Label-Free Characterization of Macrophage Polarization Using Raman Spectroscopy

Abstract

Macrophages are important cells of the innate immune system that play many different roles in host defense, a fact that is reflected by their polarization into many distinct subtypes. Depending on their function and phenotype, macrophages can be grossly classified into classically activated macrophages (pro-inflammatory M1 cells), alternatively activated macrophages (anti-inflammatory M2 cells), and non-activated cells (resting M0 cells). A fast, label-free and non-destructive characterization of macrophage phenotypes could be of importance for studying the contribution of the various subtypes to numerous pathologies. In this work, single cell Raman spectroscopic imaging was applied to visualize the characteristic phenotype as well as to discriminate between different human macrophage phenotypes without any label and in a non-destructive manner. Macrophages were derived by differentiation of peripheral blood monocytes of human healthy donors and differently treated to yield M0, M1 and M2 phenotypes, as confirmed by marker analysis using flow cytometry and fluorescence imaging. Raman images of chemically fixed cells of those three macrophage phenotypes were processed using chemometric methods of unmixing (N-FINDR) and discrimination (PCA-LDA). The discrimination models were validated using leave-one donor-out cross-validation. The results show that Raman imaging is able to discriminate between pro- and anti-inflammatory macrophage phenotypes with high accuracy in a non-invasive, non-destructive and label-free manner. The spectral differences observed can be explained by the biochemical characteristics of the different phenotypes.

Keywords: Raman spectroscopic imaging; chemometric unmixing; macrophage phenotype; principal component analysis and linear discriminant analysis (PCA-LDA).

Figure 1

Validation of macrophage differentiation and activation by flow cytometric analysis of macrophage phenotype-specific cell surface markers. (A) The graph validates the formation of M0 macrophages from isolated monocytes, stimulated with GM-CSF and M-CSF by a high abundance of CD11b⁺ cells within the M0 macrophage samples originating from three donors (Donor 1 = solid line, Donor 2 = dashed line, Donor 3 = dotted line). (B) Successful activation of functional macrophage phenotypes after addition of LPS/IFNy or IL-10 is confirmed by the scatter plot of double-stained M1 and M2 macrophage samples from three donors. (C) The graph summarizes the mean relative abundance of macrophage phenotype-specific cell surface markers and indicates significant differences (* p $\le$ 0.05, ** p $\le$ 0.01).

Figure 2

Biophotonic visualization of different human macrophage phenotypes: (A–F) fluorescence images, (G–I) Raman false-color images. Left column (A,D,G) shows naïve resting macrophages (M0) derived from blood monocytes after differentiation with GM-CSF/M-CSF, middle column (B,E,H) shows pro-inflammatory M1 macrophages activated with GM-CSF/ LPS/ INF-γ, and right column (C,F,I) shows anti-inflammatory M2 macrophages activated with M-CSF/ IL-4. (A–C) Fluorescence images. Cell nuclei are stained with DAPI (blue) and filamentous actin structures of the cytoskeleton with Phalloidin-AlexaFluor555 (green). Arrows indicate filopodia (f) and podosomes (p). (D–F) Fluorescence images. Cell nuclei are stained with DAPI (blue) and intracellular lipid droplets with Nile Red (red). (G–I) False color Raman images of different individual cells generated by N-FINDR analysis. The pixels are colored according to contribution of corresponding endmember group: red—lipids, green—proteins, blue—nuclei, white—environment. Corresponding brightfield images of the cells depicted in panel G-I can be found in Supplementary Figure S4E. Endmember spectra as well as further examples of false-color Raman images are presented in Supplementary Figure S4A–D. Scale bar: 20 µm.

Figure 3

Pre-processed mean Raman spectra of different activated macrophage phenotypes split by donors. Pre-processed Raman mean spectra colored according to their specific phenotype (M0, M1 or M2) and grouped by the donor. The figures on the left provide a detailed view of the 600–900 cm⁻¹ region having most pronounced differences between M0 and M2 phenotypes.

Figure 4

(A) 2D-PCA scores plot of the first two principal components. Scores were calculated by applying PCA to all data used for discrimination (i.e., without cross-validation) (B) PCA-LDA coefficients of the model differentiating between M1 phenotype and M0/M2 phenotypes using two principal components. The colors indicate different cross-validation iterations (olive green—donor 1 was in testing set, orange—donor 2 in testing set, violet—donor 3 in testing set). The image also shows that values are relatively stable.