Human peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats from normal donors over a Lymphoprep (Nycomed Pharma) gradient. Monocytes were purified from PBMCs by magnetic-activated cell sorting using CD14 microbeads (Miltenyi Biotech). Monocytes were cultured at 0.5 × 10⁶ cells/ml for 7 days in Roswell Park Memorial Institute (RPMI) 1640 (standard RPMI, which contains 1 mg/L folic acid) supplemented with 10% fetal calf serum, at 37°C in a humidified atmosphere with 5% CO₂, and containing GM-CSF (1,000 U/ml) to generate GM-CSF-polarized macrophages (GM-MØ) or M-CSF (10 ng/ml) to generate M-CSF-polarized macrophages (M-MØ). GM-CSF or M-CSF was added every 2 days. Low-dose PMX (50 nM), MTX (50 nM), and/or folinic acid (FA; 5-formyl tetrahydrofolate, 5 µM), pifithrin-α cyclic (25–50 μM) was added once on monocytes together with GM-CSF or M-CSF. When indicated, lipopolysaccharide (LPS) (10 ng/ml, 0111:B4 strain, smooth LPS (sLPS) that exclusively binds TLR4, InvivoGen) or LTA (5 µg/ml, from Staphylococcus aureus, InvivoGen) was added at the indicated time points onto 7-day fully differentiated macrophages without replacing the culture medium. sCD14 (20–200 ng/ml, BioLegend) was added on macrophages for intracellular signaling measurements (15 min) and IL-6 detection (3 h).

Total RNA was isolated from three independent preparations and processed at BGI (https://www.bgitechsolutions.com), where library preparation, fragmentation, and sequencing were performed using the BGISEQ-500 platform. An average of 5.41-Gb bases were generated per sample; and after filtering, clean reads were mapped to the reference (UCSC Genome assembly hg38) using Bowtie2 (average mapping ratio 93.41%) (29). Gene expression levels were calculated by using the RSEM software package (30). Differential gene expression was assessed by using DEseq2 algorithms using the parameters Fold change >2 and adjusted p-value <0.05. For gene set enrichment analysis (GSEA) (http://software.broadinstitute.org.gsea/index.jsp), the gene sets available at the website (31), as well as the “p53_Target_Gene” gene set, that contains the top 98 genes identified as p53 targets, were used (32). The data discussed in this publication have been deposited in Gene Expression Omnibus (GEO) of National Center for Biotechnology Information (NCBI) (33) and are accessible through GEO series accession numbers GSE159349 and GSE159380. For microarray data, statistical analysis for differential gene expression was carried out using empirical Bayes moderated t-test implemented in the limma package (http://www.bioconductor.org). The transcriptome of MTX-GM-MØ has been previously described (21) and is available at GEO (GSE71253).

Total RNA was retrotranscribed, and cDNA was quantified using the Universal Human Probe Roche library (Roche Diagnostics). Quantitative real-time PCR (qRT-PCR) was performed on a LightCycler^® 480 (Roche Diagnostics). Assays were made in triplicates, and results normalized according to the expression levels of TBP. Results were obtained using the ΔΔCT method for quantitation. The oligonucleotides used to quantify mRNA transcripts were (5′–3′) as follows: CD14 forward gttcggaagacttatcgaccat; CD14 reverse acaaggttctggcgtggt; IL1β forward ctgtcctgcgtgttgaaaga; IL1β reverse ttgggtaatttttgggatctaca; IL6 forward gatgagtacaaaagtcctgatcca; IL6 reverse ctgcagccactggttctgt; TBP forward cggctgtttaacttcgcttc; and TBP reverse cacacgccaagaaacagtga.

Two different siRNAs for TS (TYMS silencer select s14538 and s14539) and one control siRNA (silencer select Negative Control #1) (Ambion, Life Technologies) were used at 100 nM. GM-MØ cells were transfected with HiPerFect transfection reagent (QIAGEN), treated with PMX, and, 48 h later, tested for CD14 mRNA levels by qRT-PCR.

