Kindlin-2 Regulates the Growth of Breast Cancer Tumors by Activating CSF-1–Mediated Macrophage Infiltration

The complex interactions between cancer and stromal cells create a microenvironment where tumor cells can thrive and ultimately metastasize (1). Proinflammatory macrophages, which are significant cellular components of the tumor microenvironment, contribute to tumor initiation and progression (2). Both autocrine and paracrine signaling loops promote interactions between cancer cells and host cells within the tumor microenvironment (3, 4). In several cancers, including those originating in the breast, a paracrine signaling loop between tumor cells and macrophages involving tumor cell–derived EGF and macrophage-derived colony-stimulating factor-1 (CSF-1) conspire to promote tumor growth (5, 6). However, the molecular mechanisms that underlie the expression and secretion of both CSF-1 and EGF remain largely unknown.

Kindlins are members of the 4.1- ezrin-ridixin-moesin (FERM) domain containing proteins (7). Kindlins contain F1, F2, and F3 subdomains typical of FERM domain proteins that follow an N-terminal F0 subdomain and a pleckstrin homology (PH) subdomain that transects F2 subdomain. The three mammalian kindlin family members (Kindlin-1:FERMT1; Kindlin-2:FERMT2, and Kindlin-3:FERMT3) are highly homologous, sharing approximately 60% amino acid sequence identity (7). The adaptor functions of kindlins allow them to regulate numerous cellular responses. Of particular importance is their capacity to regulate integrin activation, and integrin-dependent cell adhesion, spreading, and migration (7–11). Kindlin-2, as well as its homologues, Kindlin-1 and Kindlin-3, has been linked to the malignancy of several types of cancers (reviewed in refs. 7, 8). However, despite a few reports implicating Kindlin-2 in breast cancer (12–15), a molecular mechanism by which Kindlin-2 contributes to the pathogenesis of breast cancer is presumed to be dependent on its activation of tumor cell integrins, but its role in regulating tumor cell–stromal interactions has not been considered.

In the current study, we report that Kindlin-2 is upregulated in human breast cancer cell lines and tumors and that its increased expression correlates with poor prognosis in breast cancer patients and with the metastatic potential of both human and mouse breast cancer progression series. Both human and mouse breast cancer tumors lacking Kindlin-2 grow slowly in mice. Importantly, tumors lacking Kindlin-2 have diminished tumor-associated macrophages (TAM). This is a consequence of Kindlin-2–mediated regulation of expression and secretion of CSF-1 by breast cancer cells, thus supporting macrophage infiltration into tumors. We further show that TGFβ signaling modulates Kindlin-2–mediated regulation of CSF-1 and EGF expression and is required for autocrine and paracrine cross-talk between cancer cells and macrophages that promotes tumor growth.

Normal murine mammary gland cells (NMuMG), murine 4T1 cells, and human MCF7, MCF10A, MDA-MB-231, T47D, SKBr3, and BT549 cells were obtained from ATCC between 2010 and 2013. We used STR DNA fingerprinting analysis for authentication of these cells in 2014 and again in March 2016. Murine 67NR and 4T07 and human MCF10Ca1h and MCF10Ca1a were obtained from Dr. Fred Miller (Wayne State University, Detroit, MI). Cells were maintained in DMEM supplemented with 10% FBS. 4T1 cells were engineered to stably express firefly luciferase by transfection with pNifty-CMV-luciferase and selection (500 μg/mL) with Zeocin (Invitrogen). Kindlin-2–deficient cells were produced by pLenti-CRISPRv2 lentiviral transduction using a scrambled single-guide RNA (sgRNA; i.e., nonsilencing sgRNA) or two independent and verified Kindlin-2–specific sgRNAs for human and mouse Kindlin-2. Stable pools of Kindlin-2–deficient 4T1 or MDA-MB-231 cells were obtained by culture over 14 days in puromycin (5 μg/mL). The extent of Kindlin-2 deficiency was determined by immunoblots.

LentiCRISPRv2 lentiviral plasmid system (Addgene) was used to specifically knock out Kindlin-2 in human MDA-MB-231 and mouse 4T1 breast cancer cells as described by Cong and colleagues (16). The human- and mouse Kindlin-2–specific sgRNA were identified on the basis of two different predictive algorithms (Chopchop; https://chopchop.rc.fas.harvard.edu, and CRISPR Design, http://crispr.mit.edu). The sgRNAs common to both algorithms were validated against human and mouse GECKOv2 sgRNA libraries (16), and only sgRNAs found in the GECKOv2 libraries were selected. sgRNA oligos were purchased from IDT-DNA and subcloned in lentiCRISPRv2 plasmid (16). Lentivirus production and cancer cell infection were performed as described previously (17–19).

