Identification of potent inhibitors of JUN N-terminal kinases for treatment of endometriosis and associated pain

To discover new JNK inhibitors with greater potency than currently available inhibitors, we conducted a selection for JNK-Is using the Baylor College of Medicine (BCM) Center for Drug Discovery (CDD) DNA-encoded chemical libraries (DECLs). Specifically, 45 unique DECLs cumulatively containing 3.94 billion compounds were screened in our DECL platform with a recombinant His-tagged JNK protein ( Fig. 1 ). During affinity selections, the library pools were incubated with the His-tagged JNK (Target) and with a no-target control (NTC) to rule out bead binders. After two to three rounds of selection, the Target and NTC samples were PCR amplified and subjected to Illumina next generation sequencing and our informatics pipeline (Supplementary Methods) to decode the chemical structures specifically bound to JNK ( 12 ). This pipeline uses an in-house normalized z-score to quantify the enrichment of n-synthons, where a higher z-score indicates a more significant enrichment in the selection. The normalized z-score is calculated on the basis of modeling the selection data as binomial distribution ( 12 ). From 4 billion compounds, 3,034 positives sharing 7 dominant chemical scaffolds were identified (hit rate = 8.7 × 10 −7 ; over 70-fold enrichment), with high affinity binding to JNK ( Fig. 1 A ). Enriched selection hits containing identical building block 3 (BB3) but different building blocks 2 (BB2) and building blocks 1 (BB1) were identified as JNK binders. Enrichment profile of BCM DNA-encoded chemical library qDOS28_1 against JNK3 at 0.05 μM. ( A ) JNK selection with DNA-encoded library containing 3.8 billion compounds. The x-axis shows the enrichment of hit molecules relative to a no-target control. Structures of hit molecules show that the same building block 3 (BB3, in black) was detected, with various building block 1 (BB1, in blue) and building block 2 (BB2, in red). The enrichment of each hit series was shown as count/z-score at 0.05 μM. ( B ) Hit compound CDD-2428 was directly obtained from the DEL library, whereas CDD-2728 and CDD-3013 are derivatives of CDD-2428. In vitro (LanthaScreen) and cell-based assays (NanoBRET) determine the binding affinities of JNK inhibitors to JNK1, JNK2, and JNK3 with hit and lead compounds (CDD-2428, CDD-2728, and CDD-3013) and discussed investigational drugs (BEND and TANZ). Kd values were obtained as described in the methods. IC50 values for the NanoBRET assays were calculated as described in the supplementary method section. ( C and D ) Promega NanoBRET assay with CDD-3013 performed at 1 µM. Kinase dendrogram was generated by online platform from KinHub, Cell Signaling technology, http://www.kinhub.org/kinmap/index.html to explore sequence similarity and evolutionary relationships of the kinase domain and compound selectivity ( C ). The percent (%) occupancy rates for the top kinases are displayed ( D ). BEND, bentamapimod; TANZ, tanzisertib. NT, not tested; ND, not detected. The DECL-hit CDD-2428 was synthesized off-DNA and confirmed as a potent binder to JNK1, JNK2, and JNK3 in a LanthaScreen binding assay performed at ThermoFisher [dissociation constant (Kd) for JNK1 = 0.5 ± 0.1 nM, JNK2 = 6.2 ± 0.9 nM, and JNK3 = 1.1 ± 0.1 nM] ( Fig. 1 B ). Despite excellent inhibition of kinase activity in vitro, CDD-2428 displayed a low kinetic solubility of 3.1 µM, requiring further optimization ( Fig. 1 B ). Substitution of the bulky N -methyl-3-(piperidin-4-yl) propanamide moiety in CDD-2428’s BB1 with a dimethylamine group, along with replacement of the methoxy (–OCH 3 ) substituent by a hydroxyl (–OH), yielded the simplified analogue CDD-2728, which maintained binding to JNK with low nM affinity in a LanthaScreen binding assay (Kd for JNK1 = 0.31 ± 0.05 nM; JNK2 = 3.79 ± 0.29 nM; JNK3 = 0.97 ± 0.11 nM) and improved its kinetic solubility (19.4 µM) ( Fig. 1 B ). Modification of CDD-2728 BB1 by replacing the dimethylamine with a pyrrolidine ring generated CDD-3013 which showed strong binding to JNK in a LanthaScreen (Kd for JNK1 = 0.12 ± 0.03 nM; JNK2 = 0.72 ± 0.1 nM; JNK3 = 0.75 ± 0.07 nM) but reduced its solubility (0.4 µM) ( SI Appendix , Fig. S1 ). Compared to BEND (Kd for JNK1 =35 ± 9 nM, JNK2 = 94 ± 9 nM, and JNK3 = 117 ± 9 nM), CDD-2728 and CDD-3013 were 150- to 300-fold more potent in biochemical assays. When tested in a NanoBRET intracellular target engagement assay, the half-maximal inhibitory concentration (IC50) for CDD-2728 was JNK1 = 0.61 ± 0.06 µM, JNK2 = 0.57 ± 0.07 µM, and JNK3 = 0.53 ± 0.09 µM, and for CDD-3013 was JNK1 = 0.15 ± 0.02 µM, JNK2 = 0.15 ± 0.01 µM, and JNK3 = 0.11 ± 0.1 µM ( Fig. 1 B ), suggesting that cell permeability limited target engagement of the JNK isoforms. Metabolic studies indicated that CDD-3013 was more stable in human liver microsomes (HLM, t 1/2 = 425.4 min) than CDD-2728 (HLM, t 1/2 = 74.6 min) ( Fig. 1 B ). To define CDD-3013’s cellular kinome-wide selectivity at 1 µM, we used a NanoBRET K192 assay (Promega) with a 192 full-length kinase panel in human embryonic kidney cells (HEK293) ( 13 ). CDD-3013 engaged the JNKs with the highest affinity (JNK1, 95.73%; JNK2, 94.1%; and JNK3, 73.9% occupancy) while other kinases were engaged with lower occupancy (SNRK, 24.8% occupancy, TIE1, 22.9% occupancy, and MAPK11, 22.7% occupancy) ( Fig. 1 D and Dataset S1 ). Thus, CDD-3013 was identified as a selective JNK-I ( Fig. 1 C and D ). To evaluate additional off-target effects, biochemical assays evaluated the specificity of CDD-2728 and CDD-3013 for MEK4, which was not included in the K192 assay. These assays showed that CDD-2728 (MEK4, Kd = 18 ± 2 nM) was 58-fold- and CDD-3013 (MEK4, Kd = 30.5 ± 1.9 nM) was 254-fold selective for JNK1 over MEK4, achieving selectivity for JNK1 and reducing off-target liabilities ( Fig. 1 B ). To visualize the binding mode, crystal structures of JNK1 bound to CDD-2728 and CDD-3013 were determined at 1.65 Å and 1.55 Å resolution ( Fig. 2 and SI Appendix , Table S1 ). Both complexes were crystallized in the same space group with nearly identical unit cell dimensions. Both crystals contain one molecule per asymmetric unit and show clear density for bound inhibitors ( Fig. 2 B ). The kinase domain used in the expression construct was visible excluding the first 7 residues and a