Chemotherapy Induces Breast Cancer Stemness in Association with Dysregulated Monocytosis

This article is featured in Highlights of This Issue, p. 2235

Cytotoxic chemotherapy is used as conventional treatment for cancer in adjuvant and neoadjuvant settings and as induction therapy in de novo stage IV breast cancer. The cancer cell population is influenced directly by the therapeutic agents and indirectly by therapy-associated changes that occur in the tumor microenvironment and possibly also at the systemic level. Compared with adjuvant chemotherapy aiming to target residual cancer cells upon surgical removal of a majority of the tumor mass, neoadjuvant therapy (NT) exerts direct and indirect effects on the entire population of cancer cells that present preoperatively. The promise of NT relies on the potentials to downstage tumors thus allowing optimal surgery, to eradicate clinically undetectable disseminated cancer cells, and to test the efficacy of therapy. In breast cancer and several major human cancers, NT has been shown to significantly improve clinical parameters and outcomes (1–3).

In the treatment of breast cancer, NT is increasingly administered to candidates for breast preservation and/or present with locally advanced or inflammatory breast cancer. Although only 10% to 30% of treated hormone receptor–positive (HR⁺) breast cancer patients exhibit pathologic complete response (pCR) in breast and in regional lymph nodes, the pCR rate of untreated breast cancers characterized as triple-negative (TN) or HER2⁺ breast cancers is over 50% (4, 5). Evidence is emerging about the association between pCR and long-term progression-free and overall survival, particularly in HR⁻ breast cancers. However, after NT, some patients with pCR and more with non-pCR relapse, or progress (in the case of non-pCR) with stage IV metastatic breast cancer, which is ultimately fatal (6). Therefore, understanding both de novo and acquired resistance to NT or induction therapy is of utmost importance. Mechanisms of breast cancer resistance are partly intrinsic to cancer cells, including altered drug metabolism and enhanced damage repair and survival capacities (7). More recently, the complex role of the tumor microenvironment has become clearer; cytokines and other secreted factors produced by stromal fibroblasts, endothelial cells, and certain tumor-infiltrating immune cells have been shown to impede the tumor response to conventional and targeted therapies (8, 9). In addition, accumulating evidence indicates the critical role of a cancer stem–like cell (CSC) phenotype in resistance to therapy.

CSCs are defined as a subset of cancer cells that can proliferate/maintain/differentiate into a new phenotypically heterogeneous tumor, resembling normal stem cells in the self-renewing and pluripotent capacities and undifferentiated gene expression patterns (10, 11). CSCs are implicated in tumor initiation [as the tumor-initiating cells (TIC)], sustained tumor growth (by undergoing self-renewal and generating cells with diverging phenotypes), therapy refractoriness (by expressing drug transporters and remaining dormant in the tumor), and metastasis (as seeds for distant colonization; ref. 12). A variety of cell surface markers and phenotypic markers have been used to enrich CSCs from bulk tumor cells; for human breast cancer, these include CD44⁺/CD24^–/low, expression or activity of aldehyde dehydrogenase 1 (ALDH1), and the ability to escape anoikis and grow into spheres in anchorage-independent conditions (10, 13, 14). The diverse and dynamic nature of CSCs has been recognized: the gene expression markers of CSCs may vary between tumors, and multiple CSC pools may exist within individual tumors; CSCs may undergo genetic evolution during cancer recurrence and metastasis; and nonstem cancer cells may reversibly switch to CSCs (12). Therefore, effective therapies against cancer stemness need to target all CSC subsets existing in the tumor and meanwhile block new CSC emergence, such as those potentially induced by therapy.

Several studies have reported that post-NT breast tumors exhibit a higher CSC frequency (15) and stemness-associated gene expression (16). However, it remains unclear if the CSC population, in addition to escaping the cytotoxic effect of NT, also undergoes therapy-induced expansion. We therefore set out to determine if and how NT affects the CSC traits, to shed light on selecting patients who may benefit from potential combination therapy targeting NT-induced cancer stemness.

