. Recent studies and clinical trials have highlighted that manipulation of NK cell activation can be used as an immunotherapy in cancer, particularly against tumor metastases [41]. As Fc dependence of antibodies is a sine qua non requirement for their use in the clinic to ensure effectiveness, NK cell activation following F1 antibody treatment is a key finding. We also found that F1 has a remarkable effect on the tumor immune microenvironment. First, it prevented the infiltration of M2-like macrophages in MDA-MB-231 tumors. Earlier studies suggested that TAMs participate in antitumor responses; however, recent evidences indicate that they support tumor progression by blocking the antitumor immunity and by secreting factors that promote angiogenesis and re-activation of the epithelial-to-mesenchymal transition, which enhance metastasis formation [39]. In BC, TAMs are the most abundant inflammatory cells and are typically M2-polarized with suppressive capacity [39] that stems from their enzymatic activities and production of anti-inflammatory cytokines, such as TGFβ [46]. In our TNBC mouse model, M2-polarized macrophages and TGFβ expression were decreased after F1 treatment. High TAM levels have been associated with poorer BC outcomes [37]. Therefore, several strategies are currently under investigation, such as the suppression of TAM recruitment, their depletion, or the switch from the protumor M2 to the antitumor M1 phenotype in patients with TNBC [47]. Our findings showing reduced macrophage infiltration and decreased M2-like macrophages in response to F1 treatment are in line with the ongoing therapeutic strategies.
MDSCs are immature myeloid cells that promote the immunosuppressive tumor microenvironment through multiple mechanisms, including expression of immunosuppressive cytokines, such as TGFβ [48]. Treatment with F1 limited MDSC infiltration in MDA-MB-231 xenografts, and downregulated TGFβ expression. MDSC expansion in tumors is a major mechanism of tumor immune escape [38], and MDSC is considered as a potential cancer biomarker and therapeutic target [48]. Accordingly, disruption of MDSC function can improve the antitumor immune responses and impair tumor growth in mice [38]. Collectively, our data indicate that F1 treatment can specifically modulate myeloid cells in the tumor microenvironment, leading to a less immunosuppressive tumor microenvironment. As TAMs and MDSCs can suppress NK responses, it is possible that their depletion upon F1 therapy favors NK activation and cytotoxic activity, particularly through TGFβ downregulation. Consequently, we suggest that treatment with the anti-cath-D F1 antibody could normalize the tumor microenvironment, by boosting the host antitumor immune response, as illustrated here by NK cell activation. Our data show that F1 Fc effector function has an impact on its antitumor effect through NK cell dependent-pathways. However, we do not know how F1 limits TAM and MDSC infiltration in tumors. As in vitro cath-D can cleave chemokines that are implicated in the immune responses and the recruitment of myeloid cells and lymphocytes [16,17], the Fab part of F1 could affect TAM and MDSC recruitment by modulating chemokine homeostasis.
In this study, to generate clinically relevant TNBC mouse models (MDA-MB-231 cell xenografts and PDXs), we used immunodeficient Foxn1 nu nude mice, instead of severely immunocompromised mice strains with little or no endogenous immune system, such as NOD/SCID, BAL-B/c-RAG2 null , or their derivatives [49]. The use of the more clinically relevant Foxn1 nu mouse model that harbors NK and myeloid cell populations allowed the discovery of F1 immunomodulatory activity. However, this model did not permit to assess F1 effect on lymphocyte populations. Thus, our future studies will focus on humanized mouse models.