Parental (scram) or Kindlin-2–deficient MDA-MB-231 cells (10⁶ cells per mouse, n = 5) were implanted into the mammary fat pads of female NSG mice (Cleveland Clinic). Tumor growth was followed by thrice weekly monitoring of tumor volume with digital Vernier calipers.

Luciferase-expressing parental (scram) or Kindlin-2–deficient 4T1 cells (10,000 cells per mouse, n = 5) were implanted into the mammary fat pads of female BALB/c mice (The Jackson Laboratory). Tumor growth was quantified using bioluminescence imaging as described previously (17). All animal studies were performed under protocols approved by the Institutional Animal Care and Use Committee.

Female BALB/c mice were injected intraperitoneally with 0.5 mL of 4% thioglycollate (20). After 72 hours, cells were isolated from peritoneal lavages, where macrophages constituted approximately 90% of all cells as determined by flow cytometry using FITC-labeled anti-F4/80 Ab (21).

Three-dimensional (3D) organotypic cultures using the “on-top” method were performed as described previously (22). Briefly, cells (2,000 cells per well) were cultured in 96-well plates onto Cultrex cushions (50 μL/well; Trevigen) in complete medium supplemented with 5% Cultrex Organoid Growth, which was monitored by bright-field microscopy (23).

TGFβ1 and CSF-1 production by 4T1 and MDA-MB-231 cell derivatives was measured using the mouse or human Quantikine TGFβ1 assays (R&D systems MB100B and DB100B for human and mouse TGFβ1, respectively) and the human and mouse CSF ELISA Kits (Thermo Fisher Scientific, EHCSF1 and EMCSF1, respectively) according to the manufacturer's instructions. 4T1 and MDA-MB-231 cell derivatives were cultured for 24 hours, and the medium was replaced with serum-free DMEM. After an additional 24 hours, conditioned medium (CM) was collected and centrifuged at 500 × g for 3 minutes. TGFβ1 (latent + active) and CSF-1 were measured in the serum-free CM. Cells remaining on the plate were lysed. Total protein concentrations in lysates were used to normalize total TGFβ1 and CSF-1.

The following primary antibodies were from Cell Signaling Technology: anti-phospho-Smad3, anti-Smad3, anti-phospho-ERK MAPK, anti-ERK MAPK, and anti-M-CSF Receptor. The FC blocker used was Mouse SeroBlock FcR, which is a rat mAb (clone FCR 4G8, Bio-Rad) that specifically recognizes mouse CD16 and CD32, which are cell surface proteins also known as FcRgIII and FcRgII, respectively. Mouse monoclonal anti-K2, clone 3A3 (1:2000, EMD Millipore), anti-F4/80 (1:50, eBioscience), anti-Gr-1 (1:50, AbD Serotec), anti-CD206 (1:50, Bio-Rad), goat horseradish peroxidase–conjugated anti-mouse IgG (1:2,000), and goat horseradish peroxidase–conjugated anti-rabbit IgG (1:2,000) were from Calbiochem. VECTASHIELD with 4′,6-diamidino-2-phenylindole was from Vector Laboratories. Gel electrophoresis reagents were from Bio-Rad. CSF-1 was from Thermo Fisher Scientific. Lipopolysaccharide (LPS) and CCL2 were from Sigma.

Total RNA was extracted from cancer cell lines or tumor tissue using TRIzol reagent (Invitrogen), following the manufacturer's instructions. cDNA was generated and used as a template for qRT-PCR as described previously (24–26). Oligonucleotide primers used for qRT-PCR were from SABiosciences.

Cell migration was assessed by wounding confluent cultures with a micropipette tip and immediately placing them in complete medium. Bright-field images were obtained immediately after wounding and after 18 hours. Wound closure was quantified by measuring wound areas from ≥6 different fields using ImageJ 1.34s (NIH, Bethesda, MD). For invasion assays, modified Boyden chambers were coated with Matrigel (1:10 dilution; BD Biosciences), the invasion of MDA-MB-231 and 4T1 cells, and their Kindlin-2–deficient derivatives in response to 10% serum were measured as described previously (17, 25). Alterations in cell proliferation of MDA-MB-231 and 4T1 cells and their Kindlin-2–deficient derivatives were determined by counting the number of viable cells. Cells were seeded into 6-well plates in complete growth medium and then harvested at 1-day intervals over 5 days, and counted in a hemocytometer. Cell viability was assessed using Trypan blue staining. Assays were performed in triplicates, and the values plotted were the average of two independent experiments.

Tissues were collected at the designated times and snap-frozen in optimal cutting temperature medium (Sakura Finetek), and 8-μm sections were prepared. Macrophages were detected with rat anti-F4/80 (eBioscience). M2 macrophages were detected with rat anti-F4/80 and rat anti-mouse CD206 (Bio-Rad). Monocytes/neutrophils were detected with anti-Gr1 (AbD Serotec), followed by Alexa 488 or Alexa 568–conjugated goat anti-rabbit IgG. Stained sections were analyzed using fluorescent or bright-field imaging microscopy (Leica) and ImagePro Plus Capture and Analysis software (Media Cybernetics). F4/80-, CD206-, and Gr1-positive areas were quantified in 15 independent fields/section using Image Pro-Plus software (25, 27).

