FormalPara Key Summary Points

Why carry out this study?

TAMs are known to play critical roles in the development, progression, and metastasis of breast cancer. However, the breast cancer microenvironment contains several subsets of macrophages, including pro-inflammatory M1, anti-inflammatory M2, and M2-like TAMs.

Previous meta-analyses have predominantly contained studies that used pan-macrophage markers, whereas more recent studies have expanded to include markers that identify macrophage subsets.

Our hypothesis was that markers of M2 macrophages and M2-like TAMs would be a better predictor of survival outcomes in breast cancer compared to non-specific macrophage markers.

What was learned from this study?

In breast cancer, a high density of CD163+ TAMs that express a similar phenotype to alternatively activated macrophages are a better predictor of poor clinical outcomes.

Future studies should use a larger number of antibodies that identify specific macrophage subsets in the breast cancer tumor microenvironment.

Introduction

Breast cancers consist of complex cellular microenvironments where interactions between tumor cells, stroma, and immune cells influence cancer cell growth and response to treatment [1]. One of the most studied immune cell populations is tumor-associated macrophages (TAMs). These play critical roles in the development, progression, and metastasis of breast cancer, and in future risk models could be included together with tumor size and nodal status to guide treatment decisions [2]. However, the lack of reliable and specific markers to identify and differentiate TAM subsets has limited the development of prognostic algorithms or development of targeted therapies. Further complicating targeting of TAMs is that specific subsets may alternatively facilitate or inhibit neoplastic transformation and metastasis [3,4,5]. All macrophages, including TAMs, show a high level of plasticity and variation in phenotype and activation state that are modified by the microenvironment [6]. These differences may explain the discrepant results of studies of TAMs in the biologic regulation of cancers [7,8,9]. It is therefore essential to identify breast cancer TAM subtypes using differentially expressed markers that can be used in association studies of survival outcomes.

Breast cancer TAMs are derived from peripheral blood monocytes that migrate into the tumor and differentiate into several subtypes [4, 10, 11]. Historically, macrophages have been classified as classically activated, pro-inflammatory M1 macrophages, and alternatively activated anti-inflammatory M2 macrophages that secrete immunosuppressive cytokines [3]. This binary model has been superseded by a spectrum model of polarization following the discovery of non M1/M2 macrophages that include TAMs [4]. Although a distinct macrophage subtype, breast cancer TAMs express a number of markers, including CD163, Fc fragment of IgG and C-type lectin domains, which are also expressed by M2 macrophages [4]. Our current understanding of the association between TAM density in breast cancer tissue and clinical outcomes relies on retrospective cohort studies that have used one or two clusters of differentiation (CD) markers that identify M1 and M2 macrophages rather than panels of markers that define TAM subsets [12]. Most studies have used the pan-macrophage marker CD68 to identify TAMs, however this marker does not discriminate between M1 and M2 macrophages and TAMs [4, 13].

A systematic review and meta-analysis performed by Zhao et al. [14] that included 16 studies performed from 1996 to 2016 found that CD68 was more highly associated with lower overall survival (OS) and disease-free survival (DFS) compared to CD163 [14]. However, this analysis was limited by the inclusion of only three studies of OS that used CD163, and one study that included CD206 and CD40. Similarly, a meta-analysis of 13 studies from 2005 to 2018 by Ni et al. [15] included only three studies that assessed 5-year OS and relapse-free survival (RFS) using CD163+ TAM density [15].

Breast cancer has traditionally been classified based on expression of the estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). Although the molecular heterogeneity of cancers that lack expression of these receptors has been well described, they are often grouped as triple-negative breast cancer (TNBC) [16]. In addition to the importance of TAM markers, it has become increasingly recognized that breast cancer subtype and microanatomical location influence TAM phenotype and behavior [17].

Following the publication of the two latest systematic reviews and meta-analyses of TAMs and breast cancer [14, 15], several studies have been published that analyzed the association between breast cancer survival outcomes, CD163+ TAM density, TAM location in the tumor, and TNBC. The present systematic review and meta-analysis was performed to update the current knowledge on the significance of breast cancer TAM phenotype and focus on studies that report associations between TAM density and OS, progression-free survival (PFS) and breast cancer-specific survival (BCSS).