segment of the activation loop (residues 175 to 179) in both structures, and a segment of the glycine rich loop (residues 37 to 38) only in the JNK1/CDD-3013 complex. CDD-2728 overlaps the adenine region of the ATP site and an adjacent pocket away from the ATP site consisting of β3-β5 and β8 but does not extend into the ATP ribose and phosphate pockets ( Fig. 2 A ). The azaindole (BB3) binds the adenine pocket near the hinge region (backbone atoms of Glu110 and Met112) through hydrogen bonds ( Fig. 2 C and D ). The benzo thiophene (BB2) is sandwiched between β3 (prior to the P-loop) and β8 (posterior to the hinge) so that it interacts with Ile33, Gly33, Ala113, Asn114, and Gln117 through van der Waals interactions. Last, the phenol dimethyl (BB1) docks to a shallow pocket formed at one side of β3-β5 and β8 while its connecting nitrogen forms a hydrogen bond with the backbone carbonyl atom of Met112 ( Fig. 2 D and E ). The phenol dimethyl interacts with Asn 28, Lys30, Ala 42, Ala43, Asn51, and Leu110 through van der Waals interactions. The dimethyl moiety is replaced with a pyrrolidine in CDD-3013 and shows almost identical binding mode to CDD-2728. The pyrrolidine docks to the same pocket that the dimethyl binds but makes additional interactions with Tyr44 through van der Waals interactions ( Fig. 2 E and SI Appendix , Fig. S2 A and B ). We used modeling studies to evaluate if the pyrrolidine group in CDD-3013 creates a steric clash with other kinases’ binding pockets, which could explain CDD-3013’s increased selectivity for JNKs. Docking models of CDD-3013 onto CLK2 and MEK4 show that the pyrrolidine pocket is missing, and the pyrrolidine group forms a steric clash with bulkier residues Glu177 and Ser55 in MEK4 (A42 in JNK1) ( SI Appendix , Fig. S2 C ). Crystal structure of the JNK1/CDD-2728 and JNK1/CDD-3013 complexes. ( A ) The electrostatic potential surface near the active site with the bound CDD-2728. CDD-2728 binds the adenine site of the ATP pocket and the adjacent pocket consisting of residues from β3-β5 without occupying the ribose and phosphate pockets. ( B ) 2 Fo - Fc density for CDD-2728 and CDD-3013 in the JNK1 inhibitor complexes contoured at 1σ. Carbon atoms are in yellow; oxygens are red, nitrogens are blue. ( C ) Detailed interaction between JNK1 and CDD-2728. Hydrogen bonds between Hinge residue backbone atoms and CDD-2728 are shown with their distances. Key interacting residues are shown as sticks. ( D ) The phenol dimethyl moiety binds to the pocket in the small lobe. The electrostatic potential surface near the pocket at the small lobe is shown with 2 Fo - Fc density contoured at 1σ for the bound CDD-2728. ( E ) The pyrrolidine replaces the dimethyl in CDD-3013 and binds to the same pocket in the small lobe. To demonstrate target engagement of JNK-Is CDD-2728 and CDD-3013 we evaluated JUN phosphorylation in 12Z peritoneal endometriotic epithelial cells ( Fig. 3 ) ( 14 ). 12Z cells treated with CDD-2728 (0.1,1 and 10 µM) for 30 min, followed by 15 min of IL1β stimulation caused dose-dependent reductions in IL1β-mediated phosphorylation of Ser73 (pSer73) and pSer63 JUN phosphorylation ( Fig. 3 A − C ). Compared to IL1β-stimulated cells, treatment with 10 µM CDD-2728 + IL1β significantly reduced JUN pSer73 (1.88 ± 0.37 vs. 0.32 ± 0.08-fold change, P 0.99; 1 µM: 1.02 ± 0.27, P = 0.22) ( Fig. 3 A and B ). Relative to IL1β-treated cells, CDD-3013 significantly reduced JUN pSer73 at 0.1 µM (0.74 ± 0.012, P = 0.02), 1 µM (0.43 ± 0.153, P = 0.002), and 10 µM (0.23 ± 0.081, P < 0.001) ( Fig. 3 A and B ). When treated with BEND or TANZ with IL1β, only 10 µM TANZ significantly decreased JUN pSer73 relative to IL1β (0.74 ± 0.29, P = 0.02) ( Fig. 3 B ). A different JNK inhibitor, SP600125 ( 15 ), showed a significant reduction in pSer73 JUN at 1 µM (0.68 ± 0.19, P = 0.01) and 10 µM (0.48 ± 0.21, P = 0.003) relative to IL1β ( Fig. 3 B ). Relative to IL1β-treated cells, IL1β with 10 µM CDD-2728 (3.63 ± 0.89 vs. 0.45 ± 0.09, P = 0.02) and 10 µM CDD-3013 (0.22 ± 0.058, P = 0.01) resulted in a significant reduction of IL1β-mediated pSer63 JUN ( Fig. 3 C ). No significant reduction was observed in pSer63 JUN when the cells were treated with BEND, TANZ, or SP600125 ( Fig. 3 C ). Thus, our results indicate that CDD-2728 and CDD-3013 suppress IL1β-mediated phosphorylation of JUN Ser73 and Ser63, at similar or lower doses than BEND, TANZ, and SP600125. CDD-2728 and CDD-3013 inhibit IL1β-induced phosphorylation of downstream effectors c-JUN and JNK involved in inflammatory response in endometriotic epithelial cells. 12Z cells were incubated with JNK-I’s for 30 min prior to addition of 10 ng/mL IL1β. Cells exposed to neither IL1β or JNK-I and cells exposed to combinations of inhibitors at 0,1 1, and 10 µM and IL1β (10 ng/mL) were cultured for 15 min. ( A ) Cell lysates were loaded onto SDS-PAGE gels and immunoblotted to detect phosphorylated Ser63 or Ser73 c-JUN and GAPDH as loading control. ( B and C ) Quantitation of pJUN Ser73 ( B ) and pJUN Ser63 ( C ) band density on western blots was normalized for the respective total JUN and expressed relative to loading control GAPDH. ( D and E ) Volcano plots showing the differentially expressed transcripts when comparing cells treated with 10 ng/mL IL1β vs. vehicle ( D ) or 1 µM CDD-3013 + 10 ng/mL IL1β vs. 10 ng/mL IL1β ( E ) using a cut-off of >1.2-fold, <0.8-fold and FDR < 0.05. ( F ) Venn diagram showing the number of genes that were suppressed by 1 µM or 10 µM CDD-2728 and 1 µM CDD-3013 + IL1β when compared to the genes that were increased by 10 ng/mL IL1β vs. vehicle. This comparison excludes transcript variants mapping to the same gene and other unmapped transcripts. ( G ) Sankey plot displaying the genes and their corresponding gene ontologies in the 1 µM CDD-2728 + IL1β vs. IL1β comparison. ( H ) Heatmap displaying a subset of inflammation-, migration-, and adhesion-related genes that are generally increased by IL1β but decreased with the CDD-2728 and CDD-3013 inhibitors. We also evaluated the impact of JNK-Is on JNK phosphorylation ( Fig. 3 A ). As expected, IL1β induced robust phosphorylation of JNK1 (12.37 ± 2.52-fold change) and JNK2 (18.79 ± 5.13-fold change) relative to vehicle controls ( Fig. 3 A and SI Appendix , Fig. S3 A and B ). While neither BEND nor TANZ had an impact on JNK phosphorylation, SP600125 showed a