The expression of Kindlin-2 gene in breast cancer was examined via Oncomine database (www.oncomine.com). The Minn dataset (28) was used to compare expression levels of Kindlin-2 between breast cancer and normal tissues. The expression values of Kindlin-2 were log-transformed, median-centered, and normalized to one per array.

The effect of Kindlin-2 expression on the prognosis of 3,554 breast cancer patients was analyzed using the Kaplan–Meier plotter online software (http://kmplot.com/analysis/). The Kaplan–Meier plotter evaluates the effect of 54,675 genes on survival using 10,188 cancer samples, including breast, lung, ovarian, and gastric cancer patients. Kindlin-2 expression and survival data were derived from Affymetrix microarray data (ID: 214212_x_at). To analyze the prognostic value of Kindlin-2 gene, the samples were divided into two groups according to the median expression of Kindlin-2. The two patient groups (high and low expression of Kindlin-2) were compared using the Kaplan–Meier survival plot. The HR with 95% confidence intervals, and the log rank P value was computed as part of the Kaplan–Meier plotter online software. Similar analysis was performed for CSF-1 expression and survival data using Affymetrix microarray (ID: 210557_x_at).

Experiments were done in triplicate and analyzed using the Student t test. In calculating two-tailed significance levels for equality of means, equal variances were assumed for the two populations. Results were considered significant at P < 0.05.

Enhanced expression of Kindlin-2 has been reported for several human cancers (7, 8), including breast cancer (12–15). However, a definitive role for Kindlin-2 in tumorigenesis remains to be elucidated. To begin to address its role in breast cancer, we analyzed Kindlin-2 expression across a panel of human breast cancer cell lines. Kindlin-2 levels were 3-fold or higher in aggressive human MDA-MB-231 and BT549 breast cancer cells compared with less aggressive breast cancer lines, T47D, MCF7, or SKBR3 (Fig. 1A). Also, Kindlin-2 was expressed at very low levels in normal human MCF10A and indolent MCF10aT1K cells relative to their high-grade and more aggressive human MCF10aCa1h and MCF10aCa1a counterparts (Fig. 1B). We also found Kindlin-2 expression to be increased in metastatic murine 4T1 and dormant 4T07 cells relative to their indolent 67NR counterparts (Fig. 1C). A similar trend was observed in the murine NMUMG breast cancer series: Kindlin-2 expression was significantly elevated in the LM2 metastatic cells compared with their less aggressive NME cells or nontumorigenic NMuMG counterparts (Fig. 1C). Next, we addressed whether this trend in Kindlin-2 expression also occurred in human breast cancer tumors by Oncomine microarray expression analyses (http://www.oncomine.org). Kindlin-2 expression was significantly upregulated in breast cancers compared with normal breast tissue (Fig. 1D; ref. 28); it was in the top 3% of upregulated genes in breast cancer (Fig. 1D). Likewise, interrogation of the Kaplan–Meier Plotter breast cancer dataset (http://kmplot.com/analysis/) determined that Kindlin-2 expression was correlated with poor outcome and reduced survival in breast cancer patients (Fig. 1E). Collectively, these findings suggest that increased Kindlin-2 expression is characteristic of invasive breast cancer.

CRISPR/Cas9 technology has emerged as a significant tool for efficient site-specific gene editing and targeting (16, 29). We designed two sgRNA, sgRNA-1 and -2, to target exon 2 and exon 5 of human Kindlin-2 (FERMT2, Fig. 2A and B), and to target exon 2 and exon 3 of mouse Kindlin-2 (Fermt2, Fig. 2C and D) and used these in human MDA-MB-231 and murine 4T1 cells as widely used breast cancer models. Both have elevated levels of Kindlin-2 (Fig. 1A and C). The sgRNAs designed against human Kindlin-2 were very efficient in human MDA-MB-231 cells compared with cells infected with a scrambled (Scram) sgRNA or the parental cells (Fig. 2B). Likewise, the sgRNAs designed against mouse Kindlin-2 resulted in a very efficient knockdown of Kindlin-2 in mouse 4T1 cells (Fig. 2D). Sequencing of individual PCR clones from the genomic DNA of sgRNA-targeted exon 2 of Kindlin-2 in MDA-MB-231 cells showed several small insertion and deletions (indels) near the target site (Fig. 2E), which resulted in frameshifts in the coding sequence (Fig. 2F).