Human specimens were obtained from voluntarily consenting breast cancer patients at the City of Hope National Medical Center (Duarte, CA; for Figs. 1, 5A, C–E; Table 1; Supplementary Fig. S1) or at the Tianjin Medical University Cancer Institute and Hospital (Tianjin, China; for Fig. 5B) under Institutional Review Board–approved protocols. Written informed consents were obtained from all patients. The studies were conducted in accordance with recognized ethical guidelines. Patients from the City of Hope were participants of clinical trials NCT01525966, NCT01730833, or NCT00295893 (ClinicalTrials.gov Identifier). Clinical information, including age, tumor stage and pathology, as well as NT starting time, regimen, and response, is summarized in Supplementary Table S1.

Breast cancer cell lines MDA-MB-231, BT474, and 4T1 as well as the monocytic cell line THP-1 were obtained from the American Type Culture Collection and cultured in DMEM (for MDA-MB-231) or RPMI-1640 (for BT474, 4T1, and THP-1) base medium supplemented with 10% FBS. The patient-derived PDX265922 cells originating from a TN breast tumor and propagated in NSG mice as well as the primary cancer-associated fibroblasts (CAF) from the same human tumor are described previously (17). All cells used herein were tested to be free of mycoplasma contamination and authenticated by using the short tandem repeat profiling method at the beginning and end of the study. Aliquots of frozen cell stocks were prepared immediately and used to replace cells in culture every 2 months. Lentiviral constructs expressing shRNAs against CCR1-3 as well as a scrambled control were purchased from GeneCopoeia to generate MDA-MB-231 cells with stable gene knockdown (CCR1: #HSH002198-LVRU6MP; CCR2: #HSH002200-LVRU6MP; CCR3: #HSH002207-LVRU6MP; control: #CSHCTR001-LVRU6MP). For CCR1-3, each set contains 4 shRNA expression constructs coded as #1–4. Production of viruses, infection, and selection of transduced cells were carried out as previously described (17). The two shRNA constructs showing greatest gene knockdown efficiency were shown in Supplementary Fig. S3A–S3C. Recombinant human CCL2, CCL7, and CCL8, as well as the neutralizing antibodies against human CCL2/7/8 and the control goat IgG were purchased from R&D Systems. The CCR2 inhibitor MK-0812 was purchased from Cayman Chemical. Doxorubicin, docetaxel, and SP600125 were purchased from Sigma-Aldrich. RO4929097 was purchased from Selleckchem.

Paired pre- and post-NT human sera were analyzed for changes in cytokine levels by using RayBio C-series human cytokine antibody array C3 (RayBiotech) following the manufacturer's protocol. Levels of MCPs in human and mouse sera were measured by the corresponding ELISA kits. The human CCL2/7/8 and the mouse CCL2/8 DuoSet ELISA kits were purchased from R&D Systems. The CCL7 mouse ELISA Kit was purchased from Cusabio.

Monocytes used in coculture assays were isolated by an EasySep mouse monocyte isolation kit (Stemcell Technologies) from the peripheral blood of C57BL/6 mice after 4 weekly treatments with doxorubicin (4 mg/kg), docetaxel (25 mg/kg), or PBS. THP-1 cells were pretreated with doxorubicin (125 nmol/L), docetaxel (4 nmol/L), or PBS for 48 hours. The coculture was set up in RPMI-1640 medium supplemented with 10% FBS by seeding breast cancer cells in the lower chamber and mouse or human monocytes in the upper chamber of a 0.4-μm transwell insert (Corning). After 48 hours, cancer cells were harvested for analyses.