dose-dependent decrease, though this was not statistically significant ( Fig. 3 A and SI Appendix , Fig. S3 A − C ). We also observed effects of CDD-2728 and CDD-3013 on JNK phosphorylation that were not statistically significant ( Fig. 3 A and SI Appendix , Fig. S3 A − C ). Treatment with TANZ and BEND had measurable but not significant increases in JNK1 and JNK2 that was associated with concerns of compensatory JNK activity. Thus, CDD-2728 and CDD-3013 decreased c-JUN phosphorylation with greater cellular potency than BEND, without invoking a compensatory response among JNK1 or JNK2. To establish the cellular impact of CDD-2728, we evaluated the dose-dependent response of MMP3 and IL8 expression in endometriotic 12Z cells treated with IL1β for 6 h ( SI Appendix , Fig. S3 D and E ). Compared to IL1β-treated cells, 5 µM CDD-2728 significantly decreased the expression of MMP3 (11.82 ± 0.503 vs. 0.917 ± 0.06, P = 0.01) and IL8 (327.2 ± 13.65 vs. 182.0 ± 12.17, P = 0.02) after 6 h ( SI Appendix , Fig. S3 D and E ). These results suggested that IL1β induced an inflammatory response attenuated by CDD-2728. To determine whether the effects of CDD-2728 and CDD-3013 on 12Z cells were from specific targeting of inflammatory networks or impacts on cell viability or apoptosis, we performed dose- and time-dependent viability and apoptotic studies ( SI Appendix , Fig. S4 ). Cell metabolic viability (ATP production, CellTiterGLO) decreased only at the highest concentration of BEND, CDD-2728, and CDD-3013 (100 µM) while apoptosis (cleaved caspase 3/7) was not significantly increased at concentrations up to 100 µM for any JNK-I ( SI Appendix , Fig. S4 A − D ). The broad-spectrum kinase inhibitor, staurosporine, decreased viability, and increased apoptosis over all concentrations from 0.1 to 100 µM ( SI Appendix , Fig. S4 C and D ). Additionally, we confirmed viability preservation in nonendometrial HepG2 cells where CDD-3013 and CDD-2728 showed no impact on viability up to 130 µM or 50 µM, respectively ( SI Appendix , Fig. S4 F ). These results indicated that CDD-2728 and CDD-3013 induced a significant decrease in proinflammatory signaling at doses that did not impact cellular metabolism, viability, or apoptosis. However, impacts on viability (BEND and CDD-2728) and apoptosis (BEND) were observed when treated at the much higher dose of 100 µM, reflecting a 100-fold safety margin between on-target expression of inflammation-related genes and general safety of JNK inhibition. When we induced inflammation in 12Z cells with 10 ng/mL IL1β, we observed that treatment with CDD-2728 and CDD-3013 changed gene expression signatures ( Fig. 3 D - H ). After a 6-hour incubation, IL1β increased 2,097 (>1.2-fold change, FDR < 0.05) and decreased 1,308 transcripts (<0.8-fold change, FDR < 0.05) ( Fig. 3 D and SI Appendix , Fig. S5 C ). Gene ontology (GO) analysis of differentially expressed transcripts by IL1β corresponded to categories of “tumor necrosis factor (TNF) signaling pathway,” and “pathways in cancer” ( SI Appendix , Fig. S5 D ). When compared to IL1β alone, IL1β + 1 µM CDD-2728 for 6 h increased 188 and decreased 145 transcripts ( SI Appendix , Fig. S5 A and C ). GO analysis of decreased transcripts revealed enrichment of “MAPK signaling pathway,” “rheumatoid arthritis,” and “IL-17 signaling pathway” ( SI Appendix , Fig. S5 E and Dataset S2 ). When compared to IL1β-treated cells, IL1β + 10 µM CDD-2728 increased 1,532 and decreased 1,362 transcripts with overrepresentation of “TNF signaling pathway,” “proteoglycans in cancer,” and “apoptosis” ( SI Appendix , Fig. S5 B , C , and F and Dataset S2 ). Similarly, cells treated with IL1β + 1 µM CDD-3013 increased 2,117 and decreased 1,335 transcripts ( Fig. 3 E and SI Appendix , Fig. S5 C and F ). We then identified 91 genes suppressed by CDD-2728 and CDD-3013 and increased by IL1β, relative to the vehicle ( Fig. 3 F and Dataset S3 ). Overrepresented categories by these 91 genes are displayed as a Sankey plot ( Fig. 3 G ) and a subset of these genes is displayed as a heatmap in Fig. 3 H . To establish potency relative to BEND, we cultured 12Z cells with IL1β ± CDD-2728, CDD-3013, BEND, and measured gene expression responses using quantitative PCR ( SI Appendix , Fig. S5 H − U ). As expected, IL1β increased the expression of genes involved in inflammation ( CXCL8, TNFRSF9, IL18 ), adhesion-related genes ( TNC ), and phosphatases ( DUSP1, -3, -8 ), while 0.1 µM, 1 µM, and 10 µM CDD-2728 and CDD-3013 significantly reduced the expression of CXCL8, TNFRSF9, IL18, TNC , and DUSP1, -3, -8 ( SI Appendix , Fig. S5 H − N ). When compared to BEND, treatment with CDD-2728 and CDD-3013 was 50-fold more potent in suppressing IL1β-induced expression of IL18, TNC, DUSP8 , CREB5 , and TRAF1 ( SI Appendix , Fig. S5 J , K , S , and U ). Overall, these studies show that CDD-2728 and CDD-3013 are more potent suppressors of inflammation than BEND. Endometriotic lesions in the ovary represent an advanced state of disease and consist of endometrial epithelial and stromal cells lining endometriotic cysts with active MAPK prosurvival signaling ( 16 ). Primary cultures of ovarian lesion endometriotic stromal cells were treated with 10 ng/mL of IL1β with or without 1 µM CDD-2728. Gene expression patterns showed that compared to vehicle-treated cells, IL1β increased 2,307 and decreased 1,840 transcripts ( Fig. 4 A and B and SI Appendix , Fig. S6 A and B and Dataset S4 ). Concurrent treatment of 1 µM CDD-2728 and IL1β decreased 266 and increased 236 transcripts compared to the IL1β-treated cells ( Fig. 4 B ). Upstream regulator analysis of the differentially expressed genes induced by CDD-2728 predicted JUND, SP1, and GATA6 as key transcription factors controlling the gene expression differences ( SI Appendix , Fig. S6 E ) ( 17 ). GO analyses showed that gene expression changes induced by CDD-2728 corresponded to pathways such as, “senescence and autophagy,” “BDNF signaling pathway,” and “TNF effects on cytokines” ( SI Appendix , Fig. S6 F and Dataset S4 ). Inhibition of JNK signaling with CDD-2728 and CDD-3013 suppresses proinflammatory signaling in ectopic stromal cells of patients with endometriosis. Transcriptomic profiling of primary endometriotic stromal cells from an endometrioma treated with vehicle (DMSO), 10 ng/mL IL1β, or 1 µM CDD-2728 + 10 ng/mL IL1β, for 6 h. ( A and B ) A