Having confirmed the efficiency of Kindlin-2 knockdown using CRISPR/Cas9, we investigated the effect of Kindlin-2 deficiency on the behavior of human and mouse breast cancer cells. First, we found that both the scrambled (Scram) and the K2-sgRNAs (K2-CRISPR) did not have a significant effect on proliferation of either MDA-MB-231 (Fig. 2G) or 4T1 (Fig. 2H) cells. Next, in a wound closure assay, we found loss of Kindlin-2 expression (K2-CRISPR) in MDA-MB-231 and 4T1 cells resulted in a significant decrease of migration into wounds as compared with parental and the control (Scram) cells (Fig. 2I and J, respectively). In Boyden chamber invasion assays, less MDA-MB-231 and 4T1 Kindlin-2–deficient (K2-CRISPER) cells traversed the Matrigel-coated inserts compared with the parental and Scram cells (Fig. 2K and L, respectively). These results were replicated in two and three different pools of Scram and K2-CRISPR cells, respectively, supporting the reproducibility of the data across multiple pools.

To assess the effects of loss of Kindlin-2 on tumor growth in vivo, mammary fat pads of NOD-scid-IL2Rgamma knockout (NSG) mice were inoculated with control (Scram) or Kindlin-2–deficient (K2-CRISPR) MDA-MB-231 cells, and tumor growth was assessed over 8 weeks. Loss of Kindlin-2 inhibited the growth of primary tumors (Fig. 3A–C). Although every mouse in the control and Kindlin-2 groups developed tumors (100% tumor incidence), after an approximately 4-week latency, tumor burden, as assessed by tumor size (Fig. 3A), weight (Fig. 3B), and volume (Fig. 3C), was significantly lower (P < 0.05) in the mice implanted with the Kindlin-2–deficient cells. Similarly, loss of the mouse Kindlin-2 in 4T1 cells also delayed tumor initiation and growth in the BALB/c mice (Fig. 3D and E). Thus, loss of Kindlin-2 inhibits the rate of primary tumor growth in vivo of both human and mouse models for breast cancer. These differences in tumor burden were not a result of decreased tumor cell proliferation by Kindlin-2 knockdown, as the number of viable cells between the control and Kindlin-2–deficient cells was similar (Fig. 2G and H).

Cross-talk between the tumor and stromal microenvironment is a prominent mechanism in primary tumor growth and subsequent invasion and metastasis (2, 6, 30, 31). We determined whether loss of Kindlin-2 affected the recruitment of TAMs into in MDA-MB-231 and 4T1 tumors. As F4/80 mAb we used for staining (32) can also react with monocytes from TAM, we double stained tumor sections for F4/80 (green) and Gr-1 (red), a marker of not only neutrophils but also inflammatory monocytes. The staining revealed abundant TAMs (F4/80⁺-Gr-1⁻ cells, green) in Scram as compared with K2-CRISPER MDA-MB-231 (Fig. 3F, top; Supplementary Fig. S1) and 4T1 (Fig. 3F, bottom; Supplementary Fig. S1) tumors. In contrast, as indicated by white arrowheads, we detected very few monocytes (F4/80⁺-Gr-1⁺, yellow cells) in all tumors. Quantification of F4/80⁺-Gr-1⁻ areas (green) in tumor sections showed an approximately 7- and 3.5-fold increase in TAMs in Scram-MDA-MB-231 (Fig. 3G) and Scram-4T1 (Fig. 3H) tumors, respectively, as compared with their K2-CRISPER counterparts (P < 0.001, n = 5). In addition, F4/80⁺-Gr-1⁺ monocytes (yellow) did not comprise more than 5% of the F4/80⁺ population in the tumors. We also double stained tumor sections for F4/80 (green) and CD206 (red), a marker for polarized M2 mouse macrophage phenotype (33). Tumors derived from Scram-MDA-MB-231 and Scram-4T1 showed massive infiltration with immunosuppressive M2 macrophages (F4/80⁺-CD206⁺, yellow and orange), compared with their K2-CRISPR counterparts (Fig. 3I, top and bottom, respectively, and Supplementary Fig. S2). Quantification of F4/80⁺-CD206⁺ areas (yellow and orange) in tumor sections show that majority of macrophages are of the M2 phenotype and their counts are enhanced by approximately 10-fold in the MDA-MB-231 and 4T1 Scram-tumors compared with the K2-CRISPR tumors (Fig. 3J and K, respectively). Similar results were found when comparing Scram-MDA-MB-231 and K2-CRISPR-MDA-MB-231 tumors of roughly similar sizes (Supplementary Fig. S3). In addition, as the average volume of tumors derived from K2-CRISPR-4T1 did not exceed 55 mm³, compared with more than 240 mm³ for the Scram-4T1 tumors (Fig. 3E), we considered whether K2-CRISPR-4T1 tumors were too small to initiate macrophage recruitment. Accordingly, we repeated this experiment allowing the tumors from the 4T1-K2-CRISPR to grow for an additional 3 weeks before excision so that K2-CRISPR tumors reached comparable sizes (∼300 mm³, Supplementary Fig. S4A) to the Scram 4T1 tumors excised 3 weeks earlier. Tumors derived from Scram-4T1 showed increased infiltration with immunosuppressive M2 macrophages (F4/80⁺-Cd206⁺, yellow and orange), compared with their K2-CRISPR counterparts (Supplementary Fig. S4B), similar to the data shown in Fig. 3I. Quantification of F4/80⁺-CD206⁺ areas (yellow and orange) in tumor sections also show that majority of macrophages were of the M2 phenotype in the 4T1 Scram tumors compared with the K2-CRISPR tumors (Supplementary Fig. S4C). Thus, K2 deficiency in cancer cells inhibits tumor infiltration by macrophages independent of the tumor size.

