The Chemokine CX3CL1 Improves Trastuzumab Efficacy in HER2 Low–Expressing Cancer In Vitro and In Vivo

The mAb trastuzumab, directed against the HER2 oncogene, has led to an unprecedented improvement in the therapy of HER2-positive early and advanced breast and metastatic gastric cancer, paving the way for many other immunotherapies to follow (1). However, despite the fact that the HER2 protein is detectable in the majority of breast cancers, only about 25% of all breast cancers express it at levels that allow successful anti-HER2 treatment. Other cancers that also express HER2 (e.g., colon cancer) do not satisfactorily benefit from an anti-HER2 therapy (2).

Apart from the inhibition of HER2 downstream proliferative signaling pathways, trastuzumab's mechanism of action encompasses labeling of the tumor cells for their recognition and destruction by natural killer (NK) cells by a process called antibody-dependent cellular cytotoxicity (ADCC; refs. 3–5). NK cells trigger specific antitumor T-cell immunity (3). The importance of these cellular immune mechanisms is clinically underscored by studies demonstrating that the number of tumor-infiltrating lymphocytes is predictive of trastuzumab sensitivity (6, 7). Thus, raising the intratumoral accumulation of NK cells and other tumor-suppressive lymphocytes might be a feasible approach to enhance trastuzumab efficacy (8).

The chemokine CX3CL1, the only member of the CX3C subfamily of chemokines, is capable of recruiting NK and T cells to the tumor microenvironment (9). It is first synthesized as a membrane-bound form that allows binding of CX3CR1-positive immune cells to CX3CL1-positive target cells; it is then cleaved, primarily by ADAM proteases (e.g., ADAM17), to exert its chemotactic function as a soluble chemokine (10). These two properties render CX3CL1 a promising molecule to improve trastuzumab-mediated immune responses in HER2-expressing tumors.

However, there are conflicting data on the functional and prognostic relevance of CX3CL1 in cancer. Although most in vivo studies support an antitumor effect through immune activation (11–13), there are also data on the involvement of CX3CL1 in early tumorigenesis, for example, in HER2-positive breast cancer, via EGF pathway transactivation (14). Overexpression of CX3CL1 is associated with enhanced immune infiltration and a better prognosis in several cancer types (15–20). In pancreatic cancer, however, the opposite effect is reported (21).

Apart from its possible role in the very early stages of tumorigenesis, the functional role of CX3CL1 in breast cancer has not yet been explored. We speculated that overexpression of CX3CL1 might be a feasible way to improve the therapeutic success of trastuzumab by enhancing both the recruitment and the binding of NK cells to the trastuzumab-tagged tumor cells. In this study, we demonstrated that CX3CL1 was a protective chemokine in an immunocompetent breast cancer mouse model and acted synergistically with trastuzumab in NK cell–mediated ADCC in vitro. In vivo, CX3CL1 did not further improve the therapeutic success in trastuzumab-sensitive HER2-overexpressing cancer. However, its overexpression alone was sufficient to enable trastuzumab therapy in HER2 low–expressing, originally trastuzumab-resistant cancer. Our results support the idea that CX3CL1 might be a predictive marker of trastuzumab sensitivity in HER2 low–expressing cancers, and that its pharmacologic induction might be a feasible adjuvant to trastuzumab therapy.

Human tumor cell lines: SKBR3 (ATCC: HTB-30 breast), MDA-MB-453 (ATCC: HTB-131 breast), BT474 (ATCC: HTB-20), MCF7 (ATCC: HTB-22 breast), MDA-MB-231 (ATCC: HTB-26 breast), and HT-29 (ATCC: HTB-38 colorectal). Mouse tumor cell lines: 4T1 (ATCC: CRL-2539 breast). Cells were cultured in RPMI medium (cell lines: BT474, MCF7, MDA-MB-231, 4T1) or McCoy's Medium (GIBCO, cell lines: SKBR3, MDA-MB-453, HT-29) provided with 10% [volume for volume (v/v)] heat-inactivated FCS, 0.272 mmol/L l-asparagine, 0.550 mmol/L arginine, and 10 mmol/L HEPES buffer at 37°C and 5% (v/v) CO₂. Cell passaging was performed with 0.025% trypsin/0.01% EDTA in PBS (treatment less than 2 minutes). Cell lines were authenticated by DNA profiling of eight short tandem repeat sequences and were further controlled for the presence of mouse and rat DNA (authentication service DSMZ). The human cell lines were validated before submission of this work. All the cell lines are regularly tested for Mycoplasma contamination via PCR.

Vector DNA (pCMV-6-mCX3CL1, pCMV-6-hCX3CL1, pCMV-6-Entry, Origene) was stably transfected in human and murine breast cancer cells with Lipofectin (Thermo Fisher Scientific). Transfected cells were selected with neomycin (Thermo Fisher Scientific) for 14 days, and positive clones were selected for further expansion and testing. Proliferation of cell clones was tested by seeding 1 × 10⁴ cells in 24-well plates and counting for 4 days. Transient transfection was performed without further neomycin selection. Cells were kept in culture for 24 hours after transfection and were then used for the in vitro experiments.

