A MALDI-TOF assay identifies nilotinib as an inhibitor of inflammation in acute myeloid leukaemia

Inflammatory responses are important in cancer, particularly in the context of monocyte-rich aggressive myeloid neoplasm. We developed a label-free cellular phenotypic drug discovery assay to identify anti-inflammatory drugs in human monocytes derived from acute myeloid leukaemia (AML), by tracking several biological features ionizing from only 2,500 cells using matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry. A proof-of-concept screen showed that the BCR-ABL inhibitor nilotinib, but not the structurally similar imatinib, blocks inflammatory responses. In order to identify the cellular (off-)targets of nilotinib, we performed thermal proteome profiling (TPP). Unlike imatinib, nilotinib and other later generation BCR-ABL inhibitors inhibit the p38α-MK2/3 signalling axis which suppressed the expression of inflammatory cytokines, cell adhesion and innate immunity markers in activated human monocytes derived from AML. Thus, our study provides a tool for the discovery of new anti-inflammatory drugs, which could contribute to the treatment of inflammation in myeloid neoplasms and other diseases. Key Points Label-free cell-based assay identifies new anti-inflammatory drugs using MALDI-TOF MS. Nilotinib reduces inflammation by inhibition of MAPK14-MK2/3 signalling axis in AML.


INTRODUCTION
Monocyte-derived macrophages play important roles in both physiological and pathological 40 processes. The activation of these cells during pathogen infection or other insults leads, in part, to the pathophysiology of inflammation (Gordon & Taylor, 2005). Monocytes can recognize various damage-associated molecular (DAMPs) and pathogen-associated molecular patterns (PAMPs) through specific receptors, such as Toll-like receptors (TLRs).
Activation of these receptors leads to the production of cytokines, chemokines, and mediators 45 that are involved in inflammation in myeloid leukemia (Binder et al, 2018;Craver et al, 2018;Stuhlmuller et al, 2000).Several hematopoietic disorders, including lymphoproliferative disorders and myelodysplastic syndromes, which possess a high risk of transformation to leukaemia, have been linked to aberrant TLR signalling (Craver et al., 2018;Hemmati et al, 2017;Monlish et al, 2016). Moreover, TLR2 and TLR4 show 50 significantly higher expression in the bone marrow of patients with myeloid leukemia (Monlish et al., 2016). TLR activation leads to intracellular signalling cascades including the mitogen-activated kinase (MAP) pathways (Kim & Choi, 2010;Suzuki et al, 2018), increasing the expression and secretion of inflammatory cytokines and chemokines including interleukins, interferons, and tumour necrosis factor alpha (TNF-α) (Mogensen, 2009). A 55 shared characteristic of many hematologic malignancies is the overproduction of inflammatory cytokines, particularly TNF-α and IL-6 in myeloid malignancies (Craver et al., 2018).
The discovery of drugs that prevent chronic or acute inflammation is a major goal of the 60 pharmaceutical industry for the treatment of infectious diseases, autoimmune diseases, and cancer (Inglese et al, 2007). The discovery and design of new compounds for inhibiting inflammation are typically achieved through high-throughput screening (HTS) approaches.
As such, there is a growing need to develop physiological phenotypic HTS assays that can be used for monocytic cell lines and primary cells to accelerate the discovery of inhibitors (Dulai 65 & Sandborn, 2016;Li et al, 2017). Mass spectrometry (MS)-based readouts in drug discovery have been largely dominated by instruments comprising of solid-phase extraction (SPE) coupled electrospray ionization (ESI; i.e. RapidFire) or surface-based MS techniques, such as matrix-assisted laser/desorption ionization (MALDI) (McLaren et al, 2021). MALDI-time of flight (MALDI-TOF) MS is a versatile, label-free technique that has the potential to 70 accelerate HTS of promising drug candidates (De Cesare et al, 2018;Guitot et al, 2017;Heap et al, 2017). MALDI-TOF MS is tolerant to a number of standard buffer components and has rapidly become popular in the field of HTS drug discovery due to its versatility, requiring very small sample quantities, and minimal sample clean-up (Chandler et al, 2017;Dreisewerd, 2014;Guitot et al., 2017;Heap et al., 2017;Ritorto et al, 2014;Wang et al, 75 2008). MS-based screening approaches offer the possibility to simultaneously track several molecules in a label-free manner, provide excellent signal-to-noise, reproducibility, assay precision, and a significantly reduced reagent cost when compared to fluorescence-based assays. Furthermore, fluorescence and chemiluminescence methodologies in primary cells and other read-outs, such as antibody-based assays, are very expensive, making full-deck 80 screens of millions of compounds difficult (Adegbola et al, 2018;McLaren et al., 2021;Neefjes & Dantuma, 2004;Wang et al., 2008). MALDI-TOF MS, similar to acoustic mist ionization mass spectrometry (Smith et al, 2021), requires little sample preparation, allows rapid screening and it has high specificity and sensitivity (McLaren et al., 2021). Whole cell analyses or cellular assays for evaluating compound efficacy affecting a cellular phenotype 85 present an interesting challenge for MALDI-TOF MS analysis as the system becomes inherently more complex. One of the attractive qualities of this type of assay for the pharmaceutical industry is that the cellular assays provide biologically relevant information about the physiology and general condition of a cell, such as cell viability and protein activity in a single cellular screen (Butcher, 2005;Michelini et al, 2010). This simultaneous 90 acquisition of information is known as multiplexing and is desirable in cellular assays, as it significantly increases the screening efficiency of the compound library (Gerets et al, 2011).
In this study, we have developed a cellular MALDI-TOF MS assay that is highly reproducible, robust, and sensitive in different cell lines, with which we can identify two 95 main signatures altered under different pro-inflammatory stimuli. Using this MALDI-TOF MS assay, we performed a blind screen of 96 compounds to assess their potential antiinflammatory effects on human monocytes derived from acute myeloid leukaemia (AML).
We discovered that nilotinib, but not the structurally related imatinib, inhibits the proinflammatory phenotype. Using thermal proteome profiling (TPP) (Becher et al, 2016;100 Franken et al, 2015;Jarzab et al, 2020;Mateus et al, 2020;Mateus et al, 2016;Miettinen et al, 2018;Reinhard et al, 2015;Savitski et al, 2014;Zinn et al, 2021), and whole proteome analyses we identified direct targets and downstream regulation of nilotinib during inflammation events. Our data showed that nilotinib and other 2 nd and 3 rd generation BCR-Abl inhibitors, but not imatinib, are capable to block the p38α MAP kinase signalling 105 pathway, which abates monocyte activation and differentiation, reduces cytokine release, and prevents inflammation in AML.

