Volume 259, Issue 3 p. 236-253
Original Article
Open Access

Stromal transdifferentiation drives lipomatosis and induces extensive vascular remodeling in the aging human lymph node

Tove Bekkhus

Tove Bekkhus

Department of Medical Biochemistry and Microbiology, Uppsala University, Uppsala, Sweden

Search for more papers by this author
Anna Olofsson

Anna Olofsson

Department of Medical Biochemistry and Microbiology, Uppsala University, Uppsala, Sweden

Search for more papers by this author
Ying Sun

Ying Sun

Department of Immunology, Genetics and Pathology, Rudbeck Laboratory, Uppsala University, Uppsala, Sweden

Search for more papers by this author
Peetra U Magnusson

Peetra U Magnusson

Department of Immunology, Genetics and Pathology, Rudbeck Laboratory, Uppsala University, Uppsala, Sweden

Search for more papers by this author
Maria H Ulvmar

Corresponding Author

Maria H Ulvmar

Department of Medical Biochemistry and Microbiology, Uppsala University, Uppsala, Sweden

Correspondence to: MH Ulvmar, Department of Medical Biochemistry and Microbiology, Uppsala University, Uppsala, Sweden.

E-mail: [email protected]

Search for more papers by this author
First published: 11 November 2022
Citations: 3

No conflicts of interest were declared.

Abstract

Lymph node (LN) lipomatosis is a common but rarely discussed phenomenon associated with aging that involves a gradual exchange of the LN parenchyma into adipose tissue. The mechanisms behind these changes and the effects on the LN are unknown. We show that LN lipomatosis starts in the medullary regions of the human LN and link the initiation of lipomatosis to transdifferentiation of LN fibroblasts into adipocytes. The latter is associated with a downregulation of lymphotoxin beta expression. We also show that isolated medullary and CD34+ fibroblasts, in contrast to the reticular cells of the T-cell zone, display an inherently higher sensitivity for adipogenesis. Progression of lipomatosis leads to a gradual loss of the medullary lymphatic network, but at later stages, collecting-like lymphatic vessels are found inside the adipose tissue. The stromal dysregulation includes a dramatic remodeling and dilation of the high endothelial venules associated with reduced density of naïve T-cells. Abnormal clustering of plasma cells is also observed. Thus, LN lipomatosis causes widespread stromal dysfunction with consequences for the immune contexture of the human LN. Our data warrant an increased awareness of LN lipomatosis as a factor contributing to decreased immune functions in the elderly and in disease. © 2022 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

With aging, the function of our immune system declines, and it is well established that vaccinations, for example, do not work as well in the elderly as in the younger population [1]. The underlying reasons for impaired immune responses in the elderly are complex and most likely multifactorial. One group of contributing factors can be linked to environmental changes in lymphoid tissues [2]. This includes involution of the thymus [3, 4] and a decline in bone marrow (BM) hematopoietic stem cells [2]. In both these cases, the healthy tissue is eventually replaced by adipocytes [2-4]. A related but less well-known and studied process is lymph node (LN) lipomatosis [5, 6]. Like the age-related accumulation of adipocytes in thymus and BM [2, 4], LN lipomatosis is a process where the normal parenchyma is replaced by adipocytes. Analysis of multiple LNs across the human body support the idea that lipomatosis increases progressively with age [5-7].

The LNs are crucial organs for the induction of protective adaptive immune responses and immunological memory in vaccination, infection, and cancer [8, 9]. They are characterized by highly structured immune cell compartments allowing blood-derived naïve B- and T-cells to become activated by lymph-derived antigens and antigen-presenting cells (APCs) [8]. The structure and function of the LN depends on its highly specialized stromal cell (SC) environment with LN-specific subsets of fibroblasts [10] and a highly specialized blood and lymphatic vasculature [8, 11-13]. The high endothelial venules (HEVs), which constitutively express a unique pattern of L-selectin ligands called peripheral node addressins (PNAd) [14], are essential for lymphocyte recruitment to the LN. The LN sinusoidal lymphatic sinuses allow transport of tissue antigens and APCs into the LN but also provide the exit route out of the LN for activated and recirculating lymphocytes [8, 15, 16]. We and others have also recently shown that the LN lymphatic vasculature displays impressive heterogeneity and niche-specific functions [11, 12, 17], reflecting the need to coordinate complex immune trafficking and responses.

Although LN lipomatosis is a common pathological change seen in the elderly [5], the mechanisms driving these changes and the potential effects on the LN microenvironment have up until now not been studied. Because LNs are always surrounded by adipose tissue [18], one possible mechanism that has been proposed is the infiltration of surrounding adipocytes into the LN [7]. Our data do not support this notion. Instead, through careful analysis of LNs with different degrees of lipomatosis, we can show that it starts from deeper parts of the medullary parenchyma and demonstrate the presence of cells that display a transitional phenotype with both fibroblast and adipocyte lineage marker expression. This suggests that LN lipomatosis is driven by transdifferentiation of medullary fibroblasts into adipocytes. We show that these changes are associated with downregulation of lymphotoxin beta (LTB) in affected areas, a factor known to counteract adipocyte differentiation of SC precursors in the early development of the LN [19]. We also provide evidence that supports the idea that medullary reticular cells (MedRCs) are inherently more prone to transdifferentiate into adipocytes than T-cell zone reticular cells (TRCs), providing an additional explanation of the observed initiation of the pathology in the medulla of the LN. Finally, we demonstrate that LN lipomatosis causes loss of the medullary stroma and extensive vascular remodeling of the HEVs and the lymphatic vasculature, changing the immune contexture of the human LN.

Materials and methods

Biobank material and ethical considerations

Formalin-fixed and paraffin embedded (FFPE) biobank LNs from breast cancer patients were used for immunostaining and RNAscope in situ hybridization. The patient cohort included patients with noninvasive ductal carcinoma in situ (DCIS) (patients: n = 22, LNs: n = 65) and invasive ductal carcinoma (IDC) of breast without LN metastasis (patients: n = 44, LNs: n = 115). Only patients with no neoadjuvant therapy and no detected LN or distant metastasis at the time of the primary surgery were included in the study. The cohorts were previously described [20, 21]. Pancreatic FFPE LNs from organ donors (ODs) were used as controls (donors: n = 14, LNs: n = 66). The approval to use biobank and OD LNs was provided by the Uppsala regional ethics committee, 2017/061 and addition 2017/061:1 and 2017/061:2 to MHU. Mice for tissue isolation were held under Ethical Permit 6009/17 approved by the Uppsala Animal Experiment Ethics Board to MHU.

