An NT5E loss-of-function variant permits tissue inflammation and hypertension in systemic lupus erythematosus Authors: Isaac Peabody, Mikhail Olferiev, Katherine A Owen, Tyson Dawson, Prathyusha Bachali, Peter Kasson, Amrie C Grammer, Mary K Crow, Peter E Lipsky Published: 2026 ABSTRACT SummaryAdenosine, generated by CD73/NT5E, is a critical endogenous regulator of immune-mediated inflammation. Genetic variants in NT5E have been linked to vascular and inflammatory phenotypes. Systemic lupus erythematosus (SLE) is a chronic autoimmune disease marked by persistent inflammation. Using a systems biology approach, we examined immunologic and genetic alterations in adenosine signaling in SLE. Whereas gene signatures associated with extracellular adenosine production were elevated in lupus-affected tissues, adenosine-mediated immunosuppressive pathways were broadly downregulated and correlated with increased proinflammatory signatures. Analysis of the NT5E single-nucleotide polymorphism rs2229524, which introduces a non-synonymous amino acid change, predicted impaired enzymatic function and was associated with reduced adenosine immunomodulation and increased rates of pericarditis and hypertension in SLE patients. These findings highlight the importance of extracellular adenosine signaling in regulating immune-mediated inflammation in SLE and highlight the role of a loss-of-function NT5E variant in contributing to tissue injury.Subject areas: molecular genetics, immunology, omics Introduction Systemic lupus erythematosus (SLE) is a clinically heterogeneous autoimmune disease characterized by innate immune system activation, altered regulation of T and B cells, autoantibody production, immune complex deposition, and multisystem organ damage.1 Growing evidence suggests that an imbalance of pro- and anti-inflammatory signals and cytokines contributes to SLE development and its clinical manifestations.2,3,4 One potential source for this imbalance is dysfunction in anti-inflammatory signaling mediated by the purine nucleoside, adenosine. Extracellular adenosine production is initiated by CD39 (ENTPD1) and CD38/CD203a that catalyze the conversion of adenosine triphosphate (ATP) to adenosine monophosphate (AMP) and nicotinamide adenine dinucleotide (NAD+) to AMP, respectively.5,6 AMP is further processed to adenosine by the extracellular phosphatase, CD73 encoded by 5′ Ectonucleotidase Ecto 1 (NT5E). Extracellular adenosine acts as a local regulator of inflammation and immune activity by stimulating the G protein-coupled A1, A2A, A2B, and A3 adenosine receptors.7,8 In particular, A2A receptor stimulation inhibits the production of TNF, IL-1β, IL-2, IL-6, and IFN-γ and increases the production of the anti-inflammatory cytokine IL-10.9,10,11 Although adenosine signaling normally functions to suppress immune activity and inflammation, relatively less is known about the role of adenosine-mediated mechanisms in SLE. Both CD39 and CD73 are highly expressed on immunomodulatory T and B regulatory cells and effector memory T cells from healthy donors,12 and previous studies demonstrated that adenosine elicits an attenuated immunosuppressive response in T cells isolated from lupus patients.13,14,15 Furthermore, in the murine pristane model of SLE, CD73 knockout (KO) animals exhibited expansion of activated B cell and plasma cell populations along with elevated anti-RNP antibodies and exaggerated neutrophil extracellular trap (NET) release.16 CD73-deficient animals were also susceptible to arterial endothelial dysfunction.16 These findings are consistent with reports of deleterious NT5E mutations in human subjects that affect splicing and/or coding regions, associated with arterial calcification in both the upper and lower extremities, aortic calcification, as well as arthritis involving periarticular calcification.17,18,19,20,21,22 Importantly, analysis of SLE Immunochip association data suggested that polymorphisms in the coding region of NT5E, including rs2229524 (p = 5.67 × 10−3, odds ratio [OR]: 1.13) that encodes an M379T transition located in the enzymatic binding pocket required for the conversion of AMP to adenosine,23 might be associated with lupus. However, since the significance of this association was modest, not achieving the genome-wide significance threshold, it remains unclear what effect genetic variation in NT5E may play in relation to SLE disease activity, damage, or organ-system manifestations. Genetic predisposition imposes important risk factors for both SLE and cardiovascular disease. Here, we applied a comprehensive systems biology approach to dissect the immunologic and genetic implications of altered adenosine signaling in SLE patients. Gene expression analysis in SLE tissues revealed fundamental defects in the regulation of anti-inflammatory responses associated with altered adenosine activity. In addition, examination of the structural implications of genetic mutations in the coding sequence of NT5E predicted impaired enzymatic function that could be directly linked to elevated immune activity and proinflammatory markers, along with a greater frequency of hypertension in a cohort of SLE patients. Together, these genetic and gene expression analyses provide additional insight into the potential role of adenosine signaling in SLE, highlighting the significance of the adenosine pathway in the regulation of autoimmunity, inflammation, and vascular integrity. Results NT5E and ENTPD1 are overexpressed in lupus-affected tissues Microarray and RNA sequencing (RNA-seq) data derived from multiple SLE whole blood (WB), cell-type-specific, and tissue-specific datasets (Table S1) were first examined for differential expression of the ectonucleotidases NT5E and ENTPD1. We observed consistent differential expression of NT5E in cutaneous and discoid lupus erythematosus (CLE and DLE, respectively) skin datasets and in lupus nephritis (LN) glomerulus and, to a lesser extent, tubulointerstitium (TI) of LN kidneys (Figure 1A). This pattern was largely replicated for ENTPD1, although expression of this gene was not detected in CLE (GSE81071-CLE). In contrast, NT5E expression was broadly downregulated in WB, peripheral blood mononuclear cells (PBMCs), naive and memory B cells, and CD4 and CD8 T cells, whereas ENTPD1 expression was variable in WB, upregulated in PBMCs, downregulated in B cells, and undetected in T cells (Figure 1A). As there is considerable transcriptomic heterogeneity among lupus patients,24,25 we next sought to confirm enrichment for NT5E and ENTPD1 expression at the patient level using gene set variation analysis (GSVA).26 Whereas expression of NT5E and ENTPD1 was highly variable in WB samples, tissue biopsies from LN glomerulus and DLE skin exhibited significantly higher GSVA enrichment scores for both genes (Figure 1B). In contrast, NT5E, but not ENTPD1, was enriched in CLE, and neither gene was enriched in the TI of LN. As a control, inflamed synovial tissue from patients with rheumatoid arthritis (RA; GSE55457) was examined and revealed enrichment in ENTPD1 but not NT5E (Figure 1B). Together, these results suggest that the canonical adenosine production pathway catalyzed by the combined enzymatic activities of CD39/ENTPD1 and CD73/NT5E may be elevated in specific lupus-inflamed tissues, including DLE and the glomerulus of LN. Elevated NT5E and ENTPD1 expression in lupus kidney and skin (A) Log2 fold differential expression values for NT5E and ENTPD1 from lupus whole blood, PBMCs, B cells, T cells, LN kidney, and skin. (B) Individual sample gene expression from the blood, kidney, skin, or synovium was analyzed by GSVA for the enrichment of NT5E and ENTPD1. Asterisks (∗) indicate p < 0.05 using a Welch's t test comparing SLE to control where ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and p < 0.0001. Single-cell RNA-seq reveals elevated NT5E expression and shared adenosine regulatory pathways in endothelial cells, fibroblasts, and pericytes within lupus skin Given the tissue-specific upregulation of NT5E in lupus patients, we next sought to determine the specific cell types with increased NT5E expression within lupus-affected tissue. Using data from GSE17963327 (Table S1), we assessed single-cell RNA-seq (scRNA-seq) profiles of resident skin cells from epidermis and dermis as well as those of infiltrating immune/inflammatory cells from SLE and DLE cutaneous lesions, along