Methods

A systematic search of Medline (from 1950 to August 2020), Embase (from 1947 to August 2020), Web of Science (from 1900 to August 2020), and Google Scholar (first 30 pages of results) was used to identify relevant articles. Search terms used were “breast cancer”, “breast tumor” or “breast neoplasm” and “macrophages” or “tumor-associated macrophage” and “survival”, “prognosis” or “progression” as text word search and medical subject headings where applicable. The references of identified articles were manually searched for additional relevant studies. The preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines were followed [18]. This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors. Consistent with The National Statement on Ethical Conduct in Human Research, this systematic review and meta-analysis of published and non-identifiable data is exempt from the requirement for ethical review.

Study Selection

We included studies that met the following inclusion criteria: (1) proven diagnosis of breast cancer by pathology; (2) reported TAM density by CD68 or CD163 using immunohistochemistry; (3) assessment of OS, PFS, or BCSS with respect to TAM density; (4) risk reported as a hazard ratio (HR); (5) 95% confidence interval (CI) reported; and (6) full-text English language studies.

Data Extraction

One author (EA) extracted the data, which were checked by a second author (JM). HRs with 95% confidence intervals were collected and checked using reported p values. The names and affiliations of authors were cross-checked to ensure duplicate data were not included.

Statistical Analysis

Subjects were divided into those with (1) a low density of TAMs (low TAM) and (2) a high density of TAMs (high TAM). The study HRs represent the risk for the “high density of TAMs” group compared to the “low density of TAMs” group. There was some clustering of HR effect sizes within studies for different TAM groups because the HR in each case used the same “low density of TAMs” reference group. Consequently, the validity of a two-level meta-analysis model was checked against a three-level model, where level 1 was individual participant information, level 2 the high density of TAMs group within a study, and level 3 the individual study. A comparison of model fit between the two- and three-level models using the Akaike information criterion (AIC) showed that the two-level model was preferred. All analyses were performed using R Statistical Software https://www.r-project.org/. The primary R package used for the analysis was the “meta” package and for the multilevel analysis, the “metafor” package. For the meta-analyses, the metagen function from the meta package was used and the restricted maximum likelihood method for random effects [19,20,21]. The Hartung–Knapp adjustment was also used for random effects analysis [22].

Results

Description of the Studies

Search results yielded 1354 articles including six studies identified through citation chaining and handsearching. Of these, 1279 were excluded based on title and abstract. After careful review of 81 full-text articles, one was excluded due to analyses of duplicate data sets and 48 were excluded that did not meet the inclusion criteria, leaving 32 articles for inclusion in the meta-analysis (Fig. 1). Where articles met inclusion criteria but did not report OS, PFS, or BCSS with HRs, the authors were contacted for clarification.

Fig. 1
figure 1

Process of study selection for the meta-analysis

Full size image

All studies were retrospective cohort studies and used immunohistochemical staining to identify TAMs in paraffin-embedded tumor samples (Table 1). Pelekanou et al. [23] and Mohamed et al. [24] included two separate populations within one article and these were analyzed separately. There was heterogeneity in methods for determining TAM density (Table 1). Furthermore, for each method, there were no standardized cut-offs to define the high density of TAMs. For example, the definition of high density for HPF ranged from > 5 to > 61 per HPF. With respect to tissue distribution of TAMs, there were 12 studies reporting density in tumor stroma and 11 studies reporting density in tumor nest.

Table 1 Characteristics of eligible studies
Full size table

High Density of TAMs Survival Outcomes

OS was reported in 20 studies that included CD68+  and/or CD163+ TAMs (Table 1). To determine whether clustering should be considered, the model fit for the full three-level model was compared to the reduced two-level model. AIC was lower for the two-level model compared to the three-level model (93.8 vs. 95.8) that indicated negligible clustering within studies [55], and the two-level model, with a lower AIC, was used for the analysis. The OS meta-analysis showed that a high density of TAMs was associated with a lower OS (HR 1.69, 95% CI 1.37–2.07) (Fig. 2A).

Fig. 2
figure 2figure 2

Forest plots of HRs for TAM density and outcomes. Forest plots and meta-analyses of studies evaluating HRs of high density of TAMs compared to low density. A OS, B PFS, C BCSS. Note: Where reported, CD marker and tumor location were analyzed separately

Full size image

PFS was reported in 19 studies. The AIC was lower for the two-level model compared to the 3-level model (83.26 vs. 85.26) and the two-level model was used for the analysis. PFS meta-analysis showed high density of TAMs was associated with a lower PFS (HR 1.64, 95% CI 1.35–1.99) (Fig. 2B).

BCSS was reported in 9 studies (Fig. 2C). The AIC for the two-level model was again lower than the three-level model and the reduced model was used for the meta-analysis. BCSS meta-analysis showed the summary HR was 1.32, however the confidence interval crossed unity (95% CI 0.82–2.13). There were fewer studies for BCSS and high heterogeneity [Q (df = 15) = 57.09, p < 0.0001].

