In this retrospective observational study, we recruited non-consecutive patients referred to the internal medicine department of the Limoges University Hospital Dupuytren, France, between January 1, 2014, and December 31, 2018. The participants underwent temporal artery biopsy (TAB) for suspected GCA, which was diagnosed based on the 2022 ACR/EULAR criteria (17) and histologically confirmed based on the presence of panarteritis with an inflammatory infiltrate (e.g., macrophages and lymphocytes) and fragmentation of the internal elastic lamella, corresponding to the most common histological pattern described by Hernández-Rodríguez et al. (18, 19). This study included patients who were treated with glucocorticoids for 1–10 days before TAB as this treatment does not affect the histopathological findings of TAB (20, 21). Participants with a negative TAB and a final diagnosis other than GCA were considered controls. We recorded the age, sex, symptoms, treatments, and outcomes of each participant at the time of diagnosis.

Unilateral TABs were performed under local anesthesia by an ophthalmologic surgeon using standard techniques (22). A 1–4-cm-long specimen of the temporal artery was obtained for histological analysis. The temporal artery specimens were analyzed using standard protocols. The specimens were subjected to gross evaluation and were cross sectioned serially. Tissues were fixed in 10% neutral buffered formalin for 24 h, followed by transfer to 70% ethanol before paraffin embedding. For each patient and control, the most representative segments were grouped into formalin-fixed paraffin-embedded (FFPE) temporal artery blocks using the tissue microarray method. Tissue microarrays were constructed from 2-mm-diameter cores punched from each FFPE tissue block.

The NanoString GeoMx DSP platform quantifies RNA abundance by counting unique indexing oligonucleotides assigned to each target of interest. Those oligonucleotides are covalently attached to mRNA hybridization probes with an ultraviolet-photocleavable linker (16). First, FFPE temporal artery blocks from GCA and control patients were sectioned to a thickness of 5 μm. The sections were mounted on an adhesive or positively charged slides (cat# S21.2113.A; Bond Plus slides; Leica Biosystems, Richmond, VA, USA) with a maximum of 24 individual tissues per slide at the San Diego Pathology Department (San Diego, CA, USA). Some slides were stored at 4°C, while others were deparaffinized and used for hematoxylin & eosin (H&E) staining (San Diego Pathology Department) for subsequent microscopy imaging (Keyence BZ-X800). A cardiovascular pathologist (M.C.B.) and two GCA specialists (S.P. and C.M.W.) reviewed all H&E-stained tissues. Overall, one GCA patient was classified as healed and excluded from this study because the artery exhibited no inflammatory infiltrate, and one control sample was also excluded because it showed tertiary lymphoid structure in the adventitia.

The freshly cut sections were stained with fluorescently labeled imaging reagents and mRNA markers as described previously (16). Briefly, sections were labeled with the cell markers, α-smooth muscle actin, CD3, and CD68, and the nuclear stain SYTO 13. The NanoString commercial Human Whole Transcriptome Atlas (GeoMx WTA) panel (18,000 protein-coding genes) was selected. Four slides (two with GCA patients and two with controls) were scanned on a GeoMx DSP instrument (NanoString Technologies Inc., Seattle, WA, USA) and individual regions of interest (ROIs) with a maximum diameter of 400 μm were created within each of the four arterial layers in GCA and controls.

In this study, we considered a sample as a control where no inflammatory infiltrate was observed, and the 4 artery layers (intima, media, adventitia, and PVAT) were distinguished on the basis of H&E staining as well as internal elastic lamina between intima and media; concentric smooth muscle cells in the media unlike the adventitia, and, finally, the presence of adipocytes in the PVAT unlike the adventitia. A representative portion of each layer was selected by two GCA specialists (S.P. and R.G.) and confirmed by a pathologist (M.C.B.) ( Supplemental Figure 1 ). Once each ROI was compartmentalized, ultraviolet-cleaved indexing oligonucleotides were collected into a 96-well plate. Libraries were prepared according to the manufacturer’s instructions (protocol 01/2019). The library quality and quantity were assessed using the Bioanalyzer DNA High Sensitivity Analysis. Samples were sequenced on the NovaSeq 6000 platform (Illumina, San Diego, CA, USA) and reads were digitally quantified and normalized using GeoMx DSP Data Analysis software (Merritt).

Data obtained from the NanoString GeoMx DSP platform were analyzed using a dedicated software (GeoMx Analysis Suite 2.3). The quality control analysis showed that all ROIs had raw read counts above 1 million and good alignment rate and sequencing saturation. The local outliers were negative control probes and were removed from the ROIs. ROIs with deduplicated probe counts Q3 (upper quartile) < 2 were excluded. Furthermore, ROIs were removed if < 1% of their genes had read counts higher than the limit of quantification (LOQ). LOQ was defined as geomean (NegProbe) × geoSD(NegProbe)² for each ROI. Target filtering was applied to retain gene targets with read counts above LOQ in at least 10% of ROIs. Q3 normalization was applied on the filtered ROIs and gene targets.

The raw data were deposited at the NCBI Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE237441) under the accession number GSE237441.