with those from healthy controls. A total of 274,002 high-quality cells were obtained for further analysis after data integration and doublet exclusion. Cells were clustered and visualized using uniform manifold approximation and projection (UMAP), revealing 15 main clusters isolated from the dermis and epidermis. Cell types were determined based on a list of cell-specific gene markers, including genes specific for myeloid and lymphocyte populations, human keratinocytes and fibroblasts, endothelial cells, and pericytes27 (Figure S1). Cluster identification was determined by those marker genes (Figure 2A; Table S2), and all cell types were present in patients and controls regardless of disease status (Figure S2A). Cellular subtype analysis identified 32 subclusters, including numerous keratinocyte, fibroblast, pericyte, and endothelial cell sub-populations, as well as several populations of B, T, and myeloid cells (Figure 2B). NT5E expression in skin cell populations from SLE, DLE, and healthy donors (A and B) Major cell populations (A) and more granular subpopulations (B) were defined based on differentially expressed transcripts characteristic of each cell type. (C) Normalized NT5E expression is visualized in the UMAP plot. (D) NT5E expression levels in the defined major populations are shown as violin plots, with FDR-corrected p values indicating significant differences between donor groups. (E) NT5E expression in subpopulations is shown only for those with significant differences between donor groups. <5% false detection rate was considered significant, with level of significance in (D) and (E) indicated by ∗∗∗p < 0.0005, ∗∗p < 0.005, or ∗p < 0.05. scRNA-seq data from GSE179633, derived from skin biopsy samples of patients with SLE, DLE, and healthy donor controls, were analyzed. Consistent with bulk RNA analysis, elevated NT5E expression was detected in lupus tissues compared to normal controls, but upregulation was confined specifically to fibroblast, pericyte, and endothelial cell subsets and not to immune cells or keratinocytes (Figures 2C–2E). Importantly, significantly increased expression of NT5E in these tissue cells was observed in skin samples from SLE patients compared to controls (Figures 2D, 2E, S2B, and S2C). To characterize the expression of adenosine pathway-related genes across the cell populations identified in SLE and DLE skin, an adenosine pathway-related gene set was identified, including genes involved in extracellular nucleotide/nucleoside metabolism, receptor-mediated signaling, downstream cAMP responses, as well as genes encoding proteins known to physically interact with NT5E. A word cloud describing the most frequent abnormally expressed adenosine pathway genes in SLE skin populations compared to healthy skin identified CD38, PNP, ADA, and NT5E (Figure S3A), and increased numbers of adenosine pathway genes were particularly differentially expressed in SLE fibroblasts, endothelial cells, and pericytes (Figure S3B). Analysis of pathway genes across skin cell populations demonstrated increased expression of ADA, NT5E, P2RX4, and PDE4B, as well as CD38, a type I interferon-inducible gene, in SLE skin. Notably, expression of the adenosine receptor ADORA2A was decreased on endothelial cells, and PNP, encoding purine nucleoside phosphorylase, was decreased in several infiltrating SLE skin cell populations (Figure S3C). Broad disruption of NT5E and the adenosine pathway in SLE skin is supported by gene set enrichment analysis (GSEA) demonstrating enrichment of genes co-expressed with NT5E in pericytes, endothelial cells, and fibroblasts, suggesting shared regulation of NT5E-associated genes in those cells (Figure S4). Imbalance between signatures of adenosine production and inflammation in lupus-inflamed tissues Adenosine has well-documented roles in the extracellular transmission of immunosuppressive signals, with both ATP- and NAD-mediated adenosine production pathways contributing to broad inhibition of immune system activators and activation of immunosuppressive cellular responses (Figure 3A).13 To examine whether these pathways may be perturbed in lupus patients, GSVA was used to examine the enrichment of gene signatures for adenosine processing and production (ADO signature) (Figures 3B and 3C; Tables S1 and S3). This eight-gene signature was specifically selected to reflect the adenosine biosynthetic process rather than the induction of downstream signaling pathways. For comparison, we also examined the enrichment of signatures for a variety of inflammatory processes (inflammasome, immunoproteasome, and inflammatory cytokines) in multiple tissue and WB datasets (Figures 3B and 3C; Tables S1 and S3).26,28 Unexpectedly, given adenosine’s immunosuppressive function, we observed enrichment of the ADO signature but also concomitant upregulation of immune and inflammatory processes in SLE tissues (Figure 3B). In WB, no enrichment of the ADO signature was observed, although many inflammatory pathways were upregulated (Figure 3C). This suggests that adenosine signaling in lupus-involved tissues may be insufficient to counteract tissue-specific inflammatory responses. In further support of this conclusion, expression of the ADO signature in inflamed tissues was associated with the presence of many inflammatory cell signatures, including germinal center B cells, T cells, and monocyte/myeloid lineage cells (Figure 3D), whereas negative or insignificant correlations were observed with resident kidney and skin-derived cells in LN and C/DLE patients, respectively. These findings are consistent with the conclusion that upregulation of genes involved in adenosine production was likely associated with the presence of local inflammation in affected tissues. Adenosine signaling gene expression is elevated in lupus tissues and is associated with immunological activity (A) Pathway representation of adenosine (ADO) formation by the activity of the ecto-enzymes CD39 and CD73. (B and C) GSVA enrichment score for the ADO and additional immune processes were generated in SLE patients and healthy control samples from kidney (B, left; GSE32591) and skin (right; GSE81071) and whole blood (C; GSE88884 and GSE45291). Asterisks (∗) indicate a p < 0.05 using Welch’s t test comparing SLE to control. (D) Heatmap showing correlation coefficients between ADO and various cell-type-specific GSVA scores. N.d. (gray boxes), cell type not detected. Asterisks (∗) indicate p values with significantly non-zero slopes; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.00001. Downstream analysis of cAMP-mediated pathways reveals impaired adenosine activity in SLE Extracellular adenosine is known to modulate levels of cyclic AMP (cAMP), a potent transcriptional regulator of innate and adaptive immunity that can limit the progression of inflammation (Figure 4A).29 To determine whether adenosine signaling is impaired in SLE patients, we examined enrichment of cAMP activity as a proxy for adenosine function using an immunoregulatory gene signature to assess the expression of immunosuppressive genes upregulated by the cAMP response element (CREB; Figure 4A; Table S3). In both the kidney glomerulus and TI, the signature for cAMP-activated immunosuppressive genes was reduced in SLE compared to controls (Figure 4B). Notably, a similar expression pattern was observed in RA synovium (Figure S5A). However, the effect size (Hedge’s g) for the CREB-regulated immunosuppression signature was substantially lower in synovial tissues (g = 1.0) compared to either the LN glomerulus (g = 1.65) or TI (g = 3.4) (Figure S5B). Expression of the ADO gene signature in lupus kidneys negatively correlated with expression of genes involved in immunosuppression (Figure 4C), suggesting a linked discordance between upregulation of the ADO synthetic pathway and the downstream CREB-mediated regulation of inflammation. Skin differed from LN, however, in that CREB-regulated immunosuppressive genes were upregulated in both DLE and CLE (Figure 4D), and there was a weak, positive correlation with the ADO signature in DLE but not CLE (Figure 4E). These results imply that adenosine regulation of inflammation differs in LN versus DLE/CLE, with more profound abnormalities apparent in LN. Adenosine function is uncoupled from downstream immunomodulatory activity (A) Pathway representation showing the generation of intracellular cAMP from extracellular