To further explore the interrelationship between Kindlin-2 in breast cancer tumor cells and macrophages, 3D organotypic outgrowth system was used. Scram or K2-CRISPR MDA-MB-231 cells were seeded onto 3D organotypic cultures, either alone or premixed with mouse peritoneal macrophages. To distinguish the cell populations within the 3D organoids, MDA-MB-231 cells were labeled with GFP. When Scram MDA-MB-231 cells were seeded alone, they readily formed 3D organoids, an established characteristic of aggressive breast cancer cells (Fig. 4A-1, top and bottom). Premixing the cancer cells with macrophages (Scram + Mφ) resulted in an approximately 3-fold increase of the 3D outgrowth (Fig. 4A-2, top and bottom and B). These 3D organoids mainly originated from MDA-MB-231 breast cancer cells, not from the macrophages, as indicated by fluorescence labeling of these structures (Fig 4A, top). Seeding of macrophages alone onto 3D organotypic cultures did not support any 3D outgrowth (not shown), confirming that cancer cells give rise to these structures. Interestingly, Kindlin-2 deficiency significantly repressed the outgrowth of breast cancer organoids in 3D cultures (Fig 4A-3 top and bottom and B), and inclusion of macrophages with the Kindlin-2-deficient breast cancer cells (K2-CRISPR + Mφ) did not overcome the repression of 3D outgrowth caused by lack of Kindlin-2 in the cancer cells (Fig. 4A, top and bottom and B).

Tumor-educated macrophages switch from an anti-inflammatory to proinflammatory phenotype that can encourage tumor growth (34). We investigated whether manipulating Kindlin-2 in cancer cells would regulate macrophage function and thereby affect the 3D outgrowth of tumor cells. Untreated macrophages (naïve Mφ) or macrophages preincubated with the CM of MDA-MB-231 cells for 48 hours (tumor-educated Mφ) were premixed with either Scram or K2-CRISPR MDA-MB-231 cells. Control MDA-MB-231 cells that were mixed with tumor-educated macrophages (Scram + tumor-educated Mφ) formed 3D organoids that were approximately 6-fold larger (P < 0.05) than those formed by the same tumor cells mixed with untreated (Scram + naïve Mφ) macrophages (Fig. 4C). The 3D organoids that developed from the K2-CRISPR tumor cells remained significantly (P < 0.05) smaller than those generated from the Scram tumor cells, regardless whether they were mixed with naïve or educated macrophages (Fig. 4C). These data clearly indicate that Kindlin-2 in cancer cells is required for both 3D outgrowth, for regulation of macrophage function, and for their ability to stimulate 3D outgrowth of cancer cells. Kindlin-2 from the tumor cells most likely mediates these biological effects as neither resting nor LPS-activated macrophages expressed Kindlin-2 as assessed by Western blots (Fig. 4D). These findings also indicate that macrophages, as a part of the tumor microenvironment, play an important role in stimulating tumor growth, as suggested from Figs. 3 and 4 and consistent with the literature (6).

We considered whether the macrophage-mediated stimulation of tumor growth depends on direct contact with the cancer cells or through secreted chemotactic and stimulatory factors. MDA-MB-231 cells were treated with TGFβ1, which stimulates outgrowth of tumor cells in 3D organotypic cultures (17, 18). While TGFβ1 stimulated 3D outgrowth of the Scram cells (Fig. 4E, top panel), it failed to exert the same effect on the K2-CRISPR cells (Fig 4E, bottom). The size of the 3D organoids was approximately 4-fold (P < 0.05) smaller in the Kindlin-2–deficient cells compared with Scram cells (Fig. 4F). Next, we investigated the effect of CM of macrophages on 3D outgrowth. The size of the 3D organoids from the Scram MDA-MB-231 cells was >4-fold larger (P < 0.05) when incubated with the CM of LPS-stimulated macrophages (Fig. 4G, top, and H), compared with those incubated with plain culture medium. The 3D outgrowth of the K2-CRISPR cells was significantly smaller (P < 0.05) than the control cells under both culture conditions (Fig. 4G, bottom, and H). These findings suggest that factors secreted from activated macrophages stimulate cancer cell growth in the 3D organotypic cultures, that is, a paracrine signaling loop. As Kindlin-2–deficient MDA-MB-231 cells incubated with the CM of LPS-activated macrophages failed to stimulate 3D outgrowth, cancer cells possess a kindlin-2–dependent autocrine signaling loop, which is also involved in stimulation of 3D outgrowth.