Sub-group Analyses

High density of CD68+  TAMs versus CD163+ TAMs

There were 19 studies that analyzed the density of CD68+  and/or CD163+ TAMs, including Pelekanou et al. [23] (Table 1, studies 21a and 21b included in the same report). Studies where the CD marker type was “Other” were omitted. The forest plot with subgrouping by CD marker type showed lower OS for a high density of CD163+ TAMs (HR 2.24; 95% CI 1.71–2.92) compared to a high density of CD68+  TAMs (HR 1.5; 95% CI 1.12–2) (Fig. 3A). The random effects test for this difference was significant (p = 0.03). There was a trend for lower PFS for high density of CD163 + TAMs (HR 2.03; 95% CI 1.51–2.73) versus CD68+  TAMs (HR 1.53; 95% CI 1.18–1.98), however this was not significant (Fig. 3B). There were nine studies that reported the density of CD68+  and/or CD163+ TAMs and BCSS, with no difference between high densities for each CD marker (not shown).

Fig. 3
figure 3

Forest plots of HRs for high density of CD68+  TAMs compared to high density of CD163+ TAMs. A OS, B PFS

Full size image

High Density of TAMs in Triple-negative Breast Cancer

TAM density in TNBC and OS was analyzed in 20 studies, including Pelekanou et al. [23]. TNBC cases with a high density of TAMs had a lower OS (HR 2.81, 95% CI 1.35–5.84) compared to non-TNBC cases (HR 1.57, 95% CI 1.27–1.93) (Fig. 4A) although this difference was not significant (p = 0.07). Similarly, there was a lower PFS in high density of TAMs in TNBC cases (HR 2.37; 95% CI 0.99–5.65) versus non-TNBC cases (HR 1.56; 95% CI 1.28–1.9) (Fig. 4B) that was not significant. There were no eligible studies that reported TAM densities and BCSS in TNBC.

Fig. 4
figure 4

Forest plots of HRs for high density of TAMs in TNBC. A OS, B PFS

Full size image

Location of TAMs in the Tumor

There were seven studies that analyzed OS by location of TAMs in the tumor nest (TN) or tumor stroma (TS). The forest plot with subgrouping by tumor location showed no difference in OS for high TAMs in the TN (HR 1.72; 95% CI 1.26–2.35) compared to the TS (HR 1.98; 95% CI 1.05–3.76) (Fig. 5A). In seven studies, there was a trend for lower PFS for high density of TAMs in the TS (HR 2.16; 95% CI 1.39–3.35) compared to the TN (HR 1.37; 95% CI 1.0–1.89), however this was not significant (p = 0.07) (Fig. 5B). There was lower BCSS in tumors with a high density of TAMs in the TS (HR 5.99; 95% CI 0.72–49.89) compared to TN (HR 0.79; 95% CI 0.22–2.77; p = 0.0004), however there were only two studies in this analysis (not shown).

Fig. 5
figure 5

Forest plots of HRs for high density of TAMs and tumor location. A OS, B PFS

Full size image

Publication Bias

To evaluate publication bias for each outcome, we used a funnel plot analysis of standard error versus random effects summary estimate. For OS, most studies were clustered towards the top indicating high precision (Fig. 6A). There were very few small studies and on inspection the plot was symmetric. The heterogeneity test was significant, with Q (df = 41) = 224.19, p < 0.0001. Compared to OS, the funnel plot analysis for PFS showed a broader horizontal scatter with slight skewing of studies to the right-hand side (Fig. 6B). The heterogeneity test was significant, with Q (df = 40) = 240.95, p < 0.0001. For BCSS, the funnel plot showed symmetry albeit with outlier studies on both sides (Fig. 6C).