A novel MALDI-TOF MS cellular assay to characterize inflammatory phenotypes 110
Recently, high-throughput MALDI-TOF MS has been used successfully for in vitro assays of specific enzymes (De Cesare et al., 2018;De Cesare et al, 2020;Guitot et al., 2017;Heap et al., 2017). Here, we tested if MALDI-TOF MS can be used for phenotypic assays to monitor inflammatory responses by identifying differences in the "fingerprint" of biomolecules ionized when whole cells are spotted onto the MALDI target ( Figure 1A). As a proof of 115 concept, we tested if we could distinguish phenotypes of THP-1 cells, a model of human monocytes derived from AML (Bosshart & Heinzelmann, 2016), upon stimulation with a number of pro-inflammatory stimuli, namely LPS, Pam 2 CSK 4 , Pam 3 CSK 4 , polyinosinic:polycytidylic acid (poly(I:C)), and polyadenylic-polyuridylic acid (poly(A:U)), which activate TLR4, TLR2/6, TLR2/1, and TLR3, respectively, as well as interferon-γ (IFN-120 We confirmed that other TLR ligands such as Pam 2 CSK 4 and Pam 3 CSK 4 , which activate 130 TLR1/2 and TLR2/6, respectively, did also increase this ratio and in a dose-dependent manner, while interferon-gamma (IFN-γ), poly(I:C) and poly(A:U) did not affect it ( Figure   1F-G and Supplementary Figure 1A). Thus, we used the feature at m/z 4632 as a resting phenotype biomarker and feature m/z 4964 as an LPS-stimulated monocyte phenotype biomarker. To verify that this phenotype was not due to cell death, we treated the cells with 135 0.5 µM staurosporine for 24h before analysis with our MALDI-TOF assay ( Figure 2D) and cell viability assay (Supplementary Figure 1L). Together, these results indicate that biomarkers identified by MALDI-TOF MS can be used to detect changes in the inflammatory phenotype downstream of TLRs in AML cells.

Figure 1. Identification of features associated with the monocyte inflammatory phenotype by MALDI-TOF MS. A)
Workflow of the MALDI-TOF MS assay: (1) AML cells were pre-treated with a drug for 1h before adding a pro-inflammatory stimulus for 24h. (2) Cells were frozen on dry ice, thawed and (3) washed with 100 mM Tris-HCl buffer at 4°C. (4) Cells were mixed with a matrix (10 mg/mL α -cyano-4-cinnamic acid in 50% acetonitrile, 0.1% trifluoroacetic acid). (5) Cells were 145 analyzed in a rapifleX PharmaPulse MALDI TOF mass spectrometer. (6) In the ionization chamber a laser is used to produce ions in the gas phase. These ions are separated according to their time-offlight (TOF) in a field-free region. The smaller ions reach the detector first, followed by the bigger ions, according to the m/z ratio. (7) The detector converts the received ions into electrical current which is amplified and digitized in m/z spectra. for 24h compared to resting cells E) Box plots of significantly changing intensities between resting and LPS-treated monocytes identified at m/z 4632 and 4964. F) Relative quantitation from three biological replicates of THP-1 cells treated with 100 ng/mL LPS, Pam 2 CSK 4 , Pam 3 CSK 4 and 100 U/mL IFN-γ for 24h compared to resting cells. G) Titration of LPS, Pam 2 CSK 4 and Pam 3 CSK 4treated cells from 10 -100 ng/mL of stimulus. Significant differences between two groups were 160 determined by Mann-Whitney U-test. The statistical significance of the comparisons with resting is indicated as follows: ns, not significant; ***, P ≤ 0.001; *, P ≤ 0.05. Error bars represent the standard deviation of three biological replicates.

The MALDI-TOF MS cellular assay can detect inhibitors of inflammation 165
Next, we used NG-25, a TAK1 inhibitor which blocks TNF-α production; BI2536, a PLK1 inhibitor that has been show to inhibit pro-inflammatory gene transcription due to inhibition of BET proteins (Malik et al, 2015); as well as MRT68601, a highly selective and potent TBK1 inhibitor which blocks production of type I interferons (IFNs) (Newman et al, 2012;Xu et al, 2020), to test if pharmacological inhibition of these pathways can be detected in our 170 assay ( Figure 2A). As LPS provided the greatest response of all inflammatory stimuli ( Figure 1F-G), this stimulus was used for further assays. THP-1 cells were pre-treated with the inhibitors for one hour before LPS-stimulation. NG-25 and BI2536 treatment, but not MRT68601, reduced TNF-α secretion ( Figure 2B). It has been published that MRT68601 suppresses the production of type I interferons but increases production of pro-inflammatory 175 cytokines (Xu et al., 2020), which we could not confirm within our experiments. Comparably to the cytokine secretion, our MALDI-TOF MS assay showed that treatment with NG-25 or BI2536 blocked the inflammatory phenotype and was indistinguishable to vehicle control cells not treated with LPS, while MRT68601-treated cells still showed an increased ratio at m/z 4964/4632, similar to LPS-treated cells ( Figure 2C and Supplementary Figure 1B). To 180 verify that this phenotype was not due to cell death, we treated the cells with 0.5 µM staurosporine for 24h before analysis with our MALDI-TOF assay. Our results showed that apoptosis leads to a distinct cluster in the PCA compared to control or LPS-treated cells ( Figure 2D, suggesting that this assay can be multiplexed to identify other phenotypes such as cell toxicity. Moreover, using all biomolecule features, PCA analysis showed that 185 treatment with NG-25 and BI2536 reverted the fingerprint back close to control ( Figure 2D).
This data shows that pharmacological intervention of inflammatory pathways can be detected by our assay and that NG-25 can be used as a positive control compound in a high-throughput screen. 190