Single-cell suspensions prepared from cell culture or tissue were analyzed by a ALDEFLUOR assay kit (#01700; Stemcell Technologies) following the manufacturer's protocol. Flow cytometry was performed using a CyAn ADP flow cytometer (Dako) and analyzed by FlowJo software (TreeStar). Cell sorting based on intensity of ALDEFLUOR was performed using a FACSAria III cell sorter (BD Biosciences). Antibodies used for stemness analysis are APC anti-human CD326 (EpCAM/ESA; #324208; BioLegend); FITC anti-human CD44 (#555478; BD Biosciences); and PE anti-human CD24 (#555428; BD Biosciences). For the flow cytometry of bone marrow (BM) hematopoietic cells, femora and tibiae were collected from C57BL/6 mice at the indicated time points, and BM cells were flushed with MACS buffer and analyzed as described (18). To characterize the hematopoietic progenitors, cells were negatively selected for Lin (Ter119-APC-eFluor 780, #47-5921-82; Gr1-APC-eFluor 780, #47-5931-82; B220-APC-eFluor 780, #47-0452-82; CD3e-APC-eFluor 780, #47-0031-82; CD11b-APC-eFluor 780, #47-0112-82; CD4-APC-eFluor 780, #47-0041-82; CD8a-APC-eFluor 780, #47-0081-82; eBioscience) and stained with c-Kit-APC (#17-1171-81), Sca-1-PE-Cy7 (#25-5981-81), CD34-FITC (#11-0341-81), and CD16/CD32-PE (FcgR-III/II; #12-0161-81) antibodies (eBioscience) for 30 minutes before being analyzed by a BD FACSCanto II flow cytometer and BD FACSDiva software (BD Biosciences). Macrophage and monocyte characterization did not include the Lin-negative selection, and cells were stained with CD3-Alexa Fluor 700 (#561388), B220-Alexa Fluor 700 (#557957; BD Biosciences), NK1.1-FITC (#11-5941-81), CD115-APC (#17-1152-82), and F4/80-PE (#12-4801-82; eBioscience) antibodies. The cell populations were identified as: HSC, Lin⁻c-Kit⁺Sca-1⁺; CMP, Lin⁻c-Kit⁺Sca-1⁻CD16/CD32^lowCD34⁺; GMP, Lin⁻c-Kit⁺Sca-1⁻CD16/CD32^highCD34⁺; MEP, Lin⁻c-Kit⁺Sca-1⁻CD16/CD32⁻CD34⁻; macrophages, CD3⁻B220⁻NK1.1⁻F4/80⁺CD115⁻, Low SSC; monocytes, CD3⁻B220⁻NK1.1⁻F4/80⁻CD115⁺. Complete blood count was performed using a Sysmex XT-2000i hematology analyzer (Sysmex Corporation).

Mammosphere culture was performed as previously described (17). Cells pretreated with MCPs for 48 hours were seeded in ultralow attachment 6-well plates (Corning). The number of spheres (diameter ≥ 70 μm) was counted on day 10, and sphere-forming efficiency was calculated based on the number of initially seeded cells.

These procedures were performed as described previously (17). Primers used are indicated in Supplementary Table S2. An annealing temperature of 55°C was used for all primers.

These procedures were performed as described previously (19). Protein extracts were separated by electrophoresis on a 10% or 12% SDS polyacrylamide gel. Protein detection was performed using the following antibodies: NICD (#4147; Cell Signaling Technology); SOX9 (#AB5535; EMD Millipore); NANOG (#3580; Cell Signaling Technology); and GAPDH (#2118; Cell Signaling Technology).

IHC staining of formaldehyde-fixed, paraffin-embedded tumor tissues was performed as previously reported (17) using the following antibodies and dilutions: Ki-67 (#M7240; Dako), 1:50 dilution; NOTCH1 (#ab52627; Abcam), 1:35 dilution; ALDH1 (#611194; BD Biosciences), 1:100 dilution; and CCR2 (#ab176390; Abcam), 1:3,000 dilution. Stained slides were scored according to intensity of staining (−: 0; +: 1; ++: 2; and +++: 3) and percentage of tumor cells staining positive for each antigen (0%: 0; 1%–29%: 1; 30%–69%: 2; and ≥70%: 3). The intensity score was multiplied by the percentage score to obtain a final score, which was used in the statistical analyses.

