Abstract

Objectives

   The accumulation of tumor‐associated macrophages (TAMs) is correlated
   with poor clinical outcome, but the mechanisms governing their
   differentiation from circulating monocytes remain unclear in humans.

Methods

   Using multicolor flow cytometry, we evaluated TAMs phenotype in 93
   breast cancer (BC) patients. Furthermore, monocytes from healthy donors
   were cultured in the presence of supernatants from dilacerated primary
   tumors to investigate their differentiation into macrophages (MΦ) in
   vitro. Additionally, we used transcriptomic analysis to evaluate BC
   patients’ blood monocytes profiles.

Results

   We observed that high intra‐tumor CD163‐expressing TAM density is
   predictive of reduced survival in BC patients. In vitro, M‐CSF, TGF‐β
   and VEGF from primary tumor supernatants skewed the differentiation of
   healthy donor blood monocytes towards CD163^highCD86^lowIL‐10^high
   M2‐like MΦ that strongly suppressed CD4^+ T‐cell expansion via PD‐L1
   and IL‐10. In addition, blood monocytes from about 40% of BC patients
   displayed an altered response to in vitro stimulation, being refractory
   to type‐1 MΦ (M1‐MΦ) differentiation and secreting higher amounts of
   immunosuppressive, metastatic‐related and angiogenic cytokines. Aside
   from showing that monocyte transcriptome is significantly altered by
   the presence of BC, we also demonstrated an overall metabolic
   de‐activation in refractory monocytes of BC patients. In contrast,
   monocytes from sensitive BC patients undergoing normal M1‐MΦ
   differentiation showed up‐regulation of IFN‐response genes and had no
   signs of metabolic alteration.

Conclusion

   Altogether, our results suggest that systemic factors skew BC patient
   blood monocytes towards a pro‐metastatic profile, resulting in the
   accumulation of further polarised CD163^high TAMs resembling type‐2 MΦ
   (M2‐MΦ) in the local BC microenvironment. These data indicate that
   monitoring circulating monocytes in BC patients may provide an
   indication of early systemic alterations induced by cancer and, thus,
   be instrumental in the development of improved personalised
   immunotherapeutic interventions.

   Keywords: breast cancer, CD163, IFN responses, IL‐10, tumor‐associated
   macrophages
     __________________________________________________________________

   The mechanisms governing suppressive tumor‐associated macrophage (TAM)
   generation from circulating monocytes remain unclear in humans. In this
   study, we identify tumor‐derived M‐CSF, TGF‐β and VEGF as key factors
   skewing monocytes towards immunosuppressive
   CD163^highCD86^lowIL‐10^high M2‐like macrophages (MΦ). We also describe
   an intrinsic metabolic de‐activation in breast cancer (BC) patient
   monocytes refractory to type‐1 MΦ (M1‐MΦ) differentiation that contrast
   with the active intrinsic IFN‐signalling pathway detected in sensitive
   BC patient monocytes undergoing normal M1‐MΦ differentiation. Thus, the
   combined local and systemic skewing of monocytes may give rise to the
   accumulation of suppressive M2‐like TAMs, contributing to the immune
   response failure and impacting on BC patient outcome.

   graphic file with name CTI2-9-e1108-g007.jpg

Introduction

   During cancer development, a complex microenvironment is formed,
   generating a unique set of signals impacting infiltrating immune cells.
   One consequence is the accumulation of tumor‐associated macrophages
   (TAMs), which are often abundantly present in malignant solid tumors
   and have been associated with tumor invasion, migration and
   angiogenesis,[58]1, [59]2 as well as worse clinical outcome.[60]3,
   [61]4

   Tumor‐associated macrophages are mostly derived from circulating
   monocytes and can be classified by an oversimplified bi‐functional
   model of M1‐MΦ (inflammatory) versus M2‐MΦ (anti‐inflammatory)
   differentiation.[62]5, [63]6 M1‐MΦ are recognised as classically
   activated MΦ endowed with anti‐tumoral properties, while M2‐MΦ
   contribute to tumor development because of their immunosuppressive and
   pro‐angiogenic features.[64]7, [65]8 The use of large‐scale single‐cell
   analyses has recently revealed a new level of diversity in TAM
   populations according to their ontogeny and functional state that
   extends beyond the M1‐ and M2‐like phenotypes.[66]9, [67]10, [68]11
   However, the mechanisms and tumor‐derived factors responsible for
   educating monocytes to TAMs with different phenotypes by tumor‐derived
   factors remain poorly characterised in humans.

   Importantly, several recent pre‐clinical and clinical data highlight
   that cancer progression is driven not only by genetic alterations in
   tumors and interactions with their local microenvironment, but also by
   complex and poorly understood systemic processes, which may have a
   profound impact on anti‐tumor immune responses.[69]12 In this context,
   we have previously shown that circulating monocytes from breast cancer
   (BC) patients fail to differentiate into functional dendritic cells
   (DCs)[70]13, [71]14 and present an altered cytokine profile in response
   to in vitro stimulation.[72]15 Furthermore, other recent studies also
   highlighted the altered profile of circulating myeloid cells in both
   human[73]16, [74]17, [75]18, [76]19 and mouse[77]20, [78]21, [79]22
   cancer‐bearing hosts, strongly suggesting a systemic role for tumors in
   skewing monocytes.

   Using biological and transcriptomic approaches, we report here that
   systemic factors skew the blood monocytes of BC patients towards an
   anti‐inflammatory/pro‐metastatic profile, which, in the local
   microenvironment, are further differentiated into immunosuppressive
   CD163^high M2‐like TAMs. Understanding the mechanisms by which
   tumor‐derived factors influence TAM phenotype, either in circulation or
   within the tumor milieu, can be critical for the development of novel
   anti‐tumor therapeutic approaches.

Results

Accumulation of CD163^+ TAMs is associated with poor survival in BC patients

   Tumor‐associated macrophage infiltrates in primary BC were
   characterised by multicolour flow cytometry, using the gating strategy
   shown in Supplementary figure [80]1a. TAMs were identified as
   CD45^+CD11b^+HLA‐DR^+CD14^+BDCA1^negCD64^+ cells and formed two
   distinct clusters, namely CD163^neg/low and CD163^high, presenting
   variable patterns among patients (Figure [81]1a). A similar profile was
   obtained by analysing a large cohort of BC patients (Supplementary
   table [82]1, n = 93). Total CD14^+ TAMs represented about 25% (± 17.6%)
   of living CD45^+ cells, with CD163^neg/low and CD163^high TAMs
   representing 15% (± 11.6%) and 9.7% (± 11%), respectively (data are
   presented as log[2] values, Figure [83]1b). FACS‐sorted TAM
   subpopulations displayed distinct morphologies, CD163^high TAMs being
   larger and more vacuolated than CD163^neg/low TAMs (Figure [84]1c).