ADO and its downstream immunomodulatory effects. (B and D) GSVA enrichment scores representing the cAMP-mediated immunosuppressive gene signature were generated in kidney (B; GSE32591) and skin (D; GSE81071) datasets from SLE patients and healthy controls. Asterisks (∗) indicate a p < 0.05 using Welch’s t test comparing SLE to control where ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. (C and E) Linear regression analysis showing the relationship between GSVA scores for the ADO signature and the cAMP-mediated immunoregulatory signature in the kidney (C) and skin (E). R2 predictive values and p values are listed. N.s., not significant. Coding region variants in NT5E are likely to impede enzymatic function Genetic mutations in the coding sequence of NT5E that alter its enzymatic activity have been linked to several clinical phenotypes in non-lupus patients, including calcification of joints and arteries.17,18,19,20,21,22 Notably, an SLE genome-wide association study (GWAS) conducted using the Immunochip identified the lupus-associated single-nucleotide polymorphism (SNP) rs2229524 encoding an M379T transition in the putative heterodimer binding interface of NT5E.23 These mutations have been shown to alter the NT5E-mediated generation of adenosine.17 To determine the structural basis of this loss of enzymatic function, apo structures of the NT5E heterodimer in the soluble open form were simulated for 300 ns for both the wild-type (WT) and M379T mutant. In the WT molecular dynamic simulations, the binding pocket underwent opening and closing motions such that the catalytically required zinc ion became solvent accessible, whereas in the M379T mutant, no such opening was detected and the zinc ion remained inaccessible (Figures 5A and 5B). This molecular modeling indicates the structural basis for the loss of function of the protein encoded by the rs2229524 risk allele. Elevated immune activity in patients harboring the rs2229524 risk genotype (A) The M379T mutation is rendered in multicolored spheres, with a critical Arg395 residue in sticks and the required zinc atoms in silver spheres. (B) Molecular dynamics simulations of the active state showing local opening and closing of the catalytic site in the wild-type and the M379T NT5E mutant as measured by Arg395 nitrogen-zinc minimum distances over the simulations. (C) Three-dimensional rendering of the open, Zn-accessible conformation in the WT compared to the locally closed, Zn-inaccessible mutant enzyme in the simulations (Zn, arrowhead). Trp381, located on the same loop as residue 379, plays a critical role in closing access to the binding site (indicated by arrows). (D) Volcano plot visualization of DEGs between SLE patients homozygous for the reference allele (T/T) and patients with at least one risk allele (T/C + C/C). Red dots indicate significant DEGs (p value < 0.05); select genes of interest are labeled. (E) Pathway enrichment results using GO terms for Biological Processes and IPA canonical pathways of significant up- and downregulated genes. (F) GSVA enrichment scores for ADO signaling and cAMP-mediated immunosuppressive pathway in SLE patients with and without the risk allele. Asterisks (∗) indicate p < 0.05 using Welch’s t test comparing the T/T to T/C + C/C genotype. Because the conformational dynamics predicted impaired NT5E enzymatic activity in the M379T mutant, we next used the Genome Analysis Toolkit (GATK4)30 to enable the detection of rs2229524 in RNA-seq data collated from multiple lupus WB datasets to assess the functional impact of the putative loss-of-function NT5E variant in individual patients (Table S1). Of the 908 samples analyzed, GATK4 identified 8 samples homozygous for the rs2229524 risk allele (C/C), 71 samples that were heterozygous (T/C), and a further 190 patients homozygous for the reference allele (T/T). Previous evidence suggested that the T/C genotype of rs2229524 was sufficient to increase the risk of aortic valve calcification.18 Therefore, samples were separated into patients with no copies of the risk allele (T/T) and patients with at least one copy of the risk allele detected (T/C + C/C). Initially, we conducted differential gene expression analysis comparing these two SLE patient cohorts, detecting a total of 2,099 differentially expressed genes (DEGs; 1,693 upregulated, 406 downregulated; Figure 5C; Table S4). Analysis of upregulated DEGs revealed enrichment of numerous immune pathways in these patients, including Ingenuity Pathway Analysis (IPA) pathways for the role of PKR in IFN induction and antiviral response and neuroinflammation signaling, and Gene Ontology (GO) terms for neutrophil-mediated immunity (GO:0002446), innate immune response (GO:0045087), and pattern recognition receptor signaling (GO:0002221) (Figure 5D). Remarkably similar results were observed after orthogonal pathway analyses examining global CD73/NT5E-driven transcriptomic changes in WT and CD73 KO mice, including overall enrichment in multiple interferon and cytokine signaling pathways (Figure S6; Table S5). Next, we analyzed the enrichment of the previously utilized ADO and cREB-regulated gene signatures, finding that subjects with at least one copy of the rs2229524 risk allele had significantly decreased expression of CREB-upregulated immunosuppressive genes (Figure 5E). Together, these results suggest that even within lupus patient populations, diminished adenosine production resulting from a loss-of-function mutation in NT5E may further limit ADO-mediated immune/inflammatory regulation. Proinflammatory pathways are upregulated in SLE patients harboring the rs2229524 risk allele To identify molecules and pathways highly associated with adenosine deficiency, we utilized unbiased Protein-protein interaction (PPI)-based clustering of up- and downregulated DEGs found in patients with at least one copy of the rs2229524 risk allele versus patients homozygous for the reference allele (Figures 6A and 6B; Table S6). Upregulated pathways included numerous hallmarks of immune activation and increased inflammation, including canonical pathways for TLR signaling and pyroptosis (cluster 2), role of PKR in interferon induction and antiviral response (cluster 5), cytokine-mediated signaling (clusters 5, 25, and 46), RA signaling (clusters 6, 9, and 25), and hypoxia signaling in the cardiovascular system (cluster 30) (Figure 6A; Table S6). Key pathways associated with the risk genotype are enriched in immune and metabolic activity (A and B) Metastructures using up- and downregulated DEGs from SLE patients with (T/C + C/C) and without (T/T) the rs2229524 risk allele were generated based on PPI networks, clustered using MCODE and visualized in Cytoscape. Cluster size indicates the number of genes per cluster, edge weight indicates the number of inter-cluster connections, and color indicates the number of intra-cluster connections. Enrichment for biological function and immune cell type was determined by BIG-C, GO terms, IPA, and I-Scope. (C and D) GSVA enrichment scores were generated in kidney and skin datasets using genes in each metacluster. PPI cluster enrichment is summarized for each cluster-tissue pair and visualized in the heatmap. Increasing and decreasing enrichment are indicated by the red and blue gradient, respectively. Asterisks (∗) indicate p < 0.05 using Welch’s t test; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. (E) Schematic illustration showing how suboptimal levels of ADO linked to deleterious mutations in exon 6 affecting NT5E enzymatic activity may contribute to the elevated frequency of hypertension in SLE patients with the risk genotype (C/C or C/T). Further analysis of these clusters revealed upregulation of several key genes linked to immune activity in SLE, including TLR9, IL18RAP, IL16, and IRF1.31,32,33,34 Downregulated pathways also included several key immunoregulatory processes, including WNT signaling (cluster 7), Th1 and Th2 activation (cluster 10), and negative regulation of lymphocyte migration (cluster 22) (Figure 6B; Table S7). Interestingly, these clusters included CREB-regulated immunosuppressive genes, such as SMAD3, KLF4, and NR4A1, supporting earlier findings that perturbation of the adenosine pathway contributes to downstream immune activation at least in part via ineffective modulation of CREB-regulated genes.35,36,37 Parallel analyses of up- and downregulated genes