Report

Direct Proteomic Quantification of the Secretome of Activated Immune Cells

See allHide authors and affiliations

Science  26 Apr 2013:
Vol. 340, Issue 6131, pp. 475-478
DOI: 10.1126/science.1232578

Tracking Secreted Proteins

The proteins secreted by cells provide a flow of information within tissues and are thus of particular interest. However, systematic detection of secreted proteins is tricky because they tend to be present in small amounts within complex mixtures of proteins where other components are very abundant. Meissner et al. (p. 475) developed a method for screening the secreted proteins from cultured mouse macrophages in response to cues that cause inflammation. The amount of contaminating proteins was reduced by culturing the cells without added serum and then sensitive mass spectrometry techniques were used to detect and quantify secretion of nearly 800 different proteins. Secretion was compared from cells lacking the signaling adaptor proteins MyD88 or TRIF, or both. Secretion of some proteins were regulated redundantly and were secreted without one of the adaptors, but others required both signals for release. Some anti-inflammatory proteins were released at later times in response to synergistic signals from both adaptor proteins, perhaps as a fail-safe mechanism to prevent excessive inflammation.

Abstract

Protein secretion allows communication of distant cells in an organism and controls a broad range of physiological functions. We describe a quantitative, high-resolution mass spectrometric workflow to detect and quantify proteins that are released from immune cells upon receptor ligation. We quantified the time-resolved release of 775 proteins, including 52 annotated cytokines from only 150,000 primary Toll-like receptor 4–activated macrophages per condition. Achieving low picogram sensitivity, we detected secreted proteins whose abundance increased by a factor of more than 10,000 upon stimulation. Secretome to transcriptome comparisons revealed the transcriptionally decoupled release of lysosomal proteins. From genetic models, we defined secretory profiles that depended on distinct intracellular signaling adaptors and showed that secretion of many proinflammatory proteins is safeguarded by redundant mechanisms, whereas signaling adaptor synergy promoted the release of anti-inflammatory proteins.

Secreted proteins function as key intercellular messengers in multicellular organisms. Such proteins include cytokines, interleukins, growth factors, hormones, and others that propagate biological information in the body to coordinate homeostasis (1). The investigations of protein secretion have primarily relied on antibody-based methods, whose large-scale use is limited by availability, specificity, and affordability. Strategies for comprehensive, unbiased, and quantitative analysis would therefore be highly desirable, especially for the systems-level characterization of regulated secretory programs (25). We therefore developed a sensitive high-resolution mass spectrometry (MS)–based method to detect proteins in complex secreted protein mixtures with sufficient speed to investigate a large number of conditions. Proteins in supernatants of primary macrophages grown in a 48-well format were digested with trypsin, and the resulting peptide mixtures were directly analyzed in a single-run liquid chromatography mass spectrometry (LC-MS) format (6, 7). We performed LC with 2-hour gradients and analyzed peptides on a benchtop quadrupole-Orbitrap instrument with very high sequencing speed and high mass accuracy in MS and tandem MS (MS/MS) modes (8). Label-free quantification of the MS data and statistical analysis was done in the MaxQuant environment (see the supplementary materials) (9, 10).

Cells of the immune system sense pathogens through specific receptors and release proteins that orchestrate the immune response. The gram-negative bacterial component lipopolysaccharide (LPS) is a potent activator of protein secretion through the stimulation of Toll-like receptor 4 (TLR4). This membrane receptor signals through the proximal adaptors Toll or interleukin (IL)–1 receptor (TIR) domain-containing proteins MyD88 and TIR domain-containing adaptor-inducing interferon (TRIF) (11, 12). Efficient activation of TLR4 requires the delivery of LPS to the receptor signaling platform by proteins also present in serum (13, 14). Key challenges for MS-based methods in secretome studies are the low abundance of immune-modulatory proteins, such as cytokines, in complex mixtures that typically contain highly abundant components. Therefore, we tested bone-marrow–derived macrophages to functionally induce the secretion of known MyD88- and TRIF-dependent proteins in the presence of sufficiently low, MS-compatible amounts of serum (fig. S1). For cells washed once with serum-free medium, we obtained time-resolved samples of wild-type (WT), MyD88-KO, TRIF-KO, and double-KO (knockout) macrophages treated with or without LPS at five time points in at least triplicates, adding up to more than 120 secretome measurements.