Figure 1.

   Figure 1
   [85]Open in a new tab

   High frequency of CD163^+ TAMs is correlated with higher risk of
   relapse in BC patients. CD163^neg/low and CD163^high TAMs within live
   CD45^+CD11b^+HLA‐DR^+CD14^+CD64^+ cells from BC suspensions were
   analysed by FACS (a) (control isotype in grey). Dot plots shown are
   representative of each TAM profile for CD163 expression (low,
   intermediate and high) among 93 BC patients analysed. (b) Presence of
   total CD14^+ TAMs, CD14^+CD163^neg/low, and CD14^+CD163^high TAMs
   subsets among total live leucocytes (n = 93; horizontal bars represent
   the mean). Values depicted were log(base2)‐transformed from percentages
   obtained by FACS. (c) One representative May–Grünwald–Giemsa staining
   for sorted CD163^neg/low and CD163^high TAMs (objective 40x) obtained
   for one BC patient out of two performed. (d) Different levels of
   CD163^+ TAM infiltration detected in TMAs: 0, low infiltration and 1–2,
   high infiltration. (e) Analysis of the PFS of the 238 BC patients
   according to their high (red line, n = 129) or low (blue line, n = 109)
   level of CD163^+ TAM infiltration. (f) Frequency of CD163^+ TAMs in
   patients according to their BC molecular subtype.

   We analysed the impact of TAMs on patient survival by IHC evaluation of
   the frequency of CD163^+ TAMs on paraffin‐embedded TMA from a
   retrospective cohort of 238 primary BC patients. According to
   pathologist evaluation, tumors were classified as presenting ‘low’ (0)
   or ‘high’ (1–2) CD163^+ TAM infiltration (Figure [86]1d). Univariate
   analysis revealed that high CD163^+ TAM infiltration was correlated
   with more aggressive tumors (high SBR grade, lymph node involvement and
   lymphatic emboli; Supplementary table [87]2) and was associated with
   poor progression‐free survival (PFS) compared to patients with
   low‐CD163 TAM expression (Figure [88]1e; *log‐rank P‐value = 0.024).
   Importantly, the negative impact of high CD163^+ TAM infiltration on
   the PFS was still observed when the analysis was focused on non‐triple
   negative BC (TNBC) patients (n = 207 patients; log‐rank
   P‐value = 0.0022). This result demonstrates that the negative impact of
   high CD163^+ TAM infiltration on PFS (Figure [89]1e) is not exclusively
   attributable to the higher infiltration of TNBC by CD163^+ TAMs (Figure
   [90]1f).

TGF‐β, M‐CSF and VEGF derived from the tumor microenvironment educate
monocytes into suppressive CD163^highCD86^lowIL‐10^high MΦ

   To decipher the impact of the tumor microenvironment on TAM
   populations, we investigated whether soluble factors present in primary
   tumor supernatants could affect the differentiation/function of healthy
   donor (HD) blood monocytes. CD14^+ monocytes purified from HD blood
   were cultured for 7 days in the presence of SNDils (supernatant from
   primary dilacerated tumors) and analysed after 24 h of LPS activation
   for the expression of surface molecules and cytokine production.
   Results were compared to those observed in control M0‐MΦ, M1‐MΦ, M2‐MΦ
   and Mo‐DCs differentiated in vitro under well‐defined conditions. All
   MΦ populations (M0‐MΦ, M1‐MΦ, M2‐MΦ and SNDil‐MΦ) were characterised as
   CD14^+CD64^+BDCA1^low, and Mo‐DCs as CD14^lowCD64^negBDCA‐1^high. Among
   the cells differentiated under controlled conditions, M2‐MΦ displayed
   the highest levels of r‐CD163, while CD163 was lost in M1‐MΦ and Mo‐DCs
   (Supplementary figure [91]1b; Figure [92]2a). Interestingly, we found
   heterogeneous levels of r‐CD163 in SNDil‐MΦ, indicating a
   tumor‐dependent phenomenon (Figure [93]2a and b). A CD163^high
   phenotype (Figure [94]2a, red dots), similar to M2‐MΦ, was obtained in
   51% (15/29) of SNDil‐MΦ, whereas the other SNDils (14/29) induced a
   CD163^neg/low phenotype (Figure [95]2a, blue dots), mostly resembling
   M0‐MΦ.

Figure 2.

   Figure 2
   [96]Open in a new tab

   Tumor environmental factors turn monocytes into
   CD163^highCD86^lowIL‐10^high MΦ. HD CD14^+ monocytes were cultured in
   the presence of 25% SNDils for 7 days, and surface markers and cytokine
   production were evaluated 24 h after addition of LPS. (a) r‐CD163 MFI
   from control APCs (n = 8 independent donors) and SNDil‐MΦ (n = 29
   independent SNDils). (b) Representative pseudocolour plots of
   CD64^+CD163^+ cells of M0‐MΦ and SNDil‐MΦ based on the median obtained
   within each group. (c) Expression of CD86 and (d) production of TNF‐α
   and IL‐10 in CD163^low and CD163^high SNDil‐MΦ or in control APCs. For
   (a–d), experiments performed with, at least, five independent donor
   monocytes and different SNDils (n = 29). (e) Correlation of r‐CD163 MFI
   and IL‐10 production among SNDil‐MΦ. (f, g) SNDil‐MΦ were cultured in
   the presence of neutralising anti‐M‐CSF, anti‐TGF‐β and anti‐VEGF or
   control antibodies during the differentiation process and were
   activated by LPS during the last 24 h to evaluate surface markers (f)
   and cytokine production (g) in CD163^highIL‐10^high SNDil‐MΦ (red dots,
   n = 4; *P ≤ 0.05) and CD163^lowIL‐10^low SNDil‐MΦ (blue dots; n = 3).

   Regardless of r‐CD163 levels, SNDil‐MΦ showed significantly lower CD86
   levels than M0‐MΦ (Figure [97]2c). However, CD163^high SNDil‐MΦ
   produced significantly higher IL‐10 levels (but not TNF‐α) than
   CD163^neg/low SNDil‐MΦ (Figure [98]2d). Of note, CD86 and TNF‐α were
   statistically higher in M1‐MΦ (Figure [99]2c and d), while a
   significant up‐regulation of IL‐10 was observed in M2‐MΦ (Figure
   [100]2d). Interestingly, we also found a positive correlation between
   r‐CD163 and IL‐10 levels considering all SNDil‐MΦ (Figure [101]2e).
   Moreover, the addition of neutralising anti‐IL‐10Rα mAb to M2‐MΦ
   cultures demonstrated that autocrine IL‐10 production is critical for
   the up‐regulation of CD163 (mean MFI ± SEM: anti‐IgG: 4538 ± 779 vs
   anti‐IL‐10Rα: 790 ± 127, n = 3) and PD‐L1 (mean MFI ± SEM anti‐IgG:
   3626 ± 361 vs anti‐IL‐10Rα: 1755 ± 410, n = 3), but did not
   significantly alter PD‐L2, CD80 or CD86 expression (Supplementary
   figure [102]2a).