volcano plot displaying the number of differentially expressed transcripts when comparing samples treated with 10 ng/mL IL1β vs. vehicle ( A ) or 1 µM CDD-2728 + IL1β vs. IL1β ( B ) using a cut-off of >1.2-fold, <0.8-fold, and FDR < 0.05. ( C ) Heatmap displaying a subset of genes that are increased by IL1β relative to the vehicle and suppressed by 1 µM CDD-2728. ( D − K ) Differentially expressed genes that were identified by RNAseq were validated by RT-qPCR in the endometriotic stromal cells isolated from the endometriomas of three additional patients. The cells were treated with 10 ng/mL IL1β, 1 µM or 10 µM CDD-2728 + 10 ng/mL IL1β, or 1 µM or 10 µM CDD-3013 + 10 ng/mL IL1β for 6 h. Each patient is displayed as a different symbol, and data are presented as mean ± SEM, analyzed with a Kruskal–Wallis test with Dunn’s post hoc test. * P < 0.0332; ** P < 0.0021; *** P < 0.0002; **** P < 0.0001. ( L ) Plots showing the top ligand–receptor interactions between human endometriotic stromal cells and endometrial macrophage populations (eM1, eM2, Monocyte), ranked by interaction score. Asterisks denote DESeq2 adjusted p-values for ligand differential expression: * P < 0.05, ** P < 0.01, *** P < 0.001. ( M ) Plot depicting putative ligand–receptor interactions between IL-1β-stimulated endometriotic stromal cells ( Left ) and 19 endometrial immune cell populations from the Human Endometrial Cell Atlas ( Right ). To identify the gene-specific impact of CDD-2728, we identified genes that were increased by IL1β treatment vs. vehicle and decreased by CDD-2728 + IL1β vs. IL1β. In total, we identified 184 genes that were both increased by IL1β and decreased by CDD-2728 ( Dataset S4 ). Key genes such as CSF3, CCL20, IL33, IL11, MMP3 , and IL16R are displayed in a heatmap ( Fig. 4 C ) summarizing IL1β-induced inflammation related genes suppressed by 1 µM CDD-2728. We further tested these gene expression changes in primary derived endometriotic stromal cells from three additional patients ( Fig. 4 D – K ), which showed a dose-dependent decrease in IL1β-induced inflammatory gene expression when treated with 1 µM or 10 µM CDD-2728 or CDD-3013. These results show that CDD-2728 and CDD-3013 reduce gene expression of inflammatory pathways in primary stromal cells from endometriotic lesions. To characterize how JNK inhibitors impacted endometriotic stromal cell communication with immune cells, we integrated bulk RNAseq datasets from endometriotic stromal cells ( Dataset S4 ) with single cell RNA-sequencing from the Human Endometrial Cell Atlas (HECA) by calculating interaction scores using the LIANA framework ( 18 , 19 ). Endometrial macrophages (eM, CD14+, CD68+, CD163+ ) were extracted from the immune cell subset resulting in eM1 (n = 3,854 cells), eM2 (n = 1,643 cells), and monocytes (n = 647 cells) ( SI Appendix , Fig. S7 A ). eM1 macrophages showed enriched proinflammatory markers ( TNF, IL1B, IL6, CXCL10 ), consistent with an activated inflammatory environment, eM2 macrophages were enriched for LYVE1 and HMOX1 , suggesting they are tissue-resident macrophages, while monocytes expressed high VCAN , a canonical marker of inflammatory monocytes, and low CSF1R , representing a spectrum of monocytes recruited from the circulation ( SI Appendix , Fig. S7 B ). The macrophage and monocyte population were used to identify ligand–receptor interactions with the endometriotic stromal cells using the IL1β vs. vehicle and IL1β + CDD-2728 vs. IL1β DESeq2 comparisons ( Fig. 4 L ). Interaction scores were calculated as the fold change for each ligand and the normalized mean receptor expression within the eM1, eM2, and monocyte populations. The strongest IL1β-induced interactions were CSF2→ CSFR1, and CSF2 → ITGB1, HGF → CD44, NRG1 → HLA-DPB1, and HAS2 → CD44, reflecting activation of macrophage survival, adhesion, and growth factor signaling axes ( Fig. 4 L ). Ligands showing the largest JNK-dependent suppression included HBEGF, MMP12 , and NRG1 , leading to significantly reduced interaction scores with their receptive macrophage receptors ( Dataset S5 ). Notably, HBEGF → CD44, and MMP12 → PLAUR, targeting eM1 inflammatory macrophages, showed the greatest JNK-dependent suppression, while CD3 → CD81 and CXCL10 → TLR4 interaction scores also decreased in eM2 tissue-resident macrophages ( Fig. 4 L ). The magnitude and directionality of the impact of IL1β and JNK inhibitors on these interactions is also visualized in a communication network plot ( SI Appendix , Fig. S7 C ). This suggested that JNK inhibitor, CDD-2728, has the potential to broadly disrupt ligand/receptor mediated communication between endometriotic stromal cells and macrophage populations represented in patients with endometriosis. We further determined whether the ligand/receptor interactions were specific to the macrophage lineage or directed toward multiple immune cell populations by extending the analysis to the 19 immune cell types in the HECA dataset ( Fig. 4 M ). Across the 957 unique ligand–receptor pairs and 7,121 interactions, the stromal MMP12 ligand showed the most macrophage-restricted signaling with PLAUR having the highest expression in monocytes and eM1 cells ( Fig. 4 M ). Similarly, HGF → CD44 interactions were the strongest in macrophage lineages, though the mast cells displayed similar interaction scores. The NRG1→HLA-DPB1 interaction scores were highest in cDC1 and cDC2 populations, with macrophages displaying lower interaction scores, indicating that NRG1 from stromal cells may primarily target dendritic cells in the lesion microenvironment. Stromal derived CSF2 was broadly distributed among several immune cell types, including peripheral dendritic cells (pDCs, via IL3RA), uNK cells, and T cells (via ITGB1), and macrophages (via CSF1R) ( Fig. 4 M and SI Appendix , Fig. S7 D ). Signaling via CXCL8 via CD79 had the strongest interaction with B cells (score = 29.1). These analyses suggest that while some JNK-dependent interactions are macrophage dependent (ie., MMP12→PLAUR, HBEGF→CD44), other IL1β-induced stromal signals (ie., NRG1, CSF2, CXCL8 ) engage multiple immune cell lineages and are not macrophage-specific. We measured several intrinsic properties of these JNK-Is to determine their oral bioavailability. Microsomal stability of CDD-3013 was more favorable than CDD-2728 in human microsomes [half-life (T 1/2 )= 425.4 min and 