Kidney Lymphatics Disrupted in Chronic Rejection

Characterization of lymphatic architecture and spatial relationships in the healthy human kidney. To characterize lymphatic architecture in the healthy human kidney, we analyzed tissue samples from 4 deceased organ donors with minimal chronic damage (<10% interstitial fibrosis or tubular atrophy, Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/JCI168962DS1) (36). Intact tissue samples (<3 mm3) were immunolabeled using a D2-40 monoclonal antibody targeting podoplanin (PDPN) (37) and imaged using confocal or light-sheet fluorescence microscopy (LSFM) (16, 38). PDPN+ vessel networks were visualized in the human kidney cortex (Figure 1A and Supplemental Video 1), and antibody-omitted controls displayed minimal autofluorescence or nonspecific binding (Supplemental Figure 1A). Mapping vessel radius revealed a hierarchical network, with small lymphatics (radius ~3.5 μm) initiating in the cortex and converging into larger vessels (radius ~50 μm) at the corticomedullary junction (Figure 1B). The cells lining these vessels expressed prospero homeobox protein 1 (PROX1) (Figure 1C), a canonical LEC transcription factor (39, 40) but showed sparse expression of lymphatic vessel endothelial hyaluronan receptor 1 (LYVE1) (Figure 1D), a glycoprotein important for leukocyte entry into lymphatics (41).

To elucidate the microanatomical localization of lymphatics in the human kidney, autofluorescent tissue signals were captured alongside PDPN labeling. Large caliber lymphatics were observed adjacent to arteries at the corticomedullary junction branching into smaller cortical vessels (Supplemental Video 2). Colabeling with Lotus tetragonolobus lectin (LTL, proximal tubules) and uromodulin (UMOD, loop of Henle) revealed PDPN+ blind-ended lymphatics in the renal cortex (Figure 1E and Supplemental Video 3) and their absence from the medulla (Figure 1F). Despite previous reports of subcapsular lymphatics (42, 43), these were not detected in 3D reconstructions (Figure 1G) or optical z-sections (Supplemental Figure 1B), even with the kidney capsule intact. In the cortex, lymphatics followed UAE-I+ arterioles toward glomeruli (Figure 1H), extending terminal branches near megalin (LRP2+) proximal tubules (Figure 1I) and calbindin 1 (CALB1+) distal nephron epithelium (Figure 1J and Supplemental Video 4). Lymphatics converged toward the kidney hilum, adjacent to medullary Dolichos biflorus agglutinin (DBA+) collecting ducts (Figure 1K) and UMOD+ medullary tubules (Supplemental Figure 1C). A model summarizing these findings is presented in Figure 1L.

Determination of the molecular identity of healthy human kidney lymphatics. Because of the rarity of lymphatics in the human kidney relative to other cell types, isolating sufficient LECs for molecular profiling is challenging. To surmount this, we leveraged published scRNA-Seq data from 59 kidneys, supplemented with 5 new samples (Supplemental Figure 2A). This integrated dataset comprised 217,411 human kidney cells, with 151,038 control samples (living donor biopsies or unaffected regions of tumour nephrectomies) and 66,373 cells from diseased samples (chronic kidney disease, CKD) and kidney allograft injury; covering both chronic rejection and non-alloimmune etiologies) (Supplemental Figure 2B). All cell types were manually annotated (Supplemental Figure 2C), revealing 38 clusters (Figure 2A), including a transcriptionally distinct LEC cluster containing 700 cells.

From control samples, we curated a transcriptional signature of healthy kidney lymphatics (Supplemental Data 1), comprising 227 differentially expressed genes (DEGs) from 295 LECs. These genes were enriched for gene ontology (GO) terms associated with lymphatic fate commitment (GO:0060838, fold-enrichment > 100, FDR = 1.66 × 10–2) and lymphangiogenesis (GO:0001946, fold-enrichment = 67.4, FDR = 8.37 × 10–3). Canonical LEC markers were identified, including PROX1 (log2FC = 2.97), PDPN (log2FC = 2.65), neuropilin 2 (NRP2, log2FC = 2.73), and C-C motif ligand CCL21 (log2FC = 7.23) (Figure 2B). We also identified genes previously linked to kidney disease (44–46), such as fatty acid binding protein 4 (FABP4, log2FC = 5.69), trefoil factor 3 (TFF3, log2FC = 5.58), and angiopoietin 2 (ANGPT2, log2FC = 2.46) (Figure 2B).