For the IHC studies, formalin-fixed, paraffin-embedded tumor specimens from 250 patients with early breast cancer treated at the Department of Gynecology and Obstetrics (Technical University of Munich, Munich, Germany) between 2005 and 2012 were used, stored in an interdisciplinary tissue bank (Department of Pathology, Technical University of Munich, Munich, Germany). Patients with distant metastases or after neoadjuvant chemotherapy at the time of surgery were excluded from the study. Detailed patient characteristics are given in Supplementary Table S1. The HER2-low (n = 178) and HER2-overexpressing cohorts (n = 25) were subgroups of this larger cohort. In another cohort, only HER2-overexpressing tumors treated neoadjuvantly with trastuzumab were included (n = 69). The study was approved by the Institutional Review Board of the Technical University of Munich (Munich, Germany; approval 227/17S) in accordance with the Declaration of Helsinki. Written informed consent was obtained from all patients.

Tissue microarrays were generated as described previously (22). Mouse tissue was fixed in 4% paraformaldehyde for 48 hours and paraffin embedded. Both human and murine tissue was processed to 3 μm sections. The slides were deparaffinized by xylene and hydrolyzed by a descending alcohol row (100%–50%). Heat-induced antigen retrieval was performed in citrate buffer (pH 6.0) before endogenous peroxidase activity was blocked with 3% (v/v) H₂O₂ (20 minutes), followed by an additional blocking step with 5% goat serum in antibody dilution solution (ZUC025-500, Zytomed). Samples were then incubated with anti-CX3CL1 (1 μg/mL in antibody dilution solution, clone AF365, R&D Systems), F4/80 antibody (1:300 in antibody dilution solution, clone D2S9R, Cell Signaling Technology), or anti-CD49b (2.5 ng/mL, BioLegend) for 1 hour at room temperature. Detection was performed using the polymer one-step system (Zytomed) and 3,3′-Diaminobenzidine (DAB; Zytomed) as substrate. Slides were counterstained with hematoxylin (Merck). Slides were washed thoroughly with TBST (0.1% Tween 20) between each step. Evaluation of the staining was performed by using a semiquantitative score based on the staining intensity of tumor tissue: absent (0), weak (1+), moderate (2+), or strong (3+; Fig. 1A). Evaluators were blinded to the clinical data. In cases with positive staining, all tumor cells were positive, and therefore, percentage of positivity was not included in our evaluation.

Tumor cell lines (SKBR3, MDA-MB-453, BT474, and HT-29) were seeded in 12-well plates (1.4 × 10⁵ cells) and grown to a confluence of 80%. Afterward, cells were washed with PBS and cultivated in serum-free medium for 24 hours before test reagents were added for another 24 hours. Final concentrations were as follows: TNFα (10 ng/mL, Peprotech), TAPI-2 (7.5 μmol/L, Tocris). For determination of the transfection efficacy, cell supernatants were taken from unstarved, unstimulated cells. For measurement of human CX3CL1, mouse CX3CL1, and human CXCL10, we used the DuoSet ELISA Kits DY365, DY472, and DY266, respectively, according to the manufacturer's recommendations (R&D Systems). Triplicates of 100 μL undiluted supernatant were analyzed (Multiskan FC, Thermo Fisher Scientific). The absolute target protein concentration was calculated based on a predefined recombinant protein standard.

Tumor cells from stimulation experiments were collected and immediately lysed with RIPA buffer [1% Triton X100, complete protease inhibitor cocktail (Roche), 1 mmol/L vanadate, 1 mmol/L glycerol phosphate, 50 mmol/L sodium fluoride, 10 mmol/L sodium pyrophosphate in TBS (pH 7.4), all chemicals from Merck]. The samples were incubated for 20 minutes on ice and then treated with ultrasound (Sonopuls, Bandelin) for 20 seconds. The concentration of total protein was determined using the BCA Protein Assay Kit (Pierce, Thermo Fisher Scientific). Protein samples (30 μg) were separated by a 10% SDS-PAGE and blotted onto a 0.45 μm nitrocellulose membrane by wet blotting. After blocking (5% milk powder in TBST for 1 hour at room temperature), incubation with the primary antibodies anti-CX3CL1 (clone MAB365, R&D Systems, 1 μg/mL), anti–HER-2 (A0485, Dako, 1:1,000), or anti-GAPDH (Millipore, 0.125 μg/mL) in 2.5% milk powder/TBST was performed for 1 hour at room temperature. Secondary anti-goat (Life Technologies) or anti-mouse horseradish peroxidase (HRP)–conjugated antibody (Jackson Immuno Research) was used 1:10,000 for 1 hour at room temperature. Detection was performed using secondary HRP-conjugated antibodies and ECL reagent (Thermo Fisher Scientific) with the Bio-Rad detection system. A thorough washing step with TBST was performed between each step.

NK cells were freshly isolated from a leukapheresis product of three different volunteer donors. CD19⁺ (Miltenyi Biotec) and CD3⁺ (Miltenyi Biotec) cells were separated from the NK cells in a negative selection approach using the cliniMACS system (Miltenyi Biotec). NK-cell enrichment was determined using FACS analysis for CD56 (BD Biosciences) and CD45 (Invitrogen). Purity of NK-cell populations was determined as described before (23, 24), for the NK cells used a purity >80% was achieved.