Immunofluorescence staining

The protocol for the immunostaining was previously described (Bekkhus et al [21]). In short, FFPE LNs were treated with xylene and an ethanol gradient to remove paraffin wax and rehydrate the tissues. Antigens were retrieved in 1 mm ethylenediaminetetraacetic acid (EDTA) (pH9) (Invitrogen, Thermofisher Scientific, Waltham, MA, USA) or 1× pH9 retrieval buffer (Dako, Agilent Technologies, Santa Clara, CA, USA) at 97 °C for 20 min. The tissues were blocked with 0.033% streptavidin (Sigma-Aldrich, St. Louis, MO, USA), 0.0033% biotin (Sigma-Aldrich, St. Louis, MO, USA), in the case of biotin–streptavidin and TSA amplification, and 5% donkey serum (Sigma-Aldrich, St. Louis, MO, USA), all diluted in PBS with 0.05% Tween20 (Merck, Burlington, MA, USA) (PBST), for 15, 15 and 20 min, respectively. The tissues incubated in primary antibodies (supplementary material, Table S1) at 4 °C overnight. The tissues were washed in PBST and incubated in secondary antibodies (supplementary material, Table S2) for 30 min at room temperature. For biotin–streptavidin amplification, the tissues also incubated in CF647 conjugated streptavidin (Biotium, Fremont, CA, USA) for 30 min at room temperature. For TSA amplification the primary antibodies for the amplified antigen incubated overnight followed by 30 min incubation with secondary antibody, 30 min incubation with HRP-conjugated streptavidin (1:500) and 10 min with TSA-Cy5 (Perkin Elmer, Waltham, MA, USA) (1:50). The antibodies were stripped from the tissue by incubation in 1× pH9 retrieval buffer (Dako) at 97 °C for 20 min. Next, the primary antibodies for the second antigen incubated for 1 h followed by 30 min incubation with secondary antibodies. For nuclear counterstaining the tissues were incubated in DAPI (Invitrogen) for 5 min, and the slides were mounted with ProLongGold (Invitrogen) and No. 1.5 coverslips.

Cell sorting for adipogenic culture

Inguinal, brachial, and axillary LNs (n = 210–240 per experiment) were harvested and digested as described by Xiang et al [11]. CD45+ immune cells were depleted from the samples using CD45 beads and LS columns following the manufacturer's instructions (Miltenyi Biotec, Bergisch Gladbach, DE, USA). The pellets were resuspended in Fc block (1:100 in fluorescence-activated cell sorting [FACS]) buffer containing 2 mM EDTA and FBS (0.5%) and incubated shortly before adding primary antibodies and then incubated on ice for 30 min (supplementary material, Table S3). The cells were washed by adding FACS buffer. Sytox Blue (Invitrogen) dead cell staining reagent was added before the sorting. Cells were sorted using a BD FACSAria III (BD Biosciences) (100 μm nozzle, 20 psi, and an acquisition rate of 500–2,000 events per second) with a high-purity sorting mask as described elsewhere [22]. Live single cells were gated using FSC-A/SSC-A followed by FSC-H/FSC-W and SSC-H/SSC-W. LN fibroblasts (CD45neg, CD11bneg, PECAM1neg, PDPN+) were sorted into three fractions (supplementary material, Figure S1) based on the markers CD34 and BP3 (BST1): BP3neg CD34neg MedRCs, BP3neg CD34pos CD34+ SCs, and BP3+CD34neg RCs (i.e., TRCs, follicular dendritic cells [FDCs], and marginal reticular cells [MRCs]), the latter also referred to as BP3+ RCs in the text. 60,000–110,000 cells were collected per population and transferred for culture in coverslip-covered petri dishes with StableCell™ MEM, alpha modification medium (Sigma-Aldrich, St. Louis, MO, USA) supplemented with 1% L-glutamine (Invitrogen), 1% penicillin–streptomycin stock solution, 10% FBS, and anti-LTΒR antibody (1:200, ab65089, Abcam). After 3 days of culture, control cells were fixed. Cultures were all confirmed to be dense and healthy, based on morphology, before treatment with adipogenic medium (0.5 mm IBMX, 10 μg/ml insulin, and 1 μm dexamethasone and 100 μm indomethacin) for 3 days and fixing in 4% PFA, followed by light microscopy and staining for perilipin and alpha smooth muscle actin (αSMA)-AF647 (supplementary material, Tables S1 and S2). For nuclear counterstaining, the cells were incubated in DAPI (Invitrogen) for 5 min.

RNAscope in situ hybridization

The RNAscope studies were performed following the manufacturer's instructions (RNAscope® Fluorescent Multiplex Kit User Manual Part 1 and 2, Advanced Cell Diagnostics, Inc., Hayward, CA, USA) using their kit for amplification. In short, FFPE LNs were sectioned at 4 μm under RNase-free conditions. Sections were melted for 1 h at 60 °C and treated with xylene and absolute ethanol to remove paraffin. Endogenous peroxidase was blocked by incubation with hydrogen peroxide for 10 min, and antigens were retrieved by incubation in 1X RNAscope Target Retrieval Reagent for 15 min at 99 °C. Sections were then treated with RNAscope Protease Plus and incubated with probes for LTΒ (310471-C2) which were left to hybridize for 2 h at 40 °C. Sections were stored in 5× Saline Sodium Citrate (SSC) buffer overnight at RT. Next, Amp1, Amp2, and Amp3 were left to hybridize for 30, 30, and 15 min, respectively, at 40 °C. The HRP-C2 signal was developed by incubation with HRP-C2 for 15 min at 40 °C followed by Opal570 (1:1500) for 30 min at 40 °C and HRP block for 15 min at 40 °C. The slides were counterstained with DAPI and mounted with ProLongGold and No. 1.5 coverslips. All tissues were imaged the next day.

Imaging

For image acquisition of immunofluorescence staining, the Vectra Polaris™ Automated Quantitative Pathology Imaging System (Akoya Biosciences, Marlborough, MA, USA) was used with its whole-slide scan function, 20× objective (5 μm/pixel), and filters for DAPI, Opal520, Opal570, Opal620, and Opal 690. For the imaging of RNAscope and cell cultures, a LSM 700 confocal microscope (Zeiss) with Plan-Apochromat 63×/1.40 Oil DIC M27 and Plan-Apochromat 20×/0.8 M27 objectives were used respectively.

Image analysis

All image analysis was performed using QuPath 0.2.3 [23], Fiji ImageJ version 1.52p [24], and Phenochart 1.0.12 (Akoya Biosciences).

In Figure 1A, the threshold for green channel (CCL21) was adjusted to also allow for the detection of autofluorescence outlining the LN structure. Since adipocytes lack autofluorescence, they show up like black holes in tissue. The proportion of patients with lipomatosis shown in Figure 1B was scored based on the presence of adipocytes within the capsule in all LNs from all patients/donors to give each patient/donor a status of being affected by lipomatosis or not.

Details are in the caption following the image
Lipomatosis is frequent in lymph nodes (LNs) from patients above 40 years of age and starts within the medullary area. (A) Immunofluorescence staining of axillary LNs with low (left, 5.4%) and high (right, 62.0%) lipomatosis burden. Staining of chemokine CCL21 in the green channel with lowered threshold for signal, also allowing structural visualization through autofluorescence. Green asterisks mark examples of adipocytes. Dashed line in magenta marks the medulla (M), paracortex (PC), and cortex (C) in left panel. Dashed line in green marks the outline of the LN in the right panel. Scale bar: 1 mm. (B) Proportion of patients with lipomatosis including organ donors (ODs, n = 14 donors), patients with ductal carcinoma in situ (DCIS, n = 22 patients), and invasive ductal carcinoma (IDC, n = 44 patients). (C) Plot of extent of lipomatosis (percentage lipomatosis area/LN area) and age of patients, including ODs (n = 12 donors, n = 57 LNs), DCIS (n = 20 patients, n = 59 LNs), IDC (n = 37 patients, n = 93 LNs). Each dot represents one LN (n = 209). The age range is similar between the groups, i.e. OD: 57 ± 14, DCIS: 61 ± 12, and IDC: 57 ± 12. (D) Proportion of affected LNs with adipocytes in medulla, medulla and paracortex, and medulla and paracortex and cortex, including LNs from ODs (n = 8 donors, n = 15 LNs), patients with DCIS (n = 21 patients, n = 54 LNs), and IDC (n = 44 patients, n = 107 LNs).