Label-free MALDI-TOF MS identifies nilotinib as an anti-inflammatory compound
Once the analytical workflow was established to identify an inflammatory phenotype in THP1 cells, we evaluated the potential of the assay to discriminate positive and negative inhibitors of inflammation in a proof-of-concept blind screen of 96 compounds (Supplementary Table 1). The assay showed very good sensitivity (z'=~0.8) and 195 reproducibility (R 2 > 0.8) (Supplementary Figure 2A-B). Only two positive hits were observed when applying a cut-off of 40% to the ratio m/z 4964/4632, as previously determined for MALDI-TOF assays (De Cesare et al., 2020;Heap et al., 2017). Unblinding revealed that these compounds were NG-25 and nilotinib (Tasigna ® ) ( Figure 2E). Nilotinib is a BCR-ABL tyrosine kinase inhibitor (Pan et al, 2019;Wei et al, 2010) and is a second-200 generation derivative of imatinib (Gleevec ® ) (Pan et al., 2019), which was also included in the panel screened. However, imatinib did not affect the phenotype of inflammatory monocytes, while nilotinib-treated monocytes did not display an LPS-induced inflammatory phenotype and showed no differences compared to non-activated monocytes in the ratio of m/z 4964/4632 ( Figure 2F and Supplementary Figure 1B). Moreover, we tested by 205 MALDI-TOF MS other tyrosine kinase inhibitors (TKIs) used in CML patients (Pophali & Patnaik, 2016): the second-generation TKI, dasatinib and the third-generation TKIs, bosutinib and ponatinib, which all also reduced the m/z 4964/4632 ratio and thus, inflammatory responses ( Figure 2F  IL-6 production. BI2536 inhibits BET proteins (Bromodomain and Extra-Terminal motif proteins) which are required for pro-inflammatory gene transcription. B) TNF-α secretion of THP-1 cells treated in either vehicle control-treated (DMSO), 100 ng/mL LPS-treated, or pre-treated with 5 µM NG-25, BI2536 or 1 µM MRT68601 for one hour before 100 ng/mL LPS-treatment for up to 24h measured by ELISA. C) MALDI-TOF MS relative quantitation of the normalized ratio m/z 220 4964/4632 in THP-1 cells as treated in B). D) PCA of all features identified by the MALDI-TOF MS of experiment in B), and 0.5 µM staurosporine-treated cells. E) Compound hit map of the mean of three biological replicates with 40% effectiveness cut off showing nilotinib and NG-25 as positive hits and imatinib as negative hit (arrows). F) MALDI-TOF MS relative quantitation of the normalized ratio m/z 4964/4632 in THP-1 cells in vehicle control-treated (DMSO), 100 ng/mL LPS-treated, or 225 pre-treated with 5 µM nilotinib, imatinib, 1 µM dasatinib, bosutinib or ponatinib for one hour before 100 ng/mL LPS-treatment for up to 24h. G) TNF-α secretion of cells treated as in F.) measured by ELISA after 24h. Significant differences between two groups were determined by a Mann-Whitney U-test. The statistical significance of the comparisons with LPS is indicated as follows: ns, not significant; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001. Error bars represent the standard deviation of Nilotinib is known to inhibit BCR-ABL and other tyrosine kinases. (Pan et al., 2019) A putative role of these kinases in inflammatory signalling downstream of TLRs is unknown. 245 Moreover, the target involved in this particular function is likely a kinase and must be a specific target for nilotinib, since the treatment with imatinib did not result in the same phenotype. In order to identify the off-targets responsible for the nilotinib-specific anti-inflammatory phenotype, we performed thermal proteome profiling (TPP) by multiplexed quantitative mass spectrometry using tandem mass tags (TMT) ( Figure 3A, Supplementary 250 Table 2, and Supplementary Table 3). This method is based on determining the changes in the thermal stability of proteins, which may be due to direct drug binding, drug-induced conformation changes, binding to other cellular components, or post-translational modifications such as phosphorylation (Becher et al., 2016;Franken et al., 2015;Mateus et al., 2020;Miettinen et al., 2018;Saei et al, 2021;Savitski et al., 2014). We pre-treated THP1 255 cells for 1 h with nilotinib or imatinib before stimulation with LPS for 15 min. We reduced the LPS treatment to 15 min in order to avoid changes in protein abundance due to transcriptional responses to LPS activation. Overall, we identified 5,565 proteins in our TPP Table 4  show any differences. We further performed a concentration compound range experiment at 56ºC to determine the amounts of nilotinib required to alter protein thermal stability of p38. 275

analysis (Supplementary
We found that nilotinib was able to stabilize p38 at an EC50 concentration of 4.6 µM (Supplementary Figure 4K-L). Moreover, we found that nilotinib is also able to affect p38 stability without LPS (Supplementary Figure 4M-N). (1) AML cells were pre-treated with drug for 1h before adding LPS for 15 min. (2) Cells were split in 8 different tubes and heated for 3 min at a range of temperatures (40, 44, 48, 52, 56, 60, 64 and 68°C). Cells were lysed by freeze-thawing and precipitates removed by ultracentrifugation. (3) Proteins in the supernatants were digested with trypsin, labelled with 16-plex tandem mass tags 285 (TMT) and (4) mixed. TMT-labelled peptides were fractionated using high-performance liquid chromatography (HPLC). (5) Peptides were analysed in an Orbitrap Fusion Lumos Tribrid mass spectrometer. (6) TPP data analysis was performed using MaxQuant and the computing environment R. (7) To validate the results, target proteins were analysed by western blot and melting temperatures (Tm) were calculated. B) Determination of the thermostability of p38 alpha (MAPK14), and C) MK3 290 (MAPKAPK3) at the indicated temperatures with 5 µM nilotinib or imatinib for one hour before 100 ng/mL LPS-treatment for up to 15 min by TPP analysis. Table insert shows melting temperature (Tm ºC) and R 2 . D) Determination of the thermostability of p38 alpha at the indicated temperatures in vehicle control (DMSO), 100 ng/mL LPS and pre-treated with 5 µM nilotinib or imatinib for one hour before 100 ng/mL LPS-treatment for up to 15 min. GAPDH served as a loading control. E)