   To identify SNDil tumor microenvironmental factors responsible for
   M2‐like MΦ differentiation, we quantified 48 cytokines and chemokines
   in SNDils by conducting a multiplexed immunoassay. Elevated amounts of
   CCL2, M‐CSF, TGF‐β1, TGF‐β3 and VEGF were detected in SNDils promoting
   CD163^highIL‐10^high MΦs compared to those promoting CD163^lowIL‐10^low
   MΦ (Supplementary figure [103]2b). The addition of specific Abs
   neutralising M‐CSF, pan TGF‐β and VEGF during differentiation
   significantly impaired CD163, PD‐L1 and IL‐10 induction and led to a
   CD86 increase in SNDils promoting CD163^highIL‐10^high MΦ, whereas no
   modulation was observed for those inducing CD163^lowIL‐10^low MΦ
   (Figure [104]2f and g).

   To investigate the functional consequences of tumor supernatants in MΦ
   differentiation, we evaluated the impact of differentiated MΦ on T‐cell
   proliferation and suppression. CD163^high SNDil‐MΦ strongly suppressed
   CD4^+ T‐cell proliferation in response to anti‐CD3/anti‐CD28 beads
   compared to CD163^neg/low SNDil‐MΦ and M0‐MΦ (Figure [105]3a, upper
   panel) and were as potent as M2‐MΦ (Figure [106]3a, lower panel).
   Furthermore, significant up‐regulation of IL‐10 was observed only in
   co‐cultures containing CD163^high SNDil‐MΦ and M2‐MΦ (Figure [107]3b).
   Mechanistically, the simultaneous blockade of IL‐10/IL‐10Rα and PD‐L1
   partly neutralised the suppressive functions of CD163^high SNDil‐MΦ and
   M2‐MΦ (Figure [108]3c and d), and partially restored IFN‐γ, GM‐CSF and
   IL‐13 production by activated‐CD4^+ T cells (Figure [109]3e). Overall,
   these findings suggest that primary BC‐derived M‐CSF, VEGF and TGF‐β
   triggered IL‐10 production by monocytes, driving their final
   differentiation into immunosuppressive
   CD163^highIL‐10^highPD‐L1^+CD86^low MΦ. Of note, we did not detect any
   specific pattern in SNDil from TNBC compared to other BC tumor
   subtypes.

Figure 3.

   Figure 3
   [110]Open in a new tab

   CD163^highCD86^lowIL‐10^high SNDil‐MΦ suppress T‐cell proliferation.
   SNDil‐MΦ and control APCs were co‐cultured for 4 days with allogeneic
   naïve CD4^+ T cells pre‐activated using expand beads. Evaluation of (a)
   T‐cell proliferation and (b) IL‐10 secretion in the co‐culture
   supernatants (SNDil‐MΦ CD163^low (blue, n = 8); SNDil‐MΦ CD163^high
   (red, n = 11); control APCs (n ≥ 5) (mean ± SEM; *P ≤ 0.05, **P ≤ 0.01;
   ***P ≤ 0.0001 to M0‐MΦ). Impact of neutralising anti‐IL‐10/IL‐10R and
   anti‐PD‐L1 antibodies on (c, d) T‐cell proliferation and (e) cytokine
   production in the same experimental settings as Figure [111]3a
   (*P ≤ 0.05). One representative experiment is shown in c and e out of
   two performed. d shows cumulative data of two independent experiments.

Blood monocytes from BC patients categorised as sensitive or refractory to
GM‐CSF/IFN‐γ M1 differentiation

   Having demonstrated that the tumor microenvironment can drive the
   differentiation of monocytes into suppressive MΦ in vitro, we next
   investigated whether it could influence patient blood monocytes at
   distance via the bloodstream. To achieve this, fresh CD14^+ blood
   monocytes from BC patients were cultured in the presence of a
   GM‐CSF + IFN‐γ (M1‐MΦ) cocktail. Although no difference was observed in
   CD163 levels at the beginning of the culture (Figure [112]4a and b,
   D0‐blood monocytes), 41% (18/44) of BC patient monocytes were
   refractory to M1‐MΦ differentiation since they did not down‐regulate
   CD163 expression under GM‐CSF/IFN‐γ culture as opposed to HD monocytes
   (n = 25; 36.1 ± 4.2% all patients vs 16 ± 2.3% HDs; Figure [113]4a and
   b). We also assessed the production of cytokines/chemokines of freshly
   isolated CD14^+ patient monocytes under LPS stimulation (Figure
   [114]4c). Both patient monocyte subgroups produced significantly lower
   levels of TNF‐α and CXCL1 than HD monocytes. Refractory BC patient
   monocytes produced higher levels of IL‐10, CCL2, TGF‐β1 and TGF‐β3, but
   less IL‐22 and LIF than HD monocytes. Interestingly, refractory BC
   patient monocytes secreted elevated amounts of CCL2, CCL4, CCL5, IL‐1α
   and VEGF, and lower levels of LIF when compared to sensitive patient
   monocytes. Furthermore, considering all soluble factors, linear
   discriminant analysis (LDA) revealed that BC patient monocyte subgroups
   (refractory and sensitive to GM‐CSF) could be separated into 2 clusters
   by a combination of twelve cytokines (Supplementary figure [115]3).

Figure 4.

   Figure 4
   [116]Open in a new tab

   Blood monocytes from BC patients can be categorised as sensitive and
   refractory to GM‐CSF/IFN‐γ and produce distinct levels of cytokines
   upon stimulation. (a, b) BC patient blood monocytes were analysed for
   CD64 and CD163 expression at day 0 (at least n = 7) and after culture
   in M1‐MΦ condition (HD, n = 25; BC patients, n = 44; **P ≤ 0.01).
   a shows one representative dot plot of CD64 vs CD163 expression for
   each group. (c) BC patients and HD CD14^+ monocytes were stimulated by
   LPS for 24 h. Levels of cytokines produced in monocyte supernatants
   from HDs (black, n = 12) and BC patients (blue: sensitive patients,
   n = 12; red: refractory patients, n = 10) are represented (*P ≤ 0.05,
   **P ≤ 0.01, ***P ≤ 0.0001).