Given the frequent use of PROX1 and LYVE1 to identify or target kidney lymphatics in preclinical studies (11, 14), we examined their expression within the human kidney in more detail. PROX1 was detected not only in LECs, but also in loop of Henle and distal convoluted tubule clusters (Figure 2B), a finding validated by PROX1 and E-cadherin (CDH1+) immunolabeling of medullary tubules (Figure 2C) (47). In contrast to mouse data (38, 48), PROX1 was not detected in vasa recta at both the transcript (Figure 2B) and protein level (Figure 2D). LYVE1, meanwhile, was expressed by macrophages (Figure 2E) as reported in mouse (49) and human kidneys (50), and it was also detected in glomerular (Figure 2F) and peritubular capillary endothelium (Supplemental Figure 1, D–E).

To probe the phenotype of human kidney lymphatics, and whether the vessels we detected included smooth muscle–lined collecting vessels (51) as in the lungs (52) and skin and mesentery (53), we costained kidneys for PDPN and α-smooth muscle actin (ACTA2). Kidney lymphatics in both hilum and cortex lacked smooth muscle coverage (Figure 2, G and H). We corroborated this using subclustering analysis of our scRNA-Seq atlas, combining 295 LECs from healthy kidney with 157 additional cells from a recent study (54). This revealed 2 transcriptionally distinct LEC subclusters (Figure 2I), which expressed LEC capillary markers PROX1, PDPN, and CCL21 (Figure 2J) with sparse LYVE1 expression, consistent with our imaging data (Figure 1D). Only rare cells, not specific for either subcluster, expressed molecular markers of lymphatic valve endothelial cells, GATA2 and FOXC2 (Figure 2J). Differential expression analysis between the 2 capillary subclusters identified 129 DEGs (Supplemental Data 2). One subcluster was enriched for adipose signaling peptide neurotensin (NTS, log2FC = 3.60) (55), and the other expressed CCL2 (log2FC = 3.30), CXCL2 (log2FC = 4.22), and ICAM1 (log2FC = 4.15) (Figure 2K), indicative of capillaries involved in immune cell egress.

A multiorgan transcriptomic atlas reveals an organ-specific kidney lymphatic profile. To investigate if, akin to their blood vascular counterparts, kidney lymphatics possess an organ-specific signature (56, 57), we created a multiorgan human LEC atlas by integrating our 452 kidney LECs with scRNA-Seq data of LECs from other organs, including skin (n = 4,765 cells) (58), breast (n = 4,991) (59), heart (n = 432) (60, 61), lung (n = 1,891) (62), and small and large intestine (n = 462 and 471, respectively) (63). The final dataset encompassed 13,454 LECs from 19 anatomical sites (Supplemental Figure 3A). We resolved 5 transcriptionally distinct subclusters (Figure 3A and Supplemental Data 3). Four subclusters expressed lymphatic capillary markers CCL21 and LYVE1, while the fifth expressed lymphatic valve markers FOXC2 and integrin alpha 9 (ITGA9) (Supplemental Figure 3B) (64). Visceral organ–derived LECs (kidney, heart, lung, intestines) were predominantly grouped within 1 subcluster (LEC1), whereas breast lymphatics were found in LEC1, LEC2, and LEC3, and skin lymphatics in LEC2, LEC3, and LEC4 (Figure 3B). This spatial segregation was reflected in predicted transcription factor activity (Supplemental Figure 3C).

Comparative analysis (Supplemental Data 4) identified 118 DEGs upregulated in kidney LECs compared with other organs (Figure 3C). The top kidney lymphatic-enriched genes included DNASE1L3 (log2FC = 3.77, P = 3.24 × 10–148), the chemokine CCL14 (log2FC = 3.03, P = 7.00 × 10–59), the netrin receptor UNC5B (log2FC = 2.26, P = 9.65 × 10–29), the growth factor midkine (MDK, log2FC = 1.98, P = 5.56 × 10–21), and the anti-protease α2 macroglobulin (A2M, log2FC = 1.80, P = 4.00 × 10–33). Most of these genes were also expressed by blood endothelia in the kidney, whereas A2M was also expressed by stromal cells (Supplemental Figure 3D). Among the 251 DEGs with lower expression in kidney LECs compared with those from other organs (Figure 3D) were LYVE1 and the major neutrophil chemoattractant CXCL8 (65), the latter of which was also absent from heart, lung, and intestinal LECs. Conversely, LECs in these visceral organs expressed the alarmin cytokine IL33 (66), which was reduced in lymphatics of the skin and breast (Figure 3D and Supplemental Figure 3E).