Differential gene expression across groups was analyzed using linear mixed effect models; differentially expressed genes (DEGs) were defined as fold-change > 2 or < −2, and Benjamini-Hochberg-adjusted p < 0.01. Analyses were conducted using R software (version 3.2.2; R Foundation for Statistical Computing, Vienna, Austria).

The upregulated and downregulated DEGs were analyzed using DAVID Bioinformatics Resources software (version 2021; https://david.ncifcrf.gov/) for functional annotations and gene set enrichment analysis. DAVID was used to obtain statistically enriched Gene Ontology (GO) terms for biological processes (GOTERM_BP_FAT). GO terms with EASE score < 0.05 were considered statistically enriched. Based on the known mechanisms in GCA, pathways of interest, signaling cascades, and molecules were selected from among the dysregulated GO terms. These selections are presented as dot plots using ggplot2 package of R software (version 3.2.2).

Dysregulated genes per layer with a fold-change > 2 or < −2 were organized according to the STRING network (https://string-db.org/). The one or two main networks were represented for each layer. Genes disconnected from the main network were removed. An interaction score with the highest confidence (0.900) was used. The interaction sources excluded text mining. Using networks generated by the STRING network, a pharmacogenomic network was constructed by targeting existing druggable genes (https://www.dgidb.org/). Drugs with Food and Drug Administration (FDA) approval were selected.

From block, 4-μm-thick slices were cut and dewaxed by successive immersion in pure xylene (Sigma). Rehydration and unmasking of antigens were carried out by sequentially immersion into 100%, 90%, 80% and 70% ethanol baths (Sigma), phosphate-buffered saline (PBS; Gibco), and citrate buffer (pH 7). Saturation of nonspecific antigenic sites was achieved by incubation in PBS containing 3% bovine serum albumin (BSA). The primary antibodies were diluted in 3% BSA and incubated overnight at 4°C. The antibodies used were as follows: anti-CD74 (1:50, mouse, Abcam), anti-CD90 (1:50, rabbit, Abcam), anti-CD3 (dilution recommended by the supplier, rabbit, Ventana). The sections were subsequently incubated with a secondary antibody coupled to Alexa Fluor 594 or 488 (Molecular Probes) for 60 min. After washing with PBS, nuclear counterstaining was carried out with 4′,6-diamidino- 2-phenylindole (Sigma) and the sections were mounted in an aqueous medium (Vectashield, Sigma). Images were acquired with Leica microscope (DMi8, Leica).

The funders of the study had no role in study design, data collection, analysis, or interpretation, or in writing the manuscript.

The study protocol (87RI21_0065) was approved by the Ethics Committee of the University Hospital of Limoges, France (23). Temporal artery tissues were obtained from the biological collection DC-2010-1079. All participants provided informed consent.

The characteristics of GCA patients (n = 9) and controls (n = 7) are presented in the Supplemental Table 1 . Among the controls, the final diagnoses were non-arteritic ischemic visual impairment (n = 4), peripheral seronegative polyarthritis (no argument for polymyalgia rheumatica) (n = 2), and infection-related encephalopathy (n = 1). Three controls had an inflammatory syndrome related to an acute infection. Among the entire cohort, one patient had diabetes (control), one had a previous cardiovascular complication (control), and two patients were active smokers (one control and one GCA). Corticoids were used before TAB for 8/9 GCA patients (mean 4.3 days before TAB, mean dose of 0.85 mg/kg/d) and 3/7 controls (mean 18.3 days before TAB, mean dose of 0.7 mg/kg/day).

All GCA patients had panarteritis involving the three arterial layers with destruction of the internal lamella on histological examination. The histological sections of the artery are presented in Supplemental Figure 1 . Then temporal arteries from the 16 participants were subjected to transcriptional analysis for each arterial layer (intima, media, and adventitia) and the PVAT following technical pipeline indicated in Figure 1 . Because the control arteries did not present CD3 and CD68 staining, layers were selected under the control of a senior pathologist depending on their macro characteristics. Four ROIs involving the media, adventitia, and PVAT were excluded from analysis due to poor quality, respectively. Altogether, we analyzed 59 ROIs, checked for quality and correspondence to each layer, from 16 participants to conduct our spatial transcriptomic study.

As attempted, correlation analysis demonstrated that disease status (GCA and control) only differentiated samples ( Figure 2A ). Interestingly, Pearson correlation coefficient showed that GCA transcriptome remains more homogeneous than in controls as illustrated by increasing intensity of blue reflecting correlation index tending toward 1.