Supernatants from GM-MØ were tested for the presence of IL-6, IL-10, TNFα, and CXCL10 (BioLegend) and IFNβ (R&D Systems). For the detection of sCD14 in supernatants, serum, and plasma of RA patients and healthy controls, the Human CD14 Quantikine ELISA Kit (R&D Systems) was used.

Cell lysates were obtained in radioimmunoprecipitation assay (RIPA) buffer containing 1 mM of PIC (Protease Inhibitor Cocktail, SIGMA), 10 mM of NaF, 1 mM of Na₃VO₄, and 0.5 mM of DTT. Cell lysate measuring 10 μg was subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto an Immobilon polyvinylidene difluoride membrane (Millipore). Protein detection was carried out using mouse monoclonal Ab against IκBα (clone L35A5, #4814; Cell Signaling), rabbit polyclonal against p-p38 (clone D3F9), pJNK (clone 81E11) and pERK (clone D13.14.4E, #9910; Cell Signaling), p-IRF3 (clone 4D4G, #4947; Cell Signaling), and p-STAT1 (clone D4A7, #7649; Cell Signaling). Protein loading was normalized using an antibody against GAPDH (clone 6C5, sc-32233, Santa Cruz Biotechnology) or against human Vinculin (clone VIN-11-5, #V4505; Sigma-Aldrich).

Phenotypic analysis was carried out by immunofluorescence using standard procedures. Mouse monoclonal antibodies used for cell-surface staining included fluorescein isothiocyanate (FITC)-labeled anti-CD14 (clone MφP9; BD) and PE-labeled anti-TLR4 (clone HTA125; BioLegend). Isotype-matched labeled antibodies were included as negative controls.

For the analysis of clinical data, Stata 14.0 for Windows (Stata Corp LP, College Station, TX, USA) was used. For the discovery cohort, a non-parametric Wilcoxon test was used to assess the statistical significance of differences between pre- and posttreatment RA plasma sCD14. For the validation cohort, most quantitative variables followed a non-normal distribution, so they were represented as median and interquartile range (IQR), and the Mann–Whitney or Kruskal–Wallis tests were used to analyze significant differences. Qualitative variables were described using a calculation of the proportions, and Fisher’s exact test was used to compare categorical variables. To assess the ability of either baseline CD14 (sCD14) or the variation in sCD14 between baseline and 6 months’ follow-up visits (ΔsCD14) to discriminate between MTX-responder and MTX-non-responder patients, we generated receiver operating characteristic (ROC) curves through the command roctab and the option graph. Each cutoff point was selected on the basis of the best trade-off values between sensitivity, specificity, cases correctly classified, and positive (LR+) and negative (LR−) likelihood ratios reported using the command roctab with the option detail. With the use of these cutoff values, new variables were generated to define patients with high or low baseline sCD14 and those with relevant sCD14 decrease. Then, we estimated the ORs and their 95% CI with the command cs of Stata, with the option or. This command also provides the statistical significance with Fisher’s exact test.

To estimate with a better precision the capability of sCD14 to discriminate between MTX-responder and non-responder patients, we performed a multivariable logistic regression analysis in which the dependent variable was response to MTX, and as independent variables, gender, age, and baseline DAS28 were included, which are well-known predictors of treatment response in RA. Then, we consecutively included sCD14 or ΔsCD14 in the model to determine their respective ORs adjusted by age, gender, and baseline DAS28.

Statistical analysis for the rest of the figures was done using GraphPad Prism, using parametric Student’s t-test, as appropriate, and one-way ANOVA test coupled with Tukey’s post-hoc test, where indicated. p-Value <0.05 was considered significant (^*p < 0.05; ^**p < 0.01; ^***p < 0.001).