MS analysis (Fig. 1A) confirmed the increased abundance of known cytokines from the cells stimulated with LPS in a time- and genotype-dependent manner, whereas protein groups annotated as cytoplasmic were retained inside the cell under all conditions. Numerous proteins from bovine serum were present in all conditions, indicating that our proteomics workflow remains effective in the presence of low amounts of serum (fig. S2, A and B). We calculated relative-change (fold-change) values for protein abundances in cell supernatants with and without LPS stimulation for each genotype and time point by taking the ratio of the individual protein abundances as described in detail in the supplementary materials. These values correlated well within biological replicates (median R = 0.84) and with protein abundance determined by enzyme-linked immunosorbent assay (ELISA) (fig. S2, C to F). MS-based quantitation readily distinguished proteins secreted after receptor ligation from basally secreted or serum-derived ones, which had ratiometric values close to one. Comparison with ELISA experiments verified that the MS sensitivity was in the picogram range.

Fig. 1 The TLR4 induced protein secretome.

(A) Schematic illustration of the experimental setup, proteomics workflow, and data analysis. (B) Number of proteins with the indicated fold change in release upon LPS treatment. Centers of bins are depicted; bars represent the median of replicates with range. (C) PCA of significantly released proteins plotted as median. Numbers indicate time points after LPS stimulation. R2 of the second-order polynomial fits of the genotypes over time: WT = 0.99, MYD-ko = 0.83 and, TRIF-ko = 0.79. PCA vectors PC1 and PC3 contributed 41.7% and 5.3%, respectively, to the genotypic variance.

Our analysis detected 775 proteins that were reproducibly released by macrophages upon TLR4 induction (Fig. 1B, fig. S2G, and database S1). Among these proteins were 364 proteins with annotated extracellular functions, including 52 known cytokines (table S1 and database S1). A third of the proteins released upon LPS treatment were annotated glycoproteins (fig. S3). For two-thirds of them, there was evidence of signal peptides or membrane localization. This still leaves many proteins that may be released by unconventional secretory mechanisms (15). Our technique may thus be particularly suited for the identification of unexpected extracellular proteins or proteins with unknown routes of cellular exit (16). Many of the released proteins had transmembrane regions and could often be divided into those potentially released by proteolytic cleavage or by shedding of membranes on the basis of the topological localization of the identified peptides in the proteins (fig. S3B). A total of 782 proteins were less abundant after TLR4 activation, but their median change was less than 2-fold and they showed no signaling adaptor–dependent dynamics. Therefore, we conclude that LPS stimulation primarily induces protein secretion (Fig. 1, B and C; figs. S4 and S5; and supplementary text).

Our time- and genotype-dependent data showed that the number and individual abundances of secreted proteins increase. It appeared to depend strictly on signal transduction of at least one of the two proximal signaling adaptors, because deficiency in both MyD88 and TRIF (double KO) abolished dynamic LPS-dependent protein secretion (Fig. 1, B and C). Time-resolved principal components analysis (PCA) indicated that the differences in the secretory profiles of the genotypes produced the greatest diversion at 8 hours and 16 hours (Fig. 1C).