The extent of lipomatosis in Figure 1C was quantified as a percentage of the total LN area using QuPath and Fiji ImageJ. The channel for DAPI and the autofluorescence channels AF488 and/or AF555 were used to create regions of interest (ROIs), defining the intracapsular LN parenchyma as the total LN area using the ‘wand’ tool in QuPath. The ROIs were exported to Fiji ImageJ. To define the lipomatosis area, a threshold was set on the DAPI channel to cover the adipocytes with limited background, and the threshold was converted to a binary mask with dark background. A threshold without dark background was set on the binary mask, and the ROI for the LN area was used to measure the lipomatosis area. To limit the detection of nonadipocytic area, separate ROIs were made for the adipocytes to measure the lipomatosis area in the LNs with high background (less dense DAPI network). LNs with no morphologically confirmed adipocytes were set to 0% lipomatosis. LNs with at leastthree adipocytes were counted as having lipomatosis. LNs with one or two adipocytes were excluded from all analysis.

The localization of the lipomatosis in Figure 1D was defined based on the pattern of the DAPI, Claudin-5 (CLDN5), CCL21, and PNAd staining. The cortex was defined by dense DAPI staining (B-cell follicles) and the absence of CCL21high fibroblasts and PNAd- and/or CLDN5-positive vessels. The paracortex was defined by CCL21high fibroblasts and PNAd-/CLDN5-positive vessels. The medulla was defined by the presence of CLDN5-positive lymphatic sinuses and the absence of CCL21high fibroblasts. All LNs were in the end categorized individually based on the localization of the lipomatosis; medulla, medulla and paracortex; or medulla and paracortex and cortex. The analysis was performed using Phenochart. LNs with less than three adipocytes were excluded from the analysis.

To determine whether perilipin was expressed by fibroblasts or LECs, the localization was first determined based on the staining of perilipin, Lymphatic vessel endothelial hyaluronan receptor 1 (LYVE-1), αSMA, and CD19. LYVE-1 marked the medullary sinuses (MSs), αSMA the fibroblasts of the medulla and paracortex, and CD19 the B-cell follicles and B-cells of the medulla. If the perilipin+ adipocytes were present in the fibroblastic compartment, they were checked for costaining with αSMA, PDGFRβ and CD34. If they were present within the MSs, they were first checked for costaining with LYVE-1, but because LYVE-1 has a high heterogeneity in the human medullary lymphatic sinuses and is a marker for medullary macrophages, these LNs were also stained for CLDN5 to rule out co-expression with the medullary lymphatic sinus LECs. For the vector-based localization analysis in Figure 2B,C and in supplementary material, Figure S4B, the ‘Plot Profile’ tool in Fiji ImageJ was applied on a vector drawn across the outline of the αSMA+/perilipin+, PDGFRβ+/perilipin+ or CD34+/Perilipin+ stained cell border.

Details are in the caption following the image
Co-expression of fibroblast and adipocyte markers suggest lipomatosis is driven by adipogenic transdifferentiation of medullary fibroblasts. Immunofluorescence staining of lymph nodes (LNs) with lipomatosis. (A) Presence of adipocytes (perilipin, magenta) and fibroblasts (alpha smooth muscle actin; αSMA, green) in medullary cords. Arrowheads indicate adipocytes present in the layer of medullary lymphatic sinus lining fibroblasts. Scale bar: 100 μm. Pictures representative of 10 analyzed LNs. (B) Colocalization of αSMA (green) and perilipin (magenta) in medullary fibroblasts. Scale bar: 50 μm. Picture representative of 10 analyzed LNs. Inset shows vector (size: 4.6 μm) used for image analysis of αSMA and perilipin signal colocalization with vector-based profile plot of αSMA and perilipin. Black line denotes position of peak of perilipin in both profiles. (C) Colocalization of PDGFRβ (green) and perilipin (magenta) in medullary fibroblasts. Scale bar: 20 μm. Picture representative of five analyzed LNs. Inset shows vector (size: 7.0 μm) used for image analysis of PDGFRβ and perilipin signal colocalization with vector-based profile plot of PDGFRβ and perilipin. Black line shows position of peak of perilipin in both profiles.

The percentage of transformed fibroblasts from the FACS-sorted cultures in Figure 3C was quantified using Fiji ImageJ. A threshold was set on the DAPI channel (7–255) to mark all the nuclei. The ‘Analyze Particles’ tool was applied, and the nuclei were added to the ROI manager, excluding particles on the edges. If several nuclei were assigned to one ROI, separate ROIs were manually added to get the correct number of ROIs corresponding to the staining. The ROIs were then applied to the perilipin channel, where a threshold was set (10–255). Each nuclear ROI was assigned as being perilipin positive or negative in the immediate surrounding area. Eight  image frames using a 20x objective, per subset of cells (i.e BP3+ RCs, MedRCs and CD34+ SCs) were analyzed containing a mean number of 118 cells per frame, and the percentage of perilipin positive cells was calculated.

Details are in the caption following the image
Cell intrinsic sensitivity to adipogenesis of CD34+ SCs and MedRCs. (A) Immunofluorescent staining of cultured fibroblast subsets; BP3+ reticular cells (RCs), medullary reticular cells (MedRCs), and CD34+ SCs, including both control (in nonadipogenic culture for 3 days) and adipogenesis stimulated cells (in nonadipogenic culture for 3 days and in adipogenic stimulation culture for 3 days). DAPI (blue), perilipin (green), and alpha smooth muscle actin (αSMA, magenta). Scale bar: 50 μm. (B) Zoom-in of inset in (A), panel CD34+ SCs/adipogenic stimulation. Transmitted light morphologically indicates presence of lipid droplets in cells positive for perilipin. (C) Quantification of percentage of perilipin-positive cells after 3 days with adipogenic stimulation. Eight frames (mean: 118 cells/frame) per subset were analyzed, and the experiment was repeated once. Lines represent mean and SD. One-way ANOVA and Holm-Sidak's multiple comparisons test was used for statistical analysis. ****p < 0.0001.

The analysis of the RNAscope images in Figure 4B was performed using Fiji ImageJ. Samples with low to intermediate lipomatosis were selected to allow analysis of medullary regions with and without adipocytes within the same LN. Five image frames, taken with the 63x objective, were analyzed per LN and per condition (i.e. medulla far from adipocytes versus medulla close to adipocytes), defined by the pattern of DAPI, and compared to our other stained sections from the same LN (with staining for CLDN5, CCL21, PNAd, aSMA, and perilipin). Areas classified as close to lipomatosis were less than 200 μm from fat. Thresholds for DAPI and LTΒ were set and the areas measured. Owing to a risk of high patient and RNA quality variations as a consequence of using human biobank samples, a paired analysis was performed within the same LN, comparing the medullary areas far from and close to adipocytes, and the thresholds for DAPI and LTΒ were adjusted for the specific sample.