All animal experiments were approved by the institutional animal care and use committee at the University of California San Diego or the City of Hope Beckman Research Institute. Six-week-old female NOD/SCID/IL2Rγ-null (NSG), BALB/c, or C57BL/6 mice were used. For the mouse serum analyses in Fig. 1E–G, tumor-free NSG mice and those with MDA-MB-231 xenograft tumors of approximately 200 mm³ in the No. 4 mammary fat pad, as well as tumor-free BALB/c mice received 3 weekly i.p. injections with doxorubicin (4 mg/kg), docetaxel (25 mg/kg), or PBS as control. Serum was collected 6 days after the last injection. For the limiting-dilution transplantation in Fig. 1H, MDA-MB-231 cells pretreated with human serum (case T1) or mouse serum (pooled from 5 tumor-free BALB/c mice) for 48 hours were injected into the mammary fat pads of NSG mice at the indicated numbers. Tumor incidence after 4 weeks was shown. The TIC frequency was estimated by extreme limiting dilution analysis (ELDA; ref. 20). For the monocyte depletion experiments in Fig. 2A–C, clodronate liposome or control (Liposoma B.V.; 200 μL for the first treatment and 100 μL thereafter) were injected into BALB/c mice through the tail vein every 2 days starting at 2 days prior to the first chemotherapy treatment. At day 6 after 3 weeks of chemotherapy, blood and BM were collected. For the complete blood count analyses in Fig. 2D and Supplementary Fig. S2, BALB/c and C57BL/6 mice received 4 times of treatment with doxorubicin, docetaxel, or PBS, on days 1, 7, 14, and 21. Blood was collected via the tail vein at each indicated time point. For the bulk cancer cell transplantation in Fig. 4A–C, 10⁶ MDA-MB-231 cells with stable knockdown of CCR2 (shCCR2 #1) or those expressing control shRNA (shCTRL) were injected into the #4 mammary fat pad of NSG mice. When tumor size reached approximately 250 mm³, mice were treated weekly with docetaxel for 3 weeks, and then left free of chemotherapy until the end of experiment. One group with MDA-MB-231-shCTRL tumors also received the CCR2 inhibitor MK-0812 (oral 30 mg/kg twice a day) starting with the chemotherapy and continuing for a total of 30 days. For the ALDEFLUOR^bright cancer cell transplantation in Fig. 4D–J, BALB/c and NSG mice were treated with doxorubicin, docetaxel, or PBS for 3 weeks and then left free of chemotherapy for 1 week, before 10³ freshly sorted ALDEFLUOR^bright 4T1 or PDX265922 cells were injected into the #4 mammary fat pad to assess tumor development. As indicated, some mice received oral MK-0812 or vehicle twice a day at 30 mg/kg starting with the chemotherapy and continuing for 30 days after breast cancer cell engraftment. Tumor volume was determined by caliper measurements.

All quantitative data are presented as mean ± SD. Two-sample two-tailed Student t tests were used for comparison of means of quantitative data between two groups. For multiple independent groups, one-way ANOVA with post hoc Tukey tests were used. Nonparametric Wilcoxon tests were used for comparison of paired pre- and post-NT patient samples. Nonparametric Mann–Whitney U tests were used for comparison between two independent patient groups. Values of P < 0.05 were considered significant. Sample size was generally chosen based on preliminary data indicating the variance within each group and the differences between groups. For animal studies, sample size was predetermined to allow an 80% power to detect a difference of 50%. Animals were randomized before treatments. For experiments in which no quantification is shown, images representative of at least three independent experiments are shown.

All materials, data, and protocols described in the article are available from the corresponding author on reasonable request.

To determine if systemic factors (e.g., cytokines), altered by anticancer therapies, may regulate the CSC phenotype, we first compared the effects of paired pre- and post-treatment sera from TN and HER2⁺ breast cancer patients who had received NT. TNBC patients received carboplatin and a taxane and those with HER2⁺ breast cancers received these agents plus a HER2-targeting therapy. Patient serum was added to MDA-MB-231 TNBC cells or BT474 HER2⁺ breast cancer cells at 10% to replace the FBS in regular medium. In both breast cancer models, the post-NT serum significantly induced cell populations that are enriched for BCSCs (the ALDEFLUOR^bright population in MDA-MB-231 and the ESA⁺CD44⁺CD24^−/low population in BT474; refs. 10, 14) compared with paired pre-NT serum (Fig. 1A; Supplementary Fig. S1A). Using a semi-quantitative cytokine array, we detected elevated levels of CCL2 (MCP-1), CCL7 (MCP-3), and CCL8 (MCP-2) in post-NT sera (Supplementary Fig. S1B). This was confirmed by ELISA (Fig. 1B). When cases were stratified by breast cancer's response to NT or by HER2 status, the blood-borne CSC-stimulating activity as well as levels of all three MCPs were significantly induced following NT in both pCR and non-pCR groups, and in both TNBC and HER2⁺ groups (Table 1), indicating these NT-associated effects occur regardless of tumor response and HER2 status. Antibody-mediated neutralization of MCPs, especially CCL2, or inhibition of the MCP receptor CCR2 that serves as a major receptor for all three MCPs using its antagonist MK-0812 (21), impaired or abolished the ability of post-NT sera to induce CSCs (Fig. 1C and D; Supplementary Fig. S1C). Therefore, MCP signaling through CCR2 plays an essential role in mediating the CSC-inducing activity in post-NT human blood.