Extracellular and secreted proteins undergo a series of posttranscriptional steps to arrive at their site of action, including translation, processing, and transport by secretory machineries. When we measured transcript abundances by microarrays, the measured dynamic range of induction was more than 10 times as high in the MS data. Intriguingly, there were clear differences in the correlation of transcript and secretome changes for different populations of genes (Fig. 2A and fig. S6A). For 290 identified proteins, we observed positive correlations between the regulation of the transcripts and secreted proteins. For some of them, the time-resolved dynamics of the transcripts agreed directly with the secretion, whereas for others, transcript abundance peaked at 4 hours and then decreased while the secreted protein continued to accumulate (Fig. 2, B and C, and fig. S6, B and C). The latter population included many cytokines and secreted proteins, such as interleukin 6 (IL6), C-X-C motif chemokine 2 (Cxcl2), plasminogen activator inhibitor 2 (Serpinb2), and serum amyloid A-3 protein (Saa3), but also a number of proteins not known to be secreted. Unexpectedly, 119 proteins whose secretion was dependent on signal transduction by either MyD88 or TRIF, transcriptome and secretome were negatively correlated (Fig. 2D and fig. S6D). Gene ontology analysis of this class of proteins revealed a strong enrichment for lysosomal content (Fig. 2E) (15, 17). We confirmed the transcriptionally independent release of lysosomal cargo proteins by Western blots and ELISA assays (fig. S7). Based on the observation that proteolytically released (as opposed to membrane-shed) proteins accumulated in the supernatant at later time points of activation, we speculate that proteases, including those of lysosomal origin, may have unanticipated extracellular functions (fig. S3, C and D).

Fig. 2 Relation of transcription to protein secretion.

(A) Frequency distribution of correlations for induced protein secretion to transcriptome. Centers of the bins are indicated. (B and C) Proteins with correlated secretory and transcriptional regulations. (B) Secreted proteins accumulate over time; transcripts decline at late time points. (C) Secreted proteins and transcripts accumulate over time. (D) Proteins with anticorrelated secretory and transcriptional regulations. [(B) to (D)] (Left) Density estimation of correlation. (Middle) Median with interquartile range of secreted proteins. (Right) Median with interquartile range of transcriptional regulation. (E) Gene ontology enrichment for cellular component (GOCC slim) of the anticorrelated proteins using the Fisher’s exact test on the complete data set, plotted versus Benjamini Hochberg false discovery rate (B.H.FDR) corrected significance.

To discover proteins with secretory signatures specific for MyD88 or TRIF, we filtered for proteins that differed significantly in their secreted abundances (fig. S8, A and B) (18). MyD88 accounted for a greater number of more highly secreted proteins than did TRIF, with the highest impact at later time points (Fig. 3, A to D) (19, 20). The proteins whose secretion depended on the adaptors also accounted for much of the secretion in WT cells, with a progressive prevalence of MyD88-mediated signal transduction over time (Fig. 3E and fig. S7, C and E). Differentially secreted proteins included known MyD88-dependent inflammatory cytokines [for instance, growth-regulated alpha protein (Cxcl1), tumor necrosis factor (Tnf), C-C motif chemokine 3 (Ccl3)], TRIF-dependent antiviral cytokines [such as interferon beta (Ifnb), C-X-C motif chemokine 10 (Cxcl10), C-C motif chemokine 8 (Ccl8)], as well as other cytokines, complement components, protease inhibitors, and acute phase proteins (Fig. 3 C and D) (21). We also quantified proteins as differentially regulated that are not known to be released by macrophages, such as the putative TNF-resistance–related protein U90926, illustrating the potential of our approach to identify proteins with unknown extracellular functions and classify their adaptor dependency.

Fig. 3 Effect of signaling adaptors on secretory signatures.

(A) Adaptor dominance ranked as maximal difference between MyD88- and TRIF-mediated secretion. (B) Number of proteins predominantly released by signal transduction through MyD88 or TRIF, respectively. Shades indicate the strength of differential regulation in log10. (C) TRIF-dominated and (D) MyD88-dominated protein secretion. (Top) Median with interquartile range for proteins whose secretion depended on the adaptor. (Bottom) Heat map of differentially regulated proteins (gray denotes missing values). (E) Contribution of MyD88 and TRIF to the secretory response induced by both adaptors. Number of proteins plotted versus the strength of the contribution for the indicated genotype.