derived from CD73 KO mice also reveal cluster-specific enrichment in a similar set of molecules and pathways highly associated with absent or reduced adenosine production, including Toll-like receptor signaling, as well as upregulation in VEGF signaling (Figure S7; Tables S8 and S9), consistent with the role of adenosine in stimulating angiogenesis via VEGF and eNOS.38 To explore altered adenosine signaling in SLE in greater detail, gene clusters composed of the up- and downregulated genes from patients harboring rs2229524 risk alleles were used as GSVA gene signatures to examine enrichment in lupus kidney and skin samples. Of note, signatures derived from upregulated clusters were broadly enriched across lupus tissues, particularly clusters 2, 5, and 9 dominated by immune and inflammatory processes, whereas clusters 19, 31, 32, 33, and 35, indicative of DNA repair, phospholipid, and metabolic function, were exclusively enriched in the kidney (Figure 6C). Similarly, gene clusters composed of downregulated genes were generally de-enriched in lupus tissues with the notable exception of cluster 10 (Th1 and Th2 activation), which was highly expressed in lupus kidney and skin (Figure 6D). Elevated frequency of pericarditis and hypertension in SLE patients with the rs2229524 risk genotype Finally, we assessed the impact of the rs2229524 risk variant on clinical manifestations of SLE. Genotyping for rs2229524 was conducted on an independent patient cohort (Hospital for Special Surgery SLE Flare study, Table 1) composed of 267 patients with established SLE and a variety of other clinical attributes, including assessment for cardiac-related manifestations.39,40 Genotyping analysis revealed 96 patients with at least 1 copy of the risk allele, along with 171 homozygous for the reference allele. Of note, the rs2229524 homozygous C/C genotype was detected among 22 SLE patients and only in combination with the proximal rs2229523 G/G, rs9450284 C/C, and WT rs139534171, suggesting the linked inheritance of the mutant allele. A significant deviation from Hardy-Weinberg equilibrium was obtained for rs2229524 (p < 0.01), implying an increased distribution of the C allele among SLE patients (Table 2). Table 1Major characteristics of SLE patients determined by rs2229524 status and adjusted to self-reported ancestryParameterAllT/TT/CC/CTotal2671717422Female, n%235 (88%)155 (91%)60 (81%)20 (91%)African, n%107 (40%)46 (27%)44 (59%)17 (77%)Asian, n%27 (10%)23 (13%)4 (5%)0 (0%)White, n%133 (50%)102 (60%)26 (35%)5 (23%)Age of diagnosis (IQR)24 (18–33.8)25 (18–33.8)20 (17.2–32)32 (21–37.2)ACR criteriaAllT/TT/CC/CORp valueSerositis128 (48%)71 (42%)40 (54%)17 (77%)1.76 (1.16–2.69)0.007Photosensitivity125 (47%)80 (47%)31 (42%)14 (64%)1.6 (1.04–2.46)0.028Neurological47 (18%)26 (16%)16 (22%)5 (23%)1.4 (0.85–2.31)0.195Immunological241 (91%)152 (90%)68 (92%)21 (95%)1.15 (0.53–2.49)0.720Hematological172 (65%)111 (66%)44 (59%)17 (77%)1.08 (0.71–1.65)0.726Malar erythema137 (52%)93 (55%)34 (46%)10 (45%)0.94 (0.62–1.41)0.759Discoid rash47 (18%)27 (16%)15 (20%)5 (23%)1.04 (0.63–1.73)0.870Arthritis239 (90%)152 (89%)67 (91%)20 (91%)1.04 (0.53–2.05)0.914Renal140 (53%)86 (51%)44 (59%)10 (45%)1.02 (0.68–1.52)0.941Oral ulcer101 (38%)64 (38%)29 (39%)8 (36%)1.01 (0.67–1.53)0.966Antinuclear Abs265 (100%)169 (100%)74 (100%)22 (100%)––ACR, median (IQR)6 (5–7)6 (5–7)6 (5–7)6.5 (5–8)––ManifestationsAllT/TT/CC/CORp valuePericarditis81 (31%)43 (26%)25 (34%)13 (59%)1.61 (1.1–2.5)0.028Thyroid46 (19%)37 (24%)8 (13%)1 (5%)0.48 (0.21–1.09)0.066Deforming or erosive arthritis26 (10%)10 (6%)11 (15%)5 (23%)1.76 (0.97–3.2)0.065Jaccoud’s arthropathy16 (6%)6 (4%)6 (8%)4 (18%)1.98 (0.96–4.1)0.068Hypertension82 (33%)44 (28%)30 (46%)8 (36%)1.98 (1.1–3.55)0.022Pleurisy95 (37%)55 (33%)29 (41%)11 (50%)1.51 (0.86–2.63)0.150Myocarditis17 (8%)8 (6%)6 (10%)3 (15%)2.23 (0.76–6.57)0.145Heart valve disease20 (9%)12 (9%)4 (7%)4 (20%)0.9 (0.32–2.49)0.834Cardiovascular event10 (5%)7 (5%)1 (2%)2 (10%)0.83 (0.19–3.69)0.807Aneurysm6 (3%)4 (3%)2 (3%)0 (0%)0.98 (0.15–6.28)0.985Atherosclerosis26 (11%)14 (9%)9 (15%)3 (14%)1.19 (0.49–2.85)0.701SLICC, median (IQR)1 (0–2)1 (0–2)1 (0–3)1 (0–2)––This table lists the major characteristics of SLE patients determined by rs2229524 status and adjusted to self-reported ancestry. Values are reported as cases and rates, or median and interquartile range employing the dominant model with correction for African ancestry. The demographic information includes details of participating patients, ACR 1997 SLE criteria, and other clinical manifestations. All patients in the study meet at least 4 ACR criteria, with a median ACR of 6. Bold face indicates p < 0.05. Major characteristics of SLE patients determined by rs2229524 status and adjusted to self-reported ancestry This table lists the major characteristics of SLE patients determined by rs2229524 status and adjusted to self-reported ancestry. Values are reported as cases and rates, or median and interquartile range employing the dominant model with correction for African ancestry. The demographic information includes details of participating patients, ACR 1997 SLE criteria, and other clinical manifestations. All patients in the study meet at least 4 ACR criteria, with a median ACR of 6. Bold face indicates p < 0.05. Table 2Frequencies of the major and minor alleles for exon 6 NT5E variants among studied SLE patientsSNPChr6 positionBase changeConsequenceMAFHWEχ2p valuers222952485489525T/CMet379Thr0.229.9190.0016rs222952385489515G/AThr376Ala0.212.1030.1469rs945028485489611G/Cintron0.480.1570.6919rs13953417185489445AAGGA/Aintron0.130.0330.8568The table depicts the minor allele frequencies (MAFs) of SNPs within NT5E exon 6 and its flanking region, as detected by Sanger sequencing in 267 SLE patients. The observed SNPs and one deletion are reported, along with their chromosomal position, associated base substitution, consequence of the substitution, MAFs among SLE patients, and deviation from Hardy-Weinberg Equilibrium (HWE). Frequencies of the major and minor alleles for exon 6 NT5E variants among studied SLE patients The table depicts the minor allele frequencies (MAFs) of SNPs within NT5E exon 6 and its flanking region, as detected by Sanger sequencing in 267 SLE patients. The observed SNPs and one deletion are reported, along with their chromosomal position, associated base substitution, consequence of the substitution, MAFs among SLE patients, and deviation from Hardy-Weinberg Equilibrium (HWE). Consistent with these findings, the risk C allele was unequally distributed among ancestries, with the proportion of lupus patients harboring a C allele highest among subjects of self-reported African ancestry (AA) (41.1% T/C and 15.9% C/C) (Table 3). Comparison with control subject data from the 1000 Genomes Project, however, revealed that AA populations in general carry the risk allele at a higher frequency (46.9% T/C and 17.2% C/C) compared to European ancestry (EA) (10.4% T/C and 0.35% C/C) or Asian subjects (13.3% T/C and 0.81% C/C) (Table 3). It is of note that while the risk genotype is relatively rare among EA patients, it was strongly associated with SLE diagnosis (OR: 2.55; p < 0.0001), whereas there was no association with SLE diagnosis in AA subjects (OR = 0.84, p = 0.25) or those of Asian ancestry (OR = 0.96, p = 0.9). Table 3Frequency of the rs2229524 minor allele among 1000 Genomes Project controls and SLE patientsGroupALLT/TT/CC/CORp valueEA, n (%)1000G850759 (89.3%)88 (10.4%)3 (0.35%)––SLE133102 (76.7%)26 (19.5%)5 (3.8%)2.550.0000036AA, n (%)1000G661237 (35.9%)310 (46.9%)114 (17.2%)––SLE10746 (43.0%)44 (41.1%)17 (15.9%)0.840.25Asian, n (%)1000G993853 (85.9%)132 (13.3%)8 (0.81%)–––SLE2723 (85.2%)4 (14.8%)0 (0%)0.960.93 Frequency of the rs2229524 minor allele among 1000 Genomes Project controls and SLE patients Next, association of the risk allele with demographics and clinical features of SLE patients was assessed. There was no difference detected in the age of diagnosis or in the overall Systemic Lupus International Collaborating Clinics (SLICC)/American College of Rheumatology (ACR) Damage Index score among the groups. However, serositis (OR: 1.76, p = 0.007; 54% with the risk allele [T/C], including 77% of the C/C genotype vs. 42% without), particularly pericarditis, as well as photosensitivity, was more frequently reported among SLE patients with the risk allele (Table 