295
Quantification of thermostability of p38 MAPK western blots from four independent experiments.

300
In order to further understand the effect of nilotinib in preventing the inflammatory response by TLR activation in human monocytes, we used E. coli to stimulate and induce an inflammatory response in THP-1 cells. We used live E. coli rather than LPS to activate multiple inflammatory pathways and not only the pathways downstream of TLR4. Similar to LPS treatment, we were able to identify the inflammatory phenotype after exposure to E. coli 305 for 24h using the MALDI-TOF MS assay (Supplementary Figure 1D and Supplementary Figure 5A). Moreover, we tested by MALDI-TOF MS the p38 inhibitor losmapimod (GW856553X), which was tested in two phase III clinical trials (ClinicalTrials.gov Identifier: NCT04511819 and NCT02145468). In these results, losmapimod reduced the proinflammatory phenotype as well as nilotinib, but not imatinib nor MRT68601 ( Figure 4A,  . Nilotinib inhibits p38 MAPK-MK2/MK3 signalling axis. A) MALDI-TOF MS relative quantitation of the ratio m/z 4964 / 4632 in THP-1 cells pre-treated with DMSO, 5 µM nilotinib, imatinib, 1 µM losmapimod or 1 µM MRT68601 for one hour before stimulation with live E. coli for up to 24h. B) Western blot analysis of the p38 MAPK pathway in THP-1 cells pre-treated with 320 DMSO, 5 µM nilotinib, 5 µM imatinib, 1 µM losmapimod or 1 µM MRT68601 for one hour before stimulation with live E. coli for up to 15 min shows loss of p38 phosphorylation in response to nilotinib. Nilotinib and losmapimod, but not imatinib, block phosphorylation of the down-stream MK2. α -tubulin serves as a loading control. A representative image with two biological replicates of four replicates is shown. Relative mobilities of reference proteins (masses in kilo Daltons) are shown 325 on the left of each blot. Relative quantification is shown on the Supplementary Figure 5C. C) TNF-α, D) IL-6, and E) IL-1β secretion was measured by ELISA at 24h. The statistical significance of the comparisons with E. coli is indicated as follows: ns, not significant; **, P ≤ 0.01; ***, P ≤ 0.001. F) TNF-α expression were determined by RT-qPCR in non-transfected or transfected HEK293 cells with MAPK14 Wt or MAPK14 M107T for 48h and pre-treated with DMSO, 5 µM nilotinib or imatinib for 330 one hour before stimulation with LPS for up to 24h. The results were analyzed using the 2 -ΔΔCt method and normalized using GAPDH and TBP as the reference genes; and non-transfected DMSO sample as the reference sample. The statistical significance between MAPK14 Wt and MAPK14 T106M is indicated as follows: ns, not significant; ***, P ≤ 0.001. The statistical significance of the comparisons with LPS non-transfected is indicated as ‡; with LPS transfected with MAPK14 Wt is 335 indicated as † with LPS transfected with MAPK14 T106M is indicated as $. Error bars represent the standard deviation of four biological replicates. Significant differences between two groups were determined by Mann-Whitney U-test.

340
In order to demonstrate that the suppressive effects of nilotinib on the pro-inflammatory phenotype were mediated by inhibition of the p38 MAPK pathway in THP-1 cells, we analysed the cytokine production mediated by p38 MAPK. In contrast to imatinib, nilotinib significantly reduced TNF-α, IL-6 and IL-1β levels and expression in monocytes stimulated with E. coli ( Figure 4C-E and Supplementary Figure 6). We also overexpressed MAPK14 345 and the drug-resistant MAPK14 T106M (Eyers et al, 1999). We found that nilotinib was able to inhibit the effect of MAPK14, but it is not of the drug-resistant MAPK14 T106M ( Figure   4F). Altogether, our results indicate that nilotinib targets p38 alpha, which inhibits its activity and affects the production of pro-inflammatory cytokines, thereby reducing inflammation.
In order to better understand differences between nilotinib and imatinib treatments on monocytes, we performed quantitative proteomics of THP-1 cells pre-treated with DMSO, 5 µM nilotinib or imatinib for one hour before stimulation with live E. coli for up to 24h. We identified and quantified a total of 3570 proteins of which 242 proteins showed significant 355 differences between nilotinib and imatinib treatments (Figure 5A, Supplementary Table 5 and Supplementary Figure 7A Table 6). We found increased levels of proteins related with actin cytoskeleton regulation, such as ROCK1 and Filamin B with nilotinib. Moreover, we found decreased levels of proteins related with cell adhesion, migration and inflammation response with nilotinib compared to imatinib or E. coli-stimulated cells, such as CD14, CD44, ICAM-1, IL-1β, MMP9, and TLR2 (gene ontology and pathway enrichment data in Supplementary Table 6  365 and Figure 5B-C). During the inflammatory response, monocyte activation leads to differentiation into macrophages and also leads to an increased expression of cell adhesion proteins (Chen et al, 2012;Kounalakis & Corbett, 2006;Ma et al, 2001;Park & Kim, 2015). Optical density of dissolved crystal violet was used to evaluate the adhesion rate. Significant differences between two groups were determined by Mann-Whitney U-test. The statistical significance of the comparisons with E. coli is indicated as follows: ns, not significant; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001. Error bars represent the standard deviation of four biological replicates.

380
To further understand how nilotinib affected activation of monocytes and their differentiation into macrophages upon stimulation with E. coli, we studied cell surface differentiation and cell adhesion markers. We validated our proteomics data by showing that MMP9 and CD44 upregulation induced by E. coli was abolished with nilotinib but not with imatinib 385 (Supplementary Figure 7C and Figure 5D). Furthermore, monocyte activation markers CD11c, CD14, CD49a (integrin α 1) and CD54 (ICAM-1) were suppressed with nilotinib treatment but not with imatinib ( Figure 5F-H). Also, we observed less cell adhesion with nilotinib treatment than with imatinib ( Figure 5I). These results indicate that nilotinib inhibits the inflammatory response of monocytes under a pro-inflammatory stimulus, 390 inhibiting their differentiation into macrophages.