In the presence of both signaling adaptors, the secretory output is determined by adaptor interplay, which can be synergistic or redundant (22). To quantitatively evaluate regulatory adaptor crosstalk, we filtered for synergistically released proteins by comparing secreted protein abundances in WT cells to the combined abundances of the single-adaptor KOs as described in detail in fig. S8 and the supplementary materials. The results provided clear evidence for synergistic and redundant mechanisms of protein secretion, and the effects increased as a function of time of LPS stimulation (Fig. 4, A to C). Redundant mechanisms were the most common mode of regulation; for a multitude of proteins, the increase of secretion in the WT and single-adaptor KOs were comparable. This may allow efficient mechanisms to achieve full response regardless of the route that the signal travels. This group included the proinflammatory rather than antiviral proteins (Fig. 4D and fig. S8, D and G), presumably because robustness of secretion of these proteins is crucial to initiate inflammation.

Fig. 4 Signaling adaptor interplay.

(A) Contribution of MyD88 and TRIF to the secretory output illustrated for redundant and synergistic adaptor interplay. (B) Number of proteins with redundant and synergistic regulation over time. Centers of the bins are indicated. (C) Progressive adaptor interplay with median and ranges. Trends for synergistic and redundant protein regulations as median with interquartile range are indicated in red and gray, respectively. (D) Adaptor interplay ranked as maximal difference between WT and a combination of MyD88 and TRIF. Proteins with a maximal redundant regulation <–1.5 are shaded. (E) Synergistic protein secretion. (Top) Median with interquartile range of proteins requiring both adaptors for maximal secretion. (Bottom) Heat map of regulated proteins. (F and G) Proteins with increased secretion in the single-adaptor KOs compared with WT. (F) Number of proteins with the indicated strength of differential regulation in log10. (G) Protein secretion of MyD88- and TRIF-dependent proteins. (Top) Median with interquartile range. (Bottom) Heat map of regulated proteins.

Proteins that require synergistic action of both adaptors—interleukin-10 (IL10), IL19, metalloproteinase inhibitor 1 (Timp1), endothelial lipase (Lipg), and plasminogen activator inhibitor 1 (Serpin1)—were released only at late time points and were at least 100-fold more strongly secreted if both adaptors were present than if only one adaptor was present (Fig. 4, D and E, and fig. S8D). IL10 and its family member IL19 have anti-inflammatory properties (23, 24). Because the absence of either of the adaptors prevents their release, increased presence of proinflammatory molecules would be expected at later time points, and this is what our data show (Fig. 4, F and G). Our data suggest that both adaptors are required for establishing regulatory feedback to control inflammation by simultaneously promoting release of anti-inflammatory proteins and preventing excessive release of proinflammatory proteins. We quantified selected proteins andmRNAs with independent techniques and observed excellent agreement (figs. S7 and S9).

In summary, we have described a combination of high-resolution MS, rigorous computational proteomics, and statistical methods to discover significantly secreted proteins in response to receptor ligation that enables a systematic dissection of signaling pathways and the identification of proteins with transcriptionally independent or unexpected extracellular functions.

Supplementary Materials

www.sciencemag.org/cgi/content/full/340/6131/475/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S9

Table S1

Database S1

References (2529)

References and Notes

  1. Acknowledgments: We thank S. Dewitz, M. Dodel, I. Wagner, K. Mayr, and I. Paron for technical assistance; C. Luber, D. Hornburg, T. Viturawong, and J. Cox for discussions; and A. Zychlinsky, T. Walther, and H. Schiller for careful reading of the manuscript. The microarray data have been deposited in NCBIs Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE41490).
View Abstract

Navigate This Article