The NK cell–mediated killing of tumor cells was assessed using a dissociation-enhanced lanthanide fluorescent immunoassay (DELFIA, PerkinElmer) with time-resolved fluorescence detection. NK cells were stimulated for 3 days with IL2 (500 U/mL, Peprotech). Stably or transiently transfected cancer cells, as well as cells treated with 10 ng/mL TNFα or 7.5 μmol/L TAPI-2, were collected, adjusted to 5 × 10⁵ cells, and stained with the 0.5% DELFIA BATDA fluorescent reagent in 200 μL RPMI medium without phenol red. Afterward, the cells were treated with trastuzumab (40 μg/mL, Roche), anti-CX3CL1/goat IgG (20 μg/mL, clone AF365, R&D Systems), and/or soluble CX3CL1 (10 ng/mL, amino acids 1–76, PeproTech) for 30 minutes. NK cells and tumor cells (5 × 10⁴ cells) were coincubated for 4 hours at three different ratios (1:10, 1:5, and 1:2.5) in V-bottom 96-well plates. After incubation, the released DELFIA BATDA was detected with europium solution using a time-resolved fluorometer (Victor Wallac 3). The spontaneous lysis was subtracted, and all the values were normalized to a corresponding maximum lysis control as described by the manufacturer (DELFIA Cell cytotoxicity assay, PerkinElmer).

Eight- to 10-week-old wild-type Balb/cAnNCrl mice (strain 028), SCID mice (CB17/Icr-Prkdc^scid/IcrIcoCrl, strain 236), and SCID Beige mice (CB17.Cg-PrkdcscidLystbg-J/Crl, strain 250) were obtained from Charles River Laboratory. Mice were maintained in a pathogen-free animal facility, and all animal procedures and experiments were approved by the Government of Upper Bavaria (Regierung von Oberbayern) and were in accordance with the institutional guidelines of the Technical University of Munich (Munich, Germany).

Cell lines (4T1, HT-29, MDA-MB-453) were used after confirmation of Cx3cl1 overexpression and negative Mycoplasma testing. Mice were subcutaneously injected with 1 × 10⁴ 4T1 cells, 2 × 10⁶ HT-29, or 2 × 10⁶ MDA-MB-453 cells in PBS into the fourth mammary fat pad. Tumor cells were washed in PBS before injection. The syngeneic 4T1 mouse experiments were terminated 21 days after tumor inoculation. Mice bearing HT-29 or MDA-MB-453 tumors were culled after the tumor diameter reached 1.5 cm. The depletion of NK cells was performed by intraperitoneal injection of 100 μL anti-asialo GM1 (986-10001, FUIFILM Wako Chemicals) or 100 μL corresponding rabbit serum as control twice a week, starting 1 day after tumor inoculation. Treatment with trastuzumab was started once the tumor diameter reached 0.5 cm. The mice were treated with 10 mg/kg trastuzumab i.p. twice a week or vehicle control. In the MDA-MB-453 model, mice were treated for 3 weeks in total. In the HT-29 model, mice were treated until the end of the experiment (tumor size). The tumor diameter was measured using calipers every 72 hours, and the tumor volume was calculated (largest diameter² × smallest diameter × 0.5). Right before the finalization of the animals, blood was taken from the vena facialis, stored on ice for 30 minutes, and centrifuged at 10,000 rcf for 10 minutes. Serum was taken and stored at −80°C for further use. Directly after the mice were euthanized, tumor tissue, spleen, kidney, and heart were removed, and stored in liquid nitrogen and 4% buffered formalin for 48 hours. Tissue for FACS analysis or colony formation was stored in PBS until further use.

Fresh-frozen tumor tissues (20 mg) from mice were mechanically disrupted, and total RNA isolated using the RNeasy Plus Kit (Qiagen). A total of 1 μg of RNA was then used for reverse transcription (Super Script IV, Thermo Fisher Scientific). Quality of the RNA was checked by agarose gel electrophoresis and by NanoDrop (Thermo Fisher Scientific). A total of 210 ng of cDNA were used for the qRT-PCR analysis applying Universal Probe system (Roche). The following combination of primers and probes was used: Cx3cl1 (Probe#80, up: 5′-GGC TTT GCT CAT CCG CTA T-3′, down: 5′- CAG AAG CGT CTG TGC TGT GT-3′), mHprt (Probe#95, up: 5′-CCT CCT CAG ACC GCT TTT T-3′. Down: 5′-AAC CTG GTT CAT CAT CGC TAA-3′). The relative expression values were compared with the housekeeping gene (Hprt) expression using the ΔΔC_T method (Lightcycler MX3005, Agilent).