We next examined if chemotherapy led to a similar result in mouse serum, and if the presence of a tumor is required for this therapy-induced effect. Female NSG mice with or without MDA-MB-231 orthotopic xenograft tumors, as well as tumor-free BALB/c mice, received 3 weeks of doxorubicin or docetaxel treatment before the serum was collected and added to MDA-MB-231 cells in vitro for evaluation of CSC markers. The results indicate that, similar to humans, mice induce a tumor-independent blood-borne activity after chemotherapy that expands ALDEFLUOR^bright breast cancer cells, and that MCPs as well as their receptor CCR2 also mediate this effect in both immuno-competent and -compromised mice (Fig. 1E–G). Importantly, when varying numbers of MDA-MB-231 cells pretreated with pre-/post-NT patient sera, or with pre-/post-docetaxel sera from tumor-free NSG mice, were injected into the mammary fat pad of NSG mice in a limiting-dilution transplantation assay, postchemotherapy sera from both human and mice exhibited enhanced ability to stimulate tumorigenicity (Fig. 1H). Therefore, the posttherapy serum activity to contain elevated MCP levels and stimulate CSCs is consistently observed in mice and humans, and does not require tumor cells or cells in the tumor microenvironment.

To determine if therapy-induced MCPs are mainly produced by monocytes, we treated BALB/c mice intravenously with the monocyte-depleting agent clodronate prior to and during chemotherapy treatment. Mice receiving clodronate showed lower basal levels of monocytes and a complete blockade of monocyte induction following chemotherapy compared with the control group, which showed monocyte induction in the blood and BM following drug treatments (Fig. 2A). The clodronate-treated mice also failed to exhibit chemotherapy-induced MCP release (Fig. 2B) and CSC-stimulating activity in the blood (Fig. 2C). In the BALB/c model, total white blood cells (WBC) showed an approximately 40% to 60% decrease after 2 to 3 times of treatment with doxorubicin or docetaxel. This was followed by a rapid restoration of WBCs to levels near or slightly above the pretreatment baseline. The dynamics of WBCs parallels lymphocyte response to treatment (Fig. 2D). In contrast, the blood monocyte population significantly expanded during the rebound phase and remained higher than the baseline for more than 20 days after the last drug treatment (Fig. 2D). Similar to monocytes, neutrophils also exhibited rebound expansion, whereas the numbers of red blood cells and platelets did not dramatically fluctuate. The leukocyte dynamics in the blood was consistent with the BM cell numbers, suggesting the posttherapy monocyte expansion in the peripheral blood results from BM reconstitution. This posttherapy rebound is not likely to result from injection-associated inflammation, because injections with PBS did not significantly alter WBC or monocyte numbers, and the two tested drugs injected at the same frequency showed differences in the strength and time course of monocyte regulation.

The postchemotherapy induction of circulating monocytes was also observed in C57BL/6 mice (Supplementary Fig. S2). Using established cell surface markers (18), we analyzed the hematopoietic regeneration in the BM of these mice following chemotherapy treatment. Compared with the control group receiving PBS, both doxorubicin and docetaxel induced an expansion of HSCs that began immediately and climaxed after the last treatment. Although little effects were observed with the common myeloid progenitors (CMP), concurrently increased granulocyte-monocyte progenitor (GMP) and decreased megakaryocyte-erythrocyte progenitor (MEP) populations were detected after 4 times of chemotherapy treatment (Fig. 2E and F). In addition, we observed a significant shift of the F4/80⁺CD115⁻ macrophages to F4/80⁻CD115⁺ monocytes starting after the first treatment and accumulating thereafter. These results suggest the involvement of a series of postchemotherapy events in the BM, including the immediately altered monocyte–macrophage differentiation as well as the stimulation of HSC expansion and their biased lineage commitment to generate more GMPs versus MEPs in a relatively later phase for sustained monocytosis.