Fig. 6
figure 6

Funnel graphs for assessment of potential publication bias in studies of density of TAMs in patients with breast cancer. A OS, B PFS, C BCSS

Full size image

Discussion

Breast cancer TAMs are derived from circulating monocytes that differentiate under the influence of the tumor environment and promote progression and metastasis [12, 56]. TAMs express distinct proteins that can be used for prognostic algorithms and provide targets for therapy [12]. To date, however, a limited number of proteins have been included in association studies of TAM density and breast cancer survival outcomes. The most used marker in survival studies has been the pan-macrophage marker CD68 that in more recent studies has been combined with CD163 in immunohistochemistry studies of cancer tissue [23, 27, 32,33,34, 36, 38, 48, 53]. To distinguish from classically activated M1 macrophages, CD163 has been identified as a marker for alternatively activated M2 macrophages that are stimulated by IL-4 and IL-13; and secrete IL-10 and transforming growth factor β [57]. Although the binary model of M1 and M2 macrophages has been superseded by a spectrum model [4], most TAM subsets also express CD163 and its expression has been associated with other breast TAM markers including CD169 [12]. Considering that the recent studies of TAM density and prognosis have included CD163, we hypothesized that TAM expression of CD163 would be a better predictor of survival outcomes compared to CD68. A systematic review and meta-analysis of association studies comparing the densities of CD68+  and/or CD163+ TAMs and survival outcomes showed that a high density of TAMs (CD68+ and/or CD163+) was associated with decreased OS and reduced PFS (Fig. 2A, B, respectively). Furthermore, a high density of CD163+ TAMs was a better predictor of shorter OS compared to CD68+ TAMs (Fig. 3A). Although there was a trend for lower PFS for high density of CD163+ TAMs, this was not superior to CD68+ (Fig. 3B). There were fewer studies for BCSS and high heterogeneity. Consequently, the summary HR CIs crossed unity for high density of TAMs (CD68+ and/or CD163) and sub-group analyses of CD163+ TAMs, TNBC cases, and TAM location in the tumor.

In relation to the better predictive value of the high density of CD163+ TAMs compared to CD68+, our finding differed from a meta-analysis by Zhao et al. that showed CD163 was not superior to CD68 [14]. Zhao et al. included studies from PubMed (MEDLINE), Web of Science, and Embase from inception to July 1, 2016. There were 16 studies that met the inclusion criteria, including TAMs expressing CD68, CD206, and/or CD163 and correlations between TAM density and OS, DFS, RFS, BCSS, and clinicopathological features. A high density of TAMs of any phenotype was associated with poor OS (HR 1.5; 95% CI 1.2–1.88), however subgroup analysis of CD markers showed CD68 (HR 1.83; 95% CI 1.41–2.38) was a better predictor compared to CD163 (HR 1.13; 95% CI 0.62–2.07). Subgroup analyses for CD markers and DFS, RFS, and BCSS were not performed [14].

A more recent meta-analysis by Ni et al. included studies through February 2018 [15]. Analyses were performed on 13 studies that included densities of CD163+ or CD68+  TAMs, survival outcomes and clinicopathological features. Only three studies reported 5-year OS and RFS based on CD163+ TAMs. Considering the limited number of studies, high density of CD163+ TAMs showed shorter 5-year OS [risk ratio (RR) 2.12; 95% CI 1.09–4.13] and RFS (RR 1.93; 95% CI 1.23–3.04). Similarly, a high density of CD68+  TAMs predicted lower 5-year OS (RR 1.58; 95% CI 1.35–1.84) and RFS (RR 1.74; 95% CI 1.44–2.11). Non-metastatic breast cancer was an eligibility criterion and risk ratios (RR) were used, however our analysis only included studies reporting the HR, which is preferred for analysis of study events over time in longitudinal studies [58].

There is evidence that TAM polarization is influenced by cancer subtype [59], and molecular subtypes of breast cancer are classified as ER+, PR+, HER2+, and TNBC based on receptor expression. This classification is used to guide prognosis and therapy and the presence of ER, PR, and HER2 overexpression has enabled the development of targeted therapies [60]. However, treatment of TNBC currently lacks an option for targeted therapy and carries a poor prognosis. Although a biologically heterogenous disease [16], in TAM studies TNBC has been studied as a single entity. There are conflicting results relating to the prognostic significance of TAMs in TNBC [26, 33, 61]. In the Zhao et al. metanalysis [14], only one study of TNBC was included that reported outcomes for high and low density of TAMs [52]. This study showed a shorter DFS for high density of CD163+ TAMs (HR 2.2; 95% CI 1.53–3.16) [52]. In the Ni et al. study [15], a subgroup analysis found associations between TNBC and both higher density of CD68+  TAMs (RR 1.90; 95% CI 1.63–2.21) and CD163+ TAMs (RR 1.46; 95% CI 0.64–3.33), although only three studies of a high density of CD163+ were included, and the summary RR confidence interval crossed unity [15]. In our study, there was a trend for lower OS associated with high density of CD163+ and/or CD68+  TAMs in TNBC (HR 2.81, 95% CI 1.35–5.84) compared to non-TNBC cases (HR 1.57, 95% CI 1.27–1.93), although this analysis was limited to four studies and only three included a high density of CD163+ TAMs [23, 27, 33, 53]. For PFS, the HR estimate for TNBC was higher at 2.37 compared to 1.56 for non-TNBC, but there was overlap in the CIs (Fig. 4B).