For analysis of tumor infiltrates, 4T1 tumor tissues were dissected, cut into small pieces, and digested in RPMI medium containing Liberase TL (0.5 mg/mL; Roche) and Dnase I (0.25 mg/mL; Roche) for 45 minutes at 37°C. Reaction was stopped using 0.5 M EDTA solution (GIPCO). After filtration through a 70 μm cell strainer, cells were counted, and 5 × 10⁶ cells were used for cell surface staining. The following fluorophore-conjugated antibodies were used: anti-mCD3 (clone 17A2, BioLegend, 2.5 μg/mL), anti-mCD4 (clone RM4-5, eBioscience, 1.3 μg/mL), anti-mCD8 (clone53-7.3, eBioscience, 1 μg/mL), anti-mCD49b (DX5, eBioscience, 1 μg/mL), anti-mFoxp3 (FJK-16s, eBioscience, 2.5 μg/mL), anti-mCD25 (PC61, BioLegend, 0.5 μg/mL), anti-mCD11b (M1/70, eBioscience, 0.8 μg/mL), anti-mF4/80 (BM8, eBioscience, 1.3 μg/mL), anti-mLy-6C (HK1.4, BioLegend, 1.3 μg/mL), and anti-mLy6G (1A8, BioLegend, 1.7 μg/mL). For analysis of CX3CL1 membrane expression of tumor cells, anti-CX3CL1 (clone AF365, R&D Systems, 200 μg/mL) was used together with an Alexa488 secondary fluorescent dye (Abcam, 6 ng/μL).

Detection of dead cells was performed with fixable viability dye eFluor 506 (eBioscience), and cell suspensions were incubated with mCD16/CD32 Fc-blocking antibody (553142, BD Biosciences) for 10 minutes. before the antibody staining was performed. Flow cytometry was performed using the FACS Canto 2 (BD Biosciences) together with the DIVA software. Data analysis was done with the FlowJoV10 software (Treestar).

Lung tissues of 4T1-inoculated mice were collected directly after finalization, and digestion was performed as described above for primary tumor tissue. The cell suspensions were then resuspended in selection RPMI medium containing 6-thioguanine (5 μg/mL; Merck) and 50 U/mL penicillin/50 μg/mL streptomycin. Increasing dilutions (1:10, 1:100, and 1:1,000) were prepared, and cells moved to 10-cm cell culture dishes containing 10 mL of selection RPMI medium. After incubation at 37°C for 2 weeks, colonies were counted after fixation with 100% methanol.

For further exploratory survival analyses, we used a publicly available Affymetrix dataset (retrievable at www.kmplot.com; refs. 25, 26). All cases of breast cancer were included in the analysis (n = 3951) using best cutoff as selected by the program.

Analysis of in vitro data was performed using a one-way ANOVA. Kaplan–Meier estimates of event-free survival were compared by log-rank tests. Hypothesis testing of in vivo experiments, in detail the differences in immune cell populations, were done with the Mann–Whitney U tests. All tests were performed on exploratory two-sided 5% levels of significance (SPSS Statistics Software, Version 25.0, SPSS Inc.).

To unravel the role of CX3CL1 in breast cancer, we first determined its expression and localization immunohistochemically in the breast tumor microenvironment (TME) of 250 cases of early breast cancer (Supplementary Table S1). CX3CL1 was mainly localized in tumor and endothelial cells, and tumor cell CX3CL1 expression was semiquantitatively scored on a four-tier scale based on the staining intensity (Fig. 1A). Within individual tumors, there was hardly any heterogeneity in CX3CL1 expression. CX3CL1 expression was evenly distributed among the classical, receptor-defined breast cancer subtypes (Fig. 1B), and there was no relevant correlation with tumor size, lymph node involvement, or nuclear grading. In CX3CL1-expressing breast cancers, CX3CL1 expression was positively associated with progression-free survival (PFS; Fig. 1C). The relative 5-year PFS was 74%, 84%, and 89% for CX3CL1 IHC 1+, 2+, and 3+ cases, and the relative 10-year PFS was 61%, 69%, and 79% for CX3CL1 IHC 1+, 2+, and 3+ cases, respectively. There was no significant difference in overall survival between the CX3CL1 expression groups. Our results are in line with the prognostic value of CX3CL1 mRNA in publicly available breast cancer datasets (Supplementary Fig. S1; ref. 25).

We next sought to define the functional impact of CX3CL1 overexpression in a fully immunocompetent mouse model of breast cancer. Because our IHC studies identified tumor cells as the major source of CX3CL1 in the TME, we stably transfected murine 4T1 breast cancer cells to express murine CX3CL1 (Cx3cl1) using a vector (Supplementary Fig. S2A) and orthotopically implanted them into the mammary fat pad of Balb/c mice. Cx3cl1 overexpression did not affect cell proliferation in vitro (Supplementary Fig. S2B). However, in vivo, Cx3cl1 overexpression resulted in delayed tumor growth (Fig. 1D) and significantly less tumor weight after 21 days (Fig. 1E) compared with vector control cells (median, 0.16 vs. 0.11 g; P = 0.04). Although tumor Cx3cl1 overexpression was still measurable after 21 days (Supplementary Fig. S2C), Cx3cl1 serum concentrations were not enhanced in the mice harboring Cx3cl1-overexpressing tumors (Supplementary Fig. S2D), suggesting rather locally restricted effects of the chemokine. Nevertheless, Cx3cl1 almost completely prevented the formation of lung metastases (median number of 6-thioguanine–resistant colonies 185 vs. 0; P = 0.004; Fig. 1F). Cx3cl1-overexpressing tumors showed a significantly enhanced infiltration by CD3⁻CD49b⁺ NK cells (0.35% vs. 0.45%; P = 0.06), CD3⁺CD4⁺ T cells (1.66% vs. 2.47%; P = 0.0015), and CD3⁺CD8⁺ cytotoxic T cells (0.35% vs. 0.64%; P = 0.04; Fig. 1G). However, regulatory T cells (Treg, CD3⁺CD4⁺FOXP3⁺) were not considerably elevated (0.55% vs. 0.71%; P = 0.22). LyG6⁺LyC6⁺ myeloid cells were significantly diminished in Cx3cl1-overexpressing tumors, whereas no change was seen in F4/80⁺MHC II⁺ macrophages or CD11b⁺ cells (Supplementary Fig. S2E). These results demonstrate the protective nature of CX3CL1 in breast cancer. The induction of NK-cell infiltration supported our hypothesis that CX3CL1 might be advantageous, especially for therapies dependent on the presence of NK cells, such as trastuzumab (3).