We next cocultured breast cancer cells with mouse monocytes isolated before and after chemotherapy or with THP-1 human monocytic cells pretreated with doxorubicin or docetaxel. In a dose-dependent manner, the postchemotherapy monocytes exhibited an enhanced ability to stimulate the CSC traits in breast cancer cells, which was abolished by CCR2 inhibition (Fig. 3A) or MCP neutralization (Fig. 3B). We further found that the chemotherapeutic agents induced the secreted levels of MCPs in monocytes but not in CAF (Fig. 3C), and that this induction occurred at the RNA level through a JNK-dependent mechanism (Fig. 3D), consistent with a previous report showing JNK-mediated upregulation of CCL2 through c-Jun–binding sites in the gene promoter (22). When MDA-MB-231 cells were individually treated with recombinant MCPs, all three MCPs, especially CCL7, induced dose-dependent increases in mammosphere formation and the population of ALDEFLUOR^bright cells (Fig. 3E). Similar results were observed with patient-derived PDX265922 TNBC cells (ref. 17; Fig. 3F). In BT474 HER2⁺ breast cancer cells, which naturally harbor a high percentage of ALDEFLUOR^bright cells (23), MCPs significantly induced mammosphere formation as well as the subset of ESA⁺CD44⁺CD24^–/low cells that are known to contain enriched CSCs (ref. 10; Fig. 3G). In all breast cancer cells tested, MCPs significantly induced the expression of stemness-related genes, including SOX9 and NANOG, at the mRNA and protein levels, and increased the protein levels of the Notch intracellular domain (NICD), indicating activation of Notch signaling (Fig. 3H and I). In support of our previously reported mechanism of CCL2 to activate Notch (17), we found that the ability of all three MCPs to induce stemness genes was completely abolished by a γ secretase inhibitor RO4929097 (Fig. 3J). These results collectively indicate that chemotherapy induces monocyte expression of MCPs, which promote CSC properties by inducing Notch signaling.

Based on the in vitro data, we hypothesized that the MCP-CCR signaling mediates the post-NT CSC induction and tumor progression. To this end, we generated MDA-MB-231 cells with individual knockdown of CCR1-3 (shCCR1-3) or those expressing control shRNA (shCTRL). Stable knockdown of CCR2 but not the other two CCRs efficiently blocked the effects of all three MCPs on inducing stemness-related gene expression and mammosphere formation (Supplementary Fig. S3A–S3C). Pharmacologic inhibition of CCR2 with MK-0812 also abolished MCP-mediated induction of NICD and stemness-related genes (Supplementary Fig. S3D). We therefore focused on CCR2 and injected 10⁶ of MDA-MB-231 cells with stable expression of shCCR2 or shCTRL into the #4 mammary fat pad of NSG mice. When tumor size reached approximately 250 mm³, mice were treated with docetaxel for 3 weeks and then left free of chemotherapy until the end of experiment (Fig. 4A). One group with MDA-MB-231-shCTRL tumors also received the CCR2 inhibitor MK-0812 starting with the chemotherapy and continuing for a total of 30 days. Tumors with CCR2 knockdown grew slower than the control tumors from the beginning and exhibited significantly suppressed regrowth after the completion of chemotherapy. Treatment with the CCR2 inhibitor also significantly reduced posttherapy tumor regrowth (Fig. 4B). The posttherapy tumors with CCR2 knockdown or treated with CCR2 inhibitor also contained a lower content of ALDEFLUOR^bright cells, compared with the control tumors (Fig. 4C).

We next examined if chemotherapy prior to breast cancer cell engraftment altered the host environment to enhance tumor formation from prospectively enriched CSCs. One thousand of ALDEFLUOR^bright 4T1 or PDX265922 cells were injected into the mammary fat pad of BALB/c or NSG mice, respectively, which had previously received three weekly injections of doxorubicin, docetaxel, or PBS (Fig. 4D). Mice pretreated with chemotherapy developed mammary tumors at an earlier time, and presented tumors with larger size, higher content of ALDEFLUOR^bright cells, and increased expression of stemness-related genes (Fig. 4E–J). These effects were suppressed by cotreatment with the CCR2 inhibitor MK-0812 (Fig. 4H–J). Thus, data from mouse tumor models collectively suggest that chemotherapy promotes CSC-mediated tumor growth through MCP-CCR2 signaling, which may be targeted by pharmacologic inhibition of CCR2.