74.6 min, respectively], while clearance in mouse microsomes indicated lower stability of CDD-2728 and CDD-3013 (T 1/2 = 85.2 and 74.7 min, respectively) than desired for oral administration in mice ( Fig. 1 B ). Permeability across Caco-2 monolayers was acceptable [Papp(a-b) = 20.87 ± 0.92 nm/s; a-b for CDD-2728 and Papp(a-b) = 9.13 ± 0.92 nm/s for CDD-3013] but efflux was higher than desired [Papp(b-a) = 87.49 ± 4.43; ER = 4.2 for CDD2728 and Papp(b-a) = 21.31 ± 1.23; ER = 2.34 for CDD-3013], attributed to low aqueous kinetic solubility of both compounds (<0.1 mg/L, PBS, pH6.5) that may limit oral bioavailability. Pharmacokinetics of CDD-2728 and CDD-3013 were obtained by oral or intraperitoneal (i.p) administration ( SI Appendix , Figs. S8-S9 ). Both CDD-2728 and CDD-3013 (50 mg/kg) had low maximum plasma concentration (C max = 1.51 ± 0.53 µmol/L and C max = 8.95 ± 1.67 µmol/L, respectively) by oral administration (low absorption), and the compounds were rapidly cleared (T 1/2 = 1.48 ± 0.32 h, 1.79 ± 0.23 h, respectively) ( SI Appendix , Figs. S8 A and B and S9 A and B ). Less than 0.5% of either compound was unbound to plasma proteins. Levels of CDD-2728 and CDD-3013 following i.p. administration were determined in separate animals ( SI Appendix , Figs. S8 C and D and S9 C and D ). The C max achieved following i.p. administration of CDD-2728 and CDD-3013 (50 mg/kg) was 0.44 ± 0.19 µmol/L and 4.25 ± 2.57 µmol/L, with T 1/2 of 14.23 ± 8.92 and 1.29 ± 0.32 h, respectively. No evidence of alerts in Mini-Ames or micronucleus evaluation was detected ( SI Appendix , Fig. S10 ). While oral bioavailability requires optimization for clinical translation, i.p. administration of JNK-I provided sufficient systemic exposure to demonstrate disease-modifying potential in preclinical models. In summary, both compounds showed poor oral absorption and rapid clearance, meaning that only small amounts reached the bloodstream and were rapidly cleared when taken by mouth. These findings are related to the solubility of the compound, limiting how much of the compounds could be absorbed by the gut and metabolic clearance of the compounds by the liver. For the purpose of confirming disease modification CDD2728 and CDD-3013 were administered i.p. and bypassed oral absorption. This allowed us to achieve higher and more sustained compound concentrations to assess on-target efficacy and safety of the JNK inhibition mechanism in a disease model. We used a mouse allograft method established in our group to evaluate potential therapeutics to determine the efficacy of CDD-2728 and CDD-3013 ( Fig. 5 ) in reducing endometriotic lesion burden ( 20 ). Established endometriosis lesions (2 wk postinduction) were subjected to either CDD-2728 or CDD-3013 every other day for an additional 2 wk ( Fig. 5 A ). We tested the JNK-I, CDD-2728, at two different doses based on pharmacokinetic studies (5 mg/kg and 25 mg/kg) and evaluated the effects on lesions. CDD-2728 at 5 mg/kg showed that the size and volume of lesions were not significantly different from vehicle-treated animals ( Fig. 5 B – E ). At 25 mg/kg, there were significant reductions in lesion number, volume, and mass, with most animals (3/5) having no visible lesions ( Fig. 5 B – E ). In a parallel set of experiments, CDD-3013 (10 mg/kg) treatment caused a significant reduction in the number, volume, or mass of lesions or their elimination ( Fig. 5 F – J ). Immunohistochemical analysis of retrieved lesions indicated reduced rates of proliferation (Ki67) and absent macrophage (F4/80) staining in mice treated with 5 mg/kg or 25 mg/kg CDD-2728 and 10 mg/kg CDD-3013 ( Fig. 5 E and I ). For both JNK-Is, there were no changes in body weight, liver weight, or kidney weight. We also determined that administration of CDD-3013 had no negative impact on the histoarchitecture of the eutopic endometrium of recipient mice and maintained normal expression of estrogen receptor alpha (ERα), progesterone receptor (PR), and the glandular endometrial marker, FOXA2 ( SI Appendix , Fig. S11 ). Administration of CDD-2728 and CDD-3013 reduces lesion number and decreases macrophage infiltration and proliferation in an induced mouse model of endometriosis. ( A ) Experimental scheme used to induce endometriosis and administer treatment with CDD-2728 and CDD-3013 JNK inhibitors (i.p.). ( B – E ) The impact of CDD-2728 was evaluated in an experimentally induced model of endometriosis in mice by treating with 5 mg/kg and 25 mg/kg of CDD-2728. ( B and C ) Representative gross morphology and summarized quantification of the endometriosis lesions obtained from mice after a treatment regimen with CDD-2728. ( D and E ) Histological evaluation of the lesions using H&E ( D ), or Ki67 to assess proliferation and F4/80 to visualize macrophages within the lesions. ( F – I ) Evaluation of lesions following CDD-3013 administration (10 mg/kg) in the experimental mouse model of endometriosis. ( F ) Representative images of the lesions, and quantification of lesion number, volume, and mass (G). ( H an I ) Histological evaluation of the lesions by H&E ( H ) and with F4/80 ( I ) to visualize macrophage infiltration. Data are presented as Mean ± SEM. * P < 0.05, ** P < 0.01, *** P < 0.001, vs. Vehicle by one-way ANOVA followed Dunnett’s Multiple Comparison Test. Assessment of pain has become the primary clinically relevant outcome for drug approval in endometriosis patients. The effects of experimental endometriosis treatments on pain behaviors require 4 to 6 wk to complete the in-life phase and additional months for the extensive analysis of the behavioral responses. To implement a more rapid and cost-effective path to prioritize compounds for subsequent intensive endometriosis assessments, we measured lipopolysaccharide (LPS)-induced pain by a dynamic weight bearing test that takes six hours to assess ( SI Appendix , Fig. S12 A ) ( 21 , 22 ). Administration of LPS caused a postural shift toward the front paws as the presence of abdominal pain discourages the animals from bearing their entire body weight on their rear paws. This pain-induced postural change was quantified by measuring weight, area, and time spent on front and rear paws and duration of rearing ( SI Appendix , Fig. S12 B – E ). The presence of pain in animals is also evaluated by the display of grimace facial expression ( SI Appendix , Fig. S12 F ). Both CDD-3013 and rofecoxib, a nonsteroidal anti-inflammatory drug, caused a significant decrease in weight, area, and time spent on front paws, suggesting a decrease in perceived abdominal pain. Rofecoxib and CDD-3013 increased rearing behaviors and decreased Grimace scores, reflecting facial expressions of discomfort caused by LPS-induced pain ( SI Appendix , Fig. S12 ).