Breast cancer progression is influenced by interactions between tumor cells, stromal cells, and other cellular elements. The TS is the source of growth factors, chemokines, and cytokines [62, 63], and the ratio of TN to TS predicts outcomes in several cancers, including breast cancer [64]. The anatomical location of TAMs within the tumor stroma and nest influences macrophage polarization [10, 12], and consequently associations between survival outcomes and TAM density in the TN and TS have been studied. However, our meta-analysis was limited by low numbers of studies that assessed OS (n = 7), PFS (n = 6), and BCSS (n = 2). There were no differences in OS and PFS for high density of TAMS in the TS or TN, however a high density of TAMs in the TS was associated with lower BCSS (HR 5.99; 95% CI 0.72–49.89) compared to TN (HR 0.79; 95% CI 0.22–2.77) (p = 0.0004). The Zhao et al. [14] meta-analysis found lower survival outcomes in cases with high density of TAMs in TS compared to TN, however like our meta-analysis, there were a limited number of studies analyzed.

Although it is well documented that the presence of CD68+  TAMs in breast cancer is associated with a poor prognosis, it is only recently that bulk- and single-cell RNA and DNA sequencing have enabled an understanding of the heterogeneity of macrophages that infiltrate breast cancer tissue [12, 65]. The aim of these studies is to identify proteins differentially expressed by pro-tumor TAMs that can be targeted by treatments. Identifying prognostic factors and protein targets while avoiding off-target toxicities requires identification of gene and protein signatures that are specific to TAM subtypes, and TAM-specific mRNA signatures have been identified that are associated with poor survival outcomes [12, 66]. A highly upregulated gene in one of these TAM signatures is SIGLEC1 that encodes CD169 [12]. In a study by Cassetta et al., high expression of SIGLEC1 corresponded with high CD163 mRNA, and CD169 protein expression combined with CD163, CSFR1, CCR2, and CD68 identified three breast cancer TAMS (CSFR1+CCR2−CD68+CD163+SIGLEC1−, CSFR1+CCR2−CD68+CD163+SIGLEC1+, and CSFR1+CCR2−CD68+CD163−SIGLEC1+) that are enriched in TNBC cases [12]. Our finding that the density of CD163+ TAMs is a better marker for survival outcomes compared to CD68+  TAMs contributes to the growing evidence that specific TAM subgroups dictate tumor progression, metastasis, and ultimately survival outcomes. A major limitation to previous studies, and consequently meta-analyses, is the lack of a definition of high versus low TAM density (Table 1). Similarly, the definition of TS and TN is highly variable amongst studies. Consequently, there is a need to standardize TAM density and tumor microarchitecture. The significance of TAM phenotype and density may vary according to the breast cancer subtype, and between the early and metastatic setting. However, only six studies included samples from metastatic stage 4 breast cancer [25, 27, 28, 44, 54, 67], and only one study included prognostic data based on stage and TAM density [54]. Zhang et al. [54] demonstrated that in both early breast cancer (stage I–II, and node negative) and invasive breast cancer (stage III), high CD68+  TAM density was associated with lower OS. Further studies are required to characterize the significance of TAM infiltration in the early and metastatic settings. A further limitation to this study was the reliance on CD68 and CD163 as markers of TAMs. There were several studies that used other immunohistochemical markers including CD204 [39], CD11c+ [34], CD206 [67], PCNA [41, 42], MMP-9 [23], and CD169 [46]. However, due to small numbers, these were not included in this analysis.

Conclusions

TAMs play critical roles in the development, progression, and metastasis of breast cancer, and could be included in future risk models and provide targets for treatment. We found that a high density of TAMs is associated with poor survival outcomes in breast cancer. Furthermore, a high density of CD163+ TAMs is a better predictor of poor outcomes, suggesting that TAMs expressing a similar phenotype to alternatively activated macrophages are responsible for tumor progression and metastasis. The studies included in this meta-analysis were limited by the low specificity of markers used to identify TAMs, and heterogeneity in definitions of TAM density and location in the tumor. It is expected that advances in microscopy technology, including the development of techniques that combine immunofluorescent imaging of several hundred proteins in a single sample at subcellular resolution, will enable identification of protein signatures of TAM subtypes that will complement existing single cell genomic studies [68].