1). Furthermore, nearly half of the C/C patients (10; 45%) reported a familial history of cardiovascular disease, and a significant proportion were hypertensive (OR: 1.98, p = 0.022; 46% with the risk allele [T/C], including 36% of the C/C genotype vs. 28% without a risk allele; Table 1; Figure S8). Discussion The adenosine pathway represents an important mechanism regulating immune activity, with wide-ranging, predominately immunosuppressive actions on multiple components of the immune system.41 CD73, the protein product encoded by NT5E (5′-Nucleotidase Ecto 1), is an ectoenzyme that plays a critical role in the adenosine pathway, generating adenosine from extracellular AMP. In this analysis, both NT5E and ENTPD1 were upregulated in lupus-affected tissues, including the kidneys and skin, and this correlated with enrichment of gene signatures for adenosine processing and production. Notably, upregulation of NT5E was limited to resident cells of lupus tissues and not in immune/inflammatory cells. Despite its important immunomodulatory role, examination of cAMP-driven pathways reveals that adenosine function is uncoupled from downstream anti-inflammatory activity, predominantly in LN patients. Genetic variants in NT5E, namely rs2229524 that encode a loss-of-function non-synonymous mutation, have been associated with important clinical phenotypes in non-lupus patients, including vascular calcification and cardiovascular disease, suggesting that NT5E mutations may have a broader influence on inflammation.17,18,42 Here, SLE patients harboring NT5E risk alleles exhibited elevated immunological activity associated with reduced capacity for cAMP-mediated immunosuppression and a higher prevalence of pericarditis and hypertension. Together, these data suggest that while adenosine signaling is generally impaired among SLE patients, genetically driven immunological changes linked to altered adenosine signaling have a significant impact on immune system function and likely contribute to the profound heterogeneity evident among SLE patient populations. Results from the Immunochip GWAS indicated that rs2229524 may be associated with SLE, although with a modest level of significance.23 The rs2229524 variant resides in exon 6 and encodes a non-synonymous amino acid change in a region of NT5E that forms part of the C-terminal domain that includes the binding site for the adenosine component of the AMP substrate.43 Structural analysis of the potential impact of the variant suggests that the mutation can affect the availability of the active enzymatic site and thus the function of CD73. Indeed, analysis of upregulated DEGs from SLE patients with at least one copy of the risk allele reveals amplified immune activity exemplified by canonical pathways for pyroptosis, TLR signaling, and cytokine-mediated signaling. Consistent with this, patients harboring the risk allele demonstrated significantly decreased expression of cREB-regulated immunosuppressive genes. In addition, recent findings have shown that NT5E transcript expression is decreased in SLE Treg cells. Impaired CD73 enzyme activity and reduced production of adenosine have also been reported in SLE B cells along with diminished accumulation of AMP,15,44 although DNA sequencing of NT5E failed to detect any NT5E mutations, either deleterious or benign.15 However, it is important to note that the minor allele frequency (MAF) distribution for common exon 6 NT5E variants, including rs2229524, varies greatly among patients of different ancestral backgrounds, and this may account for the inability to detect rs22229524 and other NT5E mutations within the small 8-patient cohort sequenced.15 Moreover, the rs22229524 C allele skews heavily in favor of AA individuals (MAF 0.36) compared to those of Asian (AsA, MAF 0.07) or European (EA, MAF 0.14) descent. Consistent with this ancestral skewing, analysis of the 267 additional SLE patients studied here indicates that the C allele was unequally distributed among ancestries, with the proportion of patients homozygous for the C/C genotype higher (17 of 22) among people of self-reported AA. This represents a significant deviation from Hardy-Weinberg equilibrium and is supportive of an ancestral bias in distribution of the disease-associated variant. Despite the skewing toward subjects of AA, there was no association of the risk allele and lupus in persons of AA. Rather, an association with lupus was limited to person of EA. Notably, the data presented here describe the structural basis whereby the rs2229524 risk allele of CD73/NT5E alters the enzymatic function of the molecule and, although rare, appears to exacerbate disease even among SLE patients. In support of this, the frequency of global distribution of the rs2229524 C allele generally coincides with those ancestral groups most affected by severe SLE disease (AA and Afro-Caribbean ancestry).45 In terms of disease manifestations, patients expressing the NT5E disease variants did not have a higher incidence of overt renal disease. However, they did experience a higher rate of hypertension as well as serositis and pericarditis. Hypertension itself is common among SLE patients, with various cross-sectional studies reporting significantly elevated levels of hypertension (30%–56%) compared to healthy age-matched controls (2.5%–29%).46,47,48 In line with this, hypertension was detected in 28% of SLE patients carrying the T/T genotype but increased to 46% and 36% in patients expressing the C risk allele in genotypes T/C and C/C, respectively. In vascular tissues, extracellular adenosine induces vasodilation via activation of A2 receptors on vascular smooth muscle cells, leading to increased blood flow and oxygenation.49 Thus, it is tempting to speculate that suboptimal levels of adenosine production linked to deleterious mutations affecting NT5E enzymatic activity may contribute to the elevated frequency of hypertension in this patient subset. It is also of note that hypertension is highly prevalent in patients with autoimmune-related kidney disease, especially LN.50 Alterations to the kidney induced by inflammation and tissue damage may further compromise blood pressure control, thereby elevating cardiovascular and all-cause mortality in a cyclic manner (see Figure 6E for model). Here, robust expression of the adenosine gene signature in lupus-affected tissues was shown to be insufficient to counteract local inflammatory responses, suggesting dysregulation in the downstream signaling mechanisms that limit the progression of inflammation, particularly in the kidneys. Indeed, lack of adenosine regulation caused by CD73 deficiency in murine models leads to the spontaneous development of autoimmune renal injury as well as enhanced kidney damage in models of kidney ischemic reperfusion injury.51 Furthermore, genetic deletion of CD73 in kidney tubules contributes to enhanced renal damage in these models. Consistent with this, in patients with LN, cAMP-mediated immunosuppressive signaling in kidney tissue was significantly reduced even as ADO expression remained elevated. Remarkably similar findings were observed in the cohort of lupus patients harboring deleterious genetic mutations in NT5E demonstrating impaired regulation of inflammatory pathways normally down-modulated by extracellular adenosine. It is, therefore, not surprising that drugs, such as methotrexate, known to raise extracellular adenosine levels at sites of inflammation, are routinely used to treat rheumatic disease, including RA and SLE.52,53 Methotrexate-induced adenosine mitigates leukocyte adherence to endothelial surfaces and decreases the expression of inflammatory cytokines and complement pathways that exacerbate atherosclerotic lesions and fatty plaque build-up.54,55,56 The findings presented here suggest that enrichment of the minor NT5E allele in SLE patients could modify and further reduce the generation of adenosine by CD73, thereby abrogating immunosuppressive and cardioprotective effects on the immune system and vascular cells, contributing to chronic immune activation, inflammation, and sustained tissue damage. Limitations of the study Gene expression studies are inherently associative and cannot establish causality; while we examined enrichment of genes encoding the cAMP-generating system as a proxy for adenosine function, this approach does not determine whether reduced anti-inflammatory signaling