To provide pathological context to the kidney lymphatic DEGs, we examined their expression in NephroSeq, a gene expression database of kidney diseases. DNASE1L3 was significantly upregulated in the tubulointerstitium of patients with lupus nephritis (n = 31) compared with controls (n = 8, mean difference in log2 expression = 1.1 ± 0.34–1.9, P = 0.0013) (Figure 3E). Conversely, MDK was significantly upregulated in several inflammatory and metabolic kidney diseases, except for minimal change disease (Figure 3F).

Collectively, our analyses demonstrated that kidney LECs have an organ-specific molecular profile, enriched for DNASE1L3, MDK, and CCL14, with reduced expression of canonical immune trafficking markers such as LYVE1 and CXCL8.

Perturbation of kidney lymphatic architecture and endothelial junctional configuration in chronic transplant rejection. Lymphangiogenesis has been observed during transplant rejection in both rodent models (23, 25, 31, 67) and humans (24, 26–28), but whether this is protective or promotes alloimmunity remains unclear. To investigate this in the human context, we profiled kidney transplants with chronic mixed rejection, a setting in which both donor-specific antibodies and T cells target HLA+ molecular expressed on tubular epithelial and blood endothelial cells.

We analyzed 3 allografts with histological features consistent with chronic mixed rejection, including T cell– and antibody-mediated injury (Supplemental Table 2), and compared them with control kidneys obtained from nontransplanted donor organs. In rejecting allografts, the lymphatic vascular network exhibited marked disorganization, with loss of the hierarchical structure observed in controls (Figure 4A). Quantitative analysis revealed a 7-fold increase in mean lymphatic vessel density (95.12 ± 49.21 vs. 690.3 ± 121.6 vessels/mm3, P = 0.0014), accompanied by reductions in the distribution of vessel lengths (median difference = 132 vs. 68.4 μm, P = 0.0001), vessel radius (9.05 vs. 4.9 μm, P < 0.0001), and branching angle (112 versus 103, P < 0.0001) (Figure 4B and Supplemental Videos 5 and 6). Notably, lymphatic vessels also infiltrated the allograft medulla, a region devoid of lymphatics in healthy kidneys (Figure 4, C and D).

LEC-cell junctions are key regulators of immune cell trafficking. In homeostasis, these junctions form discontinuous “button-like” structures that facilitate leukocyte entry into lymphatics, whereas during chronic inflammation, they transition into continuous “zipper-like” formations that impair lymphatic drainage (68–70). Given the accumulation of infiltrating lymphocytes in chronically rejecting grafts (71–73), we hypothesized that altered lymphatic junctional architecture might be a feature of rejection. To assess this, we immunostained for vascular endothelial cadherin (CDH5), a key component of endothelial junctions (Supplemental Video 7), and used PDPN to distinguish lymphatics from blood vessels (Supplemental Figure 4). Discontinuous CDH5+ LEC junctions were quantified in both control (Figure 4E) and chronic rejection (Figure 4F) samples, and values were normalized to total lymphatic network volume (Figure 4G). We observed a reduction in disconnected (button-like) junctions in rejecting allografts compared with controls (Figure 4H, mean difference = 2.7 × 105 ± 7.3 × 104 CDH5+ junctions per mm3 lymphatic vessel), consistent with a shift toward a zipper-like configuration.

Tertiary lymphoid structures form around lymphatics in chronic transplant rejection accompanied by altered intralymphatic and perilymphatic lymphocyte accumulation. Given the structural perturbation of kidney lymphatics in rejecting allografts, we next examined their spatial relationship to organized immune responses within chronic rejection. A hallmark of alloimmunity is the formation of tertiary lymphoid structures (TLSs), ectopic lymph node–like aggregations of T cells and B cells, where follicular DCs and high endothelial venules (HEVs) also develop. TLSs facilitate local antigen presentation and lymphocyte activation, and they have been associated with progressive graft injury and dysfunction (74–79).

Using triple immunolabeling, we found PDPN+ lymphatics were observed close to CD4+ T cell– and CD20+ B cell–rich aggregates (Figure 5A) in 3 rejecting allografts, consistent with previous reports (26–28). To assess the relationship between lymphatics and TLS maturation, we examined PDPN+ lymphatics relative to CD21+ follicular DCs and peripheral lymph node addressin (PNAd+) HEVs, the latter serving as a marker of mature TLS (67, 80, 81). The lymphatic network in rejecting allografts interconnected multiple mature TLSs containing HEVs (Figure 5B and Supplemental Video 8). Such connections were not detected between CD31+ vessels (Supplemental Figure 5A). Spatiotemporal analysis revealed that all identified TLSs were near lymphatic vessels (Figure 5C) (n = 9/9, 100%), whereas only half contained HEVs (n = 5/9, 55.6%, P = 0.023). In mature TLSs with HEVs, PDPN+ LECs were significantly closer to the TLS core than HEVs (Figure 5D, mean distance = 49.53 ± 23.83 μm vs. 109.6 ± 25.13 μm, 95% CI = 24.33–95.76, P = 0.0047), suggesting that lymphatics are an early feature of TLS organization.