Both CSF-1 and EGF have been implicated in the interaction between tumor cells and macrophages (4); tumor cells secrete CSF-1 and respond to EGF, while macrophages secrete EGF and respond to CSF-1 (4). In a transwell migration assay, naïve macrophages were allowed to migrate toward control culture medium, medium supplemented with chemotactic agents, or CM from control or Kindlin-2–deficient tumor cells. Macrophages attracted by CSF-1 were >7-fold higher (P < 0.05) than macrophages attracted to control medium (Fig. 5A and B), consistent with the known property of CSF-1 as a macrophage chemoattractant (3, 4). Similar numbers were obtained with CCL2, another known chemotactic agent (35), and with the macrophage activator LPS (Fig. 5A and B). The number of macrophages attracted by the CM from the K2-CRISPR-231 cells was 5 times lower (P < 0.05) than attracted to the CM from control cells (231-Scram; Fig. 5C and D), confirming that Kindlin-2 is required for the production and secretion of chemoattractants by cancer cells. These results were duplicated with murine 4T1 breast cancer cells (Fig. 5E and F). Kindlin-2 itself was not detected in the CM of the cancer cells by Western blotting (not shown).

To directly implicate tumor cell–secreted CSF-1 in macrophage chemotaxis and the requirement of Kindlin-2 for release of CSF-1 from cancer cells, we preincubated macrophages with a blocking anti-CSF1 receptor (CSF1R). Macrophages preincubated with control IgG successfully migrated toward medium supplemented with CSF-1 or the CM from MDA-MB-231 or 4T1 cells (Fig. 5G, top and H), but anti-CSF1R decreased macrophage recruitment by more than 6-fold (P < 0.05) under all experimental conditions (Fig. 5G, bottom and H). Thus, Kindlin-2 in tumor cells is required for the production and secretion of CSF-1, which in turn, mediates chemoattraction of macrophages.