We have previously described that low-dose MTX modifies the transcriptome of proinflammatory TS⁺ GM-CSF-primed macrophages (GM-MØ) (21) (GSE71253). Since PMX also influences the expression of MTX-regulated genes (CCL20, LIF, MET, and GDF15) (21), we sought to determine the whole range of transcriptional changes induced by PMX in human macrophages. Comparison of the gene signatures of GM-CSF-primed monocyte-derived macrophages in the absence (GM-MØ) or presence (PMX-GM-MØ) of PMX (50 nM) (Figure 1A) revealed a large number of significant transcriptional differences (Figure 1B) and that the gene profile of PMX-GM-MØ was enriched in terms like “HALLMARK_INFLAMMATORY_RESPONSE” and “TNFA_VIA_NFκB_SIGNALING” GSEA gene sets (31) (Figures 1C, I), supporting the notion that the continuous presence of PMX determines the acquisition of a more proinflammatory profile. Actually, PMX enhanced expression of NFκB-target genes and MTX-regulated genes (Figure 1E) and, like MTX (21), potentiated the expression of p53 target genes (Figures 1D, F). Moreover, comparison of the gene signature of MTX-GM-MØ (GSE71253) and PMX-GM-MØ revealed a large overlap of the respective transcriptional effects of both antifolates (Figures 1G–I and Supplementary Figure 1), thus emphasizing that, like MTX, PMX promotes the acquisition of a more proinflammatory and p53-dependent gene signature in GM-MØ. Regarding differences between antifolates, MTX-treated macrophages exhibit a significant over-representation of SREBP-dependent genes, while this over-representation is not seen in PMX-treated macrophages (Figure 1I). This difference is relevant because SREBP activity has been recently connected to macrophage polarization and to the capacity of macrophages to acquire a reparative profile (35).

Macrophage response to a given stimulus is dependent on extracellular cues that they have been previously exposed to (training/tolerance) (13–15). Thus, we next asked whether PMX promoted macrophage tolerance state at the transcriptional level. To that end, we determined the gene signature of LPS-stimulated (3 h) GM-MØ differentiated in the presence of PMX, FA (a reduced folate with high-affinity for RFC (36)], or PMX+FA (Figure 2A). Compared with LPS-treated GM-MØ, LPS-treated PMX-GM-MØ exhibited a significantly (|log₂FC| >1; adjp < 0.05) elevated expression of 355 genes and diminished mRNA levels of 273 genes (Figure 2B). Supporting the specificity of PMX, none of these changes were observed in LPS-treated PMX+FA-GM-MØ, whose transcriptome was indistinguishable from that of control FA-GM-MØ (Figure 2B). Interestingly, GSEA revealed a very significant reduction of the genes within the GSEA “HALLMARK_INFLAMMATORY_RESPONSE” gene set (31) (Figure 2C) in LPS-treated PMX-GM-MØ, a negative enrichment that was not seen in PMX+FA-GM-MØ (Figure 2C). Therefore, the presence of PMX during macrophage differentiation impairs the LPS-induced expression of inflammation-related genes, thus indicating that PMX promotes the acquisition of a tolerance state in macrophages. Conversely, expression of the “p53_Target_Genes” gene set was significantly and specifically enriched in LPS-treated PMX-GM-MØ (Figure 2C). Moreover, the role of p53 in the PMX-induced tolerance was suggested by using the p53 inhibitor pifithrin-α, which dose dependently diminished the expression of three previously reported PMX-responsiveness genes (GDF15, LIF, and CDKN1A mRNA) in LPS-treated PMX-GM-MØ (Figure 2D). Since the CDKN1A-encoded p53 target gene p21 determines the reparative and anti-inflammatory profile of macrophages (37), this result supports the notion that PMX conditions macrophages towards a tolerance transcriptional state.