Tumor modulates transcriptional programming of BC patient monocytes

   To further investigate the systemic effect of the tumor in modulating
   patient monocytes, we evaluated the transcriptional profile of HD
   monocytes and both GM‐CSF refractory and sensitive patient monocytes
   through microarray analysis. Interestingly, principal component
   analysis (PCA) (Figure [117]5a) segregated patient monocytes from HD
   monocytes. We next applied a filter by considering genes with a log[2]
   fold change ≥ 0.58 and FDR ≤ 0.05 comparing patient versus HD
   monocytes, and obtained 421 differentially expressed genes (DEGs)
   (Supplementary table [118]3). An unsupervised hierarchical clustering
   of these 421 DEGs confirmed the transcriptome alteration of BC patient
   monocytes compared to HD monocytes (Figure [119]5b). Running pathway
   enrichment analysis (MSigDB) (Figure [120]5c and Supplementary table
   [121]3) revealed that the most significant variations were hallmark of
   TNF‐alpha signalling via NF‐κB, down‐regulated in patient monocytes and
   hallmarks of both interferon alpha and gamma responses up‐regulated in
   patient monocytes. Additionally, Gene Ontology (GO) analysis revealed
   down‐regulation of metabolic processes, but up‐regulation in defence
   response to virus in patient monocytes, which is consistent with an
   up‐regulation of the IFN‐α response. Among the top 100 DEGs (Figure
   [122]5d and Supplementary table [123]3), 70 were down‐regulated in BC
   patient monocytes, including immune‐related genes already known to be
   involved in MΦ/DC differentiation (e.g. MAFB and ID2)[124]23, [125]24
   and immunity stimulation (e.g. MAPK6 , TNFRSF12A, CD69, PTGS2, FCAR,
   CD53, CD83 and CXCL8).[126]25, [127]26, [128]27 Thirty genes were
   up‐regulated in BC patient monocytes, comprising genes encoding GTPases
   of the immunity‐associated protein (e.g. GIMAP7 and GIMAP8), and genes
   involved in inflammasome signalling (e.g. NLRC4).[129]28 Importantly,
   using recently published transcriptomic data sets of blood monocytes
   from an independent cohort of BC,[130]18 we confirmed that genes such
   as HBEGF, CD83, CD69, ID2 and HIF1A were statistically down‐regulated,
   while DDX58, NLRC4, TNFSF10, CXC3R1 and CCR2 genes were statistically
   up‐regulated in patient monocytes compared to HD monocytes (Figure
   [131]5e). Altogether, these findings revealed important differences in
   the transcriptional profiles of BC patient and HD monocytes, strongly
   suggesting that tumor development can act systemically, modifying the
   transcriptional profile of circulating monocytes.

Figure 5.

   Figure 5
   [132]Open in a new tab

   Transcriptomic profiles of blood monocytes from BC patients differ from
   HD monocytes. FACS‐sorted CD14^+ blood monocytes from BC patients
   (n = 8) and HDs (n = 8) were submitted to transcriptome analysis. (a)
   Principal component analysis (PCA) plot from HD and patient monocytes
   considering the top 500 most variant genes. (b) Hierarchical clustering
   of all differentially expressed genes (DEGs) between BC patients and HD
   monocytes. (c) Pathway enrichment analysis (hallmark sets and Gene
   Ontology) of DEGs between BC patients and HD monocytes. (d) Bar plot of
   the top 100 DEGs between HD and patient monocytes (log[2] fold change
   ≤ 0.58 and FDR ≤ 0.05). (e) Normalised expression of selected genes
   from our study in an independent cohort of BC from Cassetta et al.
   [133]18

   We next performed GSEA stratifying patient monocytes according to their
   refractory and sensitive status to GM‐CSF/IFN‐γ responses.
   Interestingly, sensitive patient monocytes presented an enrichment in
   sets of genes associated with hallmarks of both IFN‐α and IFN‐γ
   responses when compared to refractory or HD monocytes (Figure [134]6a).
   This observation was thus attributed to the up‐regulation of genes
   including XAF1, RASD2, MX1, IFIT2, IFIT3, IFI44, DDX58, IFITM1, IFITM3
   and ISG15 in sensitive patient monocytes (Figure [135]6b). Conversely,
   refractory patient monocytes displayed a reduced expression of the
   majority of gene sets tested, including metabolic‐related gene sets,
   namely oxidative phosphorylation, fatty acid metabolism and inner
   mitochondrial membrane protein complex (Figure [136]6a and c).
   Furthermore, sensitive and refractory patient monocytes displayed
   distinct DEG modulations with exclusively down‐regulated ones
   (sensitive = 52 genes,refractory = 210 genes) and up‐regulated ones
   (sensitive = 99 genes,refractory = 82 genes), compared to HD monocytes
   (Figure [137]6d and Supplementary table [138]4). In addition, analysis
   of DEGs between refractory and sensitive patient monocytes revealed
   only 16 genes (13 up‐regulated in sensitive, 3 up‐regulated in
   refractory; Figure [139]6e and Supplementary table [140]4).
   Importantly, we extracted gene signatures of refractory and sensitive
   patients (Supplementary table [141]5) using the GeneSign tool[142]25
   with the min–max method and a minimal log fold change (LFC) threshold
   to 1.2. We performed single‐sample scoring analysis to evaluate the
   sensitive and refractory signatures, as well as the hallmark signatures
   for IFN‐α response and OXPHOS identified in Figure [143]6a. We
   demonstrated in an independent cohort of BC patients[144]18 that,
   unlike the refractory signature, the sensitive signature score was
   significantly higher (P = 0.0037) in monocytes with a high IFN
   signature score (Figure [145]6f). This sensitive signature score was
   also more significantly enriched in monocytes expressing a high OXPHOS
   score, compared to the refractory signature score, which was at the
   limit of significance when comparing monocytes with high versus low
   OXPHOS status in the Cassetta cohort (P = 0.045; Figure [146]6f).

Figure 6.