Our data this far suggest that the BCSC population may undergo a dramatic expansion in response to NT through a mechanism involving therapy-induced monocytosis and MCP-CCR2 signaling in cancer cells. To seek additional clinical evidence, we performed complete blood count in two independent cohorts of breast cancer patients to determine changes of various cell populations upon NT. Among patients treated at City of Hope, although the WBC, RBC, and platelet counts were significantly decreased after 6 to 12 weeks of NT, a significant induction of monocyte content was detected (Fig. 5A). Among patients treated at the Tianjin Medical University Cancer Hospital, for which blood was examined weekly, we observed significant induction of monocytes during the rebound phase after each time of chemotherapy (Fig. 5B). These results, together with the previously described MCP elevation in post-NT blood (Fig. 1B), indicate an upstream role of chemotherapy-induced monocytosis in the herein identified CSC regulatory mechanism.

We further analyzed paired pre- and post-NT breast tumors from the City of Hope patients by immunohistochemistry and detected significantly increased levels of nuclear Notch 1 in the tumor, together with increased percentage of tumor cells expressing the BCSC marker ALDH1, upon NT treatment (Fig. 5C and D). Unlike in pre-NT tumors where ALDH1 showed a sparse and scattered staining pattern, ALDH1⁺ cells in post-NT tumors notably existed in patches, possibly suggesting a clonal origin (Fig. 5D). When pre-NT tumors were compared for tumor cell expression of CCR2, those exhibiting pCR during the subsequent NT had significantly lower CCR2 level compared with tumors with non-pCR (Fig. 5E), suggesting that CCR2 expression level which sets tumor responsiveness to MCPs could be a factor influencing tumor response to NT.

Although conventional chemotherapy has been known to induce cycles of myelosuppression and restoration (24), how the process of therapy-induced myeloid cell homeostasis may influence tumor relapse has not been fully studied. We show that chemotherapy induces monocytosis and the consequent systemic elevation of MCP chemokines, which occurs regardless of tumor response to therapy, the expression status of HER2, or even the presence of a tumor. Compared with the HER2⁺ cases, the TNBC cases exhibited higher pre- and post-NT levels of CCL2 and CCL8 and a higher post-NT ALDEFLUOR-stimulating activity in the blood (Table 1). This could suggest the involvement of TNBC-derived factors that also influence MCPs. Our data suggest that HSCs display enhanced commitment to the granulocyte-monocyte lineage following chemotherapy (Fig. 2). During hematopoietic development and homeostasis, cytokines derived from hematopoietic and stromal cells play critical roles in the fate determination of HSCs. Lineage-specific cytokines, such as granulocyte-colony stimulating factor and granulocyte/macrophage-colony stimulating factor secreted by BM mesenchymal cells, can be induced by TNFα and IL1α produced at sites of inflammation, leading to stimulation of multiple stages of granulopoiesis (25). Secretion of inflammatory cytokines has been shown in fibroblasts in response to persistent DNA damage (26) and in peripheral blood mononuclear cells in response to apoptosis induction (27). Therefore, cytotoxic therapeutic agents, by damaging hematopoietic and stromal cells in the BM and/or inducing hematopoietic stresses, may alter the local cytokine network to affect the complex cell dynamics in the BM observed in Fig. 2E and F.

In addition to enhanced monocytosis, chemotherapy also increases the production of MCPs by monocytes but not CAF, which also secrete MCPs but at lower levels (Fig. 3C). This effect is dependent on JNK, a stress-activated protein kinase that acts through the c-Jun transcriptional factor to regulate gene expression during stress response and apoptosis (28). In turn, MCPs may recruit monocytes/macrophages to the tumor, and CCL2 has been shown to induce angiogenesis (29). These events may further regulate the cellular composition and cytokine environment of the tumor, exerting additional effects on CSCs (30). Thus, local productions of MCPs by monocytes/macrophages and nonmonocytic cells (e.g., CAF) in the tumor would likely contribute to the overall effects of MCPs before and after therapy, and be blocked by CCR2 inhibition. Although CAF does not increase MCP production upon chemotherapy treatment, previous reports show that chemotherapy increases the frequency of CAF in primary tumors and stimulates CAF to produce other cytokines that promote CSCs and chemoresistance, including IL17A, IL11, and IL6 (31–33).