Details are in the caption following the image
Lymph node (LN) lipomatosis is associated with downregulation of LTB in affected areas. (A) RNAscope of lymphotoxin beta (LTB) mRNA in medulla far from adipocytes (>200 μm, left) and close to adipocytes (<200 μm, right). Pictures taken from same LN. Dotted line and A visualize the presence of adipocytes. Scale bar: 20 μm. (B) Quantification of signal of LTB mRNA as percentage of LTB mRNA area out of DAPI area in parts of medulla far from adipocytes versus close to adipocytes within the same LNs (n = 5 LNs). Dots represent mean value of three frames per LN. Paired t-test was used for statistical analysis. **p < 0.01. (C) Immunofluorescent staining of CD19 (medullary B-cells, green) and Claudin-5 (CLDN5) (medullary lymphatic sinuses, magenta) in parts of medulla far from adipocytes (>200 μm) and close to adipocytes (i.e. frames containing adipocytes) within the same LN. Dotted lines visualize adipocytes. Scale bar: 75 μm. (D) Quantification of density of medullary B-cells as percentage of medullary area not covered by adipocytes or lymphatic sinuses within the same LN (n = 5 LNs). Dots represent mean value of three frames per LN. Paired t-test was used for statistical analysis. No significant difference was detected between the groups.

The B-cell density analysis in Figure 4D was performed using Fiji ImageJ. The same LNs as in the RNAscope analysis were stained for CD19 and CLDN5. Six ROIs were selected based on the areas used for the analysis of the RNAscope sections, three far from adipocytes and three close to adipocytes. Areas occupied by medullary lymphatic sinuses (CLDN5+ staining) and adipocytes (morphologically defined as adipocytes with a lack of staining of CLDN5, CD19, and DAPI) were excluded from the area of analysis. A threshold was set for CD19, and the percentage of CD19 area was quantified. Individual thresholds were set for different LNs owing to patient variations of the biobank samples.

The percentage of MS areas was quantified using Fiji ImageJ. The LN area was reused from the analysis of the extent of lipomatosis, and a threshold was set on the CLDN5 channel. A ROI was made for the region of the LN occupied by medullary and trabecular lymphatic sinuses, which transverse into the medulla of the human LN, and the percentage of MS area out of the total LN area was calculated.

The number of collecting-like vessels in lipomatosis-affected LNs shown in Figure 5C was quantified using Phenochart. PDPN+CLDN5+ vessels that were embedded in adipose tissue were defined as collecting-like vessels, and the number was quantified by manual counting. The collecting-like vessels were also defined as being PDPNhigh/CLDN5+/CCL21/CD36/MARCOlow-negative/LYVE-1low-negative by their staining pattern shown in Figure 5D–G.

Details are in the caption following the image
Loss of medullary sinusoidal lymphatic network in lipomatosis with compensatory establishment of collecting-like vessels inside affected areas. (A) Correlation analysis of lipomatosis burden versus percentage of medullary sinus (MS) area. n = 10 lymph nodes (LNs) from n = 6 organ donors (ODs), n = 47 LNs from n = 19 patients with ductal carcinoma in situ (DCIS), n = 80 LNs from n = 35 patients with invasive ductal carcinoma (IDC). Spearman correlation test was used to test for significance. *p < 0.05. (B) Immunofluorescence (IF) staining of axillary LN with low to intermediate lipomatosis (5.52%) localized to medulla marked by asterisks in magenta. LN stained for vascular tight junction protein Claudin-5 (CLDN5, green) marking medullary lymphatic sinuses together with DAPI (nuclei). Scale bar: 750 μm. Representative image is shown. (C) Correlation analysis of number of collecting-like vessels in lipomatosis-affected area and extent of lipomatosis. Spearman correlation test was used to test for significance. ****p < 0.0001. (D–G) IF staining of collecting-like vessels in parts of LN affected by lipomatosis. Arrowheads visualize collecting-like vessels marked by high podoplanin (PDPN, magenta) and no or low detection of CD36 (green), which, however, stain the adipocytes in the same frame (D), MARCO (green) (E) LYVE-1 (F) (green) or CCL21 (green) (G). Scale bar: 50 μm. Representative pictures are shown from 10 analyzed LNs.

For the analysis of HEV remodeling in Figure 6B–F,I, red–green–blue (RGB) snapshots of lipomatosis associated HEVs were created using Phenochart and analyzed in Fiji ImageJ. The number of PNAd+CLDN5+ vessels was quantified, and vessels within 200 μm from adipocytes were defined as lipomatosis-associated HEVs. The lumen diameter was measured as a mean of three independent measurements of the shortest diameter to reduce the effects of the angle of sectioning and as previously described [21]. Each vessel was classified as nondilated (no lumen), intermediately dilated (0–10 μm), highly dilated (>10 μm), and/or having a thin endothelial cell layer or loss of PNAd expression in parts of the vessel [21]. Briefly, to determine whether a vessel had PNAd loss, the PNAd staining pattern was compared to the CLDN5 pattern. If there was a loss of PNAd signal along parts of the perimeter of the vessel, where CLDN5 was still apparent, the HEV was defined as having PNAd loss. A vessel that had lost the classical plump, cuboidal endothelial shape of a HEV and turned thin like a normal venule was classified as having a thin endothelium.

Details are in the caption following the image
Lipomatosis causes remodeling of HEVs associated with reduced density of naïve T-cells. (A) Immunofluorescence (IF) staining of paracortical (P) and lipomatosis associated paracortical (LAP) high endothelial venules (HEVs) marked by PNAd (magenta) and Claudin-5 (CLDN5, green). Adipocytes are marked by yellow asterisks. Trabecular sinuses are marked by white asterisks in the capsular center of the sinuses. (B–D) Percentage of highly dilated HEVs, lumen > 10 μm (B), HEVs with thin endothelial cell (EC) layer (C), and HEVs with loss of PNAd (D). (E and F) Correlation analysis of percentage of LAP HEVs with high dilation versus thin EC layer (E) and LAP HEVs with high dilation versus loss of PNAd (F). n = 19/14 lymph nodes (LNs) from 19/14 patients (P/LAP) in panel (A), n = 14 LNs from 14 patients in panels (C–F). Unpaired t-test and Mann Whitney test were used for statistical analysis of P and LAP HEV remodeling, and Pearson correlation test was used for correlation analysis. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. (G) Paired, intra-LN analysis of the density of naïve lymphocytes in the immediate surrounding area (a layer of 2–3 immune cells) of HEVs in the paracortex and in the LAP. Dots represent the mean value of the TCF1/7 density from five different patients with DCIS. Lines represent paired values from the same LN. Statistical analysis was performed using paired t-test. **p < 0.01. (H) IF staining HEVs, PNAd, and naïve lymphocytes (TCF1/7) in LNs with intermediate to low lipomatosis, comparing areas of paracortex (P) and LAP within the same LN. Images are representative of five LNs (DCIS). (I) Correlation analysis of percentage of highly dilated HEVs in paracortex (more than 200 μm from adipocytes) versus percentage of lipomatosis in lymph node (n = 19 patients). Spearman correlation test was used for statistical analysis. No correlation was detected; r = 0.01888, p = 0.9388.