To discover novel chemical scaffolds for JNK-Is, we deployed a compound collection from the BCM DNA-encoded chemistry technology (DEC-Tec), which is an innovative and highly cost-effective screening platform for identification of new small molecule hits for various targets ( 23 – 25 ). The DEC-Tec allows for the screening of vast chemical libraries with diverse building blocks ( 26 ). Our group has reported several significant hit molecules for diverse targets using DEC-Tec ( 27 – 31 ), emerging with very low hit rates and high confirmation rates. Following DEC-Tec selections and chemical optimization of the hit molecule CDD-2428, CDD-2728, and CDD-3013 emerged as valuable leads for improved JNK-Is. The selectivity of CDD-2728 and CDD-3013 for JNK1,3 relative to JNK2, kinases within the proximal MAPK signaling pathway (MEK-ERK, p38MAPK, NFKβ) and the broader kinome was a feature of DECL synthesis and guidance for novelty provided by our structural biology efforts. This was reflected in the selectivity values for JNK1 relative to MEK4, where CDD-2728 displayed a 58-fold higher selectivity for JNK1 (Kd = 0.31 nM) when compared to MEK4 (Kd = 18 nM), and in CDD-3013, which was found to be 254-fold more selective for JNK1 (Kd = 0.12 nM) than MEK4 (Kd = 30.5 nM). The selectivity of CDD-2728 and CDD-3013 was initially assessed by a KinomeScan assay, which was used as the basis to further evaluate their in vitro selectivity with the LanthaScreen assay (from ThermoFisher assay), and their selectivity in intact cells with the NanoBRET K192 cell-based assay (from Promega) ( 13 ). The KinomeScan data will be included in a future paper that more closely describes the medicinal chemistry approach used to generate these compounds. The NanoBRET K192 assay showed that this compound had the highest % occupancy for JNK1 (95%), JNK2 (94%), and JNK3 (73%). The next highest % occupancy for CDD-3013 was determined to be for SNRK (24%), TIE1 (22%), MAPK11 (22%), CSNK1D (19%), and CAMK1 (19%). This kinase profiling assay showed that the JNK1, -2, and -3 isoforms were the most potent targets of CDD-3013, though the possibility for minimal off-target activity could not be ruled out. Despite showing increased affinity for JNK1/3 in vitro, selectivity for the JNK isoforms was not maintained in the NanoBRET assays, suggesting that additional factors within the cell impacted selectivity. This could be attributed to JNK isoform expression in isolated cells, or JNK isoform interactions with scaffolding proteins that mask the structural selectivity of the inhibitors displayed in vitro, where no other proteins are present. Additional factors would include minor changes to protein conformation when fused to NanoLuc, or differences in the rebinding kinetics of the JNK isoforms to the inhibitors when performed in vitro versus in cells. Collectively, the subset of Serine/Threonine (Ser/Thr) kinases share several features of ATP-binding pockets within a hinge domain. Active site inhibitors interrupt ATP binding into this active site. Allosteric kinase inhibitors are regarded as a path to obtain desirable selectivity because they target amino acids outside the conserved hinge-binding domain of Ser/Thr kinases. CDD-2728 and CDD-3013 integrate hinge-binding features with allosteric-like properties realized through targeting the shallow selectivity pocket that is unique for JNKs relative to other kinases. The cocrystal of CDD-2728 or CDD-3013 with JNK1 and JNK3 identifies occupation of the selectivity pocket space by dimethyl amino or pyrrolidine of CDD-2728 and CDD-3013 respectively. These chemical groups are anticipated to create steric clashes with other kinases that lack this fold/pocket that were confirmed through Alpha-Fold modeling of CDD-3013 fit into the similar region of MKK4. Our structural biology informed that the azaindole engages the hinge-binding domain while the benzothiophene orients the phenyl group to enable access to the shallow selectivity pocket. Optimization of each of these structural features of CDD-2728 or CDD-3013 are underway to achieve orally bioavailable JNK inhibitors that take advantage of each of the three chemical subfeatures of these selective JNK inhibitors. In endometriotic cells, we demonstrated more robust inhibition of JUN phosphorylation with CDD-2728 and CDD-3013 compared to BEND. Our results showed CDD-3013 to be far more potent than BEND at inhibiting JUN pSer73 phosphorylation and similarly potent to TANZ. Although the trends were similar for pSer73 and pSer63, the inhibition of IL1β-mediated c-JUN phosphorylation at pSer73 was significantly reduced when compared to the inhibition of pSer63. Previous results indicate that both pSer63 and pSer73 are major sites of phosphorylation by the JNKs that correlate with c-Jun transactivation, and reported differences in the kinetics and functional roles of the phosphorylation events exist between the two sites ( 32 ). For example, Ser63, but not Ser73, phosphorylation, is correlated with nitric oxide–induced apoptosis in neuroblastoma cells ( 33 ). Likewise, in Alzheimer’s disease, pSer73 events are elevated in TUNEL-negative neurons, suggesting that p73, but not pSer63 c-Jun are protective phosphorylation events against cell death ( 34 ). Our results also demonstrate a significant impact of JNK-I on gene expression in endometriotic epithelial cells treated with IL1β, a representative proinflammatory signal commonly associated with endometriosis lesions. The concentration of IL1β incorporated in these results (10 ng/mL) invokes a robust inflammatory response. Changes in inflammation-induced gene expression by CDD-3013 at the concentrations assessed (1 µM) are separated by 100-fold concentrations from generalized disruption of cell viability (measured by ATP production) or induction of apoptosis (measured by cleaved caspase 3/7 activity). Additionally, we demonstrated 50-fold improved potency of CDD-3013 relative to BEND in modifying inflammatory gene