Quantification and statistical analysis

For the experiment measured in DDA mode, MS raw files were processed by the MaxQuant version 1.5.0.38⁶¹ and fragments lists were searched against the human UniProt FASTA database (21,039 entries, August 2015)⁶² with cysteine carbamidomethylation as a fixed modification and N-terminal acetylation and methionine oxidations as variable modifications. For phosphoproteome analysis, we also added serine/threonine/tyrosine phosphorylation as variable modification. We set the false-discovery rate (FDR) to less than 1% at the peptide and protein levels and specified a minimum length of seven amino acids for peptides. Enzyme specificity was set as C-terminal to arginine and lysine as expected using trypsin and lysC as proteases and a maximum of two missed cleavages.

For experiments measured in DIA mode, MS raw files were processed by the Spectronaut software version 13 (Biognosys⁶³). First, hybrid libraries were generated in Spectronaut Pulsar by combining the DDA runs of fractionated samples of either proteome or phosphoproteome with the DIA runs of the respective experiments. Human (21,039 entries, additional 74,013 entries, 2015) and mouse uniport FASTA databases (22,220 entries, 39,693 entries, 2015) as forward databases were used. To generate phosphoproteome libraries, serine/threonine/tyrosine phosphorylation was added as variable modification to the default settings, which include cysteine carbamidomethylation as fixed modification and N-terminal acetylation and methionine oxidations as variable modifications. The maximum number of fragment ions per peptide was increased from 3 to 15. The false discovery rate (FDR) was set to less than 1% at the peptide and protein levels and a minimum length of seven amino acids for peptides was specified. Enzyme specificity was set as C-terminal to arginine and lysine as expected using trypsin and LysC as proteases and a maximum of two missed cleavages. To generate proteome libraries, default settings were used. The experimental DIA runs were then analyzed against the hybrid library by using default settings for the analysis of the proteome, and for the analysis of the phosphoproteome samples the localization cutoff was set to 0. BMDM raw files were analyzed via directDIA. For phosphoproteome analysis, serine/threonine/tyrosine phosphorylation was additionally set as variable modification and localization cutoff was set to 0.

All bioinformatics analyses were done with the Perseus software (version 1.6.2.2)⁶⁴. For phosphosite analysis spectronaut normal report output tables were collapsed to phosphosites and the localization cutoff was set to 0.75 using the peptide collapse plug-in tool for Perseus⁶⁵. It collapses phosphoions to phosphosites. Importantly, it does not sum up the intensities of a phosphosite on peptides, if different phosphorylations are also present. For example, the intensity of MAPK14_T180_M1 and MAPK14_T180_M2 (for MAPK14_T180_Y182) differs. For each phosphosite on a multiple phosphorylated peptide, we receive a row with the same intensities as these phosphorylations are localized on the same peptide. While MAPK14_T180_M1 represents the singly phosphorylated peptide, MAPK14_T180_M2 and MAPK14_Y182_M2 share the same intensity as they represent the two phosphosites on the same peptide. Different phosphosites on the same peptide can have slightly different fold changes due to imputation. Also, each collapse key (gene_position_multiplicity) is unique, which means that if a phosphosite is present on two peptides that carry a different phosphosite, just one row will be assigned. Phosphosites located on phosphopeptides with more than two phosphorylations are labeled with a multiplicity of 3 (M3). Summed intensities were log2-transformed. Samples that did not meet the measurement quality of the overall experiment were excluded. For the 8-min time point of the time-course experiment, we started with fewer replicates. Quantified proteins were filtered for at least 75% of valid values among three or four biological replicates in at least one condition. Missing values were imputed and significantly up- or downregulated proteins were determined by multiple-sample test (one-way analysis of variance (ANOVA), FDR = 0.05) and Student’s t-test (two-sided) (FDR = 0.05). For Fig. 2e, we used two imputations to additionally obtain low abundant TNF-induced phosphosites (normal imputation: width 0.3, downshift 1.8, low imputation: width 0.15, downshift 3).

n represents replicates of the same cell line stimulated separately. Further statistical details of experiments can be found in the figure legends.

The 1D annotation-enrichment analysis detects whether expression values of proteins belonging to an enrichment term (here we used keywords GOCC, GOMF, GOBP, and KEGG name) show a systematic enrichment or deenrichment compared with the distribution of all expression values⁶⁶.

Fisher’s exact tests were performed to detect the systematic enrichment or deenrichment of annotations and pathways by analyzing proteins whose levels or phosphorylation levels are significantly regulated upon different conditions (we used keywords GOCC, GOMF, GOBP, and KEGG name). The Benjamini–Hochberg FDR represents the degree of significance and the enrichment factor the level of enrichment compared with the background.

Kinase-motif enrichment analysis (Supplementary Fig. ^1f) was performed by loading significantly (FDR < 0.05) up- and downregulated phosphosites onto the website: http://phosfate.com/profiler.html⁶⁷.

Networks (Supplementary Fig. ^1e, Fig. 2d) were generated using STRING⁶⁸ by uploading significantly changing phosphosites, setting a medium-confidence score (0.4), and including the active-interaction sources: Textmining, Experiments, and Databases.