Nilotinib reduces the inflammation phenotype in AML.
Nilotinib is a second-line therapy for CML patients who failed or were intolerant to imatinib (Sasaki et al, 2016;Wei et al., 2010), but little is known about its use in other types of cancer. 395 We performed an in-silico analysis of data from the Genomics of Drug Sensitivity in Cancer (GDSC) database to explore the sensitivity of nilotinib and imatinib sensitivity in different cancer types including haematological and solid tumours. We found the sensitivity to nilotinib was significantly higher than that of imatinib in haematological malignancies and gastrointestinal tumours (Supplementary Figure 8A). Interestingly, AML, MM, and diffuse 400 large B cell lymphomas (DLBCL) showed significant differences between the sensitivities to nilotinib and imatinib treatments (Supplementary Figure 8B). These data support the interest of studying the effects of nilotinib and imatinib in myeloid malignancies as AML. In order to test whether our cellular MALDI-TOF assay can be performed in other cell lines than in THP-1 cells, and to test whether nilotinib and other TKIs can be an effective 405 treatment in other monocyte-rich aggressive myeloid neoplasms, we used another monocytic AML cell line, OCI-AML2, and the MM cell line, NCI-H929. We also observed a significant reduction in the ratio m/z 4964/4632 by MALDI-TOF MS in both hematopoietic cell lines treated with nilotinib (Supplementary Figure 8C and Supplementary Figure 1G-H).

415
In addition, we performed our MALDI-TOF assay in peripheral blood primary human monocytes and primary patient AML cells. For this, we isolated primary human CD14+ monocytes from four healthy donors and AML cells from two patients (#274 and #312). We also observed a significant reduction in the ratio m/z 4964/4632 by MALDI-TOF MS in both LPS-stimulated primary monocytes and AML samples treated with nilotinib, as well as with 420 the third generation TKI, ponatinib, and the p38 inhibitor, losmapimod ( Figure 6A-B and Supplementary Figure 1I-J), while ponatinib and bosutinib were more variable. Moreover, TNF-α secretion was reduced with nilotinib treatment in both LPS-stimulated primary monocytes and AML patient cells (Figure 6C-D). Taken together, these results show that our MALDI-TOF assay is suitable tool to detect pro-inflammatory phenotypes via specific 425 biomarkers (m/z 4632 and 4964) in AML and monocytes. Furthermore, our data show that nilotinib inhibits p38 MAPK-MK2/MK3 signalling axis during the inflammatory response, reducing the levels of the pro-inflammatory cytokines in myeloma cells ( Figure 6E). quantitation of the ratio m/z 4964/4632 in A) human primary monocytes and B) primary AML cells pre-treated with DMSO, 5 µM nilotinib, imatinib, 1 µM dasatinib, bosutinib, ponatinib or losmapimod for one hour before stimulation with 100 ng/mL LPS for up to 24h. C-D) TNF-α secretion was measured by ELISA at 24h in C) human primary monocytes and D) primary AML cells. Error bars represent the standard deviation of four or two biological replicates. Significant 435 differences between two groups were determined by Mann-Whitney U-test. The statistical significance of the comparisons with E. coli is indicated as follows: ns, not significant; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001. E) Effect of nilotinib during pro-inflammatory stimulus. Activation of TLR4 by a pro-inflammatory stimulus induces p38 MAPK signalling. Nilotinib inhibits p38 phosphorylation and MK2 phosphorylation reducing the levels of cytokines such as TNF-α, IL-6 or 440 IL-1 β , and cell adhesion molecules (CAMs).