   Figure 6
   [147]Open in a new tab

   Monocytes from BC patients sensitive to M1 differentiation process
   display up‐regulated IFN‐signalling pathways. FACS‐sorted CD14^+ blood
   monocytes from BC patients (sensitive, n = 4; refractory, n = 4) and
   HDs (n = 8) were submitted to transcriptome analysis. (a) BubbleGUM
   analysis for enriched sets of genes (GSEA) in sensitive, refractory and
   HD monocytes. Representative heat‐maps of selected sets of genes
   associated with (b) IFN‐α and IFN‐γ responses, and (c) oxidative
   phosphorylation, fatty acid metabolism and inner mitochondria membrane
   protein complex. (d) Venn diagram of commonly down‐regulated (blue
   circles) or up‐regulated (red circles) genes in monocytes from
   sensitive or refractory BC patients versus HDs. (e) Representation of
   16 DEGs between sensitive and refractory BC patient monocytes. (f)
   Single‐sample scoring analysis of the sensitive and refractory
   signatures in BC patient cohorts from Cassetta et al.
   [148]18 stratified according to their type I IFN response or oxidative
   phosphorylation score.

   Altogether, these results suggest that patient monocytes that normally
   respond to GM‐CSF/IFN‐γ cytokines exhibit an intrinsic activation of an
   IFN‐signalling set of genes that could allow patient stratification.
   Notwithstanding, refractory patient monocytes display non‐functional
   features evidenced by a down‐regulation in pathways associated with
   metabolisms.

Discussion

   We herein describe an important mechanism by which the tumor
   microenvironment educates human blood monocytes, providing a possible
   explanation for the generation/accumulation of suppressive TAMs in
   human BC. We identified soluble factors derived from the primary tumor
   microenvironment, responsible for the differentiation of suppressive
   CD163^highCD86^lowIL‐10^high MΦ. Additionally, we show that blood
   CD14^+ monocytes from BC patients are transcriptionally distinct from
   HD monocytes and could be categorised as sensitive or refractory to
   M1‐MΦ differentiation, under GM‐CSF/IFN‐γ. The transcriptional
   characterisation of these two BC patient monocyte subgroups revealed an
   intrinsic metabolic de‐activation in refractory monocytes that
   contrasted with the active intrinsic IFN‐signalling pathway detected in
   sensitive monocytes, thus allowing them to overcome the systemic
   negative tumor influence. Hence, we propose that this skewed
   differentiation capacity of BC patient monocytes, induced by the
   combined local and systemic skewing of tumor education, gives rise to
   suppressive M2‐like TAMs, contributing to the immune response failure
   and impacting BC patient outcome.

   We identified about 25% of total immune cells infiltrating BCs as
   CD14^+ TAMs that express variable levels of CD163, underlying the
   phenotypic heterogeneity in TAMs. Through high‐dimensional analysis
   approaches, several recent reports have uncovered TAM heterogeneity at
   the single‐cell level in distinct tumor types, including BC,[149]9,
   [150]29 clear cell renal cell carcinoma,[151]10 melanoma[152]30 and
   lung adenocarcinoma.[153]11, [154]31 These studies revealed a great
   variety of distinct phenotypic TAM subsets that co‐exist by sharing the
   expression of well‐established M1 and M2 markers. In agreement with our
   present work, these emerging studies suggest that TAMs may adapt to a
   variety of tumor microenvironmental clues during tumor development, by
   acquiring a large spectrum of states.[155]32 Importantly, recent
   studies have shown that the use of immune checkpoint inhibitors such as
   anti‐PD‐1 and anti‐CTLA‐4 can modify the myeloid high‐dimensional
   landscape in mouse models,[156]33, [157]34 providing new insights into
   its mechanisms of action and its clinical applications.

   In addition, our IHC analysis revealed a positive association between
   high infiltration by CD163^+ TAMs and poor prognosis, in accordance
   with other studies in BC.[158]35, [159]36, [160]37, [161]38, [162]39
   Additionally, recent reports have focused on CD163 status in specific
   well‐defined BC subtypes and under different therapies. In particular,
   CD163^+ TAMs were shown in two independent cohorts of TNBC patients, to
   be associated with worse patient prognosis after adjuvant
   chemotherapy.[163]40 Moreover, BCs harbouring the c‐Myb transcription
   factor known to be associated with a good prognosis[164]41 display a
   reduced macrophage infiltration and express lower levels of CD163
   mRNA.[165]42 This may be because of the fact that c‐Myb regulates the
   transcriptional programme of tumor cells, reducing classical cytokines
   involved in monocyte recruitment and differentiation and angiogenesis
   such as Ccl2, Csf2, Csf3, Vegfa and Vegfc. INOS^+ M1‐like TAMs, in
   contrast to CD163^+ M2‐like TAMs, were associated with improved
   survival of trastuzumab‐treated HER2 amplified metastatic BC
   patients.[166]43 Other very recent studies have demonstrated
   interesting strategies for the in vivo modulation of TAMs. In a murine
   colon adenocarcinoma model, treatment with two mAbs to concomitantly
   block CSF‐1R and stimulate CD40 resulted in the reprogramming of TAMs
   by increasing pro‐inflammatory signals such as IL‐12B, IL‐27, IL‐1β and
   CCL5.[167]44 In addition, in a melanoma tumor model, the depletion of
   CD163^+ TAMs, using an anti‐CD163 antibody conjugated to a doxorubicin
   charged liposome, led to a monocyte influx, the up‐regulation of
   IFN‐related cytokines and an anti‐tumor T‐cell response.[168]45

   Importantly, we described for the first time that conditioned‐media
   supernatants of dilacerated primary human breast tumors promote either
   highly suppressive CD163^highIL‐10^highCD86^low MΦ, resembling
   suppressive TAMs,[169]46, [170]47, [171]48 or
   CD163^lowIL‐10^lowCD86^lowMΦ, displaying moderate suppressive
   functions. Aside from its central role in T‐cell suppression, we
   further documented the contribution of autocrine IL‐10 in the
   phenotypic/functional switch of monocytes into suppressive CD163^high
   MΦ, with low IL‐12p40 production and high PD‐L1 expression, in
   accordance with previous reports.[172]7, [173]49 Indeed, both murine
   and human TAMs have been described to favor tumor progression through
   the promotion of angiogenesis[174]50, [175]51 and the inhibition of
   T‐cell responses through IL‐10 and TGF‐β[176]49, [177]52 or through the
   expression of inhibitory receptors, such as PD‐L1.[178]7, [179]53
   However, the association of immunosuppressive properties of TAMs with
   the CD163 status was poorly explored in humans.