The analysis of the TCF1/7+ cell presence in Figure 6G was performed using Fiji ImageJ. Highly dilated HEVs, surrounded by CCL21+ fibroblasts, within 200 μm from adipocytes, were selected by creating a ROI for this region. Only HEVs with a clear immune cell perimeter were included in the analysis to limit the bias of having a less dense TCF1/7 area as a consequence of an adipocyte being present in the frame. As controls, nondilated HEVs, surrounded by CCL21+ fibroblasts, more than 200 μm from adipocytes were selected by creating a ROI for this region. All HEVs were selected randomly across the paracortex, independently of the TCF1/7 staining, using a composite image of DAPI, PNAd, and CCL21 for the selection of the vessels. A threshold was set for the PNAd channel and was applied to create a binary image. Background was removed by deleting pixels outside a HEV using the freehand selection tool, and only the binary mask of the HEV was retained. If there was a loss of PNAd, the outline of the HEV was filled out using the paintbrush tool focusing on the composite image, where the outline of the vessel is structurally visible. The binary tool for filling holes was used, and the mask was added as a ROI for the HEV. The binary image was dilated by 20 iterations to increase the size of the mask of the HEV using the ‘dilate’ binary tool. Twenty iterations are equal to a layer of approximately two to three lymphocytes. The channel for TCF1/7 was selected and a threshold set. The ROI for HEV only and the 20× dilated ROI was used to measure the area of TCF1/7 outside of the LN by subtracting the area of the HEV itself from the 20× ROI area. The percentage of this area covered by TCF1/7 was used for data analysis. As with the RNAscope and CD19 analysis, due to high patient variations as a consequence of using human biobank samples, a paired analysis was performed within the same LN comparing the areas far from (>200 μm) and close to (<200 μm) adipocytes, and the thresholds for PNAd and TCF1/7 were adjusted for the specific sample. Three highly dilated HEVs and three nondilated HEVs were analyzed per patient, including five patients from the DCIS group.

To analyze the coverage of CD38+ plasma cells shown in the supplementary material, Figure S8B, LNs with low to intermediate lipomatosis were used to enable intra-LN analysis comparing medullary areas (defined by the presence of MSs and CD19+ cells) with high versus no lipomatosis. Three frames per condition were selected for analysis. Individual thresholds for the CD38 channel were set for the different LNs due to patient variations of the biobank samples. The area covered by CD38 staining was quantified as a percentage of the area of the frame. Eight LNs from six patients were included for the analysis and the mean percentage of CD38 area of each LN is presented in a dot plot.

Statistical analysis

Statistical analysis was performed with GraphPad Prism versions 7 and 8. For correlation analysis, Spearman or Pearson correlation tests were used based on the distribution of the data analyzed. For comparison between normally distributed paired groups, a paired t-test was used. In the cases of nonpaired groups, unpaired t-test or Mann–Whitney test were used based on the distribution of the data analyzed. In the case of more than two groups of normally distributed data, one-way ANOVA and HolmSidak's multiple comparisons test were used.

Bioinformatic analysis

The processed single-cell RNAseq data of LN SCs and cell annotation were obtained from a previous mouse study with publicly available data [10]. Cell types were stratified using Bst1 (Bp3) gene expression. ‘TRC’ (‘Ccl19hi TRC’), ‘Ccl19lo TRC’, ‘Cxcl9+ TRC’, ‘MRC’, and ‘FDC’ were defined as Bst1 high population; cell types of ‘Nr4a1+ SC’, ‘Inmt+ SC’, and ‘CD34+ SC’ were defined as Bst1 low population. We pooled clusters with high Bst1 expression (‘TRC’,‘Ccl19lo TRC’,‘Cxcl9+ TRC’, ‘MRC’,‘FDC’) and compared with clusters of ‘CD34+ SC’, ‘Nr4a1+ SC’, and ‘Inmt+ SC’ (expressing low Bst1), respectively. Wilcoxon rank sum test was applied to identify differentially expressed genes (DEGs) in Seurat package (version 4.1.1 PMID: 25867923). For cluster ‘CD34+ SC’ compared with clusters expressing high Bst1, 482 genes were identified as statistically significantly expressed (Bonferroni corrected P value < 0.01, fold change [log2] > 0.5 or < −0.5). For cluster ‘Nr4a1+ SC’ compared with clusters expressing high Bst1, 215 genes were identified as statistically significant expressed (P value and fold change were used same as describe above). For cluster ‘Inmt+ SC’ compared with clusters expressing high Bst1, 274 genes were identified as statistically significant expressed (p value and fold change were used same as describe previously). Heatmap visualization was produced using the ComplexHeatmap package [25]. For the analysis of scRNA-seq data of human LN stromal and dentritic cells [26], the processed data were obtained from ArrayExpress database (Accession No. E-MTAB-10206). Data processing and analysis were done as in a previous public study [26]. The UMAP plot was colored with marker genes using scaled normalized expression values.

Code

Code can be accessed from https://github.com/sherriying/lipomatosis_in_LN.

Results

Lipomatosis is frequent in LNs from patients above 40 years of age and starts within the medullary area

To analyze the pathology of LN lipomatosis, we used a collection of axillary LNs without metastasis from breast cancer patients [21], including patients with noninvasive DCIS and invasive ductal carcinoma (IDC) and pancreatic LNs from cancer-free organ donors (ODs). With two exceptions, all patients were above 40 years of age. The LNs were defined as having lipomatosis based on the presence of parenchymal adipocytes inside the LN capsule (Figure 1A). LNs with only one or two adipocytes were excluded from the analysis. A majority of the patients displayed LN lipomatosis (Figure 1B). Figure 1C illustrates inter- and intrapatient variation, where some LNs had no lipomatosis and others had lost all normal parenchyma. In contrast to lipomatosis, the analyzed LNs were rarely affected by fibrosis, only 1 out of 209 LNs displayed clear signs of fibrosis as assessed by tissue morphology, and only 1 out of 56 LNs specifically stained for alpha smooth muscle actin (aSMA) (fibroblast marker). Our data are consistent with previous data suggesting that LN lipomatosis is very common in humans above 40 years of age and becomes progressively more frequent with aging [5, 6].

It has not been established whether development of lipomatosis is more frequent within specific areas of LNs. With this in mind we analyzed the localization of the lipomatosis based on the basic structure of LNs: cortex (B-cell area), paracortex (T-cell area), and medulla (central lymphatic sinus area) (Figure 1A, left panel) [27]. Lipomatosis was most frequently found within the medullary compartment and was only found in the paracortex and/or cortex if the extent of lipomatosis was large enough to also reach these parts of the LNs (Figure 1D, for illustration see Figure 1A, right panel). In LNs with a low or intermediate degree of lipomatosis, it was exclusive to the medullary region and could be seen randomly across the medulla (Figure 1A, left panel). This supports the neoformation of adipocytes within the medulla rather than the invasion of the adipocytes surrounding the LNs. The latter would be expected to first affect peripheral areas, including the cortex, rather than deep areas of the LN parenchyma.

Co-expression of fibroblast and adipocyte markers suggests lipomatosis is driven by adipogenic transdifferentiation of medullary fibroblasts

Experimental data in mice support the idea that there exists a sensitive balance in the differentiation of LN mesenchymal cells into fibroblast lineages or adipocytes that depend on lymphotoxin beta receptor (LTΒR) signaling [19]. To determine whether the differentiation of medullary fibroblasts into adipocytes could be one contributing factor driving lipomatosis, we stained for perilipin, which is a marker of adipocyte differentiation [28], together with general markers for fibroblasts, i.e. αSMA and platelet-derived growth factor receptor beta (PDGFRβ), that are not expressed by adipocytes (Figure 2A–C and supplementary material, Figure S2B,C). In parallell perilipin was also found co-stained with Claudin-5 (CLDN5), a tight junction protein expressed by several types of endothelial cells, including the lymphatic endothelial cells (LECs) forming the MSs, and LYVE-1 expressed by the LECs of the MSs and by some subsets of macrophages but not by fibroblasts or adipocytes (supplementary material, Figure S2A) [11]. Perilipin+ adipocytes were most commonly found in the medullary cords of the medulla and rarely within the lymphatic MSs (Figure 2A) but sometimes appeared to transverse into the sinuses between the fibroblasts lining the MSs (arrowheads in Figure 2A). Perilipin and αSMA were observed to be co-expressed in individual adipocytes within all the analyzed LNs (Figure 2B) but not in the adipocytes found outside of the capsule (supplementary material, Figure S2B). Similar data were found for PDGFRβ with expression in individual parenchymal adipocytes in four out of five analyzed LNs (Figure 2C), but not in the extracapsular adipocytes (supplementary material, Figure S2C). In contrast, CLDN5 or LYVE-1 was never found to be co-expressed with perilipin (supplementary material, Figure S2A). Taken together, the expression of αSMA/PDGFRβ and perilipin in the same cells is suggestive of the presence of medullary fibroblasts in transition into adipocytes.