DISCUSSION
Inflammation has an important role in many aspects of acute myeloid leukaemia (AML) such 445 as disease progression, chemoresistance, and myelosuppression (Recher, 2021). For the last five decades, AML treatment has been limited to intensive chemotherapy with cytarabine and anthracycline. Non-standard chemotherapy or immunotherapy is indicated if certain mutations or markers are detected in cancerous cells. Recently, there has been a major effort to understand the disease on the molecular level and to develop novel drugs (Kantarjian et al, 450 2021), including targeted therapies such as tyrosine kinase inhibitors (TKIs) (Fernandez et al, 2019). To date, 89 drugs targeting protein kinases have been clinically approved by the Food and Drug Administration (FDA, April 2021), while at least 150 are being investigated in clinical trials (Bhullar et al, 2018). Development of new high-throughput phenotypic drug screening approaches is important for the identification of novel drugs for the treatment of 455 inflammation in autoimmune diseases and cancer. However, current methods for measuring pro-inflammatory cytokines are expensive as they utilize labelled antibodies, making "full deck" analyses of millions of compounds difficult (Adegbola et al., 2018;Neefjes & Dantuma, 2004). MALDI-TOF MS is a versatile technique with many different applications ranging from protein identification by peptide mass fingerprinting and small molecule 460 analysis to imaging of tissues (Aichler & Walch, 2015;Cohen & Gusev, 2002). As the technology allows very fast screening, it has already found traction in the HTS field with its application to label-free screening of in vitro assays (De Cesare et al., 2018;De Cesare et al., 2020;Haslam et al, 2016;Heap et al., 2017;McLaren et al., 2021;Simon et al, 2021;Winter et al, 2018;Winter et al, 2019). 465 In this study, we present a rapid and label-free cellular MALDI-TOF MS assay able to identify an inflammatory phenotype in monocytic cells, and to screen for anti-inflammatory drugs. We applied a screen to whole cells, allowing the identification of biomarkers that provide a specific fingerprint for the phenotype of the analysed cells (Heap et al, 2019). Two 470 features at m/z 4632 and 4964 were significantly altered upon different inflammatory stimuli, such as, LPS, Pam 2 CSK 4 , Pam 3 CSK 4 and E. coli in the monocytic AML cell line THP-1 (Bosshart & Heinzelmann, 2016). The feature m/z 4964 has been previously described as a biomarker to identify patients with Crohn's Disease, intestinal tuberculosis or rheumatoid arthritis using MALDI MS imaging (Kriegsmann et al, 2012;Zhang et al, 2016). It has been 475 associated with thymosin beta 4 (TB4), a small protein involved in binding and promoting actin polymerization, however it has also been described as having anti-inflammatory properties (Holzlechner et al, 2017). Since the protein is only 44 amino acids, TB4 was not identified in our TPP or proteomics data and an attempt to identify it by MS/MS directly from cells by MALDI TOF MS failed. FT-ICR mass spectrometry may allow the identification of 480 these biomarkers in the future. Moreover, we found that several pro-inflammatory markers (Cunha et al, 2016;Langjahr et al, 2014), such as TLR2, IL1B, CD14, MMP9 or SOD2 were dysregulated. We showed that the cellular MALDI-TOF MS assay can determine the inflammatory phenotype upon activation of the plasma membrane toll-like receptors (TLR), such as TLR1, TRL2, TLR4 and TLR6, but not upon activation of interferon-receptors or the 485 intracellular TLR3 receptor.
While we tested a range of different pro-inflammatory stimuli, we found the greatest response upon LPS-stimulation, which was used in a proof-of-concept blind screen of 96 selected compounds. Within this screen, we identified nilotinib as a drug that prevents the 490 inflammatory phenotype in stimulated AML cells by MALDI-TOF MS. However, the structurally similar drug imatinib did not have this anti-inflammatory effect. Imatinib and nilotinib are BRC-ABL tyrosine kinase inhibitors (TKIs) and the standard first-line therapy for CML (Pan et al., 2019). However, nilotinib possesses higher potency than imatinib against BCR-ABL and is active against most imatinib-resistant BCR-ABL mutations (Wei et 495 al., 2010). The second-generation TKIs, dasatinib and bosutinib as well as the thirdgeneration TKI, ponatinib, are also potent multitargeted tyrosine kinase inhibitors in the treatment of chronic myeloid leukemia (Jabbour et al, 2015). These compounds have been previously suggested to have potential anti-inflammatory effects which we confirmed in our data within a range of cell types (Akashi et al, 2011;Guo et al, 2018;Jabbour et al., 2015;500 Ozanne et al, 2015).
Leukemic blasts often secret inflammatory cytokines such as TNF-a, IL-1β, and IL-6, which induce expression of proteins necessary for their adhesion to vascular endothelium, migration to tissues, proliferation, and chemoresistance (Griffin et al, 1987;Recher, 2021). Stimulation 505 with TLR ligands activates monocytes, which leads to the production of cytokines, chemokines, and mediators that are involved in inflammation (Stuhlmuller et al., 2000;Yang et al, 1998). TLR ligation induces the formation of a signalling complex that includes IL-1Rassociated kinases (IRAKs) and TNFR-associated factor 6 (TRAF6), which mediates the K63-linked polyubiquitylation and activation of TGFβ-activated kinase 1 (TAK1). TAK1 is a 510 MAPK kinase kinase (MAP3K) upstream of p38 MAPK and Jun N-terminal kinases (JNKs).
p38 MAPK directly phosphorylates other protein kinases termed MAPK-activated protein kinases (MKs). The MKs that are phosphorylated by and functionally subordinate to p38 MAPK include MK2 and MK3, which play a versatile role in transcriptional and translational regulation, and affect inflammatory responses, triggering the production of cytokines and 515 chemokines (Arthur & Ley, 2013;Canovas & Nebreda, 2021;Suzuki et al., 2018). Here, we performed a thermal proteome profiling analysis using TMTpro 16-plex (Zinn et al., 2021), to identify the possible different targets between nilotinib and imatinib during the inflammatory response. One of the most significant and highly reproducible targets identified was p38α as a main off-target of nilotinib, when compared to imatinib. While this technology 520 is very powerful, one always needs to keep in mind that membrane proteins are difficult to detect with TPP and some other off-targets for nilotinib may have been missed. Nilotinib has been reported to inhibit p38α activity (Kitagawa et al, 2013;Li et al, 2011) and p38γ (Kitagawa et al., 2013), in an in vitro kinase assay and may stabilize p38α through direct interaction, as it has been observed in vascular endothelial cells and in ischemic reperfusion 525 injury (Hadzijusufovic et al, 2017;Ocuin et al, 2012). Furthermore, we found MK3 as thermally stabilized by nilotinib, when compared to imatinib, although this could be down to changes in the phosphorylation status. However, we did not find any evidence for p38γ inhibition in our experiments, suggesting remarkable specificity for the p38α isoform. While the loss of p38 phosphorylation would suggest an inhibition upstream of p38 MAPK, we 530 found no differences for any MAPKK and MAPKK-independent p38 phosphorylation has been described before (Beenstock et al, 2014;Ge et al, 2002;Ge et al, 2003). Furthermore, our results showed that nilotinib interferes with the downstream MAPK signalling pathway comparable to the specific p38 inhibitor, losmapimod (GW856553X), which has been studied in two phase III clinical trials. The potential of p38 MAPK inhibitors was initially explored 535 within inflammatory conditions such as cancer, acute myocardial infarction, rheumatoid arthritis and Crohn's disease, but the studies demonstrated poor clinical efficacy and unacceptable side effects (Eyers et al, 1998;Fisk et al, 2014;Lee et al, 1994;Lee et al, 2020;Wudexi et al, 2021). This suggests that nilotinib may be considered for additional applications outside of cancer, particularly as it is well tolerated by patients and has few side 540 effects (Sasaki et al., 2016;Wei et al., 2010). Within the cancer field, it appears that the antiinflammatory off-target effects from nilotinib may be beneficial as seen in tumours co-treated with doxorubicin and vincristine (Dent, 2013;Wong et al, 2014). Consistent with these results, nilotinib -but not imatinib -can reduce the inflammation in human monocytesderived from AML and other myeloid cells. 545 Dysregulation of TLR signalling has been linked with myeloproliferative disorders (Hemmati et al., 2017;Monlish et al., 2016). While TLRs are specific for exogenous PAMPs such as lipopolysaccharide (LPS) which triggers TLR4 signalling in monocytes and macrophages, they are also activated by specific endogenous DAMPs such as HSP60 or biglycan (Takeda & Akira, 2005;Takeuchi et al, 1999). The bacterial surface molecule LPS activates TLR4 550 and CD14 in monocytes, inducing pro-inflammatory (IL-1, IL-6, and TNF-α) and then antiinflammatory (IL-10, soluble TNF receptor, and IL-1 receptor antagonist) cytokines (Mogensen, 2009;Takeuchi et al., 1999). Nilotinib treatment was able to decrease the levels of IL-1β, IL-6, and TNF-α cytokines secretion upon stimulation with E. coli in AML cells. In line with our results, pre-incubation with nilotinib in murine macrophages derived from bone 555 marrow led to a decreased LPS-induced IL-6 expression (Wong et al, 2013). Furthermore, monocyte differentiation into macrophages induces an up-regulation of CD11c and CD14 surface markers (Chen et al., 2012;Ma et al., 2001), as well as cell adhesion molecules (Kounalakis & Corbett, 2006;Park & Kim, 2015). Interestingly, nilotinib was able to block the upregulation of CD11c, CD14, CD44, CD49a, and CD54 markers, thus reducing immune 560 response and cell adhesion of monocytes. We further found MMP9, which cleaves CD44 and induces cell migration and cell-cell and cell-matrix adhesion (Chetty et al, 2012), downregulated upon nilotinib treatment. A potential mechanism of action is the Nilotinib dependent increase of ROCK1 levels in E. coli-stimulated cells as deficiency in ROCK1 has been implicated in the migration and recruitment of inflammatory cells such as macrophages 565 and neutrophils during acute inflammation (Vemula et al, 2010). Furthermore, we observed a reduction in cell adhesion in activated monocytes upon nilotinib treatment, compared to imatinib. Our data therefore suggests that nilotinib may not only affect inflammatory signalling and immune cell adhesion, but also differentiation from monocytes into macrophages. 570 Our study provides a valuable tool to discover new anti-inflammatory drugs using a cellular MALDI-TOF MS assay, as we demonstrate here for nilotinib. We show that nilotinib inhibits the p38α MAPK-MK2/3 signalling axis, prevents its phosphorylation and subsequent activation, ultimately preventing the transcription of pro-inflammatory genes, cell adhesion 575 markers and innate immunity markers. In consequence, nilotinib may have therapeutic potential through the inhibition of p38 MAPK-MK2/3 axis in inflammatory diseases as well as in myeloid malignancies such as myeloid leukaemia and multiple myeloma.