To test a possible exploitation of CX3CL1 in anti-HER2 therapy, we first examined the expression and regulation of CX3CL1 in the HER2-positive cancer cell lines SKBR3, MDA-MB-453, BT474, and HT-29. These cell lines expressed HER2 to variable degrees (Fig. 2A). All cell lines, except MDA-MB-453, secreted soluble CX3CL1 upon stimulation with TNFα, a potent inducer of CX3CL1 in noncancerous cells (Fig. 2B; ref. 27). SKBR3 and BT474 cells secreted significant amounts of CX3CL1 also in the absence of an inflammatory stimulus. The fraction of secreted CX3CL1 could be reduced upon ADAM17 (a disintegrin and metalloproteinase 17) inhibition by TAPI-2, demonstrating involvement of this protease in CX3CL1 shedding from the surface of tumor cells (Fig. 2B). This effect was not due to ADAM17-mediated modifications of the TNFα response because the secretion of CXCL10, which is also induced by TNFα (28), was not considerably altered (Supplementary Fig. S3A). MDA-MB-453 cells did not release any soluble CX3CL1, neither at baseline nor upon TNFα stimulation (Fig. 2B), although these cells were principally responsive to TNFα (Supplementary Fig. S3B). FACS analyses of vital cells to determine the membrane-bound CX3CL1 and Western blot analyses of whole-cell lysates confirmed the induction of CX3CL1 by TNFα concomitant with an increase of membrane-bound CX3CL1 upon ADAM17 inhibition in SKBR3, BT474, and HT-29 cells (Fig. 2C and D). In MDA-MB-453 cells, however, apart from intracellular baseline CX3CL1 expression, still no induction by TNFα and no regulation by ADAM17 inhibition was observed. Altogether, these data demonstrate that some HER2-positive cancer cell lines intrinsically express both membrane-bound and soluble CX3CL1, some need inflammatory stimuli to induce CX3CL1 expression, and others do not express significant amounts of CX3CL1.

Having characterized CX3CL1 expression and regulation in HER2-positive cancer cell lines, we next determined its impact on NK cell–mediated cytotoxicity and trastuzumab-mediated ADCC. Because preincubation of target cells with TNFα has been shown to enhance NK cell–mediated cytotoxicity in noncancerous cells (29), we examined whether this effect was also detectable in HER2-positive tumor cells and whether it was affected by CX3CL1. To this end, tumor cells were preincubated for 24 hours with or without TNFα prior to the addition of NK cells, and cytotoxicity was measured using a standard europium release assay. In both SKBR3 and HT-29 cells, TNFα led to an increase of specific tumor cell lysis by approximately 80% and 15%, respectively, which could be partially blocked by a monoclonal anti-CX3CL1 (Fig. 3A; Supplementary Fig. S4A). In the CX3CL1-negative, TNFα-responsive MDA-MB-453 cells, TNFα did not have any effect on cell lysis (Fig. 3A). These results suggest an involvement of CX3CL1 in TNFα-enhanced NK cell–mediated tumor cell lysis.

Next, we wanted to know whether overexpression of CX3CL1 was per se sufficient to enhance NK cell–mediated tumor cell lysis and capable of acting synergistically with trastuzumab. CX3CL1 was stably overexpressed in the three cell lines SKBR3, HT-29, and MDA-MB-453, respectively (Supplementary Fig. S4B), which were exposed to NK cells in the presence or absence of trastuzumab (Fig. 3B). CX3CL1 enhanced NK cell–mediated cytotoxicity by approximately 10% in SKBR3 and HT-29 cells and by 35% in MDA-MB-453 cells. As expected, trastuzumab alone caused an increase in tumor cell lysis of about 50% in the two HER2-overexpressing cell lines SKBR3 and MDA-MB-453 and of nearly 20% in HER2-weak HT-29 cells. In all three cell lines, there was a synergistic effect of CX3CL1 overexpression and trastuzumab with a specific tumor cell lysis reaching almost 100% in HT-29 and MDA-MB-453 cells (Fig. 3B). This synergism could not be mimicked by the addition of soluble CX3CL1 (chemokine domain, amino acids 1–76) to tumor cells and NK cells in wild-type HT-29 and MDA-MB-453 cells, suggesting that the membrane-bound form of CX3CL1 is primarily responsible for the described CX3CL1 effects in these cell lines (Supplementary Fig. S4C). In line with this, the increase in specific tumor cell lysis by the addition of the ADAM17 inhibitor TAPI-2 could be partially abrogated by a CX3CL1-blocking antibody in HT-29 cells (Supplementary Fig. S4D). Taken together, these results showed that CX3CL1 conferred susceptibility of tumor cells to NK cell–mediated cytotoxicity and acted synergistically with trastuzumab in HER2-positive cancer cells in vitro.