To evaluate the functional correlate of the PMX-induced transcriptional tolerance, we analyzed the ability of PMX to modify human macrophage responses upon TLR4 engagement. In agreement with the RNAseq data, LPS-treated (3 h) PMX-GM-MØ displayed a considerably diminished IL6 and IL1B mRNA expression when compared with LPS-treated GM-MØ (Figure 3A). Moreover, PMX-GM-MØ produced significantly lower levels of LPS-induced IL-6, TNFα, and IL-10 after both short-term (3 h) and long-term (24 h, for IL-6 and IL-10) LPS exposure (Figures 3B–D). The specificity of the PMX effect was demonstrated by the absence of the inhibitory effect in PMX+FA-GM-MØ (shown in Figures 3A, E, F), thus indicating that the effects of PMX are mediated through OCM blockade. Of note, PMX-GM-MØ also produced lower level of IL-6 and IL-10 after exposure to the TLR2 ligand lipoteichoic acid (LTA), thus indicating that PMX attenuates inflammatory cytokine expression in response to either TLR4 or TLR2 ligands (Supplementary Figure 2). Altogether, these results indicate that PMX conditions macrophages for the acquisition of a tolerance state at the transcriptional and functional levels.

Previous studies have demonstrated that tolerant-macrophages exhibit a lower level of MAPK and NFκB activation upon TLR stimulation (22, 38, 39). To determine whether PMX promotes “bona fide” innate tolerance in macrophages, we first determined the levels of MAPK and IκBα activation in LPS-treated PMX-GM-MØ. Upon LPS stimulation, PMX-treated GM-MØ exhibited a reduced activation of p38 and JNK compared with untreated GM-MØ (Figure 4A). By contrast, and as described in other cellular systems (40), PMX-GM-MØ showed an increased and more sustained phosphorylation of ERK1/2 in response to LPS (Figure 4A). Besides, a slower kinetics of IκBα degradation was detected in LPS-exposed PMX-GM-MØ (Figure 4B), an effect abolished in the presence of FA (PMX-FA-GM-MØ, Figure 4C). Further and compared with GM-MØ, LPS induced lower levels of p-IRF3, p-STAT1, and IFNβ1 and CXCL10 secretion in PMX-GM-MØ (Figures 4D–G). Taken together, these results demonstrate that OCM blockade impairs TLR4-initiated intracellular signaling and the establishment of a state of LPS tolerance.

Next, we determined whether PMX affected the expression of negative regulators of TLR-induced cytokine production that contribute to LPS tolerance (38). Unlike MTX-induced tolerance, mostly mediated by increased TNFAIP3 (A20) expression (38, 39, 41, 42), no difference in TNFAIP3 expression was found between LPS-treated GM-MØ and LPS-treated PMX-GM-MØ (not shown). Thus, we next screened for genes coding for regulators of TLR-induced activation and whose expression significantly differs between GM-MØ and PMX-GM-MØ, and we found that PMX-GM-MØ cells express significantly lower levels of CD14 mRNA than GM-MØ (Figure 5A). After validation of this result in additional samples (shown in Figure 5A), we confirmed that PMX-GM-MØ cells express a significantly lower level of cell surface CD14 [membrane-bound CD14 (mCD14)] than GM-MØ (Figures 5B, C). Specifically, the percentage of CD14-positive cells was 75% in GM-MØ and 45% in PMX-GM-MØ, whereas the mean fluorescence significantly varied at 46.5 and 36, respectively (Figure 5B). Further, sCD14 was also significantly lower in the supernatant of PMX-GM-MØ than in control GM-MØ cells (Figure 5D), and the PMX-induced reduction of mCD14 and sCD14 was completely reversed in the presence of FA (Figures 5C, E). Of note, TS knockdown diminished CD14 expression without further effect of PMX (Figure 5F), thus indicating that the PMX-triggered CD14 downregulation is dependent on TS expression. Of note, the antifolate MTX exhibited the same effect, as MTX-GM-MØ showed significantly lower expression of CD14 mRNA, mCD14, and sCD14 than did GM-MØ, an effect also prevented by FA (Figures 5G–K). Therefore, OCM blockade mediates the loss of CD14 gene and protein expression in macrophages differentiated in the presence of antifolates (PMX-GM-MØ or MTX-GM-MØ). The physiological significance of these findings was further stressed by the fact that both antifolates only impaired CD14 expression in proinflammatory GM-MØ, whereas they did not modify CD14 expression in M-CSF-conditioned macrophages (M-MØ) (Figures 5G, H), which resemble anti-inflammatory tissue-resident and pro-tumoral TAM (20, 43, 44).