Notch signaling promotes self-renewal of adult stem cells (including human mammary stem cells; ref. 34) and lineage-specific proliferation of multipotent progenitor cells (e.g., expansion of luminal progenitor cells in the mammary epithelial hierarchy; ref. 35). We and others have reported that Notch signaling promotes CSC phenotypes and contributes to a higher degree of tumor malignancy (17, 36, 37). The post-NT activation of Notch and acquisition of CSC properties could potentially promote tumor progression and metastasis through the regulation of epithelial-to-mesenchymal transition, angiogenesis, and genes involved in invasiveness (38, 39). Importantly, MCP-directed regulation of Notch may also influence hematopoietic stem cells/progenitor cells (HSC/HPC), as Notch activation has been shown to induce expansion of hematopoietic stem/progenitor cells (40, 41). If true, this may contribute to a feed-forward loop in which MCPs produced by initial monocytes induce an expansion of HSCs and/or monocyte progenitors to produce more monocytes as sustained sources for MCPs.

Our study is based on clinical specimens from NT trials as well as experimental tumor models to simulate NT administered at an early tumor stage and to assess CSC-mediated tumor formation in an immediately posttherapy host environment. The stage of cancer and therapy simulated by these models represents the phase when CCR2 inhibition may have a beneficial effect by antagonizing the therapy-induced, monocytosis-associated cancer stemness. It is worth noting that the induction of CSC traits during NT may also affect a subsequent metastatic event. However, metastasis is ultimately determined by multiple factors influencing cancer dormancy, molecular evolution, reprogramming upon arrival to a new site, as well as adaptation of the metastatic niche, and can occur years after the therapy-induced monocytosis through independent mechanisms (42).

CCR2 in monocytes and HSCs/HPCs mediates the mobilization of these cells from BM to inflammatory sites (43). Therefore, CCR2 inhibitors in phase I/II clinical trials for noncancer diseases (21, 44) may act on both cancer and hematopoietic cells to efficiently block chemotherapy-induced breast cancer stemness. Although higher levels of CCL2 in primary breast cancers are associated with poorer prognosis (45) and its neutralization in mice has been shown to inhibit metastasis (46), a recent study reports that cessation of CCL2 inhibition promotes breast cancer metastasis suggesting the complication of targeting CCL2 as a monotherapy (47). Indeed, a human monoclonal antibody against CCL2 has not shown antitumor activity as a single agent in metastatic castration-resistant prostate cancer patients progressing after docetaxel-based chemotherapy (48–50). These previous results highlight the need to determine the most beneficial settings for MCP-targeting therapy, which may serve to block the therapy-induced CSC phenotype when coadministered with neoadjuvant and adjuvant chemotherapies.

G. Somlo reports receiving commercial research grants from Celgene, other commercial research support from Roche, and is a consultant/advisory board member for Pfizer and Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Liu, X. Ren, G. Somlo, S.E. Wang

Development of methodology: L. Liu, X. Ren, S.E. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Liu, L. Yang, W. Yan, J. Zhai, D.P. Pizzo, P. Chu, A.R. Chin, G. Somlo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Liu, M. Shen, C. Dong, X. Ruan, X. Ren, G. Somlo, S.E. Wang

Writing, review, and/or revision of the manuscript: D.P. Pizzo, X. Ren, G. Somlo, S.E. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Shen, C. Dong, X. Ren, S.E. Wang

Study supervision: G. Somlo, S.E. Wang

This work was supported by NIH/NCI grants R01CA206911 (to S.E. Wang and G. Somlo), R01CA218140 (to S.E. Wang), and R01CA163586 (to S.E. Wang); National Natural Science Foundation of China grants 81472471 (to X. Ren) and 81572913 (to L. Liu); and National Key Technologies R&D Program of China grant 2015BAI12B12 (to X. Ren). Research reported in this publication included work performed in Core facilities supported by the NIH/NCI under grants P30CA23100 (UCSD Cancer Center) and P30CA33572 (City of Hope Cancer Center).

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.