Blind drug screening. 620
One million THP-1 cells/mL were incubated with each compound (Supplementary Table 1) at 5 µM for one hour in technical triplicates before stimulation with 100 ng/mL LPS for 24h.
Compounds were staggered into 8 sets each with individual positive (NG-25) and negative controls (MRT68601) to enable maximum relative peak area calculations. This screening process was repeated in three biological replicates. 625

Cell infection with E. coli.
The DH5α strain of E. coli bacteria (Invitrogen) was harvested during mid-log phase.
Mammalian cells were infected with a multiplicity of infection of 2. The cells with bacteria were centrifuged at 500 xg for 5 min at 37°C and incubated at 37°C in 5% CO 2 for bacterial 630 uptake for 15 min or 30 min. Thirty min post-infection, cells were washed once with PBS and incubated for 1h with 100 µg/mL gentamicin to kill extracellular bacteria. Then, the cells were washed with PBS twice and the media replaced with 20 µg/mL gentamicin for the remainder of the experiment.
The viability of THP-1 cells was assessed for 24h using the Cell Proliferation Kit II (XTT) (Sigma-Aldrich) as per manufacturer's instructions. Following incubation, the plate was read with SpectraMax iD5 microplate reader (Molecular Devices) at a wavelength of 490 nm. The reference wavelength 655 nm was also read to control for nonspecific absorption. 640

Sample preparation for MALDI-TOF MS and data analysis.
Cell pellets were frozen on dry ice, then thawed and washed with 100 mM Tris-HCl, pH 7.5 and centrifuged at 1,000 xg for 10 min at 4°C. 2,500 cells were spotted on target with 10 mg/mL α -cyano-4-cinnamic acid in 50% acetonitrile, 0.1% trifluoroacetic acid. Automated 645 target spotting was performed using a Mosquito liquid handling robot (TTP Labtech). A RapifleX PharmaPulse MALDI TOF/TOF mass spectrometer (Bruker Daltonics) equipped with a Smartbeam 3D laser was used in positive ion mode. Samples were acquired in automatic mode (AutoXecute; Bruker Daltonics), totaling 10,000 shots at a 10 kHz frequency per spot. A random walk pattern (complete sample) on spot laser ablation pattern was used 650 with an M5 Smart beam Parameter at a 45 µm × 45 µm scan range. Spot diameter was limited to 2,000 µm and a random walk pattern movement enabled at 1000 shots per raster position. Ionization was achieved using a laser power of 75% (laser attenuator offset 14%, range 30%) with a detector gain of ×6.8 in the mass range of m/z 2,000-20,000 with a mass and, reverse: GGGTCAGTCCAGTGCCATA; TNF-α, forward: AAACTCATGAGCAGTCTGCA, and reverse: AGGAGATCTTCAGTTTCGGAGG.

ELISA. 680
Cell culture supernatant was collected at 24h post-treatment. TNF-a, IL-6, and IL-1β were measured by DuoSet ELISA kits (R&D Systems). Absorbance from four biological replicates at 450 nm was measured with the correction wavelength set at 540 nm using SpectraMax ® M3 microplate reader (Molecular Devices).

Western blot.
Cells were lysed using 5% SDS supplemented with Protease Inhibitor Cocktail and Phosphatase Inhibitor Cocktail 2 (Sigma-Aldrich). The following antibodies were purchased  experiments, cells were treated with 0.1, 0.5, 1, 5 or 10 µM of nilotinib or imatinib for 1h before treatment with 100 ng/mL LPS for 15 min. Cells were heated at 56°C for 3 min. 720 Samples were lysed with four freeze-thaw cycles using dry ice and a thermo block at 35°C.
Cell lysates were centrifuged at 20,000 xg for 2h at 4°C to separate protein aggregates from soluble proteins. Supernatants were collected and used for western blot and mass spectrometry.