Cell intrinsic sensitivity to adipogenesis of CD34+ SCs and MedRCs

The transdifferentiation of LN fibroblasts into adipocytes seen specifically in the medulla may reflect that the medullary area contains specific phenotypes of fibroblasts more prone to convert into an adipogenic phenotype and/or that the LN environment in this area changes with aging, driving adipogenic transdifferentiation of fibroblasts in this specific region, or a combination of both. A niche-specific heterogeneity is known to characterize the LN fibroblasts [10, 26]. The main subsets include the FDCs, organizing the cortical B-cell zone of the LN, and the TRCs, organizing the paracortex. Two subsets of medullary fibroblast, here referred to as MedRCs and one subset of CD34+ fibroblasts, associated with the capsule and in the adventitia of rare and large blood vessels in the hilar region and medulla, have also been identified [10]. Although the mapping and molecular characterization of the fibroblast subsets in LNs are mainly derived from studies of mouse [10, 29], recent single-cell analyses of human LN fibroblasts indicate that major functions and subsets are conserved [26, 30]. Staining normal human LNs (with low lipomatosis), we confirmed at the protein level that this includes a conserved higher CCL21 expression in TRCs compared to MedRCs (supplementary material, Figure S3) and that CD34+, like in the mouse [10], is expressed by blood endothelial cells, the capsule, and in the adventitial fibroblasts of large blood vessels outside the LNs (supplementary material, Figure S4A).

To evaluate the potential difference between LN fibroblast subsets in their sensitivity to adipogenesis, we made use of the difference in expression of the marker BP3, also known as the BM SC molecule BST1 and the marker CD34 in mouse LN fibroblasts [10, 29]. Although TRCs and FDCs express high BP3, the expression of BP3 is low in MedRCs [29]. With the addition of the marker CD34, it is possible to sort three different subsets of podoplanin (PDPN)-positive SCs from mouse LNs (supplementary material, Figure S1). These include BP3+ CD34neg PDPN+ cells, which will be dominated by TRCs but may also contain the less abundant populations of FDCs and marginal reticular cells (MRCs) (therefore henceforth referred to as BP3+ RCs); the BP3neg CD34neg PDPN+ cells, which correspond to MedRCs; and, lastly, the CD34+ BP3neg PDPN+ cells, here referred to as CD34+ SCs. Under adipogenic culture conditions, CD34+ SCs displayed the highest conversion to adipocytes, followed by MedRCs, whereas the BP3+ RCs displayed minimal conversion (Figure 3A–C). The most sensitive subset, the CD34+ SCs, are rare in the medulla but they can be present through deep invaginations of trabecular sinuses and in the adventitia of large blood vessels (supplementary material, Figure S4A). CD34 is not lineage-specific for CD34+ fibroblast but is expressed abundantly by other cell types, including blood vessels (supplementary material, Figure S4A). No direct overlap in expression with parenchymal adipocytes could be confirmed, but close association between CD34+ cells and adipocytes occurred frequently (supplementary material, Figure S4B–D). Taken together, our data support the idea that MedRCs and CD34+ SCs share an ability to convert into adipocytes, whereas BP3+ RCs display higher resistance. Both MedRCs and rare CD34+ SCs in the medulla could contribute to adipogenesis.

To further understand the molecular differences between these cell types we used the published single cell analysis by Rodda et al [10] to reanalyze the mouse LN fibroblasts and focused on differentially expressed genes in the nuclear factor (NF)-kappa B signaling pathway, induced downstream of LTBR, and in the Hippo pathway (supplementary material, Figure S5A–C). Both these pathways have been implicated in controlling cell fate decisions between LN fibroblasts and adipocytes [19, 31]. For analysis, cell types were stratified based on Bst1 (also known as Bp3). TRC subsets, MRCs, and FDCs were defined as Bst1/Bp3 high populations. Cell types of ‘Nr4a1+ SC’, ‘Inmt+ SC’, and ‘CD34+ SC’ were defined as Bst1/Bp3 low populations. All these subsets express similar levels of Ltbr (supplementary material, Figure S5A). However, the analysis shows that Bst1/Bp3+ subsets, i.e. TRCs, MRCs, and FDCs, are enriched for genes associated with NF-kappa B signaling (including e.g. Relb and NFkb2), downstream of LTBR, whereas Nr4a1+ SCs and Inmt+ SCs (both MedRCs, which have been described as being located close to the medullary cords [10]) together with CD34+ SCs have lower expression of genes in this pathway (supplementary material, Figure S5B). Note also that the TRCs, MRCs, and FDCs all display higher expression of the LTBR ligand lymphotoxin beta (Ltb) compared to MedRCs and CD34+ SCs (supplementary material, Figure S5B), which is also a target gene downstream of LTBR. Similar results were observed for genes associated with the Hippo pathway (supplementary material, Figure S5C). In silico analysis of the published dataset of PDPN+ human SCs [26], which encompass TRCs and CD34+ SCs (supplementary material, Figure S6A,C,D), confirmed low but consistent expression of LTBR in all subsets (supplementary material, Figure S6B), which is also consistent with data from cultured human tonsil-derived fibroblast [32] and searchable datasets of peripheral human fibroblasts in pathology from other organs [33] (https://www.fibroxplorer.com/home). Differential expression of BIRC3 (associated with both the Hippo and NF-kappa B pathway) and RELB (NF-kappa B pathway) were also conserved between mouse and man (i.e. enriched for expression in TRCs) (supplementary material, Figure S6E,F). Human LN SC did, however, not show detectable expression of LTB and expression of the LTBR ligand LIGHT (TNFSF14) was only detected in scattered cells (supplementary material, Figure S6G,H). Taken together these data support the idea of cell-intrinsic differences between different LN SC subsets for the NF-kappa B and Hippo pathways, both known to be associated with resistance to adipogenic transformation [19, 31].

LN lipomatosis is associated with downregulation of LTB in affected areas

The main ligands for LTBR are heterodimers formed by LTA and LTB and the cytokine LIGHT (TNFSF14) [34]. Although LTA and LIGHT are more sparsely detected in situ in the medulla ([35] and authors' own observations), LTB is more abundantly detected. B-cells are a known main source of LTB [34, 35]. We therefore focused our analysis on LTB, in LNs with early stage and low to intermediate lipomatosis, comparing areas of the medulla affected or not affected by lipomatosis within the same LNs. Lower levels of LTΒ were detected in medullary areas associated with lipomatosis compared to areas without adipocytes (Figure 4A,B). However, analysis of medullary B-cell density in lipomatosis-affected versus normal medullary areas of consecutive sections of the same LNs and areas used in the LTΒ analysis did not show a difference (Figure 4C,D). These data support the idea that downregulation of LTΒ is associated with the presence of adipocytes at these sites, but downregulation of LTΒ is not caused by reduced overall presence of B-cells.