Given the synergism of CX3CL1 overexpression and trastuzumab on tumor cell lysis in vitro, we next tested whether CX3CL1 would enhance trastuzumab therapy in HER2-positive cancer also in vivo. To this end, we used xenotransplantation models in Balb/SCID mice, which lack adaptive immunity but still retain NK-cell activity, thus allowing us to study NK cell–mediated effects. First, we tested the role of CX3CL1 in the trastuzumab-sensitive MDA-MB-453 tumor model, which lacks intrinsic CX3CL1 (Fig. 2B; Supplementary Fig. S5A). After tumor implantation, mice were randomized to either receive trastuzumab or vehicle, and NK-cell depletion using an anti-asialo GM1 or an IgG control. Basophils, which are also depleted by the anti-asialo GM1 antibody, do not express CX3CR1 and were therefore not considered relevant to the study of CX3CL1 effects (30). Anti-HER2 treatment (or vehicle control) was initiated after the tumor had reached a diameter of 5 mm. In vehicle-treated mice, overexpression of CX3CL1 generally resulted in a significantly prolonged survival time of the mice (median, 161 vs. 132 days; P = 0.002; Fig. 4A). The initial tumor growth was slightly accelerated in the CX3CL1-overexpressing tumors (time to therapy start 34 vs. 41 days; P = 0.14). Although we generally detected only isolated DX5⁺ NK cells by IHC in the MDA-MB-453 tumors (with therefore no measurable difference between the groups), NK-cell depletion significantly reduced survival in the non-CX3CL1–overexpressing group (median survival, 93 vs. 132 days; P < 0.0001). However, overexpression of CX3CL1 was able to compensate for NK-cell depletion, as there was no substantial survival difference between NK cell–depleted and nondepleted mice (median survival, 165 vs. 161 days; P = 0.63; Fig. 4A). In this model, small amounts of trastuzumab were already sufficient to induce complete tumor regression (Supplementary Fig. S5B), which is in line with prior reports in these HER2-overexpressing breast cancer models (31). Herein, NK-cell depletion had no effect on tumor growth and survival, suggesting that NK cell–mediated ADCC is less important than inhibition of HER2 signaling in these HER2-overexpressing tumors. The intratumoral frequency of macrophages, also known to express the CX3CR1 receptor, was not significantly different between the groups (Supplementary Fig. S5C).

Supporting this lack of synergism of CX3CL1 expression and trastuzumab efficacy in MDA-MB-453, we did not find a significant association of CX3CL1 expression and PFS in patients with HER2-overexpressing (IHC 2+/FISH-positive, IHC 3+), trastuzumab-treated breast cancer (Supplementary Fig. S6A). In a second cohort of patients with HER2-overexpressing breast cancer who were treated neoadjuvantly with trastuzumab and chemotherapy, we did not observe a substantial correlation of pretherapeutic CX3CL1 expression and therapy response (Supplementary Fig. S6B). However, in the subgroup of CX3CL1-positive, HER2 low–expressing breast cancers (IHC score 1+ or 2+/FISH negative), we found a significant association of CX3CL1 expression and PFS (Fig. 4B).

We therefore tested the interaction of CX3CL1 and trastuzumab in the HER2 low–expressing HT-29 model. Stable transfection of HT-29 cells with a murine CX3CL1 expression vector (Supplementary Fig. S7A and S7B) caused a significant, about 4-fold increase, in NK-cell infiltration into the tumors (Fig. 4C). Trastuzumab, however, had no impact on NK-cell infiltration. As in the 4T1 and the MDA-MB-453 models, we again observed an accelerated initial tumor growth phase in the Cx3cl1-overexpressing tumors (median time to first measurable tumor 9 vs. 12 days; P = 0.001; time to onset of therapy 12 vs. 15 days; P = 0.03). Subsequently, Cx3cl1 overexpression led to a significantly reduced tumor growth compared with control mice (Fig. 4D), and median survival after onset of therapy was prolonged by Cx3cl1 overexpression (21 vs. 14 days; P = 0.038; Fig. 4E). Although trastuzumab had no significant effect in the absence of Cx3cl1, it significantly slowed down tumor growth and improved survival in Cx3cl1-overexpressing tumors (median survival time, 36 vs. 21 days; P < 0.0001). To further explore the contribution of NK cells to these effects, we performed the same experiment in SCID Beige mice lacking NK cells. Here, virtually no intratumoral DX5⁺ NK cells were detectable by IHC. Overexpression of Cx3cl1 had no impact on tumor growth or survival anymore (Fig. 4F; Supplementary Fig. S7C). Cx3cl1 and trastuzumab still had a synergistic effect, but it was significantly reduced compared with the effect seen in SCID mice (adjusted median survival difference between Cx3cl1⁺ and Cx3cl1⁺/trastuzumab 22.5 days in SCID vs. 15.5 days in SCID Beige, P < 0.0001). These results demonstrate an involvement of NK cells in the Cx3cl1-induced survival benefit and its synergism with trastuzumab in vivo. However, they also suggest other mechanisms to contribute. Because macrophages also express CX3CR1 and facilitate antibody-dependent cellular phagocytosis (32, 33), we determined their intratumoral accumulation. In contrast to our findings in the 4T1 and the MDA-MB-453 models, in both HT-29 experiments, we detected a significant increase in F4/80-positive macrophages upon Cx3cl1 overexpression, which might also participate in the observed CX3CL1/trastuzumab synergism (Fig. 4G). In summary, our data provide evidence that CX3CL1 is capable of overcoming trastuzumab resistance in HER2 low–expressing cancer and might be a suitable biomarker for the selection of patients for trastuzumab therapy in this cohort.