   We demonstrate for the first time that the combination of M‐CSF, TGF‐β
   and VEGF, all present in the BC primary tumor microenvironment,
   converts monocytes into suppressive CD163^highCD86^lowIL‐10^high MΦ,
   whereas GM‐CSF and IFN‐γ, master M1‐MΦ inducers, were not detected in
   SNDils. This observation is concordant with (1) our pioneer works on
   M‐CSF and IL‐6 skewing monocytes into suppressive MΦ‐like cells, and
   blocking Mo‐DC differentiation[180]54, [181]55 and (2) the ability of
   TGF‐β to enhance the suppressive phenotype of monocytes[182]56,
   [183]57, [184]58 and pDC.[185]59 Of importance, blocking M‐CSF, TGF‐β
   and VEGF also impairs IL‐10 production by SNDil‐MΦ. Although
   tumor‐derived factors have been extensively studied using cancer cell
   lines, our study is unique in demonstrating that primary BC
   environment‐derived M‐CSF, VEGF and TGF‐β promote IL‐10 production by
   monocytes, driving their final differentiation into immunosuppressive
   CD163^highIL‐10^highCD86^low MΦ. Importantly, other tumor
   microenvironment‐derived factors may also play a role in the TAM
   polarisation/differentiation. In particular, it was shown that breast
   tumor cell lines with epithelial‐to‐mesenchymal transition (EMT)
   features expressed AXL and that AXL/Gas6 signalling in TNBC patients
   can modulate TAM phenotype by inducing a M2‐like phenotype.[186]40 In
   addition, cancer‐associated fibroblasts producing CXCL12 stimulate the
   migration of TAMs towards tumor‐associated vessels that may promote the
   extravasation of tumor cells to the bloodstream.[187]58 In phyllode
   malignant breast tumors, a particular BC subtype, CCL5 produced by
   tumor cells promotes the polarisation of TAMs into IL18^+ producers,
   via the AKT pathway leading to a positive feedback for tumor promotion
   and aggressiveness.[188]60 Moreover under hypoxia, murine melanoma
   tumors secrete exosomes able to up‐regulate oxidative phosphorylation
   in bone marrow‐derived macrophages, inducing their polarisation towards
   F4/80^+CD206^high M2‐MΦ, favoring tumor development.[189]61

   Outstandingly, our data suggest that tumor‐derived factors affect
   monocytes in the bloodstream of patient. Microarray analysis revealed
   that patient monocytes display a distinct transcriptome by
   down‐regulating genes involved in monocyte differentiation and immune
   stimulation, while producing fewer inflammatory cytokines than HD
   monocytes and these results have been confirmed in an independent set
   of transcriptomic data from patients with primary BC.[190]18 These
   findings are in accordance with our previous studies, showing that
   blood monocytes from primary BC give rise to Mo‐DCs that favor Treg
   responses, partially via TGF‐β.[191]13, [192]14 Furthermore, this
   monocyte skewing phenomenon depends on the tumor burden, since a
   functional recovery of Mo‐DCs is observed after tumor resection[193]62
   or after immunotherapeutic DC vaccination.[194]63 Most interestingly,
   we highlight a heterogeneity among patients, by revealing two distinct
   transcriptome profiles. Patient monocytes sensitive to GM‐CSF have
   intrinsic activation of the IFN pathway that may facilitate their
   differentiation towards M1‐MΦ. In contrast, refractory patient
   monocytes show an overall metabolic de‐activation that may preclude
   their M1‐MΦ differentiation. IFN signalling has been largely reported
   as relevant for anti‐tumor responses[195]64 and predictive of reduced
   bone marrow metastasis.[196]65 IFN signalling may also exert a
   protective role by overcoming tumor education in bloodstream as
   observed in sensitive patient monocytes, confirming previous in vitro
   studies documenting the effect of IFNs on DC or MΦ
   differentiation.[197]66, [198]67 Furthermore, previous studies also
   reported that circulating patient monocytes present alterations in
   metabolic pathways in renal cell carcinoma[199]16 and metastatic BC
   patients,[200]17 corroborating the fact that tumor cells can alter the
   monocytes in their periphery. Our study adds another level of
   complexity by identifying cancer patients with localised primary breast
   tumors presenting circulating monocytes refractory or sensitive to
   GM‐CSF differentiation into M1‐MΦs. Importantly, we validate in an
   independent cohort of BC patients that the sensitive monocyte score is
   enriched in the type I IFN^high group, but the prognostic value of this
   finding should be further validated in a retrospective cohort.

   We also found that LPS‐treated refractory patient monocytes
   preferentially produce immunosuppressive (TGF‐β1, TGF‐β3 and IL‐10),
   angiogenic (VEGF) and metastatic‐related (CCL2, CCL4 and CCL5)
   cytokines. These findings corroborate our previous studies, showing
   that blood monocytes from primary BC patients have a reduced ability to
   secrete TNF‐α in response to IFN‐α stimulation[201]15 and give rise to
   TGF‐β‐producing Mo‐DCs.[202]13 Moreover, previous studies have also
   reported that TAM differentiation at distant sites may promote
   metastasis through the production of chemokines such as CCL2 and
   CCL5.[203]68, [204]69 Thus, the abnormal down‐regulation of cellular
   metabolic pathways found in refractory monocyte transcriptomes suggests
   an altered reprogramming capacity preventing their M1‐MΦ
   differentiation and increasing their suppressive phenotype.

   Soluble factors detected in serum previously associated with worse
   prognosis for BC patients such as TGF‐β,[205]70 VEGF and IL‐6[206]71 or
   M‐CSF[207]72 may alter monocytes at a distance. Alternatively, as the
   presence of tumor cells in the bone marrow of BC patients is correlated
   with worse prognosis,[208]73, [209]74 it may suggest that bone marrow
   invading tumor cells could reprogram monocyte progeny in the bone
   marrow during their differentiation/maturation.

Conclusions

   Overall, we propose a new mechanism of tumor escape in BC patients.
   Indeed, complex tumor microenvironmental products appear to act either
   locally, at the tumor site, or systemically (through the bloodstream
   and/or bone marrow) to trigger two main monocyte profiles: (1)
   metabolic impairment, which leads to differentiation and functional
   biases that favor the tumor growth and (2) IFN enrichment, suggestive
   of patient protection. These aspects should be considered in the design
   of personalised immunotherapeutic approaches targeting autologous
   monocytes/DCs against cancer.

Methods

Biological samples

   The clinical characteristics of all patients are summarised in
   Supplementary table [210]1. Fresh untreated primary breast tumors (BC)
   (n = 93) obtained at the Centre Léon Bérard hospital (CLB) were used
   for TAM analysis. All of these samples were provided by the tissue bank
   of CLB (BB‐0033‐00050, CRB‐CLB, Lyon, France, French agreement number:
   AC‐2013‐1871), after approval from the institutional review board and
   ethics committee (L‐06‐36 and L‐11‐26) and patient written informed
   consent, in accordance with the Declaration of Helsinki. Blood samples
   from 44 patients with untreated primary BC were obtained after signed
   informed consent from CLB and from Pérola Byington Hospital (São Paulo,
   Brazil). All samples were anonymously coded in accordance with local
   ethical guidelines. Healthy donor (HD) blood samples were obtained from
   the ‘Etablissement Français du Sang’ (Lyon).