Loss of the medullary sinusoidal lymphatic network in lipomatosis with compensatory establishment of collecting-like vessels inside affected areas

In LNs with increasing degrees of lipomatosis, a progressive loss of lymphatic MS area was evident with a negative correlation between the MS area and the percentage of lipomatosis (r = [−0.213]) (Figure 5A, illustration of a LN with 5.52% lipomatosis 5B). A loss of the MS network would be expected to block the flow of lymph through LNs. Conspicuously, in LNs with established lipomatosis, we observed the presence of lymphatic vessels traversing the adipose tissue in the medulla, where the number of vessels increased with the extent of lipomatosis (Figure 5C–G). The vessels are defined by the expression of CLDN5 and high expression of PDPN and were found to increase in number with the extent of lipomatosis (Figure 5C, illustration in Figure 5D). To further evaluate which molecular features define the PDPN+ lipomatosis associated lymphatic vessels, we stained for CD36, which we have demonstrated is a specific marker for PTX3 paracortical lymphatic vessels [11] but which is also is expressed by adipocytes, visible in the picture (Figure 5D), and for MARCO (Figure 5E), which is a marker that defines the lymphatic MSs [11, 12]. Lipomatosis-associated vessels were negative for CD36 (Figure 5D) and displayed only occasional and low expression of MARCO (Figure 5E). The vessels also rarely or very weakly displayed LYVE-1(Figure 5F), a marker associated with MSs and subsets of macrophages, and were negative for CCL21 (Figure 5G), a marker of capillary initial lymphatic vessels. Lack of CD36 and low or absent MARCO, LYVE-1 and CCL21 together with a high expression of PDPN support the idea that lipomatosis-associated vessels resemble collecting lymphatic vessels. However, we could not confirm the presence of valves or smooth muscle cell coverage.

Lipomatosis causes remodeling of the HEVs associated with reduced density of naïve T-cells

Based on the dramatic loss of lymphatic vessel area and on the known potential effects of adipocytes on vascular homeostasis and inflammation [36], we next analyzed potential effects on the blood vasculature with a focus on the HEVs. HEVs are characterized by their cuboidal endothelial cells and the expression of PNAd, which can be detected by the MECA-79 antibody [14]. We previously showed that patients with invasive breast cancer displayed extensive remodeling of HEVs in the paracortex (T-cell zone) of tumor-draining LNs [21] and applied the same criteria to assess HEV remodeling in lipomatosis, i.e. vessel dilation, thinning of the endothelial cells, and loss of PNAd expression [21]. To avoid cancer-induced effects on HEVs in invasive breast cancer, only patients with noninvasive breast cancer, i.e. DCIS, were included in the analysis. Since medullary HEVs have less well-established functions and are sparse, we analyzed HEVs only in paracortex, and the area was defined by expression of CCL21 in surrounding TRCs, as previously described. When analyzing the difference between HEV dilation in HEVs more than 200 μm from a lipomatosis and lipomatosis-associated paracortex (LAP) within 200 μm from lipomatosis, we detected a clear difference between the locations with significantly higher proportion of dilated HEVs close to adipocytes (Figure 6A,B). Similar results were observed for the thinning of the HEV endothelium and for PNAd loss with a higher proportion of remodeled vessels close to adipose tissue (Figure 6C,D). A positive correlation was found between HEV dilation and thinning of the HEV endothelium as well as between HEV dilation and loss of PNAd expression, suggesting that these vascular changes are linked (Figure 6E,F). Naïve T-cells are continuously recruited through HEVs from blood and distribute across the paracortex guided by CCL21 expression from the surrounding TRCs. To determine whether lipomatosis-associated remodeled HEVs results in a dysfunction in T-cell recruitment, we stained for the nuclear marker TCF1/7, which is constitutively expressed by naïve T-cells, memory cells and in long-lived, rare, CD8+ progenitor cells [37]. Staining for TCF1/7 revealed a lower density of TCF1/7 positive cells in the immediate surrounding of highly dilated HEVs close to adipocytes compared to HEVs in the normal paracortex (Figure 6G,H). This supports the idea that HEV remodeling induced by lipomatosis interferes with lymphocyte recruitment and/or distribution within the LN parenchyma. Although extensive local effects were seen in HEVs in association with parenchymal adipocytes, the extent of LN lipomatosis did not correlate with overall paracortical HEV remodeling in the patients, supporting the notion that LN lipomatosis has local effects on HEV functions (Figure 6I).

As lipomatosis progresses, the medullary cords and sinuses are replaced by adipocytes, which eventually leads to the loss of all normal tissue in this area (Figure 1A right panel). The normal medulla of the LN is characterized by the presence of a dense network of macrophages and stationary plasma cells [38, 39]. To evaluate the possible effects on the immune cell environment in the medulla in the progression of lipomatosis, we stained for the macrophage marker CD68. Medullary areas with early and intermediate lipomatosis showed no overt changes in macrophage distribution compared to unaffected areas (supplementary material, Figure S7). However, analysis of plasma cells, defined by the marker CD38, showed abnormal clustering of cells within the adipose tissue (supplementary material, Figure S8A,B). In normal medulla, plasma cells are scattered across the tissue (supplementary material, Figure S8A).

Discussion

A summary of our findings with hypotheses are outlined in the supplementary material, Figure S9. Although LN lipomatosis is a common phenomenon in the elderly, there are only a few published articles on LN lipomatosis [5, 6], and both the underlying mechanisms and consequences have been unknown. One reason for the lack of mechanistic data is that spontaneous LN lipomatosis is rarely seen in aging mice, where fibrosis instead dominates [40, 41]. To be able to directly address lipomatosis, we therefore instead chose to approach the questions by in situ analysis of human LNs.

Analyzing LNs with different degrees of lipomatosis, we show that it starts from deeper parts of the medullary parenchyma and demonstrate the presence of cells that display a transitional phenotype with both fibroblast and adipocyte lineage marker expression. This suggests that LN lipomatosis is driven by transdifferentiation of medullary fibroblasts into adipocytes. That LN mesenchymal cells can transdifferentiate into adipocytes was previously demonstrated in experimental systems in vivo [19, 31] and for human LN-derived fibroblast in vitro [42]. We show for the first time that specific subsets of LN fibroblasts, i.e. MedRCs and CD34+ SCs, display an inherently higher sensitivity to undergo transdifferentiation into adipocytes ex vivo, compared to BP3+ RC (i.e. TRCs, FDCs, and MRCs). This offers one explanation of why lipomatosis starts in the medullary regions of LNs. Although CD34+ SCs displayed the strongest sensitivity, these cells mainly build up the capsule of the LNs and are only found around the adventitia of large blood vessels inside the LN and in association with the deeper invaginations of trabecular sinuses. At later stages of lipomatosis, it is likely that other subsets, including TRCs, also contribute to adipogensis, possibly driven by advancing disruption of the microenvironment.