Sample preparation for LC-MS/MS and label-free quantification (LFQ) data analysis.
Samples were prepared for mass spectrometry using S-Trap micro columns (Protifi) according to the manufacturer's recommended protocol. Proteins were reduced with 20 mM TCEP at 37°C for 30 min and alkylated with 20 mM N-ethylmaleimide in the dark for 30 min. A ratio of 1:10 wt:wt Trypsin TPCK Treated (Worthington-Biochem) was used to digest 730 the samples for 2h at 47°C. Eluted peptides were dried down and resuspended in loading buffer (2% acetonitrile, 0.1% trifluoroacetic acid). Peptide samples were injected on a Dionex Ultimate 3000 RSLC (Thermo Fisher Scientific) connected to an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scientific). Samples were injected on a PepMap 100 C18 LC trap column (300 µm ID x 5 mm, 5 µm, 100 Å) followed by separation on an 735 EASY-Spray column (50 cm x 75 µm ID, PepMap C18, 2 µm, 100 Å) (Thermo Fisher Scientific). Buffer A consisted of water containing 0.1% FA and Buffer B of 80% acetonitrile containing 0.1% FA. Peptides were separated with a linear gradient of 3-35% Buffer B over 180 min followed by a step from 35-90% Buffer B in 0.5 min at 250 nL/min and held at 90% for 4 min. The gradient was then decreased to 3% Buffer B in 0.5 min at 250 nL/min, and the 740 column equilibrated for 10 min before the next injection. Column temperature was controlled at 45°C. The Orbitrap Fusion Lumos Tribrid mass spectrometer was operated in datadependent, positive ion mode. Full scan spectra were acquired in the range of m/z 400 to 1,600, at a resolution of 120,000, with an automatic gain control (AGC) target of 4x10 5 and a maximum injection time of 50 ms. The Orbitrap was operated in 'stop speed' to maintain a 3 745 sec fixed duty cycle. The most intense precursor ions were isolated with a quadrupole mass filter width of 1.6 m/z and higher-energy collision-induced dissociation (HCD) fragmentation was performed in one-step collision energy of 30%. Detection of MS/MS fragments was acquired in the linear ion trap in rapid scan mode with an AGC target of 1x10 4 ions and a maximum injection time of 45 ms. An electrospray voltage of 2.0 kV and capillary 750 temperature of 275°C, with no sheath and auxiliary gas flow, was used.
All discovery proteomics RAW mass spectra were analysed using MaxQuant (version 1.6.10.43) (Cox & Mann, 2008) and searched against a SwissProt Homo sapiens database (containing 42,347 database entries with isoforms; downloaded on 21 April 2020). Trypsin/P 755 was set as proteolytic enzyme, N-ethylmaleimide on cysteine was set as fixed modification, and methionine oxidation and acetylation of protein N-termini was set as variable modifications. Two missed cleavages were allowed. A protein and peptide false discovery rate (FDR) of less than 1% was employed in MaxQuant. LFQ data analysis was performed using Perseus (version 1.6.2.3) (Tyanova et al, 2016). Reverse hits, contaminants and 760 proteins only identified by site were removed before downstream statistical and bioinformatics analysis. LFQ intensity data were transformed (log 2) and filtered to contain at least two unique peptides and at least three valid values in one group for comparisons.
Significance testing was carried out using a two-tailed unpaired Student's t-test and multiple hypothesis testing was controlled using the Benjamini-Hochberg FDR threshold of p-value 765 <0.05. Gene ontology term enrichments were performed within DAVID (version 6.8) (Huang da et al, 2009).

Thermal proteome profiling (TPP) and data analysis.
Isobaric labelling of peptides was performed using the 16-plex tandem mass tag (TMT) 770 reagents (Thermo Fisher Scientific) (Supplementary connected to an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scientific). Buffer A consisted of water containing 0.1% FA and Buffer B of 80% acetonitrile containing 0.1% FA. Peptides were separated with a linear gradient of 3-35% Buffer B over 180 min followed by a step from 35-90% Buffer B in 0.5 min at 250 nL/min and held at 90% 785 for 4 min. The gradient was then decreased to 3% Buffer B in 0.5 min at 250 nL/min, and the column equilibrated for 10 min before the next injection. The scan sequence began with an MS1 spectrum (Orbitrap analysis; resolution 120,000; mass range m/z 375-1500; automatic gain control (AGC) target 4x10E5; maximum injection time 50 ms). MS2 analysis consisted of collision-induced dissociation (CID); AGC 1x10E4; normalized collision energy (NCE) of 790 30%; maximum injection time 50 ms; and isolation window of 0.7. Following acquisition of each MS2 spectrum, MS3 precursors were fragmented by high energy collision-induced dissociation (HCD) and analysed using the Orbitrap (NCE 55%; AGC 5x10E4; maximum injection time 86 ms, resolution was 50,000) and synchronous precursor selection (SPS) was enabled to include 10 MS/MS fragment ions in the FTMS3 scan. 795 Mass spectrometry data analysis was performed similarly as described above, but TMT modification on the peptide N-termini or lysine residues were enabled for the 16-plex TMT reagents, and deamidation of asparagine and glutamine was included as variable modification in MaxQuant. TPP data analysis was performed using the TPP packages 800 (https://www.bioconductor.org/packages/release/bioc/html/TPP.html) in the R statistical programming language. The data were normalized as detailed in Miettinen et al. (Miettinen et al., 2018) (Supplementary Table 3). Proteins that changed significantly in the four replicates between nilotinib and imatinib with a standard deviation below two degrees in the melting temperature were selected as positive hits. 805

Data availability
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (Deutsch et al, 2017) via the PRIDE partner repository (Perez-Riverol et al, 2019) with the data set identifier: PXD025020. 810 Reviewer account details: Username: reviewer_pxd025020@ebi.ac.uk Password: H0T2rCOL

Statistical analysis
The Shapiro-Wilk test was used to check expression data sets for normality, and the Levene 815 test was used for homogeneity of variances. Mann-Whitney U-test was performed in all the analysis, excluding proteomics and thermal proteome profiling.