We used qRT-PCR to measure CSF-1 mRNA levels in parental MDA-MB-231 and 4T1, 2 different pools of Scam and 3 different pools of K2-CRISPR cells and found CSF-1 mRNA levels to be approximately 4-fold lower (P < 0.05) in Kindlin-2–deficient cells (Fig. 6A and B). CSF-1R levels were also significantly lower (P < 0.05) in Kindlin-2–deficient MDA-MB-231 and 4T1 cells compared with controls (Fig. 6C and D, respectively). We used ELISA to quantify the amounts of secreted CSF-1. Kindlin-2–deficient MDA-MB-231 and 4T1 cells released significantly less (∼5-fold, P < 0.05) CSF-1 than their control counterparts (Fig. 6E and F, respectively). As TGFβ signaling activates CSF-1 and CSF1R expression in tumor cells (3), we considered its role in Kindlin-2–mediated production of CSF-1. TGFβ1 ELISA assays showed that Kindlin-2–deficient MDA-MB-231 cells (Fig. 6G) and 4T1 cells (Fig. 6H) released significantly (P < 0.05) less TGFβ1 compared with their Kindlin-2–expressing counterparts. We also found mRNA expression levels of EGF (Fig. 6I and J) and its receptor EGFR (Fig. 6K and L) to be approximately 2-fold and 40% lower (P < 0.05) in the Kindlin-2–deficient MDA-MB-231 and 4T1 cells, respectively. These findings are consistent with the existence of an autocrine loop wherein tumor cells secrete their own EGF (3). Thus, our data suggest that Kindlin-2 is also involved in EGF–EGFR autocrine loop in MDA-MB-231 cells. We further confirmed that EGF⟺TGFβ signaling pathway is required for the Kindlin-2–mediated regulation of the synthesis and secretion of CSF-1 in cancer cells. We specifically targeted both EGF and TGFβ nodes of this signaling axis in MDA-MB-231 and 4T1 cells and assessed for CSF-1 secretion. TGFβ1 and EGF treatment of MDA-MB-231 cells (Fig. 7A) and 4T1 cells (Fig. 7B) activated the TGFβ and EGF downstream effectors, SMAD3 (∼6-fold increase for MDA-MB-231 and ∼20-fold increase for 4T1 cells, compared with the diluent; Fig. 7C) and ERK MAP kinase (∼10-fold increase for 231 and ∼6-fold increase for 4T1 cells compared with diluent; Fig. 7D). These treatments also stimulated secretion of CSF-1 in both cells by approximately 2-fold (Fig. 7A for MDA-MB-231 231 cells and B for 4T1 cells). Importantly, although TGFβ and EGF treatment of K2-CRISPR cells still promoted secretion of CSF-1, the basal levels of secreted CSF-1 were significantly (P < 0.05) lower in the Kindlin-2–deficient cells (∼2-fold decrease in MDA-MB-231-K2-CRISPR and ∼2-fold decrease in 4T1-K2-CRISPR cells, compared with the control cells; Fig. 7A and B). Loss of Kindlin-2 appears to inhibit maximal activation of TGFβ1 and EGF signaling and, thereby, also blunts the stimulatory effects of TGFβ1 and EGF on CSF-1 secretion. We confirmed this finding at the signaling levels: the phospho-SMAD3 levels were 4-fold (MDA-MB-231 cells) and approximately 3-fold (4T1 cells) lower in the K2-deficient (K2-CRISPR) after treatment with TGFβ1, compared with their Scram counterparts (Fig. 7C). Similarly, phospho-ERK levels were 5-fold (MDA-MB-231 cells) and 4-fold (4T1 cells) lower in the K2-deficient (K2-CRISPR) after treatment with EGF, compared with their Scram counterparts (Fig. 7C). Treatment with TGFβ or EGF signaling inhibitors (SB4131542 and ZD1839, respectively) blocked the activation of their respective downstream effectors (Fig. 7C and D) and significantly (P < 0.05) dampened CSF-1 secretion by MDA-MB-231 cells (Fig. 7A) and 4T1 cells (Fig. 7B), supporting the regulatory role of Kindlin-2 in the stimulatory effects of TGFβ1 and EGF on CSF-1 secretion. We also found both EGF and TGFβ signaling to be significantly lower in the K2-CRISPR–derived tumors compared with their Scram counterparts (Fig. 7E for MDA-MB-231- and F for 4T1-derived tumors). Expression levels of CSF-1, CSF1R, EGF, and EGFR were also significantly lower in the K2-CRISPR–derived tumors compared with their Scram counterparts (Fig. 7G for MDA-MB-231 - and H for 4T1-derived tumors; and Supplementary Fig. S4D for tumors of similar sizes), therefore supporting our in vitro findings with in vivo data. Together, our findings confirm the involvement of the EGF⟺TGFβ signaling in the secretion of CSF-1 by cancer cells (3, 4) and expand this pathway to a Kindlin-2⟺EGF⟺TGFβ signaling axis, which regulates CSF-1 release from cancer cells and induces chemotaxis and activation of macrophages within the tumor microenvironment (Fig. 7I).

This study demonstrates that Kindlin-2 supports breast cancer tumor growth and does so at least in part by controlling recruitment of macrophages to the tumor microenvironment. To our knowledge, this is the first study to show that Kindlin-2 influences the tumor microenvironment, thereby providing mechanism by which increased levels of Kindlin-2 in tumors correlates with poor prognosis in breast cancer patients. In addition, we show that Kindlin-2 enhances the invasive properties of breast cancer cells further promoting the neoplastic properties of tumors. We applied a combination of genetic and pharmacologic manipulations, as well as different biochemical and cell imaging analyses in vitro and in mouse models, to investigate the role Kindlin-2 in the modulation of the growth and progression of breast cancer tumors. This study also appears to be the first to use CRISPR/Cas9 gene editing to delete Kindlin-2 in human and murine breast cancer cell lines. We showed that (i) Kindlin-2 is overexpressed in aggressive breast cancer cell lines, both human and mouse, compared with nontumorigenic or less aggressive cell lines, and that the levels of Kindlin-2 correlates with the metastatic potential of these cells, as well as correlates with poor prognosis in patients with breast cancer; (ii) in vitro, Kindlin-2 enhances tumor cell migration and invasion; (iii) in in vivo mouse models, Kindlin-2 is required for rapid breast cancer tumor growth; (iv) the extent of tumor growth depends on the amounts of tumor infiltrating macrophages (36–38), which was severely dampened in Kindlin-2–deficient-derived tumors, suggesting that Kindlin-2 regulates recruitment of macrophages to the tumor microenvironment; (v) in tumor cells, Kindlin-2 is required for the production and secretion of CSF-1, which is necessary for macrophage chemotaxis to the tumor microenvironment; (vi) Kindlin-2 plays a key role in both macrophage–tumor cell paracrine and tumor cell autocrine signaling involving EGF and CSF-1; and (vii) this activity depends upon the capacity of Kindlin-2 to regulate TGFβ signaling. Thus, we have established that Kindlin-2⟺TGFβ signaling axis is a key regulator of CSF-1 secretion by tumor cells, which in turn directs macrophage recruitment.