Preparation of tumor supernatants (SNDil) and of total cells from BC

   Breast tumors tissues were weighed and dissociated mechanically in an
   equal volume (w/v) of RPMI‐1640 medium containing antibiotics
   (penicillin 100 IU mL^−1 and streptomycin 100 mg L^−1, Life
   Technologies, Cailloux‐sur‐Fontaines, France), and supernatants, later
   referred to in the text as ‘supernatant of dilaceration (SNDil)’, were
   harvested, filtered (0.22 µm), aliquoted and frozen at −80°C until
   further use. The mechanically disrupted BC was then enzymatically
   (Collagenase Ia (1 µg mL^−1) and DNase‐I (50 kU mL^−1), Sigma‐Aldrich,
   Saint‐Quentin‐Fallavier, France) digested for 45 min at 37°C in
   serum‐free RPMI 1640 medium with antibiotics under agitation. The final
   cell suspension extracted from BC was washed and resuspended in PBS 2%
   FCS (Eurobio, Les Ulis, France) plus 0.5 mm EDTA (Sigma‐Aldrich) for
   FACS analysis and subsequent experiments.

Isolation of TAMs by FACS and cytospin

   Tumor‐associated macrophages were isolated from BC suspensions by cell
   sorting (FACS Aria‐II, Becton Dickinson) using specific fluorescent
   antibodies and the following strategy: DAPI^neg,
   CD45^+/CD11b^+/HLA‐DR^+/CD14^+/CD64^+ CD163^neg/low or CD163^high
   (control isotypes were used to define the CD163 gate) as shown in
   Supplementary figure [211]1. TAMs were submitted to cytospin
   centrifugation at 600 rpm for 5 min on glass slides. Cytoplasm and
   nuclei morphology were then revealed by May–Grünwald/Giemsa staining.

Immunohistochemistry

   A TMA (600 µm core in triplicate) collecting paraffin‐embedded tumors
   from 238 untreated patients with pathologically confirmed primary BC at
   the CLB was developed for in situ analysis (see Supplementary table
   [212]2 for clinical characteristics). CD163 expression was analysed on
   sections with a mouse IgG1 anti‐human CD163 antibody (clone 10D6,
   0.5 µg mL^−1, reference: NCL‐L‐CD163, Novocastra, Leica Biosystems,
   Nanterre, France). Optimisation of staining was previously performed by
   the Pathology Department at the CLB using a monoclonal mouse control
   isotype (mouse IgG1‐clone MOPC21, reference: PA0996, Novocastra) in
   distinct human tissues.

Blood monocyte purification and differentiation into antigen‐presenting cells
(APCs) in vitro

   Peripheral blood mononuclear cells (PBMCs) were obtained by Ficoll
   density gradient centrifugation (Eurobio). Subsequently, HD monocytes
   were enriched on 51% Percoll density gradient (GE Healthcare Life
   Sciences, Buc, France) and CD14^+ monocytes were purified by negative
   selection (CD14 isolation kit, Miltenyi Biotec, Paris, France)
   according to the manufacturer’s instructions (> 95% pure). Isolated
   monocytes were then differentiated in complete RPMI‐1640 medium
   (containing antibiotics and 10% heat‐inactivated FCS) for 7 days into
   the different control APC populations, that is: cultured in medium
   alone for M0‐MΦ; GM‐CSF (50 ng mL^−1, Schering‐Plough, Dardilly,
   France) and IFN‐γ (20 ng mL^−1, PeproTech, Neuilly‐sur‐Seine, France)
   for M1‐MΦ; M‐CSF (50 ng mL^−1, R&D Systems, Lille, France) and IL‐4
   (20 ng mL^−1, Schering‐Plough) for M2‐MΦ; and GM‐CSF (50 ng mL^−1) and
   IL‐4 (20 ng mL^−1) for Mo‐DC. For some experiments, LPS (Escherichia
   coli 0111:B4; 100 ng mL^−1, Invivogen, Toulouse, France) was added
   during the last 24 h. For SNDil experiments, purified HD CD14^+
   monocytes were incubated with 25% (v/v) of SNDils for 7 days in
   complete RPMI medium (SNDil‐MΦ) and LPS was added at day 6 to activate
   the cells. CD163 relative expression (r‐CD163) was assessed by dividing
   CD163 MFI from each investigated cell population by CD163 MFI of the
   internal control (M0‐MΦ) of each experiment. SNDil‐MΦ
   CD163^highIL‐10^high was considered when r‐CD163 was ≥ 1.6 fold, a
   value similar to the lowest value of M2‐MΦ; and when IL‐10
   concentration was ≥ 415 pg mL^−1, the highest value of IL‐10 produced
   by M0‐MΦ.

   For experiments using BC patient blood, CD14^+ monocytes were purified
   directly from PBMCs by negative selection as above. For some
   experiments, LPS (100 ng mL^−1) was added for 24 h.

Suppression assay

   CD4^+CD45RA^+ naïve T cells purified by negative selection (Magnisort
   Human CD4 Naive T cell Enrichment Kit, ThermoFisherScientific, Waltham,
   Massachusetts, USA) and stained with CellTrace Violet (5 µm, Life
   Technologies) were cultured in U‐bottom 96‐well plates in the presence
   of expand beads (anti‐CD3/anti‐CD28 beads, expand beads, Life
   Technologies). After 30 min of pre‐incubation, LPS‐activated control
   APCs or SNDil‐MΦ were added in plates for 4 days at a ratio of 1 APC:
   2 T cells. CD4^+ T‐cell proliferation was assessed by the analysis of
   CTV dilution by flow cytometry. Condition with expand beads alone
   corresponded to the maximum of proliferation. To assess
   expansion/suppression of CD4^+ T lymphocytes, data were normalised by
   calculating the % of proliferating cells with expand beads + APCs in
   relation to CD4^+ T lymphocytes activated by expand beads alone
   (defined as T‐cell expansion).

Blocking antibodies

   Anti‐IL‐10 (10 µg mL^−1, clone JES3‐12G8, rat IgG2a, AbD Serotec,
   Marnes‐la‐Coquette, France), anti‐IL‐10Rα (10 µg mL^−1, clone 3c.5.2b,
   mouse IgG1, Schering‐Plough) and anti‐PD‐L1 (20 µg mL^−1, clone
   29E.2A3, mouse IgG2b, Biolegend, Saint‐Cyr‐L’école, France) blocking
   antibodies and their respective controls were used in the suppression
   assay and to assess the role of IL‐10 in the modulation of surface
   markers on in vitro generated M2‐MΦ. Anti‐M‐CSF (20 µg mL^−1,
   polyclonal rabbit IgG, Genzyme, Lyon, France) and anti‐TGF‐β
   (10 µg mL^−1, pan‐specific antibody, polyclonal rabbit IgG, R&D
   Systems) antibodies and clinical anti‐VEGF antibody (1 µg mL^−1, human
   IgG1, Avastin^®, Roche, Bale, Switzerland) were used in SNDil‐MΦ
   cultures.