It was previously demonstrated that LTΒR signaling can counteract the differentiation of LN SCs into adipocytes in mouse experimental systems in vivo [19, 31], as well as in cell culture of mesenchymal stem cells [43]. A common progenitor for LN fibroblasts and adipocytes was also identified in early development of the LN anlage [19]. Data from the analysis of cultured human preadipocytes also support the notion that the LTBR ligand LIGHT (TNSF14) can counteract adipogenesis [44]. In this context, it is interesting to note that we could confirm reduced expression of LTΒ in areas affected by lipomatosis. LTB forms a ligand for LTBR in complex with LTA [34]. Reduced expression of LTB suggests this could be one contributing factor driving transdifferentiation of fibroblasts into adipocytes in the aging LN. Although B-cells are known to be a major source of LTΒ [34, 35], we did not detect a reduction in B-cell density. Thus, functional changes in medullary B-cells and/or other cell types may contribute to the downregulation of LTB. Although mouse SCs all expressed Ltb, with lower expression in MedRCs (Nr4a1+ SCs and Inmt+ SCs) compared to TRCs and other BP3+ LN RCs (supplementary material, Figure S5B), we did not detect LTB in the human LN SC dataset, and LIGHT (TNFSF14) was also very sparsely expressed (supplementary material, Figure S6). LTA/Lta is not detectable in either human or mouse based on the datasets we analyzed here [10, 26] (mouse http://scorpio.ucsf.edu/LNSC/ and data not shown). It is likely that combinatorial effects driven by microenvironmental changes, such as downregulation of LTΒ, together with a higher sensitivity of MedRC and potentially also rare CD34+ SCs in the medullary areas to undergo adipogenesis, drives the initiation and progression of lipomatosis. In addition to LTBR signaling, the Hippo pathway has been implicated in the regulation of LN SC differentiation [31]. In silico analysis of the dataset published by Rodda et al [10] indeed confirmed the enrichment of genes associated both with NF-kappa B signaling and the Hippo pathway in TRCs, FDCs, and MRCs compared to MedRCs and CD34+ SCs. Differential expression of RELB (NF-kappa B pathway) and BIRC3 (associated with both NF-kappa B and Hippo pathways) in human LN SCs as well, with higher expression in CCL21+ TRCs (supplementary material, Figure S6), suggests at least partly conserved LN fibroblast phenotypes in humans and in mice.

As part of the phenotype in lipomatosis-affected LNs, we observed a dramatic reduction of lymphatic MS area, and the medullary lymphatic endothelium is progressively lost. Why the vessels disappear remains to be determined. The changes we observed in LTB may also affect LTBR signaling in lymphatic vessels, although experimental data do not support major direct effects [45], indirect regulation through signaling in immune cells, such as B-cells, has, however, been indicted, affecting the expression of, for example, PD-L1 on LN lymphatic vessels in mice [46]. Our data also do not support the idea that endothelial cells transform into adipocytes like fibroblasts, but loss of ECM interactions, which are dependent on surrounding fibroblasts, may lead to the apoptosis of LECs and contribute to the observed pathology. However, an experimental system able to mimic the pathology in the mouse would be required to evaluate this hypothesis. A degeneration of LN lymphatic sinuses is expected to result in the obstruction of lymph flow, but we observed the appearance of collecting-like lymphatic vessels inside the areas of the LN affected by lipomatosis, which may explain how LN lipomatosis can progress and affect multiple LNs without causing lymph edema in patients. These vessels were defined by a higher expression of PDPN than most other LN lymphatic vessels, lacked CD36 (marker expressed by paracortical sinuses in the LN [11]), and displayed low or absent MARCO and LYVE-1, both markers of MSs. They also have no detectable expression of CCL21, which is a marker highly expressed by initial capillary lymphatic vessels in the periphery. It is interesting to speculate that these vessels result from the ingrowth of collecting vessels through the destroyed areas of LNs.

Besides the striking effect on the lymphatic vascular MS area, we showed that lipomatosis in the analyzed human LNs also caused extensive local remodeling of HEVs. Loss of LTBR signaling has been linked to dysregulation of HEVs in mouse models [45, 47], and the downregulation of LTB in affected areas that we observed may be a contributing factor, but we cannot exclude additional cooperating mechanisms. Adipocytes express high levels of a number of growth factors [48], including vascular endothelial growth factor A (VEGFA), which might also contribute to disturbing HEV homeostasis. That the effects caused by lipomatosis are local is emphasized by the lack of correlation with paracortical HEV remodeling, a phenomenon we previously showed was common in tumor-draining LNs [20, 21]. Notably, we here also showed that the remodeling of HEVs in lipomatosis was linked to a reduced density of TCF1/7 naïve T-cells around dilated HEVs. In addition to effects on naïve T-cells, we also observed an accumulation of CD38 plasma cells within adipose tissue. CD38+ plasma cells are normally scattered across the medullary cords, and the clusters of cells seen in areas with adipocytes indicate that the cells are trapped in the adipose tissue. The effect was specific to the plasma cells and not seen for the medullary CD68+ macrophages, indicating differential responses of different immune cell subsets to changes in the environment. As lipomatosis progresses, normal LN tissue, including the immune cells, medullary cords, and sinuses, is replaced by adipocytes, which is likely in late stages to cause a complete loss of its important immunogenic functions. However, already in early stages, LN lipomatosis is likely to affect both lymphocyte recruitment into the LNs through HEVs and the recirculation of lymphocytes back to the blood through the lymphatic sinuses, both of which are fundamental functions for inducing adaptive immunity [8]. In addition, experimental data suggest that intact lymphocyte homing is also important for the maintenance of normal peripheral lymphocyte homeostasis [49]. Consequently, LN lipomatosis is highly likely to have an impact on vaccination in the elderly by reducing access to functional lymphoid tissues. Clinical studies in elderly patients with different degrees of LN lipomatosis are, however, needed to confirm this hypothesis. Of interest in this context, there are indications from older literature that LNs in different draining areas are affected to different degrees [5], so it can be speculated that the site of vaccination administration may be a way to circumvent some of the effects caused.

From the perspective of cancer development, adipose tissue has attracted considerable attention [50]. Adipocytes can contribute to a disturbed microenvironment, promoting tumor formation, and have been speculated to contribute to a premetastatic niche in, for example, the BM [51]. The BM, similar to the LN, displays infiltration of adipocytes with age and, consequently, displays a reduced output of hematopoietic stem cells [51, 52]. It cannot be excluded that LN lipomatosis contributes to a premetastatic niche in the context of nodal metastasis, so patients with high degree of LN lipomatosis may be more likely to develop growth of metastases inside LNs. Evaluation of this possibility in human cancer would be of great interest in future studies, but this is beyond the scope of this paper.

The loss of LN parenchyma due to lipomatosis is highly likely to interfere with the induction of adaptive immunity and is likely to contribute to a reduced immune status of the aging individual. This may have multitude influences on the ability to respond to infections and vaccination as well as diminished ability to mount antitumor responses in cancer, highlighting a need for continued research.

Acknowledgements

The authors thank Professor Olle Korsgren, Uppsala University, for access to pancreatic lymph nodes from organ donors and Daniel Vasiliu Bacovia, Akademiska Sjukhuset Uppsala, for selecting patients for analysis. We thank Kalyani Vemuri and Anna Dimberg, Department of Immunology, Genetics and Pathology (IGP), Uppsala University, for advice and support on antibodies for fibroblasts. We also thank Liqun He, Uppsala University, for the initial bioinformatic analysis. This research was funded by the Swedish Research Council (2016-02492), Swedish Cancer Foundation (2017/759 and 20 0970 PjF), Kjell and Märta Beijer Foundation, and PO Zetterling Foundation to Ulvmar MH.

    Author contributions statement

    MHU and TB conceived of the project. TB, YS and AO performed analysis and made the illustrations. YS performed bioinformatical analysis, PUM gave advice. MHU and TB wrote the original manuscript. All authors contributed to editing of the manuscript. MHU provided the funding for the work. All authors read and agreed to the manuscript.

    Data availability statement

    Codes are available at https://github.com/sherriying/lipomatosis_in_LN.