TGFβ signaling has been found to activate the production and secretion of CSF-1 by tumor cells (34). Our results show that deficiency of Kindlin-2 in cancer cells inhibits the expression and secretion of CSF-1, which is concomitant with a significant decrease in TGFβ1 production and secretion (Fig. 6). These findings clearly suggest that Kindlin-2 regulates TGFβ1 production and its downstream signaling effects on target genes, such as CSF-1. Wu and colleagues (39) recently showed that Kindlin-2 binds to the cytoplasmic kinase domain of TGFβ receptor 1, resulting in TGFβ signaling activation and that Kindlin-2 ablation inhibited TGFβ1-induced Smad2 phosphorylation (39). Indeed, our results (Fig. 7) demonstrated that Kindlin-2 deficiency recapitulated pharmacologic inhibition of TGFβ receptor and significantly decreased CSF-1 secretion. Therefore, our data identify Kindlin-2 as a new element in a Kindlin-2⟺TGFβ⟺CSF-1 signaling axis that is required for macrophage infiltration and tumor growth. Kindlin-2 is a coactivator of integrins (40–42). We and others have shown that loss of Kindlin-2 in many cell types, including breast cancer, inhibits integrin activation (7, 8). Thus, Kindlin-2-mediated regulation of macrophage recruitment to the tumor site may be integrin dependent, and hence, integrins may intercalate into this signaling axis.

Several studies showed the involvement of a tumor autocrine CSF-1 loop that contributes to tumor growth and metastasis (3, 4). Our data corroborate the importance of this autocrine pathway and show that Kindlin-2 not only regulates CSF-1 expression and secretion (Fig. 6), but also promotes CSF1R, EGF, and EGFR, which are all involved in the autocrine loop. Kaplan–Meier plotter analysis for the relationship between CSF-1 levels and survival of breast cancer patients showed a similar pattern to that of Kindlin-2: increased levels of CSF-1 correlate with poor outcome in breast cancer patients (P = 0.00092, Supplementary Fig. S5). Thus, our data support the model shown in Fig. 7I in which Kindlin-2 is involved in both an autocrine and a paracrine CSF-1 signaling loop that activates breast cancer tumor growth.

Cancer tissues with high levels of infiltrating macrophages are associated with poor prognosis and resistance to therapy (43, 44). Our findings demonstrate that Kindlin-2 regulates tumor infiltration by this leukocyte population. Till here, Kindlin-2 modulates the sensitivity to chemotherapeutics in several cancer types (26, 45, 46), thereby providing a link between Kindlin-2–mediated regulation of macrophage and chemoresistance. The hallmarks of cancer as delineated by Hanahan and Weinberg (47) include chemoresistance as well as cancer cell invasion, tumor angiogenesis, and interactions between the tumor and its surrounding microenvironment (47). Kindin-2 has now been associated with many of these cancer hallmarks, including tumor survival, growth, progression and metastasis, angiogenesis, and chemoresistance (7). Our study now identifies another cancer hallmark regulated by Kindlin-2, that is, regulation of tumor–stromal interactions through the modulation of macrophage recruitment raises several important directions for future studies. For example, as noted above, Kindlin-2 is likely to insert integrins into the Kindlin-2⟺TGFβ⟺CSF-1 signaling pathway, but functions have been ascribed to Kindlin-2 that are independent of integrins (48). Which of the cancer hallmarks involving Kindlin-2 are integrin dependent and/or integrin independent remains to be addressed. Nevertheless, at this juncture, our findings establish an additional interface between Kindlin-2 and tumor biology and suggest that inhibition of Kindlin-2 activity may be a breast cancer therapeutic target. Furthermore, the capacity of Kindin-2 to influence the behavior of cells via paracrine and/or autocrine stimulations broadens its reach even to cells lacking this kindlin and to other biological responses.

No potential conflicts of interest were disclosed.

Conception and design: K. Sossey-Alaoui, W.P. Schiemann, E.F. Plow

Development of methodology: K. Sossey-Alaoui, D.J. Lindner

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Sossey-Alaoui, E. Pluskota, D. Szpak, Y. Parker, C.D. Morrison, D.J. Lindner, W.P. Schiemann

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Sossey-Alaoui, E. Pluskota, K. Bialkowska, D. Szpak, Y. Parker, C.D. Morrison, E.F. Plow

Writing, review, and/or revision of the manuscript: K. Sossey-Alaoui, E. Pluskota, K. Bialkowska, C.D. Morrison, D.J. Lindner, W.P. Schiemann, E.F. Plow

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.D. Morrison, E.F. Plow

Study supervision: K. Sossey-Alaoui, W.P. Schiemann, E.F. Plow

The authors thank Dmitriy Verbovetskiy for his assistance with mouse peritoneal macrophage isolation.

This work was supported in part by NIH grants P01 HL 073311 and R01 HL096062 to E.F. Plow.

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.