Cytokine detection in supernatants

   IL‐10 and TNF‐α levels were quantified in culture supernatants from
   control APCs or SNDil‐MΦ by ELISA (Life Technologies and Biolegend,
   respectively) and analysed with SkanIt software (Thermo Scientific,
   Dardilly, France). Cytokines and soluble factors were evaluated using
   multiplex Luminex technology according to the manufacturer’s
   recommendations as follows: (1) in supernatants of suppression assay
   with or without anti‐IL10/anti‐PD‐1 blockage with a customised 14‐plex
   (Millipore, Molsheim, France); (2) in SNDils using 4 different kits:
   12plex, 9plex, 8plex and 3plex (Bio‐Rad, Marnes‐la‐Coquette, France);
   and (3) supernatants from LPS‐activated CD14^+ blood monocytes from BC
   patients and HD, with 3 different kits: 23plex, 5plex (Life
   Technologies) and 3plex (TGF‐β; Bio‐Rad). All assays were analysed on
   Bio‐Plex 200 (Bio‐Rad) and analysed with Bio‐Plex Manager software.

Microarray analysis of BC patient monocytes

   CD14^+HLA‐DR^+ monocytes were FACS‐sorted (> 98% purity) from 8 HDs’
   and 8 BC patients’ frozen PBMCs and submitted to total RNA extraction
   using the microRNeasy kit (Cat. No. 74004; Qiagen, Courtaboeuf, France)
   according to the manufacturer’s protocol. After RNA integrity number
   calculation (RIN from samples ≥ 5.7), 500 pg of total RNA (samples
   ≥ 1.3 ng µL^−1) was submitted to cDNA synthesis and DNA labelling and
   hybridisation (GeneChip™ Hybridization, Cat. No. 900720; Applied
   Biosystems, Thermo Scientific, for subsequent microarray analysis
   (Clariom™ S Assay human, Cat. No. 902927; Applied Biosystems) using
   Scanner 3000 7G (Applied Biosystems). Microarray data analysis was
   performed in R (version 3.5.2, supported by R Foundation for
   Statistical Computing, Vienna, Austria). Raw probe intensities were
   quantile normalised and log[2]‐transformed using the Robust Multi‐array
   Averaging (RMA) method implemented in the oligo R package (v.
   1.46.0).[213]75 Differentially expressed genes (DEG) between HD and
   patient monocytes were identified using DESeq2 (v. 1.22.2).[214]76 A
   heat‐map of DEGs was generated using Morpheus from the Broad Institute
   considering a false discovery rate (FDR) < 0.05 and a log[2] fold
   change > 0.58. Hierarchical clustering was based on one minus the
   Pearson correlation distance. Gene Ontology (GO) analysis was performed
   with AmiGO web browser[215]77 and functional analyses (KEGG and MsigDB
   hallmark collection) with clusterProfiler package (v. 3.10.1).[216]78
   Gene set enrichment analyses (GSEA) were represented using
   BubbleGUM.[217]79 Enriched gene sets were considered with a normalised
   enrichment scores (NES) > 1.5 and FDR < 0.05. Molecular phenotypic
   signatures were extracted from monocytes of BC patients with the
   GeneSign tool of BubbleGUM[218]79 with the min–max method and a minimal
   log fold change (LFC) threshold to 1.2. Finally, single‐sample scoring
   analysis was performed using singscore with default parameters (v
   1.2.2) and patients were stratified according to the median of
   IFN‐alpha response or OXPHOS scores.

Statistical analysis

   For in situ analysis of CD163 in IHC studies, the correlation with
   clinical parameters was assessed using either the chi‐square test or
   Fisher’s exact test. The impact on BC patient progression‐free survival
   (PFS) was assessed using the Kaplan–Meier method with SAS software,
   version 9.2 (SAS Institute, North Caroline, USA). For phenotype and
   cytokine production of monocytes, MΦ, DCs, SNDil‐MΦ and tumor analysis,
   we used the one‐way ANOVA test with the Bonferroni post hoc test
   (GraphPad Prism 6.05, GraphPad Software, San Diego, CA, USA). For
   cytokine production from LPS‐stimulated blood monocytes, we used an
   unpaired t‐test with the Mann–Whitney U‐test comparing refractory,
   sensitive and HD monocytes. Linear discriminant analysis (LDA) using
   cytokine production was applied to predict patient monocyte
   classification into refractory versus sensitive groups (XLStat
   software, Addinsoft, Paris, France).

Conflict of interest

   The authors declare no conflict of interest.

Author contributions

   RNR performed the experiments, analysed the data, wrote the paper, and
   designed and interpreted the results with CC, CMC and NBV. CR performed
   the experiments, analysed the data and participated in the
   interpretation and discussion of reported results. IT, OT, CHR, HGG, FL
   and AC participated in patients’ sample selection for fresh tumor
   analysis and IHC analysis of the retrospective cohort of primary BC
   patients. EL, SC and AC performed the statistical analysis for the
   prognostic impact of CD163 expression. MH helped in the experiments of
   cytokine measurement. MH, MA and WR performed bioinformatics analyses
   of microarray results from patients’ monocytes data. EP and JAMB helped
   in the discussion and in data interpretation and reviewed the
   manuscript. CC, CMC and NBV supervised the study and reviewed the
   manuscript.

Ethics approval and consent to participate

   All patients have consent and approved the use of their biological
   tissue. This study was approved by the ethics committee of the Centre
   Leon Bérard Hospital (Lyon, France) and from Pérola Byington Hospital
   (São Paulo, Brazil).

Supporting information


   [219]Click here for additional data file.^ (498.2KB, pdf)


   [220]Click here for additional data file.^ (213.9KB, pdf)


   [221]Click here for additional data file.^ (193KB, pdf)


   [222]Click here for additional data file.^ (61.1KB, pdf)


   [223]Click here for additional data file.^ (64.1KB, pdf)


   [224]Click here for additional data file.^ (97.1KB, xlsx)


   [225]Click here for additional data file.^ (79.3KB, xlsx)


   [226]Click here for additional data file.^ (11.5KB, xlsx)

Acknowledgments