One of the key questions in contemporary immuno-oncology is how to transform a “cold,” less inflamed tumor into a “hot” tumor, enriched in immune cells. This is a fundamental prerequisite for the success of most immunotherapies, including immune checkpoint inhibitors and mAbs directed against tumor antigens (34). The poorly explored chemokine CX3CL1 does not only attract tumor-suppressive lymphocytes, it might also enhance their binding to the target tumor cells due to its presence as a membrane-bound form (10). In this study, we provide functional evidence for this concept, exemplified by the trastuzumab treatment of HER2-expressing cancer. We found that overexpression of CX3CL1 not only improved NK- and T-cell infiltration and inhibited tumor growth and metastatic spread, but also enhanced NK-cell killing of trastuzumab-tagged tumor cells in vitro. In vivo, CX3CL1 was sufficient to overcome trastuzumab resistance in a preclinical model of HER2 low–expressing cancer.

Previous studies demonstrate that intratumoral CX3CL1 recruits NK and T cells and impairs tumor growth in preclinical models of lymphoma, hepatocellular carcinoma, colon, and lung cancer (11, 35–38). In contrast, there are reports suggesting that CX3CL1 might foster migration and metastatic spread of CX3CR1-positive tumor cells to distant, CX3CL1-enriched metastatic sites such as brain or bone (39, 40). This, however, could also add to the protective nature of intratumoral CX3CL1 expression, as local CX3CL1 might prevent CX3CR1-positive tumor cells from leaving the primary tumor. Tardáguila and colleagues (14) demonstrate that CX3CL1 participates in the tumorigenesis of HER2-positive breast cancer via EGFR pathway transactivation, but it does not affect growth of established tumors. In line with this, we observed an accelerated initial tumor growth in the CX3CL1-overexpressing tumors in both the MDA-MB-453 and the HT-29 tumor models, and only a late separation of tumor growth curves in the 4T1 model. However, in the further course of the experiments, CX3CL1 suppressed tumor growth and improved survival in all three tumor models. In support of this, we observed a significant association of tumor cell–associated CX3CL1 and PFS in our breast cancer cohort, which could be confirmed in publicly available mRNA databases (25). Park and colleagues (20) reported similar results in their breast cancer cohort, in which CX3CL1 was positively associated with the number of CD8⁺ T cells, CD57⁺ NK cells, as well as disease-free and overall survival. Several groups report a downregulation of CX3CL1 in breast cancer compared with normal breast tissue, which might promote immune escape (14, 41–43). In summary, although CX3CL1 might play a role in EGFR-mediated breast cancer tumorigenesis, our results in different breast cancer models support a protective role in established breast cancer, which is further backed by clinical data.

Forced overexpression of CX3CL1 was sufficient to augment NK-cell cytotoxicity against tumor cells in the current work. Previously, CX3CL1 is demonstrated to induce NK cell–mediated damage of endothelial cells (44, 45). CX3CL1-overexpressing HeLa cells or soluble CX3CL1 triggers IFNγ release and lytic activity of NK-92 cells in vitro (46, 47). Our results showed that the enhancement of NK-cell cytotoxicity by TNFα, which has been described before (29), can be, at least in part, traced back to CX3CL1, as the inhibition by a monoclonal anti-CX3CL1 suppressed TNFα-mediated cytotoxicity, and TNFα had no effect in CX3CL1-negative but per se TNF-responsive MDA-MB-453 cells. However, it is still not entirely clear which CX3CL1 isoform is responsible for these effects. There is evidence for a stronger stimulation of CX3CR1 downstream processes, such as Ca²⁺ influx or IFNγ secretion, in NK cells by the membrane-bound CX3CL1 form in vitro (48, 49). However, the only study in cancer that differentiates between the effects of membrane-bound and soluble CX3CL1, found antitumor activity of soluble CX3CL1 and a mixed to no effect of membrane-bound chemokine in vivo (35). We approached this question by adding recombinant soluble CX3CL1 to the tumor cell–NK cell coincubation, which did not have any substantial effect. On the other hand, the increase in NK-cell lytic activity by ADAM17 inhibition was partially abrogated by CX3CL1 blockage, again implicating a role for membrane-bound CX3CL1 in this enhanced cytotoxicity. This is in line with reports showing that ADAM17 confers resistance to trastuzumab (50, 51). It is tempting to speculate that the cleavage of membrane-bound CX3CL1 might participate in this process.