To explore lymphatic-lymphocyte relationships beyond defined TLS regions, we performed 3D imaging and spatial quantification of PDPN+ lymphatics relative to CD20+ B cells and CD4+ T cells (Figure 6, A and B, and Supplemental Video 9). Intraluminal CD20+ B cell density was reduced by half in rejecting allografts compared with controls (Figure 6C), although total B cell numbers were equivocal, suggesting that this reflects increased lymphatic volume rather than changes in B cell abundance. In contrast, total intraluminal CD4+ T cells increased in rejecting kidneys, with a 3-fold increase in CD4+ T cell density (Figure 6D) relative to controls, a markedly higher density than was detected in the surrounding allograft parenchyma.

To further assess how lymphocyte position relative to lymphatics is altered in rejection, we performed spatial statistical analysis, by computing a normalized distance metric for each B cell (Figure 6E) and T cell (Figure 6F) to its nearest lymphatic vessel, and comparing this to a null model of random spatial distribution (82). CD20+ B cells showed no significant spatial association with lymphatics in either control kidneys (n = 703 cells; P = 0.631) or rejecting allografts (n = 2,963 cells; P = 0.326) (Figure 6G). However, CD4+ T cells (n = 2,149 cells across 2 controls) had a peak distribution within 0–100 μm from the nearest lymphatic vessel and were significantly enriched near lymphatic vessels compared with a random distribution (P = 0.029). This association was lost in rejecting allografts (n = 4,382 cells, P = 0.699) (Figure 6H), indicating disrupted T cell–lymphatic proximity in the context of chronic rejection.

Molecular profiling reveals IFN-γ–driven coinhibitory remodeling and alloantibody targeting of allograft lymphatics. Having established that lymphatics are structurally perturbed and spatially associated with immune aggregates in chronic rejection, we next investigated whether LECs in this setting exhibit an altered molecular profile. To do this, we first performed comparative transcriptomic analysis of LECs from healthy kidneys, rejection, and CKD (Supplemental Data 5–7).

GO revealed that LECs from rejecting allografts were enriched for pathways related to the negative regulation of viral process (GO:0048525, fold-enrichment = 90.26, FDR = 5.95 × 10–2), including IFN-induced transmembrane proteins IFITM2 (log2FC = 1.76, P = 5.89 × 10–5) and IFITM3 (log2FC = 1.62, P = 6.86 × 10–11) (Figure 7A). IFN-γ was specifically enriched in T cells and NK cells in our scRNA-Seq dataset (Figure 7B), whereas other IFN types were not detected. We then examined an IFN-γ response signature — including levels of IFITM2, IFITM3, and the IFN-γ receptor subunits IFNGR1 and IFNGR2 — which was prominent in LECs and in blood endothelial cells and macrophages from rejecting allografts (Figure 7C). To contextualize this response, we compared the LEC profile in chronic rejection with that of HEVs, identified by enrichment for PNAd (NTAN1) and downregulation of Notch pathway genes RBPJ and JAG1 (Supplemental Figure 5B) (83, 84). Unlike LECs, HEVs lacked lymphatic markers PROX1 and PDPN (Supplemental Figure 5C). Instead, they expressed transcripts involved in leukocyte recruitment, activation, and regulation, such as CXCL16, fractalkine (CX3CL1), CD40, and IL-32 (Supplemental Figure 5D and Supplemental Data 8), highlighting a distinct immune regulatory profile compared with LECs.

We next explored potential ligand-receptor interactions between LECs and lymphocytes using CellPhoneDB (85). Predicted cell-cell communication was highest in rejecting kidneys compared with CKD or healthy controls (Supplemental Figure 6A), with most interactions occurring between LECs and T cell subsets (Supplemental Figure 6B). These included IFN-γ–IFNGR signaling from CD8+ T cells to LECs across both control and rejecting kidneys (Supplemental Figure 6C). Chemokine-based interactions included established axes such as CCL21, CCL2, and ACKR2 (Supplemental Figure 7A), although CCL14/ACKR2 signaling with CD4+ effector T cells was reduced in rejection. Many chemokine receptors for ACKR2 ligands, including CCR2, CCR5, and CCR7, were expressed by T cells (Supplemental Figure 7B).