A total of 12,076 genes were detected and 337 different genes (2.8%) were dysregulated between GCA and control participants (fold-change < −2 or > 2, p-adjusted < 0.01). Venn diagram presents unique and common DEGs for the intima, media, and adventitia ( Figure 2B ). The number of DEGs increased from external layer, adventitia with 40 genes, to media and intima with respectively 227 and 282 genes. In total, 32 common DEGs were significantly identified among the three layers (intima, media and adventitia), corresponding to the major histocompatibility complex (CD74, B2M, HLA-DRB1, HLA-A, HLA-B, HLA-C, and HLA-DRA), macrophages (CD68), immunoglobulins (IGKC and IGHG1-4), apolipoprotein (APOC1), cathepsins (CTSB and CTSZ), and tissue remodeling (ADAM15, COL1A2, COL3A1, TMSB10, and ARHGDIB). The other common genes were NPIPB6, PABPC1, and those associated with ribosomes (RPL23, RPL28, RPL31, RPL37, RPS12, RPS28, RPS29, RPS3, and RPS4X). Patterns for the four layers in each participant are shown in Figure 2C . Principal component analysis (PCA) highlighted that only intima present a distribution profile more homogeneous and reproducible than other GCA layers and control samples.

Altogether, these results highlight that the intima is the layer where the transcriptome is the most dysregulated depicting a gradient of decreased DEG from the internal to the periphery of the artery.

Overall, most DEGs were upregulated in ROIs from GCA compared to controls. The top 20 upregulated and top 5 downregulated genes per arterial layer in GCA patients compared to the controls are shown in Figure 3A . The highest numbers of DEGs were observed in the media and intima.

Figure 3B presents the top 30 dysregulated genes across arterial layers between both studied groups. In the media and intima, CD74 was the most upregulated gene between GCA and control specimens. Healthy temporal arteries do not exhibit CD74 staining on immunofluorescence analysis ( Supplemental Figure 2 ). In contrary, GCA arteries has CD74+ cells, particularly within the intima, close to the media border, as well as in the media and the adventitia. Also, cells expressing CD74 are also CD90+; however, they do not express CD3, marker for lymphocytes. Finally, migration inhibitory factor (MIF), a receptor for CD74, was significantly overexpressed in the intima of GCA compared to controls (p-adj = 0.0031).

Altogether, these results unveiled that most dysregulated genes in the three arterial layers belong to the antigen presentation gene family and vascular remodeling functions.

We classified the top 30 DEGs by arterial layers according to their main functions ( Figure 4A ). Most top DEGs per layer were associated with immune or remodeling functions. Immune function-related genes were associated with macrophages (CD68 and cathepsins), major histocompatibility complex, complement system, and immunoglobulins. Remodeling-related genes were mainly associated with metalloproteases, reactive oxygen species, cytoskeleton organization, and collagen.

Using the DAVID tool, remodeling-related pathways of interest with gene dysregulation were selected ( Figure 4B ). PVAT was excluded due to the low number of dysregulated genes between GCA and control specimens. Remodeling pathways associated with GCA primarily involve the intima and media and include the reactive oxygen species and collagen metabolic process, fibroblast proliferation, muscle cell differentiation, ossification, and angiogenesis. The dysregulated genes for each signaling pathway are detailed in Supplemental Figure 3 .

Altogether these results suggest that activated immune cells are actively and mainly present in the intima and the media layers. Likewise, vascular remodeling process following a gradient from the outside to inside part of the vessel.

The DEGs among the four layers were compared ( Figure 5A ). The meta-analysis identified 412 upregulated genes and 101 downregulated genes. Genes whose expression profiles were similar to CD74, the top DEG, are shown in Figure 5B . These genes were mainly involved in antigen presentation, including the major histocompatibility complex (B2M, HLA-B, HLA-DRA, and HLA-DRB1), cathepsin (CTSB), ferritin (FTH1), immunoglobulin (IGHG2), and lysozyme (LYZ).

As macrophages are the key immune cells in GCA, we presented the genes with a similar expression profile within the four layers like CD68, a marker for macrophages ( Figure 5C ). The DEGs included those related to cytokines (TNF), chemokines (CCL5, CXCL9, and CXCR4), YKL40 (CHI3L1), metalloproteases (MMP2 and MMP9) secreted by macrophages, membrane receptors (CD63 and CD206), and iron metabolism-related molecules (FTH1 and SOD2). Similarly, we evaluated vascular remodeling during GCA based on the two mains associated DEGs: MMP2 ( Figure 5D ) and collagen COL3A1 ( Figure 5E ). We identified the top genes with the same expression profile across the layers.

Transcriptomic profiling is an efficient and essential tool for drug target discovery. Several studies have combined gene expression profiling with systematic approaches for drug re-repurposing or biomarker discovery (24, 25). Among strategies for novel drug target identification, the network-based framework uses curated biological network topology to investigate the association across genes, diseases, and drugs (26). We constructed a pharmacogenomic network by integrating DEGs from each artery layer, a high-confidence protein-protein interaction network, and drug-gene interactions. Then, the top dysregulated genes are presented as a STRING analysis to highlight their relationships. The significantly connected genes are shown in Figure 6 . Networks show the strong relationships between several genes. Two mains separate networks exist for each layer corresponding to global immune function (yellow) and vascular remodeling (blue). Key genes are present at nodes of those interactions and connected to other genes.

Approved therapies targeting these genes are presented, of which have not been evaluated in relation to GCA previously. The spatial approach highlights therapies targeting top DEGs common to the intima and the media (red). Therapies targeting a gene only present in one network layer are also represented (black).