Since the reduced expression of sCD14 and mCD14 correlated with diminished intracellular signaling and cytokine responses to LPS in PMX-GM-MØ, we next determined whether CD14 had a role in the PMX-induced macrophage tolerance state by assaying LPS responsiveness of PMX-GM-MØ in the presence of increasing concentrations of exogenous sCD14. Addition of sCD14 increased the LPS-induced activation of p38 and JNK, as well as the LPS-induced degradation of IκBα in PMX-GM-MØ, but not in GM-MØ (Figure 6A). Noteworthy, the same results were found in MTX-GM-MØ (Figure 6B). Moreover, sCD14 dose dependently increased the LPS-induced IL-6 production in PMX-GM-MØ, but not in LPS-treated GM-MØ (Figure 6C). Therefore, exogenous sCD14 partly restores the inhibitory effect of PMX on TLR4-initiated intracellular signaling and IL-6 production and demonstrates that CD14 mediates the OCM-dependent pro-tolerant effect of PMX.

Finally, we sought to determine the therapeutic relevance of the CD14 downregulation in GM-MØ generated in the presence of antifolates. Although PMX and MTX similarly diminished mCD14 and sCD14 in GM-MØ (shown in Figure 5), only low-dose MTX is an anchor drug for RA treatment (23, 45). Therefore, we turned to RA patients treated with MTX to evaluate the clinical significance of the association between antifolate exposure and CD14 downregulation. Specifically, we determined sCD14 level in plasma from MTX-responder RA patients both at baseline and after 6 months of treatment with MTX. We found that sCD14 levels significantly decreased in patients who respond to MTX (Figure 7A). Moreover, sCD14 levels positively correlated with disease activity score DAS28 and CRP (Figures 7B–E). These results indicate that sCD14 expression is lower in MTX responder RA patients, suggesting that sCD14 might be a biomarker for MTX response. To confirm these findings, we determined sCD14 concentration in a validation cohort of early arthritis patients at baseline and during 6 months of MTX monotherapy, including both MTX-responder and non-responder patients (see Materials and Methods for definition). Interestingly, baseline sCD14 levels in MTX-non-responder patients were similar to those in control healthy donors, whereas MTX-responder patients exhibited significantly higher sCD14 serum levels (Figure 8A). Moreover, sCD14 levels diminished in MTX-responder patients but not in MTX non-responders (Figure 8B). In order to determine whether baseline sCD14 and ΔsCD14 could be MTX response biomarkers, we performed ROC analysis. Although the area under the curve (AUC) of ΔsCD14 was slightly higher than that of baseline sCD14 (Figure 8C), it did not reach statistical significance. The best cutoff to discriminate between MTX responder and non-responder patients was 2,460 ng/ml for baseline sCD14 (78% sensitivity and 70% specificity) and 188 ng/ml for ΔsCD14 (83% sensitivity and 70% specificity). The OR for having high baseline sCD14 and being MTX responder was 8.4 (p = 0.0126), whereas the OR for a decreased sCD14 was 11.1 (p = 0.0059) (Figure 8D). Furthermore, after adjustment by variables related to MTX response such as age, gender, and baseline DAS28, the OR increased to 25.5 (for sCD14) and 40.35 (for ΔsCD14) (Supplementary Table 1). Altogether, these results indicate that determination of sCD14 levels could be a valuable tool to predict or evaluate MTX response in RA patients.