Our results demonstrate that CX3CL1 acts in concert with trastuzumab by improving NK-cell attraction and cytotoxicity, which might be exemplary for other ADCC-dependent therapies. In a cohort of 53 patients with HER2-positive, trastuzumab-treated breast cancer, expression of CX3CL1 was associated with a reduced risk of relapse (52). In the trastuzumab-treated, HER2 high–expressing cohort, we did not find a correlation of trastuzumab response and CX3CL1 expression. This supports our in vivo finding of a lacking synergism in the HER2-overexpressing tumor model. However, in patients with HER2 low–expressing breast cancer, we observed a positive association with PFS. As these patients had not been treated with trastuzumab according to the current guidelines, we do not know whether CX3CL1 is also predictive for trastuzumab response. In the HER2-overexpressing MDA-MB-453 tumor model, we observed neither an improvement of trastuzumab therapy by CX3CL1 overexpression nor a dependence on NK cells because a short trastuzumab treatment already induced complete tumor regression. This could be explained by the notion that immunoregulatory effects of trastuzumab are less important in HER2-high tumors in which the mere inactivation of the HER2 receptor activity is crucial and possibly sufficient (53).

The current study focused on the role of NK cells in CX3CL1-mediated tumor suppression and trastuzumab therapy. However, our results in SCID Beige mice show that other mechanisms might also participate, for example, the enhanced recruitment of macrophages observed in the HT-29 model. Macrophages utilize the CX3CR1 receptor for chemotaxis (54) and are capable of enhancing trastuzumab action through antibody-dependent cellular phagocytosis (33). However, they have been reported to be an adverse prognostic marker in trastuzumab-treated HER2-positive breast cancer (55), possibly because their polarization state determines the impact they have on anti-HER2 treatment (56). The next step now will be to clarify their role in CX3CL1-mediated antitumor response and trastuzumab efficacy.

In conclusion, our results demonstrate that CX3CL1 overexpression improves trastuzumab efficacy in HER2 low–expressing cancer in vitro and in vivo. Besides being a potential predictor of response in these cancers that are currently not treated with trastuzumab, pharmacologic enrichment of CX3CL1 in the TME might be a feasible way to enhance the clinical efficacy of trastuzumab therapy. The next steps are now to decipher ways to increase CX3CL1 expression in cancer cells. Besides the induction of its expression, the inhibition of its proteolytic inactivation might be a promising approach (10). This could potentially open targeted anti-HER2 therapy to a large number of patients.

T.F. Dreyer reports grants from DFG during the conduct of the study. C. Stange reports grants from Deutsche Forschungsgemeinschaft during the conduct of the study. S. Seitz reports grants from DFG during the conduct of the study. A.K. Wege reports grants from German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) during the conduct of the study. W. Weichert reports personal fees from Roche, MSD, Bristol Myers Squibb, AstraZeneca, Pfizer, Merck, Lilly, Boehringer, Novartis, Takeda, Bayer, Amgen, Astellas, Illumina, Siemens, Agilent, and Molecular Health for advisory boards/speaker programs and grants from Roche, MSD, Bristol Myers Squibb, and AstraZeneca (to institution) outside the submitted work. M. Kiechle reports other support from Theraws Diagnostic GmbH and AIM GmbH, and personal fees from AstraZeneca, Celgene, Eli Lilly, Exeltis, Teva, and Myriad Genetics during the conduct of the study. H. Bronger reports grants from Deutsche Forschungsgemeinschaft (DFG), German Research Foundation during the conduct of the study, as well as personal fees from Roche, AstraZeneca, GlaxoSmithKline, Clovis, Teva, PharmaMar, and Pfizer outside the submitted work. No disclosures were reported by the other authors.

T.F. Dreyer: Conceptualization, data curation, software, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S. Kuhn: Investigation, visualization, writing–original draft, writing–review and editing. C. Stange: Validation. N. Heithorst: Data curation, validation, investigation, writing–original draft, writing–review and editing. D. Schilling: Conceptualization, resources, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. J. Jelsma: Investigation, writing–original draft, writing–review and editing. W. Sievert: Conceptualization, resources, supervision, investigation, writing–original draft, writing–review and editing. S. Seitz: Validation, writing–original draft, writing–review and editing. S. Stangl: Resources, formal analysis, investigation, writing–original draft, writing–review and editing. A. Hapfelmeier: Software, formal analysis, writing–original draft, writing–review and editing. A. Noske: Formal analysis, validation, investigation, writing–original draft, writing–review and editing. A.K. Wege: Funding acquisition, investigation, writing–original draft, writing–review and editing. W. Weichert: Resources, supervision, writing–original draft, project administration, writing–review and editing. J. Ruland: Resources, writing–original draft, project administration, writing–review and editing. M. Schmitt: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing. J. Dorn: Resources, funding acquisition, writing–original draft, writing–review and editing. M. Kiechle: Resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing. U. Reuning: Resources, writing–original draft, writing–review and editing. V. Magdolen: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing. G. Multhoff: Conceptualization, resources, supervision, writing–original draft, writing–review and editing. H. Bronger: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank Christine Huber for excellent technical assistance. This work was supported by project grants from the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG; BR4733/2-1, to H. Bronger and WE3606/3-1, to A.K. Wege).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.