expression signatures in endometriotic lesion stromal cells. This set of differentially expressed genes creates a candidate list of complementary biomarkers for further investigations and clinical validations of JNK-Is. Of particular interest in both endometrial epithelial and stromal cells were the changes in DUSP1, -3, -4, -5, -6, -7, -8, -10, -16 . Dual specificity phosphatases (DUSPs) reflect an acute survival response of a cell to extracellular factors that increase the network of serine/threonine and tyrosine kinases. This broad family of phosphatases offset kinase activity with endogenous homeostatic processes engaged by otherwise healthy cells to control acute inflammatory signals ( 35 ). Increased expression of DUSPs by IL1β and decreased expression in cells exposed to JNK-I signify a reduced demand for acute survival responses and offer a first line of evidence for JNK-I to temper chronic responses to inflammatory stress within lesions. We also integrated our transcriptomic results with single cell RNAseq data from the HECA database ( 18 ) to determine how JNK-I impacted interactions between endometriotic stromal cells and macrophages. These analyses showed that while stimulation with IL1β increased signaling between the stromal cells and endometrial macrophages, treatment with the JNK-I’s attenuated these interactions. We specifically observed that JNK-I’s decreased the ligand/receptor interaction scores between endometrial stromal cells and both type 1 (eM1) and type 2 (eM2) endometrial macrophages. For example, while IL1β treatment stimulated interactions with eM1 macrophages via MMP12/PLAUR, CD3/CD81, MMP1/CD44, and INHBA/ACTR2 signaling, the magnitude of the interaction scores significantly decreased when the cells were treated with the JNK-I, CDD-2728. Similar impacts were observed in the endometriotic stromal cells’ interactions with eM2 macrophages, where IL1β and JNK-Is significantly changed the interaction scores via the NRG1/HLA-DPB1, CD3/CD8, CXCL10/TLR4, and MMP12/PLAUR ligand/receptors. Both MMP12 and PLAUR are strongly expressed by endometrial stromal cells and macrophages, and their expression is strongly increased in proinflammatory macrophages, contributing to fibrosis and tissue remodeling ( 36 , 37 ). In macrophages, the CXCL10/TLR4 interactions are critical to driving inflammation, and promoting macrophage migration and infiltration ( 38 ). Thus, we anticipate that when administered in vivo, JNK-I’s have the potential to resolve endometriosis-associated inflammation by mitigating interactions between endometriotic stromal cells and macrophages. The ligand/receptor interaction analyses also suggest that JNK inhibitor, CDD-2728, may impact how VCAN + monocytes are recruited to lesions and impact their polarization into eM1 vs. eM2 tissue-resident macrophages, showing decreased ligand/receptor interactions scores across various interactions (i.e., NRG1/HLA-DPB1, MMP12/PLAUR, and HBEGF/CD44). Previous studies showed that the JNK inhibitor, SP600125, inhibited polarization of decidual macrophages into proinflammatory M1 macrophages, decreasing IL6 and TNFα expression, while driving polarization into M2 anti-inflammatory macrophages, with enhanced TGFβ and IL10 expression ( 39 ). Similar findings have been reported in other cell types, for example, inhibition of JNK signaling enhanced M2 macrophage polarization in osteoarthritis, and decreased M1 proinflammatory macrophage polarization in atherosclerosis ( 40 , 41 ). Additional studies would be necessary to test how JNK inhibitors, CDD-2728 and CDD-3013, impact macrophage polarization in endometriosis. Preclinical animal models of endometriosis provide glimpses on the pathophysiology of endometriosis and provide evidence to support nomination of a clinical candidate to move into clinical trials. Experiments performed in mouse endometriosis models demonstrated that CDD-2728 and CDD-3013 regressed lesions and decreased macrophage infiltration. Assessing the efficacy of CDD-2728 and CDD-3013 by evaluation of lesion regression and immunofluorescent staining of lesions was considered feasible and informative, while evaluation of behavioral pain assessments in mice bearing lesions was complicated by the current limitations of CDD-3013 metabolism following oral administration. Intraperitoneal administration every other day for 1 to 3 mo would complicate interpretation of abdominal pain assessments by mechanical hyperalgesia (von Frey apparatus), dynamic weight bearing, or spontaneous abdominal pain behaviors. The experiments performed in vivo demonstrated that a novel, specific, and highly potent JNK-I, represented by CDD-2728 and CDD-3013 caused regression of lesions and modified localization of macrophages in lesions. Macrophages were selected for this first evaluation of immune cell localization within lesions, raising anticipation that localization of other subsets of immune cells, such as NK cells, T-cells, or neutrophils, commonly associated with endometriotic lesions or peritoneal fluid of patients may be influenced by treatment with JNK-I. While these induced endometriosis mouse models do not fully recapitulate human disease, as they are not naturally menstruating and require intraperitoneal injection of minced endometrial tissues from donor mice, various models have been developed to study aspects of the disease and to test experimental drugs. Multiple investigators concede that there is no ideal rodent model to study endometriosis due to the complexity of the disease, however, using a harmonized rodent model approach is balanced by results from our in vitro human stromal cell responses describing interruption of inflammatory driver (IL1β) of endometriosis ( 42 ). Assessment of changes in postural behaviors caused by the presence of abdominal pain indicated that both rofecoxib and CDD-3013 reduced LPS-induced pain. We implemented a well-reported bioassay of LPS induced pain that involves JUN phosphorylation in vivo to assess whether LPS-induced pain could become a rapid PK/PD model during JNK-I optimization. This bioassay is intended to establish an association between drug exposure and inflammatory pain. It is not intended to replace endometriosis-induced pain models. The LPS-induced pain model requires 5 h rather than 3 mo of in-life phase to evaluate and is intended to simply demonstrate exposure-dependent