reflects diminished adenosine production or altered receptor expression. Assessment of adenosine receptor expression in bulk RNA-seq data from SLE patients with the rs2229524 risk genotype was limited by low-quality or missing reads for ADORA1, ADORA2, and ADORA2A, whereas ADORA3 was detected but not significantly differentially expressed (LFC 0.43, pADJ > 0.1). However, single-cell transcriptomic data from SLE skin demonstrated decreased ADORA2A expression on endothelial cells, potentially in response to increased expression of NT5E, CD38, and other adenosine pathway genes in stromal cells. Impaired ADORA2A expression may reduce activation of downstream immunosuppressive pathways and contribute to vascular pathology, particularly given that adenosine concentration, receptor affinity and subtype expression, and cell type influence the pro- vs. anti-inflammatory response to adenosine.57 Although the loss of CD73 enzymatic activity has been reported to increased tissue-nonspecific alkaline phosphatase levels (TNAP) and calcium phosphate crystal formation,17 we did not detect changes in TNAP or other alkaline phosphatase by risk allele status, or in CD73 KO mice, suggesting that differences in adenosine receptors or alkaline phosphatases may be subtle among SLE patient populations (or SLE preclinical models) compared to a healthy control population. Altered adenosine signaling was evident in all SLE tissues and RA synovium, supporting a broader role in autoimmunity; however, regulation differed between LN and lupus-inflamed skin, with reduced immunosuppression and increased proinflammatory mediators in the kidneys but a more robust cAMP-driven anti-inflammatory response in skin, implying tissue-specific compensatory mechanisms. For example, melanocortin receptors expressed in skin (keratinocytes, fibroblasts, and melanocytes) and on immune cells (macrophages, DCs, and T and B cells) can generate a number of second messenger molecules including cAMP that mediate protective, anti-inflammatory effects in the skin independent of CD73/extracellular adenosine receptor signaling.58 Despite these complexities, SLE patients carrying the NT5E risk allele exhibited exaggerated inflammation and hypertension risk, although the low allele frequency and high linkage disequilibrium in this region necessitate validation in larger cohorts. Ultimately, investigation of the biologic basis of clinical heterogeneity observed among individuals with SLE provides an opportunity to define patient-relevant pathologic mechanisms and the most promising therapeutic targets.59 Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Peter E. Lipsky (peterlipsky@ampelbiosolutions.com). Materials availability This study did not generate any new unique reagents. Information about data collection and data analysis is available from the lead contact upon request. Data and code availability •All associated data with this paper are listed in the key resources table. No new data have been generated by this study. This paper analyzes existing publicly available data on the NCBI’s Gene Expression Omnibus (GEO) database; all dataset accession numbers are listed and referenced in the key resources table and Table S1. All associated data with this paper are listed in the key resources table. No new data have been generated by this study. This paper analyzes existing publicly available data on the NCBI’s Gene Expression Omnibus (GEO) database; all dataset accession numbers are listed and referenced in the key resources table and Table S1. •This paper does not report original code. This paper does not report original code. •Requests for further information may be directed to the lead contact. Requests for further information may be directed to the lead contact. Acknowledgments The work presented in this paper was funded by a grant from the RILITE Research Institute. The funder provided support in the form of salaries for authors (I.P., P.B., T.D., and K.A.O.). M.K.C. was supported by funding from an endowed chair, M.O. received partial support from a research grant from Spunk Fund, Inc., and P.K. was supported by a Wallenberg Academy Fellowship and computational resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS). We thank colleagues at AMPEL/RILITE Research Foundation for providing their scientific insights through the development of this project. Author contributions A.C.G., M.K.C., and P.E.L. designed the study. I.P., M.O., K.A.O., T.D., P.B., and P.K. analyzed the data and interpreted results. I.P., K.A.O., M.K.C., and P.E.L. wrote the manuscript. Declaration of interests M.K.C. has served as a consultant for Aboleris, AbelZeta, AMPEL Biosolutions, AstraZeneca, BMS, GSK, Lilly, Novartis, and Takeda. She holds public stock in Amgen and Johnson and Johnson. Declaration of generative AI and AI-assisted technologies in the writing process Generative AI was not used in the preparation of this manuscript. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERSoftware and algorithmsR BioconductorR Bioconductorhttp://www.bioconductor.orgPython 3Python Software Companyhttps://python.orgThe Genome Analysis Toolkit (GATK)Broad Institutehttps://gatk.broadinstitute.org/hc/en-usMSigDBv2025.1GO Biological Processeshttps://www.gsea-msigdb.org/gsea/msigdbSTRING Database v12STRINGhttps://string-db.orgCytoscapeCytoscape (v3.8.2)https://cytoscape.orgIngenuity Pathway Analysis platform (IPA)Qiagenhttps://www.quiagenbioinformatics.comEnrichRChen et al.60https://maayanlab.cloud/EnrichrGraphPad PrismGraphPad (v9.4.1)https://graphpad.comI-ScopeCatalina et al.25N/ABiologically Informed Gene Clustering (BIG-C)Labonte et al.61N/AGenemania (app)Cytoscape (v3.5.3)https://genemania.org/)MCODE (app)clusterMaker2 (v1.3.1)https://cytoscape.orgUniform Manifold Approximation and Projection algorithm (UMAP)Open SourceMcInnes et al., 201862The Human Protein AtlasUhlén et al.63https://www.proteinatlas.comOtherPublicly Available Dataset[S1]GSE88884Publicly Available Dataset[S2]GSE45291Publicly Available Dataset[S3]GSE29536Publicly Available Dataset[S4]GSE39088Publicly Available DatasetNo citation; data available at: https://www.ncbi.nlm.nih.gov/geo/GSE61635DatasetObtained via personal communication with authors, who may be contacted at Shashi.amur@fda.hhs.gov.FDABMC3Publicly Available Dataset[S5]GSE50772Publicly Available Dataset[S6]GSE68878DatasetData may be obtained via communication with lead contactEMTAB2713Publicly Available Dataset[S7]GSE32591Publicly Available Dataset[S8]GSE81071Publicly Available Dataset[S9]GSE52471Publicly Available Dataset[S10]GSE72535Publicly Available Dataset[S11]GSE120809Publicly Available Dataset[S12]GSE100093Publicly Available Dataset[S13]PRJNA293549Publicly Available Dataset[S14]PRJNA294187Publicly Available Dataset[S15]PRJNA318253Publicly Available Dataset[S16]PRJNA439269Publicly Available Dataset[S17]PRJNA505280Publicly Available Dataset[S18]PRJNA579374Publicly Available Dataset[S19]PRJNA627214Publicly Available Dataset[S20]PRJNA683207Publicly Available Dataset[S21]PRJNA690778Publicly Available DatasetNo citation; data available at https://www.ncbi.nlm.nih.gov/bioprojectPRJNA717024Publicly Available Dataset[S22]GSE55457Publicly Available Dataset[S23]GSE179633Publicly Available DatasetNo citation; data available at https://www.ncbi.nlm.nih.gov/bioprojectGSE150811 Experimental model and study participant details Human and mouse gene expression datasets Raw expression data from publicly available human and murine (lupus and RA datasets were derived from the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/). Raw sequence data from publicly available human datasets were derived from the Sequence Read Archive (SRA) repository. Study populations are predominately female; given the limited number of male subjects across datasets, sex-stratified analyses were not feasible, and all findings should be interpreted within the context of this known sex imbalance in autoimmune disease. All datasets are summarized in Table S1. Hospital for Special Surgery (HSS) flare study cohort This study cohort (IRB: 2014-221) was established and continues to be augmented and maintained at HSS for investigation of the relationship between immune system cells, factors and molecular pathways and alterations in clinical disease activity and manifestations over time. This was a non-interventional study in which biospecimens were collected without assignment to