interruption of pain among tens of compounds to prioritize 1 compound for evaluation in endometriosis-induced pain models. At least 2 measures of dynamic weight bearing shared with postural behaviors suggestive of endometriosis-associated abdominal pain indicated that both rofecoxib and CDD-3013 reduced some or many measures of LPS-induced pain. The Grimace score integrates facial features of mice treated with LPS-induced pain, and in this score the highest dose of CDD-3013 given by the i.p. route reduced the Grimace score but with 50% less impact than rofecoxib administered by oral gavage. These results do not imply that LPS-induced pain can be a pharmacodynamic screening tool for all drug discovery efforts related to endometriosis, nor that LPS-induced pain reflects nociceptive pain. Our results suggest there is room for optimization of ADME properties of compounds based on response of CDD-3013 relative to rofecoxib, and the variable responses for CDD-3013 within each measure of pain may be attributed to solubility, stability, permeability, and bioavailability of the present chemical probes. According to these cell-based and in vivo stores, we conclude that JNK-Is, CDD-2728 and CDD-3013 are effective for diminishing the inflammation associated with endometriosis and thereby decrease macrophage infiltration into the lesions. These observations were supported by our RNAseq results, which highlighted changes in epithelial cell cytokine-receptor signaling including ( IL-18, TRAF1, CXCL8, TNFRSF9 ), neuroangiogenesis and pain of endometriosis ( BDNF ) ( 43 ) and transcription factors ( LZST3, KLF11, KLF13, LRRFIP1, NEDD9, KLF1, JUND, JUNB, CREB5, KLF6, WT1 ) associated with IL1β responses that were inhibited by JNK inhibitors. Reduced expression of transcription factors associated with inflammation in stromal cells (ATF2 network, CREB5, JUND ), endometriosis extracellular matrix ( TIMP3, SERPINE1, TNC, LOX, MMP3, ITGA2 ) neuroangiogenic ( NGF ) and inflammatory cytokines ( IL-31, IL-33 ), in response to JNK-I signify the potential to impact multiple axes of this disease, previously pursued individually as therapeutic targets. Results with CDD-2728 and CDD-3013 caused reduced expression of HB-EGF that is anticipated to deliver similar efficacy as reducing HB-EGF processing ( 44 ). JNK-I-mediated inhibition of ALDH1A3 by CDD-3013 can be envisioned to reduce stem cell–like properties of endometriotic lesions ( 45 , 46 ). CXCL8 (aka IL-8) has been reported to be present at higher levels in peritoneal fluid of patients ( 47 ), and CDD-2728 and CDD-3013 reduced CXCL8 secretion. JNK-I also mediated reduction of CXCL1 (a neutrophil activating protein) and CXCL2 (aka MIP2a); both were reported elevated in peritoneal fluid of endometriosis patients ( 48 ), and may lead to reduced neutrophil activation and reduced oxidative stress imposed on endometriotic lesions. CSF2 (GM-CSF) and CSF3 (G-CSF)-treated macrophages induce nociceptive pain receptor transcription ( 49 ) while IL-17, a frequently detected cytokine in peritoneal fluid is a proinflammatory activator of GM-CSF production. These results demonstrated the molecular mechanism by which inhibition of JNK signaling with CDD-2728 and CDD-3013 can mitigate the inflammation and pain in endometriosis. In summary, we report the discovery and optimization of potent and selective JNK-Is as nonhormonal therapeutics for endometriosis using DEC-Tec. Two preclinical lead compounds suppressed inflammatory signaling in patient-derived cells and reduced lesion burden and pain in mouse models. These JNK-Is show strong target engagement, metabolic stability, and in vivo efficacy, offering evidence that an optimized JNK-I delivers therapeutic reduction of inflammation and pain-related features of endometriosis. Beyond endometriosis, dysregulated JNK signaling contributes to inflammatory pathology across multiple conditions, suggesting broader therapeutic potential for selective JNK inhibitors. Additional studies are underway, for CDD-3013 and other derivatives, that continue to improve the drug profile of this new class of compounds. The ability to target inflammatory mechanisms, while avoiding disruption of hormone signaling, positions JNK inhibition as a promising disease-modifying strategy for conditions affecting women of reproductive age. These findings establish CDD-3013 a first-in-class preclinical lead that addresses root inflammatory causes rather than symptoms alone, with the potential to prevent disease progression and accumulating morbidities as inflammation becomes chronic.

Tissues were collected from patients after obtaining informed written consent and following the guidelines as approved by the Baylor College of Medicine IRB protocol (H-21138 and H-46767). Tissues and any cells derived from them were deidentified prior to use to ensure patient confidentiality guidelines. Animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Baylor College of Medicine (BCM) and guidelines established by the NIH Guide for the Care and Use of Laboratory Animals. Mice were housed under standard conditions of a 12 h light/dark cycle in a vivarium that maintained a controlled ambient temperature of 20.0 °C ± 2 °C and a relative humidity of 20 to 70%. Following RNA isolation, ~200 ng was used for library preparation and RNA sequencing with Novogene, Inc. (Sacramento, CA). Illumina sequencing was performed using the Novaseq 6000 platform with ~20 to 30 million reads per sample. The reads were processed using a standardized bioinformatics pipeline to ensure high-quality and reproducible results. Raw sequencing reads underwent quality control and preprocessing using fastp ( 50 ) to remove adapter sequences, filter low-quality reads, and correct base errors. High-quality reads were then aligned to the human reference genome (hg38) using STAR ( 51 ), a splice-aware aligner optimized for RNA-seq data, to generate sorted BAM files. Gene-level quantification was performed using featureCounts ( 52 ) to obtain a raw count matrix. Differential gene expression analysis was conducted using DESeq2 ( 53 ), which normalizes count data and applies statistical modeling to identify significantly differentially expressed genes between experimental conditions.

Appendix 01 (PDF) Dataset S01 (XLSX) Dataset S02 (XLSX) Dataset S03 (XLSX) Dataset S04 (XLSX) Dataset S05 (XLSX)