experimental an group. Blood samples suitable for isolation and sequencing of DNA are collected at the time of routine patient visits or disease flares. Extensive clinical data, including disease activity scores, are collected at the time of patient encounter and are supplemented by data from the electronic medical record (EMR). To establish a database of detailed clinical features for 267 patients on whom DNA samples were available, EMR data, including family history and past medical history, were reviewed along with the data available from the Flare cohort database. Sex-stratified analyses were not feasible, and all findings should be interpreted within the context of the known sex imbalance in SLE/autoimmune disease. All SLE cases in this study met the revised American College of Rheumatology (ACR) classification criteria for SLE.64 The protocol for study of the Flare cohort was approved by the HSS Institutional Review Board, and all study participants provided written consent prior to study enrollment. Method details NT5E exon 6 sequencing DNA was recovered from frozen stored patient samples (HSS Flare study subjects) and PCR was used to amplify the NT5E exon 6. In addition to rs2229524, exon 6 also includes a second common coding variant rs2229523 (c.1126 A>G, p.Thr376Ala) located 10 nucleotides proximal to rs2229524.65 Because of the proximity of the SNPs of interest and the possibility that they might interfere with one other in standard genotyping, we designed forward 5′-AGCAGCAGGTGAAAAGCAG and reverse 5′-TCAGGGCAAGAACCATAACAG primers to amplify the entire exon 6 in a PCR reaction. The amplified fragments were subjected to Sanger sequencing provided by Eurofins USA (eurofinsgenomics.com). In addition to the two coding SNPs, the PCR product includes common non-coding genetic variants rs139534171 and rs9450284. Genotypes were called for rs2229524, rs2229523, rs139534171 and rs9450284. DNA chromatograms were obtained from 267 SLE patients. Each chromatogram was analyzed using the sangerseqR package that determines the heterogeneous nucleotide base. In addition, all chromatograms were visually inspected to avoid signal abnormalities. All genetic variants described above were detected among SLE patients. Data were further analyzed in comparison to ancestry-based data from the 1000 Genomes project using the R package snpReady. Quality control and data normalization Microarrary data were processed as previously described using free, open source programs.61 Unnormalized arrays were inspected for visual artifacts or poor RNA hybridization using QC plots. Datasets were annotated using their native chip definition files (CDFs). Probes missing gene annotation data were discarded. Raw data (CEL files) from the Affymetrix platform were background corrected and normalized using GC robust multiarray average (GCRMA) or robust multichip average (RMA) algorithms, whereas raw data files from the Illumina chip were read and normalized using neqc (limma R package). RNA-seq data was processed from FASTQ files as previously described.66 Principal component analysis was used to inspect the raw data files from each dataset for outliers. All log2 transformed data was formatted into R expression set objects (E-sets). scRNA-seq processing, clustering and pseudobulk analysis scRNA-seq analysis of the GSE179633 dataset was carried out in Python 3, primarily using the Scanpy single-cell analysis framework.67 The available matrix files were first joined between different experiments based on common genes, followed by filtering cells and genes based on minimum and maximum thresholds to remove low-quality cells and genes. Data normalization was conducted to scale counts per cell to a target value, followed by logarithmic transformation. Highly variable genes were identified for downstream analysis, and the data were scaled to unit variance. Between-experiment differences were eliminated using the Harmony algorithm.68 Dimensionality reduction was achieved using principal component analysis (PCA), and a neighborhood graph was computed for clustering and visualization. Clustering was carried out using the Leiden algorithm at multiple resolutions, and clusters were visualized using Uniform Manifold Approximation and Projection algorithm (UMAP).62 Cell types were determined based on a list of cell specific gene markers, including genes specific for myeloid and lymphocyte populations, human keratinocytes and fibroblasts, endothelial and pericytes. Differential expression analysis identified marker genes for each cell subset using the Wilcoxon rank-sum test. Pseudobulk analysis was carried out using the decoupler package69 to aggregate single-cell data into pseudobulk profiles based on sample and cell type annotations. Generated pseudobulk samples including more than 10 cells or more than 10,000 total counts were further processed in the R package with the help of the edgeR pipeline as regular bulk RNA-seq samples. TMM normalization was applied and the glmQLT negative binomial model was used for differential expression analysis. Differential expression testing was conducted for each cell type. The adenosine-related pathway gene set representing GO Biological Process terms related to purine metabolism, purine signaling, and purine transport was collected from MSigDB70 v2025.1 (https://www.gsea-msigdb.org/gsea/msigdb) c5_goterm. The list was manually curated to retain known enzymes and receptors, including ectonucleotidases/ectophosphatase (NT5E, ENTPD1, ENPP1-3, CD38, BST1, ALPL) responsible for extracellular ATP/ADP/NAD hydrolysis and adenosine production; genes related to cytosolic nucleoside metabolism (PNP, GDA, XDH, ACP3, NT5C, NT5M), adenosine receptors (ADORA1, ADORA2A/B, ADORA3); adenosine clearance and transport genes (ADA, ADA2, ADK, PNP, SLC28A1–2, SLC29A1–2, XDH); purinergic P2 receptors (P2RX1-7, P2RY1-14); and cAMP effectors (ADCY1–3, PRKACA/B, PRKAR1A/2A, CREB1, PDE4A/B). The resultant set was designed to capture genes involved in extracellular nucleotide/nucleoside metabolism, receptor-mediated signaling, and downstream cAMP responses. In addition, NT5E protein physical interaction genes were obtained using STRING database version 1271 (https://string-db.org), NT5E interaction network genes were obtained using GENEMANIA72 release 2021-04-27 (https://genemania.org/) and the co-expressed genes with NT5E were identified with ARCHS4 RNA-seq gene-gene dataset73 (https://maayanlab.cloud/Enrichr/). Gene set enrichment analysis was performed for all of the defined cell-subpopulations where significant numbers of cells were obtained using the r fgsea package74 and gene sets generated above. Word clouds were generated in R using the ggwordcloud package (v0.6.2). Input consisted of gene symbols of the generated adenosine-related pathway gene set differentially expressed between SLE and HC samples across cell subpopulations. The frequency of each gene symbol and the number of subpopulations in which it was significant were calculated and used to determine both the font size and color intensity in the word cloud. Genes appearing more frequently across subpopulations were shown in larger size and deeper color, reflecting their broader significance in SLE-associated transcriptional changes. The bubble plot was generated in R using the ggplot2 package. Point size represents the FDR-corrected significance level, while color indicates the fold-change difference between SLE and healthy donors. The y axis shows adenosine-related pathway gene sets, and the x axis represents the corresponding target cell populations. Identification of NT5E alleles from RNAseq data The Genome Analysis Toolkit (GATK) was utilized to genotype patients across datasets (https://gatk.broadinstitute.org/hc/en-us).30 Best practices for RNA-Seq short variant discovery were followed and will henceforth be described. The STAR aligner was used to map reads to the hg38 human reference genome; the gencode.v32.annotation.gtf file was employed as a reference of genomic features.75 Picard was used to convert fastq files to unmapped bam format, add read group information, merge bam alignments, and mark duplicates (http://broadinstitute.github.io/picard). The SplitNCigarReads GATK function was then used, with the output fed to the GATK BaseRecalibrator function and then to the GATK HaplotypeCaller function. GATK’s CombineGVCFs and then GenotypeGVCFs were used to generate a combined Variant Call Format (VCF) fileacross all samples. This VCF file was then used to filter samples by the presence of specific SNPs, using methods described previously.76 Differential gene expression analysis Differential gene expression analyses were carried out using the DESeq2 pipeline. P- values were adjusted for multiple hypothesis testing using the Benjamini-Hochberg correction, which resulted in a False Discovery Rate (FDR) for each gene. An FDR<0.2 was employed to avoid falsely excluding genes of interest.77 For differential gene expression analyses conducted in mice, gene expression in CD73-knockout mice was compared to gene expression in WT animals (see Table S1 for full summaries of mouse datasets used). For differential gene expression analyses in SLE patients, patients were sorted into categories based on the presence of at least one copy of the rs2229524 risk allele using the GATK pipeline described above, with patients either labeled as rs2229524-or rs2229524+ (corresponding with the T/T genotype vs. either the T/C or C/C genotype, respectively). The DESeq2 pipeline was then used to calculate differentially-expressed genes linked to the presence of at least one copy of rs2229524. Gene set variation analysis (GSVA) The GSVA (V1.42.0) software package is an open-source package available from R/Bioconductor, and was used as a non-parametric, unsupervised method for estimating the variation of pre-defined gene sets in patient and control samples of microarray and RNA-seq expression datasets (http://www.bioconductor.org/packages/release/bioc/html/GSVA.html).26 The inputs for the GSVA algorithm were a gene expression matrix of log2 expression values for pre-defined gene sets (Table S3). An additional gene expression matrix was formulated using gene expression values for rs2229524-and rs2229524+ patients, keeping the same patient classifications used for differential expression analyses. All genes within a gene set were evaluated if the interquartile range (IQR) of their expression across the samples was greater than 0. Enrichment scores (GSVA scores) were calculated non-parametrically using a Kolmogorov Smirnoff (KS)-like random walk statistic and a negative value for a particular sample and gene set, indicating that the gene set has a lower expression than the same gene set with a positive value. The enrichment scores (ES) were the largest positive and negative random walk deviations from zero, respectively, for a particular sample and gene set. The positive and negative ES for a particular gene set depend on the expression levels of the genes that form the pre-defined gene set. GSVA calculates enrichment scores using the log2 expression values for a group of genes in each patient cohort and normalizes these scores between −1 (no enrichment) and +1 (enriched). Welch’s t test was used to calculate the significance of the gene sets between the cohorts. Significant enrichment of gene sets was determined by p value <0.05. GSVA gene sets All input gene sets used for GSVA analyses can be found in Table S3. Gene sets for adenosine production and processing, and cAMP-activated immunosuppressive genes were created from literature mining. Genes in these gene sets were further validated for detectability in human blood samples using The Human Protein Atlas (https://www.proteinatlas.org/).63 Cellular and inflammatory modules were previously derived.25 Additional hematopoietic and non-hematopoietic cellular gene sets were curated from various sources, as described previously.78 Linear regression analysis Simple linear regression between adenosine gene signature GSVA scores and various cellular and biological gene set GSVA scores were performed in GraphPad Prism Version 9.4.1. For each regression analysis, the R value and p-value of the slope are displayed. For regression analyses using the cAMP-regulated gene set GSVA scores, the Goodness of Fit is displayed as the R squared value and the p-value testing the significant of the slope is displayed. Network analysis and visualization Visualization of protein–protein interaction and relationships between genes within datasets was done using Cytoscape (version 3.8.2) software. Briefly, STRING (version 11.5) generated networks were imported into Cytoscape (version 3.8.2) and partitioned with MCODE via the clusterMaker2 (version 1.3.1) plugin. Functional gene set analysis DEGs and gene metaclusters were functionally annotated for pathway enrichment using the Ingenuity Pathway Analysis platform (IPA; https://www.qiagenbioinformatics.com) and the EnrichR web server (https://maayanlab.cloud/Enrichr/).60 Functional enrichment of gene clusters were examined using Biologically Informed Gene Clustering (BIG-C), a custom functional clustering tool developed to annotate the biological meaning of large lists of genes as previously described.61 Hematopoietic cellular enrichment was performed using I-Scope, a tool developed to identify immune cell specific genes in large gene expression analyses.25 Molecular dynamic simulations Molecular dynamics simulations of CD73 were performed as follows. PDB structure 4H2F was used as a starting model,79 with the catalytic zinc atoms placed in accordance with the structure and catalytic model discussed by Scrutton and co-workers.80 Missing residues were placed and the M379T mutant modeled using Modeller,81 yielding two starting structures, one of wild-type CD73 and one of the M379T variant. Simulations were performed using Gromacs82 and the AMBER99SB- ILDN force field,83,84 as follows. Structures were placed in a box of TIP3P water and 150 mM NaCl and energy minimized. The resulting structures were equilibrated and then started for production molecular dynamics simulations of 100 or 200 ns using 2-fs timesteps with hydrogen bonds constrained using LINCS, temperature set at 310 K using a velocity-rescaling thermostat,85 pressure maintained at 1 bar, and long-distance electrostatics treated using Particle Mesh Ewald.86 A total of 300 ns of production simulation (one 200-ns run and one 100-ns run were completed for each of the wild-type and mutant enzymes, each in dimeric form) and used for analysis. Quantification and statistical analysis Statistical analysis details, including comparator groups, type of statistical test, and p-value cut-offs can be found in the results section and within the Figure Legends. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.116017. Supplemental information Document S1. Figures S1–S8 Table S1. Human and murine lupus datasets Table S2. Marker genes used to define cell type populations Table S3. GSVA gene sets Table S4. DEGs in patients with rs2229524 T/T vs. T/C + C/C genotype Table S5. DEGs in WT vs. CD73 KO PMNs Table S6. rs2229524 risk genotype PPI-based gene clusters Table S7. rs2229524 risk genotype pathway analysis summary Table S8. CD73 KO PPI-based gene clusters Table S9. CD73 KO upregulated and downregulated cluster annotation summary References Associated Data Supplementary Materials Document S1. Figures S1–S8 Table S1. Human and murine lupus datasets Table S2. Marker genes used to define cell type populations Table S3. GSVA gene sets Table S4. DEGs in patients with rs2229524 T/T vs. T/C + C/C genotype Table S5. DEGs in WT vs. CD73 KO PMNs Table S6. rs2229524 risk genotype PPI-based gene clusters Table S7. rs2229524 risk genotype pathway analysis summary Table S8. CD73 KO PPI-based gene clusters Table S9. CD73 KO upregulated and downregulated cluster annotation summary Data Availability Statement •All associated data with this paper are listed in the key resources table. No new data have been generated by this study. This paper analyzes existing publicly available data on the NCBI’s Gene Expression Omnibus (GEO) database; all dataset accession numbers are listed and referenced in the key resources table and Table S1. All associated data with this paper are listed in the key resources table. No new data have been generated by this study. This paper analyzes existing publicly available data on the NCBI’s Gene Expression Omnibus (GEO) database; all dataset accession numbers are listed and referenced in the key resources table and Table S1. •This paper does not report original code. This paper does not report original code. •Requests for further information may be directed to the